aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86
diff options
context:
space:
mode:
authorArjan van de Ven <arjan@linux.intel.com>2008-10-17 12:20:26 -0400
committerArjan van de Ven <arjan@linux.intel.com>2008-10-17 12:20:26 -0400
commit651dab4264e4ba0e563f5ff56f748127246e9065 (patch)
tree016630974bdcb00fe529b673f96d389e0fd6dc94 /arch/x86
parent40b8606253552109815786e5d4b0de98782d31f5 (diff)
parent2e532d68a2b3e2aa6b19731501222069735c741c (diff)
Merge commit 'linus/master' into merge-linus
Conflicts: arch/x86/kvm/i8254.c
Diffstat (limited to 'arch/x86')
-rw-r--r--arch/x86/Kconfig207
-rw-r--r--arch/x86/Kconfig.cpu162
-rw-r--r--arch/x86/Kconfig.debug13
-rw-r--r--arch/x86/Makefile_32.cpu6
-rw-r--r--arch/x86/boot/Makefile5
-rw-r--r--arch/x86/boot/compressed/Makefile8
-rw-r--r--arch/x86/boot/compressed/head_32.S5
-rw-r--r--arch/x86/boot/compressed/misc.c12
-rw-r--r--arch/x86/boot/compressed/relocs.c2
-rw-r--r--arch/x86/boot/cpu.c17
-rw-r--r--arch/x86/boot/cpucheck.c8
-rw-r--r--arch/x86/boot/edd.c7
-rw-r--r--arch/x86/boot/header.S1
-rw-r--r--arch/x86/boot/mkcpustr.c40
-rw-r--r--arch/x86/boot/video-vesa.c11
-rw-r--r--arch/x86/configs/i386_defconfig21
-rw-r--r--arch/x86/configs/x86_64_defconfig31
-rw-r--r--arch/x86/crypto/Makefile2
-rw-r--r--arch/x86/crypto/crc32c-intel.c197
-rw-r--r--arch/x86/ia32/ia32_aout.c11
-rw-r--r--arch/x86/ia32/ia32_signal.c138
-rw-r--r--arch/x86/ia32/ia32entry.S30
-rw-r--r--arch/x86/ia32/sys_ia32.c108
-rw-r--r--arch/x86/kernel/Makefile17
-rw-r--r--arch/x86/kernel/acpi/boot.c41
-rw-r--r--arch/x86/kernel/alternative.c46
-rw-r--r--arch/x86/kernel/amd_iommu.c359
-rw-r--r--arch/x86/kernel/amd_iommu_init.c194
-rw-r--r--arch/x86/kernel/aperture_64.c6
-rw-r--r--arch/x86/kernel/apic_32.c441
-rw-r--r--arch/x86/kernel/apic_64.c630
-rw-r--r--arch/x86/kernel/apm_32.c4
-rw-r--r--arch/x86/kernel/asm-offsets_64.c2
-rw-r--r--arch/x86/kernel/bios_uv.c10
-rw-r--r--arch/x86/kernel/cpu/Makefile34
-rw-r--r--arch/x86/kernel/cpu/addon_cpuid_features.c88
-rw-r--r--arch/x86/kernel/cpu/amd.c547
-rw-r--r--arch/x86/kernel/cpu/amd_64.c224
-rw-r--r--arch/x86/kernel/cpu/centaur.c15
-rw-r--r--arch/x86/kernel/cpu/centaur_64.c6
-rw-r--r--arch/x86/kernel/cpu/cmpxchg.c72
-rw-r--r--arch/x86/kernel/cpu/common.c1009
-rw-r--r--arch/x86/kernel/cpu/common_64.c670
-rw-r--r--arch/x86/kernel/cpu/cpu.h19
-rw-r--r--arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c13
-rw-r--r--arch/x86/kernel/cpu/cpufreq/elanfreq.c42
-rw-r--r--arch/x86/kernel/cpu/cpufreq/p4-clockmod.c2
-rw-r--r--arch/x86/kernel/cpu/cpufreq/powernow-k6.c41
-rw-r--r--arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c2
-rw-r--r--arch/x86/kernel/cpu/cyrix.c55
-rw-r--r--arch/x86/kernel/cpu/feature_names.c83
-rw-r--r--arch/x86/kernel/cpu/intel.c365
-rw-r--r--arch/x86/kernel/cpu/intel_64.c95
-rw-r--r--arch/x86/kernel/cpu/intel_cacheinfo.c169
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_64.c2
-rw-r--r--arch/x86/kernel/cpu/mkcapflags.pl32
-rw-r--r--arch/x86/kernel/cpu/mtrr/generic.c7
-rw-r--r--arch/x86/kernel/cpu/mtrr/if.c4
-rw-r--r--arch/x86/kernel/cpu/mtrr/main.c276
-rw-r--r--arch/x86/kernel/cpu/perfctr-watchdog.c86
-rw-r--r--arch/x86/kernel/cpu/powerflags.c20
-rw-r--r--arch/x86/kernel/cpu/transmeta.c32
-rw-r--r--arch/x86/kernel/cpu/umc.c3
-rw-r--r--arch/x86/kernel/cpuid.c5
-rw-r--r--arch/x86/kernel/crash_dump_64.c13
-rw-r--r--arch/x86/kernel/doublefault_32.c2
-rw-r--r--arch/x86/kernel/ds.c954
-rw-r--r--arch/x86/kernel/dumpstack_32.c449
-rw-r--r--arch/x86/kernel/dumpstack_64.c575
-rw-r--r--arch/x86/kernel/e820.c34
-rw-r--r--arch/x86/kernel/early-quirks.c66
-rw-r--r--arch/x86/kernel/early_printk.c748
-rw-r--r--arch/x86/kernel/efi.c6
-rw-r--r--arch/x86/kernel/entry_32.S22
-rw-r--r--arch/x86/kernel/entry_64.S19
-rw-r--r--arch/x86/kernel/es7000_32.c (renamed from arch/x86/mach-es7000/es7000plat.c)115
-rw-r--r--arch/x86/kernel/genapic_64.c88
-rw-r--r--arch/x86/kernel/genapic_flat_64.c62
-rw-r--r--arch/x86/kernel/genx2apic_cluster.c159
-rw-r--r--arch/x86/kernel/genx2apic_phys.c154
-rw-r--r--arch/x86/kernel/genx2apic_uv_x.c93
-rw-r--r--arch/x86/kernel/head.c1
-rw-r--r--arch/x86/kernel/head64.c5
-rw-r--r--arch/x86/kernel/head_32.S34
-rw-r--r--arch/x86/kernel/head_64.S4
-rw-r--r--arch/x86/kernel/hpet.c25
-rw-r--r--arch/x86/kernel/i387.c168
-rw-r--r--arch/x86/kernel/i8259.c24
-rw-r--r--arch/x86/kernel/io_apic_32.c47
-rw-r--r--arch/x86/kernel/io_apic_64.c639
-rw-r--r--arch/x86/kernel/ioport.c1
-rw-r--r--arch/x86/kernel/ipi.c3
-rw-r--r--arch/x86/kernel/irq_32.c2
-rw-r--r--arch/x86/kernel/irq_64.c2
-rw-r--r--arch/x86/kernel/irqinit_32.c49
-rw-r--r--arch/x86/kernel/irqinit_64.c43
-rw-r--r--arch/x86/kernel/k8.c5
-rw-r--r--arch/x86/kernel/kdebugfs.c1
-rw-r--r--arch/x86/kernel/kgdb.c50
-rw-r--r--arch/x86/kernel/kvm.c2
-rw-r--r--arch/x86/kernel/kvmclock.c30
-rw-r--r--arch/x86/kernel/ldt.c10
-rw-r--r--arch/x86/kernel/microcode.c853
-rw-r--r--arch/x86/kernel/microcode_amd.c435
-rw-r--r--arch/x86/kernel/microcode_core.c508
-rw-r--r--arch/x86/kernel/microcode_intel.c480
-rw-r--r--arch/x86/kernel/mpparse.c2
-rw-r--r--arch/x86/kernel/msr.c4
-rw-r--r--arch/x86/kernel/nmi.c11
-rw-r--r--arch/x86/kernel/numaq_32.c7
-rw-r--r--arch/x86/kernel/olpc.c6
-rw-r--r--arch/x86/kernel/paravirt-spinlocks.c37
-rw-r--r--arch/x86/kernel/paravirt.c30
-rw-r--r--arch/x86/kernel/paravirt_patch_32.c2
-rw-r--r--arch/x86/kernel/pci-calgary_64.c36
-rw-r--r--arch/x86/kernel/pci-dma.c183
-rw-r--r--arch/x86/kernel/pci-gart_64.c170
-rw-r--r--arch/x86/kernel/pci-nommu.c10
-rw-r--r--arch/x86/kernel/pcspeaker.c13
-rw-r--r--arch/x86/kernel/process.c21
-rw-r--r--arch/x86/kernel/process_32.c105
-rw-r--r--arch/x86/kernel/process_64.c197
-rw-r--r--arch/x86/kernel/ptrace.c522
-rw-r--r--arch/x86/kernel/pvclock.c12
-rw-r--r--arch/x86/kernel/quirks.c41
-rw-r--r--arch/x86/kernel/reboot.c6
-rw-r--r--arch/x86/kernel/rtc.c22
-rw-r--r--arch/x86/kernel/setup.c229
-rw-r--r--arch/x86/kernel/setup_percpu.c9
-rw-r--r--arch/x86/kernel/sigframe.h19
-rw-r--r--arch/x86/kernel/signal_32.c273
-rw-r--r--arch/x86/kernel/signal_64.c362
-rw-r--r--arch/x86/kernel/smp.c6
-rw-r--r--arch/x86/kernel/smpboot.c224
-rw-r--r--arch/x86/kernel/summit_32.c2
-rw-r--r--arch/x86/kernel/sys_i386_32.c2
-rw-r--r--arch/x86/kernel/sys_x86_64.c44
-rw-r--r--arch/x86/kernel/syscall_64.c4
-rw-r--r--arch/x86/kernel/time_32.c8
-rw-r--r--arch/x86/kernel/time_64.c23
-rw-r--r--arch/x86/kernel/tlb_32.c8
-rw-r--r--arch/x86/kernel/tls.c1
-rw-r--r--arch/x86/kernel/traps.c (renamed from arch/x86/kernel/traps_32.c)896
-rw-r--r--arch/x86/kernel/traps_64.c1212
-rw-r--r--arch/x86/kernel/tsc.c290
-rw-r--r--arch/x86/kernel/visws_quirks.c16
-rw-r--r--arch/x86/kernel/vm86_32.c1
-rw-r--r--arch/x86/kernel/vmi_32.c16
-rw-r--r--arch/x86/kernel/vmlinux_32.lds.S9
-rw-r--r--arch/x86/kernel/vmlinux_64.lds.S9
-rw-r--r--arch/x86/kernel/vsmp_64.c2
-rw-r--r--arch/x86/kernel/xsave.c345
-rw-r--r--arch/x86/kvm/Makefile5
-rw-r--r--arch/x86/kvm/i8254.c83
-rw-r--r--arch/x86/kvm/i8254.h7
-rw-r--r--arch/x86/kvm/i8259.c53
-rw-r--r--arch/x86/kvm/irq.c3
-rw-r--r--arch/x86/kvm/irq.h6
-rw-r--r--arch/x86/kvm/kvm_cache_regs.h32
-rw-r--r--arch/x86/kvm/lapic.c43
-rw-r--r--arch/x86/kvm/mmu.c684
-rw-r--r--arch/x86/kvm/paging_tmpl.h249
-rw-r--r--arch/x86/kvm/svm.c168
-rw-r--r--arch/x86/kvm/vmx.c715
-rw-r--r--arch/x86/kvm/vmx.h20
-rw-r--r--arch/x86/kvm/x86.c554
-rw-r--r--arch/x86/kvm/x86.h22
-rw-r--r--arch/x86/kvm/x86_emulate.c170
-rw-r--r--arch/x86/lguest/boot.c38
-rw-r--r--arch/x86/lib/Makefile3
-rw-r--r--arch/x86/lib/msr-on-cpu.c78
-rw-r--r--arch/x86/lib/string_32.c42
-rw-r--r--arch/x86/lib/strstr_32.c6
-rw-r--r--arch/x86/lib/usercopy_32.c7
-rw-r--r--arch/x86/mach-default/setup.c19
-rw-r--r--arch/x86/mach-es7000/Makefile5
-rw-r--r--arch/x86/mach-es7000/es7000.h114
-rw-r--r--arch/x86/mach-generic/Makefile1
-rw-r--r--arch/x86/mach-generic/bigsmp.c9
-rw-r--r--arch/x86/mach-generic/es7000.c33
-rw-r--r--arch/x86/mach-generic/numaq.c12
-rw-r--r--arch/x86/mach-generic/summit.c11
-rw-r--r--arch/x86/mach-voyager/voyager_smp.c2
-rw-r--r--arch/x86/mm/Makefile6
-rw-r--r--arch/x86/mm/dump_pagetables.c4
-rw-r--r--arch/x86/mm/fault.c67
-rw-r--r--arch/x86/mm/gup.c10
-rw-r--r--arch/x86/mm/init_32.c94
-rw-r--r--arch/x86/mm/init_64.c184
-rw-r--r--arch/x86/mm/ioremap.c197
-rw-r--r--arch/x86/mm/numa_32.c (renamed from arch/x86/mm/discontig_32.c)2
-rw-r--r--arch/x86/mm/numa_64.c10
-rw-r--r--arch/x86/mm/pageattr-test.c9
-rw-r--r--arch/x86/mm/pageattr.c463
-rw-r--r--arch/x86/mm/pat.c132
-rw-r--r--arch/x86/mm/pgtable.c6
-rw-r--r--arch/x86/mm/pgtable_32.c3
-rw-r--r--arch/x86/mm/srat_64.c2
-rw-r--r--arch/x86/oprofile/Makefile2
-rw-r--r--arch/x86/oprofile/nmi_int.c31
-rw-r--r--arch/x86/oprofile/op_model_amd.c543
-rw-r--r--arch/x86/oprofile/op_model_athlon.c190
-rw-r--r--arch/x86/oprofile/op_model_p4.c175
-rw-r--r--arch/x86/oprofile/op_x86_model.h4
-rw-r--r--arch/x86/pci/acpi.c5
-rw-r--r--arch/x86/pci/amd_bus.c2
-rw-r--r--arch/x86/pci/fixup.c28
-rw-r--r--arch/x86/pci/i386.c3
-rw-r--r--arch/x86/pci/irq.c67
-rw-r--r--arch/x86/pci/mmconfig-shared.c12
-rw-r--r--arch/x86/power/cpu_32.c7
-rw-r--r--arch/x86/power/cpu_64.c7
-rw-r--r--arch/x86/power/hibernate_asm_32.S14
-rw-r--r--arch/x86/xen/Kconfig12
-rw-r--r--arch/x86/xen/Makefile12
-rw-r--r--arch/x86/xen/debugfs.c123
-rw-r--r--arch/x86/xen/debugfs.h10
-rw-r--r--arch/x86/xen/enlighten.c319
-rw-r--r--arch/x86/xen/irq.c143
-rw-r--r--arch/x86/xen/mmu.c314
-rw-r--r--arch/x86/xen/mmu.h3
-rw-r--r--arch/x86/xen/multicalls.c115
-rw-r--r--arch/x86/xen/setup.c2
-rw-r--r--arch/x86/xen/smp.c245
-rw-r--r--arch/x86/xen/spinlock.c428
-rw-r--r--arch/x86/xen/time.c23
-rw-r--r--arch/x86/xen/xen-asm_32.S2
-rw-r--r--arch/x86/xen/xen-asm_64.S22
-rw-r--r--arch/x86/xen/xen-ops.h8
229 files changed, 17032 insertions, 9459 deletions
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index ed92864d1325..bd3c2c53873e 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -18,6 +18,7 @@ config X86_64
18### Arch settings 18### Arch settings
19config X86 19config X86
20 def_bool y 20 def_bool y
21 select HAVE_AOUT if X86_32
21 select HAVE_UNSTABLE_SCHED_CLOCK 22 select HAVE_UNSTABLE_SCHED_CLOCK
22 select HAVE_IDE 23 select HAVE_IDE
23 select HAVE_OPROFILE 24 select HAVE_OPROFILE
@@ -29,6 +30,7 @@ config X86
29 select HAVE_FTRACE 30 select HAVE_FTRACE
30 select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64) 31 select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64)
31 select HAVE_ARCH_KGDB if !X86_VOYAGER 32 select HAVE_ARCH_KGDB if !X86_VOYAGER
33 select HAVE_ARCH_TRACEHOOK
32 select HAVE_GENERIC_DMA_COHERENT if X86_32 34 select HAVE_GENERIC_DMA_COHERENT if X86_32
33 select HAVE_EFFICIENT_UNALIGNED_ACCESS 35 select HAVE_EFFICIENT_UNALIGNED_ACCESS
34 36
@@ -37,10 +39,6 @@ config ARCH_DEFCONFIG
37 default "arch/x86/configs/i386_defconfig" if X86_32 39 default "arch/x86/configs/i386_defconfig" if X86_32
38 default "arch/x86/configs/x86_64_defconfig" if X86_64 40 default "arch/x86/configs/x86_64_defconfig" if X86_64
39 41
40
41config GENERIC_LOCKBREAK
42 def_bool n
43
44config GENERIC_TIME 42config GENERIC_TIME
45 def_bool y 43 def_bool y
46 44
@@ -93,7 +91,7 @@ config GENERIC_HWEIGHT
93 def_bool y 91 def_bool y
94 92
95config GENERIC_GPIO 93config GENERIC_GPIO
96 def_bool n 94 bool
97 95
98config ARCH_MAY_HAVE_PC_FDC 96config ARCH_MAY_HAVE_PC_FDC
99 def_bool y 97 def_bool y
@@ -104,12 +102,6 @@ config RWSEM_GENERIC_SPINLOCK
104config RWSEM_XCHGADD_ALGORITHM 102config RWSEM_XCHGADD_ALGORITHM
105 def_bool X86_XADD 103 def_bool X86_XADD
106 104
107config ARCH_HAS_ILOG2_U32
108 def_bool n
109
110config ARCH_HAS_ILOG2_U64
111 def_bool n
112
113config ARCH_HAS_CPU_IDLE_WAIT 105config ARCH_HAS_CPU_IDLE_WAIT
114 def_bool y 106 def_bool y
115 107
@@ -151,9 +143,6 @@ config AUDIT_ARCH
151 bool 143 bool
152 default X86_64 144 default X86_64
153 145
154config ARCH_SUPPORTS_AOUT
155 def_bool y
156
157config ARCH_SUPPORTS_OPTIMIZED_INLINING 146config ARCH_SUPPORTS_OPTIMIZED_INLINING
158 def_bool y 147 def_bool y
159 148
@@ -553,6 +542,7 @@ config CALGARY_IOMMU_ENABLED_BY_DEFAULT
553config AMD_IOMMU 542config AMD_IOMMU
554 bool "AMD IOMMU support" 543 bool "AMD IOMMU support"
555 select SWIOTLB 544 select SWIOTLB
545 select PCI_MSI
556 depends on X86_64 && PCI && ACPI 546 depends on X86_64 && PCI && ACPI
557 help 547 help
558 With this option you can enable support for AMD IOMMU hardware in 548 With this option you can enable support for AMD IOMMU hardware in
@@ -758,9 +748,8 @@ config I8K
758 Say N otherwise. 748 Say N otherwise.
759 749
760config X86_REBOOTFIXUPS 750config X86_REBOOTFIXUPS
761 def_bool n 751 bool "Enable X86 board specific fixups for reboot"
762 prompt "Enable X86 board specific fixups for reboot" 752 depends on X86_32
763 depends on X86_32 && X86
764 ---help--- 753 ---help---
765 This enables chipset and/or board specific fixups to be done 754 This enables chipset and/or board specific fixups to be done
766 in order to get reboot to work correctly. This is only needed on 755 in order to get reboot to work correctly. This is only needed on
@@ -776,23 +765,45 @@ config X86_REBOOTFIXUPS
776 Say N otherwise. 765 Say N otherwise.
777 766
778config MICROCODE 767config MICROCODE
779 tristate "/dev/cpu/microcode - Intel IA32 CPU microcode support" 768 tristate "/dev/cpu/microcode - microcode support"
780 select FW_LOADER 769 select FW_LOADER
781 ---help--- 770 ---help---
782 If you say Y here, you will be able to update the microcode on 771 If you say Y here, you will be able to update the microcode on
783 Intel processors in the IA32 family, e.g. Pentium Pro, Pentium II, 772 certain Intel and AMD processors. The Intel support is for the
784 Pentium III, Pentium 4, Xeon etc. You will obviously need the 773 IA32 family, e.g. Pentium Pro, Pentium II, Pentium III,
785 actual microcode binary data itself which is not shipped with the 774 Pentium 4, Xeon etc. The AMD support is for family 0x10 and
786 Linux kernel. 775 0x11 processors, e.g. Opteron, Phenom and Turion 64 Ultra.
776 You will obviously need the actual microcode binary data itself
777 which is not shipped with the Linux kernel.
787 778
788 For latest news and information on obtaining all the required 779 This option selects the general module only, you need to select
789 ingredients for this driver, check: 780 at least one vendor specific module as well.
790 <http://www.urbanmyth.org/microcode/>.
791 781
792 To compile this driver as a module, choose M here: the 782 To compile this driver as a module, choose M here: the
793 module will be called microcode. 783 module will be called microcode.
794 784
795config MICROCODE_OLD_INTERFACE 785config MICROCODE_INTEL
786 bool "Intel microcode patch loading support"
787 depends on MICROCODE
788 default MICROCODE
789 select FW_LOADER
790 --help---
791 This options enables microcode patch loading support for Intel
792 processors.
793
794 For latest news and information on obtaining all the required
795 Intel ingredients for this driver, check:
796 <http://www.urbanmyth.org/microcode/>.
797
798config MICROCODE_AMD
799 bool "AMD microcode patch loading support"
800 depends on MICROCODE
801 select FW_LOADER
802 --help---
803 If you select this option, microcode patch loading support for AMD
804 processors will be enabled.
805
806 config MICROCODE_OLD_INTERFACE
796 def_bool y 807 def_bool y
797 depends on MICROCODE 808 depends on MICROCODE
798 809
@@ -922,16 +933,17 @@ config HIGHMEM
922 depends on X86_32 && (HIGHMEM64G || HIGHMEM4G) 933 depends on X86_32 && (HIGHMEM64G || HIGHMEM4G)
923 934
924config X86_PAE 935config X86_PAE
925 def_bool n 936 bool "PAE (Physical Address Extension) Support"
926 prompt "PAE (Physical Address Extension) Support"
927 depends on X86_32 && !HIGHMEM4G 937 depends on X86_32 && !HIGHMEM4G
928 select RESOURCES_64BIT
929 help 938 help
930 PAE is required for NX support, and furthermore enables 939 PAE is required for NX support, and furthermore enables
931 larger swapspace support for non-overcommit purposes. It 940 larger swapspace support for non-overcommit purposes. It
932 has the cost of more pagetable lookup overhead, and also 941 has the cost of more pagetable lookup overhead, and also
933 consumes more pagetable space per process. 942 consumes more pagetable space per process.
934 943
944config ARCH_PHYS_ADDR_T_64BIT
945 def_bool X86_64 || X86_PAE
946
935# Common NUMA Features 947# Common NUMA Features
936config NUMA 948config NUMA
937 bool "Numa Memory Allocation and Scheduler Support (EXPERIMENTAL)" 949 bool "Numa Memory Allocation and Scheduler Support (EXPERIMENTAL)"
@@ -1020,7 +1032,7 @@ config HAVE_ARCH_ALLOC_REMAP
1020 1032
1021config ARCH_FLATMEM_ENABLE 1033config ARCH_FLATMEM_ENABLE
1022 def_bool y 1034 def_bool y
1023 depends on X86_32 && ARCH_SELECT_MEMORY_MODEL && X86_PC && !NUMA 1035 depends on X86_32 && ARCH_SELECT_MEMORY_MODEL && !NUMA
1024 1036
1025config ARCH_DISCONTIGMEM_ENABLE 1037config ARCH_DISCONTIGMEM_ENABLE
1026 def_bool y 1038 def_bool y
@@ -1036,7 +1048,7 @@ config ARCH_SPARSEMEM_DEFAULT
1036 1048
1037config ARCH_SPARSEMEM_ENABLE 1049config ARCH_SPARSEMEM_ENABLE
1038 def_bool y 1050 def_bool y
1039 depends on X86_64 || NUMA || (EXPERIMENTAL && X86_PC) 1051 depends on X86_64 || NUMA || (EXPERIMENTAL && X86_PC) || X86_GENERICARCH
1040 select SPARSEMEM_STATIC if X86_32 1052 select SPARSEMEM_STATIC if X86_32
1041 select SPARSEMEM_VMEMMAP_ENABLE if X86_64 1053 select SPARSEMEM_VMEMMAP_ENABLE if X86_64
1042 1054
@@ -1059,6 +1071,56 @@ config HIGHPTE
1059 low memory. Setting this option will put user-space page table 1071 low memory. Setting this option will put user-space page table
1060 entries in high memory. 1072 entries in high memory.
1061 1073
1074config X86_CHECK_BIOS_CORRUPTION
1075 bool "Check for low memory corruption"
1076 help
1077 Periodically check for memory corruption in low memory, which
1078 is suspected to be caused by BIOS. Even when enabled in the
1079 configuration, it is disabled at runtime. Enable it by
1080 setting "memory_corruption_check=1" on the kernel command
1081 line. By default it scans the low 64k of memory every 60
1082 seconds; see the memory_corruption_check_size and
1083 memory_corruption_check_period parameters in
1084 Documentation/kernel-parameters.txt to adjust this.
1085
1086 When enabled with the default parameters, this option has
1087 almost no overhead, as it reserves a relatively small amount
1088 of memory and scans it infrequently. It both detects corruption
1089 and prevents it from affecting the running system.
1090
1091 It is, however, intended as a diagnostic tool; if repeatable
1092 BIOS-originated corruption always affects the same memory,
1093 you can use memmap= to prevent the kernel from using that
1094 memory.
1095
1096config X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
1097 bool "Set the default setting of memory_corruption_check"
1098 depends on X86_CHECK_BIOS_CORRUPTION
1099 default y
1100 help
1101 Set whether the default state of memory_corruption_check is
1102 on or off.
1103
1104config X86_RESERVE_LOW_64K
1105 bool "Reserve low 64K of RAM on AMI/Phoenix BIOSen"
1106 default y
1107 help
1108 Reserve the first 64K of physical RAM on BIOSes that are known
1109 to potentially corrupt that memory range. A numbers of BIOSes are
1110 known to utilize this area during suspend/resume, so it must not
1111 be used by the kernel.
1112
1113 Set this to N if you are absolutely sure that you trust the BIOS
1114 to get all its memory reservations and usages right.
1115
1116 If you have doubts about the BIOS (e.g. suspend/resume does not
1117 work or there's kernel crashes after certain hardware hotplug
1118 events) and it's not AMI or Phoenix, then you might want to enable
1119 X86_CHECK_BIOS_CORRUPTION=y to allow the kernel to check typical
1120 corruption patterns.
1121
1122 Say Y if unsure.
1123
1062config MATH_EMULATION 1124config MATH_EMULATION
1063 bool 1125 bool
1064 prompt "Math emulation" if X86_32 1126 prompt "Math emulation" if X86_32
@@ -1117,10 +1179,10 @@ config MTRR
1117 You can safely say Y even if your machine doesn't have MTRRs, you'll 1179 You can safely say Y even if your machine doesn't have MTRRs, you'll
1118 just add about 9 KB to your kernel. 1180 just add about 9 KB to your kernel.
1119 1181
1120 See <file:Documentation/mtrr.txt> for more information. 1182 See <file:Documentation/x86/mtrr.txt> for more information.
1121 1183
1122config MTRR_SANITIZER 1184config MTRR_SANITIZER
1123 bool 1185 def_bool y
1124 prompt "MTRR cleanup support" 1186 prompt "MTRR cleanup support"
1125 depends on MTRR 1187 depends on MTRR
1126 help 1188 help
@@ -1131,7 +1193,7 @@ config MTRR_SANITIZER
1131 The largest mtrr entry size for a continous block can be set with 1193 The largest mtrr entry size for a continous block can be set with
1132 mtrr_chunk_size. 1194 mtrr_chunk_size.
1133 1195
1134 If unsure, say N. 1196 If unsure, say Y.
1135 1197
1136config MTRR_SANITIZER_ENABLE_DEFAULT 1198config MTRR_SANITIZER_ENABLE_DEFAULT
1137 int "MTRR cleanup enable value (0-1)" 1199 int "MTRR cleanup enable value (0-1)"
@@ -1166,8 +1228,7 @@ config X86_PAT
1166 If unsure, say Y. 1228 If unsure, say Y.
1167 1229
1168config EFI 1230config EFI
1169 def_bool n 1231 bool "EFI runtime service support"
1170 prompt "EFI runtime service support"
1171 depends on ACPI 1232 depends on ACPI
1172 ---help--- 1233 ---help---
1173 This enables the kernel to use EFI runtime services that are 1234 This enables the kernel to use EFI runtime services that are
@@ -1191,7 +1252,6 @@ config IRQBALANCE
1191config SECCOMP 1252config SECCOMP
1192 def_bool y 1253 def_bool y
1193 prompt "Enable seccomp to safely compute untrusted bytecode" 1254 prompt "Enable seccomp to safely compute untrusted bytecode"
1194 depends on PROC_FS
1195 help 1255 help
1196 This kernel feature is useful for number crunching applications 1256 This kernel feature is useful for number crunching applications
1197 that may need to compute untrusted bytecode during their 1257 that may need to compute untrusted bytecode during their
@@ -1199,7 +1259,7 @@ config SECCOMP
1199 the process as file descriptors supporting the read/write 1259 the process as file descriptors supporting the read/write
1200 syscalls, it's possible to isolate those applications in 1260 syscalls, it's possible to isolate those applications in
1201 their own address space using seccomp. Once seccomp is 1261 their own address space using seccomp. Once seccomp is
1202 enabled via /proc/<pid>/seccomp, it cannot be disabled 1262 enabled via prctl(PR_SET_SECCOMP), it cannot be disabled
1203 and the task is only allowed to execute a few safe syscalls 1263 and the task is only allowed to execute a few safe syscalls
1204 defined by each seccomp mode. 1264 defined by each seccomp mode.
1205 1265
@@ -1356,14 +1416,14 @@ config PHYSICAL_ALIGN
1356 Don't change this unless you know what you are doing. 1416 Don't change this unless you know what you are doing.
1357 1417
1358config HOTPLUG_CPU 1418config HOTPLUG_CPU
1359 bool "Support for suspend on SMP and hot-pluggable CPUs (EXPERIMENTAL)" 1419 bool "Support for hot-pluggable CPUs"
1360 depends on SMP && HOTPLUG && EXPERIMENTAL && !X86_VOYAGER 1420 depends on SMP && HOTPLUG && !X86_VOYAGER
1361 ---help--- 1421 ---help---
1362 Say Y here to experiment with turning CPUs off and on, and to 1422 Say Y here to allow turning CPUs off and on. CPUs can be
1363 enable suspend on SMP systems. CPUs can be controlled through 1423 controlled through /sys/devices/system/cpu.
1364 /sys/devices/system/cpu. 1424 ( Note: power management support will enable this option
1365 Say N if you want to disable CPU hotplug and don't need to 1425 automatically on SMP systems. )
1366 suspend. 1426 Say N if you want to disable CPU hotplug.
1367 1427
1368config COMPAT_VDSO 1428config COMPAT_VDSO
1369 def_bool y 1429 def_bool y
@@ -1378,6 +1438,51 @@ config COMPAT_VDSO
1378 1438
1379 If unsure, say Y. 1439 If unsure, say Y.
1380 1440
1441config CMDLINE_BOOL
1442 bool "Built-in kernel command line"
1443 default n
1444 help
1445 Allow for specifying boot arguments to the kernel at
1446 build time. On some systems (e.g. embedded ones), it is
1447 necessary or convenient to provide some or all of the
1448 kernel boot arguments with the kernel itself (that is,
1449 to not rely on the boot loader to provide them.)
1450
1451 To compile command line arguments into the kernel,
1452 set this option to 'Y', then fill in the
1453 the boot arguments in CONFIG_CMDLINE.
1454
1455 Systems with fully functional boot loaders (i.e. non-embedded)
1456 should leave this option set to 'N'.
1457
1458config CMDLINE
1459 string "Built-in kernel command string"
1460 depends on CMDLINE_BOOL
1461 default ""
1462 help
1463 Enter arguments here that should be compiled into the kernel
1464 image and used at boot time. If the boot loader provides a
1465 command line at boot time, it is appended to this string to
1466 form the full kernel command line, when the system boots.
1467
1468 However, you can use the CONFIG_CMDLINE_OVERRIDE option to
1469 change this behavior.
1470
1471 In most cases, the command line (whether built-in or provided
1472 by the boot loader) should specify the device for the root
1473 file system.
1474
1475config CMDLINE_OVERRIDE
1476 bool "Built-in command line overrides boot loader arguments"
1477 default n
1478 depends on CMDLINE_BOOL
1479 help
1480 Set this option to 'Y' to have the kernel ignore the boot loader
1481 command line, and use ONLY the built-in command line.
1482
1483 This is used to work around broken boot loaders. This should
1484 be set to 'N' under normal conditions.
1485
1381endmenu 1486endmenu
1382 1487
1383config ARCH_ENABLE_MEMORY_HOTPLUG 1488config ARCH_ENABLE_MEMORY_HOTPLUG
@@ -1643,6 +1748,14 @@ config DMAR_FLOPPY_WA
1643 workaround will setup a 1:1 mapping for the first 1748 workaround will setup a 1:1 mapping for the first
1644 16M to make floppy (an ISA device) work. 1749 16M to make floppy (an ISA device) work.
1645 1750
1751config INTR_REMAP
1752 bool "Support for Interrupt Remapping (EXPERIMENTAL)"
1753 depends on X86_64 && X86_IO_APIC && PCI_MSI && ACPI && EXPERIMENTAL
1754 help
1755 Supports Interrupt remapping for IO-APIC and MSI devices.
1756 To use x2apic mode in the CPU's which support x2APIC enhancements or
1757 to support platforms with CPU's having > 8 bit APIC ID, say Y.
1758
1646source "drivers/pci/pcie/Kconfig" 1759source "drivers/pci/pcie/Kconfig"
1647 1760
1648source "drivers/pci/Kconfig" 1761source "drivers/pci/Kconfig"
@@ -1759,7 +1872,7 @@ config IA32_EMULATION
1759 1872
1760config IA32_AOUT 1873config IA32_AOUT
1761 tristate "IA32 a.out support" 1874 tristate "IA32 a.out support"
1762 depends on IA32_EMULATION && ARCH_SUPPORTS_AOUT 1875 depends on IA32_EMULATION
1763 help 1876 help
1764 Support old a.out binaries in the 32bit emulation. 1877 Support old a.out binaries in the 32bit emulation.
1765 1878
@@ -1773,7 +1886,7 @@ config COMPAT_FOR_U64_ALIGNMENT
1773 1886
1774config SYSVIPC_COMPAT 1887config SYSVIPC_COMPAT
1775 def_bool y 1888 def_bool y
1776 depends on X86_64 && COMPAT && SYSVIPC 1889 depends on COMPAT && SYSVIPC
1777 1890
1778endmenu 1891endmenu
1779 1892
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index 2c518fbc52ec..0b7c4a3f0651 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -38,8 +38,7 @@ config M386
38 - "Crusoe" for the Transmeta Crusoe series. 38 - "Crusoe" for the Transmeta Crusoe series.
39 - "Efficeon" for the Transmeta Efficeon series. 39 - "Efficeon" for the Transmeta Efficeon series.
40 - "Winchip-C6" for original IDT Winchip. 40 - "Winchip-C6" for original IDT Winchip.
41 - "Winchip-2" for IDT Winchip 2. 41 - "Winchip-2" for IDT Winchips with 3dNow! capabilities.
42 - "Winchip-2A" for IDT Winchips with 3dNow! capabilities.
43 - "GeodeGX1" for Geode GX1 (Cyrix MediaGX). 42 - "GeodeGX1" for Geode GX1 (Cyrix MediaGX).
44 - "Geode GX/LX" For AMD Geode GX and LX processors. 43 - "Geode GX/LX" For AMD Geode GX and LX processors.
45 - "CyrixIII/VIA C3" for VIA Cyrix III or VIA C3. 44 - "CyrixIII/VIA C3" for VIA Cyrix III or VIA C3.
@@ -194,19 +193,11 @@ config MWINCHIPC6
194 treat this chip as a 586TSC with some extended instructions 193 treat this chip as a 586TSC with some extended instructions
195 and alignment requirements. 194 and alignment requirements.
196 195
197config MWINCHIP2
198 bool "Winchip-2"
199 depends on X86_32
200 help
201 Select this for an IDT Winchip-2. Linux and GCC
202 treat this chip as a 586TSC with some extended instructions
203 and alignment requirements.
204
205config MWINCHIP3D 196config MWINCHIP3D
206 bool "Winchip-2A/Winchip-3" 197 bool "Winchip-2/Winchip-2A/Winchip-3"
207 depends on X86_32 198 depends on X86_32
208 help 199 help
209 Select this for an IDT Winchip-2A or 3. Linux and GCC 200 Select this for an IDT Winchip-2, 2A or 3. Linux and GCC
210 treat this chip as a 586TSC with some extended instructions 201 treat this chip as a 586TSC with some extended instructions
211 and alignment requirements. Also enable out of order memory 202 and alignment requirements. Also enable out of order memory
212 stores for this CPU, which can increase performance of some 203 stores for this CPU, which can increase performance of some
@@ -318,7 +309,7 @@ config X86_L1_CACHE_SHIFT
318 int 309 int
319 default "7" if MPENTIUM4 || X86_GENERIC || GENERIC_CPU || MPSC 310 default "7" if MPENTIUM4 || X86_GENERIC || GENERIC_CPU || MPSC
320 default "4" if X86_ELAN || M486 || M386 || MGEODEGX1 311 default "4" if X86_ELAN || M486 || M386 || MGEODEGX1
321 default "5" if MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCRUSOE || MEFFICEON || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2 || MGEODE_LX 312 default "5" if MWINCHIP3D || MWINCHIPC6 || MCRUSOE || MEFFICEON || MCYRIXIII || MK6 || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || M586 || MVIAC3_2 || MGEODE_LX
322 default "6" if MK7 || MK8 || MPENTIUMM || MCORE2 || MVIAC7 313 default "6" if MK7 || MK8 || MPENTIUMM || MCORE2 || MVIAC7
323 314
324config X86_XADD 315config X86_XADD
@@ -360,7 +351,7 @@ config X86_POPAD_OK
360 351
361config X86_ALIGNMENT_16 352config X86_ALIGNMENT_16
362 def_bool y 353 def_bool y
363 depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || X86_ELAN || MK6 || M586MMX || M586TSC || M586 || M486 || MVIAC3_2 || MGEODEGX1 354 depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || X86_ELAN || MK6 || M586MMX || M586TSC || M586 || M486 || MVIAC3_2 || MGEODEGX1
364 355
365config X86_INTEL_USERCOPY 356config X86_INTEL_USERCOPY
366 def_bool y 357 def_bool y
@@ -368,7 +359,7 @@ config X86_INTEL_USERCOPY
368 359
369config X86_USE_PPRO_CHECKSUM 360config X86_USE_PPRO_CHECKSUM
370 def_bool y 361 def_bool y
371 depends on MWINCHIP3D || MWINCHIP2 || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MEFFICEON || MGEODE_LX || MCORE2 362 depends on MWINCHIP3D || MWINCHIPC6 || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MK8 || MVIAC3_2 || MEFFICEON || MGEODE_LX || MCORE2
372 363
373config X86_USE_3DNOW 364config X86_USE_3DNOW
374 def_bool y 365 def_bool y
@@ -376,24 +367,27 @@ config X86_USE_3DNOW
376 367
377config X86_OOSTORE 368config X86_OOSTORE
378 def_bool y 369 def_bool y
379 depends on (MWINCHIP3D || MWINCHIP2 || MWINCHIPC6) && MTRR 370 depends on (MWINCHIP3D || MWINCHIPC6) && MTRR
380 371
381# 372#
382# P6_NOPs are a relatively minor optimization that require a family >= 373# P6_NOPs are a relatively minor optimization that require a family >=
383# 6 processor, except that it is broken on certain VIA chips. 374# 6 processor, except that it is broken on certain VIA chips.
384# Furthermore, AMD chips prefer a totally different sequence of NOPs 375# Furthermore, AMD chips prefer a totally different sequence of NOPs
385# (which work on all CPUs). As a result, disallow these if we're 376# (which work on all CPUs). In addition, it looks like Virtual PC
386# compiling X86_GENERIC but not X86_64 (these NOPs do work on all 377# does not understand them.
387# x86-64 capable chips); the list of processors in the right-hand clause 378#
388# are the cores that benefit from this optimization. 379# As a result, disallow these if we're not compiling for X86_64 (these
380# NOPs do work on all x86-64 capable chips); the list of processors in
381# the right-hand clause are the cores that benefit from this optimization.
389# 382#
390config X86_P6_NOP 383config X86_P6_NOP
391 def_bool y 384 def_bool y
392 depends on (X86_64 || !X86_GENERIC) && (M686 || MPENTIUMII || MPENTIUMIII || MPENTIUMM || MCORE2 || MPENTIUM4 || MPSC) 385 depends on X86_64
386 depends on (MCORE2 || MPENTIUM4 || MPSC)
393 387
394config X86_TSC 388config X86_TSC
395 def_bool y 389 def_bool y
396 depends on ((MWINCHIP3D || MWINCHIP2 || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2 || MVIAC7 || MGEODEGX1 || MGEODE_LX || MCORE2) && !X86_NUMAQ) || X86_64 390 depends on ((MWINCHIP3D || MCRUSOE || MEFFICEON || MCYRIXIII || MK7 || MK6 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || M586MMX || M586TSC || MK8 || MVIAC3_2 || MVIAC7 || MGEODEGX1 || MGEODE_LX || MCORE2) && !X86_NUMAQ) || X86_64
397 391
398config X86_CMPXCHG64 392config X86_CMPXCHG64
399 def_bool y 393 def_bool y
@@ -403,7 +397,7 @@ config X86_CMPXCHG64
403# generates cmov. 397# generates cmov.
404config X86_CMOV 398config X86_CMOV
405 def_bool y 399 def_bool y
406 depends on (MK7 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || X86_64) 400 depends on (MK8 || MK7 || MCORE2 || MPENTIUM4 || MPENTIUMM || MPENTIUMIII || MPENTIUMII || M686 || MVIAC3_2 || MVIAC7 || MCRUSOE || MEFFICEON || X86_64)
407 401
408config X86_MINIMUM_CPU_FAMILY 402config X86_MINIMUM_CPU_FAMILY
409 int 403 int
@@ -414,4 +408,124 @@ config X86_MINIMUM_CPU_FAMILY
414 408
415config X86_DEBUGCTLMSR 409config X86_DEBUGCTLMSR
416 def_bool y 410 def_bool y
417 depends on !(MK6 || MWINCHIPC6 || MWINCHIP2 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 || M486 || M386) 411 depends on !(MK6 || MWINCHIPC6 || MWINCHIP3D || MCYRIXIII || M586MMX || M586TSC || M586 || M486 || M386)
412
413menuconfig PROCESSOR_SELECT
414 bool "Supported processor vendors" if EMBEDDED
415 help
416 This lets you choose what x86 vendor support code your kernel
417 will include.
418
419config CPU_SUP_INTEL
420 default y
421 bool "Support Intel processors" if PROCESSOR_SELECT
422 help
423 This enables detection, tunings and quirks for Intel processors
424
425 You need this enabled if you want your kernel to run on an
426 Intel CPU. Disabling this option on other types of CPUs
427 makes the kernel a tiny bit smaller. Disabling it on an Intel
428 CPU might render the kernel unbootable.
429
430 If unsure, say N.
431
432config CPU_SUP_CYRIX_32
433 default y
434 bool "Support Cyrix processors" if PROCESSOR_SELECT
435 depends on !64BIT
436 help
437 This enables detection, tunings and quirks for Cyrix processors
438
439 You need this enabled if you want your kernel to run on a
440 Cyrix CPU. Disabling this option on other types of CPUs
441 makes the kernel a tiny bit smaller. Disabling it on a Cyrix
442 CPU might render the kernel unbootable.
443
444 If unsure, say N.
445
446config CPU_SUP_AMD
447 default y
448 bool "Support AMD processors" if PROCESSOR_SELECT
449 help
450 This enables detection, tunings and quirks for AMD processors
451
452 You need this enabled if you want your kernel to run on an
453 AMD CPU. Disabling this option on other types of CPUs
454 makes the kernel a tiny bit smaller. Disabling it on an AMD
455 CPU might render the kernel unbootable.
456
457 If unsure, say N.
458
459config CPU_SUP_CENTAUR_32
460 default y
461 bool "Support Centaur processors" if PROCESSOR_SELECT
462 depends on !64BIT
463 help
464 This enables detection, tunings and quirks for Centaur processors
465
466 You need this enabled if you want your kernel to run on a
467 Centaur CPU. Disabling this option on other types of CPUs
468 makes the kernel a tiny bit smaller. Disabling it on a Centaur
469 CPU might render the kernel unbootable.
470
471 If unsure, say N.
472
473config CPU_SUP_CENTAUR_64
474 default y
475 bool "Support Centaur processors" if PROCESSOR_SELECT
476 depends on 64BIT
477 help
478 This enables detection, tunings and quirks for Centaur processors
479
480 You need this enabled if you want your kernel to run on a
481 Centaur CPU. Disabling this option on other types of CPUs
482 makes the kernel a tiny bit smaller. Disabling it on a Centaur
483 CPU might render the kernel unbootable.
484
485 If unsure, say N.
486
487config CPU_SUP_TRANSMETA_32
488 default y
489 bool "Support Transmeta processors" if PROCESSOR_SELECT
490 depends on !64BIT
491 help
492 This enables detection, tunings and quirks for Transmeta processors
493
494 You need this enabled if you want your kernel to run on a
495 Transmeta CPU. Disabling this option on other types of CPUs
496 makes the kernel a tiny bit smaller. Disabling it on a Transmeta
497 CPU might render the kernel unbootable.
498
499 If unsure, say N.
500
501config CPU_SUP_UMC_32
502 default y
503 bool "Support UMC processors" if PROCESSOR_SELECT
504 depends on !64BIT
505 help
506 This enables detection, tunings and quirks for UMC processors
507
508 You need this enabled if you want your kernel to run on a
509 UMC CPU. Disabling this option on other types of CPUs
510 makes the kernel a tiny bit smaller. Disabling it on a UMC
511 CPU might render the kernel unbootable.
512
513 If unsure, say N.
514
515config X86_DS
516 bool "Debug Store support"
517 default y
518 help
519 Add support for Debug Store.
520 This allows the kernel to provide a memory buffer to the hardware
521 to store various profiling and tracing events.
522
523config X86_PTRACE_BTS
524 bool "ptrace interface to Branch Trace Store"
525 default y
526 depends on (X86_DS && X86_DEBUGCTLMSR)
527 help
528 Add a ptrace interface to allow collecting an execution trace
529 of the traced task.
530 This collects control flow changes in a (cyclic) buffer and allows
531 debuggers to fill in the gaps and show an execution trace of the debuggee.
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 092f019e033a..2a3dfbd5e677 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -43,6 +43,19 @@ config EARLY_PRINTK
43 with klogd/syslogd or the X server. You should normally N here, 43 with klogd/syslogd or the X server. You should normally N here,
44 unless you want to debug such a crash. 44 unless you want to debug such a crash.
45 45
46config EARLY_PRINTK_DBGP
47 bool "Early printk via EHCI debug port"
48 default n
49 depends on EARLY_PRINTK && PCI
50 help
51 Write kernel log output directly into the EHCI debug port.
52
53 This is useful for kernel debugging when your machine crashes very
54 early before the console code is initialized. For normal operation
55 it is not recommended because it looks ugly and doesn't cooperate
56 with klogd/syslogd or the X server. You should normally N here,
57 unless you want to debug such a crash. You need usb debug device.
58
46config DEBUG_STACKOVERFLOW 59config DEBUG_STACKOVERFLOW
47 bool "Check for stack overflows" 60 bool "Check for stack overflows"
48 depends on DEBUG_KERNEL 61 depends on DEBUG_KERNEL
diff --git a/arch/x86/Makefile_32.cpu b/arch/x86/Makefile_32.cpu
index e372b584e919..80177ec052f0 100644
--- a/arch/x86/Makefile_32.cpu
+++ b/arch/x86/Makefile_32.cpu
@@ -28,7 +28,6 @@ cflags-$(CONFIG_MK8) += $(call cc-option,-march=k8,-march=athlon)
28cflags-$(CONFIG_MCRUSOE) += -march=i686 $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0 28cflags-$(CONFIG_MCRUSOE) += -march=i686 $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
29cflags-$(CONFIG_MEFFICEON) += -march=i686 $(call tune,pentium3) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0 29cflags-$(CONFIG_MEFFICEON) += -march=i686 $(call tune,pentium3) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
30cflags-$(CONFIG_MWINCHIPC6) += $(call cc-option,-march=winchip-c6,-march=i586) 30cflags-$(CONFIG_MWINCHIPC6) += $(call cc-option,-march=winchip-c6,-march=i586)
31cflags-$(CONFIG_MWINCHIP2) += $(call cc-option,-march=winchip2,-march=i586)
32cflags-$(CONFIG_MWINCHIP3D) += $(call cc-option,-march=winchip2,-march=i586) 31cflags-$(CONFIG_MWINCHIP3D) += $(call cc-option,-march=winchip2,-march=i586)
33cflags-$(CONFIG_MCYRIXIII) += $(call cc-option,-march=c3,-march=i486) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0 32cflags-$(CONFIG_MCYRIXIII) += $(call cc-option,-march=c3,-march=i486) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
34cflags-$(CONFIG_MVIAC3_2) += $(call cc-option,-march=c3-2,-march=i686) 33cflags-$(CONFIG_MVIAC3_2) += $(call cc-option,-march=c3-2,-march=i686)
@@ -45,3 +44,8 @@ cflags-$(CONFIG_MGEODEGX1) += -march=pentium-mmx
45# cpu entries 44# cpu entries
46cflags-$(CONFIG_X86_GENERIC) += $(call tune,generic,$(call tune,i686)) 45cflags-$(CONFIG_X86_GENERIC) += $(call tune,generic,$(call tune,i686))
47 46
47# Bug fix for binutils: this option is required in order to keep
48# binutils from generating NOPL instructions against our will.
49ifneq ($(CONFIG_X86_P6_NOP),y)
50cflags-y += $(call cc-option,-Wa$(comma)-mtune=generic32,)
51endif
diff --git a/arch/x86/boot/Makefile b/arch/x86/boot/Makefile
index 7ee102f9c4f8..cd48c7210016 100644
--- a/arch/x86/boot/Makefile
+++ b/arch/x86/boot/Makefile
@@ -72,9 +72,7 @@ KBUILD_CFLAGS := $(LINUXINCLUDE) -g -Os -D_SETUP -D__KERNEL__ \
72KBUILD_CFLAGS += $(call cc-option,-m32) 72KBUILD_CFLAGS += $(call cc-option,-m32)
73KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ 73KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__
74 74
75$(obj)/zImage: IMAGE_OFFSET := 0x1000
76$(obj)/zImage: asflags-y := $(SVGA_MODE) $(RAMDISK) 75$(obj)/zImage: asflags-y := $(SVGA_MODE) $(RAMDISK)
77$(obj)/bzImage: IMAGE_OFFSET := 0x100000
78$(obj)/bzImage: ccflags-y := -D__BIG_KERNEL__ 76$(obj)/bzImage: ccflags-y := -D__BIG_KERNEL__
79$(obj)/bzImage: asflags-y := $(SVGA_MODE) $(RAMDISK) -D__BIG_KERNEL__ 77$(obj)/bzImage: asflags-y := $(SVGA_MODE) $(RAMDISK) -D__BIG_KERNEL__
80$(obj)/bzImage: BUILDFLAGS := -b 78$(obj)/bzImage: BUILDFLAGS := -b
@@ -117,7 +115,7 @@ $(obj)/setup.bin: $(obj)/setup.elf FORCE
117 $(call if_changed,objcopy) 115 $(call if_changed,objcopy)
118 116
119$(obj)/compressed/vmlinux: FORCE 117$(obj)/compressed/vmlinux: FORCE
120 $(Q)$(MAKE) $(build)=$(obj)/compressed IMAGE_OFFSET=$(IMAGE_OFFSET) $@ 118 $(Q)$(MAKE) $(build)=$(obj)/compressed $@
121 119
122# Set this if you want to pass append arguments to the zdisk/fdimage/isoimage kernel 120# Set this if you want to pass append arguments to the zdisk/fdimage/isoimage kernel
123FDARGS = 121FDARGS =
@@ -181,6 +179,7 @@ isoimage: $(BOOTIMAGE)
181 mkisofs -J -r -o $(obj)/image.iso -b isolinux.bin -c boot.cat \ 179 mkisofs -J -r -o $(obj)/image.iso -b isolinux.bin -c boot.cat \
182 -no-emul-boot -boot-load-size 4 -boot-info-table \ 180 -no-emul-boot -boot-load-size 4 -boot-info-table \
183 $(obj)/isoimage 181 $(obj)/isoimage
182 isohybrid $(obj)/image.iso 2>/dev/null || true
184 rm -rf $(obj)/isoimage 183 rm -rf $(obj)/isoimage
185 184
186zlilo: $(BOOTIMAGE) 185zlilo: $(BOOTIMAGE)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
index 92fdd35bd93e..1771c804e02f 100644
--- a/arch/x86/boot/compressed/Makefile
+++ b/arch/x86/boot/compressed/Makefile
@@ -27,9 +27,8 @@ $(obj)/vmlinux.bin: vmlinux FORCE
27 $(call if_changed,objcopy) 27 $(call if_changed,objcopy)
28 28
29 29
30ifeq ($(CONFIG_X86_32),y) 30targets += vmlinux.bin.all vmlinux.relocs relocs
31targets += vmlinux.bin.all vmlinux.relocs 31hostprogs-$(CONFIG_X86_32) += relocs
32hostprogs-y := relocs
33 32
34quiet_cmd_relocs = RELOCS $@ 33quiet_cmd_relocs = RELOCS $@
35 cmd_relocs = $(obj)/relocs $< > $@;$(obj)/relocs --abs-relocs $< 34 cmd_relocs = $(obj)/relocs $< > $@;$(obj)/relocs --abs-relocs $<
@@ -43,6 +42,8 @@ quiet_cmd_relocbin = BUILD $@
43$(obj)/vmlinux.bin.all: $(vmlinux.bin.all-y) FORCE 42$(obj)/vmlinux.bin.all: $(vmlinux.bin.all-y) FORCE
44 $(call if_changed,relocbin) 43 $(call if_changed,relocbin)
45 44
45ifeq ($(CONFIG_X86_32),y)
46
46ifdef CONFIG_RELOCATABLE 47ifdef CONFIG_RELOCATABLE
47$(obj)/vmlinux.bin.gz: $(obj)/vmlinux.bin.all FORCE 48$(obj)/vmlinux.bin.gz: $(obj)/vmlinux.bin.all FORCE
48 $(call if_changed,gzip) 49 $(call if_changed,gzip)
@@ -59,6 +60,5 @@ $(obj)/vmlinux.bin.gz: $(obj)/vmlinux.bin FORCE
59LDFLAGS_piggy.o := -r --format binary --oformat elf64-x86-64 -T 60LDFLAGS_piggy.o := -r --format binary --oformat elf64-x86-64 -T
60endif 61endif
61 62
62
63$(obj)/piggy.o: $(obj)/vmlinux.scr $(obj)/vmlinux.bin.gz FORCE 63$(obj)/piggy.o: $(obj)/vmlinux.scr $(obj)/vmlinux.bin.gz FORCE
64 $(call if_changed,ld) 64 $(call if_changed,ld)
diff --git a/arch/x86/boot/compressed/head_32.S b/arch/x86/boot/compressed/head_32.S
index ba7736cf2ec7..29c5fbf08392 100644
--- a/arch/x86/boot/compressed/head_32.S
+++ b/arch/x86/boot/compressed/head_32.S
@@ -137,14 +137,15 @@ relocated:
137 */ 137 */
138 movl output_len(%ebx), %eax 138 movl output_len(%ebx), %eax
139 pushl %eax 139 pushl %eax
140 # push arguments for decompress_kernel:
140 pushl %ebp # output address 141 pushl %ebp # output address
141 movl input_len(%ebx), %eax 142 movl input_len(%ebx), %eax
142 pushl %eax # input_len 143 pushl %eax # input_len
143 leal input_data(%ebx), %eax 144 leal input_data(%ebx), %eax
144 pushl %eax # input_data 145 pushl %eax # input_data
145 leal boot_heap(%ebx), %eax 146 leal boot_heap(%ebx), %eax
146 pushl %eax # heap area as third argument 147 pushl %eax # heap area
147 pushl %esi # real mode pointer as second arg 148 pushl %esi # real mode pointer
148 call decompress_kernel 149 call decompress_kernel
149 addl $20, %esp 150 addl $20, %esp
150 popl %ecx 151 popl %ecx
diff --git a/arch/x86/boot/compressed/misc.c b/arch/x86/boot/compressed/misc.c
index 9fea73706479..5780d361105b 100644
--- a/arch/x86/boot/compressed/misc.c
+++ b/arch/x86/boot/compressed/misc.c
@@ -16,7 +16,7 @@
16 */ 16 */
17#undef CONFIG_PARAVIRT 17#undef CONFIG_PARAVIRT
18#ifdef CONFIG_X86_32 18#ifdef CONFIG_X86_32
19#define _ASM_DESC_H_ 1 19#define ASM_X86__DESC_H 1
20#endif 20#endif
21 21
22#ifdef CONFIG_X86_64 22#ifdef CONFIG_X86_64
@@ -27,7 +27,7 @@
27#include <linux/linkage.h> 27#include <linux/linkage.h>
28#include <linux/screen_info.h> 28#include <linux/screen_info.h>
29#include <linux/elf.h> 29#include <linux/elf.h>
30#include <asm/io.h> 30#include <linux/io.h>
31#include <asm/page.h> 31#include <asm/page.h>
32#include <asm/boot.h> 32#include <asm/boot.h>
33#include <asm/bootparam.h> 33#include <asm/bootparam.h>
@@ -251,7 +251,7 @@ static void __putstr(int error, const char *s)
251 y--; 251 y--;
252 } 252 }
253 } else { 253 } else {
254 vidmem [(x + cols * y) * 2] = c; 254 vidmem[(x + cols * y) * 2] = c;
255 if (++x >= cols) { 255 if (++x >= cols) {
256 x = 0; 256 x = 0;
257 if (++y >= lines) { 257 if (++y >= lines) {
@@ -277,7 +277,8 @@ static void *memset(void *s, int c, unsigned n)
277 int i; 277 int i;
278 char *ss = s; 278 char *ss = s;
279 279
280 for (i = 0; i < n; i++) ss[i] = c; 280 for (i = 0; i < n; i++)
281 ss[i] = c;
281 return s; 282 return s;
282} 283}
283 284
@@ -287,7 +288,8 @@ static void *memcpy(void *dest, const void *src, unsigned n)
287 const char *s = src; 288 const char *s = src;
288 char *d = dest; 289 char *d = dest;
289 290
290 for (i = 0; i < n; i++) d[i] = s[i]; 291 for (i = 0; i < n; i++)
292 d[i] = s[i];
291 return dest; 293 return dest;
292} 294}
293 295
diff --git a/arch/x86/boot/compressed/relocs.c b/arch/x86/boot/compressed/relocs.c
index a1310c52fc0c..857e492c571e 100644
--- a/arch/x86/boot/compressed/relocs.c
+++ b/arch/x86/boot/compressed/relocs.c
@@ -492,7 +492,7 @@ static void walk_relocs(void (*visit)(Elf32_Rel *rel, Elf32_Sym *sym))
492 continue; 492 continue;
493 } 493 }
494 sh_symtab = sec_symtab->symtab; 494 sh_symtab = sec_symtab->symtab;
495 sym_strtab = sec->link->strtab; 495 sym_strtab = sec_symtab->link->strtab;
496 for (j = 0; j < sec->shdr.sh_size/sizeof(Elf32_Rel); j++) { 496 for (j = 0; j < sec->shdr.sh_size/sizeof(Elf32_Rel); j++) {
497 Elf32_Rel *rel; 497 Elf32_Rel *rel;
498 Elf32_Sym *sym; 498 Elf32_Sym *sym;
diff --git a/arch/x86/boot/cpu.c b/arch/x86/boot/cpu.c
index 75298fe2edca..6ec6bb6e9957 100644
--- a/arch/x86/boot/cpu.c
+++ b/arch/x86/boot/cpu.c
@@ -59,17 +59,18 @@ int validate_cpu(void)
59 u32 e = err_flags[i]; 59 u32 e = err_flags[i];
60 60
61 for (j = 0; j < 32; j++) { 61 for (j = 0; j < 32; j++) {
62 int n = (i << 5)+j; 62 if (msg_strs[0] < i ||
63 if (*msg_strs < n) { 63 (msg_strs[0] == i && msg_strs[1] < j)) {
64 /* Skip to the next string */ 64 /* Skip to the next string */
65 do { 65 msg_strs += 2;
66 msg_strs++; 66 while (*msg_strs++)
67 } while (*msg_strs); 67 ;
68 msg_strs++;
69 } 68 }
70 if (e & 1) { 69 if (e & 1) {
71 if (*msg_strs == n && msg_strs[1]) 70 if (msg_strs[0] == i &&
72 printf("%s ", msg_strs+1); 71 msg_strs[1] == j &&
72 msg_strs[2])
73 printf("%s ", msg_strs+2);
73 else 74 else
74 printf("%d:%d ", i, j); 75 printf("%d:%d ", i, j);
75 } 76 }
diff --git a/arch/x86/boot/cpucheck.c b/arch/x86/boot/cpucheck.c
index 4b9ae7c56748..4d3ff037201f 100644
--- a/arch/x86/boot/cpucheck.c
+++ b/arch/x86/boot/cpucheck.c
@@ -38,12 +38,12 @@ static const u32 req_flags[NCAPINTS] =
38{ 38{
39 REQUIRED_MASK0, 39 REQUIRED_MASK0,
40 REQUIRED_MASK1, 40 REQUIRED_MASK1,
41 REQUIRED_MASK2, 41 0, /* REQUIRED_MASK2 not implemented in this file */
42 REQUIRED_MASK3, 42 0, /* REQUIRED_MASK3 not implemented in this file */
43 REQUIRED_MASK4, 43 REQUIRED_MASK4,
44 REQUIRED_MASK5, 44 0, /* REQUIRED_MASK5 not implemented in this file */
45 REQUIRED_MASK6, 45 REQUIRED_MASK6,
46 REQUIRED_MASK7, 46 0, /* REQUIRED_MASK7 not implemented in this file */
47}; 47};
48 48
49#define A32(a, b, c, d) (((d) << 24)+((c) << 16)+((b) << 8)+(a)) 49#define A32(a, b, c, d) (((d) << 24)+((c) << 16)+((b) << 8)+(a))
diff --git a/arch/x86/boot/edd.c b/arch/x86/boot/edd.c
index d93cbc6464d0..1aae8f3e5ca1 100644
--- a/arch/x86/boot/edd.c
+++ b/arch/x86/boot/edd.c
@@ -41,6 +41,7 @@ static u32 read_mbr_sig(u8 devno, struct edd_info *ei, u32 *mbrsig)
41 char *mbrbuf_ptr, *mbrbuf_end; 41 char *mbrbuf_ptr, *mbrbuf_end;
42 u32 buf_base, mbr_base; 42 u32 buf_base, mbr_base;
43 extern char _end[]; 43 extern char _end[];
44 u16 mbr_magic;
44 45
45 sector_size = ei->params.bytes_per_sector; 46 sector_size = ei->params.bytes_per_sector;
46 if (!sector_size) 47 if (!sector_size)
@@ -58,11 +59,15 @@ static u32 read_mbr_sig(u8 devno, struct edd_info *ei, u32 *mbrsig)
58 if (mbrbuf_end > (char *)(size_t)boot_params.hdr.heap_end_ptr) 59 if (mbrbuf_end > (char *)(size_t)boot_params.hdr.heap_end_ptr)
59 return -1; 60 return -1;
60 61
62 memset(mbrbuf_ptr, 0, sector_size);
61 if (read_mbr(devno, mbrbuf_ptr)) 63 if (read_mbr(devno, mbrbuf_ptr))
62 return -1; 64 return -1;
63 65
64 *mbrsig = *(u32 *)&mbrbuf_ptr[EDD_MBR_SIG_OFFSET]; 66 *mbrsig = *(u32 *)&mbrbuf_ptr[EDD_MBR_SIG_OFFSET];
65 return 0; 67 mbr_magic = *(u16 *)&mbrbuf_ptr[510];
68
69 /* check for valid MBR magic */
70 return mbr_magic == 0xAA55 ? 0 : -1;
66} 71}
67 72
68static int get_edd_info(u8 devno, struct edd_info *ei) 73static int get_edd_info(u8 devno, struct edd_info *ei)
diff --git a/arch/x86/boot/header.S b/arch/x86/boot/header.S
index af86e431acfa..b993062e9a5f 100644
--- a/arch/x86/boot/header.S
+++ b/arch/x86/boot/header.S
@@ -30,7 +30,6 @@ SYSSEG = DEF_SYSSEG /* system loaded at 0x10000 (65536) */
30SYSSIZE = DEF_SYSSIZE /* system size: # of 16-byte clicks */ 30SYSSIZE = DEF_SYSSIZE /* system size: # of 16-byte clicks */
31 /* to be loaded */ 31 /* to be loaded */
32ROOT_DEV = 0 /* ROOT_DEV is now written by "build" */ 32ROOT_DEV = 0 /* ROOT_DEV is now written by "build" */
33SWAP_DEV = 0 /* SWAP_DEV is now written by "build" */
34 33
35#ifndef SVGA_MODE 34#ifndef SVGA_MODE
36#define SVGA_MODE ASK_VGA 35#define SVGA_MODE ASK_VGA
diff --git a/arch/x86/boot/mkcpustr.c b/arch/x86/boot/mkcpustr.c
index bbe76953bae9..8ef60f20b371 100644
--- a/arch/x86/boot/mkcpustr.c
+++ b/arch/x86/boot/mkcpustr.c
@@ -15,33 +15,33 @@
15 15
16#include <stdio.h> 16#include <stdio.h>
17 17
18#include "../kernel/cpu/feature_names.c" 18#include "../kernel/cpu/capflags.c"
19
20#if NCAPFLAGS > 8
21# error "Need to adjust the boot code handling of CPUID strings"
22#endif
23 19
24int main(void) 20int main(void)
25{ 21{
26 int i; 22 int i, j;
27 const char *str; 23 const char *str;
28 24
29 printf("static const char x86_cap_strs[] = \n"); 25 printf("static const char x86_cap_strs[] = \n");
30 26
31 for (i = 0; i < NCAPINTS*32; i++) { 27 for (i = 0; i < NCAPINTS; i++) {
32 str = x86_cap_flags[i]; 28 for (j = 0; j < 32; j++) {
33 29 str = x86_cap_flags[i*32+j];
34 if (i == NCAPINTS*32-1) { 30
35 /* The last entry must be unconditional; this 31 if (i == NCAPINTS-1 && j == 31) {
36 also consumes the compiler-added null character */ 32 /* The last entry must be unconditional; this
37 if (!str) 33 also consumes the compiler-added null
38 str = ""; 34 character */
39 printf("\t\"\\x%02x\"\"%s\"\n", i, str); 35 if (!str)
40 } else if (str) { 36 str = "";
41 printf("#if REQUIRED_MASK%d & (1 << %d)\n" 37 printf("\t\"\\x%02x\\x%02x\"\"%s\"\n",
42 "\t\"\\x%02x\"\"%s\\0\"\n" 38 i, j, str);
43 "#endif\n", 39 } else if (str) {
44 i >> 5, i & 31, i, str); 40 printf("#if REQUIRED_MASK%d & (1 << %d)\n"
41 "\t\"\\x%02x\\x%02x\"\"%s\\0\"\n"
42 "#endif\n",
43 i, j, i, j, str);
44 }
45 } 45 }
46 } 46 }
47 printf("\t;\n"); 47 printf("\t;\n");
diff --git a/arch/x86/boot/video-vesa.c b/arch/x86/boot/video-vesa.c
index 401ad998ad08..99b3079dc6ab 100644
--- a/arch/x86/boot/video-vesa.c
+++ b/arch/x86/boot/video-vesa.c
@@ -88,14 +88,11 @@ static int vesa_probe(void)
88 (vminfo.memory_layout == 4 || 88 (vminfo.memory_layout == 4 ||
89 vminfo.memory_layout == 6) && 89 vminfo.memory_layout == 6) &&
90 vminfo.memory_planes == 1) { 90 vminfo.memory_planes == 1) {
91#ifdef CONFIG_FB 91#ifdef CONFIG_FB_BOOT_VESA_SUPPORT
92 /* Graphics mode, color, linear frame buffer 92 /* Graphics mode, color, linear frame buffer
93 supported. Only register the mode if 93 supported. Only register the mode if
94 if framebuffer is configured, however, 94 if framebuffer is configured, however,
95 otherwise the user will be left without a screen. 95 otherwise the user will be left without a screen. */
96 We don't require CONFIG_FB_VESA, however, since
97 some of the other framebuffer drivers can use
98 this mode-setting, too. */
99 mi = GET_HEAP(struct mode_info, 1); 96 mi = GET_HEAP(struct mode_info, 1);
100 mi->mode = mode + VIDEO_FIRST_VESA; 97 mi->mode = mode + VIDEO_FIRST_VESA;
101 mi->depth = vminfo.bpp; 98 mi->depth = vminfo.bpp;
@@ -133,10 +130,12 @@ static int vesa_set_mode(struct mode_info *mode)
133 if ((vminfo.mode_attr & 0x15) == 0x05) { 130 if ((vminfo.mode_attr & 0x15) == 0x05) {
134 /* It's a supported text mode */ 131 /* It's a supported text mode */
135 is_graphic = 0; 132 is_graphic = 0;
133#ifdef CONFIG_FB_BOOT_VESA_SUPPORT
136 } else if ((vminfo.mode_attr & 0x99) == 0x99) { 134 } else if ((vminfo.mode_attr & 0x99) == 0x99) {
137 /* It's a graphics mode with linear frame buffer */ 135 /* It's a graphics mode with linear frame buffer */
138 is_graphic = 1; 136 is_graphic = 1;
139 vesa_mode |= 0x4000; /* Request linear frame buffer */ 137 vesa_mode |= 0x4000; /* Request linear frame buffer */
138#endif
140 } else { 139 } else {
141 return -1; /* Invalid mode */ 140 return -1; /* Invalid mode */
142 } 141 }
@@ -224,7 +223,7 @@ static void vesa_store_pm_info(void)
224static void vesa_store_mode_params_graphics(void) 223static void vesa_store_mode_params_graphics(void)
225{ 224{
226 /* Tell the kernel we're in VESA graphics mode */ 225 /* Tell the kernel we're in VESA graphics mode */
227 boot_params.screen_info.orig_video_isVGA = 0x23; 226 boot_params.screen_info.orig_video_isVGA = VIDEO_TYPE_VLFB;
228 227
229 /* Mode parameters */ 228 /* Mode parameters */
230 boot_params.screen_info.vesa_attributes = vminfo.mode_attr; 229 boot_params.screen_info.vesa_attributes = vminfo.mode_attr;
diff --git a/arch/x86/configs/i386_defconfig b/arch/x86/configs/i386_defconfig
index 104275e191a8..52d0359719d7 100644
--- a/arch/x86/configs/i386_defconfig
+++ b/arch/x86/configs/i386_defconfig
@@ -1,7 +1,7 @@
1# 1#
2# Automatically generated make config: don't edit 2# Automatically generated make config: don't edit
3# Linux kernel version: 2.6.27-rc4 3# Linux kernel version: 2.6.27-rc5
4# Mon Aug 25 15:04:00 2008 4# Wed Sep 3 17:23:09 2008
5# 5#
6# CONFIG_64BIT is not set 6# CONFIG_64BIT is not set
7CONFIG_X86_32=y 7CONFIG_X86_32=y
@@ -202,7 +202,7 @@ CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER=y
202# CONFIG_M586 is not set 202# CONFIG_M586 is not set
203# CONFIG_M586TSC is not set 203# CONFIG_M586TSC is not set
204# CONFIG_M586MMX is not set 204# CONFIG_M586MMX is not set
205# CONFIG_M686 is not set 205CONFIG_M686=y
206# CONFIG_MPENTIUMII is not set 206# CONFIG_MPENTIUMII is not set
207# CONFIG_MPENTIUMIII is not set 207# CONFIG_MPENTIUMIII is not set
208# CONFIG_MPENTIUMM is not set 208# CONFIG_MPENTIUMM is not set
@@ -213,7 +213,6 @@ CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER=y
213# CONFIG_MCRUSOE is not set 213# CONFIG_MCRUSOE is not set
214# CONFIG_MEFFICEON is not set 214# CONFIG_MEFFICEON is not set
215# CONFIG_MWINCHIPC6 is not set 215# CONFIG_MWINCHIPC6 is not set
216# CONFIG_MWINCHIP2 is not set
217# CONFIG_MWINCHIP3D is not set 216# CONFIG_MWINCHIP3D is not set
218# CONFIG_MGEODEGX1 is not set 217# CONFIG_MGEODEGX1 is not set
219# CONFIG_MGEODE_LX is not set 218# CONFIG_MGEODE_LX is not set
@@ -221,13 +220,14 @@ CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER=y
221# CONFIG_MVIAC3_2 is not set 220# CONFIG_MVIAC3_2 is not set
222# CONFIG_MVIAC7 is not set 221# CONFIG_MVIAC7 is not set
223# CONFIG_MPSC is not set 222# CONFIG_MPSC is not set
224CONFIG_MCORE2=y 223# CONFIG_MCORE2 is not set
225# CONFIG_GENERIC_CPU is not set 224# CONFIG_GENERIC_CPU is not set
226CONFIG_X86_GENERIC=y 225CONFIG_X86_GENERIC=y
227CONFIG_X86_CPU=y 226CONFIG_X86_CPU=y
228CONFIG_X86_CMPXCHG=y 227CONFIG_X86_CMPXCHG=y
229CONFIG_X86_L1_CACHE_SHIFT=7 228CONFIG_X86_L1_CACHE_SHIFT=7
230CONFIG_X86_XADD=y 229CONFIG_X86_XADD=y
230# CONFIG_X86_PPRO_FENCE is not set
231CONFIG_X86_WP_WORKS_OK=y 231CONFIG_X86_WP_WORKS_OK=y
232CONFIG_X86_INVLPG=y 232CONFIG_X86_INVLPG=y
233CONFIG_X86_BSWAP=y 233CONFIG_X86_BSWAP=y
@@ -235,14 +235,15 @@ CONFIG_X86_POPAD_OK=y
235CONFIG_X86_INTEL_USERCOPY=y 235CONFIG_X86_INTEL_USERCOPY=y
236CONFIG_X86_USE_PPRO_CHECKSUM=y 236CONFIG_X86_USE_PPRO_CHECKSUM=y
237CONFIG_X86_TSC=y 237CONFIG_X86_TSC=y
238CONFIG_X86_CMOV=y
238CONFIG_X86_MINIMUM_CPU_FAMILY=4 239CONFIG_X86_MINIMUM_CPU_FAMILY=4
239CONFIG_X86_DEBUGCTLMSR=y 240CONFIG_X86_DEBUGCTLMSR=y
240CONFIG_HPET_TIMER=y 241CONFIG_HPET_TIMER=y
241CONFIG_HPET_EMULATE_RTC=y 242CONFIG_HPET_EMULATE_RTC=y
242CONFIG_DMI=y 243CONFIG_DMI=y
243# CONFIG_IOMMU_HELPER is not set 244# CONFIG_IOMMU_HELPER is not set
244CONFIG_NR_CPUS=4 245CONFIG_NR_CPUS=64
245# CONFIG_SCHED_SMT is not set 246CONFIG_SCHED_SMT=y
246CONFIG_SCHED_MC=y 247CONFIG_SCHED_MC=y
247# CONFIG_PREEMPT_NONE is not set 248# CONFIG_PREEMPT_NONE is not set
248CONFIG_PREEMPT_VOLUNTARY=y 249CONFIG_PREEMPT_VOLUNTARY=y
@@ -254,7 +255,8 @@ CONFIG_VM86=y
254# CONFIG_TOSHIBA is not set 255# CONFIG_TOSHIBA is not set
255# CONFIG_I8K is not set 256# CONFIG_I8K is not set
256CONFIG_X86_REBOOTFIXUPS=y 257CONFIG_X86_REBOOTFIXUPS=y
257# CONFIG_MICROCODE is not set 258CONFIG_MICROCODE=y
259CONFIG_MICROCODE_OLD_INTERFACE=y
258CONFIG_X86_MSR=y 260CONFIG_X86_MSR=y
259CONFIG_X86_CPUID=y 261CONFIG_X86_CPUID=y
260# CONFIG_NOHIGHMEM is not set 262# CONFIG_NOHIGHMEM is not set
@@ -1532,7 +1534,6 @@ CONFIG_BACKLIGHT_CLASS_DEVICE=y
1532CONFIG_VGA_CONSOLE=y 1534CONFIG_VGA_CONSOLE=y
1533CONFIG_VGACON_SOFT_SCROLLBACK=y 1535CONFIG_VGACON_SOFT_SCROLLBACK=y
1534CONFIG_VGACON_SOFT_SCROLLBACK_SIZE=64 1536CONFIG_VGACON_SOFT_SCROLLBACK_SIZE=64
1535CONFIG_VIDEO_SELECT=y
1536CONFIG_DUMMY_CONSOLE=y 1537CONFIG_DUMMY_CONSOLE=y
1537# CONFIG_FRAMEBUFFER_CONSOLE is not set 1538# CONFIG_FRAMEBUFFER_CONSOLE is not set
1538CONFIG_LOGO=y 1539CONFIG_LOGO=y
@@ -2115,7 +2116,7 @@ CONFIG_IO_DELAY_0X80=y
2115CONFIG_DEFAULT_IO_DELAY_TYPE=0 2116CONFIG_DEFAULT_IO_DELAY_TYPE=0
2116CONFIG_DEBUG_BOOT_PARAMS=y 2117CONFIG_DEBUG_BOOT_PARAMS=y
2117# CONFIG_CPA_DEBUG is not set 2118# CONFIG_CPA_DEBUG is not set
2118# CONFIG_OPTIMIZE_INLINING is not set 2119CONFIG_OPTIMIZE_INLINING=y
2119 2120
2120# 2121#
2121# Security options 2122# Security options
diff --git a/arch/x86/configs/x86_64_defconfig b/arch/x86/configs/x86_64_defconfig
index 678c8acefe04..f0a03d7a7d63 100644
--- a/arch/x86/configs/x86_64_defconfig
+++ b/arch/x86/configs/x86_64_defconfig
@@ -1,7 +1,7 @@
1# 1#
2# Automatically generated make config: don't edit 2# Automatically generated make config: don't edit
3# Linux kernel version: 2.6.27-rc4 3# Linux kernel version: 2.6.27-rc5
4# Mon Aug 25 14:40:46 2008 4# Wed Sep 3 17:13:39 2008
5# 5#
6CONFIG_64BIT=y 6CONFIG_64BIT=y
7# CONFIG_X86_32 is not set 7# CONFIG_X86_32 is not set
@@ -210,7 +210,6 @@ CONFIG_X86_PC=y
210# CONFIG_MCRUSOE is not set 210# CONFIG_MCRUSOE is not set
211# CONFIG_MEFFICEON is not set 211# CONFIG_MEFFICEON is not set
212# CONFIG_MWINCHIPC6 is not set 212# CONFIG_MWINCHIPC6 is not set
213# CONFIG_MWINCHIP2 is not set
214# CONFIG_MWINCHIP3D is not set 213# CONFIG_MWINCHIP3D is not set
215# CONFIG_MGEODEGX1 is not set 214# CONFIG_MGEODEGX1 is not set
216# CONFIG_MGEODE_LX is not set 215# CONFIG_MGEODE_LX is not set
@@ -218,17 +217,14 @@ CONFIG_X86_PC=y
218# CONFIG_MVIAC3_2 is not set 217# CONFIG_MVIAC3_2 is not set
219# CONFIG_MVIAC7 is not set 218# CONFIG_MVIAC7 is not set
220# CONFIG_MPSC is not set 219# CONFIG_MPSC is not set
221CONFIG_MCORE2=y 220# CONFIG_MCORE2 is not set
222# CONFIG_GENERIC_CPU is not set 221CONFIG_GENERIC_CPU=y
223CONFIG_X86_CPU=y 222CONFIG_X86_CPU=y
224CONFIG_X86_L1_CACHE_BYTES=64 223CONFIG_X86_L1_CACHE_BYTES=128
225CONFIG_X86_INTERNODE_CACHE_BYTES=64 224CONFIG_X86_INTERNODE_CACHE_BYTES=128
226CONFIG_X86_CMPXCHG=y 225CONFIG_X86_CMPXCHG=y
227CONFIG_X86_L1_CACHE_SHIFT=6 226CONFIG_X86_L1_CACHE_SHIFT=7
228CONFIG_X86_WP_WORKS_OK=y 227CONFIG_X86_WP_WORKS_OK=y
229CONFIG_X86_INTEL_USERCOPY=y
230CONFIG_X86_USE_PPRO_CHECKSUM=y
231CONFIG_X86_P6_NOP=y
232CONFIG_X86_TSC=y 228CONFIG_X86_TSC=y
233CONFIG_X86_CMPXCHG64=y 229CONFIG_X86_CMPXCHG64=y
234CONFIG_X86_CMOV=y 230CONFIG_X86_CMOV=y
@@ -243,9 +239,8 @@ CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT=y
243CONFIG_AMD_IOMMU=y 239CONFIG_AMD_IOMMU=y
244CONFIG_SWIOTLB=y 240CONFIG_SWIOTLB=y
245CONFIG_IOMMU_HELPER=y 241CONFIG_IOMMU_HELPER=y
246# CONFIG_MAXSMP is not set 242CONFIG_NR_CPUS=64
247CONFIG_NR_CPUS=4 243CONFIG_SCHED_SMT=y
248# CONFIG_SCHED_SMT is not set
249CONFIG_SCHED_MC=y 244CONFIG_SCHED_MC=y
250# CONFIG_PREEMPT_NONE is not set 245# CONFIG_PREEMPT_NONE is not set
251CONFIG_PREEMPT_VOLUNTARY=y 246CONFIG_PREEMPT_VOLUNTARY=y
@@ -254,7 +249,8 @@ CONFIG_X86_LOCAL_APIC=y
254CONFIG_X86_IO_APIC=y 249CONFIG_X86_IO_APIC=y
255# CONFIG_X86_MCE is not set 250# CONFIG_X86_MCE is not set
256# CONFIG_I8K is not set 251# CONFIG_I8K is not set
257# CONFIG_MICROCODE is not set 252CONFIG_MICROCODE=y
253CONFIG_MICROCODE_OLD_INTERFACE=y
258CONFIG_X86_MSR=y 254CONFIG_X86_MSR=y
259CONFIG_X86_CPUID=y 255CONFIG_X86_CPUID=y
260CONFIG_NUMA=y 256CONFIG_NUMA=y
@@ -290,7 +286,7 @@ CONFIG_BOUNCE=y
290CONFIG_VIRT_TO_BUS=y 286CONFIG_VIRT_TO_BUS=y
291CONFIG_MTRR=y 287CONFIG_MTRR=y
292# CONFIG_MTRR_SANITIZER is not set 288# CONFIG_MTRR_SANITIZER is not set
293# CONFIG_X86_PAT is not set 289CONFIG_X86_PAT=y
294CONFIG_EFI=y 290CONFIG_EFI=y
295CONFIG_SECCOMP=y 291CONFIG_SECCOMP=y
296# CONFIG_HZ_100 is not set 292# CONFIG_HZ_100 is not set
@@ -1508,7 +1504,6 @@ CONFIG_BACKLIGHT_CLASS_DEVICE=y
1508CONFIG_VGA_CONSOLE=y 1504CONFIG_VGA_CONSOLE=y
1509CONFIG_VGACON_SOFT_SCROLLBACK=y 1505CONFIG_VGACON_SOFT_SCROLLBACK=y
1510CONFIG_VGACON_SOFT_SCROLLBACK_SIZE=64 1506CONFIG_VGACON_SOFT_SCROLLBACK_SIZE=64
1511CONFIG_VIDEO_SELECT=y
1512CONFIG_DUMMY_CONSOLE=y 1507CONFIG_DUMMY_CONSOLE=y
1513# CONFIG_FRAMEBUFFER_CONSOLE is not set 1508# CONFIG_FRAMEBUFFER_CONSOLE is not set
1514CONFIG_LOGO=y 1509CONFIG_LOGO=y
@@ -2089,7 +2084,7 @@ CONFIG_IO_DELAY_0X80=y
2089CONFIG_DEFAULT_IO_DELAY_TYPE=0 2084CONFIG_DEFAULT_IO_DELAY_TYPE=0
2090CONFIG_DEBUG_BOOT_PARAMS=y 2085CONFIG_DEBUG_BOOT_PARAMS=y
2091# CONFIG_CPA_DEBUG is not set 2086# CONFIG_CPA_DEBUG is not set
2092# CONFIG_OPTIMIZE_INLINING is not set 2087CONFIG_OPTIMIZE_INLINING=y
2093 2088
2094# 2089#
2095# Security options 2090# Security options
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index 3874c2de5403..903de4aa5094 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -10,6 +10,8 @@ obj-$(CONFIG_CRYPTO_AES_X86_64) += aes-x86_64.o
10obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o 10obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o
11obj-$(CONFIG_CRYPTO_SALSA20_X86_64) += salsa20-x86_64.o 11obj-$(CONFIG_CRYPTO_SALSA20_X86_64) += salsa20-x86_64.o
12 12
13obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
14
13aes-i586-y := aes-i586-asm_32.o aes_glue.o 15aes-i586-y := aes-i586-asm_32.o aes_glue.o
14twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o 16twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
15salsa20-i586-y := salsa20-i586-asm_32.o salsa20_glue.o 17salsa20-i586-y := salsa20-i586-asm_32.o salsa20_glue.o
diff --git a/arch/x86/crypto/crc32c-intel.c b/arch/x86/crypto/crc32c-intel.c
new file mode 100644
index 000000000000..070afc5b6c94
--- /dev/null
+++ b/arch/x86/crypto/crc32c-intel.c
@@ -0,0 +1,197 @@
1/*
2 * Using hardware provided CRC32 instruction to accelerate the CRC32 disposal.
3 * CRC32C polynomial:0x1EDC6F41(BE)/0x82F63B78(LE)
4 * CRC32 is a new instruction in Intel SSE4.2, the reference can be found at:
5 * http://www.intel.com/products/processor/manuals/
6 * Intel(R) 64 and IA-32 Architectures Software Developer's Manual
7 * Volume 2A: Instruction Set Reference, A-M
8 *
9 * Copyright (c) 2008 Austin Zhang <austin_zhang@linux.intel.com>
10 * Copyright (c) 2008 Kent Liu <kent.liu@intel.com>
11 *
12 * This program is free software; you can redistribute it and/or modify it
13 * under the terms of the GNU General Public License as published by the Free
14 * Software Foundation; either version 2 of the License, or (at your option)
15 * any later version.
16 *
17 */
18#include <linux/init.h>
19#include <linux/module.h>
20#include <linux/string.h>
21#include <linux/kernel.h>
22#include <crypto/internal/hash.h>
23
24#include <asm/cpufeature.h>
25
26#define CHKSUM_BLOCK_SIZE 1
27#define CHKSUM_DIGEST_SIZE 4
28
29#define SCALE_F sizeof(unsigned long)
30
31#ifdef CONFIG_X86_64
32#define REX_PRE "0x48, "
33#else
34#define REX_PRE
35#endif
36
37static u32 crc32c_intel_le_hw_byte(u32 crc, unsigned char const *data, size_t length)
38{
39 while (length--) {
40 __asm__ __volatile__(
41 ".byte 0xf2, 0xf, 0x38, 0xf0, 0xf1"
42 :"=S"(crc)
43 :"0"(crc), "c"(*data)
44 );
45 data++;
46 }
47
48 return crc;
49}
50
51static u32 __pure crc32c_intel_le_hw(u32 crc, unsigned char const *p, size_t len)
52{
53 unsigned int iquotient = len / SCALE_F;
54 unsigned int iremainder = len % SCALE_F;
55 unsigned long *ptmp = (unsigned long *)p;
56
57 while (iquotient--) {
58 __asm__ __volatile__(
59 ".byte 0xf2, " REX_PRE "0xf, 0x38, 0xf1, 0xf1;"
60 :"=S"(crc)
61 :"0"(crc), "c"(*ptmp)
62 );
63 ptmp++;
64 }
65
66 if (iremainder)
67 crc = crc32c_intel_le_hw_byte(crc, (unsigned char *)ptmp,
68 iremainder);
69
70 return crc;
71}
72
73/*
74 * Setting the seed allows arbitrary accumulators and flexible XOR policy
75 * If your algorithm starts with ~0, then XOR with ~0 before you set
76 * the seed.
77 */
78static int crc32c_intel_setkey(struct crypto_ahash *hash, const u8 *key,
79 unsigned int keylen)
80{
81 u32 *mctx = crypto_ahash_ctx(hash);
82
83 if (keylen != sizeof(u32)) {
84 crypto_ahash_set_flags(hash, CRYPTO_TFM_RES_BAD_KEY_LEN);
85 return -EINVAL;
86 }
87 *mctx = le32_to_cpup((__le32 *)key);
88 return 0;
89}
90
91static int crc32c_intel_init(struct ahash_request *req)
92{
93 u32 *mctx = crypto_ahash_ctx(crypto_ahash_reqtfm(req));
94 u32 *crcp = ahash_request_ctx(req);
95
96 *crcp = *mctx;
97
98 return 0;
99}
100
101static int crc32c_intel_update(struct ahash_request *req)
102{
103 struct crypto_hash_walk walk;
104 u32 *crcp = ahash_request_ctx(req);
105 u32 crc = *crcp;
106 int nbytes;
107
108 for (nbytes = crypto_hash_walk_first(req, &walk); nbytes;
109 nbytes = crypto_hash_walk_done(&walk, 0))
110 crc = crc32c_intel_le_hw(crc, walk.data, nbytes);
111
112 *crcp = crc;
113 return 0;
114}
115
116static int crc32c_intel_final(struct ahash_request *req)
117{
118 u32 *crcp = ahash_request_ctx(req);
119
120 *(__le32 *)req->result = ~cpu_to_le32p(crcp);
121 return 0;
122}
123
124static int crc32c_intel_digest(struct ahash_request *req)
125{
126 struct crypto_hash_walk walk;
127 u32 *mctx = crypto_ahash_ctx(crypto_ahash_reqtfm(req));
128 u32 crc = *mctx;
129 int nbytes;
130
131 for (nbytes = crypto_hash_walk_first(req, &walk); nbytes;
132 nbytes = crypto_hash_walk_done(&walk, 0))
133 crc = crc32c_intel_le_hw(crc, walk.data, nbytes);
134
135 *(__le32 *)req->result = ~cpu_to_le32(crc);
136 return 0;
137}
138
139static int crc32c_intel_cra_init(struct crypto_tfm *tfm)
140{
141 u32 *key = crypto_tfm_ctx(tfm);
142
143 *key = ~0;
144
145 tfm->crt_ahash.reqsize = sizeof(u32);
146
147 return 0;
148}
149
150static struct crypto_alg alg = {
151 .cra_name = "crc32c",
152 .cra_driver_name = "crc32c-intel",
153 .cra_priority = 200,
154 .cra_flags = CRYPTO_ALG_TYPE_AHASH,
155 .cra_blocksize = CHKSUM_BLOCK_SIZE,
156 .cra_alignmask = 3,
157 .cra_ctxsize = sizeof(u32),
158 .cra_module = THIS_MODULE,
159 .cra_list = LIST_HEAD_INIT(alg.cra_list),
160 .cra_init = crc32c_intel_cra_init,
161 .cra_type = &crypto_ahash_type,
162 .cra_u = {
163 .ahash = {
164 .digestsize = CHKSUM_DIGEST_SIZE,
165 .setkey = crc32c_intel_setkey,
166 .init = crc32c_intel_init,
167 .update = crc32c_intel_update,
168 .final = crc32c_intel_final,
169 .digest = crc32c_intel_digest,
170 }
171 }
172};
173
174
175static int __init crc32c_intel_mod_init(void)
176{
177 if (cpu_has_xmm4_2)
178 return crypto_register_alg(&alg);
179 else
180 return -ENODEV;
181}
182
183static void __exit crc32c_intel_mod_fini(void)
184{
185 crypto_unregister_alg(&alg);
186}
187
188module_init(crc32c_intel_mod_init);
189module_exit(crc32c_intel_mod_fini);
190
191MODULE_AUTHOR("Austin Zhang <austin.zhang@intel.com>, Kent Liu <kent.liu@intel.com>");
192MODULE_DESCRIPTION("CRC32c (Castagnoli) optimization using Intel Hardware.");
193MODULE_LICENSE("GPL");
194
195MODULE_ALIAS("crc32c");
196MODULE_ALIAS("crc32c-intel");
197
diff --git a/arch/x86/ia32/ia32_aout.c b/arch/x86/ia32/ia32_aout.c
index a0e1dbe67dc1..127ec3f07214 100644
--- a/arch/x86/ia32/ia32_aout.c
+++ b/arch/x86/ia32/ia32_aout.c
@@ -85,8 +85,10 @@ static void dump_thread32(struct pt_regs *regs, struct user32 *dump)
85 dump->regs.ax = regs->ax; 85 dump->regs.ax = regs->ax;
86 dump->regs.ds = current->thread.ds; 86 dump->regs.ds = current->thread.ds;
87 dump->regs.es = current->thread.es; 87 dump->regs.es = current->thread.es;
88 asm("movl %%fs,%0" : "=r" (fs)); dump->regs.fs = fs; 88 savesegment(fs, fs);
89 asm("movl %%gs,%0" : "=r" (gs)); dump->regs.gs = gs; 89 dump->regs.fs = fs;
90 savesegment(gs, gs);
91 dump->regs.gs = gs;
90 dump->regs.orig_ax = regs->orig_ax; 92 dump->regs.orig_ax = regs->orig_ax;
91 dump->regs.ip = regs->ip; 93 dump->regs.ip = regs->ip;
92 dump->regs.cs = regs->cs; 94 dump->regs.cs = regs->cs;
@@ -430,8 +432,9 @@ beyond_if:
430 current->mm->start_stack = 432 current->mm->start_stack =
431 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm); 433 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
432 /* start thread */ 434 /* start thread */
433 asm volatile("movl %0,%%fs" :: "r" (0)); \ 435 loadsegment(fs, 0);
434 asm volatile("movl %0,%%es; movl %0,%%ds": :"r" (__USER32_DS)); 436 loadsegment(ds, __USER32_DS);
437 loadsegment(es, __USER32_DS);
435 load_gs_index(0); 438 load_gs_index(0);
436 (regs)->ip = ex.a_entry; 439 (regs)->ip = ex.a_entry;
437 (regs)->sp = current->mm->start_stack; 440 (regs)->sp = current->mm->start_stack;
diff --git a/arch/x86/ia32/ia32_signal.c b/arch/x86/ia32/ia32_signal.c
index 20af4c79579a..4bc02b23674b 100644
--- a/arch/x86/ia32/ia32_signal.c
+++ b/arch/x86/ia32/ia32_signal.c
@@ -179,9 +179,10 @@ struct sigframe
179 u32 pretcode; 179 u32 pretcode;
180 int sig; 180 int sig;
181 struct sigcontext_ia32 sc; 181 struct sigcontext_ia32 sc;
182 struct _fpstate_ia32 fpstate; 182 struct _fpstate_ia32 fpstate_unused; /* look at kernel/sigframe.h */
183 unsigned int extramask[_COMPAT_NSIG_WORDS-1]; 183 unsigned int extramask[_COMPAT_NSIG_WORDS-1];
184 char retcode[8]; 184 char retcode[8];
185 /* fp state follows here */
185}; 186};
186 187
187struct rt_sigframe 188struct rt_sigframe
@@ -192,8 +193,8 @@ struct rt_sigframe
192 u32 puc; 193 u32 puc;
193 compat_siginfo_t info; 194 compat_siginfo_t info;
194 struct ucontext_ia32 uc; 195 struct ucontext_ia32 uc;
195 struct _fpstate_ia32 fpstate;
196 char retcode[8]; 196 char retcode[8];
197 /* fp state follows here */
197}; 198};
198 199
199#define COPY(x) { \ 200#define COPY(x) { \
@@ -206,7 +207,7 @@ struct rt_sigframe
206 { unsigned int cur; \ 207 { unsigned int cur; \
207 unsigned short pre; \ 208 unsigned short pre; \
208 err |= __get_user(pre, &sc->seg); \ 209 err |= __get_user(pre, &sc->seg); \
209 asm volatile("movl %%" #seg ",%0" : "=r" (cur)); \ 210 savesegment(seg, cur); \
210 pre |= mask; \ 211 pre |= mask; \
211 if (pre != cur) loadsegment(seg, pre); } 212 if (pre != cur) loadsegment(seg, pre); }
212 213
@@ -215,7 +216,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
215 unsigned int *peax) 216 unsigned int *peax)
216{ 217{
217 unsigned int tmpflags, gs, oldgs, err = 0; 218 unsigned int tmpflags, gs, oldgs, err = 0;
218 struct _fpstate_ia32 __user *buf; 219 void __user *buf;
219 u32 tmp; 220 u32 tmp;
220 221
221 /* Always make any pending restarted system calls return -EINTR */ 222 /* Always make any pending restarted system calls return -EINTR */
@@ -235,7 +236,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
235 */ 236 */
236 err |= __get_user(gs, &sc->gs); 237 err |= __get_user(gs, &sc->gs);
237 gs |= 3; 238 gs |= 3;
238 asm("movl %%gs,%0" : "=r" (oldgs)); 239 savesegment(gs, oldgs);
239 if (gs != oldgs) 240 if (gs != oldgs)
240 load_gs_index(gs); 241 load_gs_index(gs);
241 242
@@ -259,26 +260,12 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
259 260
260 err |= __get_user(tmp, &sc->fpstate); 261 err |= __get_user(tmp, &sc->fpstate);
261 buf = compat_ptr(tmp); 262 buf = compat_ptr(tmp);
262 if (buf) { 263 err |= restore_i387_xstate_ia32(buf);
263 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
264 goto badframe;
265 err |= restore_i387_ia32(buf);
266 } else {
267 struct task_struct *me = current;
268
269 if (used_math()) {
270 clear_fpu(me);
271 clear_used_math();
272 }
273 }
274 264
275 err |= __get_user(tmp, &sc->ax); 265 err |= __get_user(tmp, &sc->ax);
276 *peax = tmp; 266 *peax = tmp;
277 267
278 return err; 268 return err;
279
280badframe:
281 return 1;
282} 269}
283 270
284asmlinkage long sys32_sigreturn(struct pt_regs *regs) 271asmlinkage long sys32_sigreturn(struct pt_regs *regs)
@@ -350,46 +337,42 @@ badframe:
350 */ 337 */
351 338
352static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc, 339static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc,
353 struct _fpstate_ia32 __user *fpstate, 340 void __user *fpstate,
354 struct pt_regs *regs, unsigned int mask) 341 struct pt_regs *regs, unsigned int mask)
355{ 342{
356 int tmp, err = 0; 343 int tmp, err = 0;
357 344
358 tmp = 0; 345 savesegment(gs, tmp);
359 __asm__("movl %%gs,%0" : "=r"(tmp): "0"(tmp));
360 err |= __put_user(tmp, (unsigned int __user *)&sc->gs); 346 err |= __put_user(tmp, (unsigned int __user *)&sc->gs);
361 __asm__("movl %%fs,%0" : "=r"(tmp): "0"(tmp)); 347 savesegment(fs, tmp);
362 err |= __put_user(tmp, (unsigned int __user *)&sc->fs); 348 err |= __put_user(tmp, (unsigned int __user *)&sc->fs);
363 __asm__("movl %%ds,%0" : "=r"(tmp): "0"(tmp)); 349 savesegment(ds, tmp);
364 err |= __put_user(tmp, (unsigned int __user *)&sc->ds); 350 err |= __put_user(tmp, (unsigned int __user *)&sc->ds);
365 __asm__("movl %%es,%0" : "=r"(tmp): "0"(tmp)); 351 savesegment(es, tmp);
366 err |= __put_user(tmp, (unsigned int __user *)&sc->es); 352 err |= __put_user(tmp, (unsigned int __user *)&sc->es);
367 353
368 err |= __put_user((u32)regs->di, &sc->di); 354 err |= __put_user(regs->di, &sc->di);
369 err |= __put_user((u32)regs->si, &sc->si); 355 err |= __put_user(regs->si, &sc->si);
370 err |= __put_user((u32)regs->bp, &sc->bp); 356 err |= __put_user(regs->bp, &sc->bp);
371 err |= __put_user((u32)regs->sp, &sc->sp); 357 err |= __put_user(regs->sp, &sc->sp);
372 err |= __put_user((u32)regs->bx, &sc->bx); 358 err |= __put_user(regs->bx, &sc->bx);
373 err |= __put_user((u32)regs->dx, &sc->dx); 359 err |= __put_user(regs->dx, &sc->dx);
374 err |= __put_user((u32)regs->cx, &sc->cx); 360 err |= __put_user(regs->cx, &sc->cx);
375 err |= __put_user((u32)regs->ax, &sc->ax); 361 err |= __put_user(regs->ax, &sc->ax);
376 err |= __put_user((u32)regs->cs, &sc->cs); 362 err |= __put_user(regs->cs, &sc->cs);
377 err |= __put_user((u32)regs->ss, &sc->ss); 363 err |= __put_user(regs->ss, &sc->ss);
378 err |= __put_user(current->thread.trap_no, &sc->trapno); 364 err |= __put_user(current->thread.trap_no, &sc->trapno);
379 err |= __put_user(current->thread.error_code, &sc->err); 365 err |= __put_user(current->thread.error_code, &sc->err);
380 err |= __put_user((u32)regs->ip, &sc->ip); 366 err |= __put_user(regs->ip, &sc->ip);
381 err |= __put_user((u32)regs->flags, &sc->flags); 367 err |= __put_user(regs->flags, &sc->flags);
382 err |= __put_user((u32)regs->sp, &sc->sp_at_signal); 368 err |= __put_user(regs->sp, &sc->sp_at_signal);
383 369
384 tmp = save_i387_ia32(fpstate); 370 tmp = save_i387_xstate_ia32(fpstate);
385 if (tmp < 0) 371 if (tmp < 0)
386 err = -EFAULT; 372 err = -EFAULT;
387 else { 373 else
388 clear_used_math();
389 stts();
390 err |= __put_user(ptr_to_compat(tmp ? fpstate : NULL), 374 err |= __put_user(ptr_to_compat(tmp ? fpstate : NULL),
391 &sc->fpstate); 375 &sc->fpstate);
392 }
393 376
394 /* non-iBCS2 extensions.. */ 377 /* non-iBCS2 extensions.. */
395 err |= __put_user(mask, &sc->oldmask); 378 err |= __put_user(mask, &sc->oldmask);
@@ -402,7 +385,8 @@ static int ia32_setup_sigcontext(struct sigcontext_ia32 __user *sc,
402 * Determine which stack to use.. 385 * Determine which stack to use..
403 */ 386 */
404static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, 387static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs,
405 size_t frame_size) 388 size_t frame_size,
389 void **fpstate)
406{ 390{
407 unsigned long sp; 391 unsigned long sp;
408 392
@@ -421,6 +405,11 @@ static void __user *get_sigframe(struct k_sigaction *ka, struct pt_regs *regs,
421 ka->sa.sa_restorer) 405 ka->sa.sa_restorer)
422 sp = (unsigned long) ka->sa.sa_restorer; 406 sp = (unsigned long) ka->sa.sa_restorer;
423 407
408 if (used_math()) {
409 sp = sp - sig_xstate_ia32_size;
410 *fpstate = (struct _fpstate_ia32 *) sp;
411 }
412
424 sp -= frame_size; 413 sp -= frame_size;
425 /* Align the stack pointer according to the i386 ABI, 414 /* Align the stack pointer according to the i386 ABI,
426 * i.e. so that on function entry ((sp + 4) & 15) == 0. */ 415 * i.e. so that on function entry ((sp + 4) & 15) == 0. */
@@ -434,6 +423,7 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
434 struct sigframe __user *frame; 423 struct sigframe __user *frame;
435 void __user *restorer; 424 void __user *restorer;
436 int err = 0; 425 int err = 0;
426 void __user *fpstate = NULL;
437 427
438 /* copy_to_user optimizes that into a single 8 byte store */ 428 /* copy_to_user optimizes that into a single 8 byte store */
439 static const struct { 429 static const struct {
@@ -448,25 +438,21 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
448 0, 438 0,
449 }; 439 };
450 440
451 frame = get_sigframe(ka, regs, sizeof(*frame)); 441 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
452 442
453 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 443 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
454 goto give_sigsegv; 444 return -EFAULT;
455 445
456 err |= __put_user(sig, &frame->sig); 446 if (__put_user(sig, &frame->sig))
457 if (err) 447 return -EFAULT;
458 goto give_sigsegv;
459 448
460 err |= ia32_setup_sigcontext(&frame->sc, &frame->fpstate, regs, 449 if (ia32_setup_sigcontext(&frame->sc, fpstate, regs, set->sig[0]))
461 set->sig[0]); 450 return -EFAULT;
462 if (err)
463 goto give_sigsegv;
464 451
465 if (_COMPAT_NSIG_WORDS > 1) { 452 if (_COMPAT_NSIG_WORDS > 1) {
466 err |= __copy_to_user(frame->extramask, &set->sig[1], 453 if (__copy_to_user(frame->extramask, &set->sig[1],
467 sizeof(frame->extramask)); 454 sizeof(frame->extramask)))
468 if (err) 455 return -EFAULT;
469 goto give_sigsegv;
470 } 456 }
471 457
472 if (ka->sa.sa_flags & SA_RESTORER) { 458 if (ka->sa.sa_flags & SA_RESTORER) {
@@ -487,7 +473,7 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
487 */ 473 */
488 err |= __copy_to_user(frame->retcode, &code, 8); 474 err |= __copy_to_user(frame->retcode, &code, 8);
489 if (err) 475 if (err)
490 goto give_sigsegv; 476 return -EFAULT;
491 477
492 /* Set up registers for signal handler */ 478 /* Set up registers for signal handler */
493 regs->sp = (unsigned long) frame; 479 regs->sp = (unsigned long) frame;
@@ -498,8 +484,8 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
498 regs->dx = 0; 484 regs->dx = 0;
499 regs->cx = 0; 485 regs->cx = 0;
500 486
501 asm volatile("movl %0,%%ds" :: "r" (__USER32_DS)); 487 loadsegment(ds, __USER32_DS);
502 asm volatile("movl %0,%%es" :: "r" (__USER32_DS)); 488 loadsegment(es, __USER32_DS);
503 489
504 regs->cs = __USER32_CS; 490 regs->cs = __USER32_CS;
505 regs->ss = __USER32_DS; 491 regs->ss = __USER32_DS;
@@ -510,10 +496,6 @@ int ia32_setup_frame(int sig, struct k_sigaction *ka,
510#endif 496#endif
511 497
512 return 0; 498 return 0;
513
514give_sigsegv:
515 force_sigsegv(sig, current);
516 return -EFAULT;
517} 499}
518 500
519int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 501int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
@@ -522,6 +504,7 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
522 struct rt_sigframe __user *frame; 504 struct rt_sigframe __user *frame;
523 void __user *restorer; 505 void __user *restorer;
524 int err = 0; 506 int err = 0;
507 void __user *fpstate = NULL;
525 508
526 /* __copy_to_user optimizes that into a single 8 byte store */ 509 /* __copy_to_user optimizes that into a single 8 byte store */
527 static const struct { 510 static const struct {
@@ -537,30 +520,33 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
537 0, 520 0,
538 }; 521 };
539 522
540 frame = get_sigframe(ka, regs, sizeof(*frame)); 523 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
541 524
542 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 525 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
543 goto give_sigsegv; 526 return -EFAULT;
544 527
545 err |= __put_user(sig, &frame->sig); 528 err |= __put_user(sig, &frame->sig);
546 err |= __put_user(ptr_to_compat(&frame->info), &frame->pinfo); 529 err |= __put_user(ptr_to_compat(&frame->info), &frame->pinfo);
547 err |= __put_user(ptr_to_compat(&frame->uc), &frame->puc); 530 err |= __put_user(ptr_to_compat(&frame->uc), &frame->puc);
548 err |= copy_siginfo_to_user32(&frame->info, info); 531 err |= copy_siginfo_to_user32(&frame->info, info);
549 if (err) 532 if (err)
550 goto give_sigsegv; 533 return -EFAULT;
551 534
552 /* Create the ucontext. */ 535 /* Create the ucontext. */
553 err |= __put_user(0, &frame->uc.uc_flags); 536 if (cpu_has_xsave)
537 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
538 else
539 err |= __put_user(0, &frame->uc.uc_flags);
554 err |= __put_user(0, &frame->uc.uc_link); 540 err |= __put_user(0, &frame->uc.uc_link);
555 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 541 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
556 err |= __put_user(sas_ss_flags(regs->sp), 542 err |= __put_user(sas_ss_flags(regs->sp),
557 &frame->uc.uc_stack.ss_flags); 543 &frame->uc.uc_stack.ss_flags);
558 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size); 544 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size);
559 err |= ia32_setup_sigcontext(&frame->uc.uc_mcontext, &frame->fpstate, 545 err |= ia32_setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
560 regs, set->sig[0]); 546 regs, set->sig[0]);
561 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)); 547 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
562 if (err) 548 if (err)
563 goto give_sigsegv; 549 return -EFAULT;
564 550
565 if (ka->sa.sa_flags & SA_RESTORER) 551 if (ka->sa.sa_flags & SA_RESTORER)
566 restorer = ka->sa.sa_restorer; 552 restorer = ka->sa.sa_restorer;
@@ -575,7 +561,7 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
575 */ 561 */
576 err |= __copy_to_user(frame->retcode, &code, 8); 562 err |= __copy_to_user(frame->retcode, &code, 8);
577 if (err) 563 if (err)
578 goto give_sigsegv; 564 return -EFAULT;
579 565
580 /* Set up registers for signal handler */ 566 /* Set up registers for signal handler */
581 regs->sp = (unsigned long) frame; 567 regs->sp = (unsigned long) frame;
@@ -591,8 +577,8 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
591 regs->dx = (unsigned long) &frame->info; 577 regs->dx = (unsigned long) &frame->info;
592 regs->cx = (unsigned long) &frame->uc; 578 regs->cx = (unsigned long) &frame->uc;
593 579
594 asm volatile("movl %0,%%ds" :: "r" (__USER32_DS)); 580 loadsegment(ds, __USER32_DS);
595 asm volatile("movl %0,%%es" :: "r" (__USER32_DS)); 581 loadsegment(es, __USER32_DS);
596 582
597 regs->cs = __USER32_CS; 583 regs->cs = __USER32_CS;
598 regs->ss = __USER32_DS; 584 regs->ss = __USER32_DS;
@@ -603,8 +589,4 @@ int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
603#endif 589#endif
604 590
605 return 0; 591 return 0;
606
607give_sigsegv:
608 force_sigsegv(sig, current);
609 return -EFAULT;
610} 592}
diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index ffc1bb4fed7d..256b00b61892 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -39,11 +39,11 @@
39 .endm 39 .endm
40 40
41 /* clobbers %eax */ 41 /* clobbers %eax */
42 .macro CLEAR_RREGS 42 .macro CLEAR_RREGS _r9=rax
43 xorl %eax,%eax 43 xorl %eax,%eax
44 movq %rax,R11(%rsp) 44 movq %rax,R11(%rsp)
45 movq %rax,R10(%rsp) 45 movq %rax,R10(%rsp)
46 movq %rax,R9(%rsp) 46 movq %\_r9,R9(%rsp)
47 movq %rax,R8(%rsp) 47 movq %rax,R8(%rsp)
48 .endm 48 .endm
49 49
@@ -52,11 +52,10 @@
52 * We don't reload %eax because syscall_trace_enter() returned 52 * We don't reload %eax because syscall_trace_enter() returned
53 * the value it wants us to use in the table lookup. 53 * the value it wants us to use in the table lookup.
54 */ 54 */
55 .macro LOAD_ARGS32 offset 55 .macro LOAD_ARGS32 offset, _r9=0
56 movl \offset(%rsp),%r11d 56 .if \_r9
57 movl \offset+8(%rsp),%r10d
58 movl \offset+16(%rsp),%r9d 57 movl \offset+16(%rsp),%r9d
59 movl \offset+24(%rsp),%r8d 58 .endif
60 movl \offset+40(%rsp),%ecx 59 movl \offset+40(%rsp),%ecx
61 movl \offset+48(%rsp),%edx 60 movl \offset+48(%rsp),%edx
62 movl \offset+56(%rsp),%esi 61 movl \offset+56(%rsp),%esi
@@ -145,7 +144,7 @@ ENTRY(ia32_sysenter_target)
145 SAVE_ARGS 0,0,1 144 SAVE_ARGS 0,0,1
146 /* no need to do an access_ok check here because rbp has been 145 /* no need to do an access_ok check here because rbp has been
147 32bit zero extended */ 146 32bit zero extended */
1481: movl (%rbp),%r9d 1471: movl (%rbp),%ebp
149 .section __ex_table,"a" 148 .section __ex_table,"a"
150 .quad 1b,ia32_badarg 149 .quad 1b,ia32_badarg
151 .previous 150 .previous
@@ -157,7 +156,7 @@ ENTRY(ia32_sysenter_target)
157 cmpl $(IA32_NR_syscalls-1),%eax 156 cmpl $(IA32_NR_syscalls-1),%eax
158 ja ia32_badsys 157 ja ia32_badsys
159sysenter_do_call: 158sysenter_do_call:
160 IA32_ARG_FIXUP 1 159 IA32_ARG_FIXUP
161sysenter_dispatch: 160sysenter_dispatch:
162 call *ia32_sys_call_table(,%rax,8) 161 call *ia32_sys_call_table(,%rax,8)
163 movq %rax,RAX-ARGOFFSET(%rsp) 162 movq %rax,RAX-ARGOFFSET(%rsp)
@@ -234,20 +233,17 @@ sysexit_audit:
234#endif 233#endif
235 234
236sysenter_tracesys: 235sysenter_tracesys:
237 xchgl %r9d,%ebp
238#ifdef CONFIG_AUDITSYSCALL 236#ifdef CONFIG_AUDITSYSCALL
239 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%r10) 237 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%r10)
240 jz sysenter_auditsys 238 jz sysenter_auditsys
241#endif 239#endif
242 SAVE_REST 240 SAVE_REST
243 CLEAR_RREGS 241 CLEAR_RREGS
244 movq %r9,R9(%rsp)
245 movq $-ENOSYS,RAX(%rsp)/* ptrace can change this for a bad syscall */ 242 movq $-ENOSYS,RAX(%rsp)/* ptrace can change this for a bad syscall */
246 movq %rsp,%rdi /* &pt_regs -> arg1 */ 243 movq %rsp,%rdi /* &pt_regs -> arg1 */
247 call syscall_trace_enter 244 call syscall_trace_enter
248 LOAD_ARGS32 ARGOFFSET /* reload args from stack in case ptrace changed it */ 245 LOAD_ARGS32 ARGOFFSET /* reload args from stack in case ptrace changed it */
249 RESTORE_REST 246 RESTORE_REST
250 xchgl %ebp,%r9d
251 cmpl $(IA32_NR_syscalls-1),%eax 247 cmpl $(IA32_NR_syscalls-1),%eax
252 ja int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */ 248 ja int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
253 jmp sysenter_do_call 249 jmp sysenter_do_call
@@ -314,9 +310,9 @@ ENTRY(ia32_cstar_target)
314 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%r10) 310 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%r10)
315 CFI_REMEMBER_STATE 311 CFI_REMEMBER_STATE
316 jnz cstar_tracesys 312 jnz cstar_tracesys
317cstar_do_call:
318 cmpl $IA32_NR_syscalls-1,%eax 313 cmpl $IA32_NR_syscalls-1,%eax
319 ja ia32_badsys 314 ja ia32_badsys
315cstar_do_call:
320 IA32_ARG_FIXUP 1 316 IA32_ARG_FIXUP 1
321cstar_dispatch: 317cstar_dispatch:
322 call *ia32_sys_call_table(,%rax,8) 318 call *ia32_sys_call_table(,%rax,8)
@@ -357,15 +353,13 @@ cstar_tracesys:
357#endif 353#endif
358 xchgl %r9d,%ebp 354 xchgl %r9d,%ebp
359 SAVE_REST 355 SAVE_REST
360 CLEAR_RREGS 356 CLEAR_RREGS r9
361 movq %r9,R9(%rsp)
362 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */ 357 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
363 movq %rsp,%rdi /* &pt_regs -> arg1 */ 358 movq %rsp,%rdi /* &pt_regs -> arg1 */
364 call syscall_trace_enter 359 call syscall_trace_enter
365 LOAD_ARGS32 ARGOFFSET /* reload args from stack in case ptrace changed it */ 360 LOAD_ARGS32 ARGOFFSET, 1 /* reload args from stack in case ptrace changed it */
366 RESTORE_REST 361 RESTORE_REST
367 xchgl %ebp,%r9d 362 xchgl %ebp,%r9d
368 movl RSP-ARGOFFSET(%rsp), %r8d
369 cmpl $(IA32_NR_syscalls-1),%eax 363 cmpl $(IA32_NR_syscalls-1),%eax
370 ja int_ret_from_sys_call /* cstar_tracesys has set RAX(%rsp) */ 364 ja int_ret_from_sys_call /* cstar_tracesys has set RAX(%rsp) */
371 jmp cstar_do_call 365 jmp cstar_do_call
@@ -577,8 +571,8 @@ ia32_sys_call_table:
577 .quad compat_sys_setrlimit /* 75 */ 571 .quad compat_sys_setrlimit /* 75 */
578 .quad compat_sys_old_getrlimit /* old_getrlimit */ 572 .quad compat_sys_old_getrlimit /* old_getrlimit */
579 .quad compat_sys_getrusage 573 .quad compat_sys_getrusage
580 .quad sys32_gettimeofday 574 .quad compat_sys_gettimeofday
581 .quad sys32_settimeofday 575 .quad compat_sys_settimeofday
582 .quad sys_getgroups16 /* 80 */ 576 .quad sys_getgroups16 /* 80 */
583 .quad sys_setgroups16 577 .quad sys_setgroups16
584 .quad sys32_old_select 578 .quad sys32_old_select
diff --git a/arch/x86/ia32/sys_ia32.c b/arch/x86/ia32/sys_ia32.c
index d3c64088b981..2e09dcd3c0a6 100644
--- a/arch/x86/ia32/sys_ia32.c
+++ b/arch/x86/ia32/sys_ia32.c
@@ -49,41 +49,6 @@
49 49
50#define AA(__x) ((unsigned long)(__x)) 50#define AA(__x) ((unsigned long)(__x))
51 51
52int cp_compat_stat(struct kstat *kbuf, struct compat_stat __user *ubuf)
53{
54 compat_ino_t ino;
55
56 typeof(ubuf->st_uid) uid = 0;
57 typeof(ubuf->st_gid) gid = 0;
58 SET_UID(uid, kbuf->uid);
59 SET_GID(gid, kbuf->gid);
60 if (!old_valid_dev(kbuf->dev) || !old_valid_dev(kbuf->rdev))
61 return -EOVERFLOW;
62 if (kbuf->size >= 0x7fffffff)
63 return -EOVERFLOW;
64 ino = kbuf->ino;
65 if (sizeof(ino) < sizeof(kbuf->ino) && ino != kbuf->ino)
66 return -EOVERFLOW;
67 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(struct compat_stat)) ||
68 __put_user(old_encode_dev(kbuf->dev), &ubuf->st_dev) ||
69 __put_user(ino, &ubuf->st_ino) ||
70 __put_user(kbuf->mode, &ubuf->st_mode) ||
71 __put_user(kbuf->nlink, &ubuf->st_nlink) ||
72 __put_user(uid, &ubuf->st_uid) ||
73 __put_user(gid, &ubuf->st_gid) ||
74 __put_user(old_encode_dev(kbuf->rdev), &ubuf->st_rdev) ||
75 __put_user(kbuf->size, &ubuf->st_size) ||
76 __put_user(kbuf->atime.tv_sec, &ubuf->st_atime) ||
77 __put_user(kbuf->atime.tv_nsec, &ubuf->st_atime_nsec) ||
78 __put_user(kbuf->mtime.tv_sec, &ubuf->st_mtime) ||
79 __put_user(kbuf->mtime.tv_nsec, &ubuf->st_mtime_nsec) ||
80 __put_user(kbuf->ctime.tv_sec, &ubuf->st_ctime) ||
81 __put_user(kbuf->ctime.tv_nsec, &ubuf->st_ctime_nsec) ||
82 __put_user(kbuf->blksize, &ubuf->st_blksize) ||
83 __put_user(kbuf->blocks, &ubuf->st_blocks))
84 return -EFAULT;
85 return 0;
86}
87 52
88asmlinkage long sys32_truncate64(char __user *filename, 53asmlinkage long sys32_truncate64(char __user *filename,
89 unsigned long offset_low, 54 unsigned long offset_low,
@@ -402,75 +367,11 @@ asmlinkage long sys32_rt_sigprocmask(int how, compat_sigset_t __user *set,
402 return 0; 367 return 0;
403} 368}
404 369
405static inline long get_tv32(struct timeval *o, struct compat_timeval __user *i)
406{
407 int err = -EFAULT;
408
409 if (access_ok(VERIFY_READ, i, sizeof(*i))) {
410 err = __get_user(o->tv_sec, &i->tv_sec);
411 err |= __get_user(o->tv_usec, &i->tv_usec);
412 }
413 return err;
414}
415
416static inline long put_tv32(struct compat_timeval __user *o, struct timeval *i)
417{
418 int err = -EFAULT;
419
420 if (access_ok(VERIFY_WRITE, o, sizeof(*o))) {
421 err = __put_user(i->tv_sec, &o->tv_sec);
422 err |= __put_user(i->tv_usec, &o->tv_usec);
423 }
424 return err;
425}
426
427asmlinkage long sys32_alarm(unsigned int seconds) 370asmlinkage long sys32_alarm(unsigned int seconds)
428{ 371{
429 return alarm_setitimer(seconds); 372 return alarm_setitimer(seconds);
430} 373}
431 374
432/*
433 * Translations due to time_t size differences. Which affects all
434 * sorts of things, like timeval and itimerval.
435 */
436asmlinkage long sys32_gettimeofday(struct compat_timeval __user *tv,
437 struct timezone __user *tz)
438{
439 if (tv) {
440 struct timeval ktv;
441
442 do_gettimeofday(&ktv);
443 if (put_tv32(tv, &ktv))
444 return -EFAULT;
445 }
446 if (tz) {
447 if (copy_to_user(tz, &sys_tz, sizeof(sys_tz)))
448 return -EFAULT;
449 }
450 return 0;
451}
452
453asmlinkage long sys32_settimeofday(struct compat_timeval __user *tv,
454 struct timezone __user *tz)
455{
456 struct timeval ktv;
457 struct timespec kts;
458 struct timezone ktz;
459
460 if (tv) {
461 if (get_tv32(&ktv, tv))
462 return -EFAULT;
463 kts.tv_sec = ktv.tv_sec;
464 kts.tv_nsec = ktv.tv_usec * NSEC_PER_USEC;
465 }
466 if (tz) {
467 if (copy_from_user(&ktz, tz, sizeof(ktz)))
468 return -EFAULT;
469 }
470
471 return do_sys_settimeofday(tv ? &kts : NULL, tz ? &ktz : NULL);
472}
473
474struct sel_arg_struct { 375struct sel_arg_struct {
475 unsigned int n; 376 unsigned int n;
476 unsigned int inp; 377 unsigned int inp;
@@ -556,15 +457,6 @@ asmlinkage long sys32_rt_sigqueueinfo(int pid, int sig,
556 return ret; 457 return ret;
557} 458}
558 459
559/* These are here just in case some old ia32 binary calls it. */
560asmlinkage long sys32_pause(void)
561{
562 current->state = TASK_INTERRUPTIBLE;
563 schedule();
564 return -ERESTARTNOHAND;
565}
566
567
568#ifdef CONFIG_SYSCTL_SYSCALL 460#ifdef CONFIG_SYSCTL_SYSCALL
569struct sysctl_ia32 { 461struct sysctl_ia32 {
570 unsigned int name; 462 unsigned int name;
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 3db651fc8ec5..0d41f0343dc0 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -10,7 +10,7 @@ ifdef CONFIG_FTRACE
10# Do not profile debug and lowlevel utilities 10# Do not profile debug and lowlevel utilities
11CFLAGS_REMOVE_tsc.o = -pg 11CFLAGS_REMOVE_tsc.o = -pg
12CFLAGS_REMOVE_rtc.o = -pg 12CFLAGS_REMOVE_rtc.o = -pg
13CFLAGS_REMOVE_paravirt.o = -pg 13CFLAGS_REMOVE_paravirt-spinlocks.o = -pg
14endif 14endif
15 15
16# 16#
@@ -23,7 +23,7 @@ CFLAGS_hpet.o := $(nostackp)
23CFLAGS_tsc.o := $(nostackp) 23CFLAGS_tsc.o := $(nostackp)
24 24
25obj-y := process_$(BITS).o signal_$(BITS).o entry_$(BITS).o 25obj-y := process_$(BITS).o signal_$(BITS).o entry_$(BITS).o
26obj-y += traps_$(BITS).o irq_$(BITS).o 26obj-y += traps.o irq_$(BITS).o dumpstack_$(BITS).o
27obj-y += time_$(BITS).o ioport.o ldt.o 27obj-y += time_$(BITS).o ioport.o ldt.o
28obj-y += setup.o i8259.o irqinit_$(BITS).o setup_percpu.o 28obj-y += setup.o i8259.o irqinit_$(BITS).o setup_percpu.o
29obj-$(CONFIG_X86_VISWS) += visws_quirks.o 29obj-$(CONFIG_X86_VISWS) += visws_quirks.o
@@ -38,7 +38,7 @@ obj-y += tsc.o io_delay.o rtc.o
38 38
39obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o 39obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o
40obj-y += process.o 40obj-y += process.o
41obj-y += i387.o 41obj-y += i387.o xsave.o
42obj-y += ptrace.o 42obj-y += ptrace.o
43obj-y += ds.o 43obj-y += ds.o
44obj-$(CONFIG_X86_32) += tls.o 44obj-$(CONFIG_X86_32) += tls.o
@@ -51,7 +51,6 @@ obj-$(CONFIG_X86_BIOS_REBOOT) += reboot.o
51obj-$(CONFIG_MCA) += mca_32.o 51obj-$(CONFIG_MCA) += mca_32.o
52obj-$(CONFIG_X86_MSR) += msr.o 52obj-$(CONFIG_X86_MSR) += msr.o
53obj-$(CONFIG_X86_CPUID) += cpuid.o 53obj-$(CONFIG_X86_CPUID) += cpuid.o
54obj-$(CONFIG_MICROCODE) += microcode.o
55obj-$(CONFIG_PCI) += early-quirks.o 54obj-$(CONFIG_PCI) += early-quirks.o
56apm-y := apm_32.o 55apm-y := apm_32.o
57obj-$(CONFIG_APM) += apm.o 56obj-$(CONFIG_APM) += apm.o
@@ -69,6 +68,7 @@ obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o
69obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o 68obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o
70obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o 69obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o
71obj-$(CONFIG_X86_NUMAQ) += numaq_32.o 70obj-$(CONFIG_X86_NUMAQ) += numaq_32.o
71obj-$(CONFIG_X86_ES7000) += es7000_32.o
72obj-$(CONFIG_X86_SUMMIT_NUMA) += summit_32.o 72obj-$(CONFIG_X86_SUMMIT_NUMA) += summit_32.o
73obj-y += vsmp_64.o 73obj-y += vsmp_64.o
74obj-$(CONFIG_KPROBES) += kprobes.o 74obj-$(CONFIG_KPROBES) += kprobes.o
@@ -89,7 +89,7 @@ obj-$(CONFIG_DEBUG_NX_TEST) += test_nx.o
89obj-$(CONFIG_VMI) += vmi_32.o vmiclock_32.o 89obj-$(CONFIG_VMI) += vmi_32.o vmiclock_32.o
90obj-$(CONFIG_KVM_GUEST) += kvm.o 90obj-$(CONFIG_KVM_GUEST) += kvm.o
91obj-$(CONFIG_KVM_CLOCK) += kvmclock.o 91obj-$(CONFIG_KVM_CLOCK) += kvmclock.o
92obj-$(CONFIG_PARAVIRT) += paravirt.o paravirt_patch_$(BITS).o 92obj-$(CONFIG_PARAVIRT) += paravirt.o paravirt_patch_$(BITS).o paravirt-spinlocks.o
93obj-$(CONFIG_PARAVIRT_CLOCK) += pvclock.o 93obj-$(CONFIG_PARAVIRT_CLOCK) += pvclock.o
94 94
95obj-$(CONFIG_PCSPKR_PLATFORM) += pcspeaker.o 95obj-$(CONFIG_PCSPKR_PLATFORM) += pcspeaker.o
@@ -99,11 +99,18 @@ scx200-y += scx200_32.o
99 99
100obj-$(CONFIG_OLPC) += olpc.o 100obj-$(CONFIG_OLPC) += olpc.o
101 101
102microcode-y := microcode_core.o
103microcode-$(CONFIG_MICROCODE_INTEL) += microcode_intel.o
104microcode-$(CONFIG_MICROCODE_AMD) += microcode_amd.o
105obj-$(CONFIG_MICROCODE) += microcode.o
106
102### 107###
103# 64 bit specific files 108# 64 bit specific files
104ifeq ($(CONFIG_X86_64),y) 109ifeq ($(CONFIG_X86_64),y)
105 obj-y += genapic_64.o genapic_flat_64.o genx2apic_uv_x.o tlb_uv.o 110 obj-y += genapic_64.o genapic_flat_64.o genx2apic_uv_x.o tlb_uv.o
106 obj-y += bios_uv.o 111 obj-y += bios_uv.o
112 obj-y += genx2apic_cluster.o
113 obj-y += genx2apic_phys.o
107 obj-$(CONFIG_X86_PM_TIMER) += pmtimer_64.o 114 obj-$(CONFIG_X86_PM_TIMER) += pmtimer_64.o
108 obj-$(CONFIG_AUDIT) += audit_64.o 115 obj-$(CONFIG_AUDIT) += audit_64.o
109 116
diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index bfd10fd211cd..eb875cdc7367 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -58,7 +58,6 @@ EXPORT_SYMBOL(acpi_disabled);
58#ifdef CONFIG_X86_64 58#ifdef CONFIG_X86_64
59 59
60#include <asm/proto.h> 60#include <asm/proto.h>
61#include <asm/genapic.h>
62 61
63#else /* X86 */ 62#else /* X86 */
64 63
@@ -97,8 +96,6 @@ static u64 acpi_lapic_addr __initdata = APIC_DEFAULT_PHYS_BASE;
97#warning ACPI uses CMPXCHG, i486 and later hardware 96#warning ACPI uses CMPXCHG, i486 and later hardware
98#endif 97#endif
99 98
100static int acpi_mcfg_64bit_base_addr __initdata = FALSE;
101
102/* -------------------------------------------------------------------------- 99/* --------------------------------------------------------------------------
103 Boot-time Configuration 100 Boot-time Configuration
104 -------------------------------------------------------------------------- */ 101 -------------------------------------------------------------------------- */
@@ -160,6 +157,8 @@ char *__init __acpi_map_table(unsigned long phys, unsigned long size)
160struct acpi_mcfg_allocation *pci_mmcfg_config; 157struct acpi_mcfg_allocation *pci_mmcfg_config;
161int pci_mmcfg_config_num; 158int pci_mmcfg_config_num;
162 159
160static int acpi_mcfg_64bit_base_addr __initdata = FALSE;
161
163static int __init acpi_mcfg_oem_check(struct acpi_table_mcfg *mcfg) 162static int __init acpi_mcfg_oem_check(struct acpi_table_mcfg *mcfg)
164{ 163{
165 if (!strcmp(mcfg->header.oem_id, "SGI")) 164 if (!strcmp(mcfg->header.oem_id, "SGI"))
@@ -253,10 +252,8 @@ static void __cpuinit acpi_register_lapic(int id, u8 enabled)
253 return; 252 return;
254 } 253 }
255 254
256#ifdef CONFIG_X86_32
257 if (boot_cpu_physical_apicid != -1U) 255 if (boot_cpu_physical_apicid != -1U)
258 ver = apic_version[boot_cpu_physical_apicid]; 256 ver = apic_version[boot_cpu_physical_apicid];
259#endif
260 257
261 generic_processor_info(id, ver); 258 generic_processor_info(id, ver);
262} 259}
@@ -775,11 +772,9 @@ static void __init acpi_register_lapic_address(unsigned long address)
775 772
776 set_fixmap_nocache(FIX_APIC_BASE, address); 773 set_fixmap_nocache(FIX_APIC_BASE, address);
777 if (boot_cpu_physical_apicid == -1U) { 774 if (boot_cpu_physical_apicid == -1U) {
778 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 775 boot_cpu_physical_apicid = read_apic_id();
779#ifdef CONFIG_X86_32
780 apic_version[boot_cpu_physical_apicid] = 776 apic_version[boot_cpu_physical_apicid] =
781 GET_APIC_VERSION(apic_read(APIC_LVR)); 777 GET_APIC_VERSION(apic_read(APIC_LVR));
782#endif
783 } 778 }
784} 779}
785 780
@@ -1351,7 +1346,9 @@ static void __init acpi_process_madt(void)
1351 acpi_ioapic = 1; 1346 acpi_ioapic = 1;
1352 1347
1353 smp_found_config = 1; 1348 smp_found_config = 1;
1349#ifdef CONFIG_X86_32
1354 setup_apic_routing(); 1350 setup_apic_routing();
1351#endif
1355 } 1352 }
1356 } 1353 }
1357 if (error == -EINVAL) { 1354 if (error == -EINVAL) {
@@ -1421,8 +1418,16 @@ static int __init force_acpi_ht(const struct dmi_system_id *d)
1421 */ 1418 */
1422static int __init dmi_ignore_irq0_timer_override(const struct dmi_system_id *d) 1419static int __init dmi_ignore_irq0_timer_override(const struct dmi_system_id *d)
1423{ 1420{
1424 pr_notice("%s detected: Ignoring BIOS IRQ0 pin2 override\n", d->ident); 1421 /*
1425 acpi_skip_timer_override = 1; 1422 * The ati_ixp4x0_rev() early PCI quirk should have set
1423 * the acpi_skip_timer_override flag already:
1424 */
1425 if (!acpi_skip_timer_override) {
1426 WARN(1, KERN_ERR "ati_ixp4x0 quirk not complete.\n");
1427 pr_notice("%s detected: Ignoring BIOS IRQ0 pin2 override\n",
1428 d->ident);
1429 acpi_skip_timer_override = 1;
1430 }
1426 return 0; 1431 return 0;
1427} 1432}
1428 1433
@@ -1605,6 +1610,14 @@ static struct dmi_system_id __initdata acpi_dmi_table[] = {
1605 */ 1610 */
1606 { 1611 {
1607 .callback = dmi_ignore_irq0_timer_override, 1612 .callback = dmi_ignore_irq0_timer_override,
1613 .ident = "HP nx6115 laptop",
1614 .matches = {
1615 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1616 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq nx6115"),
1617 },
1618 },
1619 {
1620 .callback = dmi_ignore_irq0_timer_override,
1608 .ident = "HP NX6125 laptop", 1621 .ident = "HP NX6125 laptop",
1609 .matches = { 1622 .matches = {
1610 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"), 1623 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
@@ -1619,6 +1632,14 @@ static struct dmi_system_id __initdata acpi_dmi_table[] = {
1619 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq nx6325"), 1632 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq nx6325"),
1620 }, 1633 },
1621 }, 1634 },
1635 {
1636 .callback = dmi_ignore_irq0_timer_override,
1637 .ident = "HP 6715b laptop",
1638 .matches = {
1639 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1640 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq 6715b"),
1641 },
1642 },
1622 {} 1643 {}
1623}; 1644};
1624 1645
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index 2763cb37b553..a84ac7b570e6 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -145,35 +145,25 @@ static const unsigned char *const p6_nops[ASM_NOP_MAX+1] = {
145extern char __vsyscall_0; 145extern char __vsyscall_0;
146const unsigned char *const *find_nop_table(void) 146const unsigned char *const *find_nop_table(void)
147{ 147{
148 return boot_cpu_data.x86_vendor != X86_VENDOR_INTEL || 148 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
149 boot_cpu_data.x86 < 6 ? k8_nops : p6_nops; 149 boot_cpu_has(X86_FEATURE_NOPL))
150 return p6_nops;
151 else
152 return k8_nops;
150} 153}
151 154
152#else /* CONFIG_X86_64 */ 155#else /* CONFIG_X86_64 */
153 156
154static const struct nop {
155 int cpuid;
156 const unsigned char *const *noptable;
157} noptypes[] = {
158 { X86_FEATURE_K8, k8_nops },
159 { X86_FEATURE_K7, k7_nops },
160 { X86_FEATURE_P4, p6_nops },
161 { X86_FEATURE_P3, p6_nops },
162 { -1, NULL }
163};
164
165const unsigned char *const *find_nop_table(void) 157const unsigned char *const *find_nop_table(void)
166{ 158{
167 const unsigned char *const *noptable = intel_nops; 159 if (boot_cpu_has(X86_FEATURE_K8))
168 int i; 160 return k8_nops;
169 161 else if (boot_cpu_has(X86_FEATURE_K7))
170 for (i = 0; noptypes[i].cpuid >= 0; i++) { 162 return k7_nops;
171 if (boot_cpu_has(noptypes[i].cpuid)) { 163 else if (boot_cpu_has(X86_FEATURE_NOPL))
172 noptable = noptypes[i].noptable; 164 return p6_nops;
173 break; 165 else
174 } 166 return intel_nops;
175 }
176 return noptable;
177} 167}
178 168
179#endif /* CONFIG_X86_64 */ 169#endif /* CONFIG_X86_64 */
@@ -241,25 +231,25 @@ static void alternatives_smp_lock(u8 **start, u8 **end, u8 *text, u8 *text_end)
241 continue; 231 continue;
242 if (*ptr > text_end) 232 if (*ptr > text_end)
243 continue; 233 continue;
244 text_poke(*ptr, ((unsigned char []){0xf0}), 1); /* add lock prefix */ 234 /* turn DS segment override prefix into lock prefix */
235 text_poke(*ptr, ((unsigned char []){0xf0}), 1);
245 }; 236 };
246} 237}
247 238
248static void alternatives_smp_unlock(u8 **start, u8 **end, u8 *text, u8 *text_end) 239static void alternatives_smp_unlock(u8 **start, u8 **end, u8 *text, u8 *text_end)
249{ 240{
250 u8 **ptr; 241 u8 **ptr;
251 char insn[1];
252 242
253 if (noreplace_smp) 243 if (noreplace_smp)
254 return; 244 return;
255 245
256 add_nops(insn, 1);
257 for (ptr = start; ptr < end; ptr++) { 246 for (ptr = start; ptr < end; ptr++) {
258 if (*ptr < text) 247 if (*ptr < text)
259 continue; 248 continue;
260 if (*ptr > text_end) 249 if (*ptr > text_end)
261 continue; 250 continue;
262 text_poke(*ptr, insn, 1); 251 /* turn lock prefix into DS segment override prefix */
252 text_poke(*ptr, ((unsigned char []){0x3E}), 1);
263 }; 253 };
264} 254}
265 255
@@ -454,7 +444,7 @@ void __init alternative_instructions(void)
454 _text, _etext); 444 _text, _etext);
455 445
456 /* Only switch to UP mode if we don't immediately boot others */ 446 /* Only switch to UP mode if we don't immediately boot others */
457 if (num_possible_cpus() == 1 || setup_max_cpus <= 1) 447 if (num_present_cpus() == 1 || setup_max_cpus <= 1)
458 alternatives_smp_switch(0); 448 alternatives_smp_switch(0);
459 } 449 }
460#endif 450#endif
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 69b4d060b21c..a8fd9ebdc8e2 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -33,6 +33,10 @@
33 33
34static DEFINE_RWLOCK(amd_iommu_devtable_lock); 34static DEFINE_RWLOCK(amd_iommu_devtable_lock);
35 35
36/* A list of preallocated protection domains */
37static LIST_HEAD(iommu_pd_list);
38static DEFINE_SPINLOCK(iommu_pd_list_lock);
39
36/* 40/*
37 * general struct to manage commands send to an IOMMU 41 * general struct to manage commands send to an IOMMU
38 */ 42 */
@@ -51,6 +55,102 @@ static int iommu_has_npcache(struct amd_iommu *iommu)
51 55
52/**************************************************************************** 56/****************************************************************************
53 * 57 *
58 * Interrupt handling functions
59 *
60 ****************************************************************************/
61
62static void iommu_print_event(void *__evt)
63{
64 u32 *event = __evt;
65 int type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK;
66 int devid = (event[0] >> EVENT_DEVID_SHIFT) & EVENT_DEVID_MASK;
67 int domid = (event[1] >> EVENT_DOMID_SHIFT) & EVENT_DOMID_MASK;
68 int flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK;
69 u64 address = (u64)(((u64)event[3]) << 32) | event[2];
70
71 printk(KERN_ERR "AMD IOMMU: Event logged [");
72
73 switch (type) {
74 case EVENT_TYPE_ILL_DEV:
75 printk("ILLEGAL_DEV_TABLE_ENTRY device=%02x:%02x.%x "
76 "address=0x%016llx flags=0x%04x]\n",
77 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
78 address, flags);
79 break;
80 case EVENT_TYPE_IO_FAULT:
81 printk("IO_PAGE_FAULT device=%02x:%02x.%x "
82 "domain=0x%04x address=0x%016llx flags=0x%04x]\n",
83 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
84 domid, address, flags);
85 break;
86 case EVENT_TYPE_DEV_TAB_ERR:
87 printk("DEV_TAB_HARDWARE_ERROR device=%02x:%02x.%x "
88 "address=0x%016llx flags=0x%04x]\n",
89 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
90 address, flags);
91 break;
92 case EVENT_TYPE_PAGE_TAB_ERR:
93 printk("PAGE_TAB_HARDWARE_ERROR device=%02x:%02x.%x "
94 "domain=0x%04x address=0x%016llx flags=0x%04x]\n",
95 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
96 domid, address, flags);
97 break;
98 case EVENT_TYPE_ILL_CMD:
99 printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address);
100 break;
101 case EVENT_TYPE_CMD_HARD_ERR:
102 printk("COMMAND_HARDWARE_ERROR address=0x%016llx "
103 "flags=0x%04x]\n", address, flags);
104 break;
105 case EVENT_TYPE_IOTLB_INV_TO:
106 printk("IOTLB_INV_TIMEOUT device=%02x:%02x.%x "
107 "address=0x%016llx]\n",
108 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
109 address);
110 break;
111 case EVENT_TYPE_INV_DEV_REQ:
112 printk("INVALID_DEVICE_REQUEST device=%02x:%02x.%x "
113 "address=0x%016llx flags=0x%04x]\n",
114 PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
115 address, flags);
116 break;
117 default:
118 printk(KERN_ERR "UNKNOWN type=0x%02x]\n", type);
119 }
120}
121
122static void iommu_poll_events(struct amd_iommu *iommu)
123{
124 u32 head, tail;
125 unsigned long flags;
126
127 spin_lock_irqsave(&iommu->lock, flags);
128
129 head = readl(iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
130 tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
131
132 while (head != tail) {
133 iommu_print_event(iommu->evt_buf + head);
134 head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size;
135 }
136
137 writel(head, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
138
139 spin_unlock_irqrestore(&iommu->lock, flags);
140}
141
142irqreturn_t amd_iommu_int_handler(int irq, void *data)
143{
144 struct amd_iommu *iommu;
145
146 list_for_each_entry(iommu, &amd_iommu_list, list)
147 iommu_poll_events(iommu);
148
149 return IRQ_HANDLED;
150}
151
152/****************************************************************************
153 *
54 * IOMMU command queuing functions 154 * IOMMU command queuing functions
55 * 155 *
56 ****************************************************************************/ 156 ****************************************************************************/
@@ -101,10 +201,10 @@ static int iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd)
101 */ 201 */
102static int iommu_completion_wait(struct amd_iommu *iommu) 202static int iommu_completion_wait(struct amd_iommu *iommu)
103{ 203{
104 int ret, ready = 0; 204 int ret = 0, ready = 0;
105 unsigned status = 0; 205 unsigned status = 0;
106 struct iommu_cmd cmd; 206 struct iommu_cmd cmd;
107 unsigned long i = 0; 207 unsigned long flags, i = 0;
108 208
109 memset(&cmd, 0, sizeof(cmd)); 209 memset(&cmd, 0, sizeof(cmd));
110 cmd.data[0] = CMD_COMPL_WAIT_INT_MASK; 210 cmd.data[0] = CMD_COMPL_WAIT_INT_MASK;
@@ -112,10 +212,12 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
112 212
113 iommu->need_sync = 0; 213 iommu->need_sync = 0;
114 214
115 ret = iommu_queue_command(iommu, &cmd); 215 spin_lock_irqsave(&iommu->lock, flags);
216
217 ret = __iommu_queue_command(iommu, &cmd);
116 218
117 if (ret) 219 if (ret)
118 return ret; 220 goto out;
119 221
120 while (!ready && (i < EXIT_LOOP_COUNT)) { 222 while (!ready && (i < EXIT_LOOP_COUNT)) {
121 ++i; 223 ++i;
@@ -130,6 +232,8 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
130 232
131 if (unlikely((i == EXIT_LOOP_COUNT) && printk_ratelimit())) 233 if (unlikely((i == EXIT_LOOP_COUNT) && printk_ratelimit()))
132 printk(KERN_WARNING "AMD IOMMU: Completion wait loop failed\n"); 234 printk(KERN_WARNING "AMD IOMMU: Completion wait loop failed\n");
235out:
236 spin_unlock_irqrestore(&iommu->lock, flags);
133 237
134 return 0; 238 return 0;
135} 239}
@@ -140,6 +244,7 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
140static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid) 244static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
141{ 245{
142 struct iommu_cmd cmd; 246 struct iommu_cmd cmd;
247 int ret;
143 248
144 BUG_ON(iommu == NULL); 249 BUG_ON(iommu == NULL);
145 250
@@ -147,9 +252,11 @@ static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
147 CMD_SET_TYPE(&cmd, CMD_INV_DEV_ENTRY); 252 CMD_SET_TYPE(&cmd, CMD_INV_DEV_ENTRY);
148 cmd.data[0] = devid; 253 cmd.data[0] = devid;
149 254
255 ret = iommu_queue_command(iommu, &cmd);
256
150 iommu->need_sync = 1; 257 iommu->need_sync = 1;
151 258
152 return iommu_queue_command(iommu, &cmd); 259 return ret;
153} 260}
154 261
155/* 262/*
@@ -159,6 +266,7 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
159 u64 address, u16 domid, int pde, int s) 266 u64 address, u16 domid, int pde, int s)
160{ 267{
161 struct iommu_cmd cmd; 268 struct iommu_cmd cmd;
269 int ret;
162 270
163 memset(&cmd, 0, sizeof(cmd)); 271 memset(&cmd, 0, sizeof(cmd));
164 address &= PAGE_MASK; 272 address &= PAGE_MASK;
@@ -171,9 +279,11 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
171 if (pde) /* PDE bit - we wan't flush everything not only the PTEs */ 279 if (pde) /* PDE bit - we wan't flush everything not only the PTEs */
172 cmd.data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK; 280 cmd.data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK;
173 281
282 ret = iommu_queue_command(iommu, &cmd);
283
174 iommu->need_sync = 1; 284 iommu->need_sync = 1;
175 285
176 return iommu_queue_command(iommu, &cmd); 286 return ret;
177} 287}
178 288
179/* 289/*
@@ -185,7 +295,7 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
185 u64 address, size_t size) 295 u64 address, size_t size)
186{ 296{
187 int s = 0; 297 int s = 0;
188 unsigned pages = iommu_num_pages(address, size); 298 unsigned pages = iommu_num_pages(address, size, PAGE_SIZE);
189 299
190 address &= PAGE_MASK; 300 address &= PAGE_MASK;
191 301
@@ -203,6 +313,14 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
203 return 0; 313 return 0;
204} 314}
205 315
316/* Flush the whole IO/TLB for a given protection domain */
317static void iommu_flush_tlb(struct amd_iommu *iommu, u16 domid)
318{
319 u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
320
321 iommu_queue_inv_iommu_pages(iommu, address, domid, 0, 1);
322}
323
206/**************************************************************************** 324/****************************************************************************
207 * 325 *
208 * The functions below are used the create the page table mappings for 326 * The functions below are used the create the page table mappings for
@@ -362,11 +480,6 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
362 * efficient allocator. 480 * efficient allocator.
363 * 481 *
364 ****************************************************************************/ 482 ****************************************************************************/
365static unsigned long dma_mask_to_pages(unsigned long mask)
366{
367 return (mask >> PAGE_SHIFT) +
368 (PAGE_ALIGN(mask & ~PAGE_MASK) >> PAGE_SHIFT);
369}
370 483
371/* 484/*
372 * The address allocator core function. 485 * The address allocator core function.
@@ -375,25 +488,31 @@ static unsigned long dma_mask_to_pages(unsigned long mask)
375 */ 488 */
376static unsigned long dma_ops_alloc_addresses(struct device *dev, 489static unsigned long dma_ops_alloc_addresses(struct device *dev,
377 struct dma_ops_domain *dom, 490 struct dma_ops_domain *dom,
378 unsigned int pages) 491 unsigned int pages,
492 unsigned long align_mask,
493 u64 dma_mask)
379{ 494{
380 unsigned long limit = dma_mask_to_pages(*dev->dma_mask); 495 unsigned long limit;
381 unsigned long address; 496 unsigned long address;
382 unsigned long size = dom->aperture_size >> PAGE_SHIFT;
383 unsigned long boundary_size; 497 unsigned long boundary_size;
384 498
385 boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, 499 boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
386 PAGE_SIZE) >> PAGE_SHIFT; 500 PAGE_SIZE) >> PAGE_SHIFT;
387 limit = limit < size ? limit : size; 501 limit = iommu_device_max_index(dom->aperture_size >> PAGE_SHIFT, 0,
502 dma_mask >> PAGE_SHIFT);
388 503
389 if (dom->next_bit >= limit) 504 if (dom->next_bit >= limit) {
390 dom->next_bit = 0; 505 dom->next_bit = 0;
506 dom->need_flush = true;
507 }
391 508
392 address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages, 509 address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
393 0 , boundary_size, 0); 510 0 , boundary_size, align_mask);
394 if (address == -1) 511 if (address == -1) {
395 address = iommu_area_alloc(dom->bitmap, limit, 0, pages, 512 address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
396 0, boundary_size, 0); 513 0, boundary_size, align_mask);
514 dom->need_flush = true;
515 }
397 516
398 if (likely(address != -1)) { 517 if (likely(address != -1)) {
399 dom->next_bit = address + pages; 518 dom->next_bit = address + pages;
@@ -459,7 +578,7 @@ static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
459 if (start_page + pages > last_page) 578 if (start_page + pages > last_page)
460 pages = last_page - start_page; 579 pages = last_page - start_page;
461 580
462 set_bit_string(dom->bitmap, start_page, pages); 581 iommu_area_reserve(dom->bitmap, start_page, pages);
463} 582}
464 583
465static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom) 584static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom)
@@ -553,12 +672,16 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
553 dma_dom->bitmap[0] = 1; 672 dma_dom->bitmap[0] = 1;
554 dma_dom->next_bit = 0; 673 dma_dom->next_bit = 0;
555 674
675 dma_dom->need_flush = false;
676 dma_dom->target_dev = 0xffff;
677
556 /* Intialize the exclusion range if necessary */ 678 /* Intialize the exclusion range if necessary */
557 if (iommu->exclusion_start && 679 if (iommu->exclusion_start &&
558 iommu->exclusion_start < dma_dom->aperture_size) { 680 iommu->exclusion_start < dma_dom->aperture_size) {
559 unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT; 681 unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
560 int pages = iommu_num_pages(iommu->exclusion_start, 682 int pages = iommu_num_pages(iommu->exclusion_start,
561 iommu->exclusion_length); 683 iommu->exclusion_length,
684 PAGE_SIZE);
562 dma_ops_reserve_addresses(dma_dom, startpage, pages); 685 dma_ops_reserve_addresses(dma_dom, startpage, pages);
563 } 686 }
564 687
@@ -623,12 +746,13 @@ static void set_device_domain(struct amd_iommu *iommu,
623 746
624 u64 pte_root = virt_to_phys(domain->pt_root); 747 u64 pte_root = virt_to_phys(domain->pt_root);
625 748
626 pte_root |= (domain->mode & 0x07) << 9; 749 pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
627 pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | 2; 750 << DEV_ENTRY_MODE_SHIFT;
751 pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;
628 752
629 write_lock_irqsave(&amd_iommu_devtable_lock, flags); 753 write_lock_irqsave(&amd_iommu_devtable_lock, flags);
630 amd_iommu_dev_table[devid].data[0] = pte_root; 754 amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
631 amd_iommu_dev_table[devid].data[1] = pte_root >> 32; 755 amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root);
632 amd_iommu_dev_table[devid].data[2] = domain->id; 756 amd_iommu_dev_table[devid].data[2] = domain->id;
633 757
634 amd_iommu_pd_table[devid] = domain; 758 amd_iommu_pd_table[devid] = domain;
@@ -646,6 +770,45 @@ static void set_device_domain(struct amd_iommu *iommu,
646 *****************************************************************************/ 770 *****************************************************************************/
647 771
648/* 772/*
773 * This function checks if the driver got a valid device from the caller to
774 * avoid dereferencing invalid pointers.
775 */
776static bool check_device(struct device *dev)
777{
778 if (!dev || !dev->dma_mask)
779 return false;
780
781 return true;
782}
783
784/*
785 * In this function the list of preallocated protection domains is traversed to
786 * find the domain for a specific device
787 */
788static struct dma_ops_domain *find_protection_domain(u16 devid)
789{
790 struct dma_ops_domain *entry, *ret = NULL;
791 unsigned long flags;
792
793 if (list_empty(&iommu_pd_list))
794 return NULL;
795
796 spin_lock_irqsave(&iommu_pd_list_lock, flags);
797
798 list_for_each_entry(entry, &iommu_pd_list, list) {
799 if (entry->target_dev == devid) {
800 ret = entry;
801 list_del(&ret->list);
802 break;
803 }
804 }
805
806 spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
807
808 return ret;
809}
810
811/*
649 * In the dma_ops path we only have the struct device. This function 812 * In the dma_ops path we only have the struct device. This function
650 * finds the corresponding IOMMU, the protection domain and the 813 * finds the corresponding IOMMU, the protection domain and the
651 * requestor id for a given device. 814 * requestor id for a given device.
@@ -661,27 +824,30 @@ static int get_device_resources(struct device *dev,
661 struct pci_dev *pcidev; 824 struct pci_dev *pcidev;
662 u16 _bdf; 825 u16 _bdf;
663 826
664 BUG_ON(!dev || dev->bus != &pci_bus_type || !dev->dma_mask); 827 *iommu = NULL;
828 *domain = NULL;
829 *bdf = 0xffff;
830
831 if (dev->bus != &pci_bus_type)
832 return 0;
665 833
666 pcidev = to_pci_dev(dev); 834 pcidev = to_pci_dev(dev);
667 _bdf = calc_devid(pcidev->bus->number, pcidev->devfn); 835 _bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
668 836
669 /* device not translated by any IOMMU in the system? */ 837 /* device not translated by any IOMMU in the system? */
670 if (_bdf > amd_iommu_last_bdf) { 838 if (_bdf > amd_iommu_last_bdf)
671 *iommu = NULL;
672 *domain = NULL;
673 *bdf = 0xffff;
674 return 0; 839 return 0;
675 }
676 840
677 *bdf = amd_iommu_alias_table[_bdf]; 841 *bdf = amd_iommu_alias_table[_bdf];
678 842
679 *iommu = amd_iommu_rlookup_table[*bdf]; 843 *iommu = amd_iommu_rlookup_table[*bdf];
680 if (*iommu == NULL) 844 if (*iommu == NULL)
681 return 0; 845 return 0;
682 dma_dom = (*iommu)->default_dom;
683 *domain = domain_for_device(*bdf); 846 *domain = domain_for_device(*bdf);
684 if (*domain == NULL) { 847 if (*domain == NULL) {
848 dma_dom = find_protection_domain(*bdf);
849 if (!dma_dom)
850 dma_dom = (*iommu)->default_dom;
685 *domain = &dma_dom->domain; 851 *domain = &dma_dom->domain;
686 set_device_domain(*iommu, *domain, *bdf); 852 set_device_domain(*iommu, *domain, *bdf);
687 printk(KERN_INFO "AMD IOMMU: Using protection domain %d for " 853 printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
@@ -760,17 +926,24 @@ static dma_addr_t __map_single(struct device *dev,
760 struct dma_ops_domain *dma_dom, 926 struct dma_ops_domain *dma_dom,
761 phys_addr_t paddr, 927 phys_addr_t paddr,
762 size_t size, 928 size_t size,
763 int dir) 929 int dir,
930 bool align,
931 u64 dma_mask)
764{ 932{
765 dma_addr_t offset = paddr & ~PAGE_MASK; 933 dma_addr_t offset = paddr & ~PAGE_MASK;
766 dma_addr_t address, start; 934 dma_addr_t address, start;
767 unsigned int pages; 935 unsigned int pages;
936 unsigned long align_mask = 0;
768 int i; 937 int i;
769 938
770 pages = iommu_num_pages(paddr, size); 939 pages = iommu_num_pages(paddr, size, PAGE_SIZE);
771 paddr &= PAGE_MASK; 940 paddr &= PAGE_MASK;
772 941
773 address = dma_ops_alloc_addresses(dev, dma_dom, pages); 942 if (align)
943 align_mask = (1UL << get_order(size)) - 1;
944
945 address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask,
946 dma_mask);
774 if (unlikely(address == bad_dma_address)) 947 if (unlikely(address == bad_dma_address))
775 goto out; 948 goto out;
776 949
@@ -782,6 +955,12 @@ static dma_addr_t __map_single(struct device *dev,
782 } 955 }
783 address += offset; 956 address += offset;
784 957
958 if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) {
959 iommu_flush_tlb(iommu, dma_dom->domain.id);
960 dma_dom->need_flush = false;
961 } else if (unlikely(iommu_has_npcache(iommu)))
962 iommu_flush_pages(iommu, dma_dom->domain.id, address, size);
963
785out: 964out:
786 return address; 965 return address;
787} 966}
@@ -802,7 +981,7 @@ static void __unmap_single(struct amd_iommu *iommu,
802 if ((dma_addr == 0) || (dma_addr + size > dma_dom->aperture_size)) 981 if ((dma_addr == 0) || (dma_addr + size > dma_dom->aperture_size))
803 return; 982 return;
804 983
805 pages = iommu_num_pages(dma_addr, size); 984 pages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
806 dma_addr &= PAGE_MASK; 985 dma_addr &= PAGE_MASK;
807 start = dma_addr; 986 start = dma_addr;
808 987
@@ -812,6 +991,9 @@ static void __unmap_single(struct amd_iommu *iommu,
812 } 991 }
813 992
814 dma_ops_free_addresses(dma_dom, dma_addr, pages); 993 dma_ops_free_addresses(dma_dom, dma_addr, pages);
994
995 if (amd_iommu_unmap_flush)
996 iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
815} 997}
816 998
817/* 999/*
@@ -825,6 +1007,12 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
825 struct protection_domain *domain; 1007 struct protection_domain *domain;
826 u16 devid; 1008 u16 devid;
827 dma_addr_t addr; 1009 dma_addr_t addr;
1010 u64 dma_mask;
1011
1012 if (!check_device(dev))
1013 return bad_dma_address;
1014
1015 dma_mask = *dev->dma_mask;
828 1016
829 get_device_resources(dev, &iommu, &domain, &devid); 1017 get_device_resources(dev, &iommu, &domain, &devid);
830 1018
@@ -833,14 +1021,12 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
833 return (dma_addr_t)paddr; 1021 return (dma_addr_t)paddr;
834 1022
835 spin_lock_irqsave(&domain->lock, flags); 1023 spin_lock_irqsave(&domain->lock, flags);
836 addr = __map_single(dev, iommu, domain->priv, paddr, size, dir); 1024 addr = __map_single(dev, iommu, domain->priv, paddr, size, dir, false,
1025 dma_mask);
837 if (addr == bad_dma_address) 1026 if (addr == bad_dma_address)
838 goto out; 1027 goto out;
839 1028
840 if (iommu_has_npcache(iommu)) 1029 if (unlikely(iommu->need_sync))
841 iommu_flush_pages(iommu, domain->id, addr, size);
842
843 if (iommu->need_sync)
844 iommu_completion_wait(iommu); 1030 iommu_completion_wait(iommu);
845 1031
846out: 1032out:
@@ -860,7 +1046,8 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
860 struct protection_domain *domain; 1046 struct protection_domain *domain;
861 u16 devid; 1047 u16 devid;
862 1048
863 if (!get_device_resources(dev, &iommu, &domain, &devid)) 1049 if (!check_device(dev) ||
1050 !get_device_resources(dev, &iommu, &domain, &devid))
864 /* device not handled by any AMD IOMMU */ 1051 /* device not handled by any AMD IOMMU */
865 return; 1052 return;
866 1053
@@ -868,9 +1055,7 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
868 1055
869 __unmap_single(iommu, domain->priv, dma_addr, size, dir); 1056 __unmap_single(iommu, domain->priv, dma_addr, size, dir);
870 1057
871 iommu_flush_pages(iommu, domain->id, dma_addr, size); 1058 if (unlikely(iommu->need_sync))
872
873 if (iommu->need_sync)
874 iommu_completion_wait(iommu); 1059 iommu_completion_wait(iommu);
875 1060
876 spin_unlock_irqrestore(&domain->lock, flags); 1061 spin_unlock_irqrestore(&domain->lock, flags);
@@ -909,6 +1094,12 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
909 struct scatterlist *s; 1094 struct scatterlist *s;
910 phys_addr_t paddr; 1095 phys_addr_t paddr;
911 int mapped_elems = 0; 1096 int mapped_elems = 0;
1097 u64 dma_mask;
1098
1099 if (!check_device(dev))
1100 return 0;
1101
1102 dma_mask = *dev->dma_mask;
912 1103
913 get_device_resources(dev, &iommu, &domain, &devid); 1104 get_device_resources(dev, &iommu, &domain, &devid);
914 1105
@@ -921,19 +1112,17 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
921 paddr = sg_phys(s); 1112 paddr = sg_phys(s);
922 1113
923 s->dma_address = __map_single(dev, iommu, domain->priv, 1114 s->dma_address = __map_single(dev, iommu, domain->priv,
924 paddr, s->length, dir); 1115 paddr, s->length, dir, false,
1116 dma_mask);
925 1117
926 if (s->dma_address) { 1118 if (s->dma_address) {
927 s->dma_length = s->length; 1119 s->dma_length = s->length;
928 mapped_elems++; 1120 mapped_elems++;
929 } else 1121 } else
930 goto unmap; 1122 goto unmap;
931 if (iommu_has_npcache(iommu))
932 iommu_flush_pages(iommu, domain->id, s->dma_address,
933 s->dma_length);
934 } 1123 }
935 1124
936 if (iommu->need_sync) 1125 if (unlikely(iommu->need_sync))
937 iommu_completion_wait(iommu); 1126 iommu_completion_wait(iommu);
938 1127
939out: 1128out:
@@ -967,7 +1156,8 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
967 u16 devid; 1156 u16 devid;
968 int i; 1157 int i;
969 1158
970 if (!get_device_resources(dev, &iommu, &domain, &devid)) 1159 if (!check_device(dev) ||
1160 !get_device_resources(dev, &iommu, &domain, &devid))
971 return; 1161 return;
972 1162
973 spin_lock_irqsave(&domain->lock, flags); 1163 spin_lock_irqsave(&domain->lock, flags);
@@ -975,12 +1165,10 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
975 for_each_sg(sglist, s, nelems, i) { 1165 for_each_sg(sglist, s, nelems, i) {
976 __unmap_single(iommu, domain->priv, s->dma_address, 1166 __unmap_single(iommu, domain->priv, s->dma_address,
977 s->dma_length, dir); 1167 s->dma_length, dir);
978 iommu_flush_pages(iommu, domain->id, s->dma_address,
979 s->dma_length);
980 s->dma_address = s->dma_length = 0; 1168 s->dma_address = s->dma_length = 0;
981 } 1169 }
982 1170
983 if (iommu->need_sync) 1171 if (unlikely(iommu->need_sync))
984 iommu_completion_wait(iommu); 1172 iommu_completion_wait(iommu);
985 1173
986 spin_unlock_irqrestore(&domain->lock, flags); 1174 spin_unlock_irqrestore(&domain->lock, flags);
@@ -998,25 +1186,33 @@ static void *alloc_coherent(struct device *dev, size_t size,
998 struct protection_domain *domain; 1186 struct protection_domain *domain;
999 u16 devid; 1187 u16 devid;
1000 phys_addr_t paddr; 1188 phys_addr_t paddr;
1189 u64 dma_mask = dev->coherent_dma_mask;
1190
1191 if (!check_device(dev))
1192 return NULL;
1001 1193
1194 if (!get_device_resources(dev, &iommu, &domain, &devid))
1195 flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
1196
1197 flag |= __GFP_ZERO;
1002 virt_addr = (void *)__get_free_pages(flag, get_order(size)); 1198 virt_addr = (void *)__get_free_pages(flag, get_order(size));
1003 if (!virt_addr) 1199 if (!virt_addr)
1004 return 0; 1200 return 0;
1005 1201
1006 memset(virt_addr, 0, size);
1007 paddr = virt_to_phys(virt_addr); 1202 paddr = virt_to_phys(virt_addr);
1008 1203
1009 get_device_resources(dev, &iommu, &domain, &devid);
1010
1011 if (!iommu || !domain) { 1204 if (!iommu || !domain) {
1012 *dma_addr = (dma_addr_t)paddr; 1205 *dma_addr = (dma_addr_t)paddr;
1013 return virt_addr; 1206 return virt_addr;
1014 } 1207 }
1015 1208
1209 if (!dma_mask)
1210 dma_mask = *dev->dma_mask;
1211
1016 spin_lock_irqsave(&domain->lock, flags); 1212 spin_lock_irqsave(&domain->lock, flags);
1017 1213
1018 *dma_addr = __map_single(dev, iommu, domain->priv, paddr, 1214 *dma_addr = __map_single(dev, iommu, domain->priv, paddr,
1019 size, DMA_BIDIRECTIONAL); 1215 size, DMA_BIDIRECTIONAL, true, dma_mask);
1020 1216
1021 if (*dma_addr == bad_dma_address) { 1217 if (*dma_addr == bad_dma_address) {
1022 free_pages((unsigned long)virt_addr, get_order(size)); 1218 free_pages((unsigned long)virt_addr, get_order(size));
@@ -1024,10 +1220,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
1024 goto out; 1220 goto out;
1025 } 1221 }
1026 1222
1027 if (iommu_has_npcache(iommu)) 1223 if (unlikely(iommu->need_sync))
1028 iommu_flush_pages(iommu, domain->id, *dma_addr, size);
1029
1030 if (iommu->need_sync)
1031 iommu_completion_wait(iommu); 1224 iommu_completion_wait(iommu);
1032 1225
1033out: 1226out:
@@ -1038,8 +1231,6 @@ out:
1038 1231
1039/* 1232/*
1040 * The exported free_coherent function for dma_ops. 1233 * The exported free_coherent function for dma_ops.
1041 * FIXME: fix the generic x86 DMA layer so that it actually calls that
1042 * function.
1043 */ 1234 */
1044static void free_coherent(struct device *dev, size_t size, 1235static void free_coherent(struct device *dev, size_t size,
1045 void *virt_addr, dma_addr_t dma_addr) 1236 void *virt_addr, dma_addr_t dma_addr)
@@ -1049,6 +1240,9 @@ static void free_coherent(struct device *dev, size_t size,
1049 struct protection_domain *domain; 1240 struct protection_domain *domain;
1050 u16 devid; 1241 u16 devid;
1051 1242
1243 if (!check_device(dev))
1244 return;
1245
1052 get_device_resources(dev, &iommu, &domain, &devid); 1246 get_device_resources(dev, &iommu, &domain, &devid);
1053 1247
1054 if (!iommu || !domain) 1248 if (!iommu || !domain)
@@ -1057,9 +1251,8 @@ static void free_coherent(struct device *dev, size_t size,
1057 spin_lock_irqsave(&domain->lock, flags); 1251 spin_lock_irqsave(&domain->lock, flags);
1058 1252
1059 __unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL); 1253 __unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
1060 iommu_flush_pages(iommu, domain->id, dma_addr, size);
1061 1254
1062 if (iommu->need_sync) 1255 if (unlikely(iommu->need_sync))
1063 iommu_completion_wait(iommu); 1256 iommu_completion_wait(iommu);
1064 1257
1065 spin_unlock_irqrestore(&domain->lock, flags); 1258 spin_unlock_irqrestore(&domain->lock, flags);
@@ -1069,6 +1262,30 @@ free_mem:
1069} 1262}
1070 1263
1071/* 1264/*
1265 * This function is called by the DMA layer to find out if we can handle a
1266 * particular device. It is part of the dma_ops.
1267 */
1268static int amd_iommu_dma_supported(struct device *dev, u64 mask)
1269{
1270 u16 bdf;
1271 struct pci_dev *pcidev;
1272
1273 /* No device or no PCI device */
1274 if (!dev || dev->bus != &pci_bus_type)
1275 return 0;
1276
1277 pcidev = to_pci_dev(dev);
1278
1279 bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
1280
1281 /* Out of our scope? */
1282 if (bdf > amd_iommu_last_bdf)
1283 return 0;
1284
1285 return 1;
1286}
1287
1288/*
1072 * The function for pre-allocating protection domains. 1289 * The function for pre-allocating protection domains.
1073 * 1290 *
1074 * If the driver core informs the DMA layer if a driver grabs a device 1291 * If the driver core informs the DMA layer if a driver grabs a device
@@ -1097,10 +1314,9 @@ void prealloc_protection_domains(void)
1097 if (!dma_dom) 1314 if (!dma_dom)
1098 continue; 1315 continue;
1099 init_unity_mappings_for_device(dma_dom, devid); 1316 init_unity_mappings_for_device(dma_dom, devid);
1100 set_device_domain(iommu, &dma_dom->domain, devid); 1317 dma_dom->target_dev = devid;
1101 printk(KERN_INFO "AMD IOMMU: Allocated domain %d for device ", 1318
1102 dma_dom->domain.id); 1319 list_add_tail(&dma_dom->list, &iommu_pd_list);
1103 print_devid(devid, 1);
1104 } 1320 }
1105} 1321}
1106 1322
@@ -1111,6 +1327,7 @@ static struct dma_mapping_ops amd_iommu_dma_ops = {
1111 .unmap_single = unmap_single, 1327 .unmap_single = unmap_single,
1112 .map_sg = map_sg, 1328 .map_sg = map_sg,
1113 .unmap_sg = unmap_sg, 1329 .unmap_sg = unmap_sg,
1330 .dma_supported = amd_iommu_dma_supported,
1114}; 1331};
1115 1332
1116/* 1333/*
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index a69cc0f52042..4cd8083c58be 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -22,6 +22,8 @@
22#include <linux/gfp.h> 22#include <linux/gfp.h>
23#include <linux/list.h> 23#include <linux/list.h>
24#include <linux/sysdev.h> 24#include <linux/sysdev.h>
25#include <linux/interrupt.h>
26#include <linux/msi.h>
25#include <asm/pci-direct.h> 27#include <asm/pci-direct.h>
26#include <asm/amd_iommu_types.h> 28#include <asm/amd_iommu_types.h>
27#include <asm/amd_iommu.h> 29#include <asm/amd_iommu.h>
@@ -30,7 +32,6 @@
30/* 32/*
31 * definitions for the ACPI scanning code 33 * definitions for the ACPI scanning code
32 */ 34 */
33#define PCI_BUS(x) (((x) >> 8) & 0xff)
34#define IVRS_HEADER_LENGTH 48 35#define IVRS_HEADER_LENGTH 48
35 36
36#define ACPI_IVHD_TYPE 0x10 37#define ACPI_IVHD_TYPE 0x10
@@ -121,6 +122,7 @@ LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
121 we find in ACPI */ 122 we find in ACPI */
122unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */ 123unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
123int amd_iommu_isolate; /* if 1, device isolation is enabled */ 124int amd_iommu_isolate; /* if 1, device isolation is enabled */
125bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
124 126
125LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the 127LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
126 system */ 128 system */
@@ -234,7 +236,7 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
234{ 236{
235 u32 ctrl; 237 u32 ctrl;
236 238
237 ctrl = (u64)readl(iommu->mmio_base + MMIO_CONTROL_OFFSET); 239 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
238 ctrl &= ~(1 << bit); 240 ctrl &= ~(1 << bit);
239 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 241 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
240} 242}
@@ -242,13 +244,23 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
242/* Function to enable the hardware */ 244/* Function to enable the hardware */
243void __init iommu_enable(struct amd_iommu *iommu) 245void __init iommu_enable(struct amd_iommu *iommu)
244{ 246{
245 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at "); 247 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU "
246 print_devid(iommu->devid, 0); 248 "at %02x:%02x.%x cap 0x%hx\n",
247 printk(" cap 0x%hx\n", iommu->cap_ptr); 249 iommu->dev->bus->number,
250 PCI_SLOT(iommu->dev->devfn),
251 PCI_FUNC(iommu->dev->devfn),
252 iommu->cap_ptr);
248 253
249 iommu_feature_enable(iommu, CONTROL_IOMMU_EN); 254 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
250} 255}
251 256
257/* Function to enable IOMMU event logging and event interrupts */
258void __init iommu_enable_event_logging(struct amd_iommu *iommu)
259{
260 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
261 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
262}
263
252/* 264/*
253 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in 265 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
254 * the system has one. 266 * the system has one.
@@ -286,6 +298,14 @@ static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
286 ****************************************************************************/ 298 ****************************************************************************/
287 299
288/* 300/*
301 * This function calculates the length of a given IVHD entry
302 */
303static inline int ivhd_entry_length(u8 *ivhd)
304{
305 return 0x04 << (*ivhd >> 6);
306}
307
308/*
289 * This function reads the last device id the IOMMU has to handle from the PCI 309 * This function reads the last device id the IOMMU has to handle from the PCI
290 * capability header for this IOMMU 310 * capability header for this IOMMU
291 */ 311 */
@@ -329,7 +349,7 @@ static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
329 default: 349 default:
330 break; 350 break;
331 } 351 }
332 p += 0x04 << (*p >> 6); 352 p += ivhd_entry_length(p);
333 } 353 }
334 354
335 WARN_ON(p != end); 355 WARN_ON(p != end);
@@ -414,7 +434,32 @@ static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
414 434
415static void __init free_command_buffer(struct amd_iommu *iommu) 435static void __init free_command_buffer(struct amd_iommu *iommu)
416{ 436{
417 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE)); 437 free_pages((unsigned long)iommu->cmd_buf,
438 get_order(iommu->cmd_buf_size));
439}
440
441/* allocates the memory where the IOMMU will log its events to */
442static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
443{
444 u64 entry;
445 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
446 get_order(EVT_BUFFER_SIZE));
447
448 if (iommu->evt_buf == NULL)
449 return NULL;
450
451 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
452 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
453 &entry, sizeof(entry));
454
455 iommu->evt_buf_size = EVT_BUFFER_SIZE;
456
457 return iommu->evt_buf;
458}
459
460static void __init free_event_buffer(struct amd_iommu *iommu)
461{
462 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
418} 463}
419 464
420/* sets a specific bit in the device table entry. */ 465/* sets a specific bit in the device table entry. */
@@ -487,19 +532,21 @@ static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
487 */ 532 */
488static void __init init_iommu_from_pci(struct amd_iommu *iommu) 533static void __init init_iommu_from_pci(struct amd_iommu *iommu)
489{ 534{
490 int bus = PCI_BUS(iommu->devid);
491 int dev = PCI_SLOT(iommu->devid);
492 int fn = PCI_FUNC(iommu->devid);
493 int cap_ptr = iommu->cap_ptr; 535 int cap_ptr = iommu->cap_ptr;
494 u32 range; 536 u32 range, misc;
495 537
496 iommu->cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_CAP_HDR_OFFSET); 538 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
539 &iommu->cap);
540 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
541 &range);
542 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
543 &misc);
497 544
498 range = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
499 iommu->first_device = calc_devid(MMIO_GET_BUS(range), 545 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
500 MMIO_GET_FD(range)); 546 MMIO_GET_FD(range));
501 iommu->last_device = calc_devid(MMIO_GET_BUS(range), 547 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
502 MMIO_GET_LD(range)); 548 MMIO_GET_LD(range));
549 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
503} 550}
504 551
505/* 552/*
@@ -604,7 +651,7 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
604 break; 651 break;
605 } 652 }
606 653
607 p += 0x04 << (e->type >> 6); 654 p += ivhd_entry_length(p);
608 } 655 }
609} 656}
610 657
@@ -622,6 +669,7 @@ static int __init init_iommu_devices(struct amd_iommu *iommu)
622static void __init free_iommu_one(struct amd_iommu *iommu) 669static void __init free_iommu_one(struct amd_iommu *iommu)
623{ 670{
624 free_command_buffer(iommu); 671 free_command_buffer(iommu);
672 free_event_buffer(iommu);
625 iommu_unmap_mmio_space(iommu); 673 iommu_unmap_mmio_space(iommu);
626} 674}
627 675
@@ -649,8 +697,12 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
649 /* 697 /*
650 * Copy data from ACPI table entry to the iommu struct 698 * Copy data from ACPI table entry to the iommu struct
651 */ 699 */
652 iommu->devid = h->devid; 700 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
701 if (!iommu->dev)
702 return 1;
703
653 iommu->cap_ptr = h->cap_ptr; 704 iommu->cap_ptr = h->cap_ptr;
705 iommu->pci_seg = h->pci_seg;
654 iommu->mmio_phys = h->mmio_phys; 706 iommu->mmio_phys = h->mmio_phys;
655 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys); 707 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
656 if (!iommu->mmio_base) 708 if (!iommu->mmio_base)
@@ -661,11 +713,17 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
661 if (!iommu->cmd_buf) 713 if (!iommu->cmd_buf)
662 return -ENOMEM; 714 return -ENOMEM;
663 715
716 iommu->evt_buf = alloc_event_buffer(iommu);
717 if (!iommu->evt_buf)
718 return -ENOMEM;
719
720 iommu->int_enabled = false;
721
664 init_iommu_from_pci(iommu); 722 init_iommu_from_pci(iommu);
665 init_iommu_from_acpi(iommu, h); 723 init_iommu_from_acpi(iommu, h);
666 init_iommu_devices(iommu); 724 init_iommu_devices(iommu);
667 725
668 return 0; 726 return pci_enable_device(iommu->dev);
669} 727}
670 728
671/* 729/*
@@ -706,6 +764,95 @@ static int __init init_iommu_all(struct acpi_table_header *table)
706 764
707/**************************************************************************** 765/****************************************************************************
708 * 766 *
767 * The following functions initialize the MSI interrupts for all IOMMUs
768 * in the system. Its a bit challenging because there could be multiple
769 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
770 * pci_dev.
771 *
772 ****************************************************************************/
773
774static int __init iommu_setup_msix(struct amd_iommu *iommu)
775{
776 struct amd_iommu *curr;
777 struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
778 int nvec = 0, i;
779
780 list_for_each_entry(curr, &amd_iommu_list, list) {
781 if (curr->dev == iommu->dev) {
782 entries[nvec].entry = curr->evt_msi_num;
783 entries[nvec].vector = 0;
784 curr->int_enabled = true;
785 nvec++;
786 }
787 }
788
789 if (pci_enable_msix(iommu->dev, entries, nvec)) {
790 pci_disable_msix(iommu->dev);
791 return 1;
792 }
793
794 for (i = 0; i < nvec; ++i) {
795 int r = request_irq(entries->vector, amd_iommu_int_handler,
796 IRQF_SAMPLE_RANDOM,
797 "AMD IOMMU",
798 NULL);
799 if (r)
800 goto out_free;
801 }
802
803 return 0;
804
805out_free:
806 for (i -= 1; i >= 0; --i)
807 free_irq(entries->vector, NULL);
808
809 pci_disable_msix(iommu->dev);
810
811 return 1;
812}
813
814static int __init iommu_setup_msi(struct amd_iommu *iommu)
815{
816 int r;
817 struct amd_iommu *curr;
818
819 list_for_each_entry(curr, &amd_iommu_list, list) {
820 if (curr->dev == iommu->dev)
821 curr->int_enabled = true;
822 }
823
824
825 if (pci_enable_msi(iommu->dev))
826 return 1;
827
828 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
829 IRQF_SAMPLE_RANDOM,
830 "AMD IOMMU",
831 NULL);
832
833 if (r) {
834 pci_disable_msi(iommu->dev);
835 return 1;
836 }
837
838 return 0;
839}
840
841static int __init iommu_init_msi(struct amd_iommu *iommu)
842{
843 if (iommu->int_enabled)
844 return 0;
845
846 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
847 return iommu_setup_msix(iommu);
848 else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
849 return iommu_setup_msi(iommu);
850
851 return 1;
852}
853
854/****************************************************************************
855 *
709 * The next functions belong to the third pass of parsing the ACPI 856 * The next functions belong to the third pass of parsing the ACPI
710 * table. In this last pass the memory mapping requirements are 857 * table. In this last pass the memory mapping requirements are
711 * gathered (like exclusion and unity mapping reanges). 858 * gathered (like exclusion and unity mapping reanges).
@@ -811,7 +958,6 @@ static void init_device_table(void)
811 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) { 958 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
812 set_dev_entry_bit(devid, DEV_ENTRY_VALID); 959 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
813 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION); 960 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
814 set_dev_entry_bit(devid, DEV_ENTRY_NO_PAGE_FAULT);
815 } 961 }
816} 962}
817 963
@@ -825,6 +971,8 @@ static void __init enable_iommus(void)
825 971
826 list_for_each_entry(iommu, &amd_iommu_list, list) { 972 list_for_each_entry(iommu, &amd_iommu_list, list) {
827 iommu_set_exclusion_range(iommu); 973 iommu_set_exclusion_range(iommu);
974 iommu_init_msi(iommu);
975 iommu_enable_event_logging(iommu);
828 iommu_enable(iommu); 976 iommu_enable(iommu);
829 } 977 }
830} 978}
@@ -995,11 +1143,17 @@ int __init amd_iommu_init(void)
995 else 1143 else
996 printk("disabled\n"); 1144 printk("disabled\n");
997 1145
1146 if (amd_iommu_unmap_flush)
1147 printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
1148 else
1149 printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
1150
998out: 1151out:
999 return ret; 1152 return ret;
1000 1153
1001free: 1154free:
1002 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 1); 1155 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1156 get_order(MAX_DOMAIN_ID/8));
1003 1157
1004 free_pages((unsigned long)amd_iommu_pd_table, 1158 free_pages((unsigned long)amd_iommu_pd_table,
1005 get_order(rlookup_table_size)); 1159 get_order(rlookup_table_size));
@@ -1057,8 +1211,10 @@ void __init amd_iommu_detect(void)
1057static int __init parse_amd_iommu_options(char *str) 1211static int __init parse_amd_iommu_options(char *str)
1058{ 1212{
1059 for (; *str; ++str) { 1213 for (; *str; ++str) {
1060 if (strcmp(str, "isolate") == 0) 1214 if (strncmp(str, "isolate", 7) == 0)
1061 amd_iommu_isolate = 1; 1215 amd_iommu_isolate = 1;
1216 if (strncmp(str, "fullflush", 11) == 0)
1217 amd_iommu_unmap_flush = true;
1062 } 1218 }
1063 1219
1064 return 1; 1220 return 1;
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 44e21826db11..9a32b37ee2ee 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -455,11 +455,11 @@ out:
455 force_iommu || 455 force_iommu ||
456 valid_agp || 456 valid_agp ||
457 fallback_aper_force) { 457 fallback_aper_force) {
458 printk(KERN_ERR 458 printk(KERN_INFO
459 "Your BIOS doesn't leave a aperture memory hole\n"); 459 "Your BIOS doesn't leave a aperture memory hole\n");
460 printk(KERN_ERR 460 printk(KERN_INFO
461 "Please enable the IOMMU option in the BIOS setup\n"); 461 "Please enable the IOMMU option in the BIOS setup\n");
462 printk(KERN_ERR 462 printk(KERN_INFO
463 "This costs you %d MB of RAM\n", 463 "This costs you %d MB of RAM\n",
464 32 << fallback_aper_order); 464 32 << fallback_aper_order);
465 465
diff --git a/arch/x86/kernel/apic_32.c b/arch/x86/kernel/apic_32.c
index f88bd0d982b0..21c831d96af3 100644
--- a/arch/x86/kernel/apic_32.c
+++ b/arch/x86/kernel/apic_32.c
@@ -60,10 +60,8 @@ unsigned long mp_lapic_addr;
60static int force_enable_local_apic; 60static int force_enable_local_apic;
61int disable_apic; 61int disable_apic;
62 62
63/* Local APIC timer verification ok */
64static int local_apic_timer_verify_ok;
65/* Disable local APIC timer from the kernel commandline or via dmi quirk */ 63/* Disable local APIC timer from the kernel commandline or via dmi quirk */
66static int local_apic_timer_disabled; 64static int disable_apic_timer __cpuinitdata;
67/* Local APIC timer works in C2 */ 65/* Local APIC timer works in C2 */
68int local_apic_timer_c2_ok; 66int local_apic_timer_c2_ok;
69EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok); 67EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
@@ -130,7 +128,11 @@ static inline int lapic_get_version(void)
130 */ 128 */
131static inline int lapic_is_integrated(void) 129static inline int lapic_is_integrated(void)
132{ 130{
131#ifdef CONFIG_X86_64
132 return 1;
133#else
133 return APIC_INTEGRATED(lapic_get_version()); 134 return APIC_INTEGRATED(lapic_get_version());
135#endif
134} 136}
135 137
136/* 138/*
@@ -145,13 +147,18 @@ static int modern_apic(void)
145 return lapic_get_version() >= 0x14; 147 return lapic_get_version() >= 0x14;
146} 148}
147 149
148void apic_wait_icr_idle(void) 150/*
151 * Paravirt kernels also might be using these below ops. So we still
152 * use generic apic_read()/apic_write(), which might be pointing to different
153 * ops in PARAVIRT case.
154 */
155void xapic_wait_icr_idle(void)
149{ 156{
150 while (apic_read(APIC_ICR) & APIC_ICR_BUSY) 157 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
151 cpu_relax(); 158 cpu_relax();
152} 159}
153 160
154u32 safe_apic_wait_icr_idle(void) 161u32 safe_xapic_wait_icr_idle(void)
155{ 162{
156 u32 send_status; 163 u32 send_status;
157 int timeout; 164 int timeout;
@@ -167,16 +174,48 @@ u32 safe_apic_wait_icr_idle(void)
167 return send_status; 174 return send_status;
168} 175}
169 176
177void xapic_icr_write(u32 low, u32 id)
178{
179 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
180 apic_write(APIC_ICR, low);
181}
182
183u64 xapic_icr_read(void)
184{
185 u32 icr1, icr2;
186
187 icr2 = apic_read(APIC_ICR2);
188 icr1 = apic_read(APIC_ICR);
189
190 return icr1 | ((u64)icr2 << 32);
191}
192
193static struct apic_ops xapic_ops = {
194 .read = native_apic_mem_read,
195 .write = native_apic_mem_write,
196 .icr_read = xapic_icr_read,
197 .icr_write = xapic_icr_write,
198 .wait_icr_idle = xapic_wait_icr_idle,
199 .safe_wait_icr_idle = safe_xapic_wait_icr_idle,
200};
201
202struct apic_ops __read_mostly *apic_ops = &xapic_ops;
203EXPORT_SYMBOL_GPL(apic_ops);
204
170/** 205/**
171 * enable_NMI_through_LVT0 - enable NMI through local vector table 0 206 * enable_NMI_through_LVT0 - enable NMI through local vector table 0
172 */ 207 */
173void __cpuinit enable_NMI_through_LVT0(void) 208void __cpuinit enable_NMI_through_LVT0(void)
174{ 209{
175 unsigned int v = APIC_DM_NMI; 210 unsigned int v;
176 211
177 /* Level triggered for 82489DX */ 212 /* unmask and set to NMI */
213 v = APIC_DM_NMI;
214
215 /* Level triggered for 82489DX (32bit mode) */
178 if (!lapic_is_integrated()) 216 if (!lapic_is_integrated())
179 v |= APIC_LVT_LEVEL_TRIGGER; 217 v |= APIC_LVT_LEVEL_TRIGGER;
218
180 apic_write(APIC_LVT0, v); 219 apic_write(APIC_LVT0, v);
181} 220}
182 221
@@ -193,9 +232,13 @@ int get_physical_broadcast(void)
193 */ 232 */
194int lapic_get_maxlvt(void) 233int lapic_get_maxlvt(void)
195{ 234{
196 unsigned int v = apic_read(APIC_LVR); 235 unsigned int v;
197 236
198 /* 82489DXs do not report # of LVT entries. */ 237 v = apic_read(APIC_LVR);
238 /*
239 * - we always have APIC integrated on 64bit mode
240 * - 82489DXs do not report # of LVT entries
241 */
199 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2; 242 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
200} 243}
201 244
@@ -203,8 +246,12 @@ int lapic_get_maxlvt(void)
203 * Local APIC timer 246 * Local APIC timer
204 */ 247 */
205 248
206/* Clock divisor is set to 16 */ 249/* Clock divisor */
250#ifdef CONFG_X86_64
251#define APIC_DIVISOR 1
252#else
207#define APIC_DIVISOR 16 253#define APIC_DIVISOR 16
254#endif
208 255
209/* 256/*
210 * This function sets up the local APIC timer, with a timeout of 257 * This function sets up the local APIC timer, with a timeout of
@@ -212,6 +259,9 @@ int lapic_get_maxlvt(void)
212 * this function twice on the boot CPU, once with a bogus timeout 259 * this function twice on the boot CPU, once with a bogus timeout
213 * value, second time for real. The other (noncalibrating) CPUs 260 * value, second time for real. The other (noncalibrating) CPUs
214 * call this function only once, with the real, calibrated value. 261 * call this function only once, with the real, calibrated value.
262 *
263 * We do reads before writes even if unnecessary, to get around the
264 * P5 APIC double write bug.
215 */ 265 */
216static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen) 266static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
217{ 267{
@@ -233,14 +283,48 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
233 */ 283 */
234 tmp_value = apic_read(APIC_TDCR); 284 tmp_value = apic_read(APIC_TDCR);
235 apic_write(APIC_TDCR, 285 apic_write(APIC_TDCR,
236 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) | 286 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
237 APIC_TDR_DIV_16); 287 APIC_TDR_DIV_16);
238 288
239 if (!oneshot) 289 if (!oneshot)
240 apic_write(APIC_TMICT, clocks / APIC_DIVISOR); 290 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
241} 291}
242 292
243/* 293/*
294 * Setup extended LVT, AMD specific (K8, family 10h)
295 *
296 * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and
297 * MCE interrupts are supported. Thus MCE offset must be set to 0.
298 *
299 * If mask=1, the LVT entry does not generate interrupts while mask=0
300 * enables the vector. See also the BKDGs.
301 */
302
303#define APIC_EILVT_LVTOFF_MCE 0
304#define APIC_EILVT_LVTOFF_IBS 1
305
306static void setup_APIC_eilvt(u8 lvt_off, u8 vector, u8 msg_type, u8 mask)
307{
308 unsigned long reg = (lvt_off << 4) + APIC_EILVT0;
309 unsigned int v = (mask << 16) | (msg_type << 8) | vector;
310
311 apic_write(reg, v);
312}
313
314u8 setup_APIC_eilvt_mce(u8 vector, u8 msg_type, u8 mask)
315{
316 setup_APIC_eilvt(APIC_EILVT_LVTOFF_MCE, vector, msg_type, mask);
317 return APIC_EILVT_LVTOFF_MCE;
318}
319
320u8 setup_APIC_eilvt_ibs(u8 vector, u8 msg_type, u8 mask)
321{
322 setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask);
323 return APIC_EILVT_LVTOFF_IBS;
324}
325EXPORT_SYMBOL_GPL(setup_APIC_eilvt_ibs);
326
327/*
244 * Program the next event, relative to now 328 * Program the next event, relative to now
245 */ 329 */
246static int lapic_next_event(unsigned long delta, 330static int lapic_next_event(unsigned long delta,
@@ -259,8 +343,8 @@ static void lapic_timer_setup(enum clock_event_mode mode,
259 unsigned long flags; 343 unsigned long flags;
260 unsigned int v; 344 unsigned int v;
261 345
262 /* Lapic used for broadcast ? */ 346 /* Lapic used as dummy for broadcast ? */
263 if (!local_apic_timer_verify_ok) 347 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
264 return; 348 return;
265 349
266 local_irq_save(flags); 350 local_irq_save(flags);
@@ -473,7 +557,7 @@ static int __init calibrate_APIC_clock(void)
473 return -1; 557 return -1;
474 } 558 }
475 559
476 local_apic_timer_verify_ok = 1; 560 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
477 561
478 /* We trust the pm timer based calibration */ 562 /* We trust the pm timer based calibration */
479 if (!pm_referenced) { 563 if (!pm_referenced) {
@@ -507,11 +591,11 @@ static int __init calibrate_APIC_clock(void)
507 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2) 591 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
508 apic_printk(APIC_VERBOSE, "... jiffies result ok\n"); 592 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
509 else 593 else
510 local_apic_timer_verify_ok = 0; 594 levt->features |= CLOCK_EVT_FEAT_DUMMY;
511 } else 595 } else
512 local_irq_enable(); 596 local_irq_enable();
513 597
514 if (!local_apic_timer_verify_ok) { 598 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
515 printk(KERN_WARNING 599 printk(KERN_WARNING
516 "APIC timer disabled due to verification failure.\n"); 600 "APIC timer disabled due to verification failure.\n");
517 return -1; 601 return -1;
@@ -533,7 +617,8 @@ void __init setup_boot_APIC_clock(void)
533 * timer as a dummy clock event source on SMP systems, so the 617 * timer as a dummy clock event source on SMP systems, so the
534 * broadcast mechanism is used. On UP systems simply ignore it. 618 * broadcast mechanism is used. On UP systems simply ignore it.
535 */ 619 */
536 if (local_apic_timer_disabled) { 620 if (disable_apic_timer) {
621 printk(KERN_INFO "Disabling APIC timer\n");
537 /* No broadcast on UP ! */ 622 /* No broadcast on UP ! */
538 if (num_possible_cpus() > 1) { 623 if (num_possible_cpus() > 1) {
539 lapic_clockevent.mult = 1; 624 lapic_clockevent.mult = 1;
@@ -602,7 +687,11 @@ static void local_apic_timer_interrupt(void)
602 /* 687 /*
603 * the NMI deadlock-detector uses this. 688 * the NMI deadlock-detector uses this.
604 */ 689 */
690#ifdef CONFIG_X86_64
691 add_pda(apic_timer_irqs, 1);
692#else
605 per_cpu(irq_stat, cpu).apic_timer_irqs++; 693 per_cpu(irq_stat, cpu).apic_timer_irqs++;
694#endif
606 695
607 evt->event_handler(evt); 696 evt->event_handler(evt);
608} 697}
@@ -642,35 +731,6 @@ int setup_profiling_timer(unsigned int multiplier)
642} 731}
643 732
644/* 733/*
645 * Setup extended LVT, AMD specific (K8, family 10h)
646 *
647 * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and
648 * MCE interrupts are supported. Thus MCE offset must be set to 0.
649 */
650
651#define APIC_EILVT_LVTOFF_MCE 0
652#define APIC_EILVT_LVTOFF_IBS 1
653
654static void setup_APIC_eilvt(u8 lvt_off, u8 vector, u8 msg_type, u8 mask)
655{
656 unsigned long reg = (lvt_off << 4) + APIC_EILVT0;
657 unsigned int v = (mask << 16) | (msg_type << 8) | vector;
658 apic_write(reg, v);
659}
660
661u8 setup_APIC_eilvt_mce(u8 vector, u8 msg_type, u8 mask)
662{
663 setup_APIC_eilvt(APIC_EILVT_LVTOFF_MCE, vector, msg_type, mask);
664 return APIC_EILVT_LVTOFF_MCE;
665}
666
667u8 setup_APIC_eilvt_ibs(u8 vector, u8 msg_type, u8 mask)
668{
669 setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask);
670 return APIC_EILVT_LVTOFF_IBS;
671}
672
673/*
674 * Local APIC start and shutdown 734 * Local APIC start and shutdown
675 */ 735 */
676 736
@@ -715,7 +775,7 @@ void clear_local_APIC(void)
715 } 775 }
716 776
717 /* lets not touch this if we didn't frob it */ 777 /* lets not touch this if we didn't frob it */
718#ifdef CONFIG_X86_MCE_P4THERMAL 778#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(X86_MCE_INTEL)
719 if (maxlvt >= 5) { 779 if (maxlvt >= 5) {
720 v = apic_read(APIC_LVTTHMR); 780 v = apic_read(APIC_LVTTHMR);
721 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED); 781 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
@@ -732,10 +792,6 @@ void clear_local_APIC(void)
732 if (maxlvt >= 4) 792 if (maxlvt >= 4)
733 apic_write(APIC_LVTPC, APIC_LVT_MASKED); 793 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
734 794
735#ifdef CONFIG_X86_MCE_P4THERMAL
736 if (maxlvt >= 5)
737 apic_write(APIC_LVTTHMR, APIC_LVT_MASKED);
738#endif
739 /* Integrated APIC (!82489DX) ? */ 795 /* Integrated APIC (!82489DX) ? */
740 if (lapic_is_integrated()) { 796 if (lapic_is_integrated()) {
741 if (maxlvt > 3) 797 if (maxlvt > 3)
@@ -750,7 +806,7 @@ void clear_local_APIC(void)
750 */ 806 */
751void disable_local_APIC(void) 807void disable_local_APIC(void)
752{ 808{
753 unsigned long value; 809 unsigned int value;
754 810
755 clear_local_APIC(); 811 clear_local_APIC();
756 812
@@ -762,6 +818,7 @@ void disable_local_APIC(void)
762 value &= ~APIC_SPIV_APIC_ENABLED; 818 value &= ~APIC_SPIV_APIC_ENABLED;
763 apic_write(APIC_SPIV, value); 819 apic_write(APIC_SPIV, value);
764 820
821#ifdef CONFIG_X86_32
765 /* 822 /*
766 * When LAPIC was disabled by the BIOS and enabled by the kernel, 823 * When LAPIC was disabled by the BIOS and enabled by the kernel,
767 * restore the disabled state. 824 * restore the disabled state.
@@ -773,6 +830,7 @@ void disable_local_APIC(void)
773 l &= ~MSR_IA32_APICBASE_ENABLE; 830 l &= ~MSR_IA32_APICBASE_ENABLE;
774 wrmsr(MSR_IA32_APICBASE, l, h); 831 wrmsr(MSR_IA32_APICBASE, l, h);
775 } 832 }
833#endif
776} 834}
777 835
778/* 836/*
@@ -789,11 +847,15 @@ void lapic_shutdown(void)
789 return; 847 return;
790 848
791 local_irq_save(flags); 849 local_irq_save(flags);
792 clear_local_APIC();
793 850
794 if (enabled_via_apicbase) 851#ifdef CONFIG_X86_32
852 if (!enabled_via_apicbase)
853 clear_local_APIC();
854 else
855#endif
795 disable_local_APIC(); 856 disable_local_APIC();
796 857
858
797 local_irq_restore(flags); 859 local_irq_restore(flags);
798} 860}
799 861
@@ -838,6 +900,12 @@ int __init verify_local_APIC(void)
838 */ 900 */
839 reg0 = apic_read(APIC_ID); 901 reg0 = apic_read(APIC_ID);
840 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0); 902 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
903 apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
904 reg1 = apic_read(APIC_ID);
905 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
906 apic_write(APIC_ID, reg0);
907 if (reg1 != (reg0 ^ APIC_ID_MASK))
908 return 0;
841 909
842 /* 910 /*
843 * The next two are just to see if we have sane values. 911 * The next two are just to see if we have sane values.
@@ -863,14 +931,15 @@ void __init sync_Arb_IDs(void)
863 */ 931 */
864 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD) 932 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
865 return; 933 return;
934
866 /* 935 /*
867 * Wait for idle. 936 * Wait for idle.
868 */ 937 */
869 apic_wait_icr_idle(); 938 apic_wait_icr_idle();
870 939
871 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n"); 940 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
872 apic_write(APIC_ICR, 941 apic_write(APIC_ICR, APIC_DEST_ALLINC |
873 APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT); 942 APIC_INT_LEVELTRIG | APIC_DM_INIT);
874} 943}
875 944
876/* 945/*
@@ -878,7 +947,7 @@ void __init sync_Arb_IDs(void)
878 */ 947 */
879void __init init_bsp_APIC(void) 948void __init init_bsp_APIC(void)
880{ 949{
881 unsigned long value; 950 unsigned int value;
882 951
883 /* 952 /*
884 * Don't do the setup now if we have a SMP BIOS as the 953 * Don't do the setup now if we have a SMP BIOS as the
@@ -899,11 +968,13 @@ void __init init_bsp_APIC(void)
899 value &= ~APIC_VECTOR_MASK; 968 value &= ~APIC_VECTOR_MASK;
900 value |= APIC_SPIV_APIC_ENABLED; 969 value |= APIC_SPIV_APIC_ENABLED;
901 970
971#ifdef CONFIG_X86_32
902 /* This bit is reserved on P4/Xeon and should be cleared */ 972 /* This bit is reserved on P4/Xeon and should be cleared */
903 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && 973 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
904 (boot_cpu_data.x86 == 15)) 974 (boot_cpu_data.x86 == 15))
905 value &= ~APIC_SPIV_FOCUS_DISABLED; 975 value &= ~APIC_SPIV_FOCUS_DISABLED;
906 else 976 else
977#endif
907 value |= APIC_SPIV_FOCUS_DISABLED; 978 value |= APIC_SPIV_FOCUS_DISABLED;
908 value |= SPURIOUS_APIC_VECTOR; 979 value |= SPURIOUS_APIC_VECTOR;
909 apic_write(APIC_SPIV, value); 980 apic_write(APIC_SPIV, value);
@@ -922,6 +993,16 @@ static void __cpuinit lapic_setup_esr(void)
922{ 993{
923 unsigned long oldvalue, value, maxlvt; 994 unsigned long oldvalue, value, maxlvt;
924 if (lapic_is_integrated() && !esr_disable) { 995 if (lapic_is_integrated() && !esr_disable) {
996 if (esr_disable) {
997 /*
998 * Something untraceable is creating bad interrupts on
999 * secondary quads ... for the moment, just leave the
1000 * ESR disabled - we can't do anything useful with the
1001 * errors anyway - mbligh
1002 */
1003 printk(KERN_INFO "Leaving ESR disabled.\n");
1004 return;
1005 }
925 /* !82489DX */ 1006 /* !82489DX */
926 maxlvt = lapic_get_maxlvt(); 1007 maxlvt = lapic_get_maxlvt();
927 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ 1008 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
@@ -942,16 +1023,7 @@ static void __cpuinit lapic_setup_esr(void)
942 "vector: 0x%08lx after: 0x%08lx\n", 1023 "vector: 0x%08lx after: 0x%08lx\n",
943 oldvalue, value); 1024 oldvalue, value);
944 } else { 1025 } else {
945 if (esr_disable) 1026 printk(KERN_INFO "No ESR for 82489DX.\n");
946 /*
947 * Something untraceable is creating bad interrupts on
948 * secondary quads ... for the moment, just leave the
949 * ESR disabled - we can't do anything useful with the
950 * errors anyway - mbligh
951 */
952 printk(KERN_INFO "Leaving ESR disabled.\n");
953 else
954 printk(KERN_INFO "No ESR for 82489DX.\n");
955 } 1027 }
956} 1028}
957 1029
@@ -1089,13 +1161,17 @@ void __cpuinit setup_local_APIC(void)
1089 1161
1090void __cpuinit end_local_APIC_setup(void) 1162void __cpuinit end_local_APIC_setup(void)
1091{ 1163{
1092 unsigned long value;
1093
1094 lapic_setup_esr(); 1164 lapic_setup_esr();
1095 /* Disable the local apic timer */ 1165
1096 value = apic_read(APIC_LVTT); 1166#ifdef CONFIG_X86_32
1097 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 1167 {
1098 apic_write(APIC_LVTT, value); 1168 unsigned int value;
1169 /* Disable the local apic timer */
1170 value = apic_read(APIC_LVTT);
1171 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1172 apic_write(APIC_LVTT, value);
1173 }
1174#endif
1099 1175
1100 setup_apic_nmi_watchdog(NULL); 1176 setup_apic_nmi_watchdog(NULL);
1101 apic_pm_activate(); 1177 apic_pm_activate();
@@ -1205,7 +1281,7 @@ void __init init_apic_mappings(void)
1205 * default configuration (or the MP table is broken). 1281 * default configuration (or the MP table is broken).
1206 */ 1282 */
1207 if (boot_cpu_physical_apicid == -1U) 1283 if (boot_cpu_physical_apicid == -1U)
1208 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1284 boot_cpu_physical_apicid = read_apic_id();
1209 1285
1210} 1286}
1211 1287
@@ -1242,7 +1318,7 @@ int __init APIC_init_uniprocessor(void)
1242 * might be zero if read from MP tables. Get it from LAPIC. 1318 * might be zero if read from MP tables. Get it from LAPIC.
1243 */ 1319 */
1244#ifdef CONFIG_CRASH_DUMP 1320#ifdef CONFIG_CRASH_DUMP
1245 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1321 boot_cpu_physical_apicid = read_apic_id();
1246#endif 1322#endif
1247 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map); 1323 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1248 1324
@@ -1321,59 +1397,12 @@ void smp_error_interrupt(struct pt_regs *regs)
1321 irq_exit(); 1397 irq_exit();
1322} 1398}
1323 1399
1324#ifdef CONFIG_SMP
1325void __init smp_intr_init(void)
1326{
1327 /*
1328 * IRQ0 must be given a fixed assignment and initialized,
1329 * because it's used before the IO-APIC is set up.
1330 */
1331 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
1332
1333 /*
1334 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1335 * IPI, driven by wakeup.
1336 */
1337 alloc_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1338
1339 /* IPI for invalidation */
1340 alloc_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1341
1342 /* IPI for generic function call */
1343 alloc_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
1344
1345 /* IPI for single call function */
1346 set_intr_gate(CALL_FUNCTION_SINGLE_VECTOR,
1347 call_function_single_interrupt);
1348}
1349#endif
1350
1351/*
1352 * Initialize APIC interrupts
1353 */
1354void __init apic_intr_init(void)
1355{
1356#ifdef CONFIG_SMP
1357 smp_intr_init();
1358#endif
1359 /* self generated IPI for local APIC timer */
1360 alloc_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
1361
1362 /* IPI vectors for APIC spurious and error interrupts */
1363 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
1364 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
1365
1366 /* thermal monitor LVT interrupt */
1367#ifdef CONFIG_X86_MCE_P4THERMAL
1368 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
1369#endif
1370}
1371
1372/** 1400/**
1373 * connect_bsp_APIC - attach the APIC to the interrupt system 1401 * connect_bsp_APIC - attach the APIC to the interrupt system
1374 */ 1402 */
1375void __init connect_bsp_APIC(void) 1403void __init connect_bsp_APIC(void)
1376{ 1404{
1405#ifdef CONFIG_X86_32
1377 if (pic_mode) { 1406 if (pic_mode) {
1378 /* 1407 /*
1379 * Do not trust the local APIC being empty at bootup. 1408 * Do not trust the local APIC being empty at bootup.
@@ -1388,6 +1417,7 @@ void __init connect_bsp_APIC(void)
1388 outb(0x70, 0x22); 1417 outb(0x70, 0x22);
1389 outb(0x01, 0x23); 1418 outb(0x01, 0x23);
1390 } 1419 }
1420#endif
1391 enable_apic_mode(); 1421 enable_apic_mode();
1392} 1422}
1393 1423
@@ -1400,6 +1430,9 @@ void __init connect_bsp_APIC(void)
1400 */ 1430 */
1401void disconnect_bsp_APIC(int virt_wire_setup) 1431void disconnect_bsp_APIC(int virt_wire_setup)
1402{ 1432{
1433 unsigned int value;
1434
1435#ifdef CONFIG_X86_32
1403 if (pic_mode) { 1436 if (pic_mode) {
1404 /* 1437 /*
1405 * Put the board back into PIC mode (has an effect only on 1438 * Put the board back into PIC mode (has an effect only on
@@ -1411,54 +1444,53 @@ void disconnect_bsp_APIC(int virt_wire_setup)
1411 "entering PIC mode.\n"); 1444 "entering PIC mode.\n");
1412 outb(0x70, 0x22); 1445 outb(0x70, 0x22);
1413 outb(0x00, 0x23); 1446 outb(0x00, 0x23);
1414 } else { 1447 return;
1415 /* Go back to Virtual Wire compatibility mode */ 1448 }
1416 unsigned long value; 1449#endif
1417 1450
1418 /* For the spurious interrupt use vector F, and enable it */ 1451 /* Go back to Virtual Wire compatibility mode */
1419 value = apic_read(APIC_SPIV);
1420 value &= ~APIC_VECTOR_MASK;
1421 value |= APIC_SPIV_APIC_ENABLED;
1422 value |= 0xf;
1423 apic_write(APIC_SPIV, value);
1424 1452
1425 if (!virt_wire_setup) { 1453 /* For the spurious interrupt use vector F, and enable it */
1426 /* 1454 value = apic_read(APIC_SPIV);
1427 * For LVT0 make it edge triggered, active high, 1455 value &= ~APIC_VECTOR_MASK;
1428 * external and enabled 1456 value |= APIC_SPIV_APIC_ENABLED;
1429 */ 1457 value |= 0xf;
1430 value = apic_read(APIC_LVT0); 1458 apic_write(APIC_SPIV, value);
1431 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1432 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1433 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1434 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1435 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1436 apic_write(APIC_LVT0, value);
1437 } else {
1438 /* Disable LVT0 */
1439 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1440 }
1441 1459
1460 if (!virt_wire_setup) {
1442 /* 1461 /*
1443 * For LVT1 make it edge triggered, active high, nmi and 1462 * For LVT0 make it edge triggered, active high,
1444 * enabled 1463 * external and enabled
1445 */ 1464 */
1446 value = apic_read(APIC_LVT1); 1465 value = apic_read(APIC_LVT0);
1447 value &= ~( 1466 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1448 APIC_MODE_MASK | APIC_SEND_PENDING |
1449 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | 1467 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1450 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); 1468 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1451 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; 1469 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1452 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI); 1470 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1453 apic_write(APIC_LVT1, value); 1471 apic_write(APIC_LVT0, value);
1472 } else {
1473 /* Disable LVT0 */
1474 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1454 } 1475 }
1476
1477 /*
1478 * For LVT1 make it edge triggered, active high,
1479 * nmi and enabled
1480 */
1481 value = apic_read(APIC_LVT1);
1482 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1483 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1484 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1485 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1486 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1487 apic_write(APIC_LVT1, value);
1455} 1488}
1456 1489
1457void __cpuinit generic_processor_info(int apicid, int version) 1490void __cpuinit generic_processor_info(int apicid, int version)
1458{ 1491{
1459 int cpu; 1492 int cpu;
1460 cpumask_t tmp_map; 1493 cpumask_t tmp_map;
1461 physid_mask_t phys_cpu;
1462 1494
1463 /* 1495 /*
1464 * Validate version 1496 * Validate version
@@ -1471,9 +1503,6 @@ void __cpuinit generic_processor_info(int apicid, int version)
1471 } 1503 }
1472 apic_version[apicid] = version; 1504 apic_version[apicid] = version;
1473 1505
1474 phys_cpu = apicid_to_cpu_present(apicid);
1475 physids_or(phys_cpu_present_map, phys_cpu_present_map, phys_cpu);
1476
1477 if (num_processors >= NR_CPUS) { 1506 if (num_processors >= NR_CPUS) {
1478 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached." 1507 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached."
1479 " Processor ignored.\n", NR_CPUS); 1508 " Processor ignored.\n", NR_CPUS);
@@ -1484,17 +1513,19 @@ void __cpuinit generic_processor_info(int apicid, int version)
1484 cpus_complement(tmp_map, cpu_present_map); 1513 cpus_complement(tmp_map, cpu_present_map);
1485 cpu = first_cpu(tmp_map); 1514 cpu = first_cpu(tmp_map);
1486 1515
1487 if (apicid == boot_cpu_physical_apicid) 1516 physid_set(apicid, phys_cpu_present_map);
1517 if (apicid == boot_cpu_physical_apicid) {
1488 /* 1518 /*
1489 * x86_bios_cpu_apicid is required to have processors listed 1519 * x86_bios_cpu_apicid is required to have processors listed
1490 * in same order as logical cpu numbers. Hence the first 1520 * in same order as logical cpu numbers. Hence the first
1491 * entry is BSP, and so on. 1521 * entry is BSP, and so on.
1492 */ 1522 */
1493 cpu = 0; 1523 cpu = 0;
1494 1524 }
1495 if (apicid > max_physical_apicid) 1525 if (apicid > max_physical_apicid)
1496 max_physical_apicid = apicid; 1526 max_physical_apicid = apicid;
1497 1527
1528#ifdef CONFIG_X86_32
1498 /* 1529 /*
1499 * Would be preferable to switch to bigsmp when CONFIG_HOTPLUG_CPU=y 1530 * Would be preferable to switch to bigsmp when CONFIG_HOTPLUG_CPU=y
1500 * but we need to work other dependencies like SMP_SUSPEND etc 1531 * but we need to work other dependencies like SMP_SUSPEND etc
@@ -1514,7 +1545,9 @@ void __cpuinit generic_processor_info(int apicid, int version)
1514 def_to_bigsmp = 1; 1545 def_to_bigsmp = 1;
1515 } 1546 }
1516 } 1547 }
1517#ifdef CONFIG_SMP 1548#endif
1549
1550#if defined(CONFIG_X86_SMP) || defined(CONFIG_X86_64)
1518 /* are we being called early in kernel startup? */ 1551 /* are we being called early in kernel startup? */
1519 if (early_per_cpu_ptr(x86_cpu_to_apicid)) { 1552 if (early_per_cpu_ptr(x86_cpu_to_apicid)) {
1520 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid); 1553 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
@@ -1527,6 +1560,7 @@ void __cpuinit generic_processor_info(int apicid, int version)
1527 per_cpu(x86_bios_cpu_apicid, cpu) = apicid; 1560 per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
1528 } 1561 }
1529#endif 1562#endif
1563
1530 cpu_set(cpu, cpu_possible_map); 1564 cpu_set(cpu, cpu_possible_map);
1531 cpu_set(cpu, cpu_present_map); 1565 cpu_set(cpu, cpu_present_map);
1532} 1566}
@@ -1537,6 +1571,11 @@ void __cpuinit generic_processor_info(int apicid, int version)
1537#ifdef CONFIG_PM 1571#ifdef CONFIG_PM
1538 1572
1539static struct { 1573static struct {
1574 /*
1575 * 'active' is true if the local APIC was enabled by us and
1576 * not the BIOS; this signifies that we are also responsible
1577 * for disabling it before entering apm/acpi suspend
1578 */
1540 int active; 1579 int active;
1541 /* r/w apic fields */ 1580 /* r/w apic fields */
1542 unsigned int apic_id; 1581 unsigned int apic_id;
@@ -1577,7 +1616,7 @@ static int lapic_suspend(struct sys_device *dev, pm_message_t state)
1577 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR); 1616 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
1578 apic_pm_state.apic_tmict = apic_read(APIC_TMICT); 1617 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
1579 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR); 1618 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
1580#ifdef CONFIG_X86_MCE_P4THERMAL 1619#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
1581 if (maxlvt >= 5) 1620 if (maxlvt >= 5)
1582 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR); 1621 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
1583#endif 1622#endif
@@ -1601,16 +1640,23 @@ static int lapic_resume(struct sys_device *dev)
1601 1640
1602 local_irq_save(flags); 1641 local_irq_save(flags);
1603 1642
1604 /* 1643#ifdef CONFIG_X86_64
1605 * Make sure the APICBASE points to the right address 1644 if (x2apic)
1606 * 1645 enable_x2apic();
1607 * FIXME! This will be wrong if we ever support suspend on 1646 else
1608 * SMP! We'll need to do this as part of the CPU restore! 1647#endif
1609 */ 1648 {
1610 rdmsr(MSR_IA32_APICBASE, l, h); 1649 /*
1611 l &= ~MSR_IA32_APICBASE_BASE; 1650 * Make sure the APICBASE points to the right address
1612 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr; 1651 *
1613 wrmsr(MSR_IA32_APICBASE, l, h); 1652 * FIXME! This will be wrong if we ever support suspend on
1653 * SMP! We'll need to do this as part of the CPU restore!
1654 */
1655 rdmsr(MSR_IA32_APICBASE, l, h);
1656 l &= ~MSR_IA32_APICBASE_BASE;
1657 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
1658 wrmsr(MSR_IA32_APICBASE, l, h);
1659 }
1614 1660
1615 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED); 1661 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
1616 apic_write(APIC_ID, apic_pm_state.apic_id); 1662 apic_write(APIC_ID, apic_pm_state.apic_id);
@@ -1620,7 +1666,7 @@ static int lapic_resume(struct sys_device *dev)
1620 apic_write(APIC_SPIV, apic_pm_state.apic_spiv); 1666 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
1621 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0); 1667 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
1622 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1); 1668 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
1623#ifdef CONFIG_X86_MCE_P4THERMAL 1669#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
1624 if (maxlvt >= 5) 1670 if (maxlvt >= 5)
1625 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr); 1671 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
1626#endif 1672#endif
@@ -1634,7 +1680,9 @@ static int lapic_resume(struct sys_device *dev)
1634 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr); 1680 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
1635 apic_write(APIC_ESR, 0); 1681 apic_write(APIC_ESR, 0);
1636 apic_read(APIC_ESR); 1682 apic_read(APIC_ESR);
1683
1637 local_irq_restore(flags); 1684 local_irq_restore(flags);
1685
1638 return 0; 1686 return 0;
1639} 1687}
1640 1688
@@ -1690,20 +1738,20 @@ static int __init parse_lapic(char *arg)
1690} 1738}
1691early_param("lapic", parse_lapic); 1739early_param("lapic", parse_lapic);
1692 1740
1693static int __init parse_nolapic(char *arg) 1741static int __init setup_disableapic(char *arg)
1694{ 1742{
1695 disable_apic = 1; 1743 disable_apic = 1;
1696 setup_clear_cpu_cap(X86_FEATURE_APIC); 1744 setup_clear_cpu_cap(X86_FEATURE_APIC);
1697 return 0; 1745 return 0;
1698} 1746}
1699early_param("nolapic", parse_nolapic); 1747early_param("disableapic", setup_disableapic);
1700 1748
1701static int __init parse_disable_lapic_timer(char *arg) 1749/* same as disableapic, for compatibility */
1750static int __init setup_nolapic(char *arg)
1702{ 1751{
1703 local_apic_timer_disabled = 1; 1752 return setup_disableapic(arg);
1704 return 0;
1705} 1753}
1706early_param("nolapic_timer", parse_disable_lapic_timer); 1754early_param("nolapic", setup_nolapic);
1707 1755
1708static int __init parse_lapic_timer_c2_ok(char *arg) 1756static int __init parse_lapic_timer_c2_ok(char *arg)
1709{ 1757{
@@ -1712,15 +1760,40 @@ static int __init parse_lapic_timer_c2_ok(char *arg)
1712} 1760}
1713early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok); 1761early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
1714 1762
1763static int __init parse_disable_apic_timer(char *arg)
1764{
1765 disable_apic_timer = 1;
1766 return 0;
1767}
1768early_param("noapictimer", parse_disable_apic_timer);
1769
1770static int __init parse_nolapic_timer(char *arg)
1771{
1772 disable_apic_timer = 1;
1773 return 0;
1774}
1775early_param("nolapic_timer", parse_nolapic_timer);
1776
1715static int __init apic_set_verbosity(char *arg) 1777static int __init apic_set_verbosity(char *arg)
1716{ 1778{
1717 if (!arg) 1779 if (!arg) {
1780#ifdef CONFIG_X86_64
1781 skip_ioapic_setup = 0;
1782 ioapic_force = 1;
1783 return 0;
1784#endif
1718 return -EINVAL; 1785 return -EINVAL;
1786 }
1719 1787
1720 if (strcmp(arg, "debug") == 0) 1788 if (strcmp("debug", arg) == 0)
1721 apic_verbosity = APIC_DEBUG; 1789 apic_verbosity = APIC_DEBUG;
1722 else if (strcmp(arg, "verbose") == 0) 1790 else if (strcmp("verbose", arg) == 0)
1723 apic_verbosity = APIC_VERBOSE; 1791 apic_verbosity = APIC_VERBOSE;
1792 else {
1793 printk(KERN_WARNING "APIC Verbosity level %s not recognised"
1794 " use apic=verbose or apic=debug\n", arg);
1795 return -EINVAL;
1796 }
1724 1797
1725 return 0; 1798 return 0;
1726} 1799}
diff --git a/arch/x86/kernel/apic_64.c b/arch/x86/kernel/apic_64.c
index 446c062e831c..94ddb69ae15e 100644
--- a/arch/x86/kernel/apic_64.c
+++ b/arch/x86/kernel/apic_64.c
@@ -27,6 +27,7 @@
27#include <linux/clockchips.h> 27#include <linux/clockchips.h>
28#include <linux/acpi_pmtmr.h> 28#include <linux/acpi_pmtmr.h>
29#include <linux/module.h> 29#include <linux/module.h>
30#include <linux/dmar.h>
30 31
31#include <asm/atomic.h> 32#include <asm/atomic.h>
32#include <asm/smp.h> 33#include <asm/smp.h>
@@ -39,13 +40,20 @@
39#include <asm/proto.h> 40#include <asm/proto.h>
40#include <asm/timex.h> 41#include <asm/timex.h>
41#include <asm/apic.h> 42#include <asm/apic.h>
43#include <asm/i8259.h>
42 44
43#include <mach_ipi.h> 45#include <mach_ipi.h>
44#include <mach_apic.h> 46#include <mach_apic.h>
45 47
48/* Disable local APIC timer from the kernel commandline or via dmi quirk */
46static int disable_apic_timer __cpuinitdata; 49static int disable_apic_timer __cpuinitdata;
47static int apic_calibrate_pmtmr __initdata; 50static int apic_calibrate_pmtmr __initdata;
48int disable_apic; 51int disable_apic;
52int disable_x2apic;
53int x2apic;
54
55/* x2apic enabled before OS handover */
56int x2apic_preenabled;
49 57
50/* Local APIC timer works in C2 */ 58/* Local APIC timer works in C2 */
51int local_apic_timer_c2_ok; 59int local_apic_timer_c2_ok;
@@ -73,6 +81,9 @@ static void lapic_timer_setup(enum clock_event_mode mode,
73static void lapic_timer_broadcast(cpumask_t mask); 81static void lapic_timer_broadcast(cpumask_t mask);
74static void apic_pm_activate(void); 82static void apic_pm_activate(void);
75 83
84/*
85 * The local apic timer can be used for any function which is CPU local.
86 */
76static struct clock_event_device lapic_clockevent = { 87static struct clock_event_device lapic_clockevent = {
77 .name = "lapic", 88 .name = "lapic",
78 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT 89 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
@@ -99,11 +110,15 @@ static inline int lapic_get_version(void)
99} 110}
100 111
101/* 112/*
102 * Check, if the APIC is integrated or a seperate chip 113 * Check, if the APIC is integrated or a separate chip
103 */ 114 */
104static inline int lapic_is_integrated(void) 115static inline int lapic_is_integrated(void)
105{ 116{
117#ifdef CONFIG_X86_64
106 return 1; 118 return 1;
119#else
120 return APIC_INTEGRATED(lapic_get_version());
121#endif
107} 122}
108 123
109/* 124/*
@@ -118,13 +133,18 @@ static int modern_apic(void)
118 return lapic_get_version() >= 0x14; 133 return lapic_get_version() >= 0x14;
119} 134}
120 135
121void apic_wait_icr_idle(void) 136/*
137 * Paravirt kernels also might be using these below ops. So we still
138 * use generic apic_read()/apic_write(), which might be pointing to different
139 * ops in PARAVIRT case.
140 */
141void xapic_wait_icr_idle(void)
122{ 142{
123 while (apic_read(APIC_ICR) & APIC_ICR_BUSY) 143 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
124 cpu_relax(); 144 cpu_relax();
125} 145}
126 146
127u32 safe_apic_wait_icr_idle(void) 147u32 safe_xapic_wait_icr_idle(void)
128{ 148{
129 u32 send_status; 149 u32 send_status;
130 int timeout; 150 int timeout;
@@ -140,6 +160,68 @@ u32 safe_apic_wait_icr_idle(void)
140 return send_status; 160 return send_status;
141} 161}
142 162
163void xapic_icr_write(u32 low, u32 id)
164{
165 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
166 apic_write(APIC_ICR, low);
167}
168
169u64 xapic_icr_read(void)
170{
171 u32 icr1, icr2;
172
173 icr2 = apic_read(APIC_ICR2);
174 icr1 = apic_read(APIC_ICR);
175
176 return icr1 | ((u64)icr2 << 32);
177}
178
179static struct apic_ops xapic_ops = {
180 .read = native_apic_mem_read,
181 .write = native_apic_mem_write,
182 .icr_read = xapic_icr_read,
183 .icr_write = xapic_icr_write,
184 .wait_icr_idle = xapic_wait_icr_idle,
185 .safe_wait_icr_idle = safe_xapic_wait_icr_idle,
186};
187
188struct apic_ops __read_mostly *apic_ops = &xapic_ops;
189EXPORT_SYMBOL_GPL(apic_ops);
190
191static void x2apic_wait_icr_idle(void)
192{
193 /* no need to wait for icr idle in x2apic */
194 return;
195}
196
197static u32 safe_x2apic_wait_icr_idle(void)
198{
199 /* no need to wait for icr idle in x2apic */
200 return 0;
201}
202
203void x2apic_icr_write(u32 low, u32 id)
204{
205 wrmsrl(APIC_BASE_MSR + (APIC_ICR >> 4), ((__u64) id) << 32 | low);
206}
207
208u64 x2apic_icr_read(void)
209{
210 unsigned long val;
211
212 rdmsrl(APIC_BASE_MSR + (APIC_ICR >> 4), val);
213 return val;
214}
215
216static struct apic_ops x2apic_ops = {
217 .read = native_apic_msr_read,
218 .write = native_apic_msr_write,
219 .icr_read = x2apic_icr_read,
220 .icr_write = x2apic_icr_write,
221 .wait_icr_idle = x2apic_wait_icr_idle,
222 .safe_wait_icr_idle = safe_x2apic_wait_icr_idle,
223};
224
143/** 225/**
144 * enable_NMI_through_LVT0 - enable NMI through local vector table 0 226 * enable_NMI_through_LVT0 - enable NMI through local vector table 0
145 */ 227 */
@@ -149,6 +231,11 @@ void __cpuinit enable_NMI_through_LVT0(void)
149 231
150 /* unmask and set to NMI */ 232 /* unmask and set to NMI */
151 v = APIC_DM_NMI; 233 v = APIC_DM_NMI;
234
235 /* Level triggered for 82489DX (32bit mode) */
236 if (!lapic_is_integrated())
237 v |= APIC_LVT_LEVEL_TRIGGER;
238
152 apic_write(APIC_LVT0, v); 239 apic_write(APIC_LVT0, v);
153} 240}
154 241
@@ -157,14 +244,28 @@ void __cpuinit enable_NMI_through_LVT0(void)
157 */ 244 */
158int lapic_get_maxlvt(void) 245int lapic_get_maxlvt(void)
159{ 246{
160 unsigned int v, maxlvt; 247 unsigned int v;
161 248
162 v = apic_read(APIC_LVR); 249 v = apic_read(APIC_LVR);
163 maxlvt = GET_APIC_MAXLVT(v); 250 /*
164 return maxlvt; 251 * - we always have APIC integrated on 64bit mode
252 * - 82489DXs do not report # of LVT entries
253 */
254 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
165} 255}
166 256
167/* 257/*
258 * Local APIC timer
259 */
260
261/* Clock divisor */
262#ifdef CONFG_X86_64
263#define APIC_DIVISOR 1
264#else
265#define APIC_DIVISOR 16
266#endif
267
268/*
168 * This function sets up the local APIC timer, with a timeout of 269 * This function sets up the local APIC timer, with a timeout of
169 * 'clocks' APIC bus clock. During calibration we actually call 270 * 'clocks' APIC bus clock. During calibration we actually call
170 * this function twice on the boot CPU, once with a bogus timeout 271 * this function twice on the boot CPU, once with a bogus timeout
@@ -174,7 +275,6 @@ int lapic_get_maxlvt(void)
174 * We do reads before writes even if unnecessary, to get around the 275 * We do reads before writes even if unnecessary, to get around the
175 * P5 APIC double write bug. 276 * P5 APIC double write bug.
176 */ 277 */
177
178static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen) 278static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
179{ 279{
180 unsigned int lvtt_value, tmp_value; 280 unsigned int lvtt_value, tmp_value;
@@ -182,6 +282,9 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
182 lvtt_value = LOCAL_TIMER_VECTOR; 282 lvtt_value = LOCAL_TIMER_VECTOR;
183 if (!oneshot) 283 if (!oneshot)
184 lvtt_value |= APIC_LVT_TIMER_PERIODIC; 284 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
285 if (!lapic_is_integrated())
286 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
287
185 if (!irqen) 288 if (!irqen)
186 lvtt_value |= APIC_LVT_MASKED; 289 lvtt_value |= APIC_LVT_MASKED;
187 290
@@ -191,12 +294,12 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
191 * Divide PICLK by 16 294 * Divide PICLK by 16
192 */ 295 */
193 tmp_value = apic_read(APIC_TDCR); 296 tmp_value = apic_read(APIC_TDCR);
194 apic_write(APIC_TDCR, (tmp_value 297 apic_write(APIC_TDCR,
195 & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) 298 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
196 | APIC_TDR_DIV_16); 299 APIC_TDR_DIV_16);
197 300
198 if (!oneshot) 301 if (!oneshot)
199 apic_write(APIC_TMICT, clocks); 302 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
200} 303}
201 304
202/* 305/*
@@ -204,6 +307,9 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
204 * 307 *
205 * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and 308 * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and
206 * MCE interrupts are supported. Thus MCE offset must be set to 0. 309 * MCE interrupts are supported. Thus MCE offset must be set to 0.
310 *
311 * If mask=1, the LVT entry does not generate interrupts while mask=0
312 * enables the vector. See also the BKDGs.
207 */ 313 */
208 314
209#define APIC_EILVT_LVTOFF_MCE 0 315#define APIC_EILVT_LVTOFF_MCE 0
@@ -228,6 +334,7 @@ u8 setup_APIC_eilvt_ibs(u8 vector, u8 msg_type, u8 mask)
228 setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask); 334 setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask);
229 return APIC_EILVT_LVTOFF_IBS; 335 return APIC_EILVT_LVTOFF_IBS;
230} 336}
337EXPORT_SYMBOL_GPL(setup_APIC_eilvt_ibs);
231 338
232/* 339/*
233 * Program the next event, relative to now 340 * Program the next event, relative to now
@@ -366,7 +473,7 @@ static int __init calibrate_APIC_clock(void)
366 lapic_clockevent.min_delta_ns = 473 lapic_clockevent.min_delta_ns =
367 clockevent_delta2ns(0xF, &lapic_clockevent); 474 clockevent_delta2ns(0xF, &lapic_clockevent);
368 475
369 calibration_result = result / HZ; 476 calibration_result = (result * APIC_DIVISOR) / HZ;
370 477
371 /* 478 /*
372 * Do a sanity check on the APIC calibration result 479 * Do a sanity check on the APIC calibration result
@@ -388,10 +495,10 @@ static int __init calibrate_APIC_clock(void)
388void __init setup_boot_APIC_clock(void) 495void __init setup_boot_APIC_clock(void)
389{ 496{
390 /* 497 /*
391 * The local apic timer can be disabled via the kernel commandline. 498 * The local apic timer can be disabled via the kernel
392 * Register the lapic timer as a dummy clock event source on SMP 499 * commandline or from the CPU detection code. Register the lapic
393 * systems, so the broadcast mechanism is used. On UP systems simply 500 * timer as a dummy clock event source on SMP systems, so the
394 * ignore it. 501 * broadcast mechanism is used. On UP systems simply ignore it.
395 */ 502 */
396 if (disable_apic_timer) { 503 if (disable_apic_timer) {
397 printk(KERN_INFO "Disabling APIC timer\n"); 504 printk(KERN_INFO "Disabling APIC timer\n");
@@ -403,7 +510,9 @@ void __init setup_boot_APIC_clock(void)
403 return; 510 return;
404 } 511 }
405 512
406 printk(KERN_INFO "Using local APIC timer interrupts.\n"); 513 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
514 "calibrating APIC timer ...\n");
515
407 if (calibrate_APIC_clock()) { 516 if (calibrate_APIC_clock()) {
408 /* No broadcast on UP ! */ 517 /* No broadcast on UP ! */
409 if (num_possible_cpus() > 1) 518 if (num_possible_cpus() > 1)
@@ -422,6 +531,7 @@ void __init setup_boot_APIC_clock(void)
422 printk(KERN_WARNING "APIC timer registered as dummy," 531 printk(KERN_WARNING "APIC timer registered as dummy,"
423 " due to nmi_watchdog=%d!\n", nmi_watchdog); 532 " due to nmi_watchdog=%d!\n", nmi_watchdog);
424 533
534 /* Setup the lapic or request the broadcast */
425 setup_APIC_timer(); 535 setup_APIC_timer();
426} 536}
427 537
@@ -460,7 +570,11 @@ static void local_apic_timer_interrupt(void)
460 /* 570 /*
461 * the NMI deadlock-detector uses this. 571 * the NMI deadlock-detector uses this.
462 */ 572 */
573#ifdef CONFIG_X86_64
463 add_pda(apic_timer_irqs, 1); 574 add_pda(apic_timer_irqs, 1);
575#else
576 per_cpu(irq_stat, cpu).apic_timer_irqs++;
577#endif
464 578
465 evt->event_handler(evt); 579 evt->event_handler(evt);
466} 580}
@@ -491,6 +605,7 @@ void smp_apic_timer_interrupt(struct pt_regs *regs)
491 irq_enter(); 605 irq_enter();
492 local_apic_timer_interrupt(); 606 local_apic_timer_interrupt();
493 irq_exit(); 607 irq_exit();
608
494 set_irq_regs(old_regs); 609 set_irq_regs(old_regs);
495} 610}
496 611
@@ -544,6 +659,13 @@ void clear_local_APIC(void)
544 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED); 659 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
545 } 660 }
546 661
662 /* lets not touch this if we didn't frob it */
663#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(X86_MCE_INTEL)
664 if (maxlvt >= 5) {
665 v = apic_read(APIC_LVTTHMR);
666 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
667 }
668#endif
547 /* 669 /*
548 * Clean APIC state for other OSs: 670 * Clean APIC state for other OSs:
549 */ 671 */
@@ -554,8 +676,14 @@ void clear_local_APIC(void)
554 apic_write(APIC_LVTERR, APIC_LVT_MASKED); 676 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
555 if (maxlvt >= 4) 677 if (maxlvt >= 4)
556 apic_write(APIC_LVTPC, APIC_LVT_MASKED); 678 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
557 apic_write(APIC_ESR, 0); 679
558 apic_read(APIC_ESR); 680 /* Integrated APIC (!82489DX) ? */
681 if (lapic_is_integrated()) {
682 if (maxlvt > 3)
683 /* Clear ESR due to Pentium errata 3AP and 11AP */
684 apic_write(APIC_ESR, 0);
685 apic_read(APIC_ESR);
686 }
559} 687}
560 688
561/** 689/**
@@ -574,8 +702,28 @@ void disable_local_APIC(void)
574 value = apic_read(APIC_SPIV); 702 value = apic_read(APIC_SPIV);
575 value &= ~APIC_SPIV_APIC_ENABLED; 703 value &= ~APIC_SPIV_APIC_ENABLED;
576 apic_write(APIC_SPIV, value); 704 apic_write(APIC_SPIV, value);
705
706#ifdef CONFIG_X86_32
707 /*
708 * When LAPIC was disabled by the BIOS and enabled by the kernel,
709 * restore the disabled state.
710 */
711 if (enabled_via_apicbase) {
712 unsigned int l, h;
713
714 rdmsr(MSR_IA32_APICBASE, l, h);
715 l &= ~MSR_IA32_APICBASE_ENABLE;
716 wrmsr(MSR_IA32_APICBASE, l, h);
717 }
718#endif
577} 719}
578 720
721/*
722 * If Linux enabled the LAPIC against the BIOS default disable it down before
723 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
724 * not power-off. Additionally clear all LVT entries before disable_local_APIC
725 * for the case where Linux didn't enable the LAPIC.
726 */
579void lapic_shutdown(void) 727void lapic_shutdown(void)
580{ 728{
581 unsigned long flags; 729 unsigned long flags;
@@ -585,7 +733,13 @@ void lapic_shutdown(void)
585 733
586 local_irq_save(flags); 734 local_irq_save(flags);
587 735
588 disable_local_APIC(); 736#ifdef CONFIG_X86_32
737 if (!enabled_via_apicbase)
738 clear_local_APIC();
739 else
740#endif
741 disable_local_APIC();
742
589 743
590 local_irq_restore(flags); 744 local_irq_restore(flags);
591} 745}
@@ -629,10 +783,10 @@ int __init verify_local_APIC(void)
629 /* 783 /*
630 * The ID register is read/write in a real APIC. 784 * The ID register is read/write in a real APIC.
631 */ 785 */
632 reg0 = read_apic_id(); 786 reg0 = apic_read(APIC_ID);
633 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0); 787 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
634 apic_write(APIC_ID, reg0 ^ APIC_ID_MASK); 788 apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
635 reg1 = read_apic_id(); 789 reg1 = apic_read(APIC_ID);
636 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1); 790 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
637 apic_write(APIC_ID, reg0); 791 apic_write(APIC_ID, reg0);
638 if (reg1 != (reg0 ^ APIC_ID_MASK)) 792 if (reg1 != (reg0 ^ APIC_ID_MASK))
@@ -656,8 +810,11 @@ int __init verify_local_APIC(void)
656 */ 810 */
657void __init sync_Arb_IDs(void) 811void __init sync_Arb_IDs(void)
658{ 812{
659 /* Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 */ 813 /*
660 if (modern_apic()) 814 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
815 * needed on AMD.
816 */
817 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
661 return; 818 return;
662 819
663 /* 820 /*
@@ -666,8 +823,8 @@ void __init sync_Arb_IDs(void)
666 apic_wait_icr_idle(); 823 apic_wait_icr_idle();
667 824
668 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n"); 825 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
669 apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG 826 apic_write(APIC_ICR, APIC_DEST_ALLINC |
670 | APIC_DM_INIT); 827 APIC_INT_LEVELTRIG | APIC_DM_INIT);
671} 828}
672 829
673/* 830/*
@@ -684,8 +841,6 @@ void __init init_bsp_APIC(void)
684 if (smp_found_config || !cpu_has_apic) 841 if (smp_found_config || !cpu_has_apic)
685 return; 842 return;
686 843
687 value = apic_read(APIC_LVR);
688
689 /* 844 /*
690 * Do not trust the local APIC being empty at bootup. 845 * Do not trust the local APIC being empty at bootup.
691 */ 846 */
@@ -697,7 +852,15 @@ void __init init_bsp_APIC(void)
697 value = apic_read(APIC_SPIV); 852 value = apic_read(APIC_SPIV);
698 value &= ~APIC_VECTOR_MASK; 853 value &= ~APIC_VECTOR_MASK;
699 value |= APIC_SPIV_APIC_ENABLED; 854 value |= APIC_SPIV_APIC_ENABLED;
700 value |= APIC_SPIV_FOCUS_DISABLED; 855
856#ifdef CONFIG_X86_32
857 /* This bit is reserved on P4/Xeon and should be cleared */
858 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
859 (boot_cpu_data.x86 == 15))
860 value &= ~APIC_SPIV_FOCUS_DISABLED;
861 else
862#endif
863 value |= APIC_SPIV_FOCUS_DISABLED;
701 value |= SPURIOUS_APIC_VECTOR; 864 value |= SPURIOUS_APIC_VECTOR;
702 apic_write(APIC_SPIV, value); 865 apic_write(APIC_SPIV, value);
703 866
@@ -706,9 +869,50 @@ void __init init_bsp_APIC(void)
706 */ 869 */
707 apic_write(APIC_LVT0, APIC_DM_EXTINT); 870 apic_write(APIC_LVT0, APIC_DM_EXTINT);
708 value = APIC_DM_NMI; 871 value = APIC_DM_NMI;
872 if (!lapic_is_integrated()) /* 82489DX */
873 value |= APIC_LVT_LEVEL_TRIGGER;
709 apic_write(APIC_LVT1, value); 874 apic_write(APIC_LVT1, value);
710} 875}
711 876
877static void __cpuinit lapic_setup_esr(void)
878{
879 unsigned long oldvalue, value, maxlvt;
880 if (lapic_is_integrated() && !esr_disable) {
881 if (esr_disable) {
882 /*
883 * Something untraceable is creating bad interrupts on
884 * secondary quads ... for the moment, just leave the
885 * ESR disabled - we can't do anything useful with the
886 * errors anyway - mbligh
887 */
888 printk(KERN_INFO "Leaving ESR disabled.\n");
889 return;
890 }
891 /* !82489DX */
892 maxlvt = lapic_get_maxlvt();
893 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
894 apic_write(APIC_ESR, 0);
895 oldvalue = apic_read(APIC_ESR);
896
897 /* enables sending errors */
898 value = ERROR_APIC_VECTOR;
899 apic_write(APIC_LVTERR, value);
900 /*
901 * spec says clear errors after enabling vector.
902 */
903 if (maxlvt > 3)
904 apic_write(APIC_ESR, 0);
905 value = apic_read(APIC_ESR);
906 if (value != oldvalue)
907 apic_printk(APIC_VERBOSE, "ESR value before enabling "
908 "vector: 0x%08lx after: 0x%08lx\n",
909 oldvalue, value);
910 } else {
911 printk(KERN_INFO "No ESR for 82489DX.\n");
912 }
913}
914
915
712/** 916/**
713 * setup_local_APIC - setup the local APIC 917 * setup_local_APIC - setup the local APIC
714 */ 918 */
@@ -814,25 +1018,143 @@ void __cpuinit setup_local_APIC(void)
814 preempt_enable(); 1018 preempt_enable();
815} 1019}
816 1020
817static void __cpuinit lapic_setup_esr(void)
818{
819 unsigned maxlvt = lapic_get_maxlvt();
820
821 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR);
822 /*
823 * spec says clear errors after enabling vector.
824 */
825 if (maxlvt > 3)
826 apic_write(APIC_ESR, 0);
827}
828
829void __cpuinit end_local_APIC_setup(void) 1021void __cpuinit end_local_APIC_setup(void)
830{ 1022{
831 lapic_setup_esr(); 1023 lapic_setup_esr();
1024
1025#ifdef CONFIG_X86_32
1026 {
1027 unsigned int value;
1028 /* Disable the local apic timer */
1029 value = apic_read(APIC_LVTT);
1030 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1031 apic_write(APIC_LVTT, value);
1032 }
1033#endif
1034
832 setup_apic_nmi_watchdog(NULL); 1035 setup_apic_nmi_watchdog(NULL);
833 apic_pm_activate(); 1036 apic_pm_activate();
834} 1037}
835 1038
1039void check_x2apic(void)
1040{
1041 int msr, msr2;
1042
1043 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1044
1045 if (msr & X2APIC_ENABLE) {
1046 printk("x2apic enabled by BIOS, switching to x2apic ops\n");
1047 x2apic_preenabled = x2apic = 1;
1048 apic_ops = &x2apic_ops;
1049 }
1050}
1051
1052void enable_x2apic(void)
1053{
1054 int msr, msr2;
1055
1056 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1057 if (!(msr & X2APIC_ENABLE)) {
1058 printk("Enabling x2apic\n");
1059 wrmsr(MSR_IA32_APICBASE, msr | X2APIC_ENABLE, 0);
1060 }
1061}
1062
1063void enable_IR_x2apic(void)
1064{
1065#ifdef CONFIG_INTR_REMAP
1066 int ret;
1067 unsigned long flags;
1068
1069 if (!cpu_has_x2apic)
1070 return;
1071
1072 if (!x2apic_preenabled && disable_x2apic) {
1073 printk(KERN_INFO
1074 "Skipped enabling x2apic and Interrupt-remapping "
1075 "because of nox2apic\n");
1076 return;
1077 }
1078
1079 if (x2apic_preenabled && disable_x2apic)
1080 panic("Bios already enabled x2apic, can't enforce nox2apic");
1081
1082 if (!x2apic_preenabled && skip_ioapic_setup) {
1083 printk(KERN_INFO
1084 "Skipped enabling x2apic and Interrupt-remapping "
1085 "because of skipping io-apic setup\n");
1086 return;
1087 }
1088
1089 ret = dmar_table_init();
1090 if (ret) {
1091 printk(KERN_INFO
1092 "dmar_table_init() failed with %d:\n", ret);
1093
1094 if (x2apic_preenabled)
1095 panic("x2apic enabled by bios. But IR enabling failed");
1096 else
1097 printk(KERN_INFO
1098 "Not enabling x2apic,Intr-remapping\n");
1099 return;
1100 }
1101
1102 local_irq_save(flags);
1103 mask_8259A();
1104 save_mask_IO_APIC_setup();
1105
1106 ret = enable_intr_remapping(1);
1107
1108 if (ret && x2apic_preenabled) {
1109 local_irq_restore(flags);
1110 panic("x2apic enabled by bios. But IR enabling failed");
1111 }
1112
1113 if (ret)
1114 goto end;
1115
1116 if (!x2apic) {
1117 x2apic = 1;
1118 apic_ops = &x2apic_ops;
1119 enable_x2apic();
1120 }
1121end:
1122 if (ret)
1123 /*
1124 * IR enabling failed
1125 */
1126 restore_IO_APIC_setup();
1127 else
1128 reinit_intr_remapped_IO_APIC(x2apic_preenabled);
1129
1130 unmask_8259A();
1131 local_irq_restore(flags);
1132
1133 if (!ret) {
1134 if (!x2apic_preenabled)
1135 printk(KERN_INFO
1136 "Enabled x2apic and interrupt-remapping\n");
1137 else
1138 printk(KERN_INFO
1139 "Enabled Interrupt-remapping\n");
1140 } else
1141 printk(KERN_ERR
1142 "Failed to enable Interrupt-remapping and x2apic\n");
1143#else
1144 if (!cpu_has_x2apic)
1145 return;
1146
1147 if (x2apic_preenabled)
1148 panic("x2apic enabled prior OS handover,"
1149 " enable CONFIG_INTR_REMAP");
1150
1151 printk(KERN_INFO "Enable CONFIG_INTR_REMAP for enabling intr-remapping "
1152 " and x2apic\n");
1153#endif
1154
1155 return;
1156}
1157
836/* 1158/*
837 * Detect and enable local APICs on non-SMP boards. 1159 * Detect and enable local APICs on non-SMP boards.
838 * Original code written by Keir Fraser. 1160 * Original code written by Keir Fraser.
@@ -872,7 +1194,7 @@ void __init early_init_lapic_mapping(void)
872 * Fetch the APIC ID of the BSP in case we have a 1194 * Fetch the APIC ID of the BSP in case we have a
873 * default configuration (or the MP table is broken). 1195 * default configuration (or the MP table is broken).
874 */ 1196 */
875 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1197 boot_cpu_physical_apicid = read_apic_id();
876} 1198}
877 1199
878/** 1200/**
@@ -880,6 +1202,11 @@ void __init early_init_lapic_mapping(void)
880 */ 1202 */
881void __init init_apic_mappings(void) 1203void __init init_apic_mappings(void)
882{ 1204{
1205 if (x2apic) {
1206 boot_cpu_physical_apicid = read_apic_id();
1207 return;
1208 }
1209
883 /* 1210 /*
884 * If no local APIC can be found then set up a fake all 1211 * If no local APIC can be found then set up a fake all
885 * zeroes page to simulate the local APIC and another 1212 * zeroes page to simulate the local APIC and another
@@ -899,13 +1226,15 @@ void __init init_apic_mappings(void)
899 * Fetch the APIC ID of the BSP in case we have a 1226 * Fetch the APIC ID of the BSP in case we have a
900 * default configuration (or the MP table is broken). 1227 * default configuration (or the MP table is broken).
901 */ 1228 */
902 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1229 boot_cpu_physical_apicid = read_apic_id();
903} 1230}
904 1231
905/* 1232/*
906 * This initializes the IO-APIC and APIC hardware if this is 1233 * This initializes the IO-APIC and APIC hardware if this is
907 * a UP kernel. 1234 * a UP kernel.
908 */ 1235 */
1236int apic_version[MAX_APICS];
1237
909int __init APIC_init_uniprocessor(void) 1238int __init APIC_init_uniprocessor(void)
910{ 1239{
911 if (disable_apic) { 1240 if (disable_apic) {
@@ -918,6 +1247,9 @@ int __init APIC_init_uniprocessor(void)
918 return -1; 1247 return -1;
919 } 1248 }
920 1249
1250 enable_IR_x2apic();
1251 setup_apic_routing();
1252
921 verify_local_APIC(); 1253 verify_local_APIC();
922 1254
923 connect_bsp_APIC(); 1255 connect_bsp_APIC();
@@ -1004,17 +1336,57 @@ asmlinkage void smp_error_interrupt(void)
1004} 1336}
1005 1337
1006/** 1338/**
1007 * * connect_bsp_APIC - attach the APIC to the interrupt system 1339 * connect_bsp_APIC - attach the APIC to the interrupt system
1008 * */ 1340 */
1009void __init connect_bsp_APIC(void) 1341void __init connect_bsp_APIC(void)
1010{ 1342{
1343#ifdef CONFIG_X86_32
1344 if (pic_mode) {
1345 /*
1346 * Do not trust the local APIC being empty at bootup.
1347 */
1348 clear_local_APIC();
1349 /*
1350 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
1351 * local APIC to INT and NMI lines.
1352 */
1353 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1354 "enabling APIC mode.\n");
1355 outb(0x70, 0x22);
1356 outb(0x01, 0x23);
1357 }
1358#endif
1011 enable_apic_mode(); 1359 enable_apic_mode();
1012} 1360}
1013 1361
1362/**
1363 * disconnect_bsp_APIC - detach the APIC from the interrupt system
1364 * @virt_wire_setup: indicates, whether virtual wire mode is selected
1365 *
1366 * Virtual wire mode is necessary to deliver legacy interrupts even when the
1367 * APIC is disabled.
1368 */
1014void disconnect_bsp_APIC(int virt_wire_setup) 1369void disconnect_bsp_APIC(int virt_wire_setup)
1015{ 1370{
1371 unsigned int value;
1372
1373#ifdef CONFIG_X86_32
1374 if (pic_mode) {
1375 /*
1376 * Put the board back into PIC mode (has an effect only on
1377 * certain older boards). Note that APIC interrupts, including
1378 * IPIs, won't work beyond this point! The only exception are
1379 * INIT IPIs.
1380 */
1381 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1382 "entering PIC mode.\n");
1383 outb(0x70, 0x22);
1384 outb(0x00, 0x23);
1385 return;
1386 }
1387#endif
1388
1016 /* Go back to Virtual Wire compatibility mode */ 1389 /* Go back to Virtual Wire compatibility mode */
1017 unsigned long value;
1018 1390
1019 /* For the spurious interrupt use vector F, and enable it */ 1391 /* For the spurious interrupt use vector F, and enable it */
1020 value = apic_read(APIC_SPIV); 1392 value = apic_read(APIC_SPIV);
@@ -1040,7 +1412,10 @@ void disconnect_bsp_APIC(int virt_wire_setup)
1040 apic_write(APIC_LVT0, APIC_LVT_MASKED); 1412 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1041 } 1413 }
1042 1414
1043 /* For LVT1 make it edge triggered, active high, nmi and enabled */ 1415 /*
1416 * For LVT1 make it edge triggered, active high,
1417 * nmi and enabled
1418 */
1044 value = apic_read(APIC_LVT1); 1419 value = apic_read(APIC_LVT1);
1045 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING | 1420 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1046 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | 1421 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
@@ -1055,9 +1430,20 @@ void __cpuinit generic_processor_info(int apicid, int version)
1055 int cpu; 1430 int cpu;
1056 cpumask_t tmp_map; 1431 cpumask_t tmp_map;
1057 1432
1433 /*
1434 * Validate version
1435 */
1436 if (version == 0x0) {
1437 printk(KERN_WARNING "BIOS bug, APIC version is 0 for CPU#%d! "
1438 "fixing up to 0x10. (tell your hw vendor)\n",
1439 version);
1440 version = 0x10;
1441 }
1442 apic_version[apicid] = version;
1443
1058 if (num_processors >= NR_CPUS) { 1444 if (num_processors >= NR_CPUS) {
1059 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached." 1445 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached."
1060 " Processor ignored.\n", NR_CPUS); 1446 " Processor ignored.\n", NR_CPUS);
1061 return; 1447 return;
1062 } 1448 }
1063 1449
@@ -1077,6 +1463,29 @@ void __cpuinit generic_processor_info(int apicid, int version)
1077 if (apicid > max_physical_apicid) 1463 if (apicid > max_physical_apicid)
1078 max_physical_apicid = apicid; 1464 max_physical_apicid = apicid;
1079 1465
1466#ifdef CONFIG_X86_32
1467 /*
1468 * Would be preferable to switch to bigsmp when CONFIG_HOTPLUG_CPU=y
1469 * but we need to work other dependencies like SMP_SUSPEND etc
1470 * before this can be done without some confusion.
1471 * if (CPU_HOTPLUG_ENABLED || num_processors > 8)
1472 * - Ashok Raj <ashok.raj@intel.com>
1473 */
1474 if (max_physical_apicid >= 8) {
1475 switch (boot_cpu_data.x86_vendor) {
1476 case X86_VENDOR_INTEL:
1477 if (!APIC_XAPIC(version)) {
1478 def_to_bigsmp = 0;
1479 break;
1480 }
1481 /* If P4 and above fall through */
1482 case X86_VENDOR_AMD:
1483 def_to_bigsmp = 1;
1484 }
1485 }
1486#endif
1487
1488#if defined(CONFIG_X86_SMP) || defined(CONFIG_X86_64)
1080 /* are we being called early in kernel startup? */ 1489 /* are we being called early in kernel startup? */
1081 if (early_per_cpu_ptr(x86_cpu_to_apicid)) { 1490 if (early_per_cpu_ptr(x86_cpu_to_apicid)) {
1082 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid); 1491 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
@@ -1088,20 +1497,28 @@ void __cpuinit generic_processor_info(int apicid, int version)
1088 per_cpu(x86_cpu_to_apicid, cpu) = apicid; 1497 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
1089 per_cpu(x86_bios_cpu_apicid, cpu) = apicid; 1498 per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
1090 } 1499 }
1500#endif
1091 1501
1092 cpu_set(cpu, cpu_possible_map); 1502 cpu_set(cpu, cpu_possible_map);
1093 cpu_set(cpu, cpu_present_map); 1503 cpu_set(cpu, cpu_present_map);
1094} 1504}
1095 1505
1506int hard_smp_processor_id(void)
1507{
1508 return read_apic_id();
1509}
1510
1096/* 1511/*
1097 * Power management 1512 * Power management
1098 */ 1513 */
1099#ifdef CONFIG_PM 1514#ifdef CONFIG_PM
1100 1515
1101static struct { 1516static struct {
1102 /* 'active' is true if the local APIC was enabled by us and 1517 /*
1103 not the BIOS; this signifies that we are also responsible 1518 * 'active' is true if the local APIC was enabled by us and
1104 for disabling it before entering apm/acpi suspend */ 1519 * not the BIOS; this signifies that we are also responsible
1520 * for disabling it before entering apm/acpi suspend
1521 */
1105 int active; 1522 int active;
1106 /* r/w apic fields */ 1523 /* r/w apic fields */
1107 unsigned int apic_id; 1524 unsigned int apic_id;
@@ -1129,7 +1546,7 @@ static int lapic_suspend(struct sys_device *dev, pm_message_t state)
1129 1546
1130 maxlvt = lapic_get_maxlvt(); 1547 maxlvt = lapic_get_maxlvt();
1131 1548
1132 apic_pm_state.apic_id = read_apic_id(); 1549 apic_pm_state.apic_id = apic_read(APIC_ID);
1133 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI); 1550 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
1134 apic_pm_state.apic_ldr = apic_read(APIC_LDR); 1551 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
1135 apic_pm_state.apic_dfr = apic_read(APIC_DFR); 1552 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
@@ -1142,10 +1559,11 @@ static int lapic_suspend(struct sys_device *dev, pm_message_t state)
1142 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR); 1559 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
1143 apic_pm_state.apic_tmict = apic_read(APIC_TMICT); 1560 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
1144 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR); 1561 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
1145#ifdef CONFIG_X86_MCE_INTEL 1562#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
1146 if (maxlvt >= 5) 1563 if (maxlvt >= 5)
1147 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR); 1564 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
1148#endif 1565#endif
1566
1149 local_irq_save(flags); 1567 local_irq_save(flags);
1150 disable_local_APIC(); 1568 disable_local_APIC();
1151 local_irq_restore(flags); 1569 local_irq_restore(flags);
@@ -1164,10 +1582,25 @@ static int lapic_resume(struct sys_device *dev)
1164 maxlvt = lapic_get_maxlvt(); 1582 maxlvt = lapic_get_maxlvt();
1165 1583
1166 local_irq_save(flags); 1584 local_irq_save(flags);
1167 rdmsr(MSR_IA32_APICBASE, l, h); 1585
1168 l &= ~MSR_IA32_APICBASE_BASE; 1586#ifdef CONFIG_X86_64
1169 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr; 1587 if (x2apic)
1170 wrmsr(MSR_IA32_APICBASE, l, h); 1588 enable_x2apic();
1589 else
1590#endif
1591 {
1592 /*
1593 * Make sure the APICBASE points to the right address
1594 *
1595 * FIXME! This will be wrong if we ever support suspend on
1596 * SMP! We'll need to do this as part of the CPU restore!
1597 */
1598 rdmsr(MSR_IA32_APICBASE, l, h);
1599 l &= ~MSR_IA32_APICBASE_BASE;
1600 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
1601 wrmsr(MSR_IA32_APICBASE, l, h);
1602 }
1603
1171 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED); 1604 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
1172 apic_write(APIC_ID, apic_pm_state.apic_id); 1605 apic_write(APIC_ID, apic_pm_state.apic_id);
1173 apic_write(APIC_DFR, apic_pm_state.apic_dfr); 1606 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
@@ -1176,7 +1609,7 @@ static int lapic_resume(struct sys_device *dev)
1176 apic_write(APIC_SPIV, apic_pm_state.apic_spiv); 1609 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
1177 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0); 1610 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
1178 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1); 1611 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
1179#ifdef CONFIG_X86_MCE_INTEL 1612#if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
1180 if (maxlvt >= 5) 1613 if (maxlvt >= 5)
1181 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr); 1614 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
1182#endif 1615#endif
@@ -1190,10 +1623,17 @@ static int lapic_resume(struct sys_device *dev)
1190 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr); 1623 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
1191 apic_write(APIC_ESR, 0); 1624 apic_write(APIC_ESR, 0);
1192 apic_read(APIC_ESR); 1625 apic_read(APIC_ESR);
1626
1193 local_irq_restore(flags); 1627 local_irq_restore(flags);
1628
1194 return 0; 1629 return 0;
1195} 1630}
1196 1631
1632/*
1633 * This device has no shutdown method - fully functioning local APICs
1634 * are needed on every CPU up until machine_halt/restart/poweroff.
1635 */
1636
1197static struct sysdev_class lapic_sysclass = { 1637static struct sysdev_class lapic_sysclass = {
1198 .name = "lapic", 1638 .name = "lapic",
1199 .resume = lapic_resume, 1639 .resume = lapic_resume,
@@ -1307,31 +1747,19 @@ __cpuinit int apic_is_clustered_box(void)
1307 return (clusters > 2); 1747 return (clusters > 2);
1308} 1748}
1309 1749
1310/* 1750static __init int setup_nox2apic(char *str)
1311 * APIC command line parameters
1312 */
1313static int __init apic_set_verbosity(char *str)
1314{ 1751{
1315 if (str == NULL) { 1752 disable_x2apic = 1;
1316 skip_ioapic_setup = 0; 1753 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_X2APIC);
1317 ioapic_force = 1;
1318 return 0;
1319 }
1320 if (strcmp("debug", str) == 0)
1321 apic_verbosity = APIC_DEBUG;
1322 else if (strcmp("verbose", str) == 0)
1323 apic_verbosity = APIC_VERBOSE;
1324 else {
1325 printk(KERN_WARNING "APIC Verbosity level %s not recognised"
1326 " use apic=verbose or apic=debug\n", str);
1327 return -EINVAL;
1328 }
1329
1330 return 0; 1754 return 0;
1331} 1755}
1332early_param("apic", apic_set_verbosity); 1756early_param("nox2apic", setup_nox2apic);
1757
1333 1758
1334static __init int setup_disableapic(char *str) 1759/*
1760 * APIC command line parameters
1761 */
1762static int __init setup_disableapic(char *arg)
1335{ 1763{
1336 disable_apic = 1; 1764 disable_apic = 1;
1337 setup_clear_cpu_cap(X86_FEATURE_APIC); 1765 setup_clear_cpu_cap(X86_FEATURE_APIC);
@@ -1340,9 +1768,9 @@ static __init int setup_disableapic(char *str)
1340early_param("disableapic", setup_disableapic); 1768early_param("disableapic", setup_disableapic);
1341 1769
1342/* same as disableapic, for compatibility */ 1770/* same as disableapic, for compatibility */
1343static __init int setup_nolapic(char *str) 1771static int __init setup_nolapic(char *arg)
1344{ 1772{
1345 return setup_disableapic(str); 1773 return setup_disableapic(arg);
1346} 1774}
1347early_param("nolapic", setup_nolapic); 1775early_param("nolapic", setup_nolapic);
1348 1776
@@ -1353,14 +1781,19 @@ static int __init parse_lapic_timer_c2_ok(char *arg)
1353} 1781}
1354early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok); 1782early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
1355 1783
1356static __init int setup_noapictimer(char *str) 1784static int __init parse_disable_apic_timer(char *arg)
1357{ 1785{
1358 if (str[0] != ' ' && str[0] != 0)
1359 return 0;
1360 disable_apic_timer = 1; 1786 disable_apic_timer = 1;
1361 return 1; 1787 return 0;
1362} 1788}
1363__setup("noapictimer", setup_noapictimer); 1789early_param("noapictimer", parse_disable_apic_timer);
1790
1791static int __init parse_nolapic_timer(char *arg)
1792{
1793 disable_apic_timer = 1;
1794 return 0;
1795}
1796early_param("nolapic_timer", parse_nolapic_timer);
1364 1797
1365static __init int setup_apicpmtimer(char *s) 1798static __init int setup_apicpmtimer(char *s)
1366{ 1799{
@@ -1370,6 +1803,31 @@ static __init int setup_apicpmtimer(char *s)
1370} 1803}
1371__setup("apicpmtimer", setup_apicpmtimer); 1804__setup("apicpmtimer", setup_apicpmtimer);
1372 1805
1806static int __init apic_set_verbosity(char *arg)
1807{
1808 if (!arg) {
1809#ifdef CONFIG_X86_64
1810 skip_ioapic_setup = 0;
1811 ioapic_force = 1;
1812 return 0;
1813#endif
1814 return -EINVAL;
1815 }
1816
1817 if (strcmp("debug", arg) == 0)
1818 apic_verbosity = APIC_DEBUG;
1819 else if (strcmp("verbose", arg) == 0)
1820 apic_verbosity = APIC_VERBOSE;
1821 else {
1822 printk(KERN_WARNING "APIC Verbosity level %s not recognised"
1823 " use apic=verbose or apic=debug\n", arg);
1824 return -EINVAL;
1825 }
1826
1827 return 0;
1828}
1829early_param("apic", apic_set_verbosity);
1830
1373static int __init lapic_insert_resource(void) 1831static int __init lapic_insert_resource(void)
1374{ 1832{
1375 if (!apic_phys) 1833 if (!apic_phys)
diff --git a/arch/x86/kernel/apm_32.c b/arch/x86/kernel/apm_32.c
index 9ee24e6bc4b0..5145a6e72bbb 100644
--- a/arch/x86/kernel/apm_32.c
+++ b/arch/x86/kernel/apm_32.c
@@ -228,12 +228,12 @@
228#include <linux/suspend.h> 228#include <linux/suspend.h>
229#include <linux/kthread.h> 229#include <linux/kthread.h>
230#include <linux/jiffies.h> 230#include <linux/jiffies.h>
231#include <linux/smp_lock.h>
232 231
233#include <asm/system.h> 232#include <asm/system.h>
234#include <asm/uaccess.h> 233#include <asm/uaccess.h>
235#include <asm/desc.h> 234#include <asm/desc.h>
236#include <asm/i8253.h> 235#include <asm/i8253.h>
236#include <asm/olpc.h>
237#include <asm/paravirt.h> 237#include <asm/paravirt.h>
238#include <asm/reboot.h> 238#include <asm/reboot.h>
239 239
@@ -2217,7 +2217,7 @@ static int __init apm_init(void)
2217 2217
2218 dmi_check_system(apm_dmi_table); 2218 dmi_check_system(apm_dmi_table);
2219 2219
2220 if (apm_info.bios.version == 0 || paravirt_enabled()) { 2220 if (apm_info.bios.version == 0 || paravirt_enabled() || machine_is_olpc()) {
2221 printk(KERN_INFO "apm: BIOS not found.\n"); 2221 printk(KERN_INFO "apm: BIOS not found.\n");
2222 return -ENODEV; 2222 return -ENODEV;
2223 } 2223 }
diff --git a/arch/x86/kernel/asm-offsets_64.c b/arch/x86/kernel/asm-offsets_64.c
index aa89387006fe..505543a75a56 100644
--- a/arch/x86/kernel/asm-offsets_64.c
+++ b/arch/x86/kernel/asm-offsets_64.c
@@ -22,7 +22,7 @@
22 22
23#define __NO_STUBS 1 23#define __NO_STUBS 1
24#undef __SYSCALL 24#undef __SYSCALL
25#undef _ASM_X86_64_UNISTD_H_ 25#undef ASM_X86__UNISTD_64_H
26#define __SYSCALL(nr, sym) [nr] = 1, 26#define __SYSCALL(nr, sym) [nr] = 1,
27static char syscalls[] = { 27static char syscalls[] = {
28#include <asm/unistd.h> 28#include <asm/unistd.h>
diff --git a/arch/x86/kernel/bios_uv.c b/arch/x86/kernel/bios_uv.c
index c639bd55391c..fdd585f9c53d 100644
--- a/arch/x86/kernel/bios_uv.c
+++ b/arch/x86/kernel/bios_uv.c
@@ -25,11 +25,11 @@ x86_bios_strerror(long status)
25{ 25{
26 const char *str; 26 const char *str;
27 switch (status) { 27 switch (status) {
28 case 0: str = "Call completed without error"; break; 28 case 0: str = "Call completed without error"; break;
29 case -1: str = "Not implemented"; break; 29 case -1: str = "Not implemented"; break;
30 case -2: str = "Invalid argument"; break; 30 case -2: str = "Invalid argument"; break;
31 case -3: str = "Call completed with error"; break; 31 case -3: str = "Call completed with error"; break;
32 default: str = "Unknown BIOS status code"; break; 32 default: str = "Unknown BIOS status code"; break;
33 } 33 }
34 return str; 34 return str;
35} 35}
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index ee76eaad3001..7f0b45a5d788 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -3,22 +3,30 @@
3# 3#
4 4
5obj-y := intel_cacheinfo.o addon_cpuid_features.o 5obj-y := intel_cacheinfo.o addon_cpuid_features.o
6obj-y += proc.o feature_names.o 6obj-y += proc.o capflags.o powerflags.o common.o
7 7
8obj-$(CONFIG_X86_32) += common.o bugs.o 8obj-$(CONFIG_X86_32) += bugs.o cmpxchg.o
9obj-$(CONFIG_X86_64) += common_64.o bugs_64.o 9obj-$(CONFIG_X86_64) += bugs_64.o
10obj-$(CONFIG_X86_32) += amd.o 10
11obj-$(CONFIG_X86_64) += amd_64.o 11obj-$(CONFIG_CPU_SUP_INTEL) += intel.o
12obj-$(CONFIG_X86_32) += cyrix.o 12obj-$(CONFIG_CPU_SUP_AMD) += amd.o
13obj-$(CONFIG_X86_32) += centaur.o 13obj-$(CONFIG_CPU_SUP_CYRIX_32) += cyrix.o
14obj-$(CONFIG_X86_64) += centaur_64.o 14obj-$(CONFIG_CPU_SUP_CENTAUR_32) += centaur.o
15obj-$(CONFIG_X86_32) += transmeta.o 15obj-$(CONFIG_CPU_SUP_CENTAUR_64) += centaur_64.o
16obj-$(CONFIG_X86_32) += intel.o 16obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o
17obj-$(CONFIG_X86_64) += intel_64.o 17obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o
18obj-$(CONFIG_X86_32) += umc.o
19 18
20obj-$(CONFIG_X86_MCE) += mcheck/ 19obj-$(CONFIG_X86_MCE) += mcheck/
21obj-$(CONFIG_MTRR) += mtrr/ 20obj-$(CONFIG_MTRR) += mtrr/
22obj-$(CONFIG_CPU_FREQ) += cpufreq/ 21obj-$(CONFIG_CPU_FREQ) += cpufreq/
23 22
24obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o 23obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o
24
25quiet_cmd_mkcapflags = MKCAP $@
26 cmd_mkcapflags = $(PERL) $(srctree)/$(src)/mkcapflags.pl $< $@
27
28cpufeature = $(src)/../../../../include/asm-x86/cpufeature.h
29
30targets += capflags.c
31$(obj)/capflags.c: $(cpufeature) $(src)/mkcapflags.pl FORCE
32 $(call if_changed,mkcapflags)
diff --git a/arch/x86/kernel/cpu/addon_cpuid_features.c b/arch/x86/kernel/cpu/addon_cpuid_features.c
index a6ef672adbba..0d9c993aa93e 100644
--- a/arch/x86/kernel/cpu/addon_cpuid_features.c
+++ b/arch/x86/kernel/cpu/addon_cpuid_features.c
@@ -7,6 +7,8 @@
7#include <asm/pat.h> 7#include <asm/pat.h>
8#include <asm/processor.h> 8#include <asm/processor.h>
9 9
10#include <mach_apic.h>
11
10struct cpuid_bit { 12struct cpuid_bit {
11 u16 feature; 13 u16 feature;
12 u8 reg; 14 u8 reg;
@@ -48,6 +50,92 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c)
48 } 50 }
49} 51}
50 52
53/* leaf 0xb SMT level */
54#define SMT_LEVEL 0
55
56/* leaf 0xb sub-leaf types */
57#define INVALID_TYPE 0
58#define SMT_TYPE 1
59#define CORE_TYPE 2
60
61#define LEAFB_SUBTYPE(ecx) (((ecx) >> 8) & 0xff)
62#define BITS_SHIFT_NEXT_LEVEL(eax) ((eax) & 0x1f)
63#define LEVEL_MAX_SIBLINGS(ebx) ((ebx) & 0xffff)
64
65/*
66 * Check for extended topology enumeration cpuid leaf 0xb and if it
67 * exists, use it for populating initial_apicid and cpu topology
68 * detection.
69 */
70void __cpuinit detect_extended_topology(struct cpuinfo_x86 *c)
71{
72#ifdef CONFIG_SMP
73 unsigned int eax, ebx, ecx, edx, sub_index;
74 unsigned int ht_mask_width, core_plus_mask_width;
75 unsigned int core_select_mask, core_level_siblings;
76
77 if (c->cpuid_level < 0xb)
78 return;
79
80 cpuid_count(0xb, SMT_LEVEL, &eax, &ebx, &ecx, &edx);
81
82 /*
83 * check if the cpuid leaf 0xb is actually implemented.
84 */
85 if (ebx == 0 || (LEAFB_SUBTYPE(ecx) != SMT_TYPE))
86 return;
87
88 set_cpu_cap(c, X86_FEATURE_XTOPOLOGY);
89
90 /*
91 * initial apic id, which also represents 32-bit extended x2apic id.
92 */
93 c->initial_apicid = edx;
94
95 /*
96 * Populate HT related information from sub-leaf level 0.
97 */
98 core_level_siblings = smp_num_siblings = LEVEL_MAX_SIBLINGS(ebx);
99 core_plus_mask_width = ht_mask_width = BITS_SHIFT_NEXT_LEVEL(eax);
100
101 sub_index = 1;
102 do {
103 cpuid_count(0xb, sub_index, &eax, &ebx, &ecx, &edx);
104
105 /*
106 * Check for the Core type in the implemented sub leaves.
107 */
108 if (LEAFB_SUBTYPE(ecx) == CORE_TYPE) {
109 core_level_siblings = LEVEL_MAX_SIBLINGS(ebx);
110 core_plus_mask_width = BITS_SHIFT_NEXT_LEVEL(eax);
111 break;
112 }
113
114 sub_index++;
115 } while (LEAFB_SUBTYPE(ecx) != INVALID_TYPE);
116
117 core_select_mask = (~(-1 << core_plus_mask_width)) >> ht_mask_width;
118
119#ifdef CONFIG_X86_32
120 c->cpu_core_id = phys_pkg_id(c->initial_apicid, ht_mask_width)
121 & core_select_mask;
122 c->phys_proc_id = phys_pkg_id(c->initial_apicid, core_plus_mask_width);
123#else
124 c->cpu_core_id = phys_pkg_id(ht_mask_width) & core_select_mask;
125 c->phys_proc_id = phys_pkg_id(core_plus_mask_width);
126#endif
127 c->x86_max_cores = (core_level_siblings / smp_num_siblings);
128
129
130 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
131 c->phys_proc_id);
132 if (c->x86_max_cores > 1)
133 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
134 c->cpu_core_id);
135 return;
136#endif
137}
138
51#ifdef CONFIG_X86_PAT 139#ifdef CONFIG_X86_PAT
52void __cpuinit validate_pat_support(struct cpuinfo_x86 *c) 140void __cpuinit validate_pat_support(struct cpuinfo_x86 *c)
53{ 141{
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index cae9cabc3031..32e73520adf7 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -1,13 +1,22 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/bitops.h> 2#include <linux/bitops.h>
3#include <linux/mm.h> 3#include <linux/mm.h>
4
4#include <asm/io.h> 5#include <asm/io.h>
5#include <asm/processor.h> 6#include <asm/processor.h>
6#include <asm/apic.h> 7#include <asm/apic.h>
7 8
9#ifdef CONFIG_X86_64
10# include <asm/numa_64.h>
11# include <asm/mmconfig.h>
12# include <asm/cacheflush.h>
13#endif
14
8#include <mach_apic.h> 15#include <mach_apic.h>
16
9#include "cpu.h" 17#include "cpu.h"
10 18
19#ifdef CONFIG_X86_32
11/* 20/*
12 * B step AMD K6 before B 9730xxxx have hardware bugs that can cause 21 * B step AMD K6 before B 9730xxxx have hardware bugs that can cause
13 * misexecution of code under Linux. Owners of such processors should 22 * misexecution of code under Linux. Owners of such processors should
@@ -24,21 +33,273 @@
24extern void vide(void); 33extern void vide(void);
25__asm__(".align 4\nvide: ret"); 34__asm__(".align 4\nvide: ret");
26 35
27static void __cpuinit early_init_amd(struct cpuinfo_x86 *c) 36static void __cpuinit init_amd_k5(struct cpuinfo_x86 *c)
28{ 37{
29 if (cpuid_eax(0x80000000) >= 0x80000007) { 38/*
30 c->x86_power = cpuid_edx(0x80000007); 39 * General Systems BIOSen alias the cpu frequency registers
31 if (c->x86_power & (1<<8)) 40 * of the Elan at 0x000df000. Unfortuantly, one of the Linux
32 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 41 * drivers subsequently pokes it, and changes the CPU speed.
42 * Workaround : Remove the unneeded alias.
43 */
44#define CBAR (0xfffc) /* Configuration Base Address (32-bit) */
45#define CBAR_ENB (0x80000000)
46#define CBAR_KEY (0X000000CB)
47 if (c->x86_model == 9 || c->x86_model == 10) {
48 if (inl (CBAR) & CBAR_ENB)
49 outl (0 | CBAR_KEY, CBAR);
33 } 50 }
34} 51}
35 52
36static void __cpuinit init_amd(struct cpuinfo_x86 *c) 53
54static void __cpuinit init_amd_k6(struct cpuinfo_x86 *c)
37{ 55{
38 u32 l, h; 56 u32 l, h;
39 int mbytes = num_physpages >> (20-PAGE_SHIFT); 57 int mbytes = num_physpages >> (20-PAGE_SHIFT);
40 int r;
41 58
59 if (c->x86_model < 6) {
60 /* Based on AMD doc 20734R - June 2000 */
61 if (c->x86_model == 0) {
62 clear_cpu_cap(c, X86_FEATURE_APIC);
63 set_cpu_cap(c, X86_FEATURE_PGE);
64 }
65 return;
66 }
67
68 if (c->x86_model == 6 && c->x86_mask == 1) {
69 const int K6_BUG_LOOP = 1000000;
70 int n;
71 void (*f_vide)(void);
72 unsigned long d, d2;
73
74 printk(KERN_INFO "AMD K6 stepping B detected - ");
75
76 /*
77 * It looks like AMD fixed the 2.6.2 bug and improved indirect
78 * calls at the same time.
79 */
80
81 n = K6_BUG_LOOP;
82 f_vide = vide;
83 rdtscl(d);
84 while (n--)
85 f_vide();
86 rdtscl(d2);
87 d = d2-d;
88
89 if (d > 20*K6_BUG_LOOP)
90 printk("system stability may be impaired when more than 32 MB are used.\n");
91 else
92 printk("probably OK (after B9730xxxx).\n");
93 printk(KERN_INFO "Please see http://membres.lycos.fr/poulot/k6bug.html\n");
94 }
95
96 /* K6 with old style WHCR */
97 if (c->x86_model < 8 ||
98 (c->x86_model == 8 && c->x86_mask < 8)) {
99 /* We can only write allocate on the low 508Mb */
100 if (mbytes > 508)
101 mbytes = 508;
102
103 rdmsr(MSR_K6_WHCR, l, h);
104 if ((l&0x0000FFFF) == 0) {
105 unsigned long flags;
106 l = (1<<0)|((mbytes/4)<<1);
107 local_irq_save(flags);
108 wbinvd();
109 wrmsr(MSR_K6_WHCR, l, h);
110 local_irq_restore(flags);
111 printk(KERN_INFO "Enabling old style K6 write allocation for %d Mb\n",
112 mbytes);
113 }
114 return;
115 }
116
117 if ((c->x86_model == 8 && c->x86_mask > 7) ||
118 c->x86_model == 9 || c->x86_model == 13) {
119 /* The more serious chips .. */
120
121 if (mbytes > 4092)
122 mbytes = 4092;
123
124 rdmsr(MSR_K6_WHCR, l, h);
125 if ((l&0xFFFF0000) == 0) {
126 unsigned long flags;
127 l = ((mbytes>>2)<<22)|(1<<16);
128 local_irq_save(flags);
129 wbinvd();
130 wrmsr(MSR_K6_WHCR, l, h);
131 local_irq_restore(flags);
132 printk(KERN_INFO "Enabling new style K6 write allocation for %d Mb\n",
133 mbytes);
134 }
135
136 return;
137 }
138
139 if (c->x86_model == 10) {
140 /* AMD Geode LX is model 10 */
141 /* placeholder for any needed mods */
142 return;
143 }
144}
145
146static void __cpuinit init_amd_k7(struct cpuinfo_x86 *c)
147{
148 u32 l, h;
149
150 /*
151 * Bit 15 of Athlon specific MSR 15, needs to be 0
152 * to enable SSE on Palomino/Morgan/Barton CPU's.
153 * If the BIOS didn't enable it already, enable it here.
154 */
155 if (c->x86_model >= 6 && c->x86_model <= 10) {
156 if (!cpu_has(c, X86_FEATURE_XMM)) {
157 printk(KERN_INFO "Enabling disabled K7/SSE Support.\n");
158 rdmsr(MSR_K7_HWCR, l, h);
159 l &= ~0x00008000;
160 wrmsr(MSR_K7_HWCR, l, h);
161 set_cpu_cap(c, X86_FEATURE_XMM);
162 }
163 }
164
165 /*
166 * It's been determined by AMD that Athlons since model 8 stepping 1
167 * are more robust with CLK_CTL set to 200xxxxx instead of 600xxxxx
168 * As per AMD technical note 27212 0.2
169 */
170 if ((c->x86_model == 8 && c->x86_mask >= 1) || (c->x86_model > 8)) {
171 rdmsr(MSR_K7_CLK_CTL, l, h);
172 if ((l & 0xfff00000) != 0x20000000) {
173 printk ("CPU: CLK_CTL MSR was %x. Reprogramming to %x\n", l,
174 ((l & 0x000fffff)|0x20000000));
175 wrmsr(MSR_K7_CLK_CTL, (l & 0x000fffff)|0x20000000, h);
176 }
177 }
178
179 set_cpu_cap(c, X86_FEATURE_K7);
180}
181#endif
182
183#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
184static int __cpuinit nearby_node(int apicid)
185{
186 int i, node;
187
188 for (i = apicid - 1; i >= 0; i--) {
189 node = apicid_to_node[i];
190 if (node != NUMA_NO_NODE && node_online(node))
191 return node;
192 }
193 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
194 node = apicid_to_node[i];
195 if (node != NUMA_NO_NODE && node_online(node))
196 return node;
197 }
198 return first_node(node_online_map); /* Shouldn't happen */
199}
200#endif
201
202/*
203 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
204 * Assumes number of cores is a power of two.
205 */
206static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
207{
208#ifdef CONFIG_X86_HT
209 unsigned bits;
210
211 bits = c->x86_coreid_bits;
212
213 /* Low order bits define the core id (index of core in socket) */
214 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
215 /* Convert the initial APIC ID into the socket ID */
216 c->phys_proc_id = c->initial_apicid >> bits;
217#endif
218}
219
220static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
221{
222#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
223 int cpu = smp_processor_id();
224 int node;
225 unsigned apicid = hard_smp_processor_id();
226
227 node = c->phys_proc_id;
228 if (apicid_to_node[apicid] != NUMA_NO_NODE)
229 node = apicid_to_node[apicid];
230 if (!node_online(node)) {
231 /* Two possibilities here:
232 - The CPU is missing memory and no node was created.
233 In that case try picking one from a nearby CPU
234 - The APIC IDs differ from the HyperTransport node IDs
235 which the K8 northbridge parsing fills in.
236 Assume they are all increased by a constant offset,
237 but in the same order as the HT nodeids.
238 If that doesn't result in a usable node fall back to the
239 path for the previous case. */
240
241 int ht_nodeid = c->initial_apicid;
242
243 if (ht_nodeid >= 0 &&
244 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
245 node = apicid_to_node[ht_nodeid];
246 /* Pick a nearby node */
247 if (!node_online(node))
248 node = nearby_node(apicid);
249 }
250 numa_set_node(cpu, node);
251
252 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
253#endif
254}
255
256static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
257{
258#ifdef CONFIG_X86_HT
259 unsigned bits, ecx;
260
261 /* Multi core CPU? */
262 if (c->extended_cpuid_level < 0x80000008)
263 return;
264
265 ecx = cpuid_ecx(0x80000008);
266
267 c->x86_max_cores = (ecx & 0xff) + 1;
268
269 /* CPU telling us the core id bits shift? */
270 bits = (ecx >> 12) & 0xF;
271
272 /* Otherwise recompute */
273 if (bits == 0) {
274 while ((1 << bits) < c->x86_max_cores)
275 bits++;
276 }
277
278 c->x86_coreid_bits = bits;
279#endif
280}
281
282static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
283{
284 early_init_amd_mc(c);
285
286 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
287 if (c->x86_power & (1<<8))
288 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
289
290#ifdef CONFIG_X86_64
291 set_cpu_cap(c, X86_FEATURE_SYSCALL32);
292#else
293 /* Set MTRR capability flag if appropriate */
294 if (c->x86 == 5)
295 if (c->x86_model == 13 || c->x86_model == 9 ||
296 (c->x86_model == 8 && c->x86_mask >= 8))
297 set_cpu_cap(c, X86_FEATURE_K6_MTRR);
298#endif
299}
300
301static void __cpuinit init_amd(struct cpuinfo_x86 *c)
302{
42#ifdef CONFIG_SMP 303#ifdef CONFIG_SMP
43 unsigned long long value; 304 unsigned long long value;
44 305
@@ -49,7 +310,7 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
49 * Errata 63 for SH-B3 steppings 310 * Errata 63 for SH-B3 steppings
50 * Errata 122 for all steppings (F+ have it disabled by default) 311 * Errata 122 for all steppings (F+ have it disabled by default)
51 */ 312 */
52 if (c->x86 == 15) { 313 if (c->x86 == 0xf) {
53 rdmsrl(MSR_K7_HWCR, value); 314 rdmsrl(MSR_K7_HWCR, value);
54 value |= 1 << 6; 315 value |= 1 << 6;
55 wrmsrl(MSR_K7_HWCR, value); 316 wrmsrl(MSR_K7_HWCR, value);
@@ -59,213 +320,119 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
59 early_init_amd(c); 320 early_init_amd(c);
60 321
61 /* 322 /*
62 * FIXME: We should handle the K5 here. Set up the write
63 * range and also turn on MSR 83 bits 4 and 31 (write alloc,
64 * no bus pipeline)
65 */
66
67 /*
68 * Bit 31 in normal CPUID used for nonstandard 3DNow ID; 323 * Bit 31 in normal CPUID used for nonstandard 3DNow ID;
69 * 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway 324 * 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway
70 */ 325 */
71 clear_cpu_cap(c, 0*32+31); 326 clear_cpu_cap(c, 0*32+31);
72 327
73 r = get_model_name(c); 328#ifdef CONFIG_X86_64
329 /* On C+ stepping K8 rep microcode works well for copy/memset */
330 if (c->x86 == 0xf) {
331 u32 level;
74 332
75 switch (c->x86) { 333 level = cpuid_eax(1);
76 case 4: 334 if((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
77 /* 335 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
78 * General Systems BIOSen alias the cpu frequency registers
79 * of the Elan at 0x000df000. Unfortuantly, one of the Linux
80 * drivers subsequently pokes it, and changes the CPU speed.
81 * Workaround : Remove the unneeded alias.
82 */
83#define CBAR (0xfffc) /* Configuration Base Address (32-bit) */
84#define CBAR_ENB (0x80000000)
85#define CBAR_KEY (0X000000CB)
86 if (c->x86_model == 9 || c->x86_model == 10) {
87 if (inl (CBAR) & CBAR_ENB)
88 outl (0 | CBAR_KEY, CBAR);
89 }
90 break;
91 case 5:
92 if (c->x86_model < 6) {
93 /* Based on AMD doc 20734R - June 2000 */
94 if (c->x86_model == 0) {
95 clear_cpu_cap(c, X86_FEATURE_APIC);
96 set_cpu_cap(c, X86_FEATURE_PGE);
97 }
98 break;
99 }
100
101 if (c->x86_model == 6 && c->x86_mask == 1) {
102 const int K6_BUG_LOOP = 1000000;
103 int n;
104 void (*f_vide)(void);
105 unsigned long d, d2;
106
107 printk(KERN_INFO "AMD K6 stepping B detected - ");
108
109 /*
110 * It looks like AMD fixed the 2.6.2 bug and improved indirect
111 * calls at the same time.
112 */
113
114 n = K6_BUG_LOOP;
115 f_vide = vide;
116 rdtscl(d);
117 while (n--)
118 f_vide();
119 rdtscl(d2);
120 d = d2-d;
121
122 if (d > 20*K6_BUG_LOOP)
123 printk("system stability may be impaired when more than 32 MB are used.\n");
124 else
125 printk("probably OK (after B9730xxxx).\n");
126 printk(KERN_INFO "Please see http://membres.lycos.fr/poulot/k6bug.html\n");
127 }
128
129 /* K6 with old style WHCR */
130 if (c->x86_model < 8 ||
131 (c->x86_model == 8 && c->x86_mask < 8)) {
132 /* We can only write allocate on the low 508Mb */
133 if (mbytes > 508)
134 mbytes = 508;
135
136 rdmsr(MSR_K6_WHCR, l, h);
137 if ((l&0x0000FFFF) == 0) {
138 unsigned long flags;
139 l = (1<<0)|((mbytes/4)<<1);
140 local_irq_save(flags);
141 wbinvd();
142 wrmsr(MSR_K6_WHCR, l, h);
143 local_irq_restore(flags);
144 printk(KERN_INFO "Enabling old style K6 write allocation for %d Mb\n",
145 mbytes);
146 }
147 break;
148 }
149
150 if ((c->x86_model == 8 && c->x86_mask > 7) ||
151 c->x86_model == 9 || c->x86_model == 13) {
152 /* The more serious chips .. */
153
154 if (mbytes > 4092)
155 mbytes = 4092;
156
157 rdmsr(MSR_K6_WHCR, l, h);
158 if ((l&0xFFFF0000) == 0) {
159 unsigned long flags;
160 l = ((mbytes>>2)<<22)|(1<<16);
161 local_irq_save(flags);
162 wbinvd();
163 wrmsr(MSR_K6_WHCR, l, h);
164 local_irq_restore(flags);
165 printk(KERN_INFO "Enabling new style K6 write allocation for %d Mb\n",
166 mbytes);
167 }
168
169 /* Set MTRR capability flag if appropriate */
170 if (c->x86_model == 13 || c->x86_model == 9 ||
171 (c->x86_model == 8 && c->x86_mask >= 8))
172 set_cpu_cap(c, X86_FEATURE_K6_MTRR);
173 break;
174 }
175
176 if (c->x86_model == 10) {
177 /* AMD Geode LX is model 10 */
178 /* placeholder for any needed mods */
179 break;
180 }
181 break;
182 case 6: /* An Athlon/Duron */
183
184 /*
185 * Bit 15 of Athlon specific MSR 15, needs to be 0
186 * to enable SSE on Palomino/Morgan/Barton CPU's.
187 * If the BIOS didn't enable it already, enable it here.
188 */
189 if (c->x86_model >= 6 && c->x86_model <= 10) {
190 if (!cpu_has(c, X86_FEATURE_XMM)) {
191 printk(KERN_INFO "Enabling disabled K7/SSE Support.\n");
192 rdmsr(MSR_K7_HWCR, l, h);
193 l &= ~0x00008000;
194 wrmsr(MSR_K7_HWCR, l, h);
195 set_cpu_cap(c, X86_FEATURE_XMM);
196 }
197 }
198
199 /*
200 * It's been determined by AMD that Athlons since model 8 stepping 1
201 * are more robust with CLK_CTL set to 200xxxxx instead of 600xxxxx
202 * As per AMD technical note 27212 0.2
203 */
204 if ((c->x86_model == 8 && c->x86_mask >= 1) || (c->x86_model > 8)) {
205 rdmsr(MSR_K7_CLK_CTL, l, h);
206 if ((l & 0xfff00000) != 0x20000000) {
207 printk ("CPU: CLK_CTL MSR was %x. Reprogramming to %x\n", l,
208 ((l & 0x000fffff)|0x20000000));
209 wrmsr(MSR_K7_CLK_CTL, (l & 0x000fffff)|0x20000000, h);
210 }
211 }
212 break;
213 } 336 }
337 if (c->x86 == 0x10 || c->x86 == 0x11)
338 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
339#else
340
341 /*
342 * FIXME: We should handle the K5 here. Set up the write
343 * range and also turn on MSR 83 bits 4 and 31 (write alloc,
344 * no bus pipeline)
345 */
214 346
215 switch (c->x86) { 347 switch (c->x86) {
216 case 15: 348 case 4:
217 /* Use K8 tuning for Fam10h and Fam11h */ 349 init_amd_k5(c);
218 case 0x10:
219 case 0x11:
220 set_cpu_cap(c, X86_FEATURE_K8);
221 break; 350 break;
222 case 6: 351 case 5:
223 set_cpu_cap(c, X86_FEATURE_K7); 352 init_amd_k6(c);
353 break;
354 case 6: /* An Athlon/Duron */
355 init_amd_k7(c);
224 break; 356 break;
225 } 357 }
358
359 /* K6s reports MCEs but don't actually have all the MSRs */
360 if (c->x86 < 6)
361 clear_cpu_cap(c, X86_FEATURE_MCE);
362#endif
363
364 /* Enable workaround for FXSAVE leak */
226 if (c->x86 >= 6) 365 if (c->x86 >= 6)
227 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK); 366 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
228 367
229 display_cacheinfo(c); 368 if (!c->x86_model_id[0]) {
230 369 switch (c->x86) {
231 if (cpuid_eax(0x80000000) >= 0x80000008) 370 case 0xf:
232 c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1; 371 /* Should distinguish Models here, but this is only
372 a fallback anyways. */
373 strcpy(c->x86_model_id, "Hammer");
374 break;
375 }
376 }
233 377
234#ifdef CONFIG_X86_HT 378 display_cacheinfo(c);
235 /*
236 * On a AMD multi core setup the lower bits of the APIC id
237 * distinguish the cores.
238 */
239 if (c->x86_max_cores > 1) {
240 int cpu = smp_processor_id();
241 unsigned bits = (cpuid_ecx(0x80000008) >> 12) & 0xf;
242 379
243 if (bits == 0) { 380 /* Multi core CPU? */
244 while ((1 << bits) < c->x86_max_cores) 381 if (c->extended_cpuid_level >= 0x80000008) {
245 bits++; 382 amd_detect_cmp(c);
246 } 383 srat_detect_node(c);
247 c->cpu_core_id = c->phys_proc_id & ((1<<bits)-1);
248 c->phys_proc_id >>= bits;
249 printk(KERN_INFO "CPU %d(%d) -> Core %d\n",
250 cpu, c->x86_max_cores, c->cpu_core_id);
251 } 384 }
385
386#ifdef CONFIG_X86_32
387 detect_ht(c);
252#endif 388#endif
253 389
254 if (cpuid_eax(0x80000000) >= 0x80000006) { 390 if (c->extended_cpuid_level >= 0x80000006) {
255 if ((c->x86 == 0x10) && (cpuid_edx(0x80000006) & 0xf000)) 391 if ((c->x86 >= 0x0f) && (cpuid_edx(0x80000006) & 0xf000))
256 num_cache_leaves = 4; 392 num_cache_leaves = 4;
257 else 393 else
258 num_cache_leaves = 3; 394 num_cache_leaves = 3;
259 } 395 }
260 396
261 /* K6s reports MCEs but don't actually have all the MSRs */ 397 if (c->x86 >= 0xf && c->x86 <= 0x11)
262 if (c->x86 < 6) 398 set_cpu_cap(c, X86_FEATURE_K8);
263 clear_cpu_cap(c, X86_FEATURE_MCE);
264 399
265 if (cpu_has_xmm2) 400 if (cpu_has_xmm2) {
401 /* MFENCE stops RDTSC speculation */
266 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC); 402 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
403 }
404
405#ifdef CONFIG_X86_64
406 if (c->x86 == 0x10) {
407 /* do this for boot cpu */
408 if (c == &boot_cpu_data)
409 check_enable_amd_mmconf_dmi();
410
411 fam10h_check_enable_mmcfg();
412 }
413
414 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
415 unsigned long long tseg;
416
417 /*
418 * Split up direct mapping around the TSEG SMM area.
419 * Don't do it for gbpages because there seems very little
420 * benefit in doing so.
421 */
422 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg)) {
423 printk(KERN_DEBUG "tseg: %010llx\n", tseg);
424 if ((tseg>>PMD_SHIFT) <
425 (max_low_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) ||
426 ((tseg>>PMD_SHIFT) <
427 (max_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) &&
428 (tseg>>PMD_SHIFT) >= (1ULL<<(32 - PMD_SHIFT))))
429 set_memory_4k((unsigned long)__va(tseg), 1);
430 }
431 }
432#endif
267} 433}
268 434
435#ifdef CONFIG_X86_32
269static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int size) 436static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int size)
270{ 437{
271 /* AMD errata T13 (order #21922) */ 438 /* AMD errata T13 (order #21922) */
@@ -278,10 +445,12 @@ static unsigned int __cpuinit amd_size_cache(struct cpuinfo_x86 *c, unsigned int
278 } 445 }
279 return size; 446 return size;
280} 447}
448#endif
281 449
282static struct cpu_dev amd_cpu_dev __cpuinitdata = { 450static struct cpu_dev amd_cpu_dev __cpuinitdata = {
283 .c_vendor = "AMD", 451 .c_vendor = "AMD",
284 .c_ident = { "AuthenticAMD" }, 452 .c_ident = { "AuthenticAMD" },
453#ifdef CONFIG_X86_32
285 .c_models = { 454 .c_models = {
286 { .vendor = X86_VENDOR_AMD, .family = 4, .model_names = 455 { .vendor = X86_VENDOR_AMD, .family = 4, .model_names =
287 { 456 {
@@ -294,9 +463,11 @@ static struct cpu_dev amd_cpu_dev __cpuinitdata = {
294 } 463 }
295 }, 464 },
296 }, 465 },
466 .c_size_cache = amd_size_cache,
467#endif
297 .c_early_init = early_init_amd, 468 .c_early_init = early_init_amd,
298 .c_init = init_amd, 469 .c_init = init_amd,
299 .c_size_cache = amd_size_cache, 470 .c_x86_vendor = X86_VENDOR_AMD,
300}; 471};
301 472
302cpu_vendor_dev_register(X86_VENDOR_AMD, &amd_cpu_dev); 473cpu_dev_register(amd_cpu_dev);
diff --git a/arch/x86/kernel/cpu/amd_64.c b/arch/x86/kernel/cpu/amd_64.c
deleted file mode 100644
index d1692b2a41ff..000000000000
--- a/arch/x86/kernel/cpu/amd_64.c
+++ /dev/null
@@ -1,224 +0,0 @@
1#include <linux/init.h>
2#include <linux/mm.h>
3
4#include <asm/numa_64.h>
5#include <asm/mmconfig.h>
6#include <asm/cacheflush.h>
7
8#include <mach_apic.h>
9
10#include "cpu.h"
11
12int force_mwait __cpuinitdata;
13
14#ifdef CONFIG_NUMA
15static int __cpuinit nearby_node(int apicid)
16{
17 int i, node;
18
19 for (i = apicid - 1; i >= 0; i--) {
20 node = apicid_to_node[i];
21 if (node != NUMA_NO_NODE && node_online(node))
22 return node;
23 }
24 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
25 node = apicid_to_node[i];
26 if (node != NUMA_NO_NODE && node_online(node))
27 return node;
28 }
29 return first_node(node_online_map); /* Shouldn't happen */
30}
31#endif
32
33/*
34 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
35 * Assumes number of cores is a power of two.
36 */
37static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
38{
39#ifdef CONFIG_SMP
40 unsigned bits;
41#ifdef CONFIG_NUMA
42 int cpu = smp_processor_id();
43 int node = 0;
44 unsigned apicid = hard_smp_processor_id();
45#endif
46 bits = c->x86_coreid_bits;
47
48 /* Low order bits define the core id (index of core in socket) */
49 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
50 /* Convert the initial APIC ID into the socket ID */
51 c->phys_proc_id = c->initial_apicid >> bits;
52
53#ifdef CONFIG_NUMA
54 node = c->phys_proc_id;
55 if (apicid_to_node[apicid] != NUMA_NO_NODE)
56 node = apicid_to_node[apicid];
57 if (!node_online(node)) {
58 /* Two possibilities here:
59 - The CPU is missing memory and no node was created.
60 In that case try picking one from a nearby CPU
61 - The APIC IDs differ from the HyperTransport node IDs
62 which the K8 northbridge parsing fills in.
63 Assume they are all increased by a constant offset,
64 but in the same order as the HT nodeids.
65 If that doesn't result in a usable node fall back to the
66 path for the previous case. */
67
68 int ht_nodeid = c->initial_apicid;
69
70 if (ht_nodeid >= 0 &&
71 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
72 node = apicid_to_node[ht_nodeid];
73 /* Pick a nearby node */
74 if (!node_online(node))
75 node = nearby_node(apicid);
76 }
77 numa_set_node(cpu, node);
78
79 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
80#endif
81#endif
82}
83
84static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
85{
86#ifdef CONFIG_SMP
87 unsigned bits, ecx;
88
89 /* Multi core CPU? */
90 if (c->extended_cpuid_level < 0x80000008)
91 return;
92
93 ecx = cpuid_ecx(0x80000008);
94
95 c->x86_max_cores = (ecx & 0xff) + 1;
96
97 /* CPU telling us the core id bits shift? */
98 bits = (ecx >> 12) & 0xF;
99
100 /* Otherwise recompute */
101 if (bits == 0) {
102 while ((1 << bits) < c->x86_max_cores)
103 bits++;
104 }
105
106 c->x86_coreid_bits = bits;
107
108#endif
109}
110
111static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
112{
113 early_init_amd_mc(c);
114
115 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
116 if (c->x86_power & (1<<8))
117 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
118
119 set_cpu_cap(c, X86_FEATURE_SYSCALL32);
120}
121
122static void __cpuinit init_amd(struct cpuinfo_x86 *c)
123{
124 unsigned level;
125
126#ifdef CONFIG_SMP
127 unsigned long value;
128
129 /*
130 * Disable TLB flush filter by setting HWCR.FFDIS on K8
131 * bit 6 of msr C001_0015
132 *
133 * Errata 63 for SH-B3 steppings
134 * Errata 122 for all steppings (F+ have it disabled by default)
135 */
136 if (c->x86 == 0xf) {
137 rdmsrl(MSR_K8_HWCR, value);
138 value |= 1 << 6;
139 wrmsrl(MSR_K8_HWCR, value);
140 }
141#endif
142
143 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
144 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
145 clear_cpu_cap(c, 0*32+31);
146
147 /* On C+ stepping K8 rep microcode works well for copy/memset */
148 if (c->x86 == 0xf) {
149 level = cpuid_eax(1);
150 if((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
151 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
152 }
153 if (c->x86 == 0x10 || c->x86 == 0x11)
154 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
155
156 /* Enable workaround for FXSAVE leak */
157 if (c->x86 >= 6)
158 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
159
160 level = get_model_name(c);
161 if (!level) {
162 switch (c->x86) {
163 case 0xf:
164 /* Should distinguish Models here, but this is only
165 a fallback anyways. */
166 strcpy(c->x86_model_id, "Hammer");
167 break;
168 }
169 }
170 display_cacheinfo(c);
171
172 /* Multi core CPU? */
173 if (c->extended_cpuid_level >= 0x80000008)
174 amd_detect_cmp(c);
175
176 if (c->extended_cpuid_level >= 0x80000006 &&
177 (cpuid_edx(0x80000006) & 0xf000))
178 num_cache_leaves = 4;
179 else
180 num_cache_leaves = 3;
181
182 if (c->x86 >= 0xf && c->x86 <= 0x11)
183 set_cpu_cap(c, X86_FEATURE_K8);
184
185 /* MFENCE stops RDTSC speculation */
186 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
187
188 if (c->x86 == 0x10) {
189 /* do this for boot cpu */
190 if (c == &boot_cpu_data)
191 check_enable_amd_mmconf_dmi();
192
193 fam10h_check_enable_mmcfg();
194 }
195
196 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
197 unsigned long long tseg;
198
199 /*
200 * Split up direct mapping around the TSEG SMM area.
201 * Don't do it for gbpages because there seems very little
202 * benefit in doing so.
203 */
204 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg)) {
205 printk(KERN_DEBUG "tseg: %010llx\n", tseg);
206 if ((tseg>>PMD_SHIFT) <
207 (max_low_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) ||
208 ((tseg>>PMD_SHIFT) <
209 (max_pfn_mapped>>(PMD_SHIFT-PAGE_SHIFT)) &&
210 (tseg>>PMD_SHIFT) >= (1ULL<<(32 - PMD_SHIFT))))
211 set_memory_4k((unsigned long)__va(tseg), 1);
212 }
213 }
214}
215
216static struct cpu_dev amd_cpu_dev __cpuinitdata = {
217 .c_vendor = "AMD",
218 .c_ident = { "AuthenticAMD" },
219 .c_early_init = early_init_amd,
220 .c_init = init_amd,
221};
222
223cpu_vendor_dev_register(X86_VENDOR_AMD, &amd_cpu_dev);
224
diff --git a/arch/x86/kernel/cpu/centaur.c b/arch/x86/kernel/cpu/centaur.c
index e0f45edd6a55..89bfdd9cacc6 100644
--- a/arch/x86/kernel/cpu/centaur.c
+++ b/arch/x86/kernel/cpu/centaur.c
@@ -289,7 +289,6 @@ static void __cpuinit init_c3(struct cpuinfo_x86 *c)
289 if (c->x86_model >= 6 && c->x86_model < 9) 289 if (c->x86_model >= 6 && c->x86_model < 9)
290 set_cpu_cap(c, X86_FEATURE_3DNOW); 290 set_cpu_cap(c, X86_FEATURE_3DNOW);
291 291
292 get_model_name(c);
293 display_cacheinfo(c); 292 display_cacheinfo(c);
294} 293}
295 294
@@ -314,6 +313,16 @@ enum {
314 EAMD3D = 1<<20, 313 EAMD3D = 1<<20,
315}; 314};
316 315
316static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
317{
318 switch (c->x86) {
319 case 5:
320 /* Emulate MTRRs using Centaur's MCR. */
321 set_cpu_cap(c, X86_FEATURE_CENTAUR_MCR);
322 break;
323 }
324}
325
317static void __cpuinit init_centaur(struct cpuinfo_x86 *c) 326static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
318{ 327{
319 328
@@ -462,8 +471,10 @@ centaur_size_cache(struct cpuinfo_x86 *c, unsigned int size)
462static struct cpu_dev centaur_cpu_dev __cpuinitdata = { 471static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
463 .c_vendor = "Centaur", 472 .c_vendor = "Centaur",
464 .c_ident = { "CentaurHauls" }, 473 .c_ident = { "CentaurHauls" },
474 .c_early_init = early_init_centaur,
465 .c_init = init_centaur, 475 .c_init = init_centaur,
466 .c_size_cache = centaur_size_cache, 476 .c_size_cache = centaur_size_cache,
477 .c_x86_vendor = X86_VENDOR_CENTAUR,
467}; 478};
468 479
469cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev); 480cpu_dev_register(centaur_cpu_dev);
diff --git a/arch/x86/kernel/cpu/centaur_64.c b/arch/x86/kernel/cpu/centaur_64.c
index 1d181c40e2e1..a1625f5a1e78 100644
--- a/arch/x86/kernel/cpu/centaur_64.c
+++ b/arch/x86/kernel/cpu/centaur_64.c
@@ -16,9 +16,10 @@ static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
16 16
17static void __cpuinit init_centaur(struct cpuinfo_x86 *c) 17static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
18{ 18{
19 early_init_centaur(c);
20
19 if (c->x86 == 0x6 && c->x86_model >= 0xf) { 21 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
20 c->x86_cache_alignment = c->x86_clflush_size * 2; 22 c->x86_cache_alignment = c->x86_clflush_size * 2;
21 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
22 set_cpu_cap(c, X86_FEATURE_REP_GOOD); 23 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
23 } 24 }
24 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC); 25 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
@@ -29,7 +30,8 @@ static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
29 .c_ident = { "CentaurHauls" }, 30 .c_ident = { "CentaurHauls" },
30 .c_early_init = early_init_centaur, 31 .c_early_init = early_init_centaur,
31 .c_init = init_centaur, 32 .c_init = init_centaur,
33 .c_x86_vendor = X86_VENDOR_CENTAUR,
32}; 34};
33 35
34cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev); 36cpu_dev_register(centaur_cpu_dev);
35 37
diff --git a/arch/x86/kernel/cpu/cmpxchg.c b/arch/x86/kernel/cpu/cmpxchg.c
new file mode 100644
index 000000000000..2056ccf572cc
--- /dev/null
+++ b/arch/x86/kernel/cpu/cmpxchg.c
@@ -0,0 +1,72 @@
1/*
2 * cmpxchg*() fallbacks for CPU not supporting these instructions
3 */
4
5#include <linux/kernel.h>
6#include <linux/smp.h>
7#include <linux/module.h>
8
9#ifndef CONFIG_X86_CMPXCHG
10unsigned long cmpxchg_386_u8(volatile void *ptr, u8 old, u8 new)
11{
12 u8 prev;
13 unsigned long flags;
14
15 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
16 local_irq_save(flags);
17 prev = *(u8 *)ptr;
18 if (prev == old)
19 *(u8 *)ptr = new;
20 local_irq_restore(flags);
21 return prev;
22}
23EXPORT_SYMBOL(cmpxchg_386_u8);
24
25unsigned long cmpxchg_386_u16(volatile void *ptr, u16 old, u16 new)
26{
27 u16 prev;
28 unsigned long flags;
29
30 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
31 local_irq_save(flags);
32 prev = *(u16 *)ptr;
33 if (prev == old)
34 *(u16 *)ptr = new;
35 local_irq_restore(flags);
36 return prev;
37}
38EXPORT_SYMBOL(cmpxchg_386_u16);
39
40unsigned long cmpxchg_386_u32(volatile void *ptr, u32 old, u32 new)
41{
42 u32 prev;
43 unsigned long flags;
44
45 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
46 local_irq_save(flags);
47 prev = *(u32 *)ptr;
48 if (prev == old)
49 *(u32 *)ptr = new;
50 local_irq_restore(flags);
51 return prev;
52}
53EXPORT_SYMBOL(cmpxchg_386_u32);
54#endif
55
56#ifndef CONFIG_X86_CMPXCHG64
57unsigned long long cmpxchg_486_u64(volatile void *ptr, u64 old, u64 new)
58{
59 u64 prev;
60 unsigned long flags;
61
62 /* Poor man's cmpxchg8b for 386 and 486. Unsuitable for SMP */
63 local_irq_save(flags);
64 prev = *(u64 *)ptr;
65 if (prev == old)
66 *(u64 *)ptr = new;
67 local_irq_restore(flags);
68 return prev;
69}
70EXPORT_SYMBOL(cmpxchg_486_u64);
71#endif
72
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 80ab20d4fa39..25581dcb280e 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1,27 +1,62 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/kernel.h>
3#include <linux/sched.h>
2#include <linux/string.h> 4#include <linux/string.h>
5#include <linux/bootmem.h>
6#include <linux/bitops.h>
7#include <linux/module.h>
8#include <linux/kgdb.h>
9#include <linux/topology.h>
3#include <linux/delay.h> 10#include <linux/delay.h>
4#include <linux/smp.h> 11#include <linux/smp.h>
5#include <linux/module.h>
6#include <linux/percpu.h> 12#include <linux/percpu.h>
7#include <linux/bootmem.h>
8#include <asm/processor.h>
9#include <asm/i387.h> 13#include <asm/i387.h>
10#include <asm/msr.h> 14#include <asm/msr.h>
11#include <asm/io.h> 15#include <asm/io.h>
16#include <asm/linkage.h>
12#include <asm/mmu_context.h> 17#include <asm/mmu_context.h>
13#include <asm/mtrr.h> 18#include <asm/mtrr.h>
14#include <asm/mce.h> 19#include <asm/mce.h>
15#include <asm/pat.h> 20#include <asm/pat.h>
21#include <asm/asm.h>
22#include <asm/numa.h>
16#ifdef CONFIG_X86_LOCAL_APIC 23#ifdef CONFIG_X86_LOCAL_APIC
17#include <asm/mpspec.h> 24#include <asm/mpspec.h>
18#include <asm/apic.h> 25#include <asm/apic.h>
19#include <mach_apic.h> 26#include <mach_apic.h>
27#include <asm/genapic.h>
20#endif 28#endif
21 29
30#include <asm/pda.h>
31#include <asm/pgtable.h>
32#include <asm/processor.h>
33#include <asm/desc.h>
34#include <asm/atomic.h>
35#include <asm/proto.h>
36#include <asm/sections.h>
37#include <asm/setup.h>
38
22#include "cpu.h" 39#include "cpu.h"
23 40
41static struct cpu_dev *this_cpu __cpuinitdata;
42
43#ifdef CONFIG_X86_64
44/* We need valid kernel segments for data and code in long mode too
45 * IRET will check the segment types kkeil 2000/10/28
46 * Also sysret mandates a special GDT layout
47 */
48/* The TLS descriptors are currently at a different place compared to i386.
49 Hopefully nobody expects them at a fixed place (Wine?) */
24DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = { 50DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
51 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
52 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
53 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
54 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
55 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
56 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
57} };
58#else
59DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
25 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } }, 60 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
26 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } }, 61 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
27 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } }, 62 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
@@ -55,17 +90,157 @@ DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
55 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } }, 90 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
56 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } }, 91 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
57} }; 92} };
93#endif
58EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); 94EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
59 95
60__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata; 96#ifdef CONFIG_X86_32
61
62static int cachesize_override __cpuinitdata = -1; 97static int cachesize_override __cpuinitdata = -1;
63static int disable_x86_serial_nr __cpuinitdata = 1; 98static int disable_x86_serial_nr __cpuinitdata = 1;
64 99
65struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; 100static int __init cachesize_setup(char *str)
101{
102 get_option(&str, &cachesize_override);
103 return 1;
104}
105__setup("cachesize=", cachesize_setup);
106
107static int __init x86_fxsr_setup(char *s)
108{
109 setup_clear_cpu_cap(X86_FEATURE_FXSR);
110 setup_clear_cpu_cap(X86_FEATURE_XMM);
111 return 1;
112}
113__setup("nofxsr", x86_fxsr_setup);
114
115static int __init x86_sep_setup(char *s)
116{
117 setup_clear_cpu_cap(X86_FEATURE_SEP);
118 return 1;
119}
120__setup("nosep", x86_sep_setup);
121
122/* Standard macro to see if a specific flag is changeable */
123static inline int flag_is_changeable_p(u32 flag)
124{
125 u32 f1, f2;
126
127 /*
128 * Cyrix and IDT cpus allow disabling of CPUID
129 * so the code below may return different results
130 * when it is executed before and after enabling
131 * the CPUID. Add "volatile" to not allow gcc to
132 * optimize the subsequent calls to this function.
133 */
134 asm volatile ("pushfl\n\t"
135 "pushfl\n\t"
136 "popl %0\n\t"
137 "movl %0,%1\n\t"
138 "xorl %2,%0\n\t"
139 "pushl %0\n\t"
140 "popfl\n\t"
141 "pushfl\n\t"
142 "popl %0\n\t"
143 "popfl\n\t"
144 : "=&r" (f1), "=&r" (f2)
145 : "ir" (flag));
146
147 return ((f1^f2) & flag) != 0;
148}
149
150/* Probe for the CPUID instruction */
151static int __cpuinit have_cpuid_p(void)
152{
153 return flag_is_changeable_p(X86_EFLAGS_ID);
154}
155
156static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
157{
158 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
159 /* Disable processor serial number */
160 unsigned long lo, hi;
161 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
162 lo |= 0x200000;
163 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
164 printk(KERN_NOTICE "CPU serial number disabled.\n");
165 clear_cpu_cap(c, X86_FEATURE_PN);
166
167 /* Disabling the serial number may affect the cpuid level */
168 c->cpuid_level = cpuid_eax(0);
169 }
170}
171
172static int __init x86_serial_nr_setup(char *s)
173{
174 disable_x86_serial_nr = 0;
175 return 1;
176}
177__setup("serialnumber", x86_serial_nr_setup);
178#else
179static inline int flag_is_changeable_p(u32 flag)
180{
181 return 1;
182}
183/* Probe for the CPUID instruction */
184static inline int have_cpuid_p(void)
185{
186 return 1;
187}
188static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
189{
190}
191#endif
192
193/*
194 * Naming convention should be: <Name> [(<Codename>)]
195 * This table only is used unless init_<vendor>() below doesn't set it;
196 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
197 *
198 */
199
200/* Look up CPU names by table lookup. */
201static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
202{
203 struct cpu_model_info *info;
204
205 if (c->x86_model >= 16)
206 return NULL; /* Range check */
207
208 if (!this_cpu)
209 return NULL;
210
211 info = this_cpu->c_models;
212
213 while (info && info->family) {
214 if (info->family == c->x86)
215 return info->model_names[c->x86_model];
216 info++;
217 }
218 return NULL; /* Not found */
219}
220
221__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
222
223/* Current gdt points %fs at the "master" per-cpu area: after this,
224 * it's on the real one. */
225void switch_to_new_gdt(void)
226{
227 struct desc_ptr gdt_descr;
228
229 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
230 gdt_descr.size = GDT_SIZE - 1;
231 load_gdt(&gdt_descr);
232#ifdef CONFIG_X86_32
233 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
234#endif
235}
236
237static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
66 238
67static void __cpuinit default_init(struct cpuinfo_x86 *c) 239static void __cpuinit default_init(struct cpuinfo_x86 *c)
68{ 240{
241#ifdef CONFIG_X86_64
242 display_cacheinfo(c);
243#else
69 /* Not much we can do here... */ 244 /* Not much we can do here... */
70 /* Check if at least it has cpuid */ 245 /* Check if at least it has cpuid */
71 if (c->cpuid_level == -1) { 246 if (c->cpuid_level == -1) {
@@ -75,28 +250,22 @@ static void __cpuinit default_init(struct cpuinfo_x86 *c)
75 else if (c->x86 == 3) 250 else if (c->x86 == 3)
76 strcpy(c->x86_model_id, "386"); 251 strcpy(c->x86_model_id, "386");
77 } 252 }
253#endif
78} 254}
79 255
80static struct cpu_dev __cpuinitdata default_cpu = { 256static struct cpu_dev __cpuinitdata default_cpu = {
81 .c_init = default_init, 257 .c_init = default_init,
82 .c_vendor = "Unknown", 258 .c_vendor = "Unknown",
259 .c_x86_vendor = X86_VENDOR_UNKNOWN,
83}; 260};
84static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
85
86static int __init cachesize_setup(char *str)
87{
88 get_option(&str, &cachesize_override);
89 return 1;
90}
91__setup("cachesize=", cachesize_setup);
92 261
93int __cpuinit get_model_name(struct cpuinfo_x86 *c) 262static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
94{ 263{
95 unsigned int *v; 264 unsigned int *v;
96 char *p, *q; 265 char *p, *q;
97 266
98 if (cpuid_eax(0x80000000) < 0x80000004) 267 if (c->extended_cpuid_level < 0x80000004)
99 return 0; 268 return;
100 269
101 v = (unsigned int *) c->x86_model_id; 270 v = (unsigned int *) c->x86_model_id;
102 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); 271 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
@@ -115,30 +284,34 @@ int __cpuinit get_model_name(struct cpuinfo_x86 *c)
115 while (q <= &c->x86_model_id[48]) 284 while (q <= &c->x86_model_id[48])
116 *q++ = '\0'; /* Zero-pad the rest */ 285 *q++ = '\0'; /* Zero-pad the rest */
117 } 286 }
118
119 return 1;
120} 287}
121 288
122
123void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c) 289void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
124{ 290{
125 unsigned int n, dummy, ecx, edx, l2size; 291 unsigned int n, dummy, ebx, ecx, edx, l2size;
126 292
127 n = cpuid_eax(0x80000000); 293 n = c->extended_cpuid_level;
128 294
129 if (n >= 0x80000005) { 295 if (n >= 0x80000005) {
130 cpuid(0x80000005, &dummy, &dummy, &ecx, &edx); 296 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
131 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n", 297 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
132 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF); 298 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
133 c->x86_cache_size = (ecx>>24)+(edx>>24); 299 c->x86_cache_size = (ecx>>24) + (edx>>24);
300#ifdef CONFIG_X86_64
301 /* On K8 L1 TLB is inclusive, so don't count it */
302 c->x86_tlbsize = 0;
303#endif
134 } 304 }
135 305
136 if (n < 0x80000006) /* Some chips just has a large L1. */ 306 if (n < 0x80000006) /* Some chips just has a large L1. */
137 return; 307 return;
138 308
139 ecx = cpuid_ecx(0x80000006); 309 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
140 l2size = ecx >> 16; 310 l2size = ecx >> 16;
141 311
312#ifdef CONFIG_X86_64
313 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
314#else
142 /* do processor-specific cache resizing */ 315 /* do processor-specific cache resizing */
143 if (this_cpu->c_size_cache) 316 if (this_cpu->c_size_cache)
144 l2size = this_cpu->c_size_cache(c, l2size); 317 l2size = this_cpu->c_size_cache(c, l2size);
@@ -149,116 +322,106 @@ void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
149 322
150 if (l2size == 0) 323 if (l2size == 0)
151 return; /* Again, no L2 cache is possible */ 324 return; /* Again, no L2 cache is possible */
325#endif
152 326
153 c->x86_cache_size = l2size; 327 c->x86_cache_size = l2size;
154 328
155 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n", 329 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
156 l2size, ecx & 0xFF); 330 l2size, ecx & 0xFF);
157} 331}
158 332
159/* 333void __cpuinit detect_ht(struct cpuinfo_x86 *c)
160 * Naming convention should be: <Name> [(<Codename>)]
161 * This table only is used unless init_<vendor>() below doesn't set it;
162 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
163 *
164 */
165
166/* Look up CPU names by table lookup. */
167static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
168{ 334{
169 struct cpu_model_info *info; 335#ifdef CONFIG_X86_HT
336 u32 eax, ebx, ecx, edx;
337 int index_msb, core_bits;
170 338
171 if (c->x86_model >= 16) 339 if (!cpu_has(c, X86_FEATURE_HT))
172 return NULL; /* Range check */ 340 return;
173 341
174 if (!this_cpu) 342 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
175 return NULL; 343 goto out;
176 344
177 info = this_cpu->c_models; 345 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
346 return;
178 347
179 while (info && info->family) { 348 cpuid(1, &eax, &ebx, &ecx, &edx);
180 if (info->family == c->x86) 349
181 return info->model_names[c->x86_model]; 350 smp_num_siblings = (ebx & 0xff0000) >> 16;
182 info++; 351
352 if (smp_num_siblings == 1) {
353 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
354 } else if (smp_num_siblings > 1) {
355
356 if (smp_num_siblings > NR_CPUS) {
357 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
358 smp_num_siblings);
359 smp_num_siblings = 1;
360 return;
361 }
362
363 index_msb = get_count_order(smp_num_siblings);
364#ifdef CONFIG_X86_64
365 c->phys_proc_id = phys_pkg_id(index_msb);
366#else
367 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
368#endif
369
370 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
371
372 index_msb = get_count_order(smp_num_siblings);
373
374 core_bits = get_count_order(c->x86_max_cores);
375
376#ifdef CONFIG_X86_64
377 c->cpu_core_id = phys_pkg_id(index_msb) &
378 ((1 << core_bits) - 1);
379#else
380 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
381 ((1 << core_bits) - 1);
382#endif
183 } 383 }
184 return NULL; /* Not found */
185}
186 384
385out:
386 if ((c->x86_max_cores * smp_num_siblings) > 1) {
387 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
388 c->phys_proc_id);
389 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
390 c->cpu_core_id);
391 }
392#endif
393}
187 394
188static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c, int early) 395static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
189{ 396{
190 char *v = c->x86_vendor_id; 397 char *v = c->x86_vendor_id;
191 int i; 398 int i;
192 static int printed; 399 static int printed;
193 400
194 for (i = 0; i < X86_VENDOR_NUM; i++) { 401 for (i = 0; i < X86_VENDOR_NUM; i++) {
195 if (cpu_devs[i]) { 402 if (!cpu_devs[i])
196 if (!strcmp(v, cpu_devs[i]->c_ident[0]) || 403 break;
197 (cpu_devs[i]->c_ident[1] && 404
198 !strcmp(v, cpu_devs[i]->c_ident[1]))) { 405 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
199 c->x86_vendor = i; 406 (cpu_devs[i]->c_ident[1] &&
200 if (!early) 407 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
201 this_cpu = cpu_devs[i]; 408 this_cpu = cpu_devs[i];
202 return; 409 c->x86_vendor = this_cpu->c_x86_vendor;
203 } 410 return;
204 } 411 }
205 } 412 }
413
206 if (!printed) { 414 if (!printed) {
207 printed++; 415 printed++;
208 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n"); 416 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
209 printk(KERN_ERR "CPU: Your system may be unstable.\n"); 417 printk(KERN_ERR "CPU: Your system may be unstable.\n");
210 } 418 }
419
211 c->x86_vendor = X86_VENDOR_UNKNOWN; 420 c->x86_vendor = X86_VENDOR_UNKNOWN;
212 this_cpu = &default_cpu; 421 this_cpu = &default_cpu;
213} 422}
214 423
215 424void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
216static int __init x86_fxsr_setup(char *s)
217{
218 setup_clear_cpu_cap(X86_FEATURE_FXSR);
219 setup_clear_cpu_cap(X86_FEATURE_XMM);
220 return 1;
221}
222__setup("nofxsr", x86_fxsr_setup);
223
224
225static int __init x86_sep_setup(char *s)
226{
227 setup_clear_cpu_cap(X86_FEATURE_SEP);
228 return 1;
229}
230__setup("nosep", x86_sep_setup);
231
232
233/* Standard macro to see if a specific flag is changeable */
234static inline int flag_is_changeable_p(u32 flag)
235{
236 u32 f1, f2;
237
238 asm("pushfl\n\t"
239 "pushfl\n\t"
240 "popl %0\n\t"
241 "movl %0,%1\n\t"
242 "xorl %2,%0\n\t"
243 "pushl %0\n\t"
244 "popfl\n\t"
245 "pushfl\n\t"
246 "popl %0\n\t"
247 "popfl\n\t"
248 : "=&r" (f1), "=&r" (f2)
249 : "ir" (flag));
250
251 return ((f1^f2) & flag) != 0;
252}
253
254
255/* Probe for the CPUID instruction */
256static int __cpuinit have_cpuid_p(void)
257{
258 return flag_is_changeable_p(X86_EFLAGS_ID);
259}
260
261void __init cpu_detect(struct cpuinfo_x86 *c)
262{ 425{
263 /* Get vendor name */ 426 /* Get vendor name */
264 cpuid(0x00000000, (unsigned int *)&c->cpuid_level, 427 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
@@ -267,50 +430,87 @@ void __init cpu_detect(struct cpuinfo_x86 *c)
267 (unsigned int *)&c->x86_vendor_id[4]); 430 (unsigned int *)&c->x86_vendor_id[4]);
268 431
269 c->x86 = 4; 432 c->x86 = 4;
433 /* Intel-defined flags: level 0x00000001 */
270 if (c->cpuid_level >= 0x00000001) { 434 if (c->cpuid_level >= 0x00000001) {
271 u32 junk, tfms, cap0, misc; 435 u32 junk, tfms, cap0, misc;
272 cpuid(0x00000001, &tfms, &misc, &junk, &cap0); 436 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
273 c->x86 = (tfms >> 8) & 15; 437 c->x86 = (tfms >> 8) & 0xf;
274 c->x86_model = (tfms >> 4) & 15; 438 c->x86_model = (tfms >> 4) & 0xf;
439 c->x86_mask = tfms & 0xf;
275 if (c->x86 == 0xf) 440 if (c->x86 == 0xf)
276 c->x86 += (tfms >> 20) & 0xff; 441 c->x86 += (tfms >> 20) & 0xff;
277 if (c->x86 >= 0x6) 442 if (c->x86 >= 0x6)
278 c->x86_model += ((tfms >> 16) & 0xF) << 4; 443 c->x86_model += ((tfms >> 16) & 0xf) << 4;
279 c->x86_mask = tfms & 15;
280 if (cap0 & (1<<19)) { 444 if (cap0 & (1<<19)) {
281 c->x86_cache_alignment = ((misc >> 8) & 0xff) * 8;
282 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; 445 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
446 c->x86_cache_alignment = c->x86_clflush_size;
283 } 447 }
284 } 448 }
285} 449}
286static void __cpuinit early_get_cap(struct cpuinfo_x86 *c) 450
451static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
287{ 452{
288 u32 tfms, xlvl; 453 u32 tfms, xlvl;
289 unsigned int ebx; 454 u32 ebx;
290 455
291 memset(&c->x86_capability, 0, sizeof c->x86_capability); 456 /* Intel-defined flags: level 0x00000001 */
292 if (have_cpuid_p()) { 457 if (c->cpuid_level >= 0x00000001) {
293 /* Intel-defined flags: level 0x00000001 */ 458 u32 capability, excap;
294 if (c->cpuid_level >= 0x00000001) { 459 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
295 u32 capability, excap; 460 c->x86_capability[0] = capability;
296 cpuid(0x00000001, &tfms, &ebx, &excap, &capability); 461 c->x86_capability[4] = excap;
297 c->x86_capability[0] = capability; 462 }
298 c->x86_capability[4] = excap;
299 }
300 463
301 /* AMD-defined flags: level 0x80000001 */ 464 /* AMD-defined flags: level 0x80000001 */
302 xlvl = cpuid_eax(0x80000000); 465 xlvl = cpuid_eax(0x80000000);
303 if ((xlvl & 0xffff0000) == 0x80000000) { 466 c->extended_cpuid_level = xlvl;
304 if (xlvl >= 0x80000001) { 467 if ((xlvl & 0xffff0000) == 0x80000000) {
305 c->x86_capability[1] = cpuid_edx(0x80000001); 468 if (xlvl >= 0x80000001) {
306 c->x86_capability[6] = cpuid_ecx(0x80000001); 469 c->x86_capability[1] = cpuid_edx(0x80000001);
307 } 470 c->x86_capability[6] = cpuid_ecx(0x80000001);
308 } 471 }
472 }
473
474#ifdef CONFIG_X86_64
475 if (c->extended_cpuid_level >= 0x80000008) {
476 u32 eax = cpuid_eax(0x80000008);
309 477
478 c->x86_virt_bits = (eax >> 8) & 0xff;
479 c->x86_phys_bits = eax & 0xff;
310 } 480 }
481#endif
482
483 if (c->extended_cpuid_level >= 0x80000007)
484 c->x86_power = cpuid_edx(0x80000007);
311 485
312} 486}
313 487
488static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
489{
490#ifdef CONFIG_X86_32
491 int i;
492
493 /*
494 * First of all, decide if this is a 486 or higher
495 * It's a 486 if we can modify the AC flag
496 */
497 if (flag_is_changeable_p(X86_EFLAGS_AC))
498 c->x86 = 4;
499 else
500 c->x86 = 3;
501
502 for (i = 0; i < X86_VENDOR_NUM; i++)
503 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
504 c->x86_vendor_id[0] = 0;
505 cpu_devs[i]->c_identify(c);
506 if (c->x86_vendor_id[0]) {
507 get_cpu_vendor(c);
508 break;
509 }
510 }
511#endif
512}
513
314/* 514/*
315 * Do minimum CPU detection early. 515 * Do minimum CPU detection early.
316 * Fields really needed: vendor, cpuid_level, family, model, mask, 516 * Fields really needed: vendor, cpuid_level, family, model, mask,
@@ -320,109 +520,113 @@ static void __cpuinit early_get_cap(struct cpuinfo_x86 *c)
320 * WARNING: this function is only called on the BP. Don't add code here 520 * WARNING: this function is only called on the BP. Don't add code here
321 * that is supposed to run on all CPUs. 521 * that is supposed to run on all CPUs.
322 */ 522 */
323static void __init early_cpu_detect(void) 523static void __init early_identify_cpu(struct cpuinfo_x86 *c)
324{ 524{
325 struct cpuinfo_x86 *c = &boot_cpu_data; 525#ifdef CONFIG_X86_64
326 526 c->x86_clflush_size = 64;
327 c->x86_cache_alignment = 32; 527#else
328 c->x86_clflush_size = 32; 528 c->x86_clflush_size = 32;
529#endif
530 c->x86_cache_alignment = c->x86_clflush_size;
531
532 memset(&c->x86_capability, 0, sizeof c->x86_capability);
533 c->extended_cpuid_level = 0;
534
535 if (!have_cpuid_p())
536 identify_cpu_without_cpuid(c);
329 537
538 /* cyrix could have cpuid enabled via c_identify()*/
330 if (!have_cpuid_p()) 539 if (!have_cpuid_p())
331 return; 540 return;
332 541
333 cpu_detect(c); 542 cpu_detect(c);
334 543
335 get_cpu_vendor(c, 1); 544 get_cpu_vendor(c);
545
546 get_cpu_cap(c);
336 547
337 if (c->x86_vendor != X86_VENDOR_UNKNOWN && 548 if (this_cpu->c_early_init)
338 cpu_devs[c->x86_vendor]->c_early_init) 549 this_cpu->c_early_init(c);
339 cpu_devs[c->x86_vendor]->c_early_init(c);
340 550
341 early_get_cap(c); 551 validate_pat_support(c);
342} 552}
343 553
344static void __cpuinit generic_identify(struct cpuinfo_x86 *c) 554void __init early_cpu_init(void)
345{ 555{
346 u32 tfms, xlvl; 556 struct cpu_dev **cdev;
347 unsigned int ebx; 557 int count = 0;
348 558
349 if (have_cpuid_p()) { 559 printk("KERNEL supported cpus:\n");
350 /* Get vendor name */ 560 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
351 cpuid(0x00000000, (unsigned int *)&c->cpuid_level, 561 struct cpu_dev *cpudev = *cdev;
352 (unsigned int *)&c->x86_vendor_id[0], 562 unsigned int j;
353 (unsigned int *)&c->x86_vendor_id[8], 563
354 (unsigned int *)&c->x86_vendor_id[4]); 564 if (count >= X86_VENDOR_NUM)
355 565 break;
356 get_cpu_vendor(c, 0); 566 cpu_devs[count] = cpudev;
357 /* Initialize the standard set of capabilities */ 567 count++;
358 /* Note that the vendor-specific code below might override */ 568
359 /* Intel-defined flags: level 0x00000001 */ 569 for (j = 0; j < 2; j++) {
360 if (c->cpuid_level >= 0x00000001) { 570 if (!cpudev->c_ident[j])
361 u32 capability, excap; 571 continue;
362 cpuid(0x00000001, &tfms, &ebx, &excap, &capability); 572 printk(" %s %s\n", cpudev->c_vendor,
363 c->x86_capability[0] = capability; 573 cpudev->c_ident[j]);
364 c->x86_capability[4] = excap;
365 c->x86 = (tfms >> 8) & 15;
366 c->x86_model = (tfms >> 4) & 15;
367 if (c->x86 == 0xf)
368 c->x86 += (tfms >> 20) & 0xff;
369 if (c->x86 >= 0x6)
370 c->x86_model += ((tfms >> 16) & 0xF) << 4;
371 c->x86_mask = tfms & 15;
372 c->initial_apicid = (ebx >> 24) & 0xFF;
373#ifdef CONFIG_X86_HT
374 c->apicid = phys_pkg_id(c->initial_apicid, 0);
375 c->phys_proc_id = c->initial_apicid;
376#else
377 c->apicid = c->initial_apicid;
378#endif
379 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
380 c->x86_clflush_size = ((ebx >> 8) & 0xff) * 8;
381 } else {
382 /* Have CPUID level 0 only - unheard of */
383 c->x86 = 4;
384 }
385
386 /* AMD-defined flags: level 0x80000001 */
387 xlvl = cpuid_eax(0x80000000);
388 if ((xlvl & 0xffff0000) == 0x80000000) {
389 if (xlvl >= 0x80000001) {
390 c->x86_capability[1] = cpuid_edx(0x80000001);
391 c->x86_capability[6] = cpuid_ecx(0x80000001);
392 }
393 if (xlvl >= 0x80000004)
394 get_model_name(c); /* Default name */
395 } 574 }
396
397 init_scattered_cpuid_features(c);
398 } 575 }
399 576
577 early_identify_cpu(&boot_cpu_data);
400} 578}
401 579
402static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 580/*
581 * The NOPL instruction is supposed to exist on all CPUs with
582 * family >= 6; unfortunately, that's not true in practice because
583 * of early VIA chips and (more importantly) broken virtualizers that
584 * are not easy to detect. In the latter case it doesn't even *fail*
585 * reliably, so probing for it doesn't even work. Disable it completely
586 * unless we can find a reliable way to detect all the broken cases.
587 */
588static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
403{ 589{
404 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) { 590 clear_cpu_cap(c, X86_FEATURE_NOPL);
405 /* Disable processor serial number */
406 unsigned long lo, hi;
407 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
408 lo |= 0x200000;
409 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
410 printk(KERN_NOTICE "CPU serial number disabled.\n");
411 clear_cpu_cap(c, X86_FEATURE_PN);
412
413 /* Disabling the serial number may affect the cpuid level */
414 c->cpuid_level = cpuid_eax(0);
415 }
416} 591}
417 592
418static int __init x86_serial_nr_setup(char *s) 593static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
419{ 594{
420 disable_x86_serial_nr = 0; 595 c->extended_cpuid_level = 0;
421 return 1; 596
422} 597 if (!have_cpuid_p())
423__setup("serialnumber", x86_serial_nr_setup); 598 identify_cpu_without_cpuid(c);
599
600 /* cyrix could have cpuid enabled via c_identify()*/
601 if (!have_cpuid_p())
602 return;
603
604 cpu_detect(c);
605
606 get_cpu_vendor(c);
424 607
608 get_cpu_cap(c);
425 609
610 if (c->cpuid_level >= 0x00000001) {
611 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
612#ifdef CONFIG_X86_32
613# ifdef CONFIG_X86_HT
614 c->apicid = phys_pkg_id(c->initial_apicid, 0);
615# else
616 c->apicid = c->initial_apicid;
617# endif
618#endif
619
620#ifdef CONFIG_X86_HT
621 c->phys_proc_id = c->initial_apicid;
622#endif
623 }
624
625 get_model_name(c); /* Default name */
626
627 init_scattered_cpuid_features(c);
628 detect_nopl(c);
629}
426 630
427/* 631/*
428 * This does the hard work of actually picking apart the CPU stuff... 632 * This does the hard work of actually picking apart the CPU stuff...
@@ -434,30 +638,29 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
434 c->loops_per_jiffy = loops_per_jiffy; 638 c->loops_per_jiffy = loops_per_jiffy;
435 c->x86_cache_size = -1; 639 c->x86_cache_size = -1;
436 c->x86_vendor = X86_VENDOR_UNKNOWN; 640 c->x86_vendor = X86_VENDOR_UNKNOWN;
437 c->cpuid_level = -1; /* CPUID not detected */
438 c->x86_model = c->x86_mask = 0; /* So far unknown... */ 641 c->x86_model = c->x86_mask = 0; /* So far unknown... */
439 c->x86_vendor_id[0] = '\0'; /* Unset */ 642 c->x86_vendor_id[0] = '\0'; /* Unset */
440 c->x86_model_id[0] = '\0'; /* Unset */ 643 c->x86_model_id[0] = '\0'; /* Unset */
441 c->x86_max_cores = 1; 644 c->x86_max_cores = 1;
645 c->x86_coreid_bits = 0;
646#ifdef CONFIG_X86_64
647 c->x86_clflush_size = 64;
648#else
649 c->cpuid_level = -1; /* CPUID not detected */
442 c->x86_clflush_size = 32; 650 c->x86_clflush_size = 32;
651#endif
652 c->x86_cache_alignment = c->x86_clflush_size;
443 memset(&c->x86_capability, 0, sizeof c->x86_capability); 653 memset(&c->x86_capability, 0, sizeof c->x86_capability);
444 654
445 if (!have_cpuid_p()) {
446 /*
447 * First of all, decide if this is a 486 or higher
448 * It's a 486 if we can modify the AC flag
449 */
450 if (flag_is_changeable_p(X86_EFLAGS_AC))
451 c->x86 = 4;
452 else
453 c->x86 = 3;
454 }
455
456 generic_identify(c); 655 generic_identify(c);
457 656
458 if (this_cpu->c_identify) 657 if (this_cpu->c_identify)
459 this_cpu->c_identify(c); 658 this_cpu->c_identify(c);
460 659
660#ifdef CONFIG_X86_64
661 c->apicid = phys_pkg_id(0);
662#endif
663
461 /* 664 /*
462 * Vendor-specific initialization. In this section we 665 * Vendor-specific initialization. In this section we
463 * canonicalize the feature flags, meaning if there are 666 * canonicalize the feature flags, meaning if there are
@@ -491,6 +694,10 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
491 c->x86, c->x86_model); 694 c->x86, c->x86_model);
492 } 695 }
493 696
697#ifdef CONFIG_X86_64
698 detect_ht(c);
699#endif
700
494 /* 701 /*
495 * On SMP, boot_cpu_data holds the common feature set between 702 * On SMP, boot_cpu_data holds the common feature set between
496 * all CPUs; so make sure that we indicate which features are 703 * all CPUs; so make sure that we indicate which features are
@@ -499,7 +706,7 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
499 */ 706 */
500 if (c != &boot_cpu_data) { 707 if (c != &boot_cpu_data) {
501 /* AND the already accumulated flags with these */ 708 /* AND the already accumulated flags with these */
502 for (i = 0 ; i < NCAPINTS ; i++) 709 for (i = 0; i < NCAPINTS; i++)
503 boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; 710 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
504 } 711 }
505 712
@@ -507,72 +714,91 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
507 for (i = 0; i < NCAPINTS; i++) 714 for (i = 0; i < NCAPINTS; i++)
508 c->x86_capability[i] &= ~cleared_cpu_caps[i]; 715 c->x86_capability[i] &= ~cleared_cpu_caps[i];
509 716
717#ifdef CONFIG_X86_MCE
510 /* Init Machine Check Exception if available. */ 718 /* Init Machine Check Exception if available. */
511 mcheck_init(c); 719 mcheck_init(c);
720#endif
512 721
513 select_idle_routine(c); 722 select_idle_routine(c);
723
724#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
725 numa_add_cpu(smp_processor_id());
726#endif
514} 727}
515 728
729#ifdef CONFIG_X86_64
730static void vgetcpu_set_mode(void)
731{
732 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
733 vgetcpu_mode = VGETCPU_RDTSCP;
734 else
735 vgetcpu_mode = VGETCPU_LSL;
736}
737#endif
738
516void __init identify_boot_cpu(void) 739void __init identify_boot_cpu(void)
517{ 740{
518 identify_cpu(&boot_cpu_data); 741 identify_cpu(&boot_cpu_data);
742#ifdef CONFIG_X86_32
519 sysenter_setup(); 743 sysenter_setup();
520 enable_sep_cpu(); 744 enable_sep_cpu();
745#else
746 vgetcpu_set_mode();
747#endif
521} 748}
522 749
523void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) 750void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
524{ 751{
525 BUG_ON(c == &boot_cpu_data); 752 BUG_ON(c == &boot_cpu_data);
526 identify_cpu(c); 753 identify_cpu(c);
754#ifdef CONFIG_X86_32
527 enable_sep_cpu(); 755 enable_sep_cpu();
756#endif
528 mtrr_ap_init(); 757 mtrr_ap_init();
529} 758}
530 759
531#ifdef CONFIG_X86_HT 760struct msr_range {
532void __cpuinit detect_ht(struct cpuinfo_x86 *c) 761 unsigned min;
533{ 762 unsigned max;
534 u32 eax, ebx, ecx, edx; 763};
535 int index_msb, core_bits;
536
537 cpuid(1, &eax, &ebx, &ecx, &edx);
538
539 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
540 return;
541
542 smp_num_siblings = (ebx & 0xff0000) >> 16;
543 764
544 if (smp_num_siblings == 1) { 765static struct msr_range msr_range_array[] __cpuinitdata = {
545 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n"); 766 { 0x00000000, 0x00000418},
546 } else if (smp_num_siblings > 1) { 767 { 0xc0000000, 0xc000040b},
768 { 0xc0010000, 0xc0010142},
769 { 0xc0011000, 0xc001103b},
770};
547 771
548 if (smp_num_siblings > NR_CPUS) { 772static void __cpuinit print_cpu_msr(void)
549 printk(KERN_WARNING "CPU: Unsupported number of the " 773{
550 "siblings %d", smp_num_siblings); 774 unsigned index;
551 smp_num_siblings = 1; 775 u64 val;
552 return; 776 int i;
777 unsigned index_min, index_max;
778
779 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
780 index_min = msr_range_array[i].min;
781 index_max = msr_range_array[i].max;
782 for (index = index_min; index < index_max; index++) {
783 if (rdmsrl_amd_safe(index, &val))
784 continue;
785 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
553 } 786 }
787 }
788}
554 789
555 index_msb = get_count_order(smp_num_siblings); 790static int show_msr __cpuinitdata;
556 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb); 791static __init int setup_show_msr(char *arg)
557 792{
558 printk(KERN_INFO "CPU: Physical Processor ID: %d\n", 793 int num;
559 c->phys_proc_id);
560
561 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
562
563 index_msb = get_count_order(smp_num_siblings) ;
564
565 core_bits = get_count_order(c->x86_max_cores);
566 794
567 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) & 795 get_option(&arg, &num);
568 ((1 << core_bits) - 1);
569 796
570 if (c->x86_max_cores > 1) 797 if (num > 0)
571 printk(KERN_INFO "CPU: Processor Core ID: %d\n", 798 show_msr = num;
572 c->cpu_core_id); 799 return 1;
573 }
574} 800}
575#endif 801__setup("show_msr=", setup_show_msr);
576 802
577static __init int setup_noclflush(char *arg) 803static __init int setup_noclflush(char *arg)
578{ 804{
@@ -590,18 +816,26 @@ void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
590 else if (c->cpuid_level >= 0) 816 else if (c->cpuid_level >= 0)
591 vendor = c->x86_vendor_id; 817 vendor = c->x86_vendor_id;
592 818
593 if (vendor && strncmp(c->x86_model_id, vendor, strlen(vendor))) 819 if (vendor && !strstr(c->x86_model_id, vendor))
594 printk("%s ", vendor); 820 printk(KERN_CONT "%s ", vendor);
595 821
596 if (!c->x86_model_id[0]) 822 if (c->x86_model_id[0])
597 printk("%d86", c->x86); 823 printk(KERN_CONT "%s", c->x86_model_id);
598 else 824 else
599 printk("%s", c->x86_model_id); 825 printk(KERN_CONT "%d86", c->x86);
600 826
601 if (c->x86_mask || c->cpuid_level >= 0) 827 if (c->x86_mask || c->cpuid_level >= 0)
602 printk(" stepping %02x\n", c->x86_mask); 828 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
603 else 829 else
604 printk("\n"); 830 printk(KERN_CONT "\n");
831
832#ifdef CONFIG_SMP
833 if (c->cpu_index < show_msr)
834 print_cpu_msr();
835#else
836 if (show_msr)
837 print_cpu_msr();
838#endif
605} 839}
606 840
607static __init int setup_disablecpuid(char *arg) 841static __init int setup_disablecpuid(char *arg)
@@ -617,19 +851,89 @@ __setup("clearcpuid=", setup_disablecpuid);
617 851
618cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE; 852cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
619 853
620void __init early_cpu_init(void) 854#ifdef CONFIG_X86_64
855struct x8664_pda **_cpu_pda __read_mostly;
856EXPORT_SYMBOL(_cpu_pda);
857
858struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
859
860char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
861
862void __cpuinit pda_init(int cpu)
621{ 863{
622 struct cpu_vendor_dev *cvdev; 864 struct x8664_pda *pda = cpu_pda(cpu);
865
866 /* Setup up data that may be needed in __get_free_pages early */
867 loadsegment(fs, 0);
868 loadsegment(gs, 0);
869 /* Memory clobbers used to order PDA accessed */
870 mb();
871 wrmsrl(MSR_GS_BASE, pda);
872 mb();
873
874 pda->cpunumber = cpu;
875 pda->irqcount = -1;
876 pda->kernelstack = (unsigned long)stack_thread_info() -
877 PDA_STACKOFFSET + THREAD_SIZE;
878 pda->active_mm = &init_mm;
879 pda->mmu_state = 0;
880
881 if (cpu == 0) {
882 /* others are initialized in smpboot.c */
883 pda->pcurrent = &init_task;
884 pda->irqstackptr = boot_cpu_stack;
885 pda->irqstackptr += IRQSTACKSIZE - 64;
886 } else {
887 if (!pda->irqstackptr) {
888 pda->irqstackptr = (char *)
889 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
890 if (!pda->irqstackptr)
891 panic("cannot allocate irqstack for cpu %d",
892 cpu);
893 pda->irqstackptr += IRQSTACKSIZE - 64;
894 }
895
896 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
897 pda->nodenumber = cpu_to_node(cpu);
898 }
899}
900
901char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
902 DEBUG_STKSZ] __page_aligned_bss;
623 903
624 for (cvdev = __x86cpuvendor_start ; 904extern asmlinkage void ignore_sysret(void);
625 cvdev < __x86cpuvendor_end ;
626 cvdev++)
627 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
628 905
629 early_cpu_detect(); 906/* May not be marked __init: used by software suspend */
630 validate_pat_support(&boot_cpu_data); 907void syscall_init(void)
908{
909 /*
910 * LSTAR and STAR live in a bit strange symbiosis.
911 * They both write to the same internal register. STAR allows to
912 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
913 */
914 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
915 wrmsrl(MSR_LSTAR, system_call);
916 wrmsrl(MSR_CSTAR, ignore_sysret);
917
918#ifdef CONFIG_IA32_EMULATION
919 syscall32_cpu_init();
920#endif
921
922 /* Flags to clear on syscall */
923 wrmsrl(MSR_SYSCALL_MASK,
924 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
631} 925}
632 926
927unsigned long kernel_eflags;
928
929/*
930 * Copies of the original ist values from the tss are only accessed during
931 * debugging, no special alignment required.
932 */
933DEFINE_PER_CPU(struct orig_ist, orig_ist);
934
935#else
936
633/* Make sure %fs is initialized properly in idle threads */ 937/* Make sure %fs is initialized properly in idle threads */
634struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs) 938struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
635{ 939{
@@ -637,25 +941,136 @@ struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
637 regs->fs = __KERNEL_PERCPU; 941 regs->fs = __KERNEL_PERCPU;
638 return regs; 942 return regs;
639} 943}
640 944#endif
641/* Current gdt points %fs at the "master" per-cpu area: after this,
642 * it's on the real one. */
643void switch_to_new_gdt(void)
644{
645 struct desc_ptr gdt_descr;
646
647 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
648 gdt_descr.size = GDT_SIZE - 1;
649 load_gdt(&gdt_descr);
650 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
651}
652 945
653/* 946/*
654 * cpu_init() initializes state that is per-CPU. Some data is already 947 * cpu_init() initializes state that is per-CPU. Some data is already
655 * initialized (naturally) in the bootstrap process, such as the GDT 948 * initialized (naturally) in the bootstrap process, such as the GDT
656 * and IDT. We reload them nevertheless, this function acts as a 949 * and IDT. We reload them nevertheless, this function acts as a
657 * 'CPU state barrier', nothing should get across. 950 * 'CPU state barrier', nothing should get across.
951 * A lot of state is already set up in PDA init for 64 bit
658 */ 952 */
953#ifdef CONFIG_X86_64
954void __cpuinit cpu_init(void)
955{
956 int cpu = stack_smp_processor_id();
957 struct tss_struct *t = &per_cpu(init_tss, cpu);
958 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
959 unsigned long v;
960 char *estacks = NULL;
961 struct task_struct *me;
962 int i;
963
964 /* CPU 0 is initialised in head64.c */
965 if (cpu != 0)
966 pda_init(cpu);
967 else
968 estacks = boot_exception_stacks;
969
970 me = current;
971
972 if (cpu_test_and_set(cpu, cpu_initialized))
973 panic("CPU#%d already initialized!\n", cpu);
974
975 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
976
977 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
978
979 /*
980 * Initialize the per-CPU GDT with the boot GDT,
981 * and set up the GDT descriptor:
982 */
983
984 switch_to_new_gdt();
985 load_idt((const struct desc_ptr *)&idt_descr);
986
987 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
988 syscall_init();
989
990 wrmsrl(MSR_FS_BASE, 0);
991 wrmsrl(MSR_KERNEL_GS_BASE, 0);
992 barrier();
993
994 check_efer();
995 if (cpu != 0 && x2apic)
996 enable_x2apic();
997
998 /*
999 * set up and load the per-CPU TSS
1000 */
1001 if (!orig_ist->ist[0]) {
1002 static const unsigned int order[N_EXCEPTION_STACKS] = {
1003 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
1004 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
1005 };
1006 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1007 if (cpu) {
1008 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
1009 if (!estacks)
1010 panic("Cannot allocate exception "
1011 "stack %ld %d\n", v, cpu);
1012 }
1013 estacks += PAGE_SIZE << order[v];
1014 orig_ist->ist[v] = t->x86_tss.ist[v] =
1015 (unsigned long)estacks;
1016 }
1017 }
1018
1019 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1020 /*
1021 * <= is required because the CPU will access up to
1022 * 8 bits beyond the end of the IO permission bitmap.
1023 */
1024 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1025 t->io_bitmap[i] = ~0UL;
1026
1027 atomic_inc(&init_mm.mm_count);
1028 me->active_mm = &init_mm;
1029 if (me->mm)
1030 BUG();
1031 enter_lazy_tlb(&init_mm, me);
1032
1033 load_sp0(t, &current->thread);
1034 set_tss_desc(cpu, t);
1035 load_TR_desc();
1036 load_LDT(&init_mm.context);
1037
1038#ifdef CONFIG_KGDB
1039 /*
1040 * If the kgdb is connected no debug regs should be altered. This
1041 * is only applicable when KGDB and a KGDB I/O module are built
1042 * into the kernel and you are using early debugging with
1043 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1044 */
1045 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1046 arch_kgdb_ops.correct_hw_break();
1047 else {
1048#endif
1049 /*
1050 * Clear all 6 debug registers:
1051 */
1052
1053 set_debugreg(0UL, 0);
1054 set_debugreg(0UL, 1);
1055 set_debugreg(0UL, 2);
1056 set_debugreg(0UL, 3);
1057 set_debugreg(0UL, 6);
1058 set_debugreg(0UL, 7);
1059#ifdef CONFIG_KGDB
1060 /* If the kgdb is connected no debug regs should be altered. */
1061 }
1062#endif
1063
1064 fpu_init();
1065
1066 raw_local_save_flags(kernel_eflags);
1067
1068 if (is_uv_system())
1069 uv_cpu_init();
1070}
1071
1072#else
1073
659void __cpuinit cpu_init(void) 1074void __cpuinit cpu_init(void)
660{ 1075{
661 int cpu = smp_processor_id(); 1076 int cpu = smp_processor_id();
@@ -709,19 +1124,21 @@ void __cpuinit cpu_init(void)
709 /* 1124 /*
710 * Force FPU initialization: 1125 * Force FPU initialization:
711 */ 1126 */
712 current_thread_info()->status = 0; 1127 if (cpu_has_xsave)
1128 current_thread_info()->status = TS_XSAVE;
1129 else
1130 current_thread_info()->status = 0;
713 clear_used_math(); 1131 clear_used_math();
714 mxcsr_feature_mask_init(); 1132 mxcsr_feature_mask_init();
715}
716 1133
717#ifdef CONFIG_HOTPLUG_CPU 1134 /*
718void __cpuinit cpu_uninit(void) 1135 * Boot processor to setup the FP and extended state context info.
719{ 1136 */
720 int cpu = raw_smp_processor_id(); 1137 if (!smp_processor_id())
721 cpu_clear(cpu, cpu_initialized); 1138 init_thread_xstate();
722 1139
723 /* lazy TLB state */ 1140 xsave_init();
724 per_cpu(cpu_tlbstate, cpu).state = 0;
725 per_cpu(cpu_tlbstate, cpu).active_mm = &init_mm;
726} 1141}
1142
1143
727#endif 1144#endif
diff --git a/arch/x86/kernel/cpu/common_64.c b/arch/x86/kernel/cpu/common_64.c
deleted file mode 100644
index dd6e3f15017e..000000000000
--- a/arch/x86/kernel/cpu/common_64.c
+++ /dev/null
@@ -1,670 +0,0 @@
1#include <linux/init.h>
2#include <linux/kernel.h>
3#include <linux/sched.h>
4#include <linux/string.h>
5#include <linux/bootmem.h>
6#include <linux/bitops.h>
7#include <linux/module.h>
8#include <linux/kgdb.h>
9#include <linux/topology.h>
10#include <linux/delay.h>
11#include <linux/smp.h>
12#include <linux/percpu.h>
13#include <asm/i387.h>
14#include <asm/msr.h>
15#include <asm/io.h>
16#include <asm/linkage.h>
17#include <asm/mmu_context.h>
18#include <asm/mtrr.h>
19#include <asm/mce.h>
20#include <asm/pat.h>
21#include <asm/numa.h>
22#ifdef CONFIG_X86_LOCAL_APIC
23#include <asm/mpspec.h>
24#include <asm/apic.h>
25#include <mach_apic.h>
26#endif
27#include <asm/pda.h>
28#include <asm/pgtable.h>
29#include <asm/processor.h>
30#include <asm/desc.h>
31#include <asm/atomic.h>
32#include <asm/proto.h>
33#include <asm/sections.h>
34#include <asm/setup.h>
35#include <asm/genapic.h>
36
37#include "cpu.h"
38
39/* We need valid kernel segments for data and code in long mode too
40 * IRET will check the segment types kkeil 2000/10/28
41 * Also sysret mandates a special GDT layout
42 */
43/* The TLS descriptors are currently at a different place compared to i386.
44 Hopefully nobody expects them at a fixed place (Wine?) */
45DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
46 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
47 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
48 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
49 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
50 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
51 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
52} };
53EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
54
55__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
56
57/* Current gdt points %fs at the "master" per-cpu area: after this,
58 * it's on the real one. */
59void switch_to_new_gdt(void)
60{
61 struct desc_ptr gdt_descr;
62
63 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
64 gdt_descr.size = GDT_SIZE - 1;
65 load_gdt(&gdt_descr);
66}
67
68struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
69
70static void __cpuinit default_init(struct cpuinfo_x86 *c)
71{
72 display_cacheinfo(c);
73}
74
75static struct cpu_dev __cpuinitdata default_cpu = {
76 .c_init = default_init,
77 .c_vendor = "Unknown",
78};
79static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
80
81int __cpuinit get_model_name(struct cpuinfo_x86 *c)
82{
83 unsigned int *v;
84
85 if (c->extended_cpuid_level < 0x80000004)
86 return 0;
87
88 v = (unsigned int *) c->x86_model_id;
89 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
90 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
91 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
92 c->x86_model_id[48] = 0;
93 return 1;
94}
95
96
97void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
98{
99 unsigned int n, dummy, ebx, ecx, edx;
100
101 n = c->extended_cpuid_level;
102
103 if (n >= 0x80000005) {
104 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
105 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
106 "D cache %dK (%d bytes/line)\n",
107 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
108 c->x86_cache_size = (ecx>>24) + (edx>>24);
109 /* On K8 L1 TLB is inclusive, so don't count it */
110 c->x86_tlbsize = 0;
111 }
112
113 if (n >= 0x80000006) {
114 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
115 ecx = cpuid_ecx(0x80000006);
116 c->x86_cache_size = ecx >> 16;
117 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
118
119 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
120 c->x86_cache_size, ecx & 0xFF);
121 }
122}
123
124void __cpuinit detect_ht(struct cpuinfo_x86 *c)
125{
126#ifdef CONFIG_SMP
127 u32 eax, ebx, ecx, edx;
128 int index_msb, core_bits;
129
130 cpuid(1, &eax, &ebx, &ecx, &edx);
131
132
133 if (!cpu_has(c, X86_FEATURE_HT))
134 return;
135 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
136 goto out;
137
138 smp_num_siblings = (ebx & 0xff0000) >> 16;
139
140 if (smp_num_siblings == 1) {
141 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
142 } else if (smp_num_siblings > 1) {
143
144 if (smp_num_siblings > NR_CPUS) {
145 printk(KERN_WARNING "CPU: Unsupported number of "
146 "siblings %d", smp_num_siblings);
147 smp_num_siblings = 1;
148 return;
149 }
150
151 index_msb = get_count_order(smp_num_siblings);
152 c->phys_proc_id = phys_pkg_id(index_msb);
153
154 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
155
156 index_msb = get_count_order(smp_num_siblings);
157
158 core_bits = get_count_order(c->x86_max_cores);
159
160 c->cpu_core_id = phys_pkg_id(index_msb) &
161 ((1 << core_bits) - 1);
162 }
163out:
164 if ((c->x86_max_cores * smp_num_siblings) > 1) {
165 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
166 c->phys_proc_id);
167 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
168 c->cpu_core_id);
169 }
170
171#endif
172}
173
174static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
175{
176 char *v = c->x86_vendor_id;
177 int i;
178 static int printed;
179
180 for (i = 0; i < X86_VENDOR_NUM; i++) {
181 if (cpu_devs[i]) {
182 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
183 (cpu_devs[i]->c_ident[1] &&
184 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
185 c->x86_vendor = i;
186 this_cpu = cpu_devs[i];
187 return;
188 }
189 }
190 }
191 if (!printed) {
192 printed++;
193 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
194 printk(KERN_ERR "CPU: Your system may be unstable.\n");
195 }
196 c->x86_vendor = X86_VENDOR_UNKNOWN;
197}
198
199static void __init early_cpu_support_print(void)
200{
201 int i,j;
202 struct cpu_dev *cpu_devx;
203
204 printk("KERNEL supported cpus:\n");
205 for (i = 0; i < X86_VENDOR_NUM; i++) {
206 cpu_devx = cpu_devs[i];
207 if (!cpu_devx)
208 continue;
209 for (j = 0; j < 2; j++) {
210 if (!cpu_devx->c_ident[j])
211 continue;
212 printk(" %s %s\n", cpu_devx->c_vendor,
213 cpu_devx->c_ident[j]);
214 }
215 }
216}
217
218static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
219
220void __init early_cpu_init(void)
221{
222 struct cpu_vendor_dev *cvdev;
223
224 for (cvdev = __x86cpuvendor_start ;
225 cvdev < __x86cpuvendor_end ;
226 cvdev++)
227 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
228 early_cpu_support_print();
229 early_identify_cpu(&boot_cpu_data);
230}
231
232/* Do some early cpuid on the boot CPU to get some parameter that are
233 needed before check_bugs. Everything advanced is in identify_cpu
234 below. */
235static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
236{
237 u32 tfms, xlvl;
238
239 c->loops_per_jiffy = loops_per_jiffy;
240 c->x86_cache_size = -1;
241 c->x86_vendor = X86_VENDOR_UNKNOWN;
242 c->x86_model = c->x86_mask = 0; /* So far unknown... */
243 c->x86_vendor_id[0] = '\0'; /* Unset */
244 c->x86_model_id[0] = '\0'; /* Unset */
245 c->x86_clflush_size = 64;
246 c->x86_cache_alignment = c->x86_clflush_size;
247 c->x86_max_cores = 1;
248 c->x86_coreid_bits = 0;
249 c->extended_cpuid_level = 0;
250 memset(&c->x86_capability, 0, sizeof c->x86_capability);
251
252 /* Get vendor name */
253 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
254 (unsigned int *)&c->x86_vendor_id[0],
255 (unsigned int *)&c->x86_vendor_id[8],
256 (unsigned int *)&c->x86_vendor_id[4]);
257
258 get_cpu_vendor(c);
259
260 /* Initialize the standard set of capabilities */
261 /* Note that the vendor-specific code below might override */
262
263 /* Intel-defined flags: level 0x00000001 */
264 if (c->cpuid_level >= 0x00000001) {
265 __u32 misc;
266 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
267 &c->x86_capability[0]);
268 c->x86 = (tfms >> 8) & 0xf;
269 c->x86_model = (tfms >> 4) & 0xf;
270 c->x86_mask = tfms & 0xf;
271 if (c->x86 == 0xf)
272 c->x86 += (tfms >> 20) & 0xff;
273 if (c->x86 >= 0x6)
274 c->x86_model += ((tfms >> 16) & 0xF) << 4;
275 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
276 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
277 } else {
278 /* Have CPUID level 0 only - unheard of */
279 c->x86 = 4;
280 }
281
282 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
283#ifdef CONFIG_SMP
284 c->phys_proc_id = c->initial_apicid;
285#endif
286 /* AMD-defined flags: level 0x80000001 */
287 xlvl = cpuid_eax(0x80000000);
288 c->extended_cpuid_level = xlvl;
289 if ((xlvl & 0xffff0000) == 0x80000000) {
290 if (xlvl >= 0x80000001) {
291 c->x86_capability[1] = cpuid_edx(0x80000001);
292 c->x86_capability[6] = cpuid_ecx(0x80000001);
293 }
294 if (xlvl >= 0x80000004)
295 get_model_name(c); /* Default name */
296 }
297
298 /* Transmeta-defined flags: level 0x80860001 */
299 xlvl = cpuid_eax(0x80860000);
300 if ((xlvl & 0xffff0000) == 0x80860000) {
301 /* Don't set x86_cpuid_level here for now to not confuse. */
302 if (xlvl >= 0x80860001)
303 c->x86_capability[2] = cpuid_edx(0x80860001);
304 }
305
306 if (c->extended_cpuid_level >= 0x80000007)
307 c->x86_power = cpuid_edx(0x80000007);
308
309 if (c->extended_cpuid_level >= 0x80000008) {
310 u32 eax = cpuid_eax(0x80000008);
311
312 c->x86_virt_bits = (eax >> 8) & 0xff;
313 c->x86_phys_bits = eax & 0xff;
314 }
315
316 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
317 cpu_devs[c->x86_vendor]->c_early_init)
318 cpu_devs[c->x86_vendor]->c_early_init(c);
319
320 validate_pat_support(c);
321}
322
323/*
324 * This does the hard work of actually picking apart the CPU stuff...
325 */
326static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
327{
328 int i;
329
330 early_identify_cpu(c);
331
332 init_scattered_cpuid_features(c);
333
334 c->apicid = phys_pkg_id(0);
335
336 /*
337 * Vendor-specific initialization. In this section we
338 * canonicalize the feature flags, meaning if there are
339 * features a certain CPU supports which CPUID doesn't
340 * tell us, CPUID claiming incorrect flags, or other bugs,
341 * we handle them here.
342 *
343 * At the end of this section, c->x86_capability better
344 * indicate the features this CPU genuinely supports!
345 */
346 if (this_cpu->c_init)
347 this_cpu->c_init(c);
348
349 detect_ht(c);
350
351 /*
352 * On SMP, boot_cpu_data holds the common feature set between
353 * all CPUs; so make sure that we indicate which features are
354 * common between the CPUs. The first time this routine gets
355 * executed, c == &boot_cpu_data.
356 */
357 if (c != &boot_cpu_data) {
358 /* AND the already accumulated flags with these */
359 for (i = 0; i < NCAPINTS; i++)
360 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
361 }
362
363 /* Clear all flags overriden by options */
364 for (i = 0; i < NCAPINTS; i++)
365 c->x86_capability[i] &= ~cleared_cpu_caps[i];
366
367#ifdef CONFIG_X86_MCE
368 mcheck_init(c);
369#endif
370 select_idle_routine(c);
371
372#ifdef CONFIG_NUMA
373 numa_add_cpu(smp_processor_id());
374#endif
375
376}
377
378void __cpuinit identify_boot_cpu(void)
379{
380 identify_cpu(&boot_cpu_data);
381}
382
383void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
384{
385 BUG_ON(c == &boot_cpu_data);
386 identify_cpu(c);
387 mtrr_ap_init();
388}
389
390static __init int setup_noclflush(char *arg)
391{
392 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
393 return 1;
394}
395__setup("noclflush", setup_noclflush);
396
397void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
398{
399 if (c->x86_model_id[0])
400 printk(KERN_CONT "%s", c->x86_model_id);
401
402 if (c->x86_mask || c->cpuid_level >= 0)
403 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
404 else
405 printk(KERN_CONT "\n");
406}
407
408static __init int setup_disablecpuid(char *arg)
409{
410 int bit;
411 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
412 setup_clear_cpu_cap(bit);
413 else
414 return 0;
415 return 1;
416}
417__setup("clearcpuid=", setup_disablecpuid);
418
419cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
420
421struct x8664_pda **_cpu_pda __read_mostly;
422EXPORT_SYMBOL(_cpu_pda);
423
424struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
425
426char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
427
428unsigned long __supported_pte_mask __read_mostly = ~0UL;
429EXPORT_SYMBOL_GPL(__supported_pte_mask);
430
431static int do_not_nx __cpuinitdata;
432
433/* noexec=on|off
434Control non executable mappings for 64bit processes.
435
436on Enable(default)
437off Disable
438*/
439static int __init nonx_setup(char *str)
440{
441 if (!str)
442 return -EINVAL;
443 if (!strncmp(str, "on", 2)) {
444 __supported_pte_mask |= _PAGE_NX;
445 do_not_nx = 0;
446 } else if (!strncmp(str, "off", 3)) {
447 do_not_nx = 1;
448 __supported_pte_mask &= ~_PAGE_NX;
449 }
450 return 0;
451}
452early_param("noexec", nonx_setup);
453
454int force_personality32;
455
456/* noexec32=on|off
457Control non executable heap for 32bit processes.
458To control the stack too use noexec=off
459
460on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
461off PROT_READ implies PROT_EXEC
462*/
463static int __init nonx32_setup(char *str)
464{
465 if (!strcmp(str, "on"))
466 force_personality32 &= ~READ_IMPLIES_EXEC;
467 else if (!strcmp(str, "off"))
468 force_personality32 |= READ_IMPLIES_EXEC;
469 return 1;
470}
471__setup("noexec32=", nonx32_setup);
472
473void pda_init(int cpu)
474{
475 struct x8664_pda *pda = cpu_pda(cpu);
476
477 /* Setup up data that may be needed in __get_free_pages early */
478 loadsegment(fs, 0);
479 loadsegment(gs, 0);
480 /* Memory clobbers used to order PDA accessed */
481 mb();
482 wrmsrl(MSR_GS_BASE, pda);
483 mb();
484
485 pda->cpunumber = cpu;
486 pda->irqcount = -1;
487 pda->kernelstack = (unsigned long)stack_thread_info() -
488 PDA_STACKOFFSET + THREAD_SIZE;
489 pda->active_mm = &init_mm;
490 pda->mmu_state = 0;
491
492 if (cpu == 0) {
493 /* others are initialized in smpboot.c */
494 pda->pcurrent = &init_task;
495 pda->irqstackptr = boot_cpu_stack;
496 } else {
497 pda->irqstackptr = (char *)
498 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
499 if (!pda->irqstackptr)
500 panic("cannot allocate irqstack for cpu %d", cpu);
501
502 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
503 pda->nodenumber = cpu_to_node(cpu);
504 }
505
506 pda->irqstackptr += IRQSTACKSIZE-64;
507}
508
509char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
510 DEBUG_STKSZ] __page_aligned_bss;
511
512extern asmlinkage void ignore_sysret(void);
513
514/* May not be marked __init: used by software suspend */
515void syscall_init(void)
516{
517 /*
518 * LSTAR and STAR live in a bit strange symbiosis.
519 * They both write to the same internal register. STAR allows to
520 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
521 */
522 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
523 wrmsrl(MSR_LSTAR, system_call);
524 wrmsrl(MSR_CSTAR, ignore_sysret);
525
526#ifdef CONFIG_IA32_EMULATION
527 syscall32_cpu_init();
528#endif
529
530 /* Flags to clear on syscall */
531 wrmsrl(MSR_SYSCALL_MASK,
532 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
533}
534
535void __cpuinit check_efer(void)
536{
537 unsigned long efer;
538
539 rdmsrl(MSR_EFER, efer);
540 if (!(efer & EFER_NX) || do_not_nx)
541 __supported_pte_mask &= ~_PAGE_NX;
542}
543
544unsigned long kernel_eflags;
545
546/*
547 * Copies of the original ist values from the tss are only accessed during
548 * debugging, no special alignment required.
549 */
550DEFINE_PER_CPU(struct orig_ist, orig_ist);
551
552/*
553 * cpu_init() initializes state that is per-CPU. Some data is already
554 * initialized (naturally) in the bootstrap process, such as the GDT
555 * and IDT. We reload them nevertheless, this function acts as a
556 * 'CPU state barrier', nothing should get across.
557 * A lot of state is already set up in PDA init.
558 */
559void __cpuinit cpu_init(void)
560{
561 int cpu = stack_smp_processor_id();
562 struct tss_struct *t = &per_cpu(init_tss, cpu);
563 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
564 unsigned long v;
565 char *estacks = NULL;
566 struct task_struct *me;
567 int i;
568
569 /* CPU 0 is initialised in head64.c */
570 if (cpu != 0)
571 pda_init(cpu);
572 else
573 estacks = boot_exception_stacks;
574
575 me = current;
576
577 if (cpu_test_and_set(cpu, cpu_initialized))
578 panic("CPU#%d already initialized!\n", cpu);
579
580 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
581
582 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
583
584 /*
585 * Initialize the per-CPU GDT with the boot GDT,
586 * and set up the GDT descriptor:
587 */
588
589 switch_to_new_gdt();
590 load_idt((const struct desc_ptr *)&idt_descr);
591
592 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
593 syscall_init();
594
595 wrmsrl(MSR_FS_BASE, 0);
596 wrmsrl(MSR_KERNEL_GS_BASE, 0);
597 barrier();
598
599 check_efer();
600
601 /*
602 * set up and load the per-CPU TSS
603 */
604 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
605 static const unsigned int order[N_EXCEPTION_STACKS] = {
606 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
607 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
608 };
609 if (cpu) {
610 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
611 if (!estacks)
612 panic("Cannot allocate exception stack %ld %d\n",
613 v, cpu);
614 }
615 estacks += PAGE_SIZE << order[v];
616 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
617 }
618
619 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
620 /*
621 * <= is required because the CPU will access up to
622 * 8 bits beyond the end of the IO permission bitmap.
623 */
624 for (i = 0; i <= IO_BITMAP_LONGS; i++)
625 t->io_bitmap[i] = ~0UL;
626
627 atomic_inc(&init_mm.mm_count);
628 me->active_mm = &init_mm;
629 if (me->mm)
630 BUG();
631 enter_lazy_tlb(&init_mm, me);
632
633 load_sp0(t, &current->thread);
634 set_tss_desc(cpu, t);
635 load_TR_desc();
636 load_LDT(&init_mm.context);
637
638#ifdef CONFIG_KGDB
639 /*
640 * If the kgdb is connected no debug regs should be altered. This
641 * is only applicable when KGDB and a KGDB I/O module are built
642 * into the kernel and you are using early debugging with
643 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
644 */
645 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
646 arch_kgdb_ops.correct_hw_break();
647 else {
648#endif
649 /*
650 * Clear all 6 debug registers:
651 */
652
653 set_debugreg(0UL, 0);
654 set_debugreg(0UL, 1);
655 set_debugreg(0UL, 2);
656 set_debugreg(0UL, 3);
657 set_debugreg(0UL, 6);
658 set_debugreg(0UL, 7);
659#ifdef CONFIG_KGDB
660 /* If the kgdb is connected no debug regs should be altered. */
661 }
662#endif
663
664 fpu_init();
665
666 raw_local_save_flags(kernel_eflags);
667
668 if (is_uv_system())
669 uv_cpu_init();
670}
diff --git a/arch/x86/kernel/cpu/cpu.h b/arch/x86/kernel/cpu/cpu.h
index 4d894e8565fe..de4094a39210 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -21,23 +21,16 @@ struct cpu_dev {
21 void (*c_init)(struct cpuinfo_x86 * c); 21 void (*c_init)(struct cpuinfo_x86 * c);
22 void (*c_identify)(struct cpuinfo_x86 * c); 22 void (*c_identify)(struct cpuinfo_x86 * c);
23 unsigned int (*c_size_cache)(struct cpuinfo_x86 * c, unsigned int size); 23 unsigned int (*c_size_cache)(struct cpuinfo_x86 * c, unsigned int size);
24 int c_x86_vendor;
24}; 25};
25 26
26extern struct cpu_dev * cpu_devs [X86_VENDOR_NUM]; 27#define cpu_dev_register(cpu_devX) \
28 static struct cpu_dev *__cpu_dev_##cpu_devX __used \
29 __attribute__((__section__(".x86_cpu_dev.init"))) = \
30 &cpu_devX;
27 31
28struct cpu_vendor_dev { 32extern struct cpu_dev *__x86_cpu_dev_start[], *__x86_cpu_dev_end[];
29 int vendor;
30 struct cpu_dev *cpu_dev;
31};
32
33#define cpu_vendor_dev_register(cpu_vendor_id, cpu_dev) \
34 static struct cpu_vendor_dev __cpu_vendor_dev_##cpu_vendor_id __used \
35 __attribute__((__section__(".x86cpuvendor.init"))) = \
36 { cpu_vendor_id, cpu_dev }
37
38extern struct cpu_vendor_dev __x86cpuvendor_start[], __x86cpuvendor_end[];
39 33
40extern int get_model_name(struct cpuinfo_x86 *c);
41extern void display_cacheinfo(struct cpuinfo_x86 *c); 34extern void display_cacheinfo(struct cpuinfo_x86 *c);
42 35
43#endif 36#endif
diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index dd097b835839..c24c4a487b7c 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -256,7 +256,8 @@ static u32 get_cur_val(const cpumask_t *mask)
256 * Only IA32_APERF/IA32_MPERF ratio is architecturally defined and 256 * Only IA32_APERF/IA32_MPERF ratio is architecturally defined and
257 * no meaning should be associated with absolute values of these MSRs. 257 * no meaning should be associated with absolute values of these MSRs.
258 */ 258 */
259static unsigned int get_measured_perf(unsigned int cpu) 259static unsigned int get_measured_perf(struct cpufreq_policy *policy,
260 unsigned int cpu)
260{ 261{
261 union { 262 union {
262 struct { 263 struct {
@@ -326,7 +327,7 @@ static unsigned int get_measured_perf(unsigned int cpu)
326 327
327#endif 328#endif
328 329
329 retval = per_cpu(drv_data, cpu)->max_freq * perf_percent / 100; 330 retval = per_cpu(drv_data, policy->cpu)->max_freq * perf_percent / 100;
330 331
331 put_cpu(); 332 put_cpu();
332 set_cpus_allowed_ptr(current, &saved_mask); 333 set_cpus_allowed_ptr(current, &saved_mask);
@@ -785,7 +786,11 @@ static int __init acpi_cpufreq_init(void)
785 if (ret) 786 if (ret)
786 return ret; 787 return ret;
787 788
788 return cpufreq_register_driver(&acpi_cpufreq_driver); 789 ret = cpufreq_register_driver(&acpi_cpufreq_driver);
790 if (ret)
791 free_percpu(acpi_perf_data);
792
793 return ret;
789} 794}
790 795
791static void __exit acpi_cpufreq_exit(void) 796static void __exit acpi_cpufreq_exit(void)
@@ -795,8 +800,6 @@ static void __exit acpi_cpufreq_exit(void)
795 cpufreq_unregister_driver(&acpi_cpufreq_driver); 800 cpufreq_unregister_driver(&acpi_cpufreq_driver);
796 801
797 free_percpu(acpi_perf_data); 802 free_percpu(acpi_perf_data);
798
799 return;
800} 803}
801 804
802module_param(acpi_pstate_strict, uint, 0644); 805module_param(acpi_pstate_strict, uint, 0644);
diff --git a/arch/x86/kernel/cpu/cpufreq/elanfreq.c b/arch/x86/kernel/cpu/cpufreq/elanfreq.c
index e4a4bf870e94..fe613c93b366 100644
--- a/arch/x86/kernel/cpu/cpufreq/elanfreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/elanfreq.c
@@ -25,8 +25,8 @@
25#include <linux/cpufreq.h> 25#include <linux/cpufreq.h>
26 26
27#include <asm/msr.h> 27#include <asm/msr.h>
28#include <asm/timex.h> 28#include <linux/timex.h>
29#include <asm/io.h> 29#include <linux/io.h>
30 30
31#define REG_CSCIR 0x22 /* Chip Setup and Control Index Register */ 31#define REG_CSCIR 0x22 /* Chip Setup and Control Index Register */
32#define REG_CSCDR 0x23 /* Chip Setup and Control Data Register */ 32#define REG_CSCDR 0x23 /* Chip Setup and Control Data Register */
@@ -82,7 +82,7 @@ static unsigned int elanfreq_get_cpu_frequency(unsigned int cpu)
82 u8 clockspeed_reg; /* Clock Speed Register */ 82 u8 clockspeed_reg; /* Clock Speed Register */
83 83
84 local_irq_disable(); 84 local_irq_disable();
85 outb_p(0x80,REG_CSCIR); 85 outb_p(0x80, REG_CSCIR);
86 clockspeed_reg = inb_p(REG_CSCDR); 86 clockspeed_reg = inb_p(REG_CSCDR);
87 local_irq_enable(); 87 local_irq_enable();
88 88
@@ -98,10 +98,10 @@ static unsigned int elanfreq_get_cpu_frequency(unsigned int cpu)
98 } 98 }
99 99
100 /* 33 MHz is not 32 MHz... */ 100 /* 33 MHz is not 32 MHz... */
101 if ((clockspeed_reg & 0xE0)==0xA0) 101 if ((clockspeed_reg & 0xE0) == 0xA0)
102 return 33000; 102 return 33000;
103 103
104 return ((1<<((clockspeed_reg & 0xE0) >> 5)) * 1000); 104 return (1<<((clockspeed_reg & 0xE0) >> 5)) * 1000;
105} 105}
106 106
107 107
@@ -117,7 +117,7 @@ static unsigned int elanfreq_get_cpu_frequency(unsigned int cpu)
117 * There is no return value. 117 * There is no return value.
118 */ 118 */
119 119
120static void elanfreq_set_cpu_state (unsigned int state) 120static void elanfreq_set_cpu_state(unsigned int state)
121{ 121{
122 struct cpufreq_freqs freqs; 122 struct cpufreq_freqs freqs;
123 123
@@ -144,20 +144,20 @@ static void elanfreq_set_cpu_state (unsigned int state)
144 */ 144 */
145 145
146 local_irq_disable(); 146 local_irq_disable();
147 outb_p(0x40,REG_CSCIR); /* Disable hyperspeed mode */ 147 outb_p(0x40, REG_CSCIR); /* Disable hyperspeed mode */
148 outb_p(0x00,REG_CSCDR); 148 outb_p(0x00, REG_CSCDR);
149 local_irq_enable(); /* wait till internal pipelines and */ 149 local_irq_enable(); /* wait till internal pipelines and */
150 udelay(1000); /* buffers have cleaned up */ 150 udelay(1000); /* buffers have cleaned up */
151 151
152 local_irq_disable(); 152 local_irq_disable();
153 153
154 /* now, set the CPU clock speed register (0x80) */ 154 /* now, set the CPU clock speed register (0x80) */
155 outb_p(0x80,REG_CSCIR); 155 outb_p(0x80, REG_CSCIR);
156 outb_p(elan_multiplier[state].val80h,REG_CSCDR); 156 outb_p(elan_multiplier[state].val80h, REG_CSCDR);
157 157
158 /* now, the hyperspeed bit in PMU Force Mode Register (0x40) */ 158 /* now, the hyperspeed bit in PMU Force Mode Register (0x40) */
159 outb_p(0x40,REG_CSCIR); 159 outb_p(0x40, REG_CSCIR);
160 outb_p(elan_multiplier[state].val40h,REG_CSCDR); 160 outb_p(elan_multiplier[state].val40h, REG_CSCDR);
161 udelay(10000); 161 udelay(10000);
162 local_irq_enable(); 162 local_irq_enable();
163 163
@@ -173,12 +173,12 @@ static void elanfreq_set_cpu_state (unsigned int state)
173 * for the hardware supported by the driver. 173 * for the hardware supported by the driver.
174 */ 174 */
175 175
176static int elanfreq_verify (struct cpufreq_policy *policy) 176static int elanfreq_verify(struct cpufreq_policy *policy)
177{ 177{
178 return cpufreq_frequency_table_verify(policy, &elanfreq_table[0]); 178 return cpufreq_frequency_table_verify(policy, &elanfreq_table[0]);
179} 179}
180 180
181static int elanfreq_target (struct cpufreq_policy *policy, 181static int elanfreq_target(struct cpufreq_policy *policy,
182 unsigned int target_freq, 182 unsigned int target_freq,
183 unsigned int relation) 183 unsigned int relation)
184{ 184{
@@ -205,7 +205,7 @@ static int elanfreq_cpu_init(struct cpufreq_policy *policy)
205 205
206 /* capability check */ 206 /* capability check */
207 if ((c->x86_vendor != X86_VENDOR_AMD) || 207 if ((c->x86_vendor != X86_VENDOR_AMD) ||
208 (c->x86 != 4) || (c->x86_model!=10)) 208 (c->x86 != 4) || (c->x86_model != 10))
209 return -ENODEV; 209 return -ENODEV;
210 210
211 /* max freq */ 211 /* max freq */
@@ -213,7 +213,7 @@ static int elanfreq_cpu_init(struct cpufreq_policy *policy)
213 max_freq = elanfreq_get_cpu_frequency(0); 213 max_freq = elanfreq_get_cpu_frequency(0);
214 214
215 /* table init */ 215 /* table init */
216 for (i=0; (elanfreq_table[i].frequency != CPUFREQ_TABLE_END); i++) { 216 for (i = 0; (elanfreq_table[i].frequency != CPUFREQ_TABLE_END); i++) {
217 if (elanfreq_table[i].frequency > max_freq) 217 if (elanfreq_table[i].frequency > max_freq)
218 elanfreq_table[i].frequency = CPUFREQ_ENTRY_INVALID; 218 elanfreq_table[i].frequency = CPUFREQ_ENTRY_INVALID;
219 } 219 }
@@ -224,7 +224,7 @@ static int elanfreq_cpu_init(struct cpufreq_policy *policy)
224 224
225 result = cpufreq_frequency_table_cpuinfo(policy, elanfreq_table); 225 result = cpufreq_frequency_table_cpuinfo(policy, elanfreq_table);
226 if (result) 226 if (result)
227 return (result); 227 return result;
228 228
229 cpufreq_frequency_table_get_attr(elanfreq_table, policy->cpu); 229 cpufreq_frequency_table_get_attr(elanfreq_table, policy->cpu);
230 return 0; 230 return 0;
@@ -260,7 +260,7 @@ __setup("elanfreq=", elanfreq_setup);
260#endif 260#endif
261 261
262 262
263static struct freq_attr* elanfreq_attr[] = { 263static struct freq_attr *elanfreq_attr[] = {
264 &cpufreq_freq_attr_scaling_available_freqs, 264 &cpufreq_freq_attr_scaling_available_freqs,
265 NULL, 265 NULL,
266}; 266};
@@ -284,9 +284,9 @@ static int __init elanfreq_init(void)
284 284
285 /* Test if we have the right hardware */ 285 /* Test if we have the right hardware */
286 if ((c->x86_vendor != X86_VENDOR_AMD) || 286 if ((c->x86_vendor != X86_VENDOR_AMD) ||
287 (c->x86 != 4) || (c->x86_model!=10)) { 287 (c->x86 != 4) || (c->x86_model != 10)) {
288 printk(KERN_INFO "elanfreq: error: no Elan processor found!\n"); 288 printk(KERN_INFO "elanfreq: error: no Elan processor found!\n");
289 return -ENODEV; 289 return -ENODEV;
290 } 290 }
291 return cpufreq_register_driver(&elanfreq_driver); 291 return cpufreq_register_driver(&elanfreq_driver);
292} 292}
@@ -298,7 +298,7 @@ static void __exit elanfreq_exit(void)
298} 298}
299 299
300 300
301module_param (max_freq, int, 0444); 301module_param(max_freq, int, 0444);
302 302
303MODULE_LICENSE("GPL"); 303MODULE_LICENSE("GPL");
304MODULE_AUTHOR("Robert Schwebel <r.schwebel@pengutronix.de>, Sven Geggus <sven@geggus.net>"); 304MODULE_AUTHOR("Robert Schwebel <r.schwebel@pengutronix.de>, Sven Geggus <sven@geggus.net>");
diff --git a/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c b/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
index f1685fb91fbd..b8e05ee4f736 100644
--- a/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
+++ b/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
@@ -171,7 +171,7 @@ static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c)
171 } 171 }
172 172
173 if (c->x86 != 0xF) { 173 if (c->x86 != 0xF) {
174 printk(KERN_WARNING PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <cpufreq@lists.linux.org.uk>\n"); 174 printk(KERN_WARNING PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <cpufreq@vger.kernel.org>\n");
175 return 0; 175 return 0;
176 } 176 }
177 177
diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k6.c b/arch/x86/kernel/cpu/cpufreq/powernow-k6.c
index eb9b62b0830c..b5ced806a316 100644
--- a/arch/x86/kernel/cpu/cpufreq/powernow-k6.c
+++ b/arch/x86/kernel/cpu/cpufreq/powernow-k6.c
@@ -15,12 +15,11 @@
15#include <linux/slab.h> 15#include <linux/slab.h>
16 16
17#include <asm/msr.h> 17#include <asm/msr.h>
18#include <asm/timex.h> 18#include <linux/timex.h>
19#include <asm/io.h> 19#include <linux/io.h>
20 20
21 21#define POWERNOW_IOPORT 0xfff0 /* it doesn't matter where, as long
22#define POWERNOW_IOPORT 0xfff0 /* it doesn't matter where, as long 22 as it is unused */
23 as it is unused */
24 23
25static unsigned int busfreq; /* FSB, in 10 kHz */ 24static unsigned int busfreq; /* FSB, in 10 kHz */
26static unsigned int max_multiplier; 25static unsigned int max_multiplier;
@@ -53,7 +52,7 @@ static int powernow_k6_get_cpu_multiplier(void)
53 52
54 msrval = POWERNOW_IOPORT + 0x1; 53 msrval = POWERNOW_IOPORT + 0x1;
55 wrmsr(MSR_K6_EPMR, msrval, 0); /* enable the PowerNow port */ 54 wrmsr(MSR_K6_EPMR, msrval, 0); /* enable the PowerNow port */
56 invalue=inl(POWERNOW_IOPORT + 0x8); 55 invalue = inl(POWERNOW_IOPORT + 0x8);
57 msrval = POWERNOW_IOPORT + 0x0; 56 msrval = POWERNOW_IOPORT + 0x0;
58 wrmsr(MSR_K6_EPMR, msrval, 0); /* disable it again */ 57 wrmsr(MSR_K6_EPMR, msrval, 0); /* disable it again */
59 58
@@ -67,9 +66,9 @@ static int powernow_k6_get_cpu_multiplier(void)
67 * 66 *
68 * Tries to change the PowerNow! multiplier 67 * Tries to change the PowerNow! multiplier
69 */ 68 */
70static void powernow_k6_set_state (unsigned int best_i) 69static void powernow_k6_set_state(unsigned int best_i)
71{ 70{
72 unsigned long outvalue=0, invalue=0; 71 unsigned long outvalue = 0, invalue = 0;
73 unsigned long msrval; 72 unsigned long msrval;
74 struct cpufreq_freqs freqs; 73 struct cpufreq_freqs freqs;
75 74
@@ -90,10 +89,10 @@ static void powernow_k6_set_state (unsigned int best_i)
90 89
91 msrval = POWERNOW_IOPORT + 0x1; 90 msrval = POWERNOW_IOPORT + 0x1;
92 wrmsr(MSR_K6_EPMR, msrval, 0); /* enable the PowerNow port */ 91 wrmsr(MSR_K6_EPMR, msrval, 0); /* enable the PowerNow port */
93 invalue=inl(POWERNOW_IOPORT + 0x8); 92 invalue = inl(POWERNOW_IOPORT + 0x8);
94 invalue = invalue & 0xf; 93 invalue = invalue & 0xf;
95 outvalue = outvalue | invalue; 94 outvalue = outvalue | invalue;
96 outl(outvalue ,(POWERNOW_IOPORT + 0x8)); 95 outl(outvalue , (POWERNOW_IOPORT + 0x8));
97 msrval = POWERNOW_IOPORT + 0x0; 96 msrval = POWERNOW_IOPORT + 0x0;
98 wrmsr(MSR_K6_EPMR, msrval, 0); /* disable it again */ 97 wrmsr(MSR_K6_EPMR, msrval, 0); /* disable it again */
99 98
@@ -124,7 +123,7 @@ static int powernow_k6_verify(struct cpufreq_policy *policy)
124 * 123 *
125 * sets a new CPUFreq policy 124 * sets a new CPUFreq policy
126 */ 125 */
127static int powernow_k6_target (struct cpufreq_policy *policy, 126static int powernow_k6_target(struct cpufreq_policy *policy,
128 unsigned int target_freq, 127 unsigned int target_freq,
129 unsigned int relation) 128 unsigned int relation)
130{ 129{
@@ -152,7 +151,7 @@ static int powernow_k6_cpu_init(struct cpufreq_policy *policy)
152 busfreq = cpu_khz / max_multiplier; 151 busfreq = cpu_khz / max_multiplier;
153 152
154 /* table init */ 153 /* table init */
155 for (i=0; (clock_ratio[i].frequency != CPUFREQ_TABLE_END); i++) { 154 for (i = 0; (clock_ratio[i].frequency != CPUFREQ_TABLE_END); i++) {
156 if (clock_ratio[i].index > max_multiplier) 155 if (clock_ratio[i].index > max_multiplier)
157 clock_ratio[i].frequency = CPUFREQ_ENTRY_INVALID; 156 clock_ratio[i].frequency = CPUFREQ_ENTRY_INVALID;
158 else 157 else
@@ -165,7 +164,7 @@ static int powernow_k6_cpu_init(struct cpufreq_policy *policy)
165 164
166 result = cpufreq_frequency_table_cpuinfo(policy, clock_ratio); 165 result = cpufreq_frequency_table_cpuinfo(policy, clock_ratio);
167 if (result) 166 if (result)
168 return (result); 167 return result;
169 168
170 cpufreq_frequency_table_get_attr(clock_ratio, policy->cpu); 169 cpufreq_frequency_table_get_attr(clock_ratio, policy->cpu);
171 170
@@ -176,8 +175,8 @@ static int powernow_k6_cpu_init(struct cpufreq_policy *policy)
176static int powernow_k6_cpu_exit(struct cpufreq_policy *policy) 175static int powernow_k6_cpu_exit(struct cpufreq_policy *policy)
177{ 176{
178 unsigned int i; 177 unsigned int i;
179 for (i=0; i<8; i++) { 178 for (i = 0; i < 8; i++) {
180 if (i==max_multiplier) 179 if (i == max_multiplier)
181 powernow_k6_set_state(i); 180 powernow_k6_set_state(i);
182 } 181 }
183 cpufreq_frequency_table_put_attr(policy->cpu); 182 cpufreq_frequency_table_put_attr(policy->cpu);
@@ -189,7 +188,7 @@ static unsigned int powernow_k6_get(unsigned int cpu)
189 return busfreq * powernow_k6_get_cpu_multiplier(); 188 return busfreq * powernow_k6_get_cpu_multiplier();
190} 189}
191 190
192static struct freq_attr* powernow_k6_attr[] = { 191static struct freq_attr *powernow_k6_attr[] = {
193 &cpufreq_freq_attr_scaling_available_freqs, 192 &cpufreq_freq_attr_scaling_available_freqs,
194 NULL, 193 NULL,
195}; 194};
@@ -227,7 +226,7 @@ static int __init powernow_k6_init(void)
227 } 226 }
228 227
229 if (cpufreq_register_driver(&powernow_k6_driver)) { 228 if (cpufreq_register_driver(&powernow_k6_driver)) {
230 release_region (POWERNOW_IOPORT, 16); 229 release_region(POWERNOW_IOPORT, 16);
231 return -EINVAL; 230 return -EINVAL;
232 } 231 }
233 232
@@ -243,13 +242,13 @@ static int __init powernow_k6_init(void)
243static void __exit powernow_k6_exit(void) 242static void __exit powernow_k6_exit(void)
244{ 243{
245 cpufreq_unregister_driver(&powernow_k6_driver); 244 cpufreq_unregister_driver(&powernow_k6_driver);
246 release_region (POWERNOW_IOPORT, 16); 245 release_region(POWERNOW_IOPORT, 16);
247} 246}
248 247
249 248
250MODULE_AUTHOR ("Arjan van de Ven <arjanv@redhat.com>, Dave Jones <davej@codemonkey.org.uk>, Dominik Brodowski <linux@brodo.de>"); 249MODULE_AUTHOR("Arjan van de Ven <arjanv@redhat.com>, Dave Jones <davej@codemonkey.org.uk>, Dominik Brodowski <linux@brodo.de>");
251MODULE_DESCRIPTION ("PowerNow! driver for AMD K6-2+ / K6-3+ processors."); 250MODULE_DESCRIPTION("PowerNow! driver for AMD K6-2+ / K6-3+ processors.");
252MODULE_LICENSE ("GPL"); 251MODULE_LICENSE("GPL");
253 252
254module_init(powernow_k6_init); 253module_init(powernow_k6_init);
255module_exit(powernow_k6_exit); 254module_exit(powernow_k6_exit);
diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
index 15e13c01cc36..3b5f06423e77 100644
--- a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
+++ b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
@@ -26,7 +26,7 @@
26#include <asm/cpufeature.h> 26#include <asm/cpufeature.h>
27 27
28#define PFX "speedstep-centrino: " 28#define PFX "speedstep-centrino: "
29#define MAINTAINER "cpufreq@lists.linux.org.uk" 29#define MAINTAINER "cpufreq@vger.kernel.org"
30 30
31#define dprintk(msg...) \ 31#define dprintk(msg...) \
32 cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "speedstep-centrino", msg) 32 cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "speedstep-centrino", msg)
diff --git a/arch/x86/kernel/cpu/cyrix.c b/arch/x86/kernel/cpu/cyrix.c
index e710a21bb6e8..ffd0f5ed071a 100644
--- a/arch/x86/kernel/cpu/cyrix.c
+++ b/arch/x86/kernel/cpu/cyrix.c
@@ -15,13 +15,11 @@
15/* 15/*
16 * Read NSC/Cyrix DEVID registers (DIR) to get more detailed info. about the CPU 16 * Read NSC/Cyrix DEVID registers (DIR) to get more detailed info. about the CPU
17 */ 17 */
18static void __cpuinit do_cyrix_devid(unsigned char *dir0, unsigned char *dir1) 18static void __cpuinit __do_cyrix_devid(unsigned char *dir0, unsigned char *dir1)
19{ 19{
20 unsigned char ccr2, ccr3; 20 unsigned char ccr2, ccr3;
21 unsigned long flags;
22 21
23 /* we test for DEVID by checking whether CCR3 is writable */ 22 /* we test for DEVID by checking whether CCR3 is writable */
24 local_irq_save(flags);
25 ccr3 = getCx86(CX86_CCR3); 23 ccr3 = getCx86(CX86_CCR3);
26 setCx86(CX86_CCR3, ccr3 ^ 0x80); 24 setCx86(CX86_CCR3, ccr3 ^ 0x80);
27 getCx86(0xc0); /* dummy to change bus */ 25 getCx86(0xc0); /* dummy to change bus */
@@ -44,9 +42,16 @@ static void __cpuinit do_cyrix_devid(unsigned char *dir0, unsigned char *dir1)
44 *dir0 = getCx86(CX86_DIR0); 42 *dir0 = getCx86(CX86_DIR0);
45 *dir1 = getCx86(CX86_DIR1); 43 *dir1 = getCx86(CX86_DIR1);
46 } 44 }
47 local_irq_restore(flags);
48} 45}
49 46
47static void __cpuinit do_cyrix_devid(unsigned char *dir0, unsigned char *dir1)
48{
49 unsigned long flags;
50
51 local_irq_save(flags);
52 __do_cyrix_devid(dir0, dir1);
53 local_irq_restore(flags);
54}
50/* 55/*
51 * Cx86_dir0_msb is a HACK needed by check_cx686_cpuid/slop in bugs.h in 56 * Cx86_dir0_msb is a HACK needed by check_cx686_cpuid/slop in bugs.h in
52 * order to identify the Cyrix CPU model after we're out of setup.c 57 * order to identify the Cyrix CPU model after we're out of setup.c
@@ -116,7 +121,7 @@ static void __cpuinit set_cx86_reorder(void)
116 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 121 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
117 122
118 /* Load/Store Serialize to mem access disable (=reorder it) */ 123 /* Load/Store Serialize to mem access disable (=reorder it) */
119 setCx86(CX86_PCR0, getCx86(CX86_PCR0) & ~0x80); 124 setCx86_old(CX86_PCR0, getCx86_old(CX86_PCR0) & ~0x80);
120 /* set load/store serialize from 1GB to 4GB */ 125 /* set load/store serialize from 1GB to 4GB */
121 ccr3 |= 0xe0; 126 ccr3 |= 0xe0;
122 setCx86(CX86_CCR3, ccr3); 127 setCx86(CX86_CCR3, ccr3);
@@ -127,11 +132,11 @@ static void __cpuinit set_cx86_memwb(void)
127 printk(KERN_INFO "Enable Memory-Write-back mode on Cyrix/NSC processor.\n"); 132 printk(KERN_INFO "Enable Memory-Write-back mode on Cyrix/NSC processor.\n");
128 133
129 /* CCR2 bit 2: unlock NW bit */ 134 /* CCR2 bit 2: unlock NW bit */
130 setCx86(CX86_CCR2, getCx86(CX86_CCR2) & ~0x04); 135 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) & ~0x04);
131 /* set 'Not Write-through' */ 136 /* set 'Not Write-through' */
132 write_cr0(read_cr0() | X86_CR0_NW); 137 write_cr0(read_cr0() | X86_CR0_NW);
133 /* CCR2 bit 2: lock NW bit and set WT1 */ 138 /* CCR2 bit 2: lock NW bit and set WT1 */
134 setCx86(CX86_CCR2, getCx86(CX86_CCR2) | 0x14); 139 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) | 0x14);
135} 140}
136 141
137/* 142/*
@@ -145,14 +150,14 @@ static void __cpuinit geode_configure(void)
145 local_irq_save(flags); 150 local_irq_save(flags);
146 151
147 /* Suspend on halt power saving and enable #SUSP pin */ 152 /* Suspend on halt power saving and enable #SUSP pin */
148 setCx86(CX86_CCR2, getCx86(CX86_CCR2) | 0x88); 153 setCx86_old(CX86_CCR2, getCx86_old(CX86_CCR2) | 0x88);
149 154
150 ccr3 = getCx86(CX86_CCR3); 155 ccr3 = getCx86(CX86_CCR3);
151 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 156 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
152 157
153 158
154 /* FPU fast, DTE cache, Mem bypass */ 159 /* FPU fast, DTE cache, Mem bypass */
155 setCx86(CX86_CCR4, getCx86(CX86_CCR4) | 0x38); 160 setCx86_old(CX86_CCR4, getCx86_old(CX86_CCR4) | 0x38);
156 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */ 161 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */
157 162
158 set_cx86_memwb(); 163 set_cx86_memwb();
@@ -161,6 +166,24 @@ static void __cpuinit geode_configure(void)
161 local_irq_restore(flags); 166 local_irq_restore(flags);
162} 167}
163 168
169static void __cpuinit early_init_cyrix(struct cpuinfo_x86 *c)
170{
171 unsigned char dir0, dir0_msn, dir1 = 0;
172
173 __do_cyrix_devid(&dir0, &dir1);
174 dir0_msn = dir0 >> 4; /* identifies CPU "family" */
175
176 switch (dir0_msn) {
177 case 3: /* 6x86/6x86L */
178 /* Emulate MTRRs using Cyrix's ARRs. */
179 set_cpu_cap(c, X86_FEATURE_CYRIX_ARR);
180 break;
181 case 5: /* 6x86MX/M II */
182 /* Emulate MTRRs using Cyrix's ARRs. */
183 set_cpu_cap(c, X86_FEATURE_CYRIX_ARR);
184 break;
185 }
186}
164 187
165static void __cpuinit init_cyrix(struct cpuinfo_x86 *c) 188static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
166{ 189{
@@ -268,7 +291,7 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
268 /* GXm supports extended cpuid levels 'ala' AMD */ 291 /* GXm supports extended cpuid levels 'ala' AMD */
269 if (c->cpuid_level == 2) { 292 if (c->cpuid_level == 2) {
270 /* Enable cxMMX extensions (GX1 Datasheet 54) */ 293 /* Enable cxMMX extensions (GX1 Datasheet 54) */
271 setCx86(CX86_CCR7, getCx86(CX86_CCR7) | 1); 294 setCx86_old(CX86_CCR7, getCx86_old(CX86_CCR7) | 1);
272 295
273 /* 296 /*
274 * GXm : 0x30 ... 0x5f GXm datasheet 51 297 * GXm : 0x30 ... 0x5f GXm datasheet 51
@@ -278,7 +301,6 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
278 */ 301 */
279 if ((0x30 <= dir1 && dir1 <= 0x6f) || (0x80 <= dir1 && dir1 <= 0x8f)) 302 if ((0x30 <= dir1 && dir1 <= 0x6f) || (0x80 <= dir1 && dir1 <= 0x8f))
280 geode_configure(); 303 geode_configure();
281 get_model_name(c); /* get CPU marketing name */
282 return; 304 return;
283 } else { /* MediaGX */ 305 } else { /* MediaGX */
284 Cx86_cb[2] = (dir0_lsn & 1) ? '3' : '4'; 306 Cx86_cb[2] = (dir0_lsn & 1) ? '3' : '4';
@@ -291,7 +313,7 @@ static void __cpuinit init_cyrix(struct cpuinfo_x86 *c)
291 if (dir1 > 7) { 313 if (dir1 > 7) {
292 dir0_msn++; /* M II */ 314 dir0_msn++; /* M II */
293 /* Enable MMX extensions (App note 108) */ 315 /* Enable MMX extensions (App note 108) */
294 setCx86(CX86_CCR7, getCx86(CX86_CCR7)|1); 316 setCx86_old(CX86_CCR7, getCx86_old(CX86_CCR7)|1);
295 } else { 317 } else {
296 c->coma_bug = 1; /* 6x86MX, it has the bug. */ 318 c->coma_bug = 1; /* 6x86MX, it has the bug. */
297 } 319 }
@@ -406,7 +428,7 @@ static void __cpuinit cyrix_identify(struct cpuinfo_x86 *c)
406 local_irq_save(flags); 428 local_irq_save(flags);
407 ccr3 = getCx86(CX86_CCR3); 429 ccr3 = getCx86(CX86_CCR3);
408 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */ 430 setCx86(CX86_CCR3, (ccr3 & 0x0f) | 0x10); /* enable MAPEN */
409 setCx86(CX86_CCR4, getCx86(CX86_CCR4) | 0x80); /* enable cpuid */ 431 setCx86_old(CX86_CCR4, getCx86_old(CX86_CCR4) | 0x80); /* enable cpuid */
410 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */ 432 setCx86(CX86_CCR3, ccr3); /* disable MAPEN */
411 local_irq_restore(flags); 433 local_irq_restore(flags);
412 } 434 }
@@ -416,16 +438,19 @@ static void __cpuinit cyrix_identify(struct cpuinfo_x86 *c)
416static struct cpu_dev cyrix_cpu_dev __cpuinitdata = { 438static struct cpu_dev cyrix_cpu_dev __cpuinitdata = {
417 .c_vendor = "Cyrix", 439 .c_vendor = "Cyrix",
418 .c_ident = { "CyrixInstead" }, 440 .c_ident = { "CyrixInstead" },
441 .c_early_init = early_init_cyrix,
419 .c_init = init_cyrix, 442 .c_init = init_cyrix,
420 .c_identify = cyrix_identify, 443 .c_identify = cyrix_identify,
444 .c_x86_vendor = X86_VENDOR_CYRIX,
421}; 445};
422 446
423cpu_vendor_dev_register(X86_VENDOR_CYRIX, &cyrix_cpu_dev); 447cpu_dev_register(cyrix_cpu_dev);
424 448
425static struct cpu_dev nsc_cpu_dev __cpuinitdata = { 449static struct cpu_dev nsc_cpu_dev __cpuinitdata = {
426 .c_vendor = "NSC", 450 .c_vendor = "NSC",
427 .c_ident = { "Geode by NSC" }, 451 .c_ident = { "Geode by NSC" },
428 .c_init = init_nsc, 452 .c_init = init_nsc,
453 .c_x86_vendor = X86_VENDOR_NSC,
429}; 454};
430 455
431cpu_vendor_dev_register(X86_VENDOR_NSC, &nsc_cpu_dev); 456cpu_dev_register(nsc_cpu_dev);
diff --git a/arch/x86/kernel/cpu/feature_names.c b/arch/x86/kernel/cpu/feature_names.c
deleted file mode 100644
index e43ad4ad4cba..000000000000
--- a/arch/x86/kernel/cpu/feature_names.c
+++ /dev/null
@@ -1,83 +0,0 @@
1/*
2 * Strings for the various x86 capability flags.
3 *
4 * This file must not contain any executable code.
5 */
6
7#include <asm/cpufeature.h>
8
9/*
10 * These flag bits must match the definitions in <asm/cpufeature.h>.
11 * NULL means this bit is undefined or reserved; either way it doesn't
12 * have meaning as far as Linux is concerned. Note that it's important
13 * to realize there is a difference between this table and CPUID -- if
14 * applications want to get the raw CPUID data, they should access
15 * /dev/cpu/<cpu_nr>/cpuid instead.
16 */
17const char * const x86_cap_flags[NCAPINTS*32] = {
18 /* Intel-defined */
19 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
20 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
21 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
22 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
23
24 /* AMD-defined */
25 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
26 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
27 NULL, NULL, NULL, "mp", "nx", NULL, "mmxext", NULL,
28 NULL, "fxsr_opt", "pdpe1gb", "rdtscp", NULL, "lm",
29 "3dnowext", "3dnow",
30
31 /* Transmeta-defined */
32 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
33 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
34 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
35 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
36
37 /* Other (Linux-defined) */
38 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
39 NULL, NULL, NULL, NULL,
40 "constant_tsc", "up", NULL, "arch_perfmon",
41 "pebs", "bts", NULL, NULL,
42 "rep_good", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
43 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
44
45 /* Intel-defined (#2) */
46 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
47 "tm2", "ssse3", "cid", NULL, NULL, "cx16", "xtpr", NULL,
48 NULL, NULL, "dca", "sse4_1", "sse4_2", NULL, NULL, "popcnt",
49 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
50
51 /* VIA/Cyrix/Centaur-defined */
52 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
53 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL, NULL,
54 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
55 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
56
57 /* AMD-defined (#2) */
58 "lahf_lm", "cmp_legacy", "svm", "extapic",
59 "cr8_legacy", "abm", "sse4a", "misalignsse",
60 "3dnowprefetch", "osvw", "ibs", "sse5",
61 "skinit", "wdt", NULL, NULL,
62 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
63 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
64
65 /* Auxiliary (Linux-defined) */
66 "ida", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
67 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
68 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
69 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
70};
71
72const char *const x86_power_flags[32] = {
73 "ts", /* temperature sensor */
74 "fid", /* frequency id control */
75 "vid", /* voltage id control */
76 "ttp", /* thermal trip */
77 "tm",
78 "stc",
79 "100mhzsteps",
80 "hwpstate",
81 "", /* tsc invariant mapped to constant_tsc */
82 /* nothing */
83};
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index b75f2569b8f8..99468dbd08da 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -15,6 +15,11 @@
15#include <asm/ds.h> 15#include <asm/ds.h>
16#include <asm/bugs.h> 16#include <asm/bugs.h>
17 17
18#ifdef CONFIG_X86_64
19#include <asm/topology.h>
20#include <asm/numa_64.h>
21#endif
22
18#include "cpu.h" 23#include "cpu.h"
19 24
20#ifdef CONFIG_X86_LOCAL_APIC 25#ifdef CONFIG_X86_LOCAL_APIC
@@ -23,23 +28,22 @@
23#include <mach_apic.h> 28#include <mach_apic.h>
24#endif 29#endif
25 30
26#ifdef CONFIG_X86_INTEL_USERCOPY
27/*
28 * Alignment at which movsl is preferred for bulk memory copies.
29 */
30struct movsl_mask movsl_mask __read_mostly;
31#endif
32
33static void __cpuinit early_init_intel(struct cpuinfo_x86 *c) 31static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
34{ 32{
35 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
36 if (c->x86 == 15 && c->x86_cache_alignment == 64)
37 c->x86_cache_alignment = 128;
38 if ((c->x86 == 0xf && c->x86_model >= 0x03) || 33 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
39 (c->x86 == 0x6 && c->x86_model >= 0x0e)) 34 (c->x86 == 0x6 && c->x86_model >= 0x0e))
40 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 35 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
36
37#ifdef CONFIG_X86_64
38 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
39#else
40 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
41 if (c->x86 == 15 && c->x86_cache_alignment == 64)
42 c->x86_cache_alignment = 128;
43#endif
41} 44}
42 45
46#ifdef CONFIG_X86_32
43/* 47/*
44 * Early probe support logic for ppro memory erratum #50 48 * Early probe support logic for ppro memory erratum #50
45 * 49 *
@@ -59,15 +63,54 @@ int __cpuinit ppro_with_ram_bug(void)
59 return 0; 63 return 0;
60} 64}
61 65
66#ifdef CONFIG_X86_F00F_BUG
67static void __cpuinit trap_init_f00f_bug(void)
68{
69 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
62 70
63/* 71 /*
64 * P4 Xeon errata 037 workaround. 72 * Update the IDT descriptor and reload the IDT so that
65 * Hardware prefetcher may cause stale data to be loaded into the cache. 73 * it uses the read-only mapped virtual address.
66 */ 74 */
67static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c) 75 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
76 load_idt(&idt_descr);
77}
78#endif
79
80static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
68{ 81{
69 unsigned long lo, hi; 82 unsigned long lo, hi;
70 83
84#ifdef CONFIG_X86_F00F_BUG
85 /*
86 * All current models of Pentium and Pentium with MMX technology CPUs
87 * have the F0 0F bug, which lets nonprivileged users lock up the system.
88 * Note that the workaround only should be initialized once...
89 */
90 c->f00f_bug = 0;
91 if (!paravirt_enabled() && c->x86 == 5) {
92 static int f00f_workaround_enabled;
93
94 c->f00f_bug = 1;
95 if (!f00f_workaround_enabled) {
96 trap_init_f00f_bug();
97 printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
98 f00f_workaround_enabled = 1;
99 }
100 }
101#endif
102
103 /*
104 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
105 * model 3 mask 3
106 */
107 if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
108 clear_cpu_cap(c, X86_FEATURE_SEP);
109
110 /*
111 * P4 Xeon errata 037 workaround.
112 * Hardware prefetcher may cause stale data to be loaded into the cache.
113 */
71 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) { 114 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
72 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi); 115 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
73 if ((lo & (1<<9)) == 0) { 116 if ((lo & (1<<9)) == 0) {
@@ -77,13 +120,68 @@ static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c)
77 wrmsr (MSR_IA32_MISC_ENABLE, lo, hi); 120 wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
78 } 121 }
79 } 122 }
123
124 /*
125 * See if we have a good local APIC by checking for buggy Pentia,
126 * i.e. all B steppings and the C2 stepping of P54C when using their
127 * integrated APIC (see 11AP erratum in "Pentium Processor
128 * Specification Update").
129 */
130 if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
131 (c->x86_mask < 0x6 || c->x86_mask == 0xb))
132 set_cpu_cap(c, X86_FEATURE_11AP);
133
134
135#ifdef CONFIG_X86_INTEL_USERCOPY
136 /*
137 * Set up the preferred alignment for movsl bulk memory moves
138 */
139 switch (c->x86) {
140 case 4: /* 486: untested */
141 break;
142 case 5: /* Old Pentia: untested */
143 break;
144 case 6: /* PII/PIII only like movsl with 8-byte alignment */
145 movsl_mask.mask = 7;
146 break;
147 case 15: /* P4 is OK down to 8-byte alignment */
148 movsl_mask.mask = 7;
149 break;
150 }
151#endif
152
153#ifdef CONFIG_X86_NUMAQ
154 numaq_tsc_disable();
155#endif
80} 156}
157#else
158static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
159{
160}
161#endif
81 162
163static void __cpuinit srat_detect_node(void)
164{
165#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
166 unsigned node;
167 int cpu = smp_processor_id();
168 int apicid = hard_smp_processor_id();
169
170 /* Don't do the funky fallback heuristics the AMD version employs
171 for now. */
172 node = apicid_to_node[apicid];
173 if (node == NUMA_NO_NODE || !node_online(node))
174 node = first_node(node_online_map);
175 numa_set_node(cpu, node);
176
177 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
178#endif
179}
82 180
83/* 181/*
84 * find out the number of processor cores on the die 182 * find out the number of processor cores on the die
85 */ 183 */
86static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c) 184static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
87{ 185{
88 unsigned int eax, ebx, ecx, edx; 186 unsigned int eax, ebx, ecx, edx;
89 187
@@ -98,45 +196,51 @@ static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c)
98 return 1; 196 return 1;
99} 197}
100 198
101#ifdef CONFIG_X86_F00F_BUG 199static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
102static void __cpuinit trap_init_f00f_bug(void)
103{ 200{
104 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO); 201 /* Intel VMX MSR indicated features */
105 202#define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW 0x00200000
106 /* 203#define X86_VMX_FEATURE_PROC_CTLS_VNMI 0x00400000
107 * Update the IDT descriptor and reload the IDT so that 204#define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS 0x80000000
108 * it uses the read-only mapped virtual address. 205#define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC 0x00000001
109 */ 206#define X86_VMX_FEATURE_PROC_CTLS2_EPT 0x00000002
110 idt_descr.address = fix_to_virt(FIX_F00F_IDT); 207#define X86_VMX_FEATURE_PROC_CTLS2_VPID 0x00000020
111 load_idt(&idt_descr); 208
209 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
210
211 clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
212 clear_cpu_cap(c, X86_FEATURE_VNMI);
213 clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
214 clear_cpu_cap(c, X86_FEATURE_EPT);
215 clear_cpu_cap(c, X86_FEATURE_VPID);
216
217 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
218 msr_ctl = vmx_msr_high | vmx_msr_low;
219 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
220 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
221 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
222 set_cpu_cap(c, X86_FEATURE_VNMI);
223 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
224 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
225 vmx_msr_low, vmx_msr_high);
226 msr_ctl2 = vmx_msr_high | vmx_msr_low;
227 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
228 (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
229 set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
230 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
231 set_cpu_cap(c, X86_FEATURE_EPT);
232 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
233 set_cpu_cap(c, X86_FEATURE_VPID);
234 }
112} 235}
113#endif
114 236
115static void __cpuinit init_intel(struct cpuinfo_x86 *c) 237static void __cpuinit init_intel(struct cpuinfo_x86 *c)
116{ 238{
117 unsigned int l2 = 0; 239 unsigned int l2 = 0;
118 char *p = NULL;
119 240
120 early_init_intel(c); 241 early_init_intel(c);
121 242
122#ifdef CONFIG_X86_F00F_BUG 243 intel_workarounds(c);
123 /*
124 * All current models of Pentium and Pentium with MMX technology CPUs
125 * have the F0 0F bug, which lets nonprivileged users lock up the system.
126 * Note that the workaround only should be initialized once...
127 */
128 c->f00f_bug = 0;
129 if (!paravirt_enabled() && c->x86 == 5) {
130 static int f00f_workaround_enabled;
131
132 c->f00f_bug = 1;
133 if (!f00f_workaround_enabled) {
134 trap_init_f00f_bug();
135 printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
136 f00f_workaround_enabled = 1;
137 }
138 }
139#endif
140 244
141 l2 = init_intel_cacheinfo(c); 245 l2 = init_intel_cacheinfo(c);
142 if (c->cpuid_level > 9) { 246 if (c->cpuid_level > 9) {
@@ -146,16 +250,32 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
146 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON); 250 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
147 } 251 }
148 252
149 /* SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until model 3 mask 3 */ 253 if (cpu_has_xmm2)
150 if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633) 254 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
151 clear_cpu_cap(c, X86_FEATURE_SEP); 255 if (cpu_has_ds) {
256 unsigned int l1;
257 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
258 if (!(l1 & (1<<11)))
259 set_cpu_cap(c, X86_FEATURE_BTS);
260 if (!(l1 & (1<<12)))
261 set_cpu_cap(c, X86_FEATURE_PEBS);
262 ds_init_intel(c);
263 }
152 264
265#ifdef CONFIG_X86_64
266 if (c->x86 == 15)
267 c->x86_cache_alignment = c->x86_clflush_size * 2;
268 if (c->x86 == 6)
269 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
270#else
153 /* 271 /*
154 * Names for the Pentium II/Celeron processors 272 * Names for the Pentium II/Celeron processors
155 * detectable only by also checking the cache size. 273 * detectable only by also checking the cache size.
156 * Dixon is NOT a Celeron. 274 * Dixon is NOT a Celeron.
157 */ 275 */
158 if (c->x86 == 6) { 276 if (c->x86 == 6) {
277 char *p = NULL;
278
159 switch (c->x86_model) { 279 switch (c->x86_model) {
160 case 5: 280 case 5:
161 if (c->x86_mask == 0) { 281 if (c->x86_mask == 0) {
@@ -178,70 +298,41 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
178 p = "Celeron (Coppermine)"; 298 p = "Celeron (Coppermine)";
179 break; 299 break;
180 } 300 }
181 }
182
183 if (p)
184 strcpy(c->x86_model_id, p);
185
186 c->x86_max_cores = num_cpu_cores(c);
187
188 detect_ht(c);
189 301
190 /* Work around errata */ 302 if (p)
191 Intel_errata_workarounds(c); 303 strcpy(c->x86_model_id, p);
192
193#ifdef CONFIG_X86_INTEL_USERCOPY
194 /*
195 * Set up the preferred alignment for movsl bulk memory moves
196 */
197 switch (c->x86) {
198 case 4: /* 486: untested */
199 break;
200 case 5: /* Old Pentia: untested */
201 break;
202 case 6: /* PII/PIII only like movsl with 8-byte alignment */
203 movsl_mask.mask = 7;
204 break;
205 case 15: /* P4 is OK down to 8-byte alignment */
206 movsl_mask.mask = 7;
207 break;
208 } 304 }
209#endif
210 305
211 if (cpu_has_xmm2) 306 if (c->x86 == 15)
212 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
213 if (c->x86 == 15) {
214 set_cpu_cap(c, X86_FEATURE_P4); 307 set_cpu_cap(c, X86_FEATURE_P4);
215 }
216 if (c->x86 == 6) 308 if (c->x86 == 6)
217 set_cpu_cap(c, X86_FEATURE_P3); 309 set_cpu_cap(c, X86_FEATURE_P3);
218 if (cpu_has_ds) {
219 unsigned int l1;
220 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
221 if (!(l1 & (1<<11)))
222 set_cpu_cap(c, X86_FEATURE_BTS);
223 if (!(l1 & (1<<12)))
224 set_cpu_cap(c, X86_FEATURE_PEBS);
225 }
226 310
227 if (cpu_has_bts) 311 if (cpu_has_bts)
228 ds_init_intel(c); 312 ptrace_bts_init_intel(c);
229 313
230 /* 314#endif
231 * See if we have a good local APIC by checking for buggy Pentia,
232 * i.e. all B steppings and the C2 stepping of P54C when using their
233 * integrated APIC (see 11AP erratum in "Pentium Processor
234 * Specification Update").
235 */
236 if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
237 (c->x86_mask < 0x6 || c->x86_mask == 0xb))
238 set_cpu_cap(c, X86_FEATURE_11AP);
239 315
240#ifdef CONFIG_X86_NUMAQ 316 detect_extended_topology(c);
241 numaq_tsc_disable(); 317 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
318 /*
319 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
320 * detection.
321 */
322 c->x86_max_cores = intel_num_cpu_cores(c);
323#ifdef CONFIG_X86_32
324 detect_ht(c);
242#endif 325#endif
326 }
327
328 /* Work around errata */
329 srat_detect_node();
330
331 if (cpu_has(c, X86_FEATURE_VMX))
332 detect_vmx_virtcap(c);
243} 333}
244 334
335#ifdef CONFIG_X86_32
245static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size) 336static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
246{ 337{
247 /* 338 /*
@@ -254,10 +345,12 @@ static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned i
254 size = 256; 345 size = 256;
255 return size; 346 return size;
256} 347}
348#endif
257 349
258static struct cpu_dev intel_cpu_dev __cpuinitdata = { 350static struct cpu_dev intel_cpu_dev __cpuinitdata = {
259 .c_vendor = "Intel", 351 .c_vendor = "Intel",
260 .c_ident = { "GenuineIntel" }, 352 .c_ident = { "GenuineIntel" },
353#ifdef CONFIG_X86_32
261 .c_models = { 354 .c_models = {
262 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names = 355 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
263 { 356 {
@@ -307,76 +400,12 @@ static struct cpu_dev intel_cpu_dev __cpuinitdata = {
307 } 400 }
308 }, 401 },
309 }, 402 },
403 .c_size_cache = intel_size_cache,
404#endif
310 .c_early_init = early_init_intel, 405 .c_early_init = early_init_intel,
311 .c_init = init_intel, 406 .c_init = init_intel,
312 .c_size_cache = intel_size_cache, 407 .c_x86_vendor = X86_VENDOR_INTEL,
313}; 408};
314 409
315cpu_vendor_dev_register(X86_VENDOR_INTEL, &intel_cpu_dev); 410cpu_dev_register(intel_cpu_dev);
316
317#ifndef CONFIG_X86_CMPXCHG
318unsigned long cmpxchg_386_u8(volatile void *ptr, u8 old, u8 new)
319{
320 u8 prev;
321 unsigned long flags;
322
323 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
324 local_irq_save(flags);
325 prev = *(u8 *)ptr;
326 if (prev == old)
327 *(u8 *)ptr = new;
328 local_irq_restore(flags);
329 return prev;
330}
331EXPORT_SYMBOL(cmpxchg_386_u8);
332
333unsigned long cmpxchg_386_u16(volatile void *ptr, u16 old, u16 new)
334{
335 u16 prev;
336 unsigned long flags;
337
338 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
339 local_irq_save(flags);
340 prev = *(u16 *)ptr;
341 if (prev == old)
342 *(u16 *)ptr = new;
343 local_irq_restore(flags);
344 return prev;
345}
346EXPORT_SYMBOL(cmpxchg_386_u16);
347
348unsigned long cmpxchg_386_u32(volatile void *ptr, u32 old, u32 new)
349{
350 u32 prev;
351 unsigned long flags;
352
353 /* Poor man's cmpxchg for 386. Unsuitable for SMP */
354 local_irq_save(flags);
355 prev = *(u32 *)ptr;
356 if (prev == old)
357 *(u32 *)ptr = new;
358 local_irq_restore(flags);
359 return prev;
360}
361EXPORT_SYMBOL(cmpxchg_386_u32);
362#endif
363
364#ifndef CONFIG_X86_CMPXCHG64
365unsigned long long cmpxchg_486_u64(volatile void *ptr, u64 old, u64 new)
366{
367 u64 prev;
368 unsigned long flags;
369
370 /* Poor man's cmpxchg8b for 386 and 486. Unsuitable for SMP */
371 local_irq_save(flags);
372 prev = *(u64 *)ptr;
373 if (prev == old)
374 *(u64 *)ptr = new;
375 local_irq_restore(flags);
376 return prev;
377}
378EXPORT_SYMBOL(cmpxchg_486_u64);
379#endif
380
381/* arch_initcall(intel_cpu_init); */
382 411
diff --git a/arch/x86/kernel/cpu/intel_64.c b/arch/x86/kernel/cpu/intel_64.c
deleted file mode 100644
index 1019c58d39f0..000000000000
--- a/arch/x86/kernel/cpu/intel_64.c
+++ /dev/null
@@ -1,95 +0,0 @@
1#include <linux/init.h>
2#include <linux/smp.h>
3#include <asm/processor.h>
4#include <asm/ptrace.h>
5#include <asm/topology.h>
6#include <asm/numa_64.h>
7
8#include "cpu.h"
9
10static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
11{
12 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
13 (c->x86 == 0x6 && c->x86_model >= 0x0e))
14 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
15
16 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
17}
18
19/*
20 * find out the number of processor cores on the die
21 */
22static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
23{
24 unsigned int eax, t;
25
26 if (c->cpuid_level < 4)
27 return 1;
28
29 cpuid_count(4, 0, &eax, &t, &t, &t);
30
31 if (eax & 0x1f)
32 return ((eax >> 26) + 1);
33 else
34 return 1;
35}
36
37static void __cpuinit srat_detect_node(void)
38{
39#ifdef CONFIG_NUMA
40 unsigned node;
41 int cpu = smp_processor_id();
42 int apicid = hard_smp_processor_id();
43
44 /* Don't do the funky fallback heuristics the AMD version employs
45 for now. */
46 node = apicid_to_node[apicid];
47 if (node == NUMA_NO_NODE || !node_online(node))
48 node = first_node(node_online_map);
49 numa_set_node(cpu, node);
50
51 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
52#endif
53}
54
55static void __cpuinit init_intel(struct cpuinfo_x86 *c)
56{
57 init_intel_cacheinfo(c);
58 if (c->cpuid_level > 9) {
59 unsigned eax = cpuid_eax(10);
60 /* Check for version and the number of counters */
61 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
62 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
63 }
64
65 if (cpu_has_ds) {
66 unsigned int l1, l2;
67 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
68 if (!(l1 & (1<<11)))
69 set_cpu_cap(c, X86_FEATURE_BTS);
70 if (!(l1 & (1<<12)))
71 set_cpu_cap(c, X86_FEATURE_PEBS);
72 }
73
74
75 if (cpu_has_bts)
76 ds_init_intel(c);
77
78 if (c->x86 == 15)
79 c->x86_cache_alignment = c->x86_clflush_size * 2;
80 if (c->x86 == 6)
81 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
82 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
83 c->x86_max_cores = intel_num_cpu_cores(c);
84
85 srat_detect_node();
86}
87
88static struct cpu_dev intel_cpu_dev __cpuinitdata = {
89 .c_vendor = "Intel",
90 .c_ident = { "GenuineIntel" },
91 .c_early_init = early_init_intel,
92 .c_init = init_intel,
93};
94cpu_vendor_dev_register(X86_VENDOR_INTEL, &intel_cpu_dev);
95
diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index 6b0a10b002f1..3f46afbb1cf1 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -1,8 +1,8 @@
1/* 1/*
2 * Routines to indentify caches on Intel CPU. 2 * Routines to indentify caches on Intel CPU.
3 * 3 *
4 * Changes: 4 * Changes:
5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4) 5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure. 6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure.
7 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD. 7 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD.
8 */ 8 */
@@ -13,6 +13,7 @@
13#include <linux/compiler.h> 13#include <linux/compiler.h>
14#include <linux/cpu.h> 14#include <linux/cpu.h>
15#include <linux/sched.h> 15#include <linux/sched.h>
16#include <linux/pci.h>
16 17
17#include <asm/processor.h> 18#include <asm/processor.h>
18#include <asm/smp.h> 19#include <asm/smp.h>
@@ -130,9 +131,18 @@ struct _cpuid4_info {
130 union _cpuid4_leaf_ebx ebx; 131 union _cpuid4_leaf_ebx ebx;
131 union _cpuid4_leaf_ecx ecx; 132 union _cpuid4_leaf_ecx ecx;
132 unsigned long size; 133 unsigned long size;
134 unsigned long can_disable;
133 cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */ 135 cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */
134}; 136};
135 137
138#ifdef CONFIG_PCI
139static struct pci_device_id k8_nb_id[] = {
140 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1103) },
141 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1203) },
142 {}
143};
144#endif
145
136unsigned short num_cache_leaves; 146unsigned short num_cache_leaves;
137 147
138/* AMD doesn't have CPUID4. Emulate it here to report the same 148/* AMD doesn't have CPUID4. Emulate it here to report the same
@@ -182,9 +192,10 @@ static unsigned short assocs[] __cpuinitdata = {
182static unsigned char levels[] __cpuinitdata = { 1, 1, 2, 3 }; 192static unsigned char levels[] __cpuinitdata = { 1, 1, 2, 3 };
183static unsigned char types[] __cpuinitdata = { 1, 2, 3, 3 }; 193static unsigned char types[] __cpuinitdata = { 1, 2, 3, 3 };
184 194
185static void __cpuinit amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax, 195static void __cpuinit
186 union _cpuid4_leaf_ebx *ebx, 196amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
187 union _cpuid4_leaf_ecx *ecx) 197 union _cpuid4_leaf_ebx *ebx,
198 union _cpuid4_leaf_ecx *ecx)
188{ 199{
189 unsigned dummy; 200 unsigned dummy;
190 unsigned line_size, lines_per_tag, assoc, size_in_kb; 201 unsigned line_size, lines_per_tag, assoc, size_in_kb;
@@ -251,27 +262,40 @@ static void __cpuinit amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
251 (ebx->split.ways_of_associativity + 1) - 1; 262 (ebx->split.ways_of_associativity + 1) - 1;
252} 263}
253 264
254static int __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf) 265static void __cpuinit
266amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf)
267{
268 if (index < 3)
269 return;
270 this_leaf->can_disable = 1;
271}
272
273static int
274__cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
255{ 275{
256 union _cpuid4_leaf_eax eax; 276 union _cpuid4_leaf_eax eax;
257 union _cpuid4_leaf_ebx ebx; 277 union _cpuid4_leaf_ebx ebx;
258 union _cpuid4_leaf_ecx ecx; 278 union _cpuid4_leaf_ecx ecx;
259 unsigned edx; 279 unsigned edx;
260 280
261 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) 281 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
262 amd_cpuid4(index, &eax, &ebx, &ecx); 282 amd_cpuid4(index, &eax, &ebx, &ecx);
263 else 283 if (boot_cpu_data.x86 >= 0x10)
264 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx); 284 amd_check_l3_disable(index, this_leaf);
285 } else {
286 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx);
287 }
288
265 if (eax.split.type == CACHE_TYPE_NULL) 289 if (eax.split.type == CACHE_TYPE_NULL)
266 return -EIO; /* better error ? */ 290 return -EIO; /* better error ? */
267 291
268 this_leaf->eax = eax; 292 this_leaf->eax = eax;
269 this_leaf->ebx = ebx; 293 this_leaf->ebx = ebx;
270 this_leaf->ecx = ecx; 294 this_leaf->ecx = ecx;
271 this_leaf->size = (ecx.split.number_of_sets + 1) * 295 this_leaf->size = (ecx.split.number_of_sets + 1) *
272 (ebx.split.coherency_line_size + 1) * 296 (ebx.split.coherency_line_size + 1) *
273 (ebx.split.physical_line_partition + 1) * 297 (ebx.split.physical_line_partition + 1) *
274 (ebx.split.ways_of_associativity + 1); 298 (ebx.split.ways_of_associativity + 1);
275 return 0; 299 return 0;
276} 300}
277 301
@@ -453,7 +477,7 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
453 477
454/* pointer to _cpuid4_info array (for each cache leaf) */ 478/* pointer to _cpuid4_info array (for each cache leaf) */
455static DEFINE_PER_CPU(struct _cpuid4_info *, cpuid4_info); 479static DEFINE_PER_CPU(struct _cpuid4_info *, cpuid4_info);
456#define CPUID4_INFO_IDX(x, y) (&((per_cpu(cpuid4_info, x))[y])) 480#define CPUID4_INFO_IDX(x, y) (&((per_cpu(cpuid4_info, x))[y]))
457 481
458#ifdef CONFIG_SMP 482#ifdef CONFIG_SMP
459static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index) 483static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
@@ -490,7 +514,7 @@ static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index)
490 514
491 this_leaf = CPUID4_INFO_IDX(cpu, index); 515 this_leaf = CPUID4_INFO_IDX(cpu, index);
492 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) { 516 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) {
493 sibling_leaf = CPUID4_INFO_IDX(sibling, index); 517 sibling_leaf = CPUID4_INFO_IDX(sibling, index);
494 cpu_clear(cpu, sibling_leaf->shared_cpu_map); 518 cpu_clear(cpu, sibling_leaf->shared_cpu_map);
495 } 519 }
496} 520}
@@ -572,7 +596,7 @@ struct _index_kobject {
572 596
573/* pointer to array of kobjects for cpuX/cache/indexY */ 597/* pointer to array of kobjects for cpuX/cache/indexY */
574static DEFINE_PER_CPU(struct _index_kobject *, index_kobject); 598static DEFINE_PER_CPU(struct _index_kobject *, index_kobject);
575#define INDEX_KOBJECT_PTR(x, y) (&((per_cpu(index_kobject, x))[y])) 599#define INDEX_KOBJECT_PTR(x, y) (&((per_cpu(index_kobject, x))[y]))
576 600
577#define show_one_plus(file_name, object, val) \ 601#define show_one_plus(file_name, object, val) \
578static ssize_t show_##file_name \ 602static ssize_t show_##file_name \
@@ -637,6 +661,99 @@ static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
637 } 661 }
638} 662}
639 663
664#define to_object(k) container_of(k, struct _index_kobject, kobj)
665#define to_attr(a) container_of(a, struct _cache_attr, attr)
666
667#ifdef CONFIG_PCI
668static struct pci_dev *get_k8_northbridge(int node)
669{
670 struct pci_dev *dev = NULL;
671 int i;
672
673 for (i = 0; i <= node; i++) {
674 do {
675 dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev);
676 if (!dev)
677 break;
678 } while (!pci_match_id(&k8_nb_id[0], dev));
679 if (!dev)
680 break;
681 }
682 return dev;
683}
684#else
685static struct pci_dev *get_k8_northbridge(int node)
686{
687 return NULL;
688}
689#endif
690
691static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf)
692{
693 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
694 struct pci_dev *dev = NULL;
695 ssize_t ret = 0;
696 int i;
697
698 if (!this_leaf->can_disable)
699 return sprintf(buf, "Feature not enabled\n");
700
701 dev = get_k8_northbridge(node);
702 if (!dev) {
703 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
704 return -EINVAL;
705 }
706
707 for (i = 0; i < 2; i++) {
708 unsigned int reg;
709
710 pci_read_config_dword(dev, 0x1BC + i * 4, &reg);
711
712 ret += sprintf(buf, "%sEntry: %d\n", buf, i);
713 ret += sprintf(buf, "%sReads: %s\tNew Entries: %s\n",
714 buf,
715 reg & 0x80000000 ? "Disabled" : "Allowed",
716 reg & 0x40000000 ? "Disabled" : "Allowed");
717 ret += sprintf(buf, "%sSubCache: %x\tIndex: %x\n",
718 buf, (reg & 0x30000) >> 16, reg & 0xfff);
719 }
720 return ret;
721}
722
723static ssize_t
724store_cache_disable(struct _cpuid4_info *this_leaf, const char *buf,
725 size_t count)
726{
727 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
728 struct pci_dev *dev = NULL;
729 unsigned int ret, index, val;
730
731 if (!this_leaf->can_disable)
732 return 0;
733
734 if (strlen(buf) > 15)
735 return -EINVAL;
736
737 ret = sscanf(buf, "%x %x", &index, &val);
738 if (ret != 2)
739 return -EINVAL;
740 if (index > 1)
741 return -EINVAL;
742
743 val |= 0xc0000000;
744 dev = get_k8_northbridge(node);
745 if (!dev) {
746 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
747 return -EINVAL;
748 }
749
750 pci_write_config_dword(dev, 0x1BC + index * 4, val & ~0x40000000);
751 wbinvd();
752 pci_write_config_dword(dev, 0x1BC + index * 4, val);
753
754 return 1;
755}
756
640struct _cache_attr { 757struct _cache_attr {
641 struct attribute attr; 758 struct attribute attr;
642 ssize_t (*show)(struct _cpuid4_info *, char *); 759 ssize_t (*show)(struct _cpuid4_info *, char *);
@@ -657,6 +774,8 @@ define_one_ro(size);
657define_one_ro(shared_cpu_map); 774define_one_ro(shared_cpu_map);
658define_one_ro(shared_cpu_list); 775define_one_ro(shared_cpu_list);
659 776
777static struct _cache_attr cache_disable = __ATTR(cache_disable, 0644, show_cache_disable, store_cache_disable);
778
660static struct attribute * default_attrs[] = { 779static struct attribute * default_attrs[] = {
661 &type.attr, 780 &type.attr,
662 &level.attr, 781 &level.attr,
@@ -667,12 +786,10 @@ static struct attribute * default_attrs[] = {
667 &size.attr, 786 &size.attr,
668 &shared_cpu_map.attr, 787 &shared_cpu_map.attr,
669 &shared_cpu_list.attr, 788 &shared_cpu_list.attr,
789 &cache_disable.attr,
670 NULL 790 NULL
671}; 791};
672 792
673#define to_object(k) container_of(k, struct _index_kobject, kobj)
674#define to_attr(a) container_of(a, struct _cache_attr, attr)
675
676static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf) 793static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
677{ 794{
678 struct _cache_attr *fattr = to_attr(attr); 795 struct _cache_attr *fattr = to_attr(attr);
@@ -682,14 +799,22 @@ static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
682 ret = fattr->show ? 799 ret = fattr->show ?
683 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index), 800 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
684 buf) : 801 buf) :
685 0; 802 0;
686 return ret; 803 return ret;
687} 804}
688 805
689static ssize_t store(struct kobject * kobj, struct attribute * attr, 806static ssize_t store(struct kobject * kobj, struct attribute * attr,
690 const char * buf, size_t count) 807 const char * buf, size_t count)
691{ 808{
692 return 0; 809 struct _cache_attr *fattr = to_attr(attr);
810 struct _index_kobject *this_leaf = to_object(kobj);
811 ssize_t ret;
812
813 ret = fattr->store ?
814 fattr->store(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
815 buf, count) :
816 0;
817 return ret;
693} 818}
694 819
695static struct sysfs_ops sysfs_ops = { 820static struct sysfs_ops sysfs_ops = {
diff --git a/arch/x86/kernel/cpu/mcheck/mce_64.c b/arch/x86/kernel/cpu/mcheck/mce_64.c
index 726a5fcdf341..4b031a4ac856 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_64.c
@@ -860,7 +860,7 @@ error:
860 return err; 860 return err;
861} 861}
862 862
863static void mce_remove_device(unsigned int cpu) 863static __cpuinit void mce_remove_device(unsigned int cpu)
864{ 864{
865 int i; 865 int i;
866 866
diff --git a/arch/x86/kernel/cpu/mkcapflags.pl b/arch/x86/kernel/cpu/mkcapflags.pl
new file mode 100644
index 000000000000..dfea390e1608
--- /dev/null
+++ b/arch/x86/kernel/cpu/mkcapflags.pl
@@ -0,0 +1,32 @@
1#!/usr/bin/perl
2#
3# Generate the x86_cap_flags[] array from include/asm-x86/cpufeature.h
4#
5
6($in, $out) = @ARGV;
7
8open(IN, "< $in\0") or die "$0: cannot open: $in: $!\n";
9open(OUT, "> $out\0") or die "$0: cannot create: $out: $!\n";
10
11print OUT "#include <asm/cpufeature.h>\n\n";
12print OUT "const char * const x86_cap_flags[NCAPINTS*32] = {\n";
13
14while (defined($line = <IN>)) {
15 if ($line =~ /^\s*\#\s*define\s+(X86_FEATURE_(\S+))\s+(.*)$/) {
16 $macro = $1;
17 $feature = $2;
18 $tail = $3;
19 if ($tail =~ /\/\*\s*\"([^"]*)\".*\*\//) {
20 $feature = $1;
21 }
22
23 if ($feature ne '') {
24 printf OUT "\t%-32s = \"%s\",\n",
25 "[$macro]", "\L$feature";
26 }
27 }
28}
29print OUT "};\n";
30
31close(IN);
32close(OUT);
diff --git a/arch/x86/kernel/cpu/mtrr/generic.c b/arch/x86/kernel/cpu/mtrr/generic.c
index cb7d3b6a80eb..4e8d77f01eeb 100644
--- a/arch/x86/kernel/cpu/mtrr/generic.c
+++ b/arch/x86/kernel/cpu/mtrr/generic.c
@@ -401,12 +401,7 @@ static void generic_get_mtrr(unsigned int reg, unsigned long *base,
401 tmp |= ~((1<<(hi - 1)) - 1); 401 tmp |= ~((1<<(hi - 1)) - 1);
402 402
403 if (tmp != mask_lo) { 403 if (tmp != mask_lo) {
404 static int once = 1; 404 WARN_ONCE(1, KERN_INFO "mtrr: your BIOS has set up an incorrect mask, fixing it up.\n");
405
406 if (once) {
407 printk(KERN_INFO "mtrr: your BIOS has set up an incorrect mask, fixing it up.\n");
408 once = 0;
409 }
410 mask_lo = tmp; 405 mask_lo = tmp;
411 } 406 }
412 } 407 }
diff --git a/arch/x86/kernel/cpu/mtrr/if.c b/arch/x86/kernel/cpu/mtrr/if.c
index 84c480bb3715..4c4214690dd1 100644
--- a/arch/x86/kernel/cpu/mtrr/if.c
+++ b/arch/x86/kernel/cpu/mtrr/if.c
@@ -405,9 +405,9 @@ static int mtrr_seq_show(struct seq_file *seq, void *offset)
405 } 405 }
406 /* RED-PEN: base can be > 32bit */ 406 /* RED-PEN: base can be > 32bit */
407 len += seq_printf(seq, 407 len += seq_printf(seq,
408 "reg%02i: base=0x%05lx000 (%4luMB), size=%4lu%cB: %s, count=%d\n", 408 "reg%02i: base=0x%06lx000 (%5luMB), size=%5lu%cB, count=%d: %s\n",
409 i, base, base >> (20 - PAGE_SHIFT), size, factor, 409 i, base, base >> (20 - PAGE_SHIFT), size, factor,
410 mtrr_attrib_to_str(type), mtrr_usage_table[i]); 410 mtrr_usage_table[i], mtrr_attrib_to_str(type));
411 } 411 }
412 } 412 }
413 return 0; 413 return 0;
diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c
index b117d7f8a564..c78c04821ea1 100644
--- a/arch/x86/kernel/cpu/mtrr/main.c
+++ b/arch/x86/kernel/cpu/mtrr/main.c
@@ -729,7 +729,7 @@ struct var_mtrr_range_state {
729 mtrr_type type; 729 mtrr_type type;
730}; 730};
731 731
732struct var_mtrr_range_state __initdata range_state[RANGE_NUM]; 732static struct var_mtrr_range_state __initdata range_state[RANGE_NUM];
733static int __initdata debug_print; 733static int __initdata debug_print;
734 734
735static int __init 735static int __init
@@ -759,7 +759,8 @@ x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
759 /* take out UC ranges */ 759 /* take out UC ranges */
760 for (i = 0; i < num_var_ranges; i++) { 760 for (i = 0; i < num_var_ranges; i++) {
761 type = range_state[i].type; 761 type = range_state[i].type;
762 if (type != MTRR_TYPE_UNCACHABLE) 762 if (type != MTRR_TYPE_UNCACHABLE &&
763 type != MTRR_TYPE_WRPROT)
763 continue; 764 continue;
764 size = range_state[i].size_pfn; 765 size = range_state[i].size_pfn;
765 if (!size) 766 if (!size)
@@ -834,7 +835,14 @@ static int __init enable_mtrr_cleanup_setup(char *str)
834 enable_mtrr_cleanup = 1; 835 enable_mtrr_cleanup = 1;
835 return 0; 836 return 0;
836} 837}
837early_param("enble_mtrr_cleanup", enable_mtrr_cleanup_setup); 838early_param("enable_mtrr_cleanup", enable_mtrr_cleanup_setup);
839
840static int __init mtrr_cleanup_debug_setup(char *str)
841{
842 debug_print = 1;
843 return 0;
844}
845early_param("mtrr_cleanup_debug", mtrr_cleanup_debug_setup);
838 846
839struct var_mtrr_state { 847struct var_mtrr_state {
840 unsigned long range_startk; 848 unsigned long range_startk;
@@ -898,6 +906,27 @@ set_var_mtrr_all(unsigned int address_bits)
898 } 906 }
899} 907}
900 908
909static unsigned long to_size_factor(unsigned long sizek, char *factorp)
910{
911 char factor;
912 unsigned long base = sizek;
913
914 if (base & ((1<<10) - 1)) {
915 /* not MB alignment */
916 factor = 'K';
917 } else if (base & ((1<<20) - 1)){
918 factor = 'M';
919 base >>= 10;
920 } else {
921 factor = 'G';
922 base >>= 20;
923 }
924
925 *factorp = factor;
926
927 return base;
928}
929
901static unsigned int __init 930static unsigned int __init
902range_to_mtrr(unsigned int reg, unsigned long range_startk, 931range_to_mtrr(unsigned int reg, unsigned long range_startk,
903 unsigned long range_sizek, unsigned char type) 932 unsigned long range_sizek, unsigned char type)
@@ -919,13 +948,21 @@ range_to_mtrr(unsigned int reg, unsigned long range_startk,
919 align = max_align; 948 align = max_align;
920 949
921 sizek = 1 << align; 950 sizek = 1 << align;
922 if (debug_print) 951 if (debug_print) {
952 char start_factor = 'K', size_factor = 'K';
953 unsigned long start_base, size_base;
954
955 start_base = to_size_factor(range_startk, &start_factor),
956 size_base = to_size_factor(sizek, &size_factor),
957
923 printk(KERN_DEBUG "Setting variable MTRR %d, " 958 printk(KERN_DEBUG "Setting variable MTRR %d, "
924 "base: %ldMB, range: %ldMB, type %s\n", 959 "base: %ld%cB, range: %ld%cB, type %s\n",
925 reg, range_startk >> 10, sizek >> 10, 960 reg, start_base, start_factor,
961 size_base, size_factor,
926 (type == MTRR_TYPE_UNCACHABLE)?"UC": 962 (type == MTRR_TYPE_UNCACHABLE)?"UC":
927 ((type == MTRR_TYPE_WRBACK)?"WB":"Other") 963 ((type == MTRR_TYPE_WRBACK)?"WB":"Other")
928 ); 964 );
965 }
929 save_var_mtrr(reg++, range_startk, sizek, type); 966 save_var_mtrr(reg++, range_startk, sizek, type);
930 range_startk += sizek; 967 range_startk += sizek;
931 range_sizek -= sizek; 968 range_sizek -= sizek;
@@ -970,6 +1007,8 @@ range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
970 /* try to append some small hole */ 1007 /* try to append some small hole */
971 range0_basek = state->range_startk; 1008 range0_basek = state->range_startk;
972 range0_sizek = ALIGN(state->range_sizek, chunk_sizek); 1009 range0_sizek = ALIGN(state->range_sizek, chunk_sizek);
1010
1011 /* no increase */
973 if (range0_sizek == state->range_sizek) { 1012 if (range0_sizek == state->range_sizek) {
974 if (debug_print) 1013 if (debug_print)
975 printk(KERN_DEBUG "rangeX: %016lx - %016lx\n", 1014 printk(KERN_DEBUG "rangeX: %016lx - %016lx\n",
@@ -980,13 +1019,40 @@ range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
980 return 0; 1019 return 0;
981 } 1020 }
982 1021
983 range0_sizek -= chunk_sizek; 1022 /* only cut back, when it is not the last */
984 if (range0_sizek && sizek) { 1023 if (sizek) {
985 while (range0_basek + range0_sizek > (basek + sizek)) { 1024 while (range0_basek + range0_sizek > (basek + sizek)) {
986 range0_sizek -= chunk_sizek; 1025 if (range0_sizek >= chunk_sizek)
987 if (!range0_sizek) 1026 range0_sizek -= chunk_sizek;
988 break; 1027 else
989 } 1028 range0_sizek = 0;
1029
1030 if (!range0_sizek)
1031 break;
1032 }
1033 }
1034
1035second_try:
1036 range_basek = range0_basek + range0_sizek;
1037
1038 /* one hole in the middle */
1039 if (range_basek > basek && range_basek <= (basek + sizek))
1040 second_sizek = range_basek - basek;
1041
1042 if (range0_sizek > state->range_sizek) {
1043
1044 /* one hole in middle or at end */
1045 hole_sizek = range0_sizek - state->range_sizek - second_sizek;
1046
1047 /* hole size should be less than half of range0 size */
1048 if (hole_sizek >= (range0_sizek >> 1) &&
1049 range0_sizek >= chunk_sizek) {
1050 range0_sizek -= chunk_sizek;
1051 second_sizek = 0;
1052 hole_sizek = 0;
1053
1054 goto second_try;
1055 }
990 } 1056 }
991 1057
992 if (range0_sizek) { 1058 if (range0_sizek) {
@@ -996,50 +1062,28 @@ range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
996 (range0_basek + range0_sizek)<<10); 1062 (range0_basek + range0_sizek)<<10);
997 state->reg = range_to_mtrr(state->reg, range0_basek, 1063 state->reg = range_to_mtrr(state->reg, range0_basek,
998 range0_sizek, MTRR_TYPE_WRBACK); 1064 range0_sizek, MTRR_TYPE_WRBACK);
999
1000 }
1001
1002 range_basek = range0_basek + range0_sizek;
1003 range_sizek = chunk_sizek;
1004
1005 if (range_basek + range_sizek > basek &&
1006 range_basek + range_sizek <= (basek + sizek)) {
1007 /* one hole */
1008 second_basek = basek;
1009 second_sizek = range_basek + range_sizek - basek;
1010 } 1065 }
1011 1066
1012 /* if last piece, only could one hole near end */ 1067 if (range0_sizek < state->range_sizek) {
1013 if ((second_basek || !basek) && 1068 /* need to handle left over */
1014 range_sizek - (state->range_sizek - range0_sizek) - second_sizek <
1015 (chunk_sizek >> 1)) {
1016 /*
1017 * one hole in middle (second_sizek is 0) or at end
1018 * (second_sizek is 0 )
1019 */
1020 hole_sizek = range_sizek - (state->range_sizek - range0_sizek)
1021 - second_sizek;
1022 hole_basek = range_basek + range_sizek - hole_sizek
1023 - second_sizek;
1024 } else {
1025 /* fallback for big hole, or several holes */
1026 range_sizek = state->range_sizek - range0_sizek; 1069 range_sizek = state->range_sizek - range0_sizek;
1027 second_basek = 0; 1070
1028 second_sizek = 0; 1071 if (debug_print)
1072 printk(KERN_DEBUG "range: %016lx - %016lx\n",
1073 range_basek<<10,
1074 (range_basek + range_sizek)<<10);
1075 state->reg = range_to_mtrr(state->reg, range_basek,
1076 range_sizek, MTRR_TYPE_WRBACK);
1029 } 1077 }
1030 1078
1031 if (debug_print)
1032 printk(KERN_DEBUG "range: %016lx - %016lx\n", range_basek<<10,
1033 (range_basek + range_sizek)<<10);
1034 state->reg = range_to_mtrr(state->reg, range_basek, range_sizek,
1035 MTRR_TYPE_WRBACK);
1036 if (hole_sizek) { 1079 if (hole_sizek) {
1080 hole_basek = range_basek - hole_sizek - second_sizek;
1037 if (debug_print) 1081 if (debug_print)
1038 printk(KERN_DEBUG "hole: %016lx - %016lx\n", 1082 printk(KERN_DEBUG "hole: %016lx - %016lx\n",
1039 hole_basek<<10, (hole_basek + hole_sizek)<<10); 1083 hole_basek<<10,
1040 state->reg = range_to_mtrr(state->reg, hole_basek, hole_sizek, 1084 (hole_basek + hole_sizek)<<10);
1041 MTRR_TYPE_UNCACHABLE); 1085 state->reg = range_to_mtrr(state->reg, hole_basek,
1042 1086 hole_sizek, MTRR_TYPE_UNCACHABLE);
1043 } 1087 }
1044 1088
1045 return second_sizek; 1089 return second_sizek;
@@ -1154,11 +1198,11 @@ struct mtrr_cleanup_result {
1154}; 1198};
1155 1199
1156/* 1200/*
1157 * gran_size: 1M, 2M, ..., 2G 1201 * gran_size: 64K, 128K, 256K, 512K, 1M, 2M, ..., 2G
1158 * chunk size: gran_size, ..., 4G 1202 * chunk size: gran_size, ..., 2G
1159 * so we need (2+13)*6 1203 * so we need (1+16)*8
1160 */ 1204 */
1161#define NUM_RESULT 90 1205#define NUM_RESULT 136
1162#define PSHIFT (PAGE_SHIFT - 10) 1206#define PSHIFT (PAGE_SHIFT - 10)
1163 1207
1164static struct mtrr_cleanup_result __initdata result[NUM_RESULT]; 1208static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
@@ -1168,13 +1212,14 @@ static unsigned long __initdata min_loss_pfn[RANGE_NUM];
1168static int __init mtrr_cleanup(unsigned address_bits) 1212static int __init mtrr_cleanup(unsigned address_bits)
1169{ 1213{
1170 unsigned long extra_remove_base, extra_remove_size; 1214 unsigned long extra_remove_base, extra_remove_size;
1171 unsigned long i, base, size, def, dummy; 1215 unsigned long base, size, def, dummy;
1172 mtrr_type type; 1216 mtrr_type type;
1173 int nr_range, nr_range_new; 1217 int nr_range, nr_range_new;
1174 u64 chunk_size, gran_size; 1218 u64 chunk_size, gran_size;
1175 unsigned long range_sums, range_sums_new; 1219 unsigned long range_sums, range_sums_new;
1176 int index_good; 1220 int index_good;
1177 int num_reg_good; 1221 int num_reg_good;
1222 int i;
1178 1223
1179 /* extra one for all 0 */ 1224 /* extra one for all 0 */
1180 int num[MTRR_NUM_TYPES + 1]; 1225 int num[MTRR_NUM_TYPES + 1];
@@ -1204,6 +1249,8 @@ static int __init mtrr_cleanup(unsigned address_bits)
1204 continue; 1249 continue;
1205 if (!size) 1250 if (!size)
1206 type = MTRR_NUM_TYPES; 1251 type = MTRR_NUM_TYPES;
1252 if (type == MTRR_TYPE_WRPROT)
1253 type = MTRR_TYPE_UNCACHABLE;
1207 num[type]++; 1254 num[type]++;
1208 } 1255 }
1209 1256
@@ -1216,23 +1263,57 @@ static int __init mtrr_cleanup(unsigned address_bits)
1216 num_var_ranges - num[MTRR_NUM_TYPES]) 1263 num_var_ranges - num[MTRR_NUM_TYPES])
1217 return 0; 1264 return 0;
1218 1265
1266 /* print original var MTRRs at first, for debugging: */
1267 printk(KERN_DEBUG "original variable MTRRs\n");
1268 for (i = 0; i < num_var_ranges; i++) {
1269 char start_factor = 'K', size_factor = 'K';
1270 unsigned long start_base, size_base;
1271
1272 size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10);
1273 if (!size_base)
1274 continue;
1275
1276 size_base = to_size_factor(size_base, &size_factor),
1277 start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
1278 start_base = to_size_factor(start_base, &start_factor),
1279 type = range_state[i].type;
1280
1281 printk(KERN_DEBUG "reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
1282 i, start_base, start_factor,
1283 size_base, size_factor,
1284 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
1285 ((type == MTRR_TYPE_WRPROT) ? "WP" :
1286 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
1287 );
1288 }
1289
1219 memset(range, 0, sizeof(range)); 1290 memset(range, 0, sizeof(range));
1220 extra_remove_size = 0; 1291 extra_remove_size = 0;
1221 if (mtrr_tom2) { 1292 extra_remove_base = 1 << (32 - PAGE_SHIFT);
1222 extra_remove_base = 1 << (32 - PAGE_SHIFT); 1293 if (mtrr_tom2)
1223 extra_remove_size = 1294 extra_remove_size =
1224 (mtrr_tom2 >> PAGE_SHIFT) - extra_remove_base; 1295 (mtrr_tom2 >> PAGE_SHIFT) - extra_remove_base;
1225 }
1226 nr_range = x86_get_mtrr_mem_range(range, 0, extra_remove_base, 1296 nr_range = x86_get_mtrr_mem_range(range, 0, extra_remove_base,
1227 extra_remove_size); 1297 extra_remove_size);
1298 /*
1299 * [0, 1M) should always be coverred by var mtrr with WB
1300 * and fixed mtrrs should take effective before var mtrr for it
1301 */
1302 nr_range = add_range_with_merge(range, nr_range, 0,
1303 (1ULL<<(20 - PAGE_SHIFT)) - 1);
1304 /* sort the ranges */
1305 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
1306
1228 range_sums = sum_ranges(range, nr_range); 1307 range_sums = sum_ranges(range, nr_range);
1229 printk(KERN_INFO "total RAM coverred: %ldM\n", 1308 printk(KERN_INFO "total RAM coverred: %ldM\n",
1230 range_sums >> (20 - PAGE_SHIFT)); 1309 range_sums >> (20 - PAGE_SHIFT));
1231 1310
1232 if (mtrr_chunk_size && mtrr_gran_size) { 1311 if (mtrr_chunk_size && mtrr_gran_size) {
1233 int num_reg; 1312 int num_reg;
1313 char gran_factor, chunk_factor, lose_factor;
1314 unsigned long gran_base, chunk_base, lose_base;
1234 1315
1235 debug_print = 1; 1316 debug_print++;
1236 /* convert ranges to var ranges state */ 1317 /* convert ranges to var ranges state */
1237 num_reg = x86_setup_var_mtrrs(range, nr_range, mtrr_chunk_size, 1318 num_reg = x86_setup_var_mtrrs(range, nr_range, mtrr_chunk_size,
1238 mtrr_gran_size); 1319 mtrr_gran_size);
@@ -1256,34 +1337,48 @@ static int __init mtrr_cleanup(unsigned address_bits)
1256 result[i].lose_cover_sizek = 1337 result[i].lose_cover_sizek =
1257 (range_sums - range_sums_new) << PSHIFT; 1338 (range_sums - range_sums_new) << PSHIFT;
1258 1339
1259 printk(KERN_INFO "%sgran_size: %ldM \tchunk_size: %ldM \t", 1340 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1260 result[i].bad?"*BAD*":" ", result[i].gran_sizek >> 10, 1341 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1261 result[i].chunk_sizek >> 10); 1342 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1262 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ldM \n", 1343 printk(KERN_INFO "%sgran_size: %ld%c \tchunk_size: %ld%c \t",
1344 result[i].bad?"*BAD*":" ",
1345 gran_base, gran_factor, chunk_base, chunk_factor);
1346 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ld%c\n",
1263 result[i].num_reg, result[i].bad?"-":"", 1347 result[i].num_reg, result[i].bad?"-":"",
1264 result[i].lose_cover_sizek >> 10); 1348 lose_base, lose_factor);
1265 if (!result[i].bad) { 1349 if (!result[i].bad) {
1266 set_var_mtrr_all(address_bits); 1350 set_var_mtrr_all(address_bits);
1267 return 1; 1351 return 1;
1268 } 1352 }
1269 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, " 1353 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, "
1270 "will find optimal one\n"); 1354 "will find optimal one\n");
1271 debug_print = 0; 1355 debug_print--;
1272 memset(result, 0, sizeof(result[0])); 1356 memset(result, 0, sizeof(result[0]));
1273 } 1357 }
1274 1358
1275 i = 0; 1359 i = 0;
1276 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn)); 1360 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
1277 memset(result, 0, sizeof(result)); 1361 memset(result, 0, sizeof(result));
1278 for (gran_size = (1ULL<<20); gran_size < (1ULL<<32); gran_size <<= 1) { 1362 for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) {
1279 for (chunk_size = gran_size; chunk_size < (1ULL<<33); 1363 char gran_factor;
1364 unsigned long gran_base;
1365
1366 if (debug_print)
1367 gran_base = to_size_factor(gran_size >> 10, &gran_factor);
1368
1369 for (chunk_size = gran_size; chunk_size < (1ULL<<32);
1280 chunk_size <<= 1) { 1370 chunk_size <<= 1) {
1281 int num_reg; 1371 int num_reg;
1282 1372
1283 if (debug_print) 1373 if (debug_print) {
1284 printk(KERN_INFO 1374 char chunk_factor;
1285 "\ngran_size: %lldM chunk_size_size: %lldM\n", 1375 unsigned long chunk_base;
1286 gran_size >> 20, chunk_size >> 20); 1376
1377 chunk_base = to_size_factor(chunk_size>>10, &chunk_factor),
1378 printk(KERN_INFO "\n");
1379 printk(KERN_INFO "gran_size: %ld%c chunk_size: %ld%c \n",
1380 gran_base, gran_factor, chunk_base, chunk_factor);
1381 }
1287 if (i >= NUM_RESULT) 1382 if (i >= NUM_RESULT)
1288 continue; 1383 continue;
1289 1384
@@ -1326,12 +1421,18 @@ static int __init mtrr_cleanup(unsigned address_bits)
1326 1421
1327 /* print out all */ 1422 /* print out all */
1328 for (i = 0; i < NUM_RESULT; i++) { 1423 for (i = 0; i < NUM_RESULT; i++) {
1329 printk(KERN_INFO "%sgran_size: %ldM \tchunk_size: %ldM \t", 1424 char gran_factor, chunk_factor, lose_factor;
1330 result[i].bad?"*BAD* ":" ", result[i].gran_sizek >> 10, 1425 unsigned long gran_base, chunk_base, lose_base;
1331 result[i].chunk_sizek >> 10); 1426
1332 printk(KERN_CONT "num_reg: %d \tlose RAM: %s%ldM\n", 1427 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1333 result[i].num_reg, result[i].bad?"-":"", 1428 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1334 result[i].lose_cover_sizek >> 10); 1429 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1430 printk(KERN_INFO "%sgran_size: %ld%c \tchunk_size: %ld%c \t",
1431 result[i].bad?"*BAD*":" ",
1432 gran_base, gran_factor, chunk_base, chunk_factor);
1433 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ld%c\n",
1434 result[i].num_reg, result[i].bad?"-":"",
1435 lose_base, lose_factor);
1335 } 1436 }
1336 1437
1337 /* try to find the optimal index */ 1438 /* try to find the optimal index */
@@ -1339,10 +1440,8 @@ static int __init mtrr_cleanup(unsigned address_bits)
1339 nr_mtrr_spare_reg = num_var_ranges - 1; 1440 nr_mtrr_spare_reg = num_var_ranges - 1;
1340 num_reg_good = -1; 1441 num_reg_good = -1;
1341 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) { 1442 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
1342 if (!min_loss_pfn[i]) { 1443 if (!min_loss_pfn[i])
1343 num_reg_good = i; 1444 num_reg_good = i;
1344 break;
1345 }
1346 } 1445 }
1347 1446
1348 index_good = -1; 1447 index_good = -1;
@@ -1358,21 +1457,26 @@ static int __init mtrr_cleanup(unsigned address_bits)
1358 } 1457 }
1359 1458
1360 if (index_good != -1) { 1459 if (index_good != -1) {
1460 char gran_factor, chunk_factor, lose_factor;
1461 unsigned long gran_base, chunk_base, lose_base;
1462
1361 printk(KERN_INFO "Found optimal setting for mtrr clean up\n"); 1463 printk(KERN_INFO "Found optimal setting for mtrr clean up\n");
1362 i = index_good; 1464 i = index_good;
1363 printk(KERN_INFO "gran_size: %ldM \tchunk_size: %ldM \t", 1465 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1364 result[i].gran_sizek >> 10, 1466 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1365 result[i].chunk_sizek >> 10); 1467 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1366 printk(KERN_CONT "num_reg: %d \tlose RAM: %ldM\n", 1468 printk(KERN_INFO "gran_size: %ld%c \tchunk_size: %ld%c \t",
1367 result[i].num_reg, 1469 gran_base, gran_factor, chunk_base, chunk_factor);
1368 result[i].lose_cover_sizek >> 10); 1470 printk(KERN_CONT "num_reg: %d \tlose RAM: %ld%c\n",
1471 result[i].num_reg, lose_base, lose_factor);
1369 /* convert ranges to var ranges state */ 1472 /* convert ranges to var ranges state */
1370 chunk_size = result[i].chunk_sizek; 1473 chunk_size = result[i].chunk_sizek;
1371 chunk_size <<= 10; 1474 chunk_size <<= 10;
1372 gran_size = result[i].gran_sizek; 1475 gran_size = result[i].gran_sizek;
1373 gran_size <<= 10; 1476 gran_size <<= 10;
1374 debug_print = 1; 1477 debug_print++;
1375 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size); 1478 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
1479 debug_print--;
1376 set_var_mtrr_all(address_bits); 1480 set_var_mtrr_all(address_bits);
1377 return 1; 1481 return 1;
1378 } 1482 }
diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c
index 05cc22dbd4ff..6bff382094f5 100644
--- a/arch/x86/kernel/cpu/perfctr-watchdog.c
+++ b/arch/x86/kernel/cpu/perfctr-watchdog.c
@@ -295,13 +295,19 @@ static int setup_k7_watchdog(unsigned nmi_hz)
295 /* setup the timer */ 295 /* setup the timer */
296 wrmsr(evntsel_msr, evntsel, 0); 296 wrmsr(evntsel_msr, evntsel, 0);
297 write_watchdog_counter(perfctr_msr, "K7_PERFCTR0",nmi_hz); 297 write_watchdog_counter(perfctr_msr, "K7_PERFCTR0",nmi_hz);
298 apic_write(APIC_LVTPC, APIC_DM_NMI);
299 evntsel |= K7_EVNTSEL_ENABLE;
300 wrmsr(evntsel_msr, evntsel, 0);
301 298
299 /* initialize the wd struct before enabling */
302 wd->perfctr_msr = perfctr_msr; 300 wd->perfctr_msr = perfctr_msr;
303 wd->evntsel_msr = evntsel_msr; 301 wd->evntsel_msr = evntsel_msr;
304 wd->cccr_msr = 0; /* unused */ 302 wd->cccr_msr = 0; /* unused */
303
304 /* ok, everything is initialized, announce that we're set */
305 cpu_nmi_set_wd_enabled();
306
307 apic_write(APIC_LVTPC, APIC_DM_NMI);
308 evntsel |= K7_EVNTSEL_ENABLE;
309 wrmsr(evntsel_msr, evntsel, 0);
310
305 return 1; 311 return 1;
306} 312}
307 313
@@ -379,13 +385,19 @@ static int setup_p6_watchdog(unsigned nmi_hz)
379 wrmsr(evntsel_msr, evntsel, 0); 385 wrmsr(evntsel_msr, evntsel, 0);
380 nmi_hz = adjust_for_32bit_ctr(nmi_hz); 386 nmi_hz = adjust_for_32bit_ctr(nmi_hz);
381 write_watchdog_counter32(perfctr_msr, "P6_PERFCTR0",nmi_hz); 387 write_watchdog_counter32(perfctr_msr, "P6_PERFCTR0",nmi_hz);
382 apic_write(APIC_LVTPC, APIC_DM_NMI);
383 evntsel |= P6_EVNTSEL0_ENABLE;
384 wrmsr(evntsel_msr, evntsel, 0);
385 388
389 /* initialize the wd struct before enabling */
386 wd->perfctr_msr = perfctr_msr; 390 wd->perfctr_msr = perfctr_msr;
387 wd->evntsel_msr = evntsel_msr; 391 wd->evntsel_msr = evntsel_msr;
388 wd->cccr_msr = 0; /* unused */ 392 wd->cccr_msr = 0; /* unused */
393
394 /* ok, everything is initialized, announce that we're set */
395 cpu_nmi_set_wd_enabled();
396
397 apic_write(APIC_LVTPC, APIC_DM_NMI);
398 evntsel |= P6_EVNTSEL0_ENABLE;
399 wrmsr(evntsel_msr, evntsel, 0);
400
389 return 1; 401 return 1;
390} 402}
391 403
@@ -432,6 +444,27 @@ static const struct wd_ops p6_wd_ops = {
432#define P4_CCCR_ENABLE (1 << 12) 444#define P4_CCCR_ENABLE (1 << 12)
433#define P4_CCCR_OVF (1 << 31) 445#define P4_CCCR_OVF (1 << 31)
434 446
447#define P4_CONTROLS 18
448static unsigned int p4_controls[18] = {
449 MSR_P4_BPU_CCCR0,
450 MSR_P4_BPU_CCCR1,
451 MSR_P4_BPU_CCCR2,
452 MSR_P4_BPU_CCCR3,
453 MSR_P4_MS_CCCR0,
454 MSR_P4_MS_CCCR1,
455 MSR_P4_MS_CCCR2,
456 MSR_P4_MS_CCCR3,
457 MSR_P4_FLAME_CCCR0,
458 MSR_P4_FLAME_CCCR1,
459 MSR_P4_FLAME_CCCR2,
460 MSR_P4_FLAME_CCCR3,
461 MSR_P4_IQ_CCCR0,
462 MSR_P4_IQ_CCCR1,
463 MSR_P4_IQ_CCCR2,
464 MSR_P4_IQ_CCCR3,
465 MSR_P4_IQ_CCCR4,
466 MSR_P4_IQ_CCCR5,
467};
435/* 468/*
436 * Set up IQ_COUNTER0 to behave like a clock, by having IQ_CCCR0 filter 469 * Set up IQ_COUNTER0 to behave like a clock, by having IQ_CCCR0 filter
437 * CRU_ESCR0 (with any non-null event selector) through a complemented 470 * CRU_ESCR0 (with any non-null event selector) through a complemented
@@ -473,6 +506,26 @@ static int setup_p4_watchdog(unsigned nmi_hz)
473 evntsel_msr = MSR_P4_CRU_ESCR0; 506 evntsel_msr = MSR_P4_CRU_ESCR0;
474 cccr_msr = MSR_P4_IQ_CCCR0; 507 cccr_msr = MSR_P4_IQ_CCCR0;
475 cccr_val = P4_CCCR_OVF_PMI0 | P4_CCCR_ESCR_SELECT(4); 508 cccr_val = P4_CCCR_OVF_PMI0 | P4_CCCR_ESCR_SELECT(4);
509
510 /*
511 * If we're on the kdump kernel or other situation, we may
512 * still have other performance counter registers set to
513 * interrupt and they'll keep interrupting forever because
514 * of the P4_CCCR_OVF quirk. So we need to ACK all the
515 * pending interrupts and disable all the registers here,
516 * before reenabling the NMI delivery. Refer to p4_rearm()
517 * about the P4_CCCR_OVF quirk.
518 */
519 if (reset_devices) {
520 unsigned int low, high;
521 int i;
522
523 for (i = 0; i < P4_CONTROLS; i++) {
524 rdmsr(p4_controls[i], low, high);
525 low &= ~(P4_CCCR_ENABLE | P4_CCCR_OVF);
526 wrmsr(p4_controls[i], low, high);
527 }
528 }
476 } else { 529 } else {
477 /* logical cpu 1 */ 530 /* logical cpu 1 */
478 perfctr_msr = MSR_P4_IQ_PERFCTR1; 531 perfctr_msr = MSR_P4_IQ_PERFCTR1;
@@ -499,12 +552,17 @@ static int setup_p4_watchdog(unsigned nmi_hz)
499 wrmsr(evntsel_msr, evntsel, 0); 552 wrmsr(evntsel_msr, evntsel, 0);
500 wrmsr(cccr_msr, cccr_val, 0); 553 wrmsr(cccr_msr, cccr_val, 0);
501 write_watchdog_counter(perfctr_msr, "P4_IQ_COUNTER0", nmi_hz); 554 write_watchdog_counter(perfctr_msr, "P4_IQ_COUNTER0", nmi_hz);
502 apic_write(APIC_LVTPC, APIC_DM_NMI); 555
503 cccr_val |= P4_CCCR_ENABLE;
504 wrmsr(cccr_msr, cccr_val, 0);
505 wd->perfctr_msr = perfctr_msr; 556 wd->perfctr_msr = perfctr_msr;
506 wd->evntsel_msr = evntsel_msr; 557 wd->evntsel_msr = evntsel_msr;
507 wd->cccr_msr = cccr_msr; 558 wd->cccr_msr = cccr_msr;
559
560 /* ok, everything is initialized, announce that we're set */
561 cpu_nmi_set_wd_enabled();
562
563 apic_write(APIC_LVTPC, APIC_DM_NMI);
564 cccr_val |= P4_CCCR_ENABLE;
565 wrmsr(cccr_msr, cccr_val, 0);
508 return 1; 566 return 1;
509} 567}
510 568
@@ -620,13 +678,17 @@ static int setup_intel_arch_watchdog(unsigned nmi_hz)
620 wrmsr(evntsel_msr, evntsel, 0); 678 wrmsr(evntsel_msr, evntsel, 0);
621 nmi_hz = adjust_for_32bit_ctr(nmi_hz); 679 nmi_hz = adjust_for_32bit_ctr(nmi_hz);
622 write_watchdog_counter32(perfctr_msr, "INTEL_ARCH_PERFCTR0", nmi_hz); 680 write_watchdog_counter32(perfctr_msr, "INTEL_ARCH_PERFCTR0", nmi_hz);
623 apic_write(APIC_LVTPC, APIC_DM_NMI);
624 evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
625 wrmsr(evntsel_msr, evntsel, 0);
626 681
627 wd->perfctr_msr = perfctr_msr; 682 wd->perfctr_msr = perfctr_msr;
628 wd->evntsel_msr = evntsel_msr; 683 wd->evntsel_msr = evntsel_msr;
629 wd->cccr_msr = 0; /* unused */ 684 wd->cccr_msr = 0; /* unused */
685
686 /* ok, everything is initialized, announce that we're set */
687 cpu_nmi_set_wd_enabled();
688
689 apic_write(APIC_LVTPC, APIC_DM_NMI);
690 evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
691 wrmsr(evntsel_msr, evntsel, 0);
630 intel_arch_wd_ops.checkbit = 1ULL << (eax.split.bit_width - 1); 692 intel_arch_wd_ops.checkbit = 1ULL << (eax.split.bit_width - 1);
631 return 1; 693 return 1;
632} 694}
diff --git a/arch/x86/kernel/cpu/powerflags.c b/arch/x86/kernel/cpu/powerflags.c
new file mode 100644
index 000000000000..5abbea297e0c
--- /dev/null
+++ b/arch/x86/kernel/cpu/powerflags.c
@@ -0,0 +1,20 @@
1/*
2 * Strings for the various x86 power flags
3 *
4 * This file must not contain any executable code.
5 */
6
7#include <asm/cpufeature.h>
8
9const char *const x86_power_flags[32] = {
10 "ts", /* temperature sensor */
11 "fid", /* frequency id control */
12 "vid", /* voltage id control */
13 "ttp", /* thermal trip */
14 "tm",
15 "stc",
16 "100mhzsteps",
17 "hwpstate",
18 "", /* tsc invariant mapped to constant_tsc */
19 /* nothing */
20};
diff --git a/arch/x86/kernel/cpu/transmeta.c b/arch/x86/kernel/cpu/transmeta.c
index b911a2c61b8f..52b3fefbd5af 100644
--- a/arch/x86/kernel/cpu/transmeta.c
+++ b/arch/x86/kernel/cpu/transmeta.c
@@ -5,6 +5,18 @@
5#include <asm/msr.h> 5#include <asm/msr.h>
6#include "cpu.h" 6#include "cpu.h"
7 7
8static void __cpuinit early_init_transmeta(struct cpuinfo_x86 *c)
9{
10 u32 xlvl;
11
12 /* Transmeta-defined flags: level 0x80860001 */
13 xlvl = cpuid_eax(0x80860000);
14 if ((xlvl & 0xffff0000) == 0x80860000) {
15 if (xlvl >= 0x80860001)
16 c->x86_capability[2] = cpuid_edx(0x80860001);
17 }
18}
19
8static void __cpuinit init_transmeta(struct cpuinfo_x86 *c) 20static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
9{ 21{
10 unsigned int cap_mask, uk, max, dummy; 22 unsigned int cap_mask, uk, max, dummy;
@@ -12,7 +24,8 @@ static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
12 unsigned int cpu_rev, cpu_freq = 0, cpu_flags, new_cpu_rev; 24 unsigned int cpu_rev, cpu_freq = 0, cpu_flags, new_cpu_rev;
13 char cpu_info[65]; 25 char cpu_info[65];
14 26
15 get_model_name(c); /* Same as AMD/Cyrix */ 27 early_init_transmeta(c);
28
16 display_cacheinfo(c); 29 display_cacheinfo(c);
17 30
18 /* Print CMS and CPU revision */ 31 /* Print CMS and CPU revision */
@@ -85,23 +98,12 @@ static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
85#endif 98#endif
86} 99}
87 100
88static void __cpuinit transmeta_identify(struct cpuinfo_x86 *c)
89{
90 u32 xlvl;
91
92 /* Transmeta-defined flags: level 0x80860001 */
93 xlvl = cpuid_eax(0x80860000);
94 if ((xlvl & 0xffff0000) == 0x80860000) {
95 if (xlvl >= 0x80860001)
96 c->x86_capability[2] = cpuid_edx(0x80860001);
97 }
98}
99
100static struct cpu_dev transmeta_cpu_dev __cpuinitdata = { 101static struct cpu_dev transmeta_cpu_dev __cpuinitdata = {
101 .c_vendor = "Transmeta", 102 .c_vendor = "Transmeta",
102 .c_ident = { "GenuineTMx86", "TransmetaCPU" }, 103 .c_ident = { "GenuineTMx86", "TransmetaCPU" },
104 .c_early_init = early_init_transmeta,
103 .c_init = init_transmeta, 105 .c_init = init_transmeta,
104 .c_identify = transmeta_identify, 106 .c_x86_vendor = X86_VENDOR_TRANSMETA,
105}; 107};
106 108
107cpu_vendor_dev_register(X86_VENDOR_TRANSMETA, &transmeta_cpu_dev); 109cpu_dev_register(transmeta_cpu_dev);
diff --git a/arch/x86/kernel/cpu/umc.c b/arch/x86/kernel/cpu/umc.c
index b1fc90989d75..e777f79e0960 100644
--- a/arch/x86/kernel/cpu/umc.c
+++ b/arch/x86/kernel/cpu/umc.c
@@ -19,7 +19,8 @@ static struct cpu_dev umc_cpu_dev __cpuinitdata = {
19 } 19 }
20 }, 20 },
21 }, 21 },
22 .c_x86_vendor = X86_VENDOR_UMC,
22}; 23};
23 24
24cpu_vendor_dev_register(X86_VENDOR_UMC, &umc_cpu_dev); 25cpu_dev_register(umc_cpu_dev);
25 26
diff --git a/arch/x86/kernel/cpuid.c b/arch/x86/kernel/cpuid.c
index 8e9cd6a8ec12..72cefd1e649b 100644
--- a/arch/x86/kernel/cpuid.c
+++ b/arch/x86/kernel/cpuid.c
@@ -36,7 +36,6 @@
36#include <linux/smp_lock.h> 36#include <linux/smp_lock.h>
37#include <linux/major.h> 37#include <linux/major.h>
38#include <linux/fs.h> 38#include <linux/fs.h>
39#include <linux/smp_lock.h>
40#include <linux/device.h> 39#include <linux/device.h>
41#include <linux/cpu.h> 40#include <linux/cpu.h>
42#include <linux/notifier.h> 41#include <linux/notifier.h>
@@ -148,8 +147,8 @@ static __cpuinit int cpuid_device_create(int cpu)
148{ 147{
149 struct device *dev; 148 struct device *dev;
150 149
151 dev = device_create_drvdata(cpuid_class, NULL, MKDEV(CPUID_MAJOR, cpu), 150 dev = device_create(cpuid_class, NULL, MKDEV(CPUID_MAJOR, cpu), NULL,
152 NULL, "cpu%d", cpu); 151 "cpu%d", cpu);
153 return IS_ERR(dev) ? PTR_ERR(dev) : 0; 152 return IS_ERR(dev) ? PTR_ERR(dev) : 0;
154} 153}
155 154
diff --git a/arch/x86/kernel/crash_dump_64.c b/arch/x86/kernel/crash_dump_64.c
index 15e6c6bc4a46..e90a60ef10c2 100644
--- a/arch/x86/kernel/crash_dump_64.c
+++ b/arch/x86/kernel/crash_dump_64.c
@@ -7,9 +7,8 @@
7 7
8#include <linux/errno.h> 8#include <linux/errno.h>
9#include <linux/crash_dump.h> 9#include <linux/crash_dump.h>
10 10#include <linux/uaccess.h>
11#include <asm/uaccess.h> 11#include <linux/io.h>
12#include <asm/io.h>
13 12
14/** 13/**
15 * copy_oldmem_page - copy one page from "oldmem" 14 * copy_oldmem_page - copy one page from "oldmem"
@@ -25,7 +24,7 @@
25 * in the current kernel. We stitch up a pte, similar to kmap_atomic. 24 * in the current kernel. We stitch up a pte, similar to kmap_atomic.
26 */ 25 */
27ssize_t copy_oldmem_page(unsigned long pfn, char *buf, 26ssize_t copy_oldmem_page(unsigned long pfn, char *buf,
28 size_t csize, unsigned long offset, int userbuf) 27 size_t csize, unsigned long offset, int userbuf)
29{ 28{
30 void *vaddr; 29 void *vaddr;
31 30
@@ -33,14 +32,16 @@ ssize_t copy_oldmem_page(unsigned long pfn, char *buf,
33 return 0; 32 return 0;
34 33
35 vaddr = ioremap(pfn << PAGE_SHIFT, PAGE_SIZE); 34 vaddr = ioremap(pfn << PAGE_SHIFT, PAGE_SIZE);
35 if (!vaddr)
36 return -ENOMEM;
36 37
37 if (userbuf) { 38 if (userbuf) {
38 if (copy_to_user(buf, (vaddr + offset), csize)) { 39 if (copy_to_user(buf, vaddr + offset, csize)) {
39 iounmap(vaddr); 40 iounmap(vaddr);
40 return -EFAULT; 41 return -EFAULT;
41 } 42 }
42 } else 43 } else
43 memcpy(buf, (vaddr + offset), csize); 44 memcpy(buf, vaddr + offset, csize);
44 45
45 iounmap(vaddr); 46 iounmap(vaddr);
46 return csize; 47 return csize;
diff --git a/arch/x86/kernel/doublefault_32.c b/arch/x86/kernel/doublefault_32.c
index a47798b59f07..b4f14c6c09d9 100644
--- a/arch/x86/kernel/doublefault_32.c
+++ b/arch/x86/kernel/doublefault_32.c
@@ -66,6 +66,6 @@ struct tss_struct doublefault_tss __cacheline_aligned = {
66 .ds = __USER_DS, 66 .ds = __USER_DS,
67 .fs = __KERNEL_PERCPU, 67 .fs = __KERNEL_PERCPU,
68 68
69 .__cr3 = __pa(swapper_pg_dir) 69 .__cr3 = __pa_nodebug(swapper_pg_dir),
70 } 70 }
71}; 71};
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 11c11b8ec48d..2b69994fd3a8 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -2,26 +2,49 @@
2 * Debug Store support 2 * Debug Store support
3 * 3 *
4 * This provides a low-level interface to the hardware's Debug Store 4 * This provides a low-level interface to the hardware's Debug Store
5 * feature that is used for last branch recording (LBR) and 5 * feature that is used for branch trace store (BTS) and
6 * precise-event based sampling (PEBS). 6 * precise-event based sampling (PEBS).
7 * 7 *
8 * Different architectures use a different DS layout/pointer size. 8 * It manages:
9 * The below functions therefore work on a void*. 9 * - per-thread and per-cpu allocation of BTS and PEBS
10 * - buffer memory allocation (optional)
11 * - buffer overflow handling
12 * - buffer access
10 * 13 *
14 * It assumes:
15 * - get_task_struct on all parameter tasks
16 * - current is allowed to trace parameter tasks
11 * 17 *
12 * Since there is no user for PEBS, yet, only LBR (or branch
13 * trace store, BTS) is supported.
14 * 18 *
15 * 19 * Copyright (C) 2007-2008 Intel Corporation.
16 * Copyright (C) 2007 Intel Corporation. 20 * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008
17 * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
18 */ 21 */
19 22
23
24#ifdef CONFIG_X86_DS
25
20#include <asm/ds.h> 26#include <asm/ds.h>
21 27
22#include <linux/errno.h> 28#include <linux/errno.h>
23#include <linux/string.h> 29#include <linux/string.h>
24#include <linux/slab.h> 30#include <linux/slab.h>
31#include <linux/sched.h>
32#include <linux/mm.h>
33
34
35/*
36 * The configuration for a particular DS hardware implementation.
37 */
38struct ds_configuration {
39 /* the size of the DS structure in bytes */
40 unsigned char sizeof_ds;
41 /* the size of one pointer-typed field in the DS structure in bytes;
42 this covers the first 8 fields related to buffer management. */
43 unsigned char sizeof_field;
44 /* the size of a BTS/PEBS record in bytes */
45 unsigned char sizeof_rec[2];
46};
47static struct ds_configuration ds_cfg;
25 48
26 49
27/* 50/*
@@ -44,378 +67,747 @@
44 * (interrupt occurs when write pointer passes interrupt pointer) 67 * (interrupt occurs when write pointer passes interrupt pointer)
45 * - value to which counter is reset following counter overflow 68 * - value to which counter is reset following counter overflow
46 * 69 *
47 * On later architectures, the last branch recording hardware uses 70 * Later architectures use 64bit pointers throughout, whereas earlier
48 * 64bit pointers even in 32bit mode. 71 * architectures use 32bit pointers in 32bit mode.
49 *
50 *
51 * Branch Trace Store (BTS) records store information about control
52 * flow changes. They at least provide the following information:
53 * - source linear address
54 * - destination linear address
55 * 72 *
56 * Netburst supported a predicated bit that had been dropped in later
57 * architectures. We do not suppor it.
58 * 73 *
74 * We compute the base address for the first 8 fields based on:
75 * - the field size stored in the DS configuration
76 * - the relative field position
77 * - an offset giving the start of the respective region
59 * 78 *
60 * In order to abstract from the actual DS and BTS layout, we describe 79 * This offset is further used to index various arrays holding
61 * the access to the relevant fields. 80 * information for BTS and PEBS at the respective index.
62 * Thanks to Andi Kleen for proposing this design.
63 * 81 *
64 * The implementation, however, is not as general as it might seem. In 82 * On later 32bit processors, we only access the lower 32bit of the
65 * order to stay somewhat simple and efficient, we assume an 83 * 64bit pointer fields. The upper halves will be zeroed out.
66 * underlying unsigned type (mostly a pointer type) and we expect the
67 * field to be at least as big as that type.
68 */ 84 */
69 85
70/* 86enum ds_field {
71 * A special from_ip address to indicate that the BTS record is an 87 ds_buffer_base = 0,
72 * info record that needs to be interpreted or skipped. 88 ds_index,
73 */ 89 ds_absolute_maximum,
74#define BTS_ESCAPE_ADDRESS (-1) 90 ds_interrupt_threshold,
91};
75 92
76/* 93enum ds_qualifier {
77 * A field access descriptor 94 ds_bts = 0,
78 */ 95 ds_pebs
79struct access_desc {
80 unsigned char offset;
81 unsigned char size;
82}; 96};
83 97
98static inline unsigned long ds_get(const unsigned char *base,
99 enum ds_qualifier qual, enum ds_field field)
100{
101 base += (ds_cfg.sizeof_field * (field + (4 * qual)));
102 return *(unsigned long *)base;
103}
104
105static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
106 enum ds_field field, unsigned long value)
107{
108 base += (ds_cfg.sizeof_field * (field + (4 * qual)));
109 (*(unsigned long *)base) = value;
110}
111
112
84/* 113/*
85 * The configuration for a particular DS/BTS hardware implementation. 114 * Locking is done only for allocating BTS or PEBS resources and for
115 * guarding context and buffer memory allocation.
116 *
117 * Most functions require the current task to own the ds context part
118 * they are going to access. All the locking is done when validating
119 * access to the context.
86 */ 120 */
87struct ds_configuration { 121static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock);
88 /* the DS configuration */
89 unsigned char sizeof_ds;
90 struct access_desc bts_buffer_base;
91 struct access_desc bts_index;
92 struct access_desc bts_absolute_maximum;
93 struct access_desc bts_interrupt_threshold;
94 /* the BTS configuration */
95 unsigned char sizeof_bts;
96 struct access_desc from_ip;
97 struct access_desc to_ip;
98 /* BTS variants used to store additional information like
99 timestamps */
100 struct access_desc info_type;
101 struct access_desc info_data;
102 unsigned long debugctl_mask;
103};
104 122
105/* 123/*
106 * The global configuration used by the below accessor functions 124 * Validate that the current task is allowed to access the BTS/PEBS
125 * buffer of the parameter task.
126 *
127 * Returns 0, if access is granted; -Eerrno, otherwise.
107 */ 128 */
108static struct ds_configuration ds_cfg; 129static inline int ds_validate_access(struct ds_context *context,
130 enum ds_qualifier qual)
131{
132 if (!context)
133 return -EPERM;
134
135 if (context->owner[qual] == current)
136 return 0;
137
138 return -EPERM;
139}
140
109 141
110/* 142/*
111 * Accessor functions for some DS and BTS fields using the above 143 * We either support (system-wide) per-cpu or per-thread allocation.
112 * global ptrace_bts_cfg. 144 * We distinguish the two based on the task_struct pointer, where a
145 * NULL pointer indicates per-cpu allocation for the current cpu.
146 *
147 * Allocations are use-counted. As soon as resources are allocated,
148 * further allocations must be of the same type (per-cpu or
149 * per-thread). We model this by counting allocations (i.e. the number
150 * of tracers of a certain type) for one type negatively:
151 * =0 no tracers
152 * >0 number of per-thread tracers
153 * <0 number of per-cpu tracers
154 *
155 * The below functions to get and put tracers and to check the
156 * allocation type require the ds_lock to be held by the caller.
157 *
158 * Tracers essentially gives the number of ds contexts for a certain
159 * type of allocation.
113 */ 160 */
114static inline unsigned long get_bts_buffer_base(char *base) 161static long tracers;
162
163static inline void get_tracer(struct task_struct *task)
115{ 164{
116 return *(unsigned long *)(base + ds_cfg.bts_buffer_base.offset); 165 tracers += (task ? 1 : -1);
117} 166}
118static inline void set_bts_buffer_base(char *base, unsigned long value) 167
168static inline void put_tracer(struct task_struct *task)
119{ 169{
120 (*(unsigned long *)(base + ds_cfg.bts_buffer_base.offset)) = value; 170 tracers -= (task ? 1 : -1);
121} 171}
122static inline unsigned long get_bts_index(char *base) 172
173static inline int check_tracer(struct task_struct *task)
123{ 174{
124 return *(unsigned long *)(base + ds_cfg.bts_index.offset); 175 return (task ? (tracers >= 0) : (tracers <= 0));
125} 176}
126static inline void set_bts_index(char *base, unsigned long value) 177
178
179/*
180 * The DS context is either attached to a thread or to a cpu:
181 * - in the former case, the thread_struct contains a pointer to the
182 * attached context.
183 * - in the latter case, we use a static array of per-cpu context
184 * pointers.
185 *
186 * Contexts are use-counted. They are allocated on first access and
187 * deallocated when the last user puts the context.
188 *
189 * We distinguish between an allocating and a non-allocating get of a
190 * context:
191 * - the allocating get is used for requesting BTS/PEBS resources. It
192 * requires the caller to hold the global ds_lock.
193 * - the non-allocating get is used for all other cases. A
194 * non-existing context indicates an error. It acquires and releases
195 * the ds_lock itself for obtaining the context.
196 *
197 * A context and its DS configuration are allocated and deallocated
198 * together. A context always has a DS configuration of the
199 * appropriate size.
200 */
201static DEFINE_PER_CPU(struct ds_context *, system_context);
202
203#define this_system_context per_cpu(system_context, smp_processor_id())
204
205/*
206 * Returns the pointer to the parameter task's context or to the
207 * system-wide context, if task is NULL.
208 *
209 * Increases the use count of the returned context, if not NULL.
210 */
211static inline struct ds_context *ds_get_context(struct task_struct *task)
127{ 212{
128 (*(unsigned long *)(base + ds_cfg.bts_index.offset)) = value; 213 struct ds_context *context;
214
215 spin_lock(&ds_lock);
216
217 context = (task ? task->thread.ds_ctx : this_system_context);
218 if (context)
219 context->count++;
220
221 spin_unlock(&ds_lock);
222
223 return context;
129} 224}
130static inline unsigned long get_bts_absolute_maximum(char *base) 225
226/*
227 * Same as ds_get_context, but allocates the context and it's DS
228 * structure, if necessary; returns NULL; if out of memory.
229 *
230 * pre: requires ds_lock to be held
231 */
232static inline struct ds_context *ds_alloc_context(struct task_struct *task)
131{ 233{
132 return *(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset); 234 struct ds_context **p_context =
235 (task ? &task->thread.ds_ctx : &this_system_context);
236 struct ds_context *context = *p_context;
237
238 if (!context) {
239 context = kzalloc(sizeof(*context), GFP_KERNEL);
240
241 if (!context)
242 return NULL;
243
244 context->ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL);
245 if (!context->ds) {
246 kfree(context);
247 return NULL;
248 }
249
250 *p_context = context;
251
252 context->this = p_context;
253 context->task = task;
254
255 if (task)
256 set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
257
258 if (!task || (task == current))
259 wrmsr(MSR_IA32_DS_AREA, (unsigned long)context->ds, 0);
260
261 get_tracer(task);
262 }
263
264 context->count++;
265
266 return context;
133} 267}
134static inline void set_bts_absolute_maximum(char *base, unsigned long value) 268
269/*
270 * Decreases the use count of the parameter context, if not NULL.
271 * Deallocates the context, if the use count reaches zero.
272 */
273static inline void ds_put_context(struct ds_context *context)
135{ 274{
136 (*(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset)) = value; 275 if (!context)
276 return;
277
278 spin_lock(&ds_lock);
279
280 if (--context->count)
281 goto out;
282
283 *(context->this) = NULL;
284
285 if (context->task)
286 clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
287
288 if (!context->task || (context->task == current))
289 wrmsrl(MSR_IA32_DS_AREA, 0);
290
291 put_tracer(context->task);
292
293 /* free any leftover buffers from tracers that did not
294 * deallocate them properly. */
295 kfree(context->buffer[ds_bts]);
296 kfree(context->buffer[ds_pebs]);
297 kfree(context->ds);
298 kfree(context);
299 out:
300 spin_unlock(&ds_lock);
137} 301}
138static inline unsigned long get_bts_interrupt_threshold(char *base) 302
303
304/*
305 * Handle a buffer overflow
306 *
307 * task: the task whose buffers are overflowing;
308 * NULL for a buffer overflow on the current cpu
309 * context: the ds context
310 * qual: the buffer type
311 */
312static void ds_overflow(struct task_struct *task, struct ds_context *context,
313 enum ds_qualifier qual)
139{ 314{
140 return *(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset); 315 if (!context)
316 return;
317
318 if (context->callback[qual])
319 (*context->callback[qual])(task);
320
321 /* todo: do some more overflow handling */
141} 322}
142static inline void set_bts_interrupt_threshold(char *base, unsigned long value) 323
324
325/*
326 * Allocate a non-pageable buffer of the parameter size.
327 * Checks the memory and the locked memory rlimit.
328 *
329 * Returns the buffer, if successful;
330 * NULL, if out of memory or rlimit exceeded.
331 *
332 * size: the requested buffer size in bytes
333 * pages (out): if not NULL, contains the number of pages reserved
334 */
335static inline void *ds_allocate_buffer(size_t size, unsigned int *pages)
143{ 336{
144 (*(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset)) = value; 337 unsigned long rlim, vm, pgsz;
338 void *buffer;
339
340 pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
341
342 rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
343 vm = current->mm->total_vm + pgsz;
344 if (rlim < vm)
345 return NULL;
346
347 rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
348 vm = current->mm->locked_vm + pgsz;
349 if (rlim < vm)
350 return NULL;
351
352 buffer = kzalloc(size, GFP_KERNEL);
353 if (!buffer)
354 return NULL;
355
356 current->mm->total_vm += pgsz;
357 current->mm->locked_vm += pgsz;
358
359 if (pages)
360 *pages = pgsz;
361
362 return buffer;
145} 363}
146static inline unsigned long get_from_ip(char *base) 364
365static int ds_request(struct task_struct *task, void *base, size_t size,
366 ds_ovfl_callback_t ovfl, enum ds_qualifier qual)
147{ 367{
148 return *(unsigned long *)(base + ds_cfg.from_ip.offset); 368 struct ds_context *context;
369 unsigned long buffer, adj;
370 const unsigned long alignment = (1 << 3);
371 int error = 0;
372
373 if (!ds_cfg.sizeof_ds)
374 return -EOPNOTSUPP;
375
376 /* we require some space to do alignment adjustments below */
377 if (size < (alignment + ds_cfg.sizeof_rec[qual]))
378 return -EINVAL;
379
380 /* buffer overflow notification is not yet implemented */
381 if (ovfl)
382 return -EOPNOTSUPP;
383
384
385 spin_lock(&ds_lock);
386
387 if (!check_tracer(task))
388 return -EPERM;
389
390 error = -ENOMEM;
391 context = ds_alloc_context(task);
392 if (!context)
393 goto out_unlock;
394
395 error = -EALREADY;
396 if (context->owner[qual] == current)
397 goto out_unlock;
398 error = -EPERM;
399 if (context->owner[qual] != NULL)
400 goto out_unlock;
401 context->owner[qual] = current;
402
403 spin_unlock(&ds_lock);
404
405
406 error = -ENOMEM;
407 if (!base) {
408 base = ds_allocate_buffer(size, &context->pages[qual]);
409 if (!base)
410 goto out_release;
411
412 context->buffer[qual] = base;
413 }
414 error = 0;
415
416 context->callback[qual] = ovfl;
417
418 /* adjust the buffer address and size to meet alignment
419 * constraints:
420 * - buffer is double-word aligned
421 * - size is multiple of record size
422 *
423 * We checked the size at the very beginning; we have enough
424 * space to do the adjustment.
425 */
426 buffer = (unsigned long)base;
427
428 adj = ALIGN(buffer, alignment) - buffer;
429 buffer += adj;
430 size -= adj;
431
432 size /= ds_cfg.sizeof_rec[qual];
433 size *= ds_cfg.sizeof_rec[qual];
434
435 ds_set(context->ds, qual, ds_buffer_base, buffer);
436 ds_set(context->ds, qual, ds_index, buffer);
437 ds_set(context->ds, qual, ds_absolute_maximum, buffer + size);
438
439 if (ovfl) {
440 /* todo: select a suitable interrupt threshold */
441 } else
442 ds_set(context->ds, qual,
443 ds_interrupt_threshold, buffer + size + 1);
444
445 /* we keep the context until ds_release */
446 return error;
447
448 out_release:
449 context->owner[qual] = NULL;
450 ds_put_context(context);
451 return error;
452
453 out_unlock:
454 spin_unlock(&ds_lock);
455 ds_put_context(context);
456 return error;
149} 457}
150static inline void set_from_ip(char *base, unsigned long value) 458
459int ds_request_bts(struct task_struct *task, void *base, size_t size,
460 ds_ovfl_callback_t ovfl)
151{ 461{
152 (*(unsigned long *)(base + ds_cfg.from_ip.offset)) = value; 462 return ds_request(task, base, size, ovfl, ds_bts);
153} 463}
154static inline unsigned long get_to_ip(char *base) 464
465int ds_request_pebs(struct task_struct *task, void *base, size_t size,
466 ds_ovfl_callback_t ovfl)
155{ 467{
156 return *(unsigned long *)(base + ds_cfg.to_ip.offset); 468 return ds_request(task, base, size, ovfl, ds_pebs);
157} 469}
158static inline void set_to_ip(char *base, unsigned long value) 470
471static int ds_release(struct task_struct *task, enum ds_qualifier qual)
159{ 472{
160 (*(unsigned long *)(base + ds_cfg.to_ip.offset)) = value; 473 struct ds_context *context;
474 int error;
475
476 context = ds_get_context(task);
477 error = ds_validate_access(context, qual);
478 if (error < 0)
479 goto out;
480
481 kfree(context->buffer[qual]);
482 context->buffer[qual] = NULL;
483
484 current->mm->total_vm -= context->pages[qual];
485 current->mm->locked_vm -= context->pages[qual];
486 context->pages[qual] = 0;
487 context->owner[qual] = NULL;
488
489 /*
490 * we put the context twice:
491 * once for the ds_get_context
492 * once for the corresponding ds_request
493 */
494 ds_put_context(context);
495 out:
496 ds_put_context(context);
497 return error;
161} 498}
162static inline unsigned char get_info_type(char *base) 499
500int ds_release_bts(struct task_struct *task)
163{ 501{
164 return *(unsigned char *)(base + ds_cfg.info_type.offset); 502 return ds_release(task, ds_bts);
165} 503}
166static inline void set_info_type(char *base, unsigned char value) 504
505int ds_release_pebs(struct task_struct *task)
167{ 506{
168 (*(unsigned char *)(base + ds_cfg.info_type.offset)) = value; 507 return ds_release(task, ds_pebs);
169} 508}
170static inline unsigned long get_info_data(char *base) 509
510static int ds_get_index(struct task_struct *task, size_t *pos,
511 enum ds_qualifier qual)
171{ 512{
172 return *(unsigned long *)(base + ds_cfg.info_data.offset); 513 struct ds_context *context;
514 unsigned long base, index;
515 int error;
516
517 context = ds_get_context(task);
518 error = ds_validate_access(context, qual);
519 if (error < 0)
520 goto out;
521
522 base = ds_get(context->ds, qual, ds_buffer_base);
523 index = ds_get(context->ds, qual, ds_index);
524
525 error = ((index - base) / ds_cfg.sizeof_rec[qual]);
526 if (pos)
527 *pos = error;
528 out:
529 ds_put_context(context);
530 return error;
173} 531}
174static inline void set_info_data(char *base, unsigned long value) 532
533int ds_get_bts_index(struct task_struct *task, size_t *pos)
175{ 534{
176 (*(unsigned long *)(base + ds_cfg.info_data.offset)) = value; 535 return ds_get_index(task, pos, ds_bts);
177} 536}
178 537
538int ds_get_pebs_index(struct task_struct *task, size_t *pos)
539{
540 return ds_get_index(task, pos, ds_pebs);
541}
179 542
180int ds_allocate(void **dsp, size_t bts_size_in_bytes) 543static int ds_get_end(struct task_struct *task, size_t *pos,
544 enum ds_qualifier qual)
181{ 545{
182 size_t bts_size_in_records; 546 struct ds_context *context;
183 unsigned long bts; 547 unsigned long base, end;
184 void *ds; 548 int error;
549
550 context = ds_get_context(task);
551 error = ds_validate_access(context, qual);
552 if (error < 0)
553 goto out;
554
555 base = ds_get(context->ds, qual, ds_buffer_base);
556 end = ds_get(context->ds, qual, ds_absolute_maximum);
557
558 error = ((end - base) / ds_cfg.sizeof_rec[qual]);
559 if (pos)
560 *pos = error;
561 out:
562 ds_put_context(context);
563 return error;
564}
185 565
186 if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) 566int ds_get_bts_end(struct task_struct *task, size_t *pos)
187 return -EOPNOTSUPP; 567{
568 return ds_get_end(task, pos, ds_bts);
569}
188 570
189 if (bts_size_in_bytes < 0) 571int ds_get_pebs_end(struct task_struct *task, size_t *pos)
190 return -EINVAL; 572{
573 return ds_get_end(task, pos, ds_pebs);
574}
191 575
192 bts_size_in_records = 576static int ds_access(struct task_struct *task, size_t index,
193 bts_size_in_bytes / ds_cfg.sizeof_bts; 577 const void **record, enum ds_qualifier qual)
194 bts_size_in_bytes = 578{
195 bts_size_in_records * ds_cfg.sizeof_bts; 579 struct ds_context *context;
580 unsigned long base, idx;
581 int error;
196 582
197 if (bts_size_in_bytes <= 0) 583 if (!record)
198 return -EINVAL; 584 return -EINVAL;
199 585
200 bts = (unsigned long)kzalloc(bts_size_in_bytes, GFP_KERNEL); 586 context = ds_get_context(task);
201 587 error = ds_validate_access(context, qual);
202 if (!bts) 588 if (error < 0)
203 return -ENOMEM; 589 goto out;
204 590
205 ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL); 591 base = ds_get(context->ds, qual, ds_buffer_base);
592 idx = base + (index * ds_cfg.sizeof_rec[qual]);
206 593
207 if (!ds) { 594 error = -EINVAL;
208 kfree((void *)bts); 595 if (idx > ds_get(context->ds, qual, ds_absolute_maximum))
209 return -ENOMEM; 596 goto out;
210 }
211
212 set_bts_buffer_base(ds, bts);
213 set_bts_index(ds, bts);
214 set_bts_absolute_maximum(ds, bts + bts_size_in_bytes);
215 set_bts_interrupt_threshold(ds, bts + bts_size_in_bytes + 1);
216 597
217 *dsp = ds; 598 *record = (const void *)idx;
218 return 0; 599 error = ds_cfg.sizeof_rec[qual];
600 out:
601 ds_put_context(context);
602 return error;
219} 603}
220 604
221int ds_free(void **dsp) 605int ds_access_bts(struct task_struct *task, size_t index, const void **record)
222{ 606{
223 if (*dsp) { 607 return ds_access(task, index, record, ds_bts);
224 kfree((void *)get_bts_buffer_base(*dsp));
225 kfree(*dsp);
226 *dsp = NULL;
227 }
228 return 0;
229} 608}
230 609
231int ds_get_bts_size(void *ds) 610int ds_access_pebs(struct task_struct *task, size_t index, const void **record)
232{ 611{
233 int size_in_bytes; 612 return ds_access(task, index, record, ds_pebs);
234
235 if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
236 return -EOPNOTSUPP;
237
238 if (!ds)
239 return 0;
240
241 size_in_bytes =
242 get_bts_absolute_maximum(ds) -
243 get_bts_buffer_base(ds);
244 return size_in_bytes;
245} 613}
246 614
247int ds_get_bts_end(void *ds) 615static int ds_write(struct task_struct *task, const void *record, size_t size,
616 enum ds_qualifier qual, int force)
248{ 617{
249 int size_in_bytes = ds_get_bts_size(ds); 618 struct ds_context *context;
250 619 int error;
251 if (size_in_bytes <= 0)
252 return size_in_bytes;
253 620
254 return size_in_bytes / ds_cfg.sizeof_bts; 621 if (!record)
255} 622 return -EINVAL;
256 623
257int ds_get_bts_index(void *ds) 624 error = -EPERM;
258{ 625 context = ds_get_context(task);
259 int index_offset_in_bytes; 626 if (!context)
627 goto out;
260 628
261 if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) 629 if (!force) {
262 return -EOPNOTSUPP; 630 error = ds_validate_access(context, qual);
631 if (error < 0)
632 goto out;
633 }
263 634
264 index_offset_in_bytes = 635 error = 0;
265 get_bts_index(ds) - 636 while (size) {
266 get_bts_buffer_base(ds); 637 unsigned long base, index, end, write_end, int_th;
638 unsigned long write_size, adj_write_size;
639
640 /*
641 * write as much as possible without producing an
642 * overflow interrupt.
643 *
644 * interrupt_threshold must either be
645 * - bigger than absolute_maximum or
646 * - point to a record between buffer_base and absolute_maximum
647 *
648 * index points to a valid record.
649 */
650 base = ds_get(context->ds, qual, ds_buffer_base);
651 index = ds_get(context->ds, qual, ds_index);
652 end = ds_get(context->ds, qual, ds_absolute_maximum);
653 int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
654
655 write_end = min(end, int_th);
656
657 /* if we are already beyond the interrupt threshold,
658 * we fill the entire buffer */
659 if (write_end <= index)
660 write_end = end;
661
662 if (write_end <= index)
663 goto out;
664
665 write_size = min((unsigned long) size, write_end - index);
666 memcpy((void *)index, record, write_size);
667
668 record = (const char *)record + write_size;
669 size -= write_size;
670 error += write_size;
671
672 adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
673 adj_write_size *= ds_cfg.sizeof_rec[qual];
674
675 /* zero out trailing bytes */
676 memset((char *)index + write_size, 0,
677 adj_write_size - write_size);
678 index += adj_write_size;
679
680 if (index >= end)
681 index = base;
682 ds_set(context->ds, qual, ds_index, index);
683
684 if (index >= int_th)
685 ds_overflow(task, context, qual);
686 }
267 687
268 return index_offset_in_bytes / ds_cfg.sizeof_bts; 688 out:
689 ds_put_context(context);
690 return error;
269} 691}
270 692
271int ds_set_overflow(void *ds, int method) 693int ds_write_bts(struct task_struct *task, const void *record, size_t size)
272{ 694{
273 switch (method) { 695 return ds_write(task, record, size, ds_bts, /* force = */ 0);
274 case DS_O_SIGNAL:
275 return -EOPNOTSUPP;
276 case DS_O_WRAP:
277 return 0;
278 default:
279 return -EINVAL;
280 }
281} 696}
282 697
283int ds_get_overflow(void *ds) 698int ds_write_pebs(struct task_struct *task, const void *record, size_t size)
284{ 699{
285 return DS_O_WRAP; 700 return ds_write(task, record, size, ds_pebs, /* force = */ 0);
286} 701}
287 702
288int ds_clear(void *ds) 703int ds_unchecked_write_bts(struct task_struct *task,
704 const void *record, size_t size)
289{ 705{
290 int bts_size = ds_get_bts_size(ds); 706 return ds_write(task, record, size, ds_bts, /* force = */ 1);
291 unsigned long bts_base;
292
293 if (bts_size <= 0)
294 return bts_size;
295
296 bts_base = get_bts_buffer_base(ds);
297 memset((void *)bts_base, 0, bts_size);
298
299 set_bts_index(ds, bts_base);
300 return 0;
301} 707}
302 708
303int ds_read_bts(void *ds, int index, struct bts_struct *out) 709int ds_unchecked_write_pebs(struct task_struct *task,
710 const void *record, size_t size)
304{ 711{
305 void *bts; 712 return ds_write(task, record, size, ds_pebs, /* force = */ 1);
713}
306 714
307 if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) 715static int ds_reset_or_clear(struct task_struct *task,
308 return -EOPNOTSUPP; 716 enum ds_qualifier qual, int clear)
717{
718 struct ds_context *context;
719 unsigned long base, end;
720 int error;
309 721
310 if (index < 0) 722 context = ds_get_context(task);
311 return -EINVAL; 723 error = ds_validate_access(context, qual);
724 if (error < 0)
725 goto out;
312 726
313 if (index >= ds_get_bts_size(ds)) 727 base = ds_get(context->ds, qual, ds_buffer_base);
314 return -EINVAL; 728 end = ds_get(context->ds, qual, ds_absolute_maximum);
315 729
316 bts = (void *)(get_bts_buffer_base(ds) + (index * ds_cfg.sizeof_bts)); 730 if (clear)
731 memset((void *)base, 0, end - base);
317 732
318 memset(out, 0, sizeof(*out)); 733 ds_set(context->ds, qual, ds_index, base);
319 if (get_from_ip(bts) == BTS_ESCAPE_ADDRESS) {
320 out->qualifier = get_info_type(bts);
321 out->variant.jiffies = get_info_data(bts);
322 } else {
323 out->qualifier = BTS_BRANCH;
324 out->variant.lbr.from_ip = get_from_ip(bts);
325 out->variant.lbr.to_ip = get_to_ip(bts);
326 }
327 734
328 return sizeof(*out);; 735 error = 0;
736 out:
737 ds_put_context(context);
738 return error;
329} 739}
330 740
331int ds_write_bts(void *ds, const struct bts_struct *in) 741int ds_reset_bts(struct task_struct *task)
332{ 742{
333 unsigned long bts; 743 return ds_reset_or_clear(task, ds_bts, /* clear = */ 0);
334 744}
335 if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
336 return -EOPNOTSUPP;
337
338 if (ds_get_bts_size(ds) <= 0)
339 return -ENXIO;
340 745
341 bts = get_bts_index(ds); 746int ds_reset_pebs(struct task_struct *task)
747{
748 return ds_reset_or_clear(task, ds_pebs, /* clear = */ 0);
749}
342 750
343 memset((void *)bts, 0, ds_cfg.sizeof_bts); 751int ds_clear_bts(struct task_struct *task)
344 switch (in->qualifier) { 752{
345 case BTS_INVALID: 753 return ds_reset_or_clear(task, ds_bts, /* clear = */ 1);
346 break; 754}
347 755
348 case BTS_BRANCH: 756int ds_clear_pebs(struct task_struct *task)
349 set_from_ip((void *)bts, in->variant.lbr.from_ip); 757{
350 set_to_ip((void *)bts, in->variant.lbr.to_ip); 758 return ds_reset_or_clear(task, ds_pebs, /* clear = */ 1);
351 break; 759}
352 760
353 case BTS_TASK_ARRIVES: 761int ds_get_pebs_reset(struct task_struct *task, u64 *value)
354 case BTS_TASK_DEPARTS: 762{
355 set_from_ip((void *)bts, BTS_ESCAPE_ADDRESS); 763 struct ds_context *context;
356 set_info_type((void *)bts, in->qualifier); 764 int error;
357 set_info_data((void *)bts, in->variant.jiffies);
358 break;
359 765
360 default: 766 if (!value)
361 return -EINVAL; 767 return -EINVAL;
362 }
363 768
364 bts = bts + ds_cfg.sizeof_bts; 769 context = ds_get_context(task);
365 if (bts >= get_bts_absolute_maximum(ds)) 770 error = ds_validate_access(context, ds_pebs);
366 bts = get_bts_buffer_base(ds); 771 if (error < 0)
367 set_bts_index(ds, bts); 772 goto out;
368 773
369 return ds_cfg.sizeof_bts; 774 *value = *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8));
775
776 error = 0;
777 out:
778 ds_put_context(context);
779 return error;
370} 780}
371 781
372unsigned long ds_debugctl_mask(void) 782int ds_set_pebs_reset(struct task_struct *task, u64 value)
373{ 783{
374 return ds_cfg.debugctl_mask; 784 struct ds_context *context;
375} 785 int error;
376 786
377#ifdef __i386__ 787 context = ds_get_context(task);
378static const struct ds_configuration ds_cfg_netburst = { 788 error = ds_validate_access(context, ds_pebs);
379 .sizeof_ds = 9 * 4, 789 if (error < 0)
380 .bts_buffer_base = { 0, 4 }, 790 goto out;
381 .bts_index = { 4, 4 },
382 .bts_absolute_maximum = { 8, 4 },
383 .bts_interrupt_threshold = { 12, 4 },
384 .sizeof_bts = 3 * 4,
385 .from_ip = { 0, 4 },
386 .to_ip = { 4, 4 },
387 .info_type = { 4, 1 },
388 .info_data = { 8, 4 },
389 .debugctl_mask = (1<<2)|(1<<3)
390};
391 791
392static const struct ds_configuration ds_cfg_pentium_m = { 792 *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8)) = value;
393 .sizeof_ds = 9 * 4, 793
394 .bts_buffer_base = { 0, 4 }, 794 error = 0;
395 .bts_index = { 4, 4 }, 795 out:
396 .bts_absolute_maximum = { 8, 4 }, 796 ds_put_context(context);
397 .bts_interrupt_threshold = { 12, 4 }, 797 return error;
398 .sizeof_bts = 3 * 4, 798}
399 .from_ip = { 0, 4 }, 799
400 .to_ip = { 4, 4 }, 800static const struct ds_configuration ds_cfg_var = {
401 .info_type = { 4, 1 }, 801 .sizeof_ds = sizeof(long) * 12,
402 .info_data = { 8, 4 }, 802 .sizeof_field = sizeof(long),
403 .debugctl_mask = (1<<6)|(1<<7) 803 .sizeof_rec[ds_bts] = sizeof(long) * 3,
804 .sizeof_rec[ds_pebs] = sizeof(long) * 10
404}; 805};
405#endif /* _i386_ */ 806static const struct ds_configuration ds_cfg_64 = {
406 807 .sizeof_ds = 8 * 12,
407static const struct ds_configuration ds_cfg_core2 = { 808 .sizeof_field = 8,
408 .sizeof_ds = 9 * 8, 809 .sizeof_rec[ds_bts] = 8 * 3,
409 .bts_buffer_base = { 0, 8 }, 810 .sizeof_rec[ds_pebs] = 8 * 10
410 .bts_index = { 8, 8 },
411 .bts_absolute_maximum = { 16, 8 },
412 .bts_interrupt_threshold = { 24, 8 },
413 .sizeof_bts = 3 * 8,
414 .from_ip = { 0, 8 },
415 .to_ip = { 8, 8 },
416 .info_type = { 8, 1 },
417 .info_data = { 16, 8 },
418 .debugctl_mask = (1<<6)|(1<<7)|(1<<9)
419}; 811};
420 812
421static inline void 813static inline void
@@ -429,14 +821,13 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
429 switch (c->x86) { 821 switch (c->x86) {
430 case 0x6: 822 case 0x6:
431 switch (c->x86_model) { 823 switch (c->x86_model) {
432#ifdef __i386__
433 case 0xD: 824 case 0xD:
434 case 0xE: /* Pentium M */ 825 case 0xE: /* Pentium M */
435 ds_configure(&ds_cfg_pentium_m); 826 ds_configure(&ds_cfg_var);
436 break; 827 break;
437#endif /* _i386_ */
438 case 0xF: /* Core2 */ 828 case 0xF: /* Core2 */
439 ds_configure(&ds_cfg_core2); 829 case 0x1C: /* Atom */
830 ds_configure(&ds_cfg_64);
440 break; 831 break;
441 default: 832 default:
442 /* sorry, don't know about them */ 833 /* sorry, don't know about them */
@@ -445,13 +836,11 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
445 break; 836 break;
446 case 0xF: 837 case 0xF:
447 switch (c->x86_model) { 838 switch (c->x86_model) {
448#ifdef __i386__
449 case 0x0: 839 case 0x0:
450 case 0x1: 840 case 0x1:
451 case 0x2: /* Netburst */ 841 case 0x2: /* Netburst */
452 ds_configure(&ds_cfg_netburst); 842 ds_configure(&ds_cfg_var);
453 break; 843 break;
454#endif /* _i386_ */
455 default: 844 default:
456 /* sorry, don't know about them */ 845 /* sorry, don't know about them */
457 break; 846 break;
@@ -462,3 +851,14 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
462 break; 851 break;
463 } 852 }
464} 853}
854
855void ds_free(struct ds_context *context)
856{
857 /* This is called when the task owning the parameter context
858 * is dying. There should not be any user of that context left
859 * to disturb us, anymore. */
860 unsigned long leftovers = context->count;
861 while (leftovers--)
862 ds_put_context(context);
863}
864#endif /* CONFIG_X86_DS */
diff --git a/arch/x86/kernel/dumpstack_32.c b/arch/x86/kernel/dumpstack_32.c
new file mode 100644
index 000000000000..1a78180f08d3
--- /dev/null
+++ b/arch/x86/kernel/dumpstack_32.c
@@ -0,0 +1,449 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 */
5#include <linux/kallsyms.h>
6#include <linux/kprobes.h>
7#include <linux/uaccess.h>
8#include <linux/utsname.h>
9#include <linux/hardirq.h>
10#include <linux/kdebug.h>
11#include <linux/module.h>
12#include <linux/ptrace.h>
13#include <linux/kexec.h>
14#include <linux/bug.h>
15#include <linux/nmi.h>
16#include <linux/sysfs.h>
17
18#include <asm/stacktrace.h>
19
20#define STACKSLOTS_PER_LINE 8
21#define get_bp(bp) asm("movl %%ebp, %0" : "=r" (bp) :)
22
23int panic_on_unrecovered_nmi;
24int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
25static unsigned int code_bytes = 64;
26static int die_counter;
27
28void printk_address(unsigned long address, int reliable)
29{
30 printk(" [<%p>] %s%pS\n", (void *) address,
31 reliable ? "" : "? ", (void *) address);
32}
33
34static inline int valid_stack_ptr(struct thread_info *tinfo,
35 void *p, unsigned int size, void *end)
36{
37 void *t = tinfo;
38 if (end) {
39 if (p < end && p >= (end-THREAD_SIZE))
40 return 1;
41 else
42 return 0;
43 }
44 return p > t && p < t + THREAD_SIZE - size;
45}
46
47/* The form of the top of the frame on the stack */
48struct stack_frame {
49 struct stack_frame *next_frame;
50 unsigned long return_address;
51};
52
53static inline unsigned long
54print_context_stack(struct thread_info *tinfo,
55 unsigned long *stack, unsigned long bp,
56 const struct stacktrace_ops *ops, void *data,
57 unsigned long *end)
58{
59 struct stack_frame *frame = (struct stack_frame *)bp;
60
61 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
62 unsigned long addr;
63
64 addr = *stack;
65 if (__kernel_text_address(addr)) {
66 if ((unsigned long) stack == bp + sizeof(long)) {
67 ops->address(data, addr, 1);
68 frame = frame->next_frame;
69 bp = (unsigned long) frame;
70 } else {
71 ops->address(data, addr, bp == 0);
72 }
73 }
74 stack++;
75 }
76 return bp;
77}
78
79void dump_trace(struct task_struct *task, struct pt_regs *regs,
80 unsigned long *stack, unsigned long bp,
81 const struct stacktrace_ops *ops, void *data)
82{
83 if (!task)
84 task = current;
85
86 if (!stack) {
87 unsigned long dummy;
88 stack = &dummy;
89 if (task && task != current)
90 stack = (unsigned long *)task->thread.sp;
91 }
92
93#ifdef CONFIG_FRAME_POINTER
94 if (!bp) {
95 if (task == current) {
96 /* Grab bp right from our regs */
97 get_bp(bp);
98 } else {
99 /* bp is the last reg pushed by switch_to */
100 bp = *(unsigned long *) task->thread.sp;
101 }
102 }
103#endif
104
105 for (;;) {
106 struct thread_info *context;
107
108 context = (struct thread_info *)
109 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
110 bp = print_context_stack(context, stack, bp, ops, data, NULL);
111
112 stack = (unsigned long *)context->previous_esp;
113 if (!stack)
114 break;
115 if (ops->stack(data, "IRQ") < 0)
116 break;
117 touch_nmi_watchdog();
118 }
119}
120EXPORT_SYMBOL(dump_trace);
121
122static void
123print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
124{
125 printk(data);
126 print_symbol(msg, symbol);
127 printk("\n");
128}
129
130static void print_trace_warning(void *data, char *msg)
131{
132 printk("%s%s\n", (char *)data, msg);
133}
134
135static int print_trace_stack(void *data, char *name)
136{
137 printk("%s <%s> ", (char *)data, name);
138 return 0;
139}
140
141/*
142 * Print one address/symbol entries per line.
143 */
144static void print_trace_address(void *data, unsigned long addr, int reliable)
145{
146 touch_nmi_watchdog();
147 printk(data);
148 printk_address(addr, reliable);
149}
150
151static const struct stacktrace_ops print_trace_ops = {
152 .warning = print_trace_warning,
153 .warning_symbol = print_trace_warning_symbol,
154 .stack = print_trace_stack,
155 .address = print_trace_address,
156};
157
158static void
159show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
160 unsigned long *stack, unsigned long bp, char *log_lvl)
161{
162 printk("%sCall Trace:\n", log_lvl);
163 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
164}
165
166void show_trace(struct task_struct *task, struct pt_regs *regs,
167 unsigned long *stack, unsigned long bp)
168{
169 show_trace_log_lvl(task, regs, stack, bp, "");
170}
171
172static void
173show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
174 unsigned long *sp, unsigned long bp, char *log_lvl)
175{
176 unsigned long *stack;
177 int i;
178
179 if (sp == NULL) {
180 if (task)
181 sp = (unsigned long *)task->thread.sp;
182 else
183 sp = (unsigned long *)&sp;
184 }
185
186 stack = sp;
187 for (i = 0; i < kstack_depth_to_print; i++) {
188 if (kstack_end(stack))
189 break;
190 if (i && ((i % STACKSLOTS_PER_LINE) == 0))
191 printk("\n%s", log_lvl);
192 printk(" %08lx", *stack++);
193 touch_nmi_watchdog();
194 }
195 printk("\n");
196 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
197}
198
199void show_stack(struct task_struct *task, unsigned long *sp)
200{
201 show_stack_log_lvl(task, NULL, sp, 0, "");
202}
203
204/*
205 * The architecture-independent dump_stack generator
206 */
207void dump_stack(void)
208{
209 unsigned long bp = 0;
210 unsigned long stack;
211
212#ifdef CONFIG_FRAME_POINTER
213 if (!bp)
214 get_bp(bp);
215#endif
216
217 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
218 current->pid, current->comm, print_tainted(),
219 init_utsname()->release,
220 (int)strcspn(init_utsname()->version, " "),
221 init_utsname()->version);
222 show_trace(NULL, NULL, &stack, bp);
223}
224
225EXPORT_SYMBOL(dump_stack);
226
227void show_registers(struct pt_regs *regs)
228{
229 int i;
230
231 print_modules();
232 __show_regs(regs, 0);
233
234 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)\n",
235 TASK_COMM_LEN, current->comm, task_pid_nr(current),
236 current_thread_info(), current, task_thread_info(current));
237 /*
238 * When in-kernel, we also print out the stack and code at the
239 * time of the fault..
240 */
241 if (!user_mode_vm(regs)) {
242 unsigned int code_prologue = code_bytes * 43 / 64;
243 unsigned int code_len = code_bytes;
244 unsigned char c;
245 u8 *ip;
246
247 printk(KERN_EMERG "Stack:\n");
248 show_stack_log_lvl(NULL, regs, &regs->sp,
249 0, KERN_EMERG);
250
251 printk(KERN_EMERG "Code: ");
252
253 ip = (u8 *)regs->ip - code_prologue;
254 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
255 /* try starting at IP */
256 ip = (u8 *)regs->ip;
257 code_len = code_len - code_prologue + 1;
258 }
259 for (i = 0; i < code_len; i++, ip++) {
260 if (ip < (u8 *)PAGE_OFFSET ||
261 probe_kernel_address(ip, c)) {
262 printk(" Bad EIP value.");
263 break;
264 }
265 if (ip == (u8 *)regs->ip)
266 printk("<%02x> ", c);
267 else
268 printk("%02x ", c);
269 }
270 }
271 printk("\n");
272}
273
274int is_valid_bugaddr(unsigned long ip)
275{
276 unsigned short ud2;
277
278 if (ip < PAGE_OFFSET)
279 return 0;
280 if (probe_kernel_address((unsigned short *)ip, ud2))
281 return 0;
282
283 return ud2 == 0x0b0f;
284}
285
286static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
287static int die_owner = -1;
288static unsigned int die_nest_count;
289
290unsigned __kprobes long oops_begin(void)
291{
292 unsigned long flags;
293
294 oops_enter();
295
296 if (die_owner != raw_smp_processor_id()) {
297 console_verbose();
298 raw_local_irq_save(flags);
299 __raw_spin_lock(&die_lock);
300 die_owner = smp_processor_id();
301 die_nest_count = 0;
302 bust_spinlocks(1);
303 } else {
304 raw_local_irq_save(flags);
305 }
306 die_nest_count++;
307 return flags;
308}
309
310void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
311{
312 bust_spinlocks(0);
313 die_owner = -1;
314 add_taint(TAINT_DIE);
315 __raw_spin_unlock(&die_lock);
316 raw_local_irq_restore(flags);
317
318 if (!regs)
319 return;
320
321 if (kexec_should_crash(current))
322 crash_kexec(regs);
323 if (in_interrupt())
324 panic("Fatal exception in interrupt");
325 if (panic_on_oops)
326 panic("Fatal exception");
327 oops_exit();
328 do_exit(signr);
329}
330
331int __kprobes __die(const char *str, struct pt_regs *regs, long err)
332{
333 unsigned short ss;
334 unsigned long sp;
335
336 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
337#ifdef CONFIG_PREEMPT
338 printk("PREEMPT ");
339#endif
340#ifdef CONFIG_SMP
341 printk("SMP ");
342#endif
343#ifdef CONFIG_DEBUG_PAGEALLOC
344 printk("DEBUG_PAGEALLOC");
345#endif
346 printk("\n");
347 sysfs_printk_last_file();
348 if (notify_die(DIE_OOPS, str, regs, err,
349 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
350 return 1;
351
352 show_registers(regs);
353 /* Executive summary in case the oops scrolled away */
354 sp = (unsigned long) (&regs->sp);
355 savesegment(ss, ss);
356 if (user_mode(regs)) {
357 sp = regs->sp;
358 ss = regs->ss & 0xffff;
359 }
360 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
361 print_symbol("%s", regs->ip);
362 printk(" SS:ESP %04x:%08lx\n", ss, sp);
363 return 0;
364}
365
366/*
367 * This is gone through when something in the kernel has done something bad
368 * and is about to be terminated:
369 */
370void die(const char *str, struct pt_regs *regs, long err)
371{
372 unsigned long flags = oops_begin();
373
374 if (die_nest_count < 3) {
375 report_bug(regs->ip, regs);
376
377 if (__die(str, regs, err))
378 regs = NULL;
379 } else {
380 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
381 }
382
383 oops_end(flags, regs, SIGSEGV);
384}
385
386static DEFINE_SPINLOCK(nmi_print_lock);
387
388void notrace __kprobes
389die_nmi(char *str, struct pt_regs *regs, int do_panic)
390{
391 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
392 return;
393
394 spin_lock(&nmi_print_lock);
395 /*
396 * We are in trouble anyway, lets at least try
397 * to get a message out:
398 */
399 bust_spinlocks(1);
400 printk(KERN_EMERG "%s", str);
401 printk(" on CPU%d, ip %08lx, registers:\n",
402 smp_processor_id(), regs->ip);
403 show_registers(regs);
404 if (do_panic)
405 panic("Non maskable interrupt");
406 console_silent();
407 spin_unlock(&nmi_print_lock);
408 bust_spinlocks(0);
409
410 /*
411 * If we are in kernel we are probably nested up pretty bad
412 * and might aswell get out now while we still can:
413 */
414 if (!user_mode_vm(regs)) {
415 current->thread.trap_no = 2;
416 crash_kexec(regs);
417 }
418
419 do_exit(SIGSEGV);
420}
421
422static int __init oops_setup(char *s)
423{
424 if (!s)
425 return -EINVAL;
426 if (!strcmp(s, "panic"))
427 panic_on_oops = 1;
428 return 0;
429}
430early_param("oops", oops_setup);
431
432static int __init kstack_setup(char *s)
433{
434 if (!s)
435 return -EINVAL;
436 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
437 return 0;
438}
439early_param("kstack", kstack_setup);
440
441static int __init code_bytes_setup(char *s)
442{
443 code_bytes = simple_strtoul(s, NULL, 0);
444 if (code_bytes > 8192)
445 code_bytes = 8192;
446
447 return 1;
448}
449__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/dumpstack_64.c b/arch/x86/kernel/dumpstack_64.c
new file mode 100644
index 000000000000..96a5db7da8a7
--- /dev/null
+++ b/arch/x86/kernel/dumpstack_64.c
@@ -0,0 +1,575 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 */
5#include <linux/kallsyms.h>
6#include <linux/kprobes.h>
7#include <linux/uaccess.h>
8#include <linux/utsname.h>
9#include <linux/hardirq.h>
10#include <linux/kdebug.h>
11#include <linux/module.h>
12#include <linux/ptrace.h>
13#include <linux/kexec.h>
14#include <linux/bug.h>
15#include <linux/nmi.h>
16#include <linux/sysfs.h>
17
18#include <asm/stacktrace.h>
19
20#define STACKSLOTS_PER_LINE 4
21#define get_bp(bp) asm("movq %%rbp, %0" : "=r" (bp) :)
22
23int panic_on_unrecovered_nmi;
24int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
25static unsigned int code_bytes = 64;
26static int die_counter;
27
28void printk_address(unsigned long address, int reliable)
29{
30 printk(" [<%p>] %s%pS\n", (void *) address,
31 reliable ? "" : "? ", (void *) address);
32}
33
34static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
35 unsigned *usedp, char **idp)
36{
37 static char ids[][8] = {
38 [DEBUG_STACK - 1] = "#DB",
39 [NMI_STACK - 1] = "NMI",
40 [DOUBLEFAULT_STACK - 1] = "#DF",
41 [STACKFAULT_STACK - 1] = "#SS",
42 [MCE_STACK - 1] = "#MC",
43#if DEBUG_STKSZ > EXCEPTION_STKSZ
44 [N_EXCEPTION_STACKS ...
45 N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
46#endif
47 };
48 unsigned k;
49
50 /*
51 * Iterate over all exception stacks, and figure out whether
52 * 'stack' is in one of them:
53 */
54 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
55 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
56 /*
57 * Is 'stack' above this exception frame's end?
58 * If yes then skip to the next frame.
59 */
60 if (stack >= end)
61 continue;
62 /*
63 * Is 'stack' above this exception frame's start address?
64 * If yes then we found the right frame.
65 */
66 if (stack >= end - EXCEPTION_STKSZ) {
67 /*
68 * Make sure we only iterate through an exception
69 * stack once. If it comes up for the second time
70 * then there's something wrong going on - just
71 * break out and return NULL:
72 */
73 if (*usedp & (1U << k))
74 break;
75 *usedp |= 1U << k;
76 *idp = ids[k];
77 return (unsigned long *)end;
78 }
79 /*
80 * If this is a debug stack, and if it has a larger size than
81 * the usual exception stacks, then 'stack' might still
82 * be within the lower portion of the debug stack:
83 */
84#if DEBUG_STKSZ > EXCEPTION_STKSZ
85 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
86 unsigned j = N_EXCEPTION_STACKS - 1;
87
88 /*
89 * Black magic. A large debug stack is composed of
90 * multiple exception stack entries, which we
91 * iterate through now. Dont look:
92 */
93 do {
94 ++j;
95 end -= EXCEPTION_STKSZ;
96 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
97 } while (stack < end - EXCEPTION_STKSZ);
98 if (*usedp & (1U << j))
99 break;
100 *usedp |= 1U << j;
101 *idp = ids[j];
102 return (unsigned long *)end;
103 }
104#endif
105 }
106 return NULL;
107}
108
109/*
110 * x86-64 can have up to three kernel stacks:
111 * process stack
112 * interrupt stack
113 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
114 */
115
116static inline int valid_stack_ptr(struct thread_info *tinfo,
117 void *p, unsigned int size, void *end)
118{
119 void *t = tinfo;
120 if (end) {
121 if (p < end && p >= (end-THREAD_SIZE))
122 return 1;
123 else
124 return 0;
125 }
126 return p > t && p < t + THREAD_SIZE - size;
127}
128
129/* The form of the top of the frame on the stack */
130struct stack_frame {
131 struct stack_frame *next_frame;
132 unsigned long return_address;
133};
134
135static inline unsigned long
136print_context_stack(struct thread_info *tinfo,
137 unsigned long *stack, unsigned long bp,
138 const struct stacktrace_ops *ops, void *data,
139 unsigned long *end)
140{
141 struct stack_frame *frame = (struct stack_frame *)bp;
142
143 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
144 unsigned long addr;
145
146 addr = *stack;
147 if (__kernel_text_address(addr)) {
148 if ((unsigned long) stack == bp + sizeof(long)) {
149 ops->address(data, addr, 1);
150 frame = frame->next_frame;
151 bp = (unsigned long) frame;
152 } else {
153 ops->address(data, addr, bp == 0);
154 }
155 }
156 stack++;
157 }
158 return bp;
159}
160
161void dump_trace(struct task_struct *task, struct pt_regs *regs,
162 unsigned long *stack, unsigned long bp,
163 const struct stacktrace_ops *ops, void *data)
164{
165 const unsigned cpu = get_cpu();
166 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
167 unsigned used = 0;
168 struct thread_info *tinfo;
169
170 if (!task)
171 task = current;
172
173 if (!stack) {
174 unsigned long dummy;
175 stack = &dummy;
176 if (task && task != current)
177 stack = (unsigned long *)task->thread.sp;
178 }
179
180#ifdef CONFIG_FRAME_POINTER
181 if (!bp) {
182 if (task == current) {
183 /* Grab bp right from our regs */
184 get_bp(bp);
185 } else {
186 /* bp is the last reg pushed by switch_to */
187 bp = *(unsigned long *) task->thread.sp;
188 }
189 }
190#endif
191
192 /*
193 * Print function call entries in all stacks, starting at the
194 * current stack address. If the stacks consist of nested
195 * exceptions
196 */
197 tinfo = task_thread_info(task);
198 for (;;) {
199 char *id;
200 unsigned long *estack_end;
201 estack_end = in_exception_stack(cpu, (unsigned long)stack,
202 &used, &id);
203
204 if (estack_end) {
205 if (ops->stack(data, id) < 0)
206 break;
207
208 bp = print_context_stack(tinfo, stack, bp, ops,
209 data, estack_end);
210 ops->stack(data, "<EOE>");
211 /*
212 * We link to the next stack via the
213 * second-to-last pointer (index -2 to end) in the
214 * exception stack:
215 */
216 stack = (unsigned long *) estack_end[-2];
217 continue;
218 }
219 if (irqstack_end) {
220 unsigned long *irqstack;
221 irqstack = irqstack_end -
222 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
223
224 if (stack >= irqstack && stack < irqstack_end) {
225 if (ops->stack(data, "IRQ") < 0)
226 break;
227 bp = print_context_stack(tinfo, stack, bp,
228 ops, data, irqstack_end);
229 /*
230 * We link to the next stack (which would be
231 * the process stack normally) the last
232 * pointer (index -1 to end) in the IRQ stack:
233 */
234 stack = (unsigned long *) (irqstack_end[-1]);
235 irqstack_end = NULL;
236 ops->stack(data, "EOI");
237 continue;
238 }
239 }
240 break;
241 }
242
243 /*
244 * This handles the process stack:
245 */
246 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL);
247 put_cpu();
248}
249EXPORT_SYMBOL(dump_trace);
250
251static void
252print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
253{
254 printk(data);
255 print_symbol(msg, symbol);
256 printk("\n");
257}
258
259static void print_trace_warning(void *data, char *msg)
260{
261 printk("%s%s\n", (char *)data, msg);
262}
263
264static int print_trace_stack(void *data, char *name)
265{
266 printk("%s <%s> ", (char *)data, name);
267 return 0;
268}
269
270/*
271 * Print one address/symbol entries per line.
272 */
273static void print_trace_address(void *data, unsigned long addr, int reliable)
274{
275 touch_nmi_watchdog();
276 printk(data);
277 printk_address(addr, reliable);
278}
279
280static const struct stacktrace_ops print_trace_ops = {
281 .warning = print_trace_warning,
282 .warning_symbol = print_trace_warning_symbol,
283 .stack = print_trace_stack,
284 .address = print_trace_address,
285};
286
287static void
288show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
289 unsigned long *stack, unsigned long bp, char *log_lvl)
290{
291 printk("%sCall Trace:\n", log_lvl);
292 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
293}
294
295void show_trace(struct task_struct *task, struct pt_regs *regs,
296 unsigned long *stack, unsigned long bp)
297{
298 show_trace_log_lvl(task, regs, stack, bp, "");
299}
300
301static void
302show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
303 unsigned long *sp, unsigned long bp, char *log_lvl)
304{
305 unsigned long *stack;
306 int i;
307 const int cpu = smp_processor_id();
308 unsigned long *irqstack_end =
309 (unsigned long *) (cpu_pda(cpu)->irqstackptr);
310 unsigned long *irqstack =
311 (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
312
313 /*
314 * debugging aid: "show_stack(NULL, NULL);" prints the
315 * back trace for this cpu.
316 */
317
318 if (sp == NULL) {
319 if (task)
320 sp = (unsigned long *)task->thread.sp;
321 else
322 sp = (unsigned long *)&sp;
323 }
324
325 stack = sp;
326 for (i = 0; i < kstack_depth_to_print; i++) {
327 if (stack >= irqstack && stack <= irqstack_end) {
328 if (stack == irqstack_end) {
329 stack = (unsigned long *) (irqstack_end[-1]);
330 printk(" <EOI> ");
331 }
332 } else {
333 if (((long) stack & (THREAD_SIZE-1)) == 0)
334 break;
335 }
336 if (i && ((i % STACKSLOTS_PER_LINE) == 0))
337 printk("\n%s", log_lvl);
338 printk(" %016lx", *stack++);
339 touch_nmi_watchdog();
340 }
341 printk("\n");
342 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
343}
344
345void show_stack(struct task_struct *task, unsigned long *sp)
346{
347 show_stack_log_lvl(task, NULL, sp, 0, "");
348}
349
350/*
351 * The architecture-independent dump_stack generator
352 */
353void dump_stack(void)
354{
355 unsigned long bp = 0;
356 unsigned long stack;
357
358#ifdef CONFIG_FRAME_POINTER
359 if (!bp)
360 get_bp(bp);
361#endif
362
363 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
364 current->pid, current->comm, print_tainted(),
365 init_utsname()->release,
366 (int)strcspn(init_utsname()->version, " "),
367 init_utsname()->version);
368 show_trace(NULL, NULL, &stack, bp);
369}
370EXPORT_SYMBOL(dump_stack);
371
372void show_registers(struct pt_regs *regs)
373{
374 int i;
375 unsigned long sp;
376 const int cpu = smp_processor_id();
377 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
378
379 sp = regs->sp;
380 printk("CPU %d ", cpu);
381 __show_regs(regs, 1);
382 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
383 cur->comm, cur->pid, task_thread_info(cur), cur);
384
385 /*
386 * When in-kernel, we also print out the stack and code at the
387 * time of the fault..
388 */
389 if (!user_mode(regs)) {
390 unsigned int code_prologue = code_bytes * 43 / 64;
391 unsigned int code_len = code_bytes;
392 unsigned char c;
393 u8 *ip;
394
395 printk(KERN_EMERG "Stack:\n");
396 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
397 regs->bp, KERN_EMERG);
398
399 printk(KERN_EMERG "Code: ");
400
401 ip = (u8 *)regs->ip - code_prologue;
402 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
403 /* try starting at IP */
404 ip = (u8 *)regs->ip;
405 code_len = code_len - code_prologue + 1;
406 }
407 for (i = 0; i < code_len; i++, ip++) {
408 if (ip < (u8 *)PAGE_OFFSET ||
409 probe_kernel_address(ip, c)) {
410 printk(" Bad RIP value.");
411 break;
412 }
413 if (ip == (u8 *)regs->ip)
414 printk("<%02x> ", c);
415 else
416 printk("%02x ", c);
417 }
418 }
419 printk("\n");
420}
421
422int is_valid_bugaddr(unsigned long ip)
423{
424 unsigned short ud2;
425
426 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
427 return 0;
428
429 return ud2 == 0x0b0f;
430}
431
432static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
433static int die_owner = -1;
434static unsigned int die_nest_count;
435
436unsigned __kprobes long oops_begin(void)
437{
438 int cpu;
439 unsigned long flags;
440
441 oops_enter();
442
443 /* racy, but better than risking deadlock. */
444 raw_local_irq_save(flags);
445 cpu = smp_processor_id();
446 if (!__raw_spin_trylock(&die_lock)) {
447 if (cpu == die_owner)
448 /* nested oops. should stop eventually */;
449 else
450 __raw_spin_lock(&die_lock);
451 }
452 die_nest_count++;
453 die_owner = cpu;
454 console_verbose();
455 bust_spinlocks(1);
456 return flags;
457}
458
459void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
460{
461 die_owner = -1;
462 bust_spinlocks(0);
463 die_nest_count--;
464 if (!die_nest_count)
465 /* Nest count reaches zero, release the lock. */
466 __raw_spin_unlock(&die_lock);
467 raw_local_irq_restore(flags);
468 if (!regs) {
469 oops_exit();
470 return;
471 }
472 if (in_interrupt())
473 panic("Fatal exception in interrupt");
474 if (panic_on_oops)
475 panic("Fatal exception");
476 oops_exit();
477 do_exit(signr);
478}
479
480int __kprobes __die(const char *str, struct pt_regs *regs, long err)
481{
482 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
483#ifdef CONFIG_PREEMPT
484 printk("PREEMPT ");
485#endif
486#ifdef CONFIG_SMP
487 printk("SMP ");
488#endif
489#ifdef CONFIG_DEBUG_PAGEALLOC
490 printk("DEBUG_PAGEALLOC");
491#endif
492 printk("\n");
493 sysfs_printk_last_file();
494 if (notify_die(DIE_OOPS, str, regs, err,
495 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
496 return 1;
497
498 show_registers(regs);
499 add_taint(TAINT_DIE);
500 /* Executive summary in case the oops scrolled away */
501 printk(KERN_ALERT "RIP ");
502 printk_address(regs->ip, 1);
503 printk(" RSP <%016lx>\n", regs->sp);
504 if (kexec_should_crash(current))
505 crash_kexec(regs);
506 return 0;
507}
508
509void die(const char *str, struct pt_regs *regs, long err)
510{
511 unsigned long flags = oops_begin();
512
513 if (!user_mode(regs))
514 report_bug(regs->ip, regs);
515
516 if (__die(str, regs, err))
517 regs = NULL;
518 oops_end(flags, regs, SIGSEGV);
519}
520
521notrace __kprobes void
522die_nmi(char *str, struct pt_regs *regs, int do_panic)
523{
524 unsigned long flags;
525
526 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
527 return;
528
529 flags = oops_begin();
530 /*
531 * We are in trouble anyway, lets at least try
532 * to get a message out.
533 */
534 printk(KERN_EMERG "%s", str);
535 printk(" on CPU%d, ip %08lx, registers:\n",
536 smp_processor_id(), regs->ip);
537 show_registers(regs);
538 if (kexec_should_crash(current))
539 crash_kexec(regs);
540 if (do_panic || panic_on_oops)
541 panic("Non maskable interrupt");
542 oops_end(flags, NULL, SIGBUS);
543 nmi_exit();
544 local_irq_enable();
545 do_exit(SIGBUS);
546}
547
548static int __init oops_setup(char *s)
549{
550 if (!s)
551 return -EINVAL;
552 if (!strcmp(s, "panic"))
553 panic_on_oops = 1;
554 return 0;
555}
556early_param("oops", oops_setup);
557
558static int __init kstack_setup(char *s)
559{
560 if (!s)
561 return -EINVAL;
562 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
563 return 0;
564}
565early_param("kstack", kstack_setup);
566
567static int __init code_bytes_setup(char *s)
568{
569 code_bytes = simple_strtoul(s, NULL, 0);
570 if (code_bytes > 8192)
571 code_bytes = 8192;
572
573 return 1;
574}
575__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
index 9af89078f7bb..ce97bf3bed12 100644
--- a/arch/x86/kernel/e820.c
+++ b/arch/x86/kernel/e820.c
@@ -148,6 +148,9 @@ void __init e820_print_map(char *who)
148 case E820_NVS: 148 case E820_NVS:
149 printk(KERN_CONT "(ACPI NVS)\n"); 149 printk(KERN_CONT "(ACPI NVS)\n");
150 break; 150 break;
151 case E820_UNUSABLE:
152 printk("(unusable)\n");
153 break;
151 default: 154 default:
152 printk(KERN_CONT "type %u\n", e820.map[i].type); 155 printk(KERN_CONT "type %u\n", e820.map[i].type);
153 break; 156 break;
@@ -1203,7 +1206,7 @@ static int __init parse_memmap_opt(char *p)
1203 if (!p) 1206 if (!p)
1204 return -EINVAL; 1207 return -EINVAL;
1205 1208
1206 if (!strcmp(p, "exactmap")) { 1209 if (!strncmp(p, "exactmap", 8)) {
1207#ifdef CONFIG_CRASH_DUMP 1210#ifdef CONFIG_CRASH_DUMP
1208 /* 1211 /*
1209 * If we are doing a crash dump, we still need to know 1212 * If we are doing a crash dump, we still need to know
@@ -1260,6 +1263,7 @@ static inline const char *e820_type_to_string(int e820_type)
1260 case E820_RAM: return "System RAM"; 1263 case E820_RAM: return "System RAM";
1261 case E820_ACPI: return "ACPI Tables"; 1264 case E820_ACPI: return "ACPI Tables";
1262 case E820_NVS: return "ACPI Non-volatile Storage"; 1265 case E820_NVS: return "ACPI Non-volatile Storage";
1266 case E820_UNUSABLE: return "Unusable memory";
1263 default: return "reserved"; 1267 default: return "reserved";
1264 } 1268 }
1265} 1269}
@@ -1267,6 +1271,7 @@ static inline const char *e820_type_to_string(int e820_type)
1267/* 1271/*
1268 * Mark e820 reserved areas as busy for the resource manager. 1272 * Mark e820 reserved areas as busy for the resource manager.
1269 */ 1273 */
1274static struct resource __initdata *e820_res;
1270void __init e820_reserve_resources(void) 1275void __init e820_reserve_resources(void)
1271{ 1276{
1272 int i; 1277 int i;
@@ -1274,20 +1279,26 @@ void __init e820_reserve_resources(void)
1274 u64 end; 1279 u64 end;
1275 1280
1276 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map); 1281 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1282 e820_res = res;
1277 for (i = 0; i < e820.nr_map; i++) { 1283 for (i = 0; i < e820.nr_map; i++) {
1278 end = e820.map[i].addr + e820.map[i].size - 1; 1284 end = e820.map[i].addr + e820.map[i].size - 1;
1279#ifndef CONFIG_RESOURCES_64BIT 1285 if (end != (resource_size_t)end) {
1280 if (end > 0x100000000ULL) {
1281 res++; 1286 res++;
1282 continue; 1287 continue;
1283 } 1288 }
1284#endif
1285 res->name = e820_type_to_string(e820.map[i].type); 1289 res->name = e820_type_to_string(e820.map[i].type);
1286 res->start = e820.map[i].addr; 1290 res->start = e820.map[i].addr;
1287 res->end = end; 1291 res->end = end;
1288 1292
1289 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; 1293 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1290 insert_resource(&iomem_resource, res); 1294
1295 /*
1296 * don't register the region that could be conflicted with
1297 * pci device BAR resource and insert them later in
1298 * pcibios_resource_survey()
1299 */
1300 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20))
1301 insert_resource(&iomem_resource, res);
1291 res++; 1302 res++;
1292 } 1303 }
1293 1304
@@ -1299,6 +1310,19 @@ void __init e820_reserve_resources(void)
1299 } 1310 }
1300} 1311}
1301 1312
1313void __init e820_reserve_resources_late(void)
1314{
1315 int i;
1316 struct resource *res;
1317
1318 res = e820_res;
1319 for (i = 0; i < e820.nr_map; i++) {
1320 if (!res->parent && res->end)
1321 reserve_region_with_split(&iomem_resource, res->start, res->end, res->name);
1322 res++;
1323 }
1324}
1325
1302char *__init default_machine_specific_memory_setup(void) 1326char *__init default_machine_specific_memory_setup(void)
1303{ 1327{
1304 char *who = "BIOS-e820"; 1328 char *who = "BIOS-e820";
diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
index 4353cf5e6fac..733c4f8d42ea 100644
--- a/arch/x86/kernel/early-quirks.c
+++ b/arch/x86/kernel/early-quirks.c
@@ -95,6 +95,66 @@ static void __init nvidia_bugs(int num, int slot, int func)
95 95
96} 96}
97 97
98static u32 ati_ixp4x0_rev(int num, int slot, int func)
99{
100 u32 d;
101 u8 b;
102
103 b = read_pci_config_byte(num, slot, func, 0xac);
104 b &= ~(1<<5);
105 write_pci_config_byte(num, slot, func, 0xac, b);
106
107 d = read_pci_config(num, slot, func, 0x70);
108 d |= 1<<8;
109 write_pci_config(num, slot, func, 0x70, d);
110
111 d = read_pci_config(num, slot, func, 0x8);
112 d &= 0xff;
113 return d;
114}
115
116static void __init ati_bugs(int num, int slot, int func)
117{
118#if defined(CONFIG_ACPI) && defined (CONFIG_X86_IO_APIC)
119 u32 d;
120 u8 b;
121
122 if (acpi_use_timer_override)
123 return;
124
125 d = ati_ixp4x0_rev(num, slot, func);
126 if (d < 0x82)
127 acpi_skip_timer_override = 1;
128 else {
129 /* check for IRQ0 interrupt swap */
130 outb(0x72, 0xcd6); b = inb(0xcd7);
131 if (!(b & 0x2))
132 acpi_skip_timer_override = 1;
133 }
134
135 if (acpi_skip_timer_override) {
136 printk(KERN_INFO "SB4X0 revision 0x%x\n", d);
137 printk(KERN_INFO "Ignoring ACPI timer override.\n");
138 printk(KERN_INFO "If you got timer trouble "
139 "try acpi_use_timer_override\n");
140 }
141#endif
142}
143
144#ifdef CONFIG_DMAR
145static void __init intel_g33_dmar(int num, int slot, int func)
146{
147 struct acpi_table_header *dmar_tbl;
148 acpi_status status;
149
150 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
151 if (ACPI_SUCCESS(status)) {
152 printk(KERN_INFO "BIOS BUG: DMAR advertised on Intel G31/G33 chipset -- ignoring\n");
153 dmar_disabled = 1;
154 }
155}
156#endif
157
98#define QFLAG_APPLY_ONCE 0x1 158#define QFLAG_APPLY_ONCE 0x1
99#define QFLAG_APPLIED 0x2 159#define QFLAG_APPLIED 0x2
100#define QFLAG_DONE (QFLAG_APPLY_ONCE|QFLAG_APPLIED) 160#define QFLAG_DONE (QFLAG_APPLY_ONCE|QFLAG_APPLIED)
@@ -114,6 +174,12 @@ static struct chipset early_qrk[] __initdata = {
114 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, via_bugs }, 174 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, via_bugs },
115 { PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB, 175 { PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB,
116 PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, fix_hypertransport_config }, 176 PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, fix_hypertransport_config },
177 { PCI_VENDOR_ID_ATI, PCI_DEVICE_ID_ATI_IXP400_SMBUS,
178 PCI_CLASS_SERIAL_SMBUS, PCI_ANY_ID, 0, ati_bugs },
179#ifdef CONFIG_DMAR
180 { PCI_VENDOR_ID_INTEL, 0x29c0,
181 PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, intel_g33_dmar },
182#endif
117 {} 183 {}
118}; 184};
119 185
diff --git a/arch/x86/kernel/early_printk.c b/arch/x86/kernel/early_printk.c
index ff9e7350da54..34ad997d3834 100644
--- a/arch/x86/kernel/early_printk.c
+++ b/arch/x86/kernel/early_printk.c
@@ -3,11 +3,19 @@
3#include <linux/init.h> 3#include <linux/init.h>
4#include <linux/string.h> 4#include <linux/string.h>
5#include <linux/screen_info.h> 5#include <linux/screen_info.h>
6#include <linux/usb/ch9.h>
7#include <linux/pci_regs.h>
8#include <linux/pci_ids.h>
9#include <linux/errno.h>
6#include <asm/io.h> 10#include <asm/io.h>
7#include <asm/processor.h> 11#include <asm/processor.h>
8#include <asm/fcntl.h> 12#include <asm/fcntl.h>
9#include <asm/setup.h> 13#include <asm/setup.h>
10#include <xen/hvc-console.h> 14#include <xen/hvc-console.h>
15#include <asm/pci-direct.h>
16#include <asm/pgtable.h>
17#include <asm/fixmap.h>
18#include <linux/usb/ehci_def.h>
11 19
12/* Simple VGA output */ 20/* Simple VGA output */
13#define VGABASE (__ISA_IO_base + 0xb8000) 21#define VGABASE (__ISA_IO_base + 0xb8000)
@@ -78,6 +86,7 @@ static int early_serial_base = 0x3f8; /* ttyS0 */
78static int early_serial_putc(unsigned char ch) 86static int early_serial_putc(unsigned char ch)
79{ 87{
80 unsigned timeout = 0xffff; 88 unsigned timeout = 0xffff;
89
81 while ((inb(early_serial_base + LSR) & XMTRDY) == 0 && --timeout) 90 while ((inb(early_serial_base + LSR) & XMTRDY) == 0 && --timeout)
82 cpu_relax(); 91 cpu_relax();
83 outb(ch, early_serial_base + TXR); 92 outb(ch, early_serial_base + TXR);
@@ -111,7 +120,7 @@ static __init void early_serial_init(char *s)
111 if (!strncmp(s, "0x", 2)) { 120 if (!strncmp(s, "0x", 2)) {
112 early_serial_base = simple_strtoul(s, &e, 16); 121 early_serial_base = simple_strtoul(s, &e, 16);
113 } else { 122 } else {
114 static int bases[] = { 0x3f8, 0x2f8 }; 123 static const int __initconst bases[] = { 0x3f8, 0x2f8 };
115 124
116 if (!strncmp(s, "ttyS", 4)) 125 if (!strncmp(s, "ttyS", 4))
117 s += 4; 126 s += 4;
@@ -151,6 +160,721 @@ static struct console early_serial_console = {
151 .index = -1, 160 .index = -1,
152}; 161};
153 162
163#ifdef CONFIG_EARLY_PRINTK_DBGP
164
165static struct ehci_caps __iomem *ehci_caps;
166static struct ehci_regs __iomem *ehci_regs;
167static struct ehci_dbg_port __iomem *ehci_debug;
168static unsigned int dbgp_endpoint_out;
169
170struct ehci_dev {
171 u32 bus;
172 u32 slot;
173 u32 func;
174};
175
176static struct ehci_dev ehci_dev;
177
178#define USB_DEBUG_DEVNUM 127
179
180#define DBGP_DATA_TOGGLE 0x8800
181
182static inline u32 dbgp_pid_update(u32 x, u32 tok)
183{
184 return ((x ^ DBGP_DATA_TOGGLE) & 0xffff00) | (tok & 0xff);
185}
186
187static inline u32 dbgp_len_update(u32 x, u32 len)
188{
189 return (x & ~0x0f) | (len & 0x0f);
190}
191
192/*
193 * USB Packet IDs (PIDs)
194 */
195
196/* token */
197#define USB_PID_OUT 0xe1
198#define USB_PID_IN 0x69
199#define USB_PID_SOF 0xa5
200#define USB_PID_SETUP 0x2d
201/* handshake */
202#define USB_PID_ACK 0xd2
203#define USB_PID_NAK 0x5a
204#define USB_PID_STALL 0x1e
205#define USB_PID_NYET 0x96
206/* data */
207#define USB_PID_DATA0 0xc3
208#define USB_PID_DATA1 0x4b
209#define USB_PID_DATA2 0x87
210#define USB_PID_MDATA 0x0f
211/* Special */
212#define USB_PID_PREAMBLE 0x3c
213#define USB_PID_ERR 0x3c
214#define USB_PID_SPLIT 0x78
215#define USB_PID_PING 0xb4
216#define USB_PID_UNDEF_0 0xf0
217
218#define USB_PID_DATA_TOGGLE 0x88
219#define DBGP_CLAIM (DBGP_OWNER | DBGP_ENABLED | DBGP_INUSE)
220
221#define PCI_CAP_ID_EHCI_DEBUG 0xa
222
223#define HUB_ROOT_RESET_TIME 50 /* times are in msec */
224#define HUB_SHORT_RESET_TIME 10
225#define HUB_LONG_RESET_TIME 200
226#define HUB_RESET_TIMEOUT 500
227
228#define DBGP_MAX_PACKET 8
229
230static int dbgp_wait_until_complete(void)
231{
232 u32 ctrl;
233 int loop = 0x100000;
234
235 do {
236 ctrl = readl(&ehci_debug->control);
237 /* Stop when the transaction is finished */
238 if (ctrl & DBGP_DONE)
239 break;
240 } while (--loop > 0);
241
242 if (!loop)
243 return -1;
244
245 /*
246 * Now that we have observed the completed transaction,
247 * clear the done bit.
248 */
249 writel(ctrl | DBGP_DONE, &ehci_debug->control);
250 return (ctrl & DBGP_ERROR) ? -DBGP_ERRCODE(ctrl) : DBGP_LEN(ctrl);
251}
252
253static void dbgp_mdelay(int ms)
254{
255 int i;
256
257 while (ms--) {
258 for (i = 0; i < 1000; i++)
259 outb(0x1, 0x80);
260 }
261}
262
263static void dbgp_breath(void)
264{
265 /* Sleep to give the debug port a chance to breathe */
266}
267
268static int dbgp_wait_until_done(unsigned ctrl)
269{
270 u32 pids, lpid;
271 int ret;
272 int loop = 3;
273
274retry:
275 writel(ctrl | DBGP_GO, &ehci_debug->control);
276 ret = dbgp_wait_until_complete();
277 pids = readl(&ehci_debug->pids);
278 lpid = DBGP_PID_GET(pids);
279
280 if (ret < 0)
281 return ret;
282
283 /*
284 * If the port is getting full or it has dropped data
285 * start pacing ourselves, not necessary but it's friendly.
286 */
287 if ((lpid == USB_PID_NAK) || (lpid == USB_PID_NYET))
288 dbgp_breath();
289
290 /* If I get a NACK reissue the transmission */
291 if (lpid == USB_PID_NAK) {
292 if (--loop > 0)
293 goto retry;
294 }
295
296 return ret;
297}
298
299static void dbgp_set_data(const void *buf, int size)
300{
301 const unsigned char *bytes = buf;
302 u32 lo, hi;
303 int i;
304
305 lo = hi = 0;
306 for (i = 0; i < 4 && i < size; i++)
307 lo |= bytes[i] << (8*i);
308 for (; i < 8 && i < size; i++)
309 hi |= bytes[i] << (8*(i - 4));
310 writel(lo, &ehci_debug->data03);
311 writel(hi, &ehci_debug->data47);
312}
313
314static void dbgp_get_data(void *buf, int size)
315{
316 unsigned char *bytes = buf;
317 u32 lo, hi;
318 int i;
319
320 lo = readl(&ehci_debug->data03);
321 hi = readl(&ehci_debug->data47);
322 for (i = 0; i < 4 && i < size; i++)
323 bytes[i] = (lo >> (8*i)) & 0xff;
324 for (; i < 8 && i < size; i++)
325 bytes[i] = (hi >> (8*(i - 4))) & 0xff;
326}
327
328static int dbgp_bulk_write(unsigned devnum, unsigned endpoint,
329 const char *bytes, int size)
330{
331 u32 pids, addr, ctrl;
332 int ret;
333
334 if (size > DBGP_MAX_PACKET)
335 return -1;
336
337 addr = DBGP_EPADDR(devnum, endpoint);
338
339 pids = readl(&ehci_debug->pids);
340 pids = dbgp_pid_update(pids, USB_PID_OUT);
341
342 ctrl = readl(&ehci_debug->control);
343 ctrl = dbgp_len_update(ctrl, size);
344 ctrl |= DBGP_OUT;
345 ctrl |= DBGP_GO;
346
347 dbgp_set_data(bytes, size);
348 writel(addr, &ehci_debug->address);
349 writel(pids, &ehci_debug->pids);
350
351 ret = dbgp_wait_until_done(ctrl);
352 if (ret < 0)
353 return ret;
354
355 return ret;
356}
357
358static int dbgp_bulk_read(unsigned devnum, unsigned endpoint, void *data,
359 int size)
360{
361 u32 pids, addr, ctrl;
362 int ret;
363
364 if (size > DBGP_MAX_PACKET)
365 return -1;
366
367 addr = DBGP_EPADDR(devnum, endpoint);
368
369 pids = readl(&ehci_debug->pids);
370 pids = dbgp_pid_update(pids, USB_PID_IN);
371
372 ctrl = readl(&ehci_debug->control);
373 ctrl = dbgp_len_update(ctrl, size);
374 ctrl &= ~DBGP_OUT;
375 ctrl |= DBGP_GO;
376
377 writel(addr, &ehci_debug->address);
378 writel(pids, &ehci_debug->pids);
379 ret = dbgp_wait_until_done(ctrl);
380 if (ret < 0)
381 return ret;
382
383 if (size > ret)
384 size = ret;
385 dbgp_get_data(data, size);
386 return ret;
387}
388
389static int dbgp_control_msg(unsigned devnum, int requesttype, int request,
390 int value, int index, void *data, int size)
391{
392 u32 pids, addr, ctrl;
393 struct usb_ctrlrequest req;
394 int read;
395 int ret;
396
397 read = (requesttype & USB_DIR_IN) != 0;
398 if (size > (read ? DBGP_MAX_PACKET:0))
399 return -1;
400
401 /* Compute the control message */
402 req.bRequestType = requesttype;
403 req.bRequest = request;
404 req.wValue = cpu_to_le16(value);
405 req.wIndex = cpu_to_le16(index);
406 req.wLength = cpu_to_le16(size);
407
408 pids = DBGP_PID_SET(USB_PID_DATA0, USB_PID_SETUP);
409 addr = DBGP_EPADDR(devnum, 0);
410
411 ctrl = readl(&ehci_debug->control);
412 ctrl = dbgp_len_update(ctrl, sizeof(req));
413 ctrl |= DBGP_OUT;
414 ctrl |= DBGP_GO;
415
416 /* Send the setup message */
417 dbgp_set_data(&req, sizeof(req));
418 writel(addr, &ehci_debug->address);
419 writel(pids, &ehci_debug->pids);
420 ret = dbgp_wait_until_done(ctrl);
421 if (ret < 0)
422 return ret;
423
424 /* Read the result */
425 return dbgp_bulk_read(devnum, 0, data, size);
426}
427
428
429/* Find a PCI capability */
430static u32 __init find_cap(u32 num, u32 slot, u32 func, int cap)
431{
432 u8 pos;
433 int bytes;
434
435 if (!(read_pci_config_16(num, slot, func, PCI_STATUS) &
436 PCI_STATUS_CAP_LIST))
437 return 0;
438
439 pos = read_pci_config_byte(num, slot, func, PCI_CAPABILITY_LIST);
440 for (bytes = 0; bytes < 48 && pos >= 0x40; bytes++) {
441 u8 id;
442
443 pos &= ~3;
444 id = read_pci_config_byte(num, slot, func, pos+PCI_CAP_LIST_ID);
445 if (id == 0xff)
446 break;
447 if (id == cap)
448 return pos;
449
450 pos = read_pci_config_byte(num, slot, func,
451 pos+PCI_CAP_LIST_NEXT);
452 }
453 return 0;
454}
455
456static u32 __init __find_dbgp(u32 bus, u32 slot, u32 func)
457{
458 u32 class;
459
460 class = read_pci_config(bus, slot, func, PCI_CLASS_REVISION);
461 if ((class >> 8) != PCI_CLASS_SERIAL_USB_EHCI)
462 return 0;
463
464 return find_cap(bus, slot, func, PCI_CAP_ID_EHCI_DEBUG);
465}
466
467static u32 __init find_dbgp(int ehci_num, u32 *rbus, u32 *rslot, u32 *rfunc)
468{
469 u32 bus, slot, func;
470
471 for (bus = 0; bus < 256; bus++) {
472 for (slot = 0; slot < 32; slot++) {
473 for (func = 0; func < 8; func++) {
474 unsigned cap;
475
476 cap = __find_dbgp(bus, slot, func);
477
478 if (!cap)
479 continue;
480 if (ehci_num-- != 0)
481 continue;
482 *rbus = bus;
483 *rslot = slot;
484 *rfunc = func;
485 return cap;
486 }
487 }
488 }
489 return 0;
490}
491
492static int ehci_reset_port(int port)
493{
494 u32 portsc;
495 u32 delay_time, delay;
496 int loop;
497
498 /* Reset the usb debug port */
499 portsc = readl(&ehci_regs->port_status[port - 1]);
500 portsc &= ~PORT_PE;
501 portsc |= PORT_RESET;
502 writel(portsc, &ehci_regs->port_status[port - 1]);
503
504 delay = HUB_ROOT_RESET_TIME;
505 for (delay_time = 0; delay_time < HUB_RESET_TIMEOUT;
506 delay_time += delay) {
507 dbgp_mdelay(delay);
508
509 portsc = readl(&ehci_regs->port_status[port - 1]);
510 if (portsc & PORT_RESET) {
511 /* force reset to complete */
512 loop = 2;
513 writel(portsc & ~(PORT_RWC_BITS | PORT_RESET),
514 &ehci_regs->port_status[port - 1]);
515 do {
516 portsc = readl(&ehci_regs->port_status[port-1]);
517 } while ((portsc & PORT_RESET) && (--loop > 0));
518 }
519
520 /* Device went away? */
521 if (!(portsc & PORT_CONNECT))
522 return -ENOTCONN;
523
524 /* bomb out completely if something weird happend */
525 if ((portsc & PORT_CSC))
526 return -EINVAL;
527
528 /* If we've finished resetting, then break out of the loop */
529 if (!(portsc & PORT_RESET) && (portsc & PORT_PE))
530 return 0;
531 }
532 return -EBUSY;
533}
534
535static int ehci_wait_for_port(int port)
536{
537 u32 status;
538 int ret, reps;
539
540 for (reps = 0; reps < 3; reps++) {
541 dbgp_mdelay(100);
542 status = readl(&ehci_regs->status);
543 if (status & STS_PCD) {
544 ret = ehci_reset_port(port);
545 if (ret == 0)
546 return 0;
547 }
548 }
549 return -ENOTCONN;
550}
551
552#ifdef DBGP_DEBUG
553# define dbgp_printk early_printk
554#else
555static inline void dbgp_printk(const char *fmt, ...) { }
556#endif
557
558typedef void (*set_debug_port_t)(int port);
559
560static void default_set_debug_port(int port)
561{
562}
563
564static set_debug_port_t set_debug_port = default_set_debug_port;
565
566static void nvidia_set_debug_port(int port)
567{
568 u32 dword;
569 dword = read_pci_config(ehci_dev.bus, ehci_dev.slot, ehci_dev.func,
570 0x74);
571 dword &= ~(0x0f<<12);
572 dword |= ((port & 0x0f)<<12);
573 write_pci_config(ehci_dev.bus, ehci_dev.slot, ehci_dev.func, 0x74,
574 dword);
575 dbgp_printk("set debug port to %d\n", port);
576}
577
578static void __init detect_set_debug_port(void)
579{
580 u32 vendorid;
581
582 vendorid = read_pci_config(ehci_dev.bus, ehci_dev.slot, ehci_dev.func,
583 0x00);
584
585 if ((vendorid & 0xffff) == 0x10de) {
586 dbgp_printk("using nvidia set_debug_port\n");
587 set_debug_port = nvidia_set_debug_port;
588 }
589}
590
591static int __init ehci_setup(void)
592{
593 struct usb_debug_descriptor dbgp_desc;
594 u32 cmd, ctrl, status, portsc, hcs_params;
595 u32 debug_port, new_debug_port = 0, n_ports;
596 u32 devnum;
597 int ret, i;
598 int loop;
599 int port_map_tried;
600 int playtimes = 3;
601
602try_next_time:
603 port_map_tried = 0;
604
605try_next_port:
606
607 hcs_params = readl(&ehci_caps->hcs_params);
608 debug_port = HCS_DEBUG_PORT(hcs_params);
609 n_ports = HCS_N_PORTS(hcs_params);
610
611 dbgp_printk("debug_port: %d\n", debug_port);
612 dbgp_printk("n_ports: %d\n", n_ports);
613
614 for (i = 1; i <= n_ports; i++) {
615 portsc = readl(&ehci_regs->port_status[i-1]);
616 dbgp_printk("portstatus%d: %08x\n", i, portsc);
617 }
618
619 if (port_map_tried && (new_debug_port != debug_port)) {
620 if (--playtimes) {
621 set_debug_port(new_debug_port);
622 goto try_next_time;
623 }
624 return -1;
625 }
626
627 loop = 10;
628 /* Reset the EHCI controller */
629 cmd = readl(&ehci_regs->command);
630 cmd |= CMD_RESET;
631 writel(cmd, &ehci_regs->command);
632 do {
633 cmd = readl(&ehci_regs->command);
634 } while ((cmd & CMD_RESET) && (--loop > 0));
635
636 if (!loop) {
637 dbgp_printk("can not reset ehci\n");
638 return -1;
639 }
640 dbgp_printk("ehci reset done\n");
641
642 /* Claim ownership, but do not enable yet */
643 ctrl = readl(&ehci_debug->control);
644 ctrl |= DBGP_OWNER;
645 ctrl &= ~(DBGP_ENABLED | DBGP_INUSE);
646 writel(ctrl, &ehci_debug->control);
647
648 /* Start the ehci running */
649 cmd = readl(&ehci_regs->command);
650 cmd &= ~(CMD_LRESET | CMD_IAAD | CMD_PSE | CMD_ASE | CMD_RESET);
651 cmd |= CMD_RUN;
652 writel(cmd, &ehci_regs->command);
653
654 /* Ensure everything is routed to the EHCI */
655 writel(FLAG_CF, &ehci_regs->configured_flag);
656
657 /* Wait until the controller is no longer halted */
658 loop = 10;
659 do {
660 status = readl(&ehci_regs->status);
661 } while ((status & STS_HALT) && (--loop > 0));
662
663 if (!loop) {
664 dbgp_printk("ehci can be started\n");
665 return -1;
666 }
667 dbgp_printk("ehci started\n");
668
669 /* Wait for a device to show up in the debug port */
670 ret = ehci_wait_for_port(debug_port);
671 if (ret < 0) {
672 dbgp_printk("No device found in debug port\n");
673 goto next_debug_port;
674 }
675 dbgp_printk("ehci wait for port done\n");
676
677 /* Enable the debug port */
678 ctrl = readl(&ehci_debug->control);
679 ctrl |= DBGP_CLAIM;
680 writel(ctrl, &ehci_debug->control);
681 ctrl = readl(&ehci_debug->control);
682 if ((ctrl & DBGP_CLAIM) != DBGP_CLAIM) {
683 dbgp_printk("No device in debug port\n");
684 writel(ctrl & ~DBGP_CLAIM, &ehci_debug->control);
685 goto err;
686 }
687 dbgp_printk("debug ported enabled\n");
688
689 /* Completely transfer the debug device to the debug controller */
690 portsc = readl(&ehci_regs->port_status[debug_port - 1]);
691 portsc &= ~PORT_PE;
692 writel(portsc, &ehci_regs->port_status[debug_port - 1]);
693
694 dbgp_mdelay(100);
695
696 /* Find the debug device and make it device number 127 */
697 for (devnum = 0; devnum <= 127; devnum++) {
698 ret = dbgp_control_msg(devnum,
699 USB_DIR_IN | USB_TYPE_STANDARD | USB_RECIP_DEVICE,
700 USB_REQ_GET_DESCRIPTOR, (USB_DT_DEBUG << 8), 0,
701 &dbgp_desc, sizeof(dbgp_desc));
702 if (ret > 0)
703 break;
704 }
705 if (devnum > 127) {
706 dbgp_printk("Could not find attached debug device\n");
707 goto err;
708 }
709 if (ret < 0) {
710 dbgp_printk("Attached device is not a debug device\n");
711 goto err;
712 }
713 dbgp_endpoint_out = dbgp_desc.bDebugOutEndpoint;
714
715 /* Move the device to 127 if it isn't already there */
716 if (devnum != USB_DEBUG_DEVNUM) {
717 ret = dbgp_control_msg(devnum,
718 USB_DIR_OUT | USB_TYPE_STANDARD | USB_RECIP_DEVICE,
719 USB_REQ_SET_ADDRESS, USB_DEBUG_DEVNUM, 0, NULL, 0);
720 if (ret < 0) {
721 dbgp_printk("Could not move attached device to %d\n",
722 USB_DEBUG_DEVNUM);
723 goto err;
724 }
725 devnum = USB_DEBUG_DEVNUM;
726 dbgp_printk("debug device renamed to 127\n");
727 }
728
729 /* Enable the debug interface */
730 ret = dbgp_control_msg(USB_DEBUG_DEVNUM,
731 USB_DIR_OUT | USB_TYPE_STANDARD | USB_RECIP_DEVICE,
732 USB_REQ_SET_FEATURE, USB_DEVICE_DEBUG_MODE, 0, NULL, 0);
733 if (ret < 0) {
734 dbgp_printk(" Could not enable the debug device\n");
735 goto err;
736 }
737 dbgp_printk("debug interface enabled\n");
738
739 /* Perform a small write to get the even/odd data state in sync
740 */
741 ret = dbgp_bulk_write(USB_DEBUG_DEVNUM, dbgp_endpoint_out, " ", 1);
742 if (ret < 0) {
743 dbgp_printk("dbgp_bulk_write failed: %d\n", ret);
744 goto err;
745 }
746 dbgp_printk("small write doned\n");
747
748 return 0;
749err:
750 /* Things didn't work so remove my claim */
751 ctrl = readl(&ehci_debug->control);
752 ctrl &= ~(DBGP_CLAIM | DBGP_OUT);
753 writel(ctrl, &ehci_debug->control);
754 return -1;
755
756next_debug_port:
757 port_map_tried |= (1<<(debug_port - 1));
758 new_debug_port = ((debug_port-1+1)%n_ports) + 1;
759 if (port_map_tried != ((1<<n_ports) - 1)) {
760 set_debug_port(new_debug_port);
761 goto try_next_port;
762 }
763 if (--playtimes) {
764 set_debug_port(new_debug_port);
765 goto try_next_time;
766 }
767
768 return -1;
769}
770
771static int __init early_dbgp_init(char *s)
772{
773 u32 debug_port, bar, offset;
774 u32 bus, slot, func, cap;
775 void __iomem *ehci_bar;
776 u32 dbgp_num;
777 u32 bar_val;
778 char *e;
779 int ret;
780 u8 byte;
781
782 if (!early_pci_allowed())
783 return -1;
784
785 dbgp_num = 0;
786 if (*s)
787 dbgp_num = simple_strtoul(s, &e, 10);
788 dbgp_printk("dbgp_num: %d\n", dbgp_num);
789
790 cap = find_dbgp(dbgp_num, &bus, &slot, &func);
791 if (!cap)
792 return -1;
793
794 dbgp_printk("Found EHCI debug port on %02x:%02x.%1x\n", bus, slot,
795 func);
796
797 debug_port = read_pci_config(bus, slot, func, cap);
798 bar = (debug_port >> 29) & 0x7;
799 bar = (bar * 4) + 0xc;
800 offset = (debug_port >> 16) & 0xfff;
801 dbgp_printk("bar: %02x offset: %03x\n", bar, offset);
802 if (bar != PCI_BASE_ADDRESS_0) {
803 dbgp_printk("only debug ports on bar 1 handled.\n");
804
805 return -1;
806 }
807
808 bar_val = read_pci_config(bus, slot, func, PCI_BASE_ADDRESS_0);
809 dbgp_printk("bar_val: %02x offset: %03x\n", bar_val, offset);
810 if (bar_val & ~PCI_BASE_ADDRESS_MEM_MASK) {
811 dbgp_printk("only simple 32bit mmio bars supported\n");
812
813 return -1;
814 }
815
816 /* double check if the mem space is enabled */
817 byte = read_pci_config_byte(bus, slot, func, 0x04);
818 if (!(byte & 0x2)) {
819 byte |= 0x02;
820 write_pci_config_byte(bus, slot, func, 0x04, byte);
821 dbgp_printk("mmio for ehci enabled\n");
822 }
823
824 /*
825 * FIXME I don't have the bar size so just guess PAGE_SIZE is more
826 * than enough. 1K is the biggest I have seen.
827 */
828 set_fixmap_nocache(FIX_DBGP_BASE, bar_val & PAGE_MASK);
829 ehci_bar = (void __iomem *)__fix_to_virt(FIX_DBGP_BASE);
830 ehci_bar += bar_val & ~PAGE_MASK;
831 dbgp_printk("ehci_bar: %p\n", ehci_bar);
832
833 ehci_caps = ehci_bar;
834 ehci_regs = ehci_bar + HC_LENGTH(readl(&ehci_caps->hc_capbase));
835 ehci_debug = ehci_bar + offset;
836 ehci_dev.bus = bus;
837 ehci_dev.slot = slot;
838 ehci_dev.func = func;
839
840 detect_set_debug_port();
841
842 ret = ehci_setup();
843 if (ret < 0) {
844 dbgp_printk("ehci_setup failed\n");
845 ehci_debug = NULL;
846
847 return -1;
848 }
849
850 return 0;
851}
852
853static void early_dbgp_write(struct console *con, const char *str, u32 n)
854{
855 int chunk, ret;
856
857 if (!ehci_debug)
858 return;
859 while (n > 0) {
860 chunk = n;
861 if (chunk > DBGP_MAX_PACKET)
862 chunk = DBGP_MAX_PACKET;
863 ret = dbgp_bulk_write(USB_DEBUG_DEVNUM,
864 dbgp_endpoint_out, str, chunk);
865 str += chunk;
866 n -= chunk;
867 }
868}
869
870static struct console early_dbgp_console = {
871 .name = "earlydbg",
872 .write = early_dbgp_write,
873 .flags = CON_PRINTBUFFER,
874 .index = -1,
875};
876#endif
877
154/* Console interface to a host file on AMD's SimNow! */ 878/* Console interface to a host file on AMD's SimNow! */
155 879
156static int simnow_fd; 880static int simnow_fd;
@@ -165,6 +889,7 @@ enum {
165static noinline long simnow(long cmd, long a, long b, long c) 889static noinline long simnow(long cmd, long a, long b, long c)
166{ 890{
167 long ret; 891 long ret;
892
168 asm volatile("cpuid" : 893 asm volatile("cpuid" :
169 "=a" (ret) : 894 "=a" (ret) :
170 "b" (a), "c" (b), "d" (c), "0" (MAGIC1), "D" (cmd + MAGIC2)); 895 "b" (a), "c" (b), "d" (c), "0" (MAGIC1), "D" (cmd + MAGIC2));
@@ -174,6 +899,7 @@ static noinline long simnow(long cmd, long a, long b, long c)
174static void __init simnow_init(char *str) 899static void __init simnow_init(char *str)
175{ 900{
176 char *fn = "klog"; 901 char *fn = "klog";
902
177 if (*str == '=') 903 if (*str == '=')
178 fn = ++str; 904 fn = ++str;
179 /* error ignored */ 905 /* error ignored */
@@ -194,7 +920,7 @@ static struct console simnow_console = {
194 920
195/* Direct interface for emergencies */ 921/* Direct interface for emergencies */
196static struct console *early_console = &early_vga_console; 922static struct console *early_console = &early_vga_console;
197static int early_console_initialized; 923static int __initdata early_console_initialized;
198 924
199asmlinkage void early_printk(const char *fmt, ...) 925asmlinkage void early_printk(const char *fmt, ...)
200{ 926{
@@ -208,10 +934,11 @@ asmlinkage void early_printk(const char *fmt, ...)
208 va_end(ap); 934 va_end(ap);
209} 935}
210 936
211static int __initdata keep_early;
212 937
213static int __init setup_early_printk(char *buf) 938static int __init setup_early_printk(char *buf)
214{ 939{
940 int keep_early;
941
215 if (!buf) 942 if (!buf)
216 return 0; 943 return 0;
217 944
@@ -219,8 +946,7 @@ static int __init setup_early_printk(char *buf)
219 return 0; 946 return 0;
220 early_console_initialized = 1; 947 early_console_initialized = 1;
221 948
222 if (strstr(buf, "keep")) 949 keep_early = (strstr(buf, "keep") != NULL);
223 keep_early = 1;
224 950
225 if (!strncmp(buf, "serial", 6)) { 951 if (!strncmp(buf, "serial", 6)) {
226 early_serial_init(buf + 6); 952 early_serial_init(buf + 6);
@@ -238,6 +964,17 @@ static int __init setup_early_printk(char *buf)
238 simnow_init(buf + 6); 964 simnow_init(buf + 6);
239 early_console = &simnow_console; 965 early_console = &simnow_console;
240 keep_early = 1; 966 keep_early = 1;
967#ifdef CONFIG_EARLY_PRINTK_DBGP
968 } else if (!strncmp(buf, "dbgp", 4)) {
969 if (early_dbgp_init(buf+4) < 0)
970 return 0;
971 early_console = &early_dbgp_console;
972 /*
973 * usb subsys will reset ehci controller, so don't keep
974 * that early console
975 */
976 keep_early = 0;
977#endif
241#ifdef CONFIG_HVC_XEN 978#ifdef CONFIG_HVC_XEN
242 } else if (!strncmp(buf, "xen", 3)) { 979 } else if (!strncmp(buf, "xen", 3)) {
243 early_console = &xenboot_console; 980 early_console = &xenboot_console;
@@ -251,4 +988,5 @@ static int __init setup_early_printk(char *buf)
251 register_console(early_console); 988 register_console(early_console);
252 return 0; 989 return 0;
253} 990}
991
254early_param("earlyprintk", setup_early_printk); 992early_param("earlyprintk", setup_early_printk);
diff --git a/arch/x86/kernel/efi.c b/arch/x86/kernel/efi.c
index 06cc8d4254b1..945a31cdd81f 100644
--- a/arch/x86/kernel/efi.c
+++ b/arch/x86/kernel/efi.c
@@ -414,9 +414,11 @@ void __init efi_init(void)
414 if (memmap.map == NULL) 414 if (memmap.map == NULL)
415 printk(KERN_ERR "Could not map the EFI memory map!\n"); 415 printk(KERN_ERR "Could not map the EFI memory map!\n");
416 memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size); 416 memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
417
417 if (memmap.desc_size != sizeof(efi_memory_desc_t)) 418 if (memmap.desc_size != sizeof(efi_memory_desc_t))
418 printk(KERN_WARNING "Kernel-defined memdesc" 419 printk(KERN_WARNING
419 "doesn't match the one from EFI!\n"); 420 "Kernel-defined memdesc doesn't match the one from EFI!\n");
421
420 if (add_efi_memmap) 422 if (add_efi_memmap)
421 do_add_efi_memmap(); 423 do_add_efi_memmap();
422 424
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index 109792bc7cfa..b21fbfaffe39 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -730,6 +730,7 @@ error_code:
730 movl $(__USER_DS), %ecx 730 movl $(__USER_DS), %ecx
731 movl %ecx, %ds 731 movl %ecx, %ds
732 movl %ecx, %es 732 movl %ecx, %es
733 TRACE_IRQS_OFF
733 movl %esp,%eax # pt_regs pointer 734 movl %esp,%eax # pt_regs pointer
734 call *%edi 735 call *%edi
735 jmp ret_from_exception 736 jmp ret_from_exception
@@ -760,20 +761,9 @@ ENTRY(device_not_available)
760 RING0_INT_FRAME 761 RING0_INT_FRAME
761 pushl $-1 # mark this as an int 762 pushl $-1 # mark this as an int
762 CFI_ADJUST_CFA_OFFSET 4 763 CFI_ADJUST_CFA_OFFSET 4
763 SAVE_ALL 764 pushl $do_device_not_available
764 GET_CR0_INTO_EAX
765 testl $0x4, %eax # EM (math emulation bit)
766 jne device_not_available_emulate
767 preempt_stop(CLBR_ANY)
768 call math_state_restore
769 jmp ret_from_exception
770device_not_available_emulate:
771 pushl $0 # temporary storage for ORIG_EIP
772 CFI_ADJUST_CFA_OFFSET 4 765 CFI_ADJUST_CFA_OFFSET 4
773 call math_emulate 766 jmp error_code
774 addl $4, %esp
775 CFI_ADJUST_CFA_OFFSET -4
776 jmp ret_from_exception
777 CFI_ENDPROC 767 CFI_ENDPROC
778END(device_not_available) 768END(device_not_available)
779 769
@@ -814,6 +804,7 @@ debug_stack_correct:
814 pushl $-1 # mark this as an int 804 pushl $-1 # mark this as an int
815 CFI_ADJUST_CFA_OFFSET 4 805 CFI_ADJUST_CFA_OFFSET 4
816 SAVE_ALL 806 SAVE_ALL
807 TRACE_IRQS_OFF
817 xorl %edx,%edx # error code 0 808 xorl %edx,%edx # error code 0
818 movl %esp,%eax # pt_regs pointer 809 movl %esp,%eax # pt_regs pointer
819 call do_debug 810 call do_debug
@@ -858,6 +849,7 @@ nmi_stack_correct:
858 pushl %eax 849 pushl %eax
859 CFI_ADJUST_CFA_OFFSET 4 850 CFI_ADJUST_CFA_OFFSET 4
860 SAVE_ALL 851 SAVE_ALL
852 TRACE_IRQS_OFF
861 xorl %edx,%edx # zero error code 853 xorl %edx,%edx # zero error code
862 movl %esp,%eax # pt_regs pointer 854 movl %esp,%eax # pt_regs pointer
863 call do_nmi 855 call do_nmi
@@ -898,6 +890,7 @@ nmi_espfix_stack:
898 pushl %eax 890 pushl %eax
899 CFI_ADJUST_CFA_OFFSET 4 891 CFI_ADJUST_CFA_OFFSET 4
900 SAVE_ALL 892 SAVE_ALL
893 TRACE_IRQS_OFF
901 FIXUP_ESPFIX_STACK # %eax == %esp 894 FIXUP_ESPFIX_STACK # %eax == %esp
902 xorl %edx,%edx # zero error code 895 xorl %edx,%edx # zero error code
903 call do_nmi 896 call do_nmi
@@ -928,6 +921,7 @@ KPROBE_ENTRY(int3)
928 pushl $-1 # mark this as an int 921 pushl $-1 # mark this as an int
929 CFI_ADJUST_CFA_OFFSET 4 922 CFI_ADJUST_CFA_OFFSET 4
930 SAVE_ALL 923 SAVE_ALL
924 TRACE_IRQS_OFF
931 xorl %edx,%edx # zero error code 925 xorl %edx,%edx # zero error code
932 movl %esp,%eax # pt_regs pointer 926 movl %esp,%eax # pt_regs pointer
933 call do_int3 927 call do_int3
@@ -1030,7 +1024,7 @@ ENTRY(machine_check)
1030 RING0_INT_FRAME 1024 RING0_INT_FRAME
1031 pushl $0 1025 pushl $0
1032 CFI_ADJUST_CFA_OFFSET 4 1026 CFI_ADJUST_CFA_OFFSET 4
1033 pushl machine_check_vector 1027 pushl $do_machine_check
1034 CFI_ADJUST_CFA_OFFSET 4 1028 CFI_ADJUST_CFA_OFFSET 4
1035 jmp error_code 1029 jmp error_code
1036 CFI_ENDPROC 1030 CFI_ENDPROC
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 89434d439605..1db6ce4314e1 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -275,9 +275,9 @@ ENTRY(native_usergs_sysret64)
275ENTRY(ret_from_fork) 275ENTRY(ret_from_fork)
276 CFI_DEFAULT_STACK 276 CFI_DEFAULT_STACK
277 push kernel_eflags(%rip) 277 push kernel_eflags(%rip)
278 CFI_ADJUST_CFA_OFFSET 4 278 CFI_ADJUST_CFA_OFFSET 8
279 popf # reset kernel eflags 279 popf # reset kernel eflags
280 CFI_ADJUST_CFA_OFFSET -4 280 CFI_ADJUST_CFA_OFFSET -8
281 call schedule_tail 281 call schedule_tail
282 GET_THREAD_INFO(%rcx) 282 GET_THREAD_INFO(%rcx)
283 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),TI_flags(%rcx) 283 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
@@ -667,6 +667,13 @@ END(stub_rt_sigreturn)
667 SAVE_ARGS 667 SAVE_ARGS
668 leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler 668 leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler
669 pushq %rbp 669 pushq %rbp
670 /*
671 * Save rbp twice: One is for marking the stack frame, as usual, and the
672 * other, to fill pt_regs properly. This is because bx comes right
673 * before the last saved register in that structure, and not bp. If the
674 * base pointer were in the place bx is today, this would not be needed.
675 */
676 movq %rbp, -8(%rsp)
670 CFI_ADJUST_CFA_OFFSET 8 677 CFI_ADJUST_CFA_OFFSET 8
671 CFI_REL_OFFSET rbp, 0 678 CFI_REL_OFFSET rbp, 0
672 movq %rsp,%rbp 679 movq %rsp,%rbp
@@ -932,6 +939,9 @@ END(spurious_interrupt)
932 .if \ist 939 .if \ist
933 movq %gs:pda_data_offset, %rbp 940 movq %gs:pda_data_offset, %rbp
934 .endif 941 .endif
942 .if \irqtrace
943 TRACE_IRQS_OFF
944 .endif
935 movq %rsp,%rdi 945 movq %rsp,%rdi
936 movq ORIG_RAX(%rsp),%rsi 946 movq ORIG_RAX(%rsp),%rsi
937 movq $-1,ORIG_RAX(%rsp) 947 movq $-1,ORIG_RAX(%rsp)
@@ -1058,7 +1068,8 @@ KPROBE_ENTRY(error_entry)
1058 je error_kernelspace 1068 je error_kernelspace
1059error_swapgs: 1069error_swapgs:
1060 SWAPGS 1070 SWAPGS
1061error_sti: 1071error_sti:
1072 TRACE_IRQS_OFF
1062 movq %rdi,RDI(%rsp) 1073 movq %rdi,RDI(%rsp)
1063 CFI_REL_OFFSET rdi,RDI 1074 CFI_REL_OFFSET rdi,RDI
1064 movq %rsp,%rdi 1075 movq %rsp,%rdi
@@ -1232,7 +1243,7 @@ ENTRY(simd_coprocessor_error)
1232END(simd_coprocessor_error) 1243END(simd_coprocessor_error)
1233 1244
1234ENTRY(device_not_available) 1245ENTRY(device_not_available)
1235 zeroentry math_state_restore 1246 zeroentry do_device_not_available
1236END(device_not_available) 1247END(device_not_available)
1237 1248
1238 /* runs on exception stack */ 1249 /* runs on exception stack */
diff --git a/arch/x86/mach-es7000/es7000plat.c b/arch/x86/kernel/es7000_32.c
index 50189af14b85..f454c78fcef6 100644
--- a/arch/x86/mach-es7000/es7000plat.c
+++ b/arch/x86/kernel/es7000_32.c
@@ -39,10 +39,94 @@
39#include <asm/nmi.h> 39#include <asm/nmi.h>
40#include <asm/smp.h> 40#include <asm/smp.h>
41#include <asm/apicdef.h> 41#include <asm/apicdef.h>
42#include "es7000.h"
43#include <mach_mpparse.h> 42#include <mach_mpparse.h>
44 43
45/* 44/*
45 * ES7000 chipsets
46 */
47
48#define NON_UNISYS 0
49#define ES7000_CLASSIC 1
50#define ES7000_ZORRO 2
51
52
53#define MIP_REG 1
54#define MIP_PSAI_REG 4
55
56#define MIP_BUSY 1
57#define MIP_SPIN 0xf0000
58#define MIP_VALID 0x0100000000000000ULL
59#define MIP_PORT(VALUE) ((VALUE >> 32) & 0xffff)
60
61#define MIP_RD_LO(VALUE) (VALUE & 0xffffffff)
62
63struct mip_reg_info {
64 unsigned long long mip_info;
65 unsigned long long delivery_info;
66 unsigned long long host_reg;
67 unsigned long long mip_reg;
68};
69
70struct part_info {
71 unsigned char type;
72 unsigned char length;
73 unsigned char part_id;
74 unsigned char apic_mode;
75 unsigned long snum;
76 char ptype[16];
77 char sname[64];
78 char pname[64];
79};
80
81struct psai {
82 unsigned long long entry_type;
83 unsigned long long addr;
84 unsigned long long bep_addr;
85};
86
87struct es7000_mem_info {
88 unsigned char type;
89 unsigned char length;
90 unsigned char resv[6];
91 unsigned long long start;
92 unsigned long long size;
93};
94
95struct es7000_oem_table {
96 unsigned long long hdr;
97 struct mip_reg_info mip;
98 struct part_info pif;
99 struct es7000_mem_info shm;
100 struct psai psai;
101};
102
103#ifdef CONFIG_ACPI
104
105struct oem_table {
106 struct acpi_table_header Header;
107 u32 OEMTableAddr;
108 u32 OEMTableSize;
109};
110
111extern int find_unisys_acpi_oem_table(unsigned long *oem_addr);
112extern void unmap_unisys_acpi_oem_table(unsigned long oem_addr);
113#endif
114
115struct mip_reg {
116 unsigned long long off_0;
117 unsigned long long off_8;
118 unsigned long long off_10;
119 unsigned long long off_18;
120 unsigned long long off_20;
121 unsigned long long off_28;
122 unsigned long long off_30;
123 unsigned long long off_38;
124};
125
126#define MIP_SW_APIC 0x1020b
127#define MIP_FUNC(VALUE) (VALUE & 0xff)
128
129/*
46 * ES7000 Globals 130 * ES7000 Globals
47 */ 131 */
48 132
@@ -72,7 +156,7 @@ es7000_rename_gsi(int ioapic, int gsi)
72 base += nr_ioapic_registers[i]; 156 base += nr_ioapic_registers[i];
73 } 157 }
74 158
75 if (!ioapic && (gsi < 16)) 159 if (!ioapic && (gsi < 16))
76 gsi += base; 160 gsi += base;
77 return gsi; 161 return gsi;
78} 162}
@@ -160,21 +244,38 @@ parse_unisys_oem (char *oemptr)
160} 244}
161 245
162#ifdef CONFIG_ACPI 246#ifdef CONFIG_ACPI
163int __init 247static unsigned long oem_addrX;
164find_unisys_acpi_oem_table(unsigned long *oem_addr) 248static unsigned long oem_size;
249int __init find_unisys_acpi_oem_table(unsigned long *oem_addr)
165{ 250{
166 struct acpi_table_header *header = NULL; 251 struct acpi_table_header *header = NULL;
167 int i = 0; 252 int i = 0;
168 while (ACPI_SUCCESS(acpi_get_table("OEM1", i++, &header))) { 253 acpi_size tbl_size;
254
255 while (ACPI_SUCCESS(acpi_get_table_with_size("OEM1", i++, &header, &tbl_size))) {
169 if (!memcmp((char *) &header->oem_id, "UNISYS", 6)) { 256 if (!memcmp((char *) &header->oem_id, "UNISYS", 6)) {
170 struct oem_table *t = (struct oem_table *)header; 257 struct oem_table *t = (struct oem_table *)header;
171 *oem_addr = (unsigned long)__acpi_map_table(t->OEMTableAddr, 258
172 t->OEMTableSize); 259 oem_addrX = t->OEMTableAddr;
260 oem_size = t->OEMTableSize;
261 early_acpi_os_unmap_memory(header, tbl_size);
262
263 *oem_addr = (unsigned long)__acpi_map_table(oem_addrX,
264 oem_size);
173 return 0; 265 return 0;
174 } 266 }
267 early_acpi_os_unmap_memory(header, tbl_size);
175 } 268 }
176 return -1; 269 return -1;
177} 270}
271
272void __init unmap_unisys_acpi_oem_table(unsigned long oem_addr)
273{
274 if (!oem_addr)
275 return;
276
277 __acpi_unmap_table((char *)oem_addr, oem_size);
278}
178#endif 279#endif
179 280
180static void 281static void
diff --git a/arch/x86/kernel/genapic_64.c b/arch/x86/kernel/genapic_64.c
index eaff0bbb1444..6c9bfc9e1e95 100644
--- a/arch/x86/kernel/genapic_64.c
+++ b/arch/x86/kernel/genapic_64.c
@@ -16,87 +16,63 @@
16#include <linux/ctype.h> 16#include <linux/ctype.h>
17#include <linux/init.h> 17#include <linux/init.h>
18#include <linux/hardirq.h> 18#include <linux/hardirq.h>
19#include <linux/dmar.h>
19 20
20#include <asm/smp.h> 21#include <asm/smp.h>
21#include <asm/ipi.h> 22#include <asm/ipi.h>
22#include <asm/genapic.h> 23#include <asm/genapic.h>
23 24
24#ifdef CONFIG_ACPI 25extern struct genapic apic_flat;
25#include <acpi/acpi_bus.h> 26extern struct genapic apic_physflat;
26#endif 27extern struct genapic apic_x2xpic_uv_x;
27 28extern struct genapic apic_x2apic_phys;
28DEFINE_PER_CPU(int, x2apic_extra_bits); 29extern struct genapic apic_x2apic_cluster;
29 30
30struct genapic __read_mostly *genapic = &apic_flat; 31struct genapic __read_mostly *genapic = &apic_flat;
31 32
32static enum uv_system_type uv_system_type; 33static struct genapic *apic_probe[] __initdata = {
34 &apic_x2apic_uv_x,
35 &apic_x2apic_phys,
36 &apic_x2apic_cluster,
37 &apic_physflat,
38 NULL,
39};
33 40
34/* 41/*
35 * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode. 42 * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode.
36 */ 43 */
37void __init setup_apic_routing(void) 44void __init setup_apic_routing(void)
38{ 45{
39 if (uv_system_type == UV_NON_UNIQUE_APIC) 46 if (genapic == &apic_x2apic_phys || genapic == &apic_x2apic_cluster) {
40 genapic = &apic_x2apic_uv_x; 47 if (!intr_remapping_enabled)
41 else 48 genapic = &apic_flat;
42#ifdef CONFIG_ACPI 49 }
43 /*
44 * Quirk: some x86_64 machines can only use physical APIC mode
45 * regardless of how many processors are present (x86_64 ES7000
46 * is an example).
47 */
48 if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID &&
49 (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL))
50 genapic = &apic_physflat;
51 else
52#endif
53
54 if (max_physical_apicid < 8)
55 genapic = &apic_flat;
56 else
57 genapic = &apic_physflat;
58 50
59 printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name); 51 if (genapic == &apic_flat) {
52 if (max_physical_apicid >= 8)
53 genapic = &apic_physflat;
54 printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name);
55 }
60} 56}
61 57
62/* Same for both flat and physical. */ 58/* Same for both flat and physical. */
63 59
64void send_IPI_self(int vector) 60void apic_send_IPI_self(int vector)
65{ 61{
66 __send_IPI_shortcut(APIC_DEST_SELF, vector, APIC_DEST_PHYSICAL); 62 __send_IPI_shortcut(APIC_DEST_SELF, vector, APIC_DEST_PHYSICAL);
67} 63}
68 64
69int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id) 65int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id)
70{ 66{
71 if (!strcmp(oem_id, "SGI")) { 67 int i;
72 if (!strcmp(oem_table_id, "UVL")) 68
73 uv_system_type = UV_LEGACY_APIC; 69 for (i = 0; apic_probe[i]; ++i) {
74 else if (!strcmp(oem_table_id, "UVX")) 70 if (apic_probe[i]->acpi_madt_oem_check(oem_id, oem_table_id)) {
75 uv_system_type = UV_X2APIC; 71 genapic = apic_probe[i];
76 else if (!strcmp(oem_table_id, "UVH")) 72 printk(KERN_INFO "Setting APIC routing to %s.\n",
77 uv_system_type = UV_NON_UNIQUE_APIC; 73 genapic->name);
74 return 1;
75 }
78 } 76 }
79 return 0; 77 return 0;
80} 78}
81
82unsigned int read_apic_id(void)
83{
84 unsigned int id;
85
86 WARN_ON(preemptible() && num_online_cpus() > 1);
87 id = apic_read(APIC_ID);
88 if (uv_system_type >= UV_X2APIC)
89 id |= __get_cpu_var(x2apic_extra_bits);
90 return id;
91}
92
93enum uv_system_type get_uv_system_type(void)
94{
95 return uv_system_type;
96}
97
98int is_uv_system(void)
99{
100 return uv_system_type != UV_NONE;
101}
102EXPORT_SYMBOL_GPL(is_uv_system);
diff --git a/arch/x86/kernel/genapic_flat_64.c b/arch/x86/kernel/genapic_flat_64.c
index 786548a62d38..9eca5ba7a6b1 100644
--- a/arch/x86/kernel/genapic_flat_64.c
+++ b/arch/x86/kernel/genapic_flat_64.c
@@ -15,9 +15,20 @@
15#include <linux/kernel.h> 15#include <linux/kernel.h>
16#include <linux/ctype.h> 16#include <linux/ctype.h>
17#include <linux/init.h> 17#include <linux/init.h>
18#include <linux/hardirq.h>
18#include <asm/smp.h> 19#include <asm/smp.h>
19#include <asm/ipi.h> 20#include <asm/ipi.h>
20#include <asm/genapic.h> 21#include <asm/genapic.h>
22#include <mach_apicdef.h>
23
24#ifdef CONFIG_ACPI
25#include <acpi/acpi_bus.h>
26#endif
27
28static int __init flat_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
29{
30 return 1;
31}
21 32
22static cpumask_t flat_target_cpus(void) 33static cpumask_t flat_target_cpus(void)
23{ 34{
@@ -95,9 +106,33 @@ static void flat_send_IPI_all(int vector)
95 __send_IPI_shortcut(APIC_DEST_ALLINC, vector, APIC_DEST_LOGICAL); 106 __send_IPI_shortcut(APIC_DEST_ALLINC, vector, APIC_DEST_LOGICAL);
96} 107}
97 108
109static unsigned int get_apic_id(unsigned long x)
110{
111 unsigned int id;
112
113 id = (((x)>>24) & 0xFFu);
114 return id;
115}
116
117static unsigned long set_apic_id(unsigned int id)
118{
119 unsigned long x;
120
121 x = ((id & 0xFFu)<<24);
122 return x;
123}
124
125static unsigned int read_xapic_id(void)
126{
127 unsigned int id;
128
129 id = get_apic_id(apic_read(APIC_ID));
130 return id;
131}
132
98static int flat_apic_id_registered(void) 133static int flat_apic_id_registered(void)
99{ 134{
100 return physid_isset(GET_APIC_ID(read_apic_id()), phys_cpu_present_map); 135 return physid_isset(read_xapic_id(), phys_cpu_present_map);
101} 136}
102 137
103static unsigned int flat_cpu_mask_to_apicid(cpumask_t cpumask) 138static unsigned int flat_cpu_mask_to_apicid(cpumask_t cpumask)
@@ -112,6 +147,7 @@ static unsigned int phys_pkg_id(int index_msb)
112 147
113struct genapic apic_flat = { 148struct genapic apic_flat = {
114 .name = "flat", 149 .name = "flat",
150 .acpi_madt_oem_check = flat_acpi_madt_oem_check,
115 .int_delivery_mode = dest_LowestPrio, 151 .int_delivery_mode = dest_LowestPrio,
116 .int_dest_mode = (APIC_DEST_LOGICAL != 0), 152 .int_dest_mode = (APIC_DEST_LOGICAL != 0),
117 .target_cpus = flat_target_cpus, 153 .target_cpus = flat_target_cpus,
@@ -121,8 +157,12 @@ struct genapic apic_flat = {
121 .send_IPI_all = flat_send_IPI_all, 157 .send_IPI_all = flat_send_IPI_all,
122 .send_IPI_allbutself = flat_send_IPI_allbutself, 158 .send_IPI_allbutself = flat_send_IPI_allbutself,
123 .send_IPI_mask = flat_send_IPI_mask, 159 .send_IPI_mask = flat_send_IPI_mask,
160 .send_IPI_self = apic_send_IPI_self,
124 .cpu_mask_to_apicid = flat_cpu_mask_to_apicid, 161 .cpu_mask_to_apicid = flat_cpu_mask_to_apicid,
125 .phys_pkg_id = phys_pkg_id, 162 .phys_pkg_id = phys_pkg_id,
163 .get_apic_id = get_apic_id,
164 .set_apic_id = set_apic_id,
165 .apic_id_mask = (0xFFu<<24),
126}; 166};
127 167
128/* 168/*
@@ -130,6 +170,21 @@ struct genapic apic_flat = {
130 * We cannot use logical delivery in this case because the mask 170 * We cannot use logical delivery in this case because the mask
131 * overflows, so use physical mode. 171 * overflows, so use physical mode.
132 */ 172 */
173static int __init physflat_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
174{
175#ifdef CONFIG_ACPI
176 /*
177 * Quirk: some x86_64 machines can only use physical APIC mode
178 * regardless of how many processors are present (x86_64 ES7000
179 * is an example).
180 */
181 if (acpi_gbl_FADT.header.revision > FADT2_REVISION_ID &&
182 (acpi_gbl_FADT.flags & ACPI_FADT_APIC_PHYSICAL))
183 return 1;
184#endif
185
186 return 0;
187}
133 188
134static cpumask_t physflat_target_cpus(void) 189static cpumask_t physflat_target_cpus(void)
135{ 190{
@@ -176,6 +231,7 @@ static unsigned int physflat_cpu_mask_to_apicid(cpumask_t cpumask)
176 231
177struct genapic apic_physflat = { 232struct genapic apic_physflat = {
178 .name = "physical flat", 233 .name = "physical flat",
234 .acpi_madt_oem_check = physflat_acpi_madt_oem_check,
179 .int_delivery_mode = dest_Fixed, 235 .int_delivery_mode = dest_Fixed,
180 .int_dest_mode = (APIC_DEST_PHYSICAL != 0), 236 .int_dest_mode = (APIC_DEST_PHYSICAL != 0),
181 .target_cpus = physflat_target_cpus, 237 .target_cpus = physflat_target_cpus,
@@ -185,6 +241,10 @@ struct genapic apic_physflat = {
185 .send_IPI_all = physflat_send_IPI_all, 241 .send_IPI_all = physflat_send_IPI_all,
186 .send_IPI_allbutself = physflat_send_IPI_allbutself, 242 .send_IPI_allbutself = physflat_send_IPI_allbutself,
187 .send_IPI_mask = physflat_send_IPI_mask, 243 .send_IPI_mask = physflat_send_IPI_mask,
244 .send_IPI_self = apic_send_IPI_self,
188 .cpu_mask_to_apicid = physflat_cpu_mask_to_apicid, 245 .cpu_mask_to_apicid = physflat_cpu_mask_to_apicid,
189 .phys_pkg_id = phys_pkg_id, 246 .phys_pkg_id = phys_pkg_id,
247 .get_apic_id = get_apic_id,
248 .set_apic_id = set_apic_id,
249 .apic_id_mask = (0xFFu<<24),
190}; 250};
diff --git a/arch/x86/kernel/genx2apic_cluster.c b/arch/x86/kernel/genx2apic_cluster.c
new file mode 100644
index 000000000000..e4bf2cc0d743
--- /dev/null
+++ b/arch/x86/kernel/genx2apic_cluster.c
@@ -0,0 +1,159 @@
1#include <linux/threads.h>
2#include <linux/cpumask.h>
3#include <linux/string.h>
4#include <linux/kernel.h>
5#include <linux/ctype.h>
6#include <linux/init.h>
7#include <linux/dmar.h>
8
9#include <asm/smp.h>
10#include <asm/ipi.h>
11#include <asm/genapic.h>
12
13DEFINE_PER_CPU(u32, x86_cpu_to_logical_apicid);
14
15static int __init x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
16{
17 if (cpu_has_x2apic)
18 return 1;
19
20 return 0;
21}
22
23/* Start with all IRQs pointing to boot CPU. IRQ balancing will shift them. */
24
25static cpumask_t x2apic_target_cpus(void)
26{
27 return cpumask_of_cpu(0);
28}
29
30/*
31 * for now each logical cpu is in its own vector allocation domain.
32 */
33static cpumask_t x2apic_vector_allocation_domain(int cpu)
34{
35 cpumask_t domain = CPU_MASK_NONE;
36 cpu_set(cpu, domain);
37 return domain;
38}
39
40static void __x2apic_send_IPI_dest(unsigned int apicid, int vector,
41 unsigned int dest)
42{
43 unsigned long cfg;
44
45 cfg = __prepare_ICR(0, vector, dest);
46
47 /*
48 * send the IPI.
49 */
50 x2apic_icr_write(cfg, apicid);
51}
52
53/*
54 * for now, we send the IPI's one by one in the cpumask.
55 * TBD: Based on the cpu mask, we can send the IPI's to the cluster group
56 * at once. We have 16 cpu's in a cluster. This will minimize IPI register
57 * writes.
58 */
59static void x2apic_send_IPI_mask(cpumask_t mask, int vector)
60{
61 unsigned long flags;
62 unsigned long query_cpu;
63
64 local_irq_save(flags);
65 for_each_cpu_mask(query_cpu, mask) {
66 __x2apic_send_IPI_dest(per_cpu(x86_cpu_to_logical_apicid, query_cpu),
67 vector, APIC_DEST_LOGICAL);
68 }
69 local_irq_restore(flags);
70}
71
72static void x2apic_send_IPI_allbutself(int vector)
73{
74 cpumask_t mask = cpu_online_map;
75
76 cpu_clear(smp_processor_id(), mask);
77
78 if (!cpus_empty(mask))
79 x2apic_send_IPI_mask(mask, vector);
80}
81
82static void x2apic_send_IPI_all(int vector)
83{
84 x2apic_send_IPI_mask(cpu_online_map, vector);
85}
86
87static int x2apic_apic_id_registered(void)
88{
89 return 1;
90}
91
92static unsigned int x2apic_cpu_mask_to_apicid(cpumask_t cpumask)
93{
94 int cpu;
95
96 /*
97 * We're using fixed IRQ delivery, can only return one phys APIC ID.
98 * May as well be the first.
99 */
100 cpu = first_cpu(cpumask);
101 if ((unsigned)cpu < NR_CPUS)
102 return per_cpu(x86_cpu_to_logical_apicid, cpu);
103 else
104 return BAD_APICID;
105}
106
107static unsigned int get_apic_id(unsigned long x)
108{
109 unsigned int id;
110
111 id = x;
112 return id;
113}
114
115static unsigned long set_apic_id(unsigned int id)
116{
117 unsigned long x;
118
119 x = id;
120 return x;
121}
122
123static unsigned int phys_pkg_id(int index_msb)
124{
125 return current_cpu_data.initial_apicid >> index_msb;
126}
127
128static void x2apic_send_IPI_self(int vector)
129{
130 apic_write(APIC_SELF_IPI, vector);
131}
132
133static void init_x2apic_ldr(void)
134{
135 int cpu = smp_processor_id();
136
137 per_cpu(x86_cpu_to_logical_apicid, cpu) = apic_read(APIC_LDR);
138 return;
139}
140
141struct genapic apic_x2apic_cluster = {
142 .name = "cluster x2apic",
143 .acpi_madt_oem_check = x2apic_acpi_madt_oem_check,
144 .int_delivery_mode = dest_LowestPrio,
145 .int_dest_mode = (APIC_DEST_LOGICAL != 0),
146 .target_cpus = x2apic_target_cpus,
147 .vector_allocation_domain = x2apic_vector_allocation_domain,
148 .apic_id_registered = x2apic_apic_id_registered,
149 .init_apic_ldr = init_x2apic_ldr,
150 .send_IPI_all = x2apic_send_IPI_all,
151 .send_IPI_allbutself = x2apic_send_IPI_allbutself,
152 .send_IPI_mask = x2apic_send_IPI_mask,
153 .send_IPI_self = x2apic_send_IPI_self,
154 .cpu_mask_to_apicid = x2apic_cpu_mask_to_apicid,
155 .phys_pkg_id = phys_pkg_id,
156 .get_apic_id = get_apic_id,
157 .set_apic_id = set_apic_id,
158 .apic_id_mask = (0xFFFFFFFFu),
159};
diff --git a/arch/x86/kernel/genx2apic_phys.c b/arch/x86/kernel/genx2apic_phys.c
new file mode 100644
index 000000000000..8f1343df2627
--- /dev/null
+++ b/arch/x86/kernel/genx2apic_phys.c
@@ -0,0 +1,154 @@
1#include <linux/threads.h>
2#include <linux/cpumask.h>
3#include <linux/string.h>
4#include <linux/kernel.h>
5#include <linux/ctype.h>
6#include <linux/init.h>
7#include <linux/dmar.h>
8
9#include <asm/smp.h>
10#include <asm/ipi.h>
11#include <asm/genapic.h>
12
13static int x2apic_phys;
14
15static int set_x2apic_phys_mode(char *arg)
16{
17 x2apic_phys = 1;
18 return 0;
19}
20early_param("x2apic_phys", set_x2apic_phys_mode);
21
22static int __init x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
23{
24 if (cpu_has_x2apic && x2apic_phys)
25 return 1;
26
27 return 0;
28}
29
30/* Start with all IRQs pointing to boot CPU. IRQ balancing will shift them. */
31
32static cpumask_t x2apic_target_cpus(void)
33{
34 return cpumask_of_cpu(0);
35}
36
37static cpumask_t x2apic_vector_allocation_domain(int cpu)
38{
39 cpumask_t domain = CPU_MASK_NONE;
40 cpu_set(cpu, domain);
41 return domain;
42}
43
44static void __x2apic_send_IPI_dest(unsigned int apicid, int vector,
45 unsigned int dest)
46{
47 unsigned long cfg;
48
49 cfg = __prepare_ICR(0, vector, dest);
50
51 /*
52 * send the IPI.
53 */
54 x2apic_icr_write(cfg, apicid);
55}
56
57static void x2apic_send_IPI_mask(cpumask_t mask, int vector)
58{
59 unsigned long flags;
60 unsigned long query_cpu;
61
62 local_irq_save(flags);
63 for_each_cpu_mask(query_cpu, mask) {
64 __x2apic_send_IPI_dest(per_cpu(x86_cpu_to_apicid, query_cpu),
65 vector, APIC_DEST_PHYSICAL);
66 }
67 local_irq_restore(flags);
68}
69
70static void x2apic_send_IPI_allbutself(int vector)
71{
72 cpumask_t mask = cpu_online_map;
73
74 cpu_clear(smp_processor_id(), mask);
75
76 if (!cpus_empty(mask))
77 x2apic_send_IPI_mask(mask, vector);
78}
79
80static void x2apic_send_IPI_all(int vector)
81{
82 x2apic_send_IPI_mask(cpu_online_map, vector);
83}
84
85static int x2apic_apic_id_registered(void)
86{
87 return 1;
88}
89
90static unsigned int x2apic_cpu_mask_to_apicid(cpumask_t cpumask)
91{
92 int cpu;
93
94 /*
95 * We're using fixed IRQ delivery, can only return one phys APIC ID.
96 * May as well be the first.
97 */
98 cpu = first_cpu(cpumask);
99 if ((unsigned)cpu < NR_CPUS)
100 return per_cpu(x86_cpu_to_apicid, cpu);
101 else
102 return BAD_APICID;
103}
104
105static unsigned int get_apic_id(unsigned long x)
106{
107 unsigned int id;
108
109 id = x;
110 return id;
111}
112
113static unsigned long set_apic_id(unsigned int id)
114{
115 unsigned long x;
116
117 x = id;
118 return x;
119}
120
121static unsigned int phys_pkg_id(int index_msb)
122{
123 return current_cpu_data.initial_apicid >> index_msb;
124}
125
126void x2apic_send_IPI_self(int vector)
127{
128 apic_write(APIC_SELF_IPI, vector);
129}
130
131void init_x2apic_ldr(void)
132{
133 return;
134}
135
136struct genapic apic_x2apic_phys = {
137 .name = "physical x2apic",
138 .acpi_madt_oem_check = x2apic_acpi_madt_oem_check,
139 .int_delivery_mode = dest_Fixed,
140 .int_dest_mode = (APIC_DEST_PHYSICAL != 0),
141 .target_cpus = x2apic_target_cpus,
142 .vector_allocation_domain = x2apic_vector_allocation_domain,
143 .apic_id_registered = x2apic_apic_id_registered,
144 .init_apic_ldr = init_x2apic_ldr,
145 .send_IPI_all = x2apic_send_IPI_all,
146 .send_IPI_allbutself = x2apic_send_IPI_allbutself,
147 .send_IPI_mask = x2apic_send_IPI_mask,
148 .send_IPI_self = x2apic_send_IPI_self,
149 .cpu_mask_to_apicid = x2apic_cpu_mask_to_apicid,
150 .phys_pkg_id = phys_pkg_id,
151 .get_apic_id = get_apic_id,
152 .set_apic_id = set_apic_id,
153 .apic_id_mask = (0xFFFFFFFFu),
154};
diff --git a/arch/x86/kernel/genx2apic_uv_x.c b/arch/x86/kernel/genx2apic_uv_x.c
index bfa837cb16be..33581d94a90e 100644
--- a/arch/x86/kernel/genx2apic_uv_x.c
+++ b/arch/x86/kernel/genx2apic_uv_x.c
@@ -12,12 +12,12 @@
12#include <linux/threads.h> 12#include <linux/threads.h>
13#include <linux/cpumask.h> 13#include <linux/cpumask.h>
14#include <linux/string.h> 14#include <linux/string.h>
15#include <linux/kernel.h>
16#include <linux/ctype.h> 15#include <linux/ctype.h>
17#include <linux/init.h> 16#include <linux/init.h>
18#include <linux/sched.h> 17#include <linux/sched.h>
19#include <linux/bootmem.h> 18#include <linux/bootmem.h>
20#include <linux/module.h> 19#include <linux/module.h>
20#include <linux/hardirq.h>
21#include <asm/smp.h> 21#include <asm/smp.h>
22#include <asm/ipi.h> 22#include <asm/ipi.h>
23#include <asm/genapic.h> 23#include <asm/genapic.h>
@@ -26,6 +26,36 @@
26#include <asm/uv/uv_hub.h> 26#include <asm/uv/uv_hub.h>
27#include <asm/uv/bios.h> 27#include <asm/uv/bios.h>
28 28
29DEFINE_PER_CPU(int, x2apic_extra_bits);
30
31static enum uv_system_type uv_system_type;
32
33static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
34{
35 if (!strcmp(oem_id, "SGI")) {
36 if (!strcmp(oem_table_id, "UVL"))
37 uv_system_type = UV_LEGACY_APIC;
38 else if (!strcmp(oem_table_id, "UVX"))
39 uv_system_type = UV_X2APIC;
40 else if (!strcmp(oem_table_id, "UVH")) {
41 uv_system_type = UV_NON_UNIQUE_APIC;
42 return 1;
43 }
44 }
45 return 0;
46}
47
48enum uv_system_type get_uv_system_type(void)
49{
50 return uv_system_type;
51}
52
53int is_uv_system(void)
54{
55 return uv_system_type != UV_NONE;
56}
57EXPORT_SYMBOL_GPL(is_uv_system);
58
29DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info); 59DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
30EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info); 60EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
31 61
@@ -84,7 +114,7 @@ static void uv_send_IPI_one(int cpu, int vector)
84 unsigned long val, apicid, lapicid; 114 unsigned long val, apicid, lapicid;
85 int pnode; 115 int pnode;
86 116
87 apicid = per_cpu(x86_cpu_to_apicid, cpu); /* ZZZ - cache node-local ? */ 117 apicid = per_cpu(x86_cpu_to_apicid, cpu);
88 lapicid = apicid & 0x3f; /* ZZZ macro needed */ 118 lapicid = apicid & 0x3f; /* ZZZ macro needed */
89 pnode = uv_apicid_to_pnode(apicid); 119 pnode = uv_apicid_to_pnode(apicid);
90 val = 120 val =
@@ -123,6 +153,10 @@ static int uv_apic_id_registered(void)
123 return 1; 153 return 1;
124} 154}
125 155
156static void uv_init_apic_ldr(void)
157{
158}
159
126static unsigned int uv_cpu_mask_to_apicid(cpumask_t cpumask) 160static unsigned int uv_cpu_mask_to_apicid(cpumask_t cpumask)
127{ 161{
128 int cpu; 162 int cpu;
@@ -138,31 +172,59 @@ static unsigned int uv_cpu_mask_to_apicid(cpumask_t cpumask)
138 return BAD_APICID; 172 return BAD_APICID;
139} 173}
140 174
175static unsigned int get_apic_id(unsigned long x)
176{
177 unsigned int id;
178
179 WARN_ON(preemptible() && num_online_cpus() > 1);
180 id = x | __get_cpu_var(x2apic_extra_bits);
181
182 return id;
183}
184
185static unsigned long set_apic_id(unsigned int id)
186{
187 unsigned long x;
188
189 /* maskout x2apic_extra_bits ? */
190 x = id;
191 return x;
192}
193
194static unsigned int uv_read_apic_id(void)
195{
196
197 return get_apic_id(apic_read(APIC_ID));
198}
199
141static unsigned int phys_pkg_id(int index_msb) 200static unsigned int phys_pkg_id(int index_msb)
142{ 201{
143 return GET_APIC_ID(read_apic_id()) >> index_msb; 202 return uv_read_apic_id() >> index_msb;
144} 203}
145 204
146#ifdef ZZZ /* Needs x2apic patch */
147static void uv_send_IPI_self(int vector) 205static void uv_send_IPI_self(int vector)
148{ 206{
149 apic_write(APIC_SELF_IPI, vector); 207 apic_write(APIC_SELF_IPI, vector);
150} 208}
151#endif
152 209
153struct genapic apic_x2apic_uv_x = { 210struct genapic apic_x2apic_uv_x = {
154 .name = "UV large system", 211 .name = "UV large system",
212 .acpi_madt_oem_check = uv_acpi_madt_oem_check,
155 .int_delivery_mode = dest_Fixed, 213 .int_delivery_mode = dest_Fixed,
156 .int_dest_mode = (APIC_DEST_PHYSICAL != 0), 214 .int_dest_mode = (APIC_DEST_PHYSICAL != 0),
157 .target_cpus = uv_target_cpus, 215 .target_cpus = uv_target_cpus,
158 .vector_allocation_domain = uv_vector_allocation_domain,/* Fixme ZZZ */ 216 .vector_allocation_domain = uv_vector_allocation_domain,
159 .apic_id_registered = uv_apic_id_registered, 217 .apic_id_registered = uv_apic_id_registered,
218 .init_apic_ldr = uv_init_apic_ldr,
160 .send_IPI_all = uv_send_IPI_all, 219 .send_IPI_all = uv_send_IPI_all,
161 .send_IPI_allbutself = uv_send_IPI_allbutself, 220 .send_IPI_allbutself = uv_send_IPI_allbutself,
162 .send_IPI_mask = uv_send_IPI_mask, 221 .send_IPI_mask = uv_send_IPI_mask,
163 /* ZZZ.send_IPI_self = uv_send_IPI_self, */ 222 .send_IPI_self = uv_send_IPI_self,
164 .cpu_mask_to_apicid = uv_cpu_mask_to_apicid, 223 .cpu_mask_to_apicid = uv_cpu_mask_to_apicid,
165 .phys_pkg_id = phys_pkg_id, /* Fixme ZZZ */ 224 .phys_pkg_id = phys_pkg_id,
225 .get_apic_id = get_apic_id,
226 .set_apic_id = set_apic_id,
227 .apic_id_mask = (0xFFFFFFFFu),
166}; 228};
167 229
168static __cpuinit void set_x2apic_extra_bits(int pnode) 230static __cpuinit void set_x2apic_extra_bits(int pnode)
@@ -222,12 +284,13 @@ static __init void map_low_mmrs(void)
222 284
223enum map_type {map_wb, map_uc}; 285enum map_type {map_wb, map_uc};
224 286
225static __init void map_high(char *id, unsigned long base, int shift, enum map_type map_type) 287static __init void map_high(char *id, unsigned long base, int shift,
288 int max_pnode, enum map_type map_type)
226{ 289{
227 unsigned long bytes, paddr; 290 unsigned long bytes, paddr;
228 291
229 paddr = base << shift; 292 paddr = base << shift;
230 bytes = (1UL << shift); 293 bytes = (1UL << shift) * (max_pnode + 1);
231 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, 294 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
232 paddr + bytes); 295 paddr + bytes);
233 if (map_type == map_uc) 296 if (map_type == map_uc)
@@ -243,7 +306,7 @@ static __init void map_gru_high(int max_pnode)
243 306
244 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR); 307 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
245 if (gru.s.enable) 308 if (gru.s.enable)
246 map_high("GRU", gru.s.base, shift, map_wb); 309 map_high("GRU", gru.s.base, shift, max_pnode, map_wb);
247} 310}
248 311
249static __init void map_config_high(int max_pnode) 312static __init void map_config_high(int max_pnode)
@@ -253,7 +316,7 @@ static __init void map_config_high(int max_pnode)
253 316
254 cfg.v = uv_read_local_mmr(UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR); 317 cfg.v = uv_read_local_mmr(UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR);
255 if (cfg.s.enable) 318 if (cfg.s.enable)
256 map_high("CONFIG", cfg.s.base, shift, map_uc); 319 map_high("CONFIG", cfg.s.base, shift, max_pnode, map_uc);
257} 320}
258 321
259static __init void map_mmr_high(int max_pnode) 322static __init void map_mmr_high(int max_pnode)
@@ -263,7 +326,7 @@ static __init void map_mmr_high(int max_pnode)
263 326
264 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR); 327 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
265 if (mmr.s.enable) 328 if (mmr.s.enable)
266 map_high("MMR", mmr.s.base, shift, map_uc); 329 map_high("MMR", mmr.s.base, shift, max_pnode, map_uc);
267} 330}
268 331
269static __init void map_mmioh_high(int max_pnode) 332static __init void map_mmioh_high(int max_pnode)
@@ -273,7 +336,7 @@ static __init void map_mmioh_high(int max_pnode)
273 336
274 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR); 337 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
275 if (mmioh.s.enable) 338 if (mmioh.s.enable)
276 map_high("MMIOH", mmioh.s.base, shift, map_uc); 339 map_high("MMIOH", mmioh.s.base, shift, max_pnode, map_uc);
277} 340}
278 341
279static __init void uv_rtc_init(void) 342static __init void uv_rtc_init(void)
@@ -401,3 +464,5 @@ void __cpuinit uv_cpu_init(void)
401 if (get_uv_system_type() == UV_NON_UNIQUE_APIC) 464 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
402 set_x2apic_extra_bits(uv_hub_info->pnode); 465 set_x2apic_extra_bits(uv_hub_info->pnode);
403} 466}
467
468
diff --git a/arch/x86/kernel/head.c b/arch/x86/kernel/head.c
index 3e66bd364a9d..1dcb0f13897e 100644
--- a/arch/x86/kernel/head.c
+++ b/arch/x86/kernel/head.c
@@ -35,6 +35,7 @@ void __init reserve_ebda_region(void)
35 35
36 /* start of EBDA area */ 36 /* start of EBDA area */
37 ebda_addr = get_bios_ebda(); 37 ebda_addr = get_bios_ebda();
38 printk(KERN_INFO "BIOS EBDA/lowmem at: %08x/%08x\n", ebda_addr, lowmem);
38 39
39 /* Fixup: bios puts an EBDA in the top 64K segment */ 40 /* Fixup: bios puts an EBDA in the top 64K segment */
40 /* of conventional memory, but does not adjust lowmem. */ 41 /* of conventional memory, but does not adjust lowmem. */
diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index 9bfc4d72fb2e..d16084f90649 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -108,12 +108,11 @@ void __init x86_64_start_kernel(char * real_mode_data)
108 } 108 }
109 load_idt((const struct desc_ptr *)&idt_descr); 109 load_idt((const struct desc_ptr *)&idt_descr);
110 110
111 early_printk("Kernel alive\n"); 111 if (console_loglevel == 10)
112 early_printk("Kernel alive\n");
112 113
113 x86_64_init_pda(); 114 x86_64_init_pda();
114 115
115 early_printk("Kernel really alive\n");
116
117 x86_64_start_reservations(real_mode_data); 116 x86_64_start_reservations(real_mode_data);
118} 117}
119 118
diff --git a/arch/x86/kernel/head_32.S b/arch/x86/kernel/head_32.S
index a7010c3a377a..e835b4eea70b 100644
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -172,10 +172,6 @@ num_subarch_entries = (. - subarch_entries) / 4
172 * 172 *
173 * Note that the stack is not yet set up! 173 * Note that the stack is not yet set up!
174 */ 174 */
175#define PTE_ATTR 0x007 /* PRESENT+RW+USER */
176#define PDE_ATTR 0x067 /* PRESENT+RW+USER+DIRTY+ACCESSED */
177#define PGD_ATTR 0x001 /* PRESENT (no other attributes) */
178
179default_entry: 175default_entry:
180#ifdef CONFIG_X86_PAE 176#ifdef CONFIG_X86_PAE
181 177
@@ -196,9 +192,9 @@ default_entry:
196 movl $pa(pg0), %edi 192 movl $pa(pg0), %edi
197 movl %edi, pa(init_pg_tables_start) 193 movl %edi, pa(init_pg_tables_start)
198 movl $pa(swapper_pg_pmd), %edx 194 movl $pa(swapper_pg_pmd), %edx
199 movl $PTE_ATTR, %eax 195 movl $PTE_IDENT_ATTR, %eax
20010: 19610:
201 leal PDE_ATTR(%edi),%ecx /* Create PMD entry */ 197 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
202 movl %ecx,(%edx) /* Store PMD entry */ 198 movl %ecx,(%edx) /* Store PMD entry */
203 /* Upper half already zero */ 199 /* Upper half already zero */
204 addl $8,%edx 200 addl $8,%edx
@@ -215,7 +211,7 @@ default_entry:
215 * End condition: we must map up to and including INIT_MAP_BEYOND_END 211 * End condition: we must map up to and including INIT_MAP_BEYOND_END
216 * bytes beyond the end of our own page tables. 212 * bytes beyond the end of our own page tables.
217 */ 213 */
218 leal (INIT_MAP_BEYOND_END+PTE_ATTR)(%edi),%ebp 214 leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
219 cmpl %ebp,%eax 215 cmpl %ebp,%eax
220 jb 10b 216 jb 10b
2211: 2171:
@@ -224,7 +220,7 @@ default_entry:
224 movl %eax, pa(max_pfn_mapped) 220 movl %eax, pa(max_pfn_mapped)
225 221
226 /* Do early initialization of the fixmap area */ 222 /* Do early initialization of the fixmap area */
227 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax 223 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
228 movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8) 224 movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8)
229#else /* Not PAE */ 225#else /* Not PAE */
230 226
@@ -233,9 +229,9 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
233 movl $pa(pg0), %edi 229 movl $pa(pg0), %edi
234 movl %edi, pa(init_pg_tables_start) 230 movl %edi, pa(init_pg_tables_start)
235 movl $pa(swapper_pg_dir), %edx 231 movl $pa(swapper_pg_dir), %edx
236 movl $PTE_ATTR, %eax 232 movl $PTE_IDENT_ATTR, %eax
23710: 23310:
238 leal PDE_ATTR(%edi),%ecx /* Create PDE entry */ 234 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
239 movl %ecx,(%edx) /* Store identity PDE entry */ 235 movl %ecx,(%edx) /* Store identity PDE entry */
240 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */ 236 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
241 addl $4,%edx 237 addl $4,%edx
@@ -249,7 +245,7 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
249 * bytes beyond the end of our own page tables; the +0x007 is 245 * bytes beyond the end of our own page tables; the +0x007 is
250 * the attribute bits 246 * the attribute bits
251 */ 247 */
252 leal (INIT_MAP_BEYOND_END+PTE_ATTR)(%edi),%ebp 248 leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
253 cmpl %ebp,%eax 249 cmpl %ebp,%eax
254 jb 10b 250 jb 10b
255 movl %edi,pa(init_pg_tables_end) 251 movl %edi,pa(init_pg_tables_end)
@@ -257,7 +253,7 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
257 movl %eax, pa(max_pfn_mapped) 253 movl %eax, pa(max_pfn_mapped)
258 254
259 /* Do early initialization of the fixmap area */ 255 /* Do early initialization of the fixmap area */
260 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax 256 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
261 movl %eax,pa(swapper_pg_dir+0xffc) 257 movl %eax,pa(swapper_pg_dir+0xffc)
262#endif 258#endif
263 jmp 3f 259 jmp 3f
@@ -634,19 +630,19 @@ ENTRY(empty_zero_page)
634 /* Page-aligned for the benefit of paravirt? */ 630 /* Page-aligned for the benefit of paravirt? */
635 .align PAGE_SIZE_asm 631 .align PAGE_SIZE_asm
636ENTRY(swapper_pg_dir) 632ENTRY(swapper_pg_dir)
637 .long pa(swapper_pg_pmd+PGD_ATTR),0 /* low identity map */ 633 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
638# if KPMDS == 3 634# if KPMDS == 3
639 .long pa(swapper_pg_pmd+PGD_ATTR),0 635 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
640 .long pa(swapper_pg_pmd+PGD_ATTR+0x1000),0 636 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
641 .long pa(swapper_pg_pmd+PGD_ATTR+0x2000),0 637 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x2000),0
642# elif KPMDS == 2 638# elif KPMDS == 2
643 .long 0,0 639 .long 0,0
644 .long pa(swapper_pg_pmd+PGD_ATTR),0 640 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
645 .long pa(swapper_pg_pmd+PGD_ATTR+0x1000),0 641 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
646# elif KPMDS == 1 642# elif KPMDS == 1
647 .long 0,0 643 .long 0,0
648 .long 0,0 644 .long 0,0
649 .long pa(swapper_pg_pmd+PGD_ATTR),0 645 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
650# else 646# else
651# error "Kernel PMDs should be 1, 2 or 3" 647# error "Kernel PMDs should be 1, 2 or 3"
652# endif 648# endif
diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index db3280afe886..26cfdc1d7c7f 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -110,7 +110,7 @@ startup_64:
110 movq %rdi, %rax 110 movq %rdi, %rax
111 shrq $PMD_SHIFT, %rax 111 shrq $PMD_SHIFT, %rax
112 andq $(PTRS_PER_PMD - 1), %rax 112 andq $(PTRS_PER_PMD - 1), %rax
113 leaq __PAGE_KERNEL_LARGE_EXEC(%rdi), %rdx 113 leaq __PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
114 leaq level2_spare_pgt(%rip), %rbx 114 leaq level2_spare_pgt(%rip), %rbx
115 movq %rdx, 0(%rbx, %rax, 8) 115 movq %rdx, 0(%rbx, %rax, 8)
116ident_complete: 116ident_complete:
@@ -374,7 +374,7 @@ NEXT_PAGE(level2_ident_pgt)
374 /* Since I easily can, map the first 1G. 374 /* Since I easily can, map the first 1G.
375 * Don't set NX because code runs from these pages. 375 * Don't set NX because code runs from these pages.
376 */ 376 */
377 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD) 377 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
378 378
379NEXT_PAGE(level2_kernel_pgt) 379NEXT_PAGE(level2_kernel_pgt)
380 /* 380 /*
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index 59fd3b6b1303..acf62fc233da 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -115,13 +115,17 @@ static void hpet_reserve_platform_timers(unsigned long id)
115 hd.hd_phys_address = hpet_address; 115 hd.hd_phys_address = hpet_address;
116 hd.hd_address = hpet; 116 hd.hd_address = hpet;
117 hd.hd_nirqs = nrtimers; 117 hd.hd_nirqs = nrtimers;
118 hd.hd_flags = HPET_DATA_PLATFORM;
119 hpet_reserve_timer(&hd, 0); 118 hpet_reserve_timer(&hd, 0);
120 119
121#ifdef CONFIG_HPET_EMULATE_RTC 120#ifdef CONFIG_HPET_EMULATE_RTC
122 hpet_reserve_timer(&hd, 1); 121 hpet_reserve_timer(&hd, 1);
123#endif 122#endif
124 123
124 /*
125 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
126 * is wrong for i8259!) not the output IRQ. Many BIOS writers
127 * don't bother configuring *any* comparator interrupts.
128 */
125 hd.hd_irq[0] = HPET_LEGACY_8254; 129 hd.hd_irq[0] = HPET_LEGACY_8254;
126 hd.hd_irq[1] = HPET_LEGACY_RTC; 130 hd.hd_irq[1] = HPET_LEGACY_RTC;
127 131
@@ -210,8 +214,8 @@ static void hpet_legacy_clockevent_register(void)
210 /* Calculate the min / max delta */ 214 /* Calculate the min / max delta */
211 hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, 215 hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
212 &hpet_clockevent); 216 &hpet_clockevent);
213 hpet_clockevent.min_delta_ns = clockevent_delta2ns(0x30, 217 /* 5 usec minimum reprogramming delta. */
214 &hpet_clockevent); 218 hpet_clockevent.min_delta_ns = 5000;
215 219
216 /* 220 /*
217 * Start hpet with the boot cpu mask and make it 221 * Start hpet with the boot cpu mask and make it
@@ -270,15 +274,22 @@ static void hpet_legacy_set_mode(enum clock_event_mode mode,
270} 274}
271 275
272static int hpet_legacy_next_event(unsigned long delta, 276static int hpet_legacy_next_event(unsigned long delta,
273 struct clock_event_device *evt) 277 struct clock_event_device *evt)
274{ 278{
275 unsigned long cnt; 279 u32 cnt;
276 280
277 cnt = hpet_readl(HPET_COUNTER); 281 cnt = hpet_readl(HPET_COUNTER);
278 cnt += delta; 282 cnt += (u32) delta;
279 hpet_writel(cnt, HPET_T0_CMP); 283 hpet_writel(cnt, HPET_T0_CMP);
280 284
281 return ((long)(hpet_readl(HPET_COUNTER) - cnt ) > 0) ? -ETIME : 0; 285 /*
286 * We need to read back the CMP register to make sure that
287 * what we wrote hit the chip before we compare it to the
288 * counter.
289 */
290 WARN_ON((u32)hpet_readl(HPET_T0_CMP) != cnt);
291
292 return (s32)((u32)hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
282} 293}
283 294
284/* 295/*
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index eb9ddd8efb82..1f20608d4ca8 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -21,9 +21,12 @@
21# include <asm/sigcontext32.h> 21# include <asm/sigcontext32.h>
22# include <asm/user32.h> 22# include <asm/user32.h>
23#else 23#else
24# define save_i387_ia32 save_i387 24# define save_i387_xstate_ia32 save_i387_xstate
25# define restore_i387_ia32 restore_i387 25# define restore_i387_xstate_ia32 restore_i387_xstate
26# define _fpstate_ia32 _fpstate 26# define _fpstate_ia32 _fpstate
27# define _xstate_ia32 _xstate
28# define sig_xstate_ia32_size sig_xstate_size
29# define fx_sw_reserved_ia32 fx_sw_reserved
27# define user_i387_ia32_struct user_i387_struct 30# define user_i387_ia32_struct user_i387_struct
28# define user32_fxsr_struct user_fxsr_struct 31# define user32_fxsr_struct user_fxsr_struct
29#endif 32#endif
@@ -36,6 +39,7 @@
36 39
37static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; 40static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
38unsigned int xstate_size; 41unsigned int xstate_size;
42unsigned int sig_xstate_ia32_size = sizeof(struct _fpstate_ia32);
39static struct i387_fxsave_struct fx_scratch __cpuinitdata; 43static struct i387_fxsave_struct fx_scratch __cpuinitdata;
40 44
41void __cpuinit mxcsr_feature_mask_init(void) 45void __cpuinit mxcsr_feature_mask_init(void)
@@ -61,6 +65,11 @@ void __init init_thread_xstate(void)
61 return; 65 return;
62 } 66 }
63 67
68 if (cpu_has_xsave) {
69 xsave_cntxt_init();
70 return;
71 }
72
64 if (cpu_has_fxsr) 73 if (cpu_has_fxsr)
65 xstate_size = sizeof(struct i387_fxsave_struct); 74 xstate_size = sizeof(struct i387_fxsave_struct);
66#ifdef CONFIG_X86_32 75#ifdef CONFIG_X86_32
@@ -83,9 +92,19 @@ void __cpuinit fpu_init(void)
83 92
84 write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */ 93 write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */
85 94
95 /*
96 * Boot processor to setup the FP and extended state context info.
97 */
98 if (!smp_processor_id())
99 init_thread_xstate();
100 xsave_init();
101
86 mxcsr_feature_mask_init(); 102 mxcsr_feature_mask_init();
87 /* clean state in init */ 103 /* clean state in init */
88 current_thread_info()->status = 0; 104 if (cpu_has_xsave)
105 current_thread_info()->status = TS_XSAVE;
106 else
107 current_thread_info()->status = 0;
89 clear_used_math(); 108 clear_used_math();
90} 109}
91#endif /* CONFIG_X86_64 */ 110#endif /* CONFIG_X86_64 */
@@ -195,6 +214,13 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
195 */ 214 */
196 target->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask; 215 target->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
197 216
217 /*
218 * update the header bits in the xsave header, indicating the
219 * presence of FP and SSE state.
220 */
221 if (cpu_has_xsave)
222 target->thread.xstate->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
223
198 return ret; 224 return ret;
199} 225}
200 226
@@ -395,6 +421,12 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,
395 if (!ret) 421 if (!ret)
396 convert_to_fxsr(target, &env); 422 convert_to_fxsr(target, &env);
397 423
424 /*
425 * update the header bit in the xsave header, indicating the
426 * presence of FP.
427 */
428 if (cpu_has_xsave)
429 target->thread.xstate->xsave.xsave_hdr.xstate_bv |= XSTATE_FP;
398 return ret; 430 return ret;
399} 431}
400 432
@@ -407,7 +439,6 @@ static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf)
407 struct task_struct *tsk = current; 439 struct task_struct *tsk = current;
408 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave; 440 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave;
409 441
410 unlazy_fpu(tsk);
411 fp->status = fp->swd; 442 fp->status = fp->swd;
412 if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct))) 443 if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct)))
413 return -1; 444 return -1;
@@ -421,8 +452,6 @@ static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
421 struct user_i387_ia32_struct env; 452 struct user_i387_ia32_struct env;
422 int err = 0; 453 int err = 0;
423 454
424 unlazy_fpu(tsk);
425
426 convert_from_fxsr(&env, tsk); 455 convert_from_fxsr(&env, tsk);
427 if (__copy_to_user(buf, &env, sizeof(env))) 456 if (__copy_to_user(buf, &env, sizeof(env)))
428 return -1; 457 return -1;
@@ -432,16 +461,54 @@ static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
432 if (err) 461 if (err)
433 return -1; 462 return -1;
434 463
435 if (__copy_to_user(&buf->_fxsr_env[0], fx, 464 if (__copy_to_user(&buf->_fxsr_env[0], fx, xstate_size))
436 sizeof(struct i387_fxsave_struct))) 465 return -1;
466 return 1;
467}
468
469static int save_i387_xsave(void __user *buf)
470{
471 struct task_struct *tsk = current;
472 struct _fpstate_ia32 __user *fx = buf;
473 int err = 0;
474
475 /*
476 * For legacy compatible, we always set FP/SSE bits in the bit
477 * vector while saving the state to the user context.
478 * This will enable us capturing any changes(during sigreturn) to
479 * the FP/SSE bits by the legacy applications which don't touch
480 * xstate_bv in the xsave header.
481 *
482 * xsave aware applications can change the xstate_bv in the xsave
483 * header as well as change any contents in the memory layout.
484 * xrestore as part of sigreturn will capture all the changes.
485 */
486 tsk->thread.xstate->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
487
488 if (save_i387_fxsave(fx) < 0)
489 return -1;
490
491 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved_ia32,
492 sizeof(struct _fpx_sw_bytes));
493 err |= __put_user(FP_XSTATE_MAGIC2,
494 (__u32 __user *) (buf + sig_xstate_ia32_size
495 - FP_XSTATE_MAGIC2_SIZE));
496 if (err)
437 return -1; 497 return -1;
498
438 return 1; 499 return 1;
439} 500}
440 501
441int save_i387_ia32(struct _fpstate_ia32 __user *buf) 502int save_i387_xstate_ia32(void __user *buf)
442{ 503{
504 struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
505 struct task_struct *tsk = current;
506
443 if (!used_math()) 507 if (!used_math())
444 return 0; 508 return 0;
509
510 if (!access_ok(VERIFY_WRITE, buf, sig_xstate_ia32_size))
511 return -EACCES;
445 /* 512 /*
446 * This will cause a "finit" to be triggered by the next 513 * This will cause a "finit" to be triggered by the next
447 * attempted FPU operation by the 'current' process. 514 * attempted FPU operation by the 'current' process.
@@ -451,13 +518,17 @@ int save_i387_ia32(struct _fpstate_ia32 __user *buf)
451 if (!HAVE_HWFP) { 518 if (!HAVE_HWFP) {
452 return fpregs_soft_get(current, NULL, 519 return fpregs_soft_get(current, NULL,
453 0, sizeof(struct user_i387_ia32_struct), 520 0, sizeof(struct user_i387_ia32_struct),
454 NULL, buf) ? -1 : 1; 521 NULL, fp) ? -1 : 1;
455 } 522 }
456 523
524 unlazy_fpu(tsk);
525
526 if (cpu_has_xsave)
527 return save_i387_xsave(fp);
457 if (cpu_has_fxsr) 528 if (cpu_has_fxsr)
458 return save_i387_fxsave(buf); 529 return save_i387_fxsave(fp);
459 else 530 else
460 return save_i387_fsave(buf); 531 return save_i387_fsave(fp);
461} 532}
462 533
463static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf) 534static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
@@ -468,14 +539,15 @@ static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
468 sizeof(struct i387_fsave_struct)); 539 sizeof(struct i387_fsave_struct));
469} 540}
470 541
471static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf) 542static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf,
543 unsigned int size)
472{ 544{
473 struct task_struct *tsk = current; 545 struct task_struct *tsk = current;
474 struct user_i387_ia32_struct env; 546 struct user_i387_ia32_struct env;
475 int err; 547 int err;
476 548
477 err = __copy_from_user(&tsk->thread.xstate->fxsave, &buf->_fxsr_env[0], 549 err = __copy_from_user(&tsk->thread.xstate->fxsave, &buf->_fxsr_env[0],
478 sizeof(struct i387_fxsave_struct)); 550 size);
479 /* mxcsr reserved bits must be masked to zero for security reasons */ 551 /* mxcsr reserved bits must be masked to zero for security reasons */
480 tsk->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask; 552 tsk->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
481 if (err || __copy_from_user(&env, buf, sizeof(env))) 553 if (err || __copy_from_user(&env, buf, sizeof(env)))
@@ -485,14 +557,69 @@ static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf)
485 return 0; 557 return 0;
486} 558}
487 559
488int restore_i387_ia32(struct _fpstate_ia32 __user *buf) 560static int restore_i387_xsave(void __user *buf)
561{
562 struct _fpx_sw_bytes fx_sw_user;
563 struct _fpstate_ia32 __user *fx_user =
564 ((struct _fpstate_ia32 __user *) buf);
565 struct i387_fxsave_struct __user *fx =
566 (struct i387_fxsave_struct __user *) &fx_user->_fxsr_env[0];
567 struct xsave_hdr_struct *xsave_hdr =
568 &current->thread.xstate->xsave.xsave_hdr;
569 u64 mask;
570 int err;
571
572 if (check_for_xstate(fx, buf, &fx_sw_user))
573 goto fx_only;
574
575 mask = fx_sw_user.xstate_bv;
576
577 err = restore_i387_fxsave(buf, fx_sw_user.xstate_size);
578
579 xsave_hdr->xstate_bv &= pcntxt_mask;
580 /*
581 * These bits must be zero.
582 */
583 xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0;
584
585 /*
586 * Init the state that is not present in the memory layout
587 * and enabled by the OS.
588 */
589 mask = ~(pcntxt_mask & ~mask);
590 xsave_hdr->xstate_bv &= mask;
591
592 return err;
593fx_only:
594 /*
595 * Couldn't find the extended state information in the memory
596 * layout. Restore the FP/SSE and init the other extended state
597 * enabled by the OS.
598 */
599 xsave_hdr->xstate_bv = XSTATE_FPSSE;
600 return restore_i387_fxsave(buf, sizeof(struct i387_fxsave_struct));
601}
602
603int restore_i387_xstate_ia32(void __user *buf)
489{ 604{
490 int err; 605 int err;
491 struct task_struct *tsk = current; 606 struct task_struct *tsk = current;
607 struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf;
492 608
493 if (HAVE_HWFP) 609 if (HAVE_HWFP)
494 clear_fpu(tsk); 610 clear_fpu(tsk);
495 611
612 if (!buf) {
613 if (used_math()) {
614 clear_fpu(tsk);
615 clear_used_math();
616 }
617
618 return 0;
619 } else
620 if (!access_ok(VERIFY_READ, buf, sig_xstate_ia32_size))
621 return -EACCES;
622
496 if (!used_math()) { 623 if (!used_math()) {
497 err = init_fpu(tsk); 624 err = init_fpu(tsk);
498 if (err) 625 if (err)
@@ -500,14 +627,17 @@ int restore_i387_ia32(struct _fpstate_ia32 __user *buf)
500 } 627 }
501 628
502 if (HAVE_HWFP) { 629 if (HAVE_HWFP) {
503 if (cpu_has_fxsr) 630 if (cpu_has_xsave)
504 err = restore_i387_fxsave(buf); 631 err = restore_i387_xsave(buf);
632 else if (cpu_has_fxsr)
633 err = restore_i387_fxsave(fp, sizeof(struct
634 i387_fxsave_struct));
505 else 635 else
506 err = restore_i387_fsave(buf); 636 err = restore_i387_fsave(fp);
507 } else { 637 } else {
508 err = fpregs_soft_set(current, NULL, 638 err = fpregs_soft_set(current, NULL,
509 0, sizeof(struct user_i387_ia32_struct), 639 0, sizeof(struct user_i387_ia32_struct),
510 NULL, buf) != 0; 640 NULL, fp) != 0;
511 } 641 }
512 set_used_math(); 642 set_used_math();
513 643
diff --git a/arch/x86/kernel/i8259.c b/arch/x86/kernel/i8259.c
index dc92b49d9204..4b8a53d841f7 100644
--- a/arch/x86/kernel/i8259.c
+++ b/arch/x86/kernel/i8259.c
@@ -282,6 +282,30 @@ static int __init i8259A_init_sysfs(void)
282 282
283device_initcall(i8259A_init_sysfs); 283device_initcall(i8259A_init_sysfs);
284 284
285void mask_8259A(void)
286{
287 unsigned long flags;
288
289 spin_lock_irqsave(&i8259A_lock, flags);
290
291 outb(0xff, PIC_MASTER_IMR); /* mask all of 8259A-1 */
292 outb(0xff, PIC_SLAVE_IMR); /* mask all of 8259A-2 */
293
294 spin_unlock_irqrestore(&i8259A_lock, flags);
295}
296
297void unmask_8259A(void)
298{
299 unsigned long flags;
300
301 spin_lock_irqsave(&i8259A_lock, flags);
302
303 outb(cached_master_mask, PIC_MASTER_IMR); /* restore master IRQ mask */
304 outb(cached_slave_mask, PIC_SLAVE_IMR); /* restore slave IRQ mask */
305
306 spin_unlock_irqrestore(&i8259A_lock, flags);
307}
308
285void init_8259A(int auto_eoi) 309void init_8259A(int auto_eoi)
286{ 310{
287 unsigned long flags; 311 unsigned long flags;
diff --git a/arch/x86/kernel/io_apic_32.c b/arch/x86/kernel/io_apic_32.c
index 09cddb57bec4..e710289f673e 100644
--- a/arch/x86/kernel/io_apic_32.c
+++ b/arch/x86/kernel/io_apic_32.c
@@ -46,10 +46,13 @@
46#include <asm/nmi.h> 46#include <asm/nmi.h>
47#include <asm/msidef.h> 47#include <asm/msidef.h>
48#include <asm/hypertransport.h> 48#include <asm/hypertransport.h>
49#include <asm/setup.h>
49 50
50#include <mach_apic.h> 51#include <mach_apic.h>
51#include <mach_apicdef.h> 52#include <mach_apicdef.h>
52 53
54#define __apicdebuginit(type) static type __init
55
53int (*ioapic_renumber_irq)(int ioapic, int irq); 56int (*ioapic_renumber_irq)(int ioapic, int irq);
54atomic_t irq_mis_count; 57atomic_t irq_mis_count;
55 58
@@ -1341,7 +1344,8 @@ static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1341 ioapic_write_entry(apic, pin, entry); 1344 ioapic_write_entry(apic, pin, entry);
1342} 1345}
1343 1346
1344void __init print_IO_APIC(void) 1347
1348__apicdebuginit(void) print_IO_APIC(void)
1345{ 1349{
1346 int apic, i; 1350 int apic, i;
1347 union IO_APIC_reg_00 reg_00; 1351 union IO_APIC_reg_00 reg_00;
@@ -1456,9 +1460,7 @@ void __init print_IO_APIC(void)
1456 return; 1460 return;
1457} 1461}
1458 1462
1459#if 0 1463__apicdebuginit(void) print_APIC_bitfield(int base)
1460
1461static void print_APIC_bitfield(int base)
1462{ 1464{
1463 unsigned int v; 1465 unsigned int v;
1464 int i, j; 1466 int i, j;
@@ -1479,9 +1481,10 @@ static void print_APIC_bitfield(int base)
1479 } 1481 }
1480} 1482}
1481 1483
1482void /*__init*/ print_local_APIC(void *dummy) 1484__apicdebuginit(void) print_local_APIC(void *dummy)
1483{ 1485{
1484 unsigned int v, ver, maxlvt; 1486 unsigned int v, ver, maxlvt;
1487 u64 icr;
1485 1488
1486 if (apic_verbosity == APIC_QUIET) 1489 if (apic_verbosity == APIC_QUIET)
1487 return; 1490 return;
@@ -1490,7 +1493,7 @@ void /*__init*/ print_local_APIC(void *dummy)
1490 smp_processor_id(), hard_smp_processor_id()); 1493 smp_processor_id(), hard_smp_processor_id());
1491 v = apic_read(APIC_ID); 1494 v = apic_read(APIC_ID);
1492 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, 1495 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v,
1493 GET_APIC_ID(read_apic_id())); 1496 GET_APIC_ID(v));
1494 v = apic_read(APIC_LVR); 1497 v = apic_read(APIC_LVR);
1495 printk(KERN_INFO "... APIC VERSION: %08x\n", v); 1498 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1496 ver = GET_APIC_VERSION(v); 1499 ver = GET_APIC_VERSION(v);
@@ -1532,10 +1535,9 @@ void /*__init*/ print_local_APIC(void *dummy)
1532 printk(KERN_DEBUG "... APIC ESR: %08x\n", v); 1535 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1533 } 1536 }
1534 1537
1535 v = apic_read(APIC_ICR); 1538 icr = apic_icr_read();
1536 printk(KERN_DEBUG "... APIC ICR: %08x\n", v); 1539 printk(KERN_DEBUG "... APIC ICR: %08x\n", icr);
1537 v = apic_read(APIC_ICR2); 1540 printk(KERN_DEBUG "... APIC ICR2: %08x\n", icr >> 32);
1538 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1539 1541
1540 v = apic_read(APIC_LVTT); 1542 v = apic_read(APIC_LVTT);
1541 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v); 1543 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
@@ -1563,12 +1565,12 @@ void /*__init*/ print_local_APIC(void *dummy)
1563 printk("\n"); 1565 printk("\n");
1564} 1566}
1565 1567
1566void print_all_local_APICs(void) 1568__apicdebuginit(void) print_all_local_APICs(void)
1567{ 1569{
1568 on_each_cpu(print_local_APIC, NULL, 1); 1570 on_each_cpu(print_local_APIC, NULL, 1);
1569} 1571}
1570 1572
1571void /*__init*/ print_PIC(void) 1573__apicdebuginit(void) print_PIC(void)
1572{ 1574{
1573 unsigned int v; 1575 unsigned int v;
1574 unsigned long flags; 1576 unsigned long flags;
@@ -1600,7 +1602,17 @@ void /*__init*/ print_PIC(void)
1600 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v); 1602 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1601} 1603}
1602 1604
1603#endif /* 0 */ 1605__apicdebuginit(int) print_all_ICs(void)
1606{
1607 print_PIC();
1608 print_all_local_APICs();
1609 print_IO_APIC();
1610
1611 return 0;
1612}
1613
1614fs_initcall(print_all_ICs);
1615
1604 1616
1605static void __init enable_IO_APIC(void) 1617static void __init enable_IO_APIC(void)
1606{ 1618{
@@ -1698,8 +1710,7 @@ void disable_IO_APIC(void)
1698 entry.dest_mode = 0; /* Physical */ 1710 entry.dest_mode = 0; /* Physical */
1699 entry.delivery_mode = dest_ExtINT; /* ExtInt */ 1711 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1700 entry.vector = 0; 1712 entry.vector = 0;
1701 entry.dest.physical.physical_dest = 1713 entry.dest.physical.physical_dest = read_apic_id();
1702 GET_APIC_ID(read_apic_id());
1703 1714
1704 /* 1715 /*
1705 * Add it to the IO-APIC irq-routing table: 1716 * Add it to the IO-APIC irq-routing table:
@@ -1725,10 +1736,8 @@ static void __init setup_ioapic_ids_from_mpc(void)
1725 unsigned char old_id; 1736 unsigned char old_id;
1726 unsigned long flags; 1737 unsigned long flags;
1727 1738
1728#ifdef CONFIG_X86_NUMAQ 1739 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
1729 if (found_numaq)
1730 return; 1740 return;
1731#endif
1732 1741
1733 /* 1742 /*
1734 * Don't check I/O APIC IDs for xAPIC systems. They have 1743 * Don't check I/O APIC IDs for xAPIC systems. They have
@@ -2329,8 +2338,6 @@ void __init setup_IO_APIC(void)
2329 setup_IO_APIC_irqs(); 2338 setup_IO_APIC_irqs();
2330 init_IO_APIC_traps(); 2339 init_IO_APIC_traps();
2331 check_timer(); 2340 check_timer();
2332 if (!acpi_ioapic)
2333 print_IO_APIC();
2334} 2341}
2335 2342
2336/* 2343/*
diff --git a/arch/x86/kernel/io_apic_64.c b/arch/x86/kernel/io_apic_64.c
index 61a83b70c18f..02063ae042f7 100644
--- a/arch/x86/kernel/io_apic_64.c
+++ b/arch/x86/kernel/io_apic_64.c
@@ -37,6 +37,7 @@
37#include <acpi/acpi_bus.h> 37#include <acpi/acpi_bus.h>
38#endif 38#endif
39#include <linux/bootmem.h> 39#include <linux/bootmem.h>
40#include <linux/dmar.h>
40 41
41#include <asm/idle.h> 42#include <asm/idle.h>
42#include <asm/io.h> 43#include <asm/io.h>
@@ -49,10 +50,13 @@
49#include <asm/nmi.h> 50#include <asm/nmi.h>
50#include <asm/msidef.h> 51#include <asm/msidef.h>
51#include <asm/hypertransport.h> 52#include <asm/hypertransport.h>
53#include <asm/irq_remapping.h>
52 54
53#include <mach_ipi.h> 55#include <mach_ipi.h>
54#include <mach_apic.h> 56#include <mach_apic.h>
55 57
58#define __apicdebuginit(type) static type __init
59
56struct irq_cfg { 60struct irq_cfg {
57 cpumask_t domain; 61 cpumask_t domain;
58 cpumask_t old_domain; 62 cpumask_t old_domain;
@@ -87,8 +91,6 @@ int first_system_vector = 0xfe;
87 91
88char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE}; 92char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};
89 93
90#define __apicdebuginit __init
91
92int sis_apic_bug; /* not actually supported, dummy for compile */ 94int sis_apic_bug; /* not actually supported, dummy for compile */
93 95
94static int no_timer_check; 96static int no_timer_check;
@@ -108,6 +110,9 @@ static DEFINE_SPINLOCK(vector_lock);
108 */ 110 */
109int nr_ioapic_registers[MAX_IO_APICS]; 111int nr_ioapic_registers[MAX_IO_APICS];
110 112
113/* I/O APIC RTE contents at the OS boot up */
114struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
115
111/* I/O APIC entries */ 116/* I/O APIC entries */
112struct mp_config_ioapic mp_ioapics[MAX_IO_APICS]; 117struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
113int nr_ioapics; 118int nr_ioapics;
@@ -303,7 +308,12 @@ static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
303 pin = entry->pin; 308 pin = entry->pin;
304 if (pin == -1) 309 if (pin == -1)
305 break; 310 break;
306 io_apic_write(apic, 0x11 + pin*2, dest); 311 /*
312 * With interrupt-remapping, destination information comes
313 * from interrupt-remapping table entry.
314 */
315 if (!irq_remapped(irq))
316 io_apic_write(apic, 0x11 + pin*2, dest);
307 reg = io_apic_read(apic, 0x10 + pin*2); 317 reg = io_apic_read(apic, 0x10 + pin*2);
308 reg &= ~IO_APIC_REDIR_VECTOR_MASK; 318 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
309 reg |= vector; 319 reg |= vector;
@@ -440,6 +450,69 @@ static void clear_IO_APIC (void)
440 clear_IO_APIC_pin(apic, pin); 450 clear_IO_APIC_pin(apic, pin);
441} 451}
442 452
453/*
454 * Saves and masks all the unmasked IO-APIC RTE's
455 */
456int save_mask_IO_APIC_setup(void)
457{
458 union IO_APIC_reg_01 reg_01;
459 unsigned long flags;
460 int apic, pin;
461
462 /*
463 * The number of IO-APIC IRQ registers (== #pins):
464 */
465 for (apic = 0; apic < nr_ioapics; apic++) {
466 spin_lock_irqsave(&ioapic_lock, flags);
467 reg_01.raw = io_apic_read(apic, 1);
468 spin_unlock_irqrestore(&ioapic_lock, flags);
469 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
470 }
471
472 for (apic = 0; apic < nr_ioapics; apic++) {
473 early_ioapic_entries[apic] =
474 kzalloc(sizeof(struct IO_APIC_route_entry) *
475 nr_ioapic_registers[apic], GFP_KERNEL);
476 if (!early_ioapic_entries[apic])
477 return -ENOMEM;
478 }
479
480 for (apic = 0; apic < nr_ioapics; apic++)
481 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
482 struct IO_APIC_route_entry entry;
483
484 entry = early_ioapic_entries[apic][pin] =
485 ioapic_read_entry(apic, pin);
486 if (!entry.mask) {
487 entry.mask = 1;
488 ioapic_write_entry(apic, pin, entry);
489 }
490 }
491 return 0;
492}
493
494void restore_IO_APIC_setup(void)
495{
496 int apic, pin;
497
498 for (apic = 0; apic < nr_ioapics; apic++)
499 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
500 ioapic_write_entry(apic, pin,
501 early_ioapic_entries[apic][pin]);
502}
503
504void reinit_intr_remapped_IO_APIC(int intr_remapping)
505{
506 /*
507 * for now plain restore of previous settings.
508 * TBD: In the case of OS enabling interrupt-remapping,
509 * IO-APIC RTE's need to be setup to point to interrupt-remapping
510 * table entries. for now, do a plain restore, and wait for
511 * the setup_IO_APIC_irqs() to do proper initialization.
512 */
513 restore_IO_APIC_setup();
514}
515
443int skip_ioapic_setup; 516int skip_ioapic_setup;
444int ioapic_force; 517int ioapic_force;
445 518
@@ -839,18 +912,98 @@ void __setup_vector_irq(int cpu)
839} 912}
840 913
841static struct irq_chip ioapic_chip; 914static struct irq_chip ioapic_chip;
915#ifdef CONFIG_INTR_REMAP
916static struct irq_chip ir_ioapic_chip;
917#endif
842 918
843static void ioapic_register_intr(int irq, unsigned long trigger) 919static void ioapic_register_intr(int irq, unsigned long trigger)
844{ 920{
845 if (trigger) { 921 if (trigger)
846 irq_desc[irq].status |= IRQ_LEVEL; 922 irq_desc[irq].status |= IRQ_LEVEL;
847 set_irq_chip_and_handler_name(irq, &ioapic_chip, 923 else
848 handle_fasteoi_irq, "fasteoi");
849 } else {
850 irq_desc[irq].status &= ~IRQ_LEVEL; 924 irq_desc[irq].status &= ~IRQ_LEVEL;
925
926#ifdef CONFIG_INTR_REMAP
927 if (irq_remapped(irq)) {
928 irq_desc[irq].status |= IRQ_MOVE_PCNTXT;
929 if (trigger)
930 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
931 handle_fasteoi_irq,
932 "fasteoi");
933 else
934 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
935 handle_edge_irq, "edge");
936 return;
937 }
938#endif
939 if (trigger)
940 set_irq_chip_and_handler_name(irq, &ioapic_chip,
941 handle_fasteoi_irq,
942 "fasteoi");
943 else
851 set_irq_chip_and_handler_name(irq, &ioapic_chip, 944 set_irq_chip_and_handler_name(irq, &ioapic_chip,
852 handle_edge_irq, "edge"); 945 handle_edge_irq, "edge");
946}
947
948static int setup_ioapic_entry(int apic, int irq,
949 struct IO_APIC_route_entry *entry,
950 unsigned int destination, int trigger,
951 int polarity, int vector)
952{
953 /*
954 * add it to the IO-APIC irq-routing table:
955 */
956 memset(entry,0,sizeof(*entry));
957
958#ifdef CONFIG_INTR_REMAP
959 if (intr_remapping_enabled) {
960 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
961 struct irte irte;
962 struct IR_IO_APIC_route_entry *ir_entry =
963 (struct IR_IO_APIC_route_entry *) entry;
964 int index;
965
966 if (!iommu)
967 panic("No mapping iommu for ioapic %d\n", apic);
968
969 index = alloc_irte(iommu, irq, 1);
970 if (index < 0)
971 panic("Failed to allocate IRTE for ioapic %d\n", apic);
972
973 memset(&irte, 0, sizeof(irte));
974
975 irte.present = 1;
976 irte.dst_mode = INT_DEST_MODE;
977 irte.trigger_mode = trigger;
978 irte.dlvry_mode = INT_DELIVERY_MODE;
979 irte.vector = vector;
980 irte.dest_id = IRTE_DEST(destination);
981
982 modify_irte(irq, &irte);
983
984 ir_entry->index2 = (index >> 15) & 0x1;
985 ir_entry->zero = 0;
986 ir_entry->format = 1;
987 ir_entry->index = (index & 0x7fff);
988 } else
989#endif
990 {
991 entry->delivery_mode = INT_DELIVERY_MODE;
992 entry->dest_mode = INT_DEST_MODE;
993 entry->dest = destination;
853 } 994 }
995
996 entry->mask = 0; /* enable IRQ */
997 entry->trigger = trigger;
998 entry->polarity = polarity;
999 entry->vector = vector;
1000
1001 /* Mask level triggered irqs.
1002 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1003 */
1004 if (trigger)
1005 entry->mask = 1;
1006 return 0;
854} 1007}
855 1008
856static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, 1009static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
@@ -875,24 +1028,15 @@ static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
875 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector, 1028 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
876 irq, trigger, polarity); 1029 irq, trigger, polarity);
877 1030
878 /*
879 * add it to the IO-APIC irq-routing table:
880 */
881 memset(&entry,0,sizeof(entry));
882
883 entry.delivery_mode = INT_DELIVERY_MODE;
884 entry.dest_mode = INT_DEST_MODE;
885 entry.dest = cpu_mask_to_apicid(mask);
886 entry.mask = 0; /* enable IRQ */
887 entry.trigger = trigger;
888 entry.polarity = polarity;
889 entry.vector = cfg->vector;
890 1031
891 /* Mask level triggered irqs. 1032 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
892 * Use IRQ_DELAYED_DISABLE for edge triggered irqs. 1033 cpu_mask_to_apicid(mask), trigger, polarity,
893 */ 1034 cfg->vector)) {
894 if (trigger) 1035 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
895 entry.mask = 1; 1036 mp_ioapics[apic].mp_apicid, pin);
1037 __clear_irq_vector(irq);
1038 return;
1039 }
896 1040
897 ioapic_register_intr(irq, trigger); 1041 ioapic_register_intr(irq, trigger);
898 if (irq < 16) 1042 if (irq < 16)
@@ -944,6 +1088,9 @@ static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
944{ 1088{
945 struct IO_APIC_route_entry entry; 1089 struct IO_APIC_route_entry entry;
946 1090
1091 if (intr_remapping_enabled)
1092 return;
1093
947 memset(&entry, 0, sizeof(entry)); 1094 memset(&entry, 0, sizeof(entry));
948 1095
949 /* 1096 /*
@@ -970,7 +1117,8 @@ static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
970 ioapic_write_entry(apic, pin, entry); 1117 ioapic_write_entry(apic, pin, entry);
971} 1118}
972 1119
973void __apicdebuginit print_IO_APIC(void) 1120
1121__apicdebuginit(void) print_IO_APIC(void)
974{ 1122{
975 int apic, i; 1123 int apic, i;
976 union IO_APIC_reg_00 reg_00; 1124 union IO_APIC_reg_00 reg_00;
@@ -1064,9 +1212,7 @@ void __apicdebuginit print_IO_APIC(void)
1064 return; 1212 return;
1065} 1213}
1066 1214
1067#if 0 1215__apicdebuginit(void) print_APIC_bitfield(int base)
1068
1069static __apicdebuginit void print_APIC_bitfield (int base)
1070{ 1216{
1071 unsigned int v; 1217 unsigned int v;
1072 int i, j; 1218 int i, j;
@@ -1087,9 +1233,10 @@ static __apicdebuginit void print_APIC_bitfield (int base)
1087 } 1233 }
1088} 1234}
1089 1235
1090void __apicdebuginit print_local_APIC(void * dummy) 1236__apicdebuginit(void) print_local_APIC(void *dummy)
1091{ 1237{
1092 unsigned int v, ver, maxlvt; 1238 unsigned int v, ver, maxlvt;
1239 unsigned long icr;
1093 1240
1094 if (apic_verbosity == APIC_QUIET) 1241 if (apic_verbosity == APIC_QUIET)
1095 return; 1242 return;
@@ -1097,7 +1244,7 @@ void __apicdebuginit print_local_APIC(void * dummy)
1097 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n", 1244 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1098 smp_processor_id(), hard_smp_processor_id()); 1245 smp_processor_id(), hard_smp_processor_id());
1099 v = apic_read(APIC_ID); 1246 v = apic_read(APIC_ID);
1100 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id())); 1247 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1101 v = apic_read(APIC_LVR); 1248 v = apic_read(APIC_LVR);
1102 printk(KERN_INFO "... APIC VERSION: %08x\n", v); 1249 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1103 ver = GET_APIC_VERSION(v); 1250 ver = GET_APIC_VERSION(v);
@@ -1133,10 +1280,9 @@ void __apicdebuginit print_local_APIC(void * dummy)
1133 v = apic_read(APIC_ESR); 1280 v = apic_read(APIC_ESR);
1134 printk(KERN_DEBUG "... APIC ESR: %08x\n", v); 1281 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1135 1282
1136 v = apic_read(APIC_ICR); 1283 icr = apic_icr_read();
1137 printk(KERN_DEBUG "... APIC ICR: %08x\n", v); 1284 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1138 v = apic_read(APIC_ICR2); 1285 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1139 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1140 1286
1141 v = apic_read(APIC_LVTT); 1287 v = apic_read(APIC_LVTT);
1142 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v); 1288 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
@@ -1164,12 +1310,12 @@ void __apicdebuginit print_local_APIC(void * dummy)
1164 printk("\n"); 1310 printk("\n");
1165} 1311}
1166 1312
1167void print_all_local_APICs (void) 1313__apicdebuginit(void) print_all_local_APICs(void)
1168{ 1314{
1169 on_each_cpu(print_local_APIC, NULL, 1); 1315 on_each_cpu(print_local_APIC, NULL, 1);
1170} 1316}
1171 1317
1172void __apicdebuginit print_PIC(void) 1318__apicdebuginit(void) print_PIC(void)
1173{ 1319{
1174 unsigned int v; 1320 unsigned int v;
1175 unsigned long flags; 1321 unsigned long flags;
@@ -1201,7 +1347,17 @@ void __apicdebuginit print_PIC(void)
1201 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v); 1347 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1202} 1348}
1203 1349
1204#endif /* 0 */ 1350__apicdebuginit(int) print_all_ICs(void)
1351{
1352 print_PIC();
1353 print_all_local_APICs();
1354 print_IO_APIC();
1355
1356 return 0;
1357}
1358
1359fs_initcall(print_all_ICs);
1360
1205 1361
1206void __init enable_IO_APIC(void) 1362void __init enable_IO_APIC(void)
1207{ 1363{
@@ -1291,7 +1447,7 @@ void disable_IO_APIC(void)
1291 entry.dest_mode = 0; /* Physical */ 1447 entry.dest_mode = 0; /* Physical */
1292 entry.delivery_mode = dest_ExtINT; /* ExtInt */ 1448 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1293 entry.vector = 0; 1449 entry.vector = 0;
1294 entry.dest = GET_APIC_ID(read_apic_id()); 1450 entry.dest = read_apic_id();
1295 1451
1296 /* 1452 /*
1297 * Add it to the IO-APIC irq-routing table: 1453 * Add it to the IO-APIC irq-routing table:
@@ -1397,6 +1553,147 @@ static int ioapic_retrigger_irq(unsigned int irq)
1397 */ 1553 */
1398 1554
1399#ifdef CONFIG_SMP 1555#ifdef CONFIG_SMP
1556
1557#ifdef CONFIG_INTR_REMAP
1558static void ir_irq_migration(struct work_struct *work);
1559
1560static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
1561
1562/*
1563 * Migrate the IO-APIC irq in the presence of intr-remapping.
1564 *
1565 * For edge triggered, irq migration is a simple atomic update(of vector
1566 * and cpu destination) of IRTE and flush the hardware cache.
1567 *
1568 * For level triggered, we need to modify the io-apic RTE aswell with the update
1569 * vector information, along with modifying IRTE with vector and destination.
1570 * So irq migration for level triggered is little bit more complex compared to
1571 * edge triggered migration. But the good news is, we use the same algorithm
1572 * for level triggered migration as we have today, only difference being,
1573 * we now initiate the irq migration from process context instead of the
1574 * interrupt context.
1575 *
1576 * In future, when we do a directed EOI (combined with cpu EOI broadcast
1577 * suppression) to the IO-APIC, level triggered irq migration will also be
1578 * as simple as edge triggered migration and we can do the irq migration
1579 * with a simple atomic update to IO-APIC RTE.
1580 */
1581static void migrate_ioapic_irq(int irq, cpumask_t mask)
1582{
1583 struct irq_cfg *cfg = irq_cfg + irq;
1584 struct irq_desc *desc = irq_desc + irq;
1585 cpumask_t tmp, cleanup_mask;
1586 struct irte irte;
1587 int modify_ioapic_rte = desc->status & IRQ_LEVEL;
1588 unsigned int dest;
1589 unsigned long flags;
1590
1591 cpus_and(tmp, mask, cpu_online_map);
1592 if (cpus_empty(tmp))
1593 return;
1594
1595 if (get_irte(irq, &irte))
1596 return;
1597
1598 if (assign_irq_vector(irq, mask))
1599 return;
1600
1601 cpus_and(tmp, cfg->domain, mask);
1602 dest = cpu_mask_to_apicid(tmp);
1603
1604 if (modify_ioapic_rte) {
1605 spin_lock_irqsave(&ioapic_lock, flags);
1606 __target_IO_APIC_irq(irq, dest, cfg->vector);
1607 spin_unlock_irqrestore(&ioapic_lock, flags);
1608 }
1609
1610 irte.vector = cfg->vector;
1611 irte.dest_id = IRTE_DEST(dest);
1612
1613 /*
1614 * Modified the IRTE and flushes the Interrupt entry cache.
1615 */
1616 modify_irte(irq, &irte);
1617
1618 if (cfg->move_in_progress) {
1619 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1620 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1621 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1622 cfg->move_in_progress = 0;
1623 }
1624
1625 irq_desc[irq].affinity = mask;
1626}
1627
1628static int migrate_irq_remapped_level(int irq)
1629{
1630 int ret = -1;
1631
1632 mask_IO_APIC_irq(irq);
1633
1634 if (io_apic_level_ack_pending(irq)) {
1635 /*
1636 * Interrupt in progress. Migrating irq now will change the
1637 * vector information in the IO-APIC RTE and that will confuse
1638 * the EOI broadcast performed by cpu.
1639 * So, delay the irq migration to the next instance.
1640 */
1641 schedule_delayed_work(&ir_migration_work, 1);
1642 goto unmask;
1643 }
1644
1645 /* everthing is clear. we have right of way */
1646 migrate_ioapic_irq(irq, irq_desc[irq].pending_mask);
1647
1648 ret = 0;
1649 irq_desc[irq].status &= ~IRQ_MOVE_PENDING;
1650 cpus_clear(irq_desc[irq].pending_mask);
1651
1652unmask:
1653 unmask_IO_APIC_irq(irq);
1654 return ret;
1655}
1656
1657static void ir_irq_migration(struct work_struct *work)
1658{
1659 int irq;
1660
1661 for (irq = 0; irq < NR_IRQS; irq++) {
1662 struct irq_desc *desc = irq_desc + irq;
1663 if (desc->status & IRQ_MOVE_PENDING) {
1664 unsigned long flags;
1665
1666 spin_lock_irqsave(&desc->lock, flags);
1667 if (!desc->chip->set_affinity ||
1668 !(desc->status & IRQ_MOVE_PENDING)) {
1669 desc->status &= ~IRQ_MOVE_PENDING;
1670 spin_unlock_irqrestore(&desc->lock, flags);
1671 continue;
1672 }
1673
1674 desc->chip->set_affinity(irq,
1675 irq_desc[irq].pending_mask);
1676 spin_unlock_irqrestore(&desc->lock, flags);
1677 }
1678 }
1679}
1680
1681/*
1682 * Migrates the IRQ destination in the process context.
1683 */
1684static void set_ir_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
1685{
1686 if (irq_desc[irq].status & IRQ_LEVEL) {
1687 irq_desc[irq].status |= IRQ_MOVE_PENDING;
1688 irq_desc[irq].pending_mask = mask;
1689 migrate_irq_remapped_level(irq);
1690 return;
1691 }
1692
1693 migrate_ioapic_irq(irq, mask);
1694}
1695#endif
1696
1400asmlinkage void smp_irq_move_cleanup_interrupt(void) 1697asmlinkage void smp_irq_move_cleanup_interrupt(void)
1401{ 1698{
1402 unsigned vector, me; 1699 unsigned vector, me;
@@ -1453,6 +1750,17 @@ static void irq_complete_move(unsigned int irq)
1453#else 1750#else
1454static inline void irq_complete_move(unsigned int irq) {} 1751static inline void irq_complete_move(unsigned int irq) {}
1455#endif 1752#endif
1753#ifdef CONFIG_INTR_REMAP
1754static void ack_x2apic_level(unsigned int irq)
1755{
1756 ack_x2APIC_irq();
1757}
1758
1759static void ack_x2apic_edge(unsigned int irq)
1760{
1761 ack_x2APIC_irq();
1762}
1763#endif
1456 1764
1457static void ack_apic_edge(unsigned int irq) 1765static void ack_apic_edge(unsigned int irq)
1458{ 1766{
@@ -1527,6 +1835,21 @@ static struct irq_chip ioapic_chip __read_mostly = {
1527 .retrigger = ioapic_retrigger_irq, 1835 .retrigger = ioapic_retrigger_irq,
1528}; 1836};
1529 1837
1838#ifdef CONFIG_INTR_REMAP
1839static struct irq_chip ir_ioapic_chip __read_mostly = {
1840 .name = "IR-IO-APIC",
1841 .startup = startup_ioapic_irq,
1842 .mask = mask_IO_APIC_irq,
1843 .unmask = unmask_IO_APIC_irq,
1844 .ack = ack_x2apic_edge,
1845 .eoi = ack_x2apic_level,
1846#ifdef CONFIG_SMP
1847 .set_affinity = set_ir_ioapic_affinity_irq,
1848#endif
1849 .retrigger = ioapic_retrigger_irq,
1850};
1851#endif
1852
1530static inline void init_IO_APIC_traps(void) 1853static inline void init_IO_APIC_traps(void)
1531{ 1854{
1532 int irq; 1855 int irq;
@@ -1712,6 +2035,8 @@ static inline void __init check_timer(void)
1712 * 8259A. 2035 * 8259A.
1713 */ 2036 */
1714 if (pin1 == -1) { 2037 if (pin1 == -1) {
2038 if (intr_remapping_enabled)
2039 panic("BIOS bug: timer not connected to IO-APIC");
1715 pin1 = pin2; 2040 pin1 = pin2;
1716 apic1 = apic2; 2041 apic1 = apic2;
1717 no_pin1 = 1; 2042 no_pin1 = 1;
@@ -1738,6 +2063,8 @@ static inline void __init check_timer(void)
1738 clear_IO_APIC_pin(0, pin1); 2063 clear_IO_APIC_pin(0, pin1);
1739 goto out; 2064 goto out;
1740 } 2065 }
2066 if (intr_remapping_enabled)
2067 panic("timer doesn't work through Interrupt-remapped IO-APIC");
1741 clear_IO_APIC_pin(apic1, pin1); 2068 clear_IO_APIC_pin(apic1, pin1);
1742 if (!no_pin1) 2069 if (!no_pin1)
1743 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: " 2070 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
@@ -1854,8 +2181,6 @@ void __init setup_IO_APIC(void)
1854 setup_IO_APIC_irqs(); 2181 setup_IO_APIC_irqs();
1855 init_IO_APIC_traps(); 2182 init_IO_APIC_traps();
1856 check_timer(); 2183 check_timer();
1857 if (!acpi_ioapic)
1858 print_IO_APIC();
1859} 2184}
1860 2185
1861struct sysfs_ioapic_data { 2186struct sysfs_ioapic_data {
@@ -1977,6 +2302,9 @@ void destroy_irq(unsigned int irq)
1977 2302
1978 dynamic_irq_cleanup(irq); 2303 dynamic_irq_cleanup(irq);
1979 2304
2305#ifdef CONFIG_INTR_REMAP
2306 free_irte(irq);
2307#endif
1980 spin_lock_irqsave(&vector_lock, flags); 2308 spin_lock_irqsave(&vector_lock, flags);
1981 __clear_irq_vector(irq); 2309 __clear_irq_vector(irq);
1982 spin_unlock_irqrestore(&vector_lock, flags); 2310 spin_unlock_irqrestore(&vector_lock, flags);
@@ -1995,11 +2323,42 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
1995 2323
1996 tmp = TARGET_CPUS; 2324 tmp = TARGET_CPUS;
1997 err = assign_irq_vector(irq, tmp); 2325 err = assign_irq_vector(irq, tmp);
1998 if (!err) { 2326 if (err)
1999 cpus_and(tmp, cfg->domain, tmp); 2327 return err;
2000 dest = cpu_mask_to_apicid(tmp); 2328
2329 cpus_and(tmp, cfg->domain, tmp);
2330 dest = cpu_mask_to_apicid(tmp);
2331
2332#ifdef CONFIG_INTR_REMAP
2333 if (irq_remapped(irq)) {
2334 struct irte irte;
2335 int ir_index;
2336 u16 sub_handle;
2337
2338 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
2339 BUG_ON(ir_index == -1);
2340
2341 memset (&irte, 0, sizeof(irte));
2342
2343 irte.present = 1;
2344 irte.dst_mode = INT_DEST_MODE;
2345 irte.trigger_mode = 0; /* edge */
2346 irte.dlvry_mode = INT_DELIVERY_MODE;
2347 irte.vector = cfg->vector;
2348 irte.dest_id = IRTE_DEST(dest);
2349
2350 modify_irte(irq, &irte);
2001 2351
2002 msg->address_hi = MSI_ADDR_BASE_HI; 2352 msg->address_hi = MSI_ADDR_BASE_HI;
2353 msg->data = sub_handle;
2354 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
2355 MSI_ADDR_IR_SHV |
2356 MSI_ADDR_IR_INDEX1(ir_index) |
2357 MSI_ADDR_IR_INDEX2(ir_index);
2358 } else
2359#endif
2360 {
2361 msg->address_hi = MSI_ADDR_BASE_HI;
2003 msg->address_lo = 2362 msg->address_lo =
2004 MSI_ADDR_BASE_LO | 2363 MSI_ADDR_BASE_LO |
2005 ((INT_DEST_MODE == 0) ? 2364 ((INT_DEST_MODE == 0) ?
@@ -2049,6 +2408,55 @@ static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2049 write_msi_msg(irq, &msg); 2408 write_msi_msg(irq, &msg);
2050 irq_desc[irq].affinity = mask; 2409 irq_desc[irq].affinity = mask;
2051} 2410}
2411
2412#ifdef CONFIG_INTR_REMAP
2413/*
2414 * Migrate the MSI irq to another cpumask. This migration is
2415 * done in the process context using interrupt-remapping hardware.
2416 */
2417static void ir_set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2418{
2419 struct irq_cfg *cfg = irq_cfg + irq;
2420 unsigned int dest;
2421 cpumask_t tmp, cleanup_mask;
2422 struct irte irte;
2423
2424 cpus_and(tmp, mask, cpu_online_map);
2425 if (cpus_empty(tmp))
2426 return;
2427
2428 if (get_irte(irq, &irte))
2429 return;
2430
2431 if (assign_irq_vector(irq, mask))
2432 return;
2433
2434 cpus_and(tmp, cfg->domain, mask);
2435 dest = cpu_mask_to_apicid(tmp);
2436
2437 irte.vector = cfg->vector;
2438 irte.dest_id = IRTE_DEST(dest);
2439
2440 /*
2441 * atomically update the IRTE with the new destination and vector.
2442 */
2443 modify_irte(irq, &irte);
2444
2445 /*
2446 * After this point, all the interrupts will start arriving
2447 * at the new destination. So, time to cleanup the previous
2448 * vector allocation.
2449 */
2450 if (cfg->move_in_progress) {
2451 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
2452 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
2453 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2454 cfg->move_in_progress = 0;
2455 }
2456
2457 irq_desc[irq].affinity = mask;
2458}
2459#endif
2052#endif /* CONFIG_SMP */ 2460#endif /* CONFIG_SMP */
2053 2461
2054/* 2462/*
@@ -2066,26 +2474,157 @@ static struct irq_chip msi_chip = {
2066 .retrigger = ioapic_retrigger_irq, 2474 .retrigger = ioapic_retrigger_irq,
2067}; 2475};
2068 2476
2069int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc) 2477#ifdef CONFIG_INTR_REMAP
2478static struct irq_chip msi_ir_chip = {
2479 .name = "IR-PCI-MSI",
2480 .unmask = unmask_msi_irq,
2481 .mask = mask_msi_irq,
2482 .ack = ack_x2apic_edge,
2483#ifdef CONFIG_SMP
2484 .set_affinity = ir_set_msi_irq_affinity,
2485#endif
2486 .retrigger = ioapic_retrigger_irq,
2487};
2488
2489/*
2490 * Map the PCI dev to the corresponding remapping hardware unit
2491 * and allocate 'nvec' consecutive interrupt-remapping table entries
2492 * in it.
2493 */
2494static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
2070{ 2495{
2496 struct intel_iommu *iommu;
2497 int index;
2498
2499 iommu = map_dev_to_ir(dev);
2500 if (!iommu) {
2501 printk(KERN_ERR
2502 "Unable to map PCI %s to iommu\n", pci_name(dev));
2503 return -ENOENT;
2504 }
2505
2506 index = alloc_irte(iommu, irq, nvec);
2507 if (index < 0) {
2508 printk(KERN_ERR
2509 "Unable to allocate %d IRTE for PCI %s\n", nvec,
2510 pci_name(dev));
2511 return -ENOSPC;
2512 }
2513 return index;
2514}
2515#endif
2516
2517static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc, int irq)
2518{
2519 int ret;
2071 struct msi_msg msg; 2520 struct msi_msg msg;
2521
2522 ret = msi_compose_msg(dev, irq, &msg);
2523 if (ret < 0)
2524 return ret;
2525
2526 set_irq_msi(irq, desc);
2527 write_msi_msg(irq, &msg);
2528
2529#ifdef CONFIG_INTR_REMAP
2530 if (irq_remapped(irq)) {
2531 struct irq_desc *desc = irq_desc + irq;
2532 /*
2533 * irq migration in process context
2534 */
2535 desc->status |= IRQ_MOVE_PCNTXT;
2536 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
2537 } else
2538#endif
2539 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2540
2541 return 0;
2542}
2543
2544int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2545{
2072 int irq, ret; 2546 int irq, ret;
2547
2073 irq = create_irq(); 2548 irq = create_irq();
2074 if (irq < 0) 2549 if (irq < 0)
2075 return irq; 2550 return irq;
2076 2551
2077 ret = msi_compose_msg(dev, irq, &msg); 2552#ifdef CONFIG_INTR_REMAP
2553 if (!intr_remapping_enabled)
2554 goto no_ir;
2555
2556 ret = msi_alloc_irte(dev, irq, 1);
2557 if (ret < 0)
2558 goto error;
2559no_ir:
2560#endif
2561 ret = setup_msi_irq(dev, desc, irq);
2078 if (ret < 0) { 2562 if (ret < 0) {
2079 destroy_irq(irq); 2563 destroy_irq(irq);
2080 return ret; 2564 return ret;
2081 } 2565 }
2566 return 0;
2082 2567
2083 set_irq_msi(irq, desc); 2568#ifdef CONFIG_INTR_REMAP
2084 write_msi_msg(irq, &msg); 2569error:
2570 destroy_irq(irq);
2571 return ret;
2572#endif
2573}
2085 2574
2086 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge"); 2575int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
2576{
2577 int irq, ret, sub_handle;
2578 struct msi_desc *desc;
2579#ifdef CONFIG_INTR_REMAP
2580 struct intel_iommu *iommu = 0;
2581 int index = 0;
2582#endif
2583
2584 sub_handle = 0;
2585 list_for_each_entry(desc, &dev->msi_list, list) {
2586 irq = create_irq();
2587 if (irq < 0)
2588 return irq;
2589#ifdef CONFIG_INTR_REMAP
2590 if (!intr_remapping_enabled)
2591 goto no_ir;
2087 2592
2593 if (!sub_handle) {
2594 /*
2595 * allocate the consecutive block of IRTE's
2596 * for 'nvec'
2597 */
2598 index = msi_alloc_irte(dev, irq, nvec);
2599 if (index < 0) {
2600 ret = index;
2601 goto error;
2602 }
2603 } else {
2604 iommu = map_dev_to_ir(dev);
2605 if (!iommu) {
2606 ret = -ENOENT;
2607 goto error;
2608 }
2609 /*
2610 * setup the mapping between the irq and the IRTE
2611 * base index, the sub_handle pointing to the
2612 * appropriate interrupt remap table entry.
2613 */
2614 set_irte_irq(irq, iommu, index, sub_handle);
2615 }
2616no_ir:
2617#endif
2618 ret = setup_msi_irq(dev, desc, irq);
2619 if (ret < 0)
2620 goto error;
2621 sub_handle++;
2622 }
2088 return 0; 2623 return 0;
2624
2625error:
2626 destroy_irq(irq);
2627 return ret;
2089} 2628}
2090 2629
2091void arch_teardown_msi_irq(unsigned int irq) 2630void arch_teardown_msi_irq(unsigned int irq)
@@ -2333,6 +2872,10 @@ void __init setup_ioapic_dest(void)
2333 setup_IO_APIC_irq(ioapic, pin, irq, 2872 setup_IO_APIC_irq(ioapic, pin, irq,
2334 irq_trigger(irq_entry), 2873 irq_trigger(irq_entry),
2335 irq_polarity(irq_entry)); 2874 irq_polarity(irq_entry));
2875#ifdef CONFIG_INTR_REMAP
2876 else if (intr_remapping_enabled)
2877 set_ir_ioapic_affinity_irq(irq, TARGET_CPUS);
2878#endif
2336 else 2879 else
2337 set_ioapic_affinity_irq(irq, TARGET_CPUS); 2880 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2338 } 2881 }
diff --git a/arch/x86/kernel/ioport.c b/arch/x86/kernel/ioport.c
index 50e5e4a31c85..191914302744 100644
--- a/arch/x86/kernel/ioport.c
+++ b/arch/x86/kernel/ioport.c
@@ -14,6 +14,7 @@
14#include <linux/slab.h> 14#include <linux/slab.h>
15#include <linux/thread_info.h> 15#include <linux/thread_info.h>
16#include <linux/syscalls.h> 16#include <linux/syscalls.h>
17#include <asm/syscalls.h>
17 18
18/* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */ 19/* Set EXTENT bits starting at BASE in BITMAP to value TURN_ON. */
19static void set_bitmap(unsigned long *bitmap, unsigned int base, 20static void set_bitmap(unsigned long *bitmap, unsigned int base,
diff --git a/arch/x86/kernel/ipi.c b/arch/x86/kernel/ipi.c
index 3f7537b669d3..f1c688e46f35 100644
--- a/arch/x86/kernel/ipi.c
+++ b/arch/x86/kernel/ipi.c
@@ -20,6 +20,8 @@
20 20
21#ifdef CONFIG_X86_32 21#ifdef CONFIG_X86_32
22#include <mach_apic.h> 22#include <mach_apic.h>
23#include <mach_ipi.h>
24
23/* 25/*
24 * the following functions deal with sending IPIs between CPUs. 26 * the following functions deal with sending IPIs between CPUs.
25 * 27 *
@@ -147,7 +149,6 @@ void send_IPI_mask_sequence(cpumask_t mask, int vector)
147} 149}
148 150
149/* must come after the send_IPI functions above for inlining */ 151/* must come after the send_IPI functions above for inlining */
150#include <mach_ipi.h>
151static int convert_apicid_to_cpu(int apic_id) 152static int convert_apicid_to_cpu(int apic_id)
152{ 153{
153 int i; 154 int i;
diff --git a/arch/x86/kernel/irq_32.c b/arch/x86/kernel/irq_32.c
index 1cf8c1fcc088..b71e02d42f4f 100644
--- a/arch/x86/kernel/irq_32.c
+++ b/arch/x86/kernel/irq_32.c
@@ -325,7 +325,7 @@ skip:
325 for_each_online_cpu(j) 325 for_each_online_cpu(j)
326 seq_printf(p, "%10u ", 326 seq_printf(p, "%10u ",
327 per_cpu(irq_stat,j).irq_call_count); 327 per_cpu(irq_stat,j).irq_call_count);
328 seq_printf(p, " function call interrupts\n"); 328 seq_printf(p, " Function call interrupts\n");
329 seq_printf(p, "TLB: "); 329 seq_printf(p, "TLB: ");
330 for_each_online_cpu(j) 330 for_each_online_cpu(j)
331 seq_printf(p, "%10u ", 331 seq_printf(p, "%10u ",
diff --git a/arch/x86/kernel/irq_64.c b/arch/x86/kernel/irq_64.c
index 1f78b238d8d2..f065fe9071b9 100644
--- a/arch/x86/kernel/irq_64.c
+++ b/arch/x86/kernel/irq_64.c
@@ -129,7 +129,7 @@ skip:
129 seq_printf(p, "CAL: "); 129 seq_printf(p, "CAL: ");
130 for_each_online_cpu(j) 130 for_each_online_cpu(j)
131 seq_printf(p, "%10u ", cpu_pda(j)->irq_call_count); 131 seq_printf(p, "%10u ", cpu_pda(j)->irq_call_count);
132 seq_printf(p, " function call interrupts\n"); 132 seq_printf(p, " Function call interrupts\n");
133 seq_printf(p, "TLB: "); 133 seq_printf(p, "TLB: ");
134 for_each_online_cpu(j) 134 for_each_online_cpu(j)
135 seq_printf(p, "%10u ", cpu_pda(j)->irq_tlb_count); 135 seq_printf(p, "%10u ", cpu_pda(j)->irq_tlb_count);
diff --git a/arch/x86/kernel/irqinit_32.c b/arch/x86/kernel/irqinit_32.c
index d66914287ee1..9200a1e2752d 100644
--- a/arch/x86/kernel/irqinit_32.c
+++ b/arch/x86/kernel/irqinit_32.c
@@ -74,6 +74,15 @@ void __init init_ISA_irqs (void)
74 } 74 }
75} 75}
76 76
77/*
78 * IRQ2 is cascade interrupt to second interrupt controller
79 */
80static struct irqaction irq2 = {
81 .handler = no_action,
82 .mask = CPU_MASK_NONE,
83 .name = "cascade",
84};
85
77/* Overridden in paravirt.c */ 86/* Overridden in paravirt.c */
78void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ"))); 87void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
79 88
@@ -98,6 +107,46 @@ void __init native_init_IRQ(void)
98 set_intr_gate(vector, interrupt[i]); 107 set_intr_gate(vector, interrupt[i]);
99 } 108 }
100 109
110#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_SMP)
111 /*
112 * IRQ0 must be given a fixed assignment and initialized,
113 * because it's used before the IO-APIC is set up.
114 */
115 set_intr_gate(FIRST_DEVICE_VECTOR, interrupt[0]);
116
117 /*
118 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
119 * IPI, driven by wakeup.
120 */
121 alloc_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
122
123 /* IPI for invalidation */
124 alloc_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
125
126 /* IPI for generic function call */
127 alloc_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
128
129 /* IPI for single call function */
130 set_intr_gate(CALL_FUNCTION_SINGLE_VECTOR, call_function_single_interrupt);
131#endif
132
133#ifdef CONFIG_X86_LOCAL_APIC
134 /* self generated IPI for local APIC timer */
135 alloc_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
136
137 /* IPI vectors for APIC spurious and error interrupts */
138 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
139 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
140#endif
141
142#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86_MCE_P4THERMAL)
143 /* thermal monitor LVT interrupt */
144 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
145#endif
146
147 if (!acpi_ioapic)
148 setup_irq(2, &irq2);
149
101 /* setup after call gates are initialised (usually add in 150 /* setup after call gates are initialised (usually add in
102 * the architecture specific gates) 151 * the architecture specific gates)
103 */ 152 */
diff --git a/arch/x86/kernel/irqinit_64.c b/arch/x86/kernel/irqinit_64.c
index 1f26fd9ec4f4..5b5be9d43c2a 100644
--- a/arch/x86/kernel/irqinit_64.c
+++ b/arch/x86/kernel/irqinit_64.c
@@ -135,7 +135,7 @@ DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
135 [IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1 135 [IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
136}; 136};
137 137
138static void __init init_ISA_irqs (void) 138void __init init_ISA_irqs(void)
139{ 139{
140 int i; 140 int i;
141 141
@@ -164,22 +164,8 @@ static void __init init_ISA_irqs (void)
164 164
165void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ"))); 165void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
166 166
167void __init native_init_IRQ(void) 167static void __init smp_intr_init(void)
168{ 168{
169 int i;
170
171 init_ISA_irqs();
172 /*
173 * Cover the whole vector space, no vector can escape
174 * us. (some of these will be overridden and become
175 * 'special' SMP interrupts)
176 */
177 for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
178 int vector = FIRST_EXTERNAL_VECTOR + i;
179 if (vector != IA32_SYSCALL_VECTOR)
180 set_intr_gate(vector, interrupt[i]);
181 }
182
183#ifdef CONFIG_SMP 169#ifdef CONFIG_SMP
184 /* 170 /*
185 * The reschedule interrupt is a CPU-to-CPU reschedule-helper 171 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
@@ -207,6 +193,12 @@ void __init native_init_IRQ(void)
207 /* Low priority IPI to cleanup after moving an irq */ 193 /* Low priority IPI to cleanup after moving an irq */
208 set_intr_gate(IRQ_MOVE_CLEANUP_VECTOR, irq_move_cleanup_interrupt); 194 set_intr_gate(IRQ_MOVE_CLEANUP_VECTOR, irq_move_cleanup_interrupt);
209#endif 195#endif
196}
197
198static void __init apic_intr_init(void)
199{
200 smp_intr_init();
201
210 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt); 202 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
211 alloc_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt); 203 alloc_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt);
212 204
@@ -216,6 +208,25 @@ void __init native_init_IRQ(void)
216 /* IPI vectors for APIC spurious and error interrupts */ 208 /* IPI vectors for APIC spurious and error interrupts */
217 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt); 209 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
218 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt); 210 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
211}
212
213void __init native_init_IRQ(void)
214{
215 int i;
216
217 init_ISA_irqs();
218 /*
219 * Cover the whole vector space, no vector can escape
220 * us. (some of these will be overridden and become
221 * 'special' SMP interrupts)
222 */
223 for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
224 int vector = FIRST_EXTERNAL_VECTOR + i;
225 if (vector != IA32_SYSCALL_VECTOR)
226 set_intr_gate(vector, interrupt[i]);
227 }
228
229 apic_intr_init();
219 230
220 if (!acpi_ioapic) 231 if (!acpi_ioapic)
221 setup_irq(2, &irq2); 232 setup_irq(2, &irq2);
diff --git a/arch/x86/kernel/k8.c b/arch/x86/kernel/k8.c
index 7377ccb21335..304d8bad6559 100644
--- a/arch/x86/kernel/k8.c
+++ b/arch/x86/kernel/k8.c
@@ -16,8 +16,9 @@ EXPORT_SYMBOL(num_k8_northbridges);
16static u32 *flush_words; 16static u32 *flush_words;
17 17
18struct pci_device_id k8_nb_ids[] = { 18struct pci_device_id k8_nb_ids[] = {
19 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1103) }, 19 { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB_MISC) },
20 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1203) }, 20 { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_10H_NB_MISC) },
21 { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_11H_NB_MISC) },
21 {} 22 {}
22}; 23};
23EXPORT_SYMBOL(k8_nb_ids); 24EXPORT_SYMBOL(k8_nb_ids);
diff --git a/arch/x86/kernel/kdebugfs.c b/arch/x86/kernel/kdebugfs.c
index f2d43bc75514..ff7d3b0124f1 100644
--- a/arch/x86/kernel/kdebugfs.c
+++ b/arch/x86/kernel/kdebugfs.c
@@ -139,6 +139,7 @@ static int __init create_setup_data_nodes(struct dentry *parent)
139 if (PageHighMem(pg)) { 139 if (PageHighMem(pg)) {
140 data = ioremap_cache(pa_data, sizeof(*data)); 140 data = ioremap_cache(pa_data, sizeof(*data));
141 if (!data) { 141 if (!data) {
142 kfree(node);
142 error = -ENXIO; 143 error = -ENXIO;
143 goto err_dir; 144 goto err_dir;
144 } 145 }
diff --git a/arch/x86/kernel/kgdb.c b/arch/x86/kernel/kgdb.c
index f47f0eb886b8..10435a120d22 100644
--- a/arch/x86/kernel/kgdb.c
+++ b/arch/x86/kernel/kgdb.c
@@ -69,6 +69,9 @@ static int gdb_x86vector = -1;
69 */ 69 */
70void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs) 70void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs)
71{ 71{
72#ifndef CONFIG_X86_32
73 u32 *gdb_regs32 = (u32 *)gdb_regs;
74#endif
72 gdb_regs[GDB_AX] = regs->ax; 75 gdb_regs[GDB_AX] = regs->ax;
73 gdb_regs[GDB_BX] = regs->bx; 76 gdb_regs[GDB_BX] = regs->bx;
74 gdb_regs[GDB_CX] = regs->cx; 77 gdb_regs[GDB_CX] = regs->cx;
@@ -76,9 +79,9 @@ void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs)
76 gdb_regs[GDB_SI] = regs->si; 79 gdb_regs[GDB_SI] = regs->si;
77 gdb_regs[GDB_DI] = regs->di; 80 gdb_regs[GDB_DI] = regs->di;
78 gdb_regs[GDB_BP] = regs->bp; 81 gdb_regs[GDB_BP] = regs->bp;
79 gdb_regs[GDB_PS] = regs->flags;
80 gdb_regs[GDB_PC] = regs->ip; 82 gdb_regs[GDB_PC] = regs->ip;
81#ifdef CONFIG_X86_32 83#ifdef CONFIG_X86_32
84 gdb_regs[GDB_PS] = regs->flags;
82 gdb_regs[GDB_DS] = regs->ds; 85 gdb_regs[GDB_DS] = regs->ds;
83 gdb_regs[GDB_ES] = regs->es; 86 gdb_regs[GDB_ES] = regs->es;
84 gdb_regs[GDB_CS] = regs->cs; 87 gdb_regs[GDB_CS] = regs->cs;
@@ -94,6 +97,9 @@ void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs)
94 gdb_regs[GDB_R13] = regs->r13; 97 gdb_regs[GDB_R13] = regs->r13;
95 gdb_regs[GDB_R14] = regs->r14; 98 gdb_regs[GDB_R14] = regs->r14;
96 gdb_regs[GDB_R15] = regs->r15; 99 gdb_regs[GDB_R15] = regs->r15;
100 gdb_regs32[GDB_PS] = regs->flags;
101 gdb_regs32[GDB_CS] = regs->cs;
102 gdb_regs32[GDB_SS] = regs->ss;
97#endif 103#endif
98 gdb_regs[GDB_SP] = regs->sp; 104 gdb_regs[GDB_SP] = regs->sp;
99} 105}
@@ -112,6 +118,9 @@ void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs)
112 */ 118 */
113void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *p) 119void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *p)
114{ 120{
121#ifndef CONFIG_X86_32
122 u32 *gdb_regs32 = (u32 *)gdb_regs;
123#endif
115 gdb_regs[GDB_AX] = 0; 124 gdb_regs[GDB_AX] = 0;
116 gdb_regs[GDB_BX] = 0; 125 gdb_regs[GDB_BX] = 0;
117 gdb_regs[GDB_CX] = 0; 126 gdb_regs[GDB_CX] = 0;
@@ -129,8 +138,10 @@ void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *p)
129 gdb_regs[GDB_FS] = 0xFFFF; 138 gdb_regs[GDB_FS] = 0xFFFF;
130 gdb_regs[GDB_GS] = 0xFFFF; 139 gdb_regs[GDB_GS] = 0xFFFF;
131#else 140#else
132 gdb_regs[GDB_PS] = *(unsigned long *)(p->thread.sp + 8); 141 gdb_regs32[GDB_PS] = *(unsigned long *)(p->thread.sp + 8);
133 gdb_regs[GDB_PC] = 0; 142 gdb_regs32[GDB_CS] = __KERNEL_CS;
143 gdb_regs32[GDB_SS] = __KERNEL_DS;
144 gdb_regs[GDB_PC] = p->thread.ip;
134 gdb_regs[GDB_R8] = 0; 145 gdb_regs[GDB_R8] = 0;
135 gdb_regs[GDB_R9] = 0; 146 gdb_regs[GDB_R9] = 0;
136 gdb_regs[GDB_R10] = 0; 147 gdb_regs[GDB_R10] = 0;
@@ -153,6 +164,9 @@ void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *p)
153 */ 164 */
154void gdb_regs_to_pt_regs(unsigned long *gdb_regs, struct pt_regs *regs) 165void gdb_regs_to_pt_regs(unsigned long *gdb_regs, struct pt_regs *regs)
155{ 166{
167#ifndef CONFIG_X86_32
168 u32 *gdb_regs32 = (u32 *)gdb_regs;
169#endif
156 regs->ax = gdb_regs[GDB_AX]; 170 regs->ax = gdb_regs[GDB_AX];
157 regs->bx = gdb_regs[GDB_BX]; 171 regs->bx = gdb_regs[GDB_BX];
158 regs->cx = gdb_regs[GDB_CX]; 172 regs->cx = gdb_regs[GDB_CX];
@@ -160,9 +174,9 @@ void gdb_regs_to_pt_regs(unsigned long *gdb_regs, struct pt_regs *regs)
160 regs->si = gdb_regs[GDB_SI]; 174 regs->si = gdb_regs[GDB_SI];
161 regs->di = gdb_regs[GDB_DI]; 175 regs->di = gdb_regs[GDB_DI];
162 regs->bp = gdb_regs[GDB_BP]; 176 regs->bp = gdb_regs[GDB_BP];
163 regs->flags = gdb_regs[GDB_PS];
164 regs->ip = gdb_regs[GDB_PC]; 177 regs->ip = gdb_regs[GDB_PC];
165#ifdef CONFIG_X86_32 178#ifdef CONFIG_X86_32
179 regs->flags = gdb_regs[GDB_PS];
166 regs->ds = gdb_regs[GDB_DS]; 180 regs->ds = gdb_regs[GDB_DS];
167 regs->es = gdb_regs[GDB_ES]; 181 regs->es = gdb_regs[GDB_ES];
168 regs->cs = gdb_regs[GDB_CS]; 182 regs->cs = gdb_regs[GDB_CS];
@@ -175,6 +189,9 @@ void gdb_regs_to_pt_regs(unsigned long *gdb_regs, struct pt_regs *regs)
175 regs->r13 = gdb_regs[GDB_R13]; 189 regs->r13 = gdb_regs[GDB_R13];
176 regs->r14 = gdb_regs[GDB_R14]; 190 regs->r14 = gdb_regs[GDB_R14];
177 regs->r15 = gdb_regs[GDB_R15]; 191 regs->r15 = gdb_regs[GDB_R15];
192 regs->flags = gdb_regs32[GDB_PS];
193 regs->cs = gdb_regs32[GDB_CS];
194 regs->ss = gdb_regs32[GDB_SS];
178#endif 195#endif
179} 196}
180 197
@@ -378,10 +395,8 @@ int kgdb_arch_handle_exception(int e_vector, int signo, int err_code,
378 if (remcomInBuffer[0] == 's') { 395 if (remcomInBuffer[0] == 's') {
379 linux_regs->flags |= X86_EFLAGS_TF; 396 linux_regs->flags |= X86_EFLAGS_TF;
380 kgdb_single_step = 1; 397 kgdb_single_step = 1;
381 if (kgdb_contthread) { 398 atomic_set(&kgdb_cpu_doing_single_step,
382 atomic_set(&kgdb_cpu_doing_single_step, 399 raw_smp_processor_id());
383 raw_smp_processor_id());
384 }
385 } 400 }
386 401
387 get_debugreg(dr6, 6); 402 get_debugreg(dr6, 6);
@@ -440,12 +455,7 @@ static int __kgdb_notify(struct die_args *args, unsigned long cmd)
440 return NOTIFY_DONE; 455 return NOTIFY_DONE;
441 456
442 case DIE_NMI_IPI: 457 case DIE_NMI_IPI:
443 if (atomic_read(&kgdb_active) != -1) { 458 /* Just ignore, we will handle the roundup on DIE_NMI. */
444 /* KGDB CPU roundup */
445 kgdb_nmicallback(raw_smp_processor_id(), regs);
446 was_in_debug_nmi[raw_smp_processor_id()] = 1;
447 touch_nmi_watchdog();
448 }
449 return NOTIFY_DONE; 459 return NOTIFY_DONE;
450 460
451 case DIE_NMIUNKNOWN: 461 case DIE_NMIUNKNOWN:
@@ -466,9 +476,15 @@ static int __kgdb_notify(struct die_args *args, unsigned long cmd)
466 476
467 case DIE_DEBUG: 477 case DIE_DEBUG:
468 if (atomic_read(&kgdb_cpu_doing_single_step) == 478 if (atomic_read(&kgdb_cpu_doing_single_step) ==
469 raw_smp_processor_id() && 479 raw_smp_processor_id()) {
470 user_mode(regs)) 480 if (user_mode(regs))
471 return single_step_cont(regs, args); 481 return single_step_cont(regs, args);
482 break;
483 } else if (test_thread_flag(TIF_SINGLESTEP))
484 /* This means a user thread is single stepping
485 * a system call which should be ignored
486 */
487 return NOTIFY_DONE;
472 /* fall through */ 488 /* fall through */
473 default: 489 default:
474 if (user_mode(regs)) 490 if (user_mode(regs))
diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
index 8b7a3cf37d2b..478bca986eca 100644
--- a/arch/x86/kernel/kvm.c
+++ b/arch/x86/kernel/kvm.c
@@ -178,7 +178,7 @@ static void kvm_flush_tlb(void)
178 kvm_deferred_mmu_op(&ftlb, sizeof ftlb); 178 kvm_deferred_mmu_op(&ftlb, sizeof ftlb);
179} 179}
180 180
181static void kvm_release_pt(u32 pfn) 181static void kvm_release_pt(unsigned long pfn)
182{ 182{
183 struct kvm_mmu_op_release_pt rpt = { 183 struct kvm_mmu_op_release_pt rpt = {
184 .header.op = KVM_MMU_OP_RELEASE_PT, 184 .header.op = KVM_MMU_OP_RELEASE_PT,
diff --git a/arch/x86/kernel/kvmclock.c b/arch/x86/kernel/kvmclock.c
index d02def06ca91..774ac4991568 100644
--- a/arch/x86/kernel/kvmclock.c
+++ b/arch/x86/kernel/kvmclock.c
@@ -78,6 +78,34 @@ static cycle_t kvm_clock_read(void)
78 return ret; 78 return ret;
79} 79}
80 80
81/*
82 * If we don't do that, there is the possibility that the guest
83 * will calibrate under heavy load - thus, getting a lower lpj -
84 * and execute the delays themselves without load. This is wrong,
85 * because no delay loop can finish beforehand.
86 * Any heuristics is subject to fail, because ultimately, a large
87 * poll of guests can be running and trouble each other. So we preset
88 * lpj here
89 */
90static unsigned long kvm_get_tsc_khz(void)
91{
92 return preset_lpj;
93}
94
95static void kvm_get_preset_lpj(void)
96{
97 struct pvclock_vcpu_time_info *src;
98 unsigned long khz;
99 u64 lpj;
100
101 src = &per_cpu(hv_clock, 0);
102 khz = pvclock_tsc_khz(src);
103
104 lpj = ((u64)khz * 1000);
105 do_div(lpj, HZ);
106 preset_lpj = lpj;
107}
108
81static struct clocksource kvm_clock = { 109static struct clocksource kvm_clock = {
82 .name = "kvm-clock", 110 .name = "kvm-clock",
83 .read = kvm_clock_read, 111 .read = kvm_clock_read,
@@ -153,6 +181,7 @@ void __init kvmclock_init(void)
153 pv_time_ops.get_wallclock = kvm_get_wallclock; 181 pv_time_ops.get_wallclock = kvm_get_wallclock;
154 pv_time_ops.set_wallclock = kvm_set_wallclock; 182 pv_time_ops.set_wallclock = kvm_set_wallclock;
155 pv_time_ops.sched_clock = kvm_clock_read; 183 pv_time_ops.sched_clock = kvm_clock_read;
184 pv_time_ops.get_tsc_khz = kvm_get_tsc_khz;
156#ifdef CONFIG_X86_LOCAL_APIC 185#ifdef CONFIG_X86_LOCAL_APIC
157 pv_apic_ops.setup_secondary_clock = kvm_setup_secondary_clock; 186 pv_apic_ops.setup_secondary_clock = kvm_setup_secondary_clock;
158#endif 187#endif
@@ -163,6 +192,7 @@ void __init kvmclock_init(void)
163#ifdef CONFIG_KEXEC 192#ifdef CONFIG_KEXEC
164 machine_ops.crash_shutdown = kvm_crash_shutdown; 193 machine_ops.crash_shutdown = kvm_crash_shutdown;
165#endif 194#endif
195 kvm_get_preset_lpj();
166 clocksource_register(&kvm_clock); 196 clocksource_register(&kvm_clock);
167 } 197 }
168} 198}
diff --git a/arch/x86/kernel/ldt.c b/arch/x86/kernel/ldt.c
index b68e21f06f4f..eee32b43fee3 100644
--- a/arch/x86/kernel/ldt.c
+++ b/arch/x86/kernel/ldt.c
@@ -18,6 +18,7 @@
18#include <asm/ldt.h> 18#include <asm/ldt.h>
19#include <asm/desc.h> 19#include <asm/desc.h>
20#include <asm/mmu_context.h> 20#include <asm/mmu_context.h>
21#include <asm/syscalls.h>
21 22
22#ifdef CONFIG_SMP 23#ifdef CONFIG_SMP
23static void flush_ldt(void *current_mm) 24static void flush_ldt(void *current_mm)
@@ -51,6 +52,8 @@ static int alloc_ldt(mm_context_t *pc, int mincount, int reload)
51 memset(newldt + oldsize * LDT_ENTRY_SIZE, 0, 52 memset(newldt + oldsize * LDT_ENTRY_SIZE, 0,
52 (mincount - oldsize) * LDT_ENTRY_SIZE); 53 (mincount - oldsize) * LDT_ENTRY_SIZE);
53 54
55 paravirt_alloc_ldt(newldt, mincount);
56
54#ifdef CONFIG_X86_64 57#ifdef CONFIG_X86_64
55 /* CHECKME: Do we really need this ? */ 58 /* CHECKME: Do we really need this ? */
56 wmb(); 59 wmb();
@@ -73,6 +76,7 @@ static int alloc_ldt(mm_context_t *pc, int mincount, int reload)
73#endif 76#endif
74 } 77 }
75 if (oldsize) { 78 if (oldsize) {
79 paravirt_free_ldt(oldldt, oldsize);
76 if (oldsize * LDT_ENTRY_SIZE > PAGE_SIZE) 80 if (oldsize * LDT_ENTRY_SIZE > PAGE_SIZE)
77 vfree(oldldt); 81 vfree(oldldt);
78 else 82 else
@@ -84,10 +88,13 @@ static int alloc_ldt(mm_context_t *pc, int mincount, int reload)
84static inline int copy_ldt(mm_context_t *new, mm_context_t *old) 88static inline int copy_ldt(mm_context_t *new, mm_context_t *old)
85{ 89{
86 int err = alloc_ldt(new, old->size, 0); 90 int err = alloc_ldt(new, old->size, 0);
91 int i;
87 92
88 if (err < 0) 93 if (err < 0)
89 return err; 94 return err;
90 memcpy(new->ldt, old->ldt, old->size * LDT_ENTRY_SIZE); 95
96 for(i = 0; i < old->size; i++)
97 write_ldt_entry(new->ldt, i, old->ldt + i * LDT_ENTRY_SIZE);
91 return 0; 98 return 0;
92} 99}
93 100
@@ -124,6 +131,7 @@ void destroy_context(struct mm_struct *mm)
124 if (mm == current->active_mm) 131 if (mm == current->active_mm)
125 clear_LDT(); 132 clear_LDT();
126#endif 133#endif
134 paravirt_free_ldt(mm->context.ldt, mm->context.size);
127 if (mm->context.size * LDT_ENTRY_SIZE > PAGE_SIZE) 135 if (mm->context.size * LDT_ENTRY_SIZE > PAGE_SIZE)
128 vfree(mm->context.ldt); 136 vfree(mm->context.ldt);
129 else 137 else
diff --git a/arch/x86/kernel/microcode.c b/arch/x86/kernel/microcode.c
deleted file mode 100644
index 652fa5c38ebe..000000000000
--- a/arch/x86/kernel/microcode.c
+++ /dev/null
@@ -1,853 +0,0 @@
1/*
2 * Intel CPU Microcode Update Driver for Linux
3 *
4 * Copyright (C) 2000-2006 Tigran Aivazian <tigran@aivazian.fsnet.co.uk>
5 * 2006 Shaohua Li <shaohua.li@intel.com>
6 *
7 * This driver allows to upgrade microcode on Intel processors
8 * belonging to IA-32 family - PentiumPro, Pentium II,
9 * Pentium III, Xeon, Pentium 4, etc.
10 *
11 * Reference: Section 8.11 of Volume 3a, IA-32 Intel? Architecture
12 * Software Developer's Manual
13 * Order Number 253668 or free download from:
14 *
15 * http://developer.intel.com/design/pentium4/manuals/253668.htm
16 *
17 * For more information, go to http://www.urbanmyth.org/microcode
18 *
19 * This program is free software; you can redistribute it and/or
20 * modify it under the terms of the GNU General Public License
21 * as published by the Free Software Foundation; either version
22 * 2 of the License, or (at your option) any later version.
23 *
24 * 1.0 16 Feb 2000, Tigran Aivazian <tigran@sco.com>
25 * Initial release.
26 * 1.01 18 Feb 2000, Tigran Aivazian <tigran@sco.com>
27 * Added read() support + cleanups.
28 * 1.02 21 Feb 2000, Tigran Aivazian <tigran@sco.com>
29 * Added 'device trimming' support. open(O_WRONLY) zeroes
30 * and frees the saved copy of applied microcode.
31 * 1.03 29 Feb 2000, Tigran Aivazian <tigran@sco.com>
32 * Made to use devfs (/dev/cpu/microcode) + cleanups.
33 * 1.04 06 Jun 2000, Simon Trimmer <simon@veritas.com>
34 * Added misc device support (now uses both devfs and misc).
35 * Added MICROCODE_IOCFREE ioctl to clear memory.
36 * 1.05 09 Jun 2000, Simon Trimmer <simon@veritas.com>
37 * Messages for error cases (non Intel & no suitable microcode).
38 * 1.06 03 Aug 2000, Tigran Aivazian <tigran@veritas.com>
39 * Removed ->release(). Removed exclusive open and status bitmap.
40 * Added microcode_rwsem to serialize read()/write()/ioctl().
41 * Removed global kernel lock usage.
42 * 1.07 07 Sep 2000, Tigran Aivazian <tigran@veritas.com>
43 * Write 0 to 0x8B msr and then cpuid before reading revision,
44 * so that it works even if there were no update done by the
45 * BIOS. Otherwise, reading from 0x8B gives junk (which happened
46 * to be 0 on my machine which is why it worked even when I
47 * disabled update by the BIOS)
48 * Thanks to Eric W. Biederman <ebiederman@lnxi.com> for the fix.
49 * 1.08 11 Dec 2000, Richard Schaal <richard.schaal@intel.com> and
50 * Tigran Aivazian <tigran@veritas.com>
51 * Intel Pentium 4 processor support and bugfixes.
52 * 1.09 30 Oct 2001, Tigran Aivazian <tigran@veritas.com>
53 * Bugfix for HT (Hyper-Threading) enabled processors
54 * whereby processor resources are shared by all logical processors
55 * in a single CPU package.
56 * 1.10 28 Feb 2002 Asit K Mallick <asit.k.mallick@intel.com> and
57 * Tigran Aivazian <tigran@veritas.com>,
58 * Serialize updates as required on HT processors due to speculative
59 * nature of implementation.
60 * 1.11 22 Mar 2002 Tigran Aivazian <tigran@veritas.com>
61 * Fix the panic when writing zero-length microcode chunk.
62 * 1.12 29 Sep 2003 Nitin Kamble <nitin.a.kamble@intel.com>,
63 * Jun Nakajima <jun.nakajima@intel.com>
64 * Support for the microcode updates in the new format.
65 * 1.13 10 Oct 2003 Tigran Aivazian <tigran@veritas.com>
66 * Removed ->read() method and obsoleted MICROCODE_IOCFREE ioctl
67 * because we no longer hold a copy of applied microcode
68 * in kernel memory.
69 * 1.14 25 Jun 2004 Tigran Aivazian <tigran@veritas.com>
70 * Fix sigmatch() macro to handle old CPUs with pf == 0.
71 * Thanks to Stuart Swales for pointing out this bug.
72 */
73
74//#define DEBUG /* pr_debug */
75#include <linux/capability.h>
76#include <linux/kernel.h>
77#include <linux/init.h>
78#include <linux/sched.h>
79#include <linux/smp_lock.h>
80#include <linux/cpumask.h>
81#include <linux/module.h>
82#include <linux/slab.h>
83#include <linux/vmalloc.h>
84#include <linux/miscdevice.h>
85#include <linux/spinlock.h>
86#include <linux/mm.h>
87#include <linux/fs.h>
88#include <linux/mutex.h>
89#include <linux/cpu.h>
90#include <linux/firmware.h>
91#include <linux/platform_device.h>
92
93#include <asm/msr.h>
94#include <asm/uaccess.h>
95#include <asm/processor.h>
96
97MODULE_DESCRIPTION("Intel CPU (IA-32) Microcode Update Driver");
98MODULE_AUTHOR("Tigran Aivazian <tigran@aivazian.fsnet.co.uk>");
99MODULE_LICENSE("GPL");
100
101#define MICROCODE_VERSION "1.14a"
102
103#define DEFAULT_UCODE_DATASIZE (2000) /* 2000 bytes */
104#define MC_HEADER_SIZE (sizeof (microcode_header_t)) /* 48 bytes */
105#define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE) /* 2048 bytes */
106#define EXT_HEADER_SIZE (sizeof (struct extended_sigtable)) /* 20 bytes */
107#define EXT_SIGNATURE_SIZE (sizeof (struct extended_signature)) /* 12 bytes */
108#define DWSIZE (sizeof (u32))
109#define get_totalsize(mc) \
110 (((microcode_t *)mc)->hdr.totalsize ? \
111 ((microcode_t *)mc)->hdr.totalsize : DEFAULT_UCODE_TOTALSIZE)
112#define get_datasize(mc) \
113 (((microcode_t *)mc)->hdr.datasize ? \
114 ((microcode_t *)mc)->hdr.datasize : DEFAULT_UCODE_DATASIZE)
115
116#define sigmatch(s1, s2, p1, p2) \
117 (((s1) == (s2)) && (((p1) & (p2)) || (((p1) == 0) && ((p2) == 0))))
118
119#define exttable_size(et) ((et)->count * EXT_SIGNATURE_SIZE + EXT_HEADER_SIZE)
120
121/* serialize access to the physical write to MSR 0x79 */
122static DEFINE_SPINLOCK(microcode_update_lock);
123
124/* no concurrent ->write()s are allowed on /dev/cpu/microcode */
125static DEFINE_MUTEX(microcode_mutex);
126
127static struct ucode_cpu_info {
128 int valid;
129 unsigned int sig;
130 unsigned int pf;
131 unsigned int rev;
132 microcode_t *mc;
133} ucode_cpu_info[NR_CPUS];
134
135static void collect_cpu_info(int cpu_num)
136{
137 struct cpuinfo_x86 *c = &cpu_data(cpu_num);
138 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num;
139 unsigned int val[2];
140
141 /* We should bind the task to the CPU */
142 BUG_ON(raw_smp_processor_id() != cpu_num);
143 uci->pf = uci->rev = 0;
144 uci->mc = NULL;
145 uci->valid = 1;
146
147 if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 < 6 ||
148 cpu_has(c, X86_FEATURE_IA64)) {
149 printk(KERN_ERR "microcode: CPU%d not a capable Intel "
150 "processor\n", cpu_num);
151 uci->valid = 0;
152 return;
153 }
154
155 uci->sig = cpuid_eax(0x00000001);
156
157 if ((c->x86_model >= 5) || (c->x86 > 6)) {
158 /* get processor flags from MSR 0x17 */
159 rdmsr(MSR_IA32_PLATFORM_ID, val[0], val[1]);
160 uci->pf = 1 << ((val[1] >> 18) & 7);
161 }
162
163 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
164 /* see notes above for revision 1.07. Apparent chip bug */
165 sync_core();
166 /* get the current revision from MSR 0x8B */
167 rdmsr(MSR_IA32_UCODE_REV, val[0], uci->rev);
168 pr_debug("microcode: collect_cpu_info : sig=0x%x, pf=0x%x, rev=0x%x\n",
169 uci->sig, uci->pf, uci->rev);
170}
171
172static inline int microcode_update_match(int cpu_num,
173 microcode_header_t *mc_header, int sig, int pf)
174{
175 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num;
176
177 if (!sigmatch(sig, uci->sig, pf, uci->pf)
178 || mc_header->rev <= uci->rev)
179 return 0;
180 return 1;
181}
182
183static int microcode_sanity_check(void *mc)
184{
185 microcode_header_t *mc_header = mc;
186 struct extended_sigtable *ext_header = NULL;
187 struct extended_signature *ext_sig;
188 unsigned long total_size, data_size, ext_table_size;
189 int sum, orig_sum, ext_sigcount = 0, i;
190
191 total_size = get_totalsize(mc_header);
192 data_size = get_datasize(mc_header);
193 if (data_size + MC_HEADER_SIZE > total_size) {
194 printk(KERN_ERR "microcode: error! "
195 "Bad data size in microcode data file\n");
196 return -EINVAL;
197 }
198
199 if (mc_header->ldrver != 1 || mc_header->hdrver != 1) {
200 printk(KERN_ERR "microcode: error! "
201 "Unknown microcode update format\n");
202 return -EINVAL;
203 }
204 ext_table_size = total_size - (MC_HEADER_SIZE + data_size);
205 if (ext_table_size) {
206 if ((ext_table_size < EXT_HEADER_SIZE)
207 || ((ext_table_size - EXT_HEADER_SIZE) % EXT_SIGNATURE_SIZE)) {
208 printk(KERN_ERR "microcode: error! "
209 "Small exttable size in microcode data file\n");
210 return -EINVAL;
211 }
212 ext_header = mc + MC_HEADER_SIZE + data_size;
213 if (ext_table_size != exttable_size(ext_header)) {
214 printk(KERN_ERR "microcode: error! "
215 "Bad exttable size in microcode data file\n");
216 return -EFAULT;
217 }
218 ext_sigcount = ext_header->count;
219 }
220
221 /* check extended table checksum */
222 if (ext_table_size) {
223 int ext_table_sum = 0;
224 int *ext_tablep = (int *)ext_header;
225
226 i = ext_table_size / DWSIZE;
227 while (i--)
228 ext_table_sum += ext_tablep[i];
229 if (ext_table_sum) {
230 printk(KERN_WARNING "microcode: aborting, "
231 "bad extended signature table checksum\n");
232 return -EINVAL;
233 }
234 }
235
236 /* calculate the checksum */
237 orig_sum = 0;
238 i = (MC_HEADER_SIZE + data_size) / DWSIZE;
239 while (i--)
240 orig_sum += ((int *)mc)[i];
241 if (orig_sum) {
242 printk(KERN_ERR "microcode: aborting, bad checksum\n");
243 return -EINVAL;
244 }
245 if (!ext_table_size)
246 return 0;
247 /* check extended signature checksum */
248 for (i = 0; i < ext_sigcount; i++) {
249 ext_sig = (void *)ext_header + EXT_HEADER_SIZE +
250 EXT_SIGNATURE_SIZE * i;
251 sum = orig_sum
252 - (mc_header->sig + mc_header->pf + mc_header->cksum)
253 + (ext_sig->sig + ext_sig->pf + ext_sig->cksum);
254 if (sum) {
255 printk(KERN_ERR "microcode: aborting, bad checksum\n");
256 return -EINVAL;
257 }
258 }
259 return 0;
260}
261
262/*
263 * return 0 - no update found
264 * return 1 - found update
265 * return < 0 - error
266 */
267static int get_maching_microcode(void *mc, int cpu)
268{
269 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
270 microcode_header_t *mc_header = mc;
271 struct extended_sigtable *ext_header;
272 unsigned long total_size = get_totalsize(mc_header);
273 int ext_sigcount, i;
274 struct extended_signature *ext_sig;
275 void *new_mc;
276
277 if (microcode_update_match(cpu, mc_header,
278 mc_header->sig, mc_header->pf))
279 goto find;
280
281 if (total_size <= get_datasize(mc_header) + MC_HEADER_SIZE)
282 return 0;
283
284 ext_header = mc + get_datasize(mc_header) + MC_HEADER_SIZE;
285 ext_sigcount = ext_header->count;
286 ext_sig = (void *)ext_header + EXT_HEADER_SIZE;
287 for (i = 0; i < ext_sigcount; i++) {
288 if (microcode_update_match(cpu, mc_header,
289 ext_sig->sig, ext_sig->pf))
290 goto find;
291 ext_sig++;
292 }
293 return 0;
294find:
295 pr_debug("microcode: CPU%d found a matching microcode update with"
296 " version 0x%x (current=0x%x)\n", cpu, mc_header->rev,uci->rev);
297 new_mc = vmalloc(total_size);
298 if (!new_mc) {
299 printk(KERN_ERR "microcode: error! Can not allocate memory\n");
300 return -ENOMEM;
301 }
302
303 /* free previous update file */
304 vfree(uci->mc);
305
306 memcpy(new_mc, mc, total_size);
307 uci->mc = new_mc;
308 return 1;
309}
310
311static void apply_microcode(int cpu)
312{
313 unsigned long flags;
314 unsigned int val[2];
315 int cpu_num = raw_smp_processor_id();
316 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num;
317
318 /* We should bind the task to the CPU */
319 BUG_ON(cpu_num != cpu);
320
321 if (uci->mc == NULL)
322 return;
323
324 /* serialize access to the physical write to MSR 0x79 */
325 spin_lock_irqsave(&microcode_update_lock, flags);
326
327 /* write microcode via MSR 0x79 */
328 wrmsr(MSR_IA32_UCODE_WRITE,
329 (unsigned long) uci->mc->bits,
330 (unsigned long) uci->mc->bits >> 16 >> 16);
331 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
332
333 /* see notes above for revision 1.07. Apparent chip bug */
334 sync_core();
335
336 /* get the current revision from MSR 0x8B */
337 rdmsr(MSR_IA32_UCODE_REV, val[0], val[1]);
338
339 spin_unlock_irqrestore(&microcode_update_lock, flags);
340 if (val[1] != uci->mc->hdr.rev) {
341 printk(KERN_ERR "microcode: CPU%d update from revision "
342 "0x%x to 0x%x failed\n", cpu_num, uci->rev, val[1]);
343 return;
344 }
345 printk(KERN_INFO "microcode: CPU%d updated from revision "
346 "0x%x to 0x%x, date = %08x \n",
347 cpu_num, uci->rev, val[1], uci->mc->hdr.date);
348 uci->rev = val[1];
349}
350
351#ifdef CONFIG_MICROCODE_OLD_INTERFACE
352static void __user *user_buffer; /* user area microcode data buffer */
353static unsigned int user_buffer_size; /* it's size */
354
355static long get_next_ucode(void **mc, long offset)
356{
357 microcode_header_t mc_header;
358 unsigned long total_size;
359
360 /* No more data */
361 if (offset >= user_buffer_size)
362 return 0;
363 if (copy_from_user(&mc_header, user_buffer + offset, MC_HEADER_SIZE)) {
364 printk(KERN_ERR "microcode: error! Can not read user data\n");
365 return -EFAULT;
366 }
367 total_size = get_totalsize(&mc_header);
368 if (offset + total_size > user_buffer_size) {
369 printk(KERN_ERR "microcode: error! Bad total size in microcode "
370 "data file\n");
371 return -EINVAL;
372 }
373 *mc = vmalloc(total_size);
374 if (!*mc)
375 return -ENOMEM;
376 if (copy_from_user(*mc, user_buffer + offset, total_size)) {
377 printk(KERN_ERR "microcode: error! Can not read user data\n");
378 vfree(*mc);
379 return -EFAULT;
380 }
381 return offset + total_size;
382}
383
384static int do_microcode_update (void)
385{
386 long cursor = 0;
387 int error = 0;
388 void *new_mc = NULL;
389 int cpu;
390 cpumask_t old;
391
392 old = current->cpus_allowed;
393
394 while ((cursor = get_next_ucode(&new_mc, cursor)) > 0) {
395 error = microcode_sanity_check(new_mc);
396 if (error)
397 goto out;
398 /*
399 * It's possible the data file has multiple matching ucode,
400 * lets keep searching till the latest version
401 */
402 for_each_online_cpu(cpu) {
403 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
404
405 if (!uci->valid)
406 continue;
407 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
408 error = get_maching_microcode(new_mc, cpu);
409 if (error < 0)
410 goto out;
411 if (error == 1)
412 apply_microcode(cpu);
413 }
414 vfree(new_mc);
415 }
416out:
417 if (cursor > 0)
418 vfree(new_mc);
419 if (cursor < 0)
420 error = cursor;
421 set_cpus_allowed_ptr(current, &old);
422 return error;
423}
424
425static int microcode_open (struct inode *unused1, struct file *unused2)
426{
427 cycle_kernel_lock();
428 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
429}
430
431static ssize_t microcode_write (struct file *file, const char __user *buf, size_t len, loff_t *ppos)
432{
433 ssize_t ret;
434
435 if ((len >> PAGE_SHIFT) > num_physpages) {
436 printk(KERN_ERR "microcode: too much data (max %ld pages)\n", num_physpages);
437 return -EINVAL;
438 }
439
440 get_online_cpus();
441 mutex_lock(&microcode_mutex);
442
443 user_buffer = (void __user *) buf;
444 user_buffer_size = (int) len;
445
446 ret = do_microcode_update();
447 if (!ret)
448 ret = (ssize_t)len;
449
450 mutex_unlock(&microcode_mutex);
451 put_online_cpus();
452
453 return ret;
454}
455
456static const struct file_operations microcode_fops = {
457 .owner = THIS_MODULE,
458 .write = microcode_write,
459 .open = microcode_open,
460};
461
462static struct miscdevice microcode_dev = {
463 .minor = MICROCODE_MINOR,
464 .name = "microcode",
465 .fops = &microcode_fops,
466};
467
468static int __init microcode_dev_init (void)
469{
470 int error;
471
472 error = misc_register(&microcode_dev);
473 if (error) {
474 printk(KERN_ERR
475 "microcode: can't misc_register on minor=%d\n",
476 MICROCODE_MINOR);
477 return error;
478 }
479
480 return 0;
481}
482
483static void microcode_dev_exit (void)
484{
485 misc_deregister(&microcode_dev);
486}
487
488MODULE_ALIAS_MISCDEV(MICROCODE_MINOR);
489#else
490#define microcode_dev_init() 0
491#define microcode_dev_exit() do { } while(0)
492#endif
493
494static long get_next_ucode_from_buffer(void **mc, const u8 *buf,
495 unsigned long size, long offset)
496{
497 microcode_header_t *mc_header;
498 unsigned long total_size;
499
500 /* No more data */
501 if (offset >= size)
502 return 0;
503 mc_header = (microcode_header_t *)(buf + offset);
504 total_size = get_totalsize(mc_header);
505
506 if (offset + total_size > size) {
507 printk(KERN_ERR "microcode: error! Bad data in microcode data file\n");
508 return -EINVAL;
509 }
510
511 *mc = vmalloc(total_size);
512 if (!*mc) {
513 printk(KERN_ERR "microcode: error! Can not allocate memory\n");
514 return -ENOMEM;
515 }
516 memcpy(*mc, buf + offset, total_size);
517 return offset + total_size;
518}
519
520/* fake device for request_firmware */
521static struct platform_device *microcode_pdev;
522
523static int cpu_request_microcode(int cpu)
524{
525 char name[30];
526 struct cpuinfo_x86 *c = &cpu_data(cpu);
527 const struct firmware *firmware;
528 const u8 *buf;
529 unsigned long size;
530 long offset = 0;
531 int error;
532 void *mc;
533
534 /* We should bind the task to the CPU */
535 BUG_ON(cpu != raw_smp_processor_id());
536 sprintf(name,"intel-ucode/%02x-%02x-%02x",
537 c->x86, c->x86_model, c->x86_mask);
538 error = request_firmware(&firmware, name, &microcode_pdev->dev);
539 if (error) {
540 pr_debug("microcode: data file %s load failed\n", name);
541 return error;
542 }
543 buf = firmware->data;
544 size = firmware->size;
545 while ((offset = get_next_ucode_from_buffer(&mc, buf, size, offset))
546 > 0) {
547 error = microcode_sanity_check(mc);
548 if (error)
549 break;
550 error = get_maching_microcode(mc, cpu);
551 if (error < 0)
552 break;
553 /*
554 * It's possible the data file has multiple matching ucode,
555 * lets keep searching till the latest version
556 */
557 if (error == 1) {
558 apply_microcode(cpu);
559 error = 0;
560 }
561 vfree(mc);
562 }
563 if (offset > 0)
564 vfree(mc);
565 if (offset < 0)
566 error = offset;
567 release_firmware(firmware);
568
569 return error;
570}
571
572static int apply_microcode_check_cpu(int cpu)
573{
574 struct cpuinfo_x86 *c = &cpu_data(cpu);
575 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
576 cpumask_t old;
577 unsigned int val[2];
578 int err = 0;
579
580 /* Check if the microcode is available */
581 if (!uci->mc)
582 return 0;
583
584 old = current->cpus_allowed;
585 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
586
587 /* Check if the microcode we have in memory matches the CPU */
588 if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 < 6 ||
589 cpu_has(c, X86_FEATURE_IA64) || uci->sig != cpuid_eax(0x00000001))
590 err = -EINVAL;
591
592 if (!err && ((c->x86_model >= 5) || (c->x86 > 6))) {
593 /* get processor flags from MSR 0x17 */
594 rdmsr(MSR_IA32_PLATFORM_ID, val[0], val[1]);
595 if (uci->pf != (1 << ((val[1] >> 18) & 7)))
596 err = -EINVAL;
597 }
598
599 if (!err) {
600 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
601 /* see notes above for revision 1.07. Apparent chip bug */
602 sync_core();
603 /* get the current revision from MSR 0x8B */
604 rdmsr(MSR_IA32_UCODE_REV, val[0], val[1]);
605 if (uci->rev != val[1])
606 err = -EINVAL;
607 }
608
609 if (!err)
610 apply_microcode(cpu);
611 else
612 printk(KERN_ERR "microcode: Could not apply microcode to CPU%d:"
613 " sig=0x%x, pf=0x%x, rev=0x%x\n",
614 cpu, uci->sig, uci->pf, uci->rev);
615
616 set_cpus_allowed_ptr(current, &old);
617 return err;
618}
619
620static void microcode_init_cpu(int cpu, int resume)
621{
622 cpumask_t old;
623 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
624
625 old = current->cpus_allowed;
626
627 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
628 mutex_lock(&microcode_mutex);
629 collect_cpu_info(cpu);
630 if (uci->valid && system_state == SYSTEM_RUNNING && !resume)
631 cpu_request_microcode(cpu);
632 mutex_unlock(&microcode_mutex);
633 set_cpus_allowed_ptr(current, &old);
634}
635
636static void microcode_fini_cpu(int cpu)
637{
638 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
639
640 mutex_lock(&microcode_mutex);
641 uci->valid = 0;
642 vfree(uci->mc);
643 uci->mc = NULL;
644 mutex_unlock(&microcode_mutex);
645}
646
647static ssize_t reload_store(struct sys_device *dev,
648 struct sysdev_attribute *attr,
649 const char *buf, size_t sz)
650{
651 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
652 char *end;
653 unsigned long val = simple_strtoul(buf, &end, 0);
654 int err = 0;
655 int cpu = dev->id;
656
657 if (end == buf)
658 return -EINVAL;
659 if (val == 1) {
660 cpumask_t old = current->cpus_allowed;
661
662 get_online_cpus();
663 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
664
665 mutex_lock(&microcode_mutex);
666 if (uci->valid)
667 err = cpu_request_microcode(cpu);
668 mutex_unlock(&microcode_mutex);
669 put_online_cpus();
670 set_cpus_allowed_ptr(current, &old);
671 }
672 if (err)
673 return err;
674 return sz;
675}
676
677static ssize_t version_show(struct sys_device *dev,
678 struct sysdev_attribute *attr, char *buf)
679{
680 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
681
682 return sprintf(buf, "0x%x\n", uci->rev);
683}
684
685static ssize_t pf_show(struct sys_device *dev,
686 struct sysdev_attribute *attr, char *buf)
687{
688 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
689
690 return sprintf(buf, "0x%x\n", uci->pf);
691}
692
693static SYSDEV_ATTR(reload, 0200, NULL, reload_store);
694static SYSDEV_ATTR(version, 0400, version_show, NULL);
695static SYSDEV_ATTR(processor_flags, 0400, pf_show, NULL);
696
697static struct attribute *mc_default_attrs[] = {
698 &attr_reload.attr,
699 &attr_version.attr,
700 &attr_processor_flags.attr,
701 NULL
702};
703
704static struct attribute_group mc_attr_group = {
705 .attrs = mc_default_attrs,
706 .name = "microcode",
707};
708
709static int __mc_sysdev_add(struct sys_device *sys_dev, int resume)
710{
711 int err, cpu = sys_dev->id;
712 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
713
714 if (!cpu_online(cpu))
715 return 0;
716
717 pr_debug("microcode: CPU%d added\n", cpu);
718 memset(uci, 0, sizeof(*uci));
719
720 err = sysfs_create_group(&sys_dev->kobj, &mc_attr_group);
721 if (err)
722 return err;
723
724 microcode_init_cpu(cpu, resume);
725
726 return 0;
727}
728
729static int mc_sysdev_add(struct sys_device *sys_dev)
730{
731 return __mc_sysdev_add(sys_dev, 0);
732}
733
734static int mc_sysdev_remove(struct sys_device *sys_dev)
735{
736 int cpu = sys_dev->id;
737
738 if (!cpu_online(cpu))
739 return 0;
740
741 pr_debug("microcode: CPU%d removed\n", cpu);
742 microcode_fini_cpu(cpu);
743 sysfs_remove_group(&sys_dev->kobj, &mc_attr_group);
744 return 0;
745}
746
747static int mc_sysdev_resume(struct sys_device *dev)
748{
749 int cpu = dev->id;
750
751 if (!cpu_online(cpu))
752 return 0;
753 pr_debug("microcode: CPU%d resumed\n", cpu);
754 /* only CPU 0 will apply ucode here */
755 apply_microcode(0);
756 return 0;
757}
758
759static struct sysdev_driver mc_sysdev_driver = {
760 .add = mc_sysdev_add,
761 .remove = mc_sysdev_remove,
762 .resume = mc_sysdev_resume,
763};
764
765static __cpuinit int
766mc_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
767{
768 unsigned int cpu = (unsigned long)hcpu;
769 struct sys_device *sys_dev;
770
771 sys_dev = get_cpu_sysdev(cpu);
772 switch (action) {
773 case CPU_UP_CANCELED_FROZEN:
774 /* The CPU refused to come up during a system resume */
775 microcode_fini_cpu(cpu);
776 break;
777 case CPU_ONLINE:
778 case CPU_DOWN_FAILED:
779 mc_sysdev_add(sys_dev);
780 break;
781 case CPU_ONLINE_FROZEN:
782 /* System-wide resume is in progress, try to apply microcode */
783 if (apply_microcode_check_cpu(cpu)) {
784 /* The application of microcode failed */
785 microcode_fini_cpu(cpu);
786 __mc_sysdev_add(sys_dev, 1);
787 break;
788 }
789 case CPU_DOWN_FAILED_FROZEN:
790 if (sysfs_create_group(&sys_dev->kobj, &mc_attr_group))
791 printk(KERN_ERR "microcode: Failed to create the sysfs "
792 "group for CPU%d\n", cpu);
793 break;
794 case CPU_DOWN_PREPARE:
795 mc_sysdev_remove(sys_dev);
796 break;
797 case CPU_DOWN_PREPARE_FROZEN:
798 /* Suspend is in progress, only remove the interface */
799 sysfs_remove_group(&sys_dev->kobj, &mc_attr_group);
800 break;
801 }
802 return NOTIFY_OK;
803}
804
805static struct notifier_block __refdata mc_cpu_notifier = {
806 .notifier_call = mc_cpu_callback,
807};
808
809static int __init microcode_init (void)
810{
811 int error;
812
813 printk(KERN_INFO
814 "IA-32 Microcode Update Driver: v" MICROCODE_VERSION " <tigran@aivazian.fsnet.co.uk>\n");
815
816 error = microcode_dev_init();
817 if (error)
818 return error;
819 microcode_pdev = platform_device_register_simple("microcode", -1,
820 NULL, 0);
821 if (IS_ERR(microcode_pdev)) {
822 microcode_dev_exit();
823 return PTR_ERR(microcode_pdev);
824 }
825
826 get_online_cpus();
827 error = sysdev_driver_register(&cpu_sysdev_class, &mc_sysdev_driver);
828 put_online_cpus();
829 if (error) {
830 microcode_dev_exit();
831 platform_device_unregister(microcode_pdev);
832 return error;
833 }
834
835 register_hotcpu_notifier(&mc_cpu_notifier);
836 return 0;
837}
838
839static void __exit microcode_exit (void)
840{
841 microcode_dev_exit();
842
843 unregister_hotcpu_notifier(&mc_cpu_notifier);
844
845 get_online_cpus();
846 sysdev_driver_unregister(&cpu_sysdev_class, &mc_sysdev_driver);
847 put_online_cpus();
848
849 platform_device_unregister(microcode_pdev);
850}
851
852module_init(microcode_init)
853module_exit(microcode_exit)
diff --git a/arch/x86/kernel/microcode_amd.c b/arch/x86/kernel/microcode_amd.c
new file mode 100644
index 000000000000..7a1f8eeac2c7
--- /dev/null
+++ b/arch/x86/kernel/microcode_amd.c
@@ -0,0 +1,435 @@
1/*
2 * AMD CPU Microcode Update Driver for Linux
3 * Copyright (C) 2008 Advanced Micro Devices Inc.
4 *
5 * Author: Peter Oruba <peter.oruba@amd.com>
6 *
7 * Based on work by:
8 * Tigran Aivazian <tigran@aivazian.fsnet.co.uk>
9 *
10 * This driver allows to upgrade microcode on AMD
11 * family 0x10 and 0x11 processors.
12 *
13 * Licensed unter the terms of the GNU General Public
14 * License version 2. See file COPYING for details.
15*/
16
17#include <linux/capability.h>
18#include <linux/kernel.h>
19#include <linux/init.h>
20#include <linux/sched.h>
21#include <linux/cpumask.h>
22#include <linux/module.h>
23#include <linux/slab.h>
24#include <linux/vmalloc.h>
25#include <linux/miscdevice.h>
26#include <linux/spinlock.h>
27#include <linux/mm.h>
28#include <linux/fs.h>
29#include <linux/mutex.h>
30#include <linux/cpu.h>
31#include <linux/firmware.h>
32#include <linux/platform_device.h>
33#include <linux/pci.h>
34#include <linux/pci_ids.h>
35
36#include <asm/msr.h>
37#include <asm/uaccess.h>
38#include <asm/processor.h>
39#include <asm/microcode.h>
40
41MODULE_DESCRIPTION("AMD Microcode Update Driver");
42MODULE_AUTHOR("Peter Oruba <peter.oruba@amd.com>");
43MODULE_LICENSE("GPL v2");
44
45#define UCODE_MAGIC 0x00414d44
46#define UCODE_EQUIV_CPU_TABLE_TYPE 0x00000000
47#define UCODE_UCODE_TYPE 0x00000001
48
49struct equiv_cpu_entry {
50 unsigned int installed_cpu;
51 unsigned int fixed_errata_mask;
52 unsigned int fixed_errata_compare;
53 unsigned int equiv_cpu;
54};
55
56struct microcode_header_amd {
57 unsigned int data_code;
58 unsigned int patch_id;
59 unsigned char mc_patch_data_id[2];
60 unsigned char mc_patch_data_len;
61 unsigned char init_flag;
62 unsigned int mc_patch_data_checksum;
63 unsigned int nb_dev_id;
64 unsigned int sb_dev_id;
65 unsigned char processor_rev_id[2];
66 unsigned char nb_rev_id;
67 unsigned char sb_rev_id;
68 unsigned char bios_api_rev;
69 unsigned char reserved1[3];
70 unsigned int match_reg[8];
71};
72
73struct microcode_amd {
74 struct microcode_header_amd hdr;
75 unsigned int mpb[0];
76};
77
78#define UCODE_MAX_SIZE (2048)
79#define DEFAULT_UCODE_DATASIZE (896)
80#define MC_HEADER_SIZE (sizeof(struct microcode_header_amd))
81#define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE)
82#define DWSIZE (sizeof(u32))
83/* For now we support a fixed ucode total size only */
84#define get_totalsize(mc) \
85 ((((struct microcode_amd *)mc)->hdr.mc_patch_data_len * 28) \
86 + MC_HEADER_SIZE)
87
88/* serialize access to the physical write */
89static DEFINE_SPINLOCK(microcode_update_lock);
90
91static struct equiv_cpu_entry *equiv_cpu_table;
92
93static int collect_cpu_info_amd(int cpu, struct cpu_signature *csig)
94{
95 struct cpuinfo_x86 *c = &cpu_data(cpu);
96
97 memset(csig, 0, sizeof(*csig));
98
99 if (c->x86_vendor != X86_VENDOR_AMD || c->x86 < 0x10) {
100 printk(KERN_ERR "microcode: CPU%d not a capable AMD processor\n",
101 cpu);
102 return -1;
103 }
104
105 asm volatile("movl %1, %%ecx; rdmsr"
106 : "=a" (csig->rev)
107 : "i" (0x0000008B) : "ecx");
108
109 printk(KERN_INFO "microcode: collect_cpu_info_amd : patch_id=0x%x\n",
110 csig->rev);
111
112 return 0;
113}
114
115static int get_matching_microcode(int cpu, void *mc, int rev)
116{
117 struct microcode_header_amd *mc_header = mc;
118 struct pci_dev *nb_pci_dev, *sb_pci_dev;
119 unsigned int current_cpu_id;
120 unsigned int equiv_cpu_id = 0x00;
121 unsigned int i = 0;
122
123 BUG_ON(equiv_cpu_table == NULL);
124 current_cpu_id = cpuid_eax(0x00000001);
125
126 while (equiv_cpu_table[i].installed_cpu != 0) {
127 if (current_cpu_id == equiv_cpu_table[i].installed_cpu) {
128 equiv_cpu_id = equiv_cpu_table[i].equiv_cpu;
129 break;
130 }
131 i++;
132 }
133
134 if (!equiv_cpu_id) {
135 printk(KERN_ERR "microcode: CPU%d cpu_id "
136 "not found in equivalent cpu table \n", cpu);
137 return 0;
138 }
139
140 if ((mc_header->processor_rev_id[0]) != (equiv_cpu_id & 0xff)) {
141 printk(KERN_ERR
142 "microcode: CPU%d patch does not match "
143 "(patch is %x, cpu extended is %x) \n",
144 cpu, mc_header->processor_rev_id[0],
145 (equiv_cpu_id & 0xff));
146 return 0;
147 }
148
149 if ((mc_header->processor_rev_id[1]) != ((equiv_cpu_id >> 16) & 0xff)) {
150 printk(KERN_ERR "microcode: CPU%d patch does not match "
151 "(patch is %x, cpu base id is %x) \n",
152 cpu, mc_header->processor_rev_id[1],
153 ((equiv_cpu_id >> 16) & 0xff));
154
155 return 0;
156 }
157
158 /* ucode may be northbridge specific */
159 if (mc_header->nb_dev_id) {
160 nb_pci_dev = pci_get_device(PCI_VENDOR_ID_AMD,
161 (mc_header->nb_dev_id & 0xff),
162 NULL);
163 if ((!nb_pci_dev) ||
164 (mc_header->nb_rev_id != nb_pci_dev->revision)) {
165 printk(KERN_ERR "microcode: CPU%d NB mismatch \n", cpu);
166 pci_dev_put(nb_pci_dev);
167 return 0;
168 }
169 pci_dev_put(nb_pci_dev);
170 }
171
172 /* ucode may be southbridge specific */
173 if (mc_header->sb_dev_id) {
174 sb_pci_dev = pci_get_device(PCI_VENDOR_ID_AMD,
175 (mc_header->sb_dev_id & 0xff),
176 NULL);
177 if ((!sb_pci_dev) ||
178 (mc_header->sb_rev_id != sb_pci_dev->revision)) {
179 printk(KERN_ERR "microcode: CPU%d SB mismatch \n", cpu);
180 pci_dev_put(sb_pci_dev);
181 return 0;
182 }
183 pci_dev_put(sb_pci_dev);
184 }
185
186 if (mc_header->patch_id <= rev)
187 return 0;
188
189 return 1;
190}
191
192static void apply_microcode_amd(int cpu)
193{
194 unsigned long flags;
195 unsigned int eax, edx;
196 unsigned int rev;
197 int cpu_num = raw_smp_processor_id();
198 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num;
199 struct microcode_amd *mc_amd = uci->mc;
200 unsigned long addr;
201
202 /* We should bind the task to the CPU */
203 BUG_ON(cpu_num != cpu);
204
205 if (mc_amd == NULL)
206 return;
207
208 spin_lock_irqsave(&microcode_update_lock, flags);
209
210 addr = (unsigned long)&mc_amd->hdr.data_code;
211 edx = (unsigned int)(((unsigned long)upper_32_bits(addr)));
212 eax = (unsigned int)(((unsigned long)lower_32_bits(addr)));
213
214 asm volatile("movl %0, %%ecx; wrmsr" :
215 : "i" (0xc0010020), "a" (eax), "d" (edx) : "ecx");
216
217 /* get patch id after patching */
218 asm volatile("movl %1, %%ecx; rdmsr"
219 : "=a" (rev)
220 : "i" (0x0000008B) : "ecx");
221
222 spin_unlock_irqrestore(&microcode_update_lock, flags);
223
224 /* check current patch id and patch's id for match */
225 if (rev != mc_amd->hdr.patch_id) {
226 printk(KERN_ERR "microcode: CPU%d update from revision "
227 "0x%x to 0x%x failed\n", cpu_num,
228 mc_amd->hdr.patch_id, rev);
229 return;
230 }
231
232 printk(KERN_INFO "microcode: CPU%d updated from revision "
233 "0x%x to 0x%x \n",
234 cpu_num, uci->cpu_sig.rev, mc_amd->hdr.patch_id);
235
236 uci->cpu_sig.rev = rev;
237}
238
239static void * get_next_ucode(u8 *buf, unsigned int size,
240 int (*get_ucode_data)(void *, const void *, size_t),
241 unsigned int *mc_size)
242{
243 unsigned int total_size;
244#define UCODE_CONTAINER_SECTION_HDR 8
245 u8 section_hdr[UCODE_CONTAINER_SECTION_HDR];
246 void *mc;
247
248 if (get_ucode_data(section_hdr, buf, UCODE_CONTAINER_SECTION_HDR))
249 return NULL;
250
251 if (section_hdr[0] != UCODE_UCODE_TYPE) {
252 printk(KERN_ERR "microcode: error! "
253 "Wrong microcode payload type field\n");
254 return NULL;
255 }
256
257 total_size = (unsigned long) (section_hdr[4] + (section_hdr[5] << 8));
258
259 printk(KERN_INFO "microcode: size %u, total_size %u\n",
260 size, total_size);
261
262 if (total_size > size || total_size > UCODE_MAX_SIZE) {
263 printk(KERN_ERR "microcode: error! Bad data in microcode data file\n");
264 return NULL;
265 }
266
267 mc = vmalloc(UCODE_MAX_SIZE);
268 if (mc) {
269 memset(mc, 0, UCODE_MAX_SIZE);
270 if (get_ucode_data(mc, buf + UCODE_CONTAINER_SECTION_HDR, total_size)) {
271 vfree(mc);
272 mc = NULL;
273 } else
274 *mc_size = total_size + UCODE_CONTAINER_SECTION_HDR;
275 }
276#undef UCODE_CONTAINER_SECTION_HDR
277 return mc;
278}
279
280
281static int install_equiv_cpu_table(u8 *buf,
282 int (*get_ucode_data)(void *, const void *, size_t))
283{
284#define UCODE_CONTAINER_HEADER_SIZE 12
285 u8 *container_hdr[UCODE_CONTAINER_HEADER_SIZE];
286 unsigned int *buf_pos = (unsigned int *)container_hdr;
287 unsigned long size;
288
289 if (get_ucode_data(&container_hdr, buf, UCODE_CONTAINER_HEADER_SIZE))
290 return 0;
291
292 size = buf_pos[2];
293
294 if (buf_pos[1] != UCODE_EQUIV_CPU_TABLE_TYPE || !size) {
295 printk(KERN_ERR "microcode: error! "
296 "Wrong microcode equivalnet cpu table\n");
297 return 0;
298 }
299
300 equiv_cpu_table = (struct equiv_cpu_entry *) vmalloc(size);
301 if (!equiv_cpu_table) {
302 printk(KERN_ERR "microcode: error, can't allocate memory for equiv CPU table\n");
303 return 0;
304 }
305
306 buf += UCODE_CONTAINER_HEADER_SIZE;
307 if (get_ucode_data(equiv_cpu_table, buf, size)) {
308 vfree(equiv_cpu_table);
309 return 0;
310 }
311
312 return size + UCODE_CONTAINER_HEADER_SIZE; /* add header length */
313#undef UCODE_CONTAINER_HEADER_SIZE
314}
315
316static void free_equiv_cpu_table(void)
317{
318 if (equiv_cpu_table) {
319 vfree(equiv_cpu_table);
320 equiv_cpu_table = NULL;
321 }
322}
323
324static int generic_load_microcode(int cpu, void *data, size_t size,
325 int (*get_ucode_data)(void *, const void *, size_t))
326{
327 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
328 u8 *ucode_ptr = data, *new_mc = NULL, *mc;
329 int new_rev = uci->cpu_sig.rev;
330 unsigned int leftover;
331 unsigned long offset;
332
333 offset = install_equiv_cpu_table(ucode_ptr, get_ucode_data);
334 if (!offset) {
335 printk(KERN_ERR "microcode: installing equivalent cpu table failed\n");
336 return -EINVAL;
337 }
338
339 ucode_ptr += offset;
340 leftover = size - offset;
341
342 while (leftover) {
343 unsigned int uninitialized_var(mc_size);
344 struct microcode_header_amd *mc_header;
345
346 mc = get_next_ucode(ucode_ptr, leftover, get_ucode_data, &mc_size);
347 if (!mc)
348 break;
349
350 mc_header = (struct microcode_header_amd *)mc;
351 if (get_matching_microcode(cpu, mc, new_rev)) {
352 if (new_mc)
353 vfree(new_mc);
354 new_rev = mc_header->patch_id;
355 new_mc = mc;
356 } else
357 vfree(mc);
358
359 ucode_ptr += mc_size;
360 leftover -= mc_size;
361 }
362
363 if (new_mc) {
364 if (!leftover) {
365 if (uci->mc)
366 vfree(uci->mc);
367 uci->mc = new_mc;
368 pr_debug("microcode: CPU%d found a matching microcode update with"
369 " version 0x%x (current=0x%x)\n",
370 cpu, new_rev, uci->cpu_sig.rev);
371 } else
372 vfree(new_mc);
373 }
374
375 free_equiv_cpu_table();
376
377 return (int)leftover;
378}
379
380static int get_ucode_fw(void *to, const void *from, size_t n)
381{
382 memcpy(to, from, n);
383 return 0;
384}
385
386static int request_microcode_fw(int cpu, struct device *device)
387{
388 const char *fw_name = "amd-ucode/microcode_amd.bin";
389 const struct firmware *firmware;
390 int ret;
391
392 /* We should bind the task to the CPU */
393 BUG_ON(cpu != raw_smp_processor_id());
394
395 ret = request_firmware(&firmware, fw_name, device);
396 if (ret) {
397 printk(KERN_ERR "microcode: ucode data file %s load failed\n", fw_name);
398 return ret;
399 }
400
401 ret = generic_load_microcode(cpu, (void*)firmware->data, firmware->size,
402 &get_ucode_fw);
403
404 release_firmware(firmware);
405
406 return ret;
407}
408
409static int request_microcode_user(int cpu, const void __user *buf, size_t size)
410{
411 printk(KERN_WARNING "microcode: AMD microcode update via /dev/cpu/microcode"
412 "is not supported\n");
413 return -1;
414}
415
416static void microcode_fini_cpu_amd(int cpu)
417{
418 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
419
420 vfree(uci->mc);
421 uci->mc = NULL;
422}
423
424static struct microcode_ops microcode_amd_ops = {
425 .request_microcode_user = request_microcode_user,
426 .request_microcode_fw = request_microcode_fw,
427 .collect_cpu_info = collect_cpu_info_amd,
428 .apply_microcode = apply_microcode_amd,
429 .microcode_fini_cpu = microcode_fini_cpu_amd,
430};
431
432struct microcode_ops * __init init_amd_microcode(void)
433{
434 return &microcode_amd_ops;
435}
diff --git a/arch/x86/kernel/microcode_core.c b/arch/x86/kernel/microcode_core.c
new file mode 100644
index 000000000000..936d8d55f230
--- /dev/null
+++ b/arch/x86/kernel/microcode_core.c
@@ -0,0 +1,508 @@
1/*
2 * Intel CPU Microcode Update Driver for Linux
3 *
4 * Copyright (C) 2000-2006 Tigran Aivazian <tigran@aivazian.fsnet.co.uk>
5 * 2006 Shaohua Li <shaohua.li@intel.com>
6 *
7 * This driver allows to upgrade microcode on Intel processors
8 * belonging to IA-32 family - PentiumPro, Pentium II,
9 * Pentium III, Xeon, Pentium 4, etc.
10 *
11 * Reference: Section 8.11 of Volume 3a, IA-32 Intel? Architecture
12 * Software Developer's Manual
13 * Order Number 253668 or free download from:
14 *
15 * http://developer.intel.com/design/pentium4/manuals/253668.htm
16 *
17 * For more information, go to http://www.urbanmyth.org/microcode
18 *
19 * This program is free software; you can redistribute it and/or
20 * modify it under the terms of the GNU General Public License
21 * as published by the Free Software Foundation; either version
22 * 2 of the License, or (at your option) any later version.
23 *
24 * 1.0 16 Feb 2000, Tigran Aivazian <tigran@sco.com>
25 * Initial release.
26 * 1.01 18 Feb 2000, Tigran Aivazian <tigran@sco.com>
27 * Added read() support + cleanups.
28 * 1.02 21 Feb 2000, Tigran Aivazian <tigran@sco.com>
29 * Added 'device trimming' support. open(O_WRONLY) zeroes
30 * and frees the saved copy of applied microcode.
31 * 1.03 29 Feb 2000, Tigran Aivazian <tigran@sco.com>
32 * Made to use devfs (/dev/cpu/microcode) + cleanups.
33 * 1.04 06 Jun 2000, Simon Trimmer <simon@veritas.com>
34 * Added misc device support (now uses both devfs and misc).
35 * Added MICROCODE_IOCFREE ioctl to clear memory.
36 * 1.05 09 Jun 2000, Simon Trimmer <simon@veritas.com>
37 * Messages for error cases (non Intel & no suitable microcode).
38 * 1.06 03 Aug 2000, Tigran Aivazian <tigran@veritas.com>
39 * Removed ->release(). Removed exclusive open and status bitmap.
40 * Added microcode_rwsem to serialize read()/write()/ioctl().
41 * Removed global kernel lock usage.
42 * 1.07 07 Sep 2000, Tigran Aivazian <tigran@veritas.com>
43 * Write 0 to 0x8B msr and then cpuid before reading revision,
44 * so that it works even if there were no update done by the
45 * BIOS. Otherwise, reading from 0x8B gives junk (which happened
46 * to be 0 on my machine which is why it worked even when I
47 * disabled update by the BIOS)
48 * Thanks to Eric W. Biederman <ebiederman@lnxi.com> for the fix.
49 * 1.08 11 Dec 2000, Richard Schaal <richard.schaal@intel.com> and
50 * Tigran Aivazian <tigran@veritas.com>
51 * Intel Pentium 4 processor support and bugfixes.
52 * 1.09 30 Oct 2001, Tigran Aivazian <tigran@veritas.com>
53 * Bugfix for HT (Hyper-Threading) enabled processors
54 * whereby processor resources are shared by all logical processors
55 * in a single CPU package.
56 * 1.10 28 Feb 2002 Asit K Mallick <asit.k.mallick@intel.com> and
57 * Tigran Aivazian <tigran@veritas.com>,
58 * Serialize updates as required on HT processors due to
59 * speculative nature of implementation.
60 * 1.11 22 Mar 2002 Tigran Aivazian <tigran@veritas.com>
61 * Fix the panic when writing zero-length microcode chunk.
62 * 1.12 29 Sep 2003 Nitin Kamble <nitin.a.kamble@intel.com>,
63 * Jun Nakajima <jun.nakajima@intel.com>
64 * Support for the microcode updates in the new format.
65 * 1.13 10 Oct 2003 Tigran Aivazian <tigran@veritas.com>
66 * Removed ->read() method and obsoleted MICROCODE_IOCFREE ioctl
67 * because we no longer hold a copy of applied microcode
68 * in kernel memory.
69 * 1.14 25 Jun 2004 Tigran Aivazian <tigran@veritas.com>
70 * Fix sigmatch() macro to handle old CPUs with pf == 0.
71 * Thanks to Stuart Swales for pointing out this bug.
72 */
73#include <linux/capability.h>
74#include <linux/kernel.h>
75#include <linux/init.h>
76#include <linux/sched.h>
77#include <linux/smp_lock.h>
78#include <linux/cpumask.h>
79#include <linux/module.h>
80#include <linux/slab.h>
81#include <linux/vmalloc.h>
82#include <linux/miscdevice.h>
83#include <linux/spinlock.h>
84#include <linux/mm.h>
85#include <linux/fs.h>
86#include <linux/mutex.h>
87#include <linux/cpu.h>
88#include <linux/firmware.h>
89#include <linux/platform_device.h>
90
91#include <asm/msr.h>
92#include <asm/uaccess.h>
93#include <asm/processor.h>
94#include <asm/microcode.h>
95
96MODULE_DESCRIPTION("Microcode Update Driver");
97MODULE_AUTHOR("Tigran Aivazian <tigran@aivazian.fsnet.co.uk>");
98MODULE_LICENSE("GPL");
99
100#define MICROCODE_VERSION "2.00"
101
102struct microcode_ops *microcode_ops;
103
104/* no concurrent ->write()s are allowed on /dev/cpu/microcode */
105static DEFINE_MUTEX(microcode_mutex);
106
107struct ucode_cpu_info ucode_cpu_info[NR_CPUS];
108EXPORT_SYMBOL_GPL(ucode_cpu_info);
109
110#ifdef CONFIG_MICROCODE_OLD_INTERFACE
111static int do_microcode_update(const void __user *buf, size_t size)
112{
113 cpumask_t old;
114 int error = 0;
115 int cpu;
116
117 old = current->cpus_allowed;
118
119 for_each_online_cpu(cpu) {
120 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
121
122 if (!uci->valid)
123 continue;
124
125 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
126 error = microcode_ops->request_microcode_user(cpu, buf, size);
127 if (error < 0)
128 goto out;
129 if (!error)
130 microcode_ops->apply_microcode(cpu);
131 }
132out:
133 set_cpus_allowed_ptr(current, &old);
134 return error;
135}
136
137static int microcode_open(struct inode *unused1, struct file *unused2)
138{
139 cycle_kernel_lock();
140 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
141}
142
143static ssize_t microcode_write(struct file *file, const char __user *buf,
144 size_t len, loff_t *ppos)
145{
146 ssize_t ret;
147
148 if ((len >> PAGE_SHIFT) > num_physpages) {
149 printk(KERN_ERR "microcode: too much data (max %ld pages)\n",
150 num_physpages);
151 return -EINVAL;
152 }
153
154 get_online_cpus();
155 mutex_lock(&microcode_mutex);
156
157 ret = do_microcode_update(buf, len);
158 if (!ret)
159 ret = (ssize_t)len;
160
161 mutex_unlock(&microcode_mutex);
162 put_online_cpus();
163
164 return ret;
165}
166
167static const struct file_operations microcode_fops = {
168 .owner = THIS_MODULE,
169 .write = microcode_write,
170 .open = microcode_open,
171};
172
173static struct miscdevice microcode_dev = {
174 .minor = MICROCODE_MINOR,
175 .name = "microcode",
176 .fops = &microcode_fops,
177};
178
179static int __init microcode_dev_init(void)
180{
181 int error;
182
183 error = misc_register(&microcode_dev);
184 if (error) {
185 printk(KERN_ERR
186 "microcode: can't misc_register on minor=%d\n",
187 MICROCODE_MINOR);
188 return error;
189 }
190
191 return 0;
192}
193
194static void microcode_dev_exit(void)
195{
196 misc_deregister(&microcode_dev);
197}
198
199MODULE_ALIAS_MISCDEV(MICROCODE_MINOR);
200#else
201#define microcode_dev_init() 0
202#define microcode_dev_exit() do { } while (0)
203#endif
204
205/* fake device for request_firmware */
206struct platform_device *microcode_pdev;
207
208static ssize_t reload_store(struct sys_device *dev,
209 struct sysdev_attribute *attr,
210 const char *buf, size_t sz)
211{
212 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
213 char *end;
214 unsigned long val = simple_strtoul(buf, &end, 0);
215 int err = 0;
216 int cpu = dev->id;
217
218 if (end == buf)
219 return -EINVAL;
220 if (val == 1) {
221 cpumask_t old = current->cpus_allowed;
222
223 get_online_cpus();
224 if (cpu_online(cpu)) {
225 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
226 mutex_lock(&microcode_mutex);
227 if (uci->valid) {
228 err = microcode_ops->request_microcode_fw(cpu,
229 &microcode_pdev->dev);
230 if (!err)
231 microcode_ops->apply_microcode(cpu);
232 }
233 mutex_unlock(&microcode_mutex);
234 set_cpus_allowed_ptr(current, &old);
235 }
236 put_online_cpus();
237 }
238 if (err)
239 return err;
240 return sz;
241}
242
243static ssize_t version_show(struct sys_device *dev,
244 struct sysdev_attribute *attr, char *buf)
245{
246 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
247
248 return sprintf(buf, "0x%x\n", uci->cpu_sig.rev);
249}
250
251static ssize_t pf_show(struct sys_device *dev,
252 struct sysdev_attribute *attr, char *buf)
253{
254 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
255
256 return sprintf(buf, "0x%x\n", uci->cpu_sig.pf);
257}
258
259static SYSDEV_ATTR(reload, 0200, NULL, reload_store);
260static SYSDEV_ATTR(version, 0400, version_show, NULL);
261static SYSDEV_ATTR(processor_flags, 0400, pf_show, NULL);
262
263static struct attribute *mc_default_attrs[] = {
264 &attr_reload.attr,
265 &attr_version.attr,
266 &attr_processor_flags.attr,
267 NULL
268};
269
270static struct attribute_group mc_attr_group = {
271 .attrs = mc_default_attrs,
272 .name = "microcode",
273};
274
275static void microcode_fini_cpu(int cpu)
276{
277 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
278
279 mutex_lock(&microcode_mutex);
280 microcode_ops->microcode_fini_cpu(cpu);
281 uci->valid = 0;
282 mutex_unlock(&microcode_mutex);
283}
284
285static void collect_cpu_info(int cpu)
286{
287 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
288
289 memset(uci, 0, sizeof(*uci));
290 if (!microcode_ops->collect_cpu_info(cpu, &uci->cpu_sig))
291 uci->valid = 1;
292}
293
294static int microcode_resume_cpu(int cpu)
295{
296 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
297 struct cpu_signature nsig;
298
299 pr_debug("microcode: CPU%d resumed\n", cpu);
300
301 if (!uci->mc)
302 return 1;
303
304 /*
305 * Let's verify that the 'cached' ucode does belong
306 * to this cpu (a bit of paranoia):
307 */
308 if (microcode_ops->collect_cpu_info(cpu, &nsig)) {
309 microcode_fini_cpu(cpu);
310 return -1;
311 }
312
313 if (memcmp(&nsig, &uci->cpu_sig, sizeof(nsig))) {
314 microcode_fini_cpu(cpu);
315 /* Should we look for a new ucode here? */
316 return 1;
317 }
318
319 return 0;
320}
321
322void microcode_update_cpu(int cpu)
323{
324 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
325 int err = 0;
326
327 /*
328 * Check if the system resume is in progress (uci->valid != NULL),
329 * otherwise just request a firmware:
330 */
331 if (uci->valid) {
332 err = microcode_resume_cpu(cpu);
333 } else {
334 collect_cpu_info(cpu);
335 if (uci->valid && system_state == SYSTEM_RUNNING)
336 err = microcode_ops->request_microcode_fw(cpu,
337 &microcode_pdev->dev);
338 }
339 if (!err)
340 microcode_ops->apply_microcode(cpu);
341}
342
343static void microcode_init_cpu(int cpu)
344{
345 cpumask_t old = current->cpus_allowed;
346
347 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
348 /* We should bind the task to the CPU */
349 BUG_ON(raw_smp_processor_id() != cpu);
350
351 mutex_lock(&microcode_mutex);
352 microcode_update_cpu(cpu);
353 mutex_unlock(&microcode_mutex);
354
355 set_cpus_allowed_ptr(current, &old);
356}
357
358static int mc_sysdev_add(struct sys_device *sys_dev)
359{
360 int err, cpu = sys_dev->id;
361 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
362
363 if (!cpu_online(cpu))
364 return 0;
365
366 pr_debug("microcode: CPU%d added\n", cpu);
367 memset(uci, 0, sizeof(*uci));
368
369 err = sysfs_create_group(&sys_dev->kobj, &mc_attr_group);
370 if (err)
371 return err;
372
373 microcode_init_cpu(cpu);
374 return 0;
375}
376
377static int mc_sysdev_remove(struct sys_device *sys_dev)
378{
379 int cpu = sys_dev->id;
380
381 if (!cpu_online(cpu))
382 return 0;
383
384 pr_debug("microcode: CPU%d removed\n", cpu);
385 microcode_fini_cpu(cpu);
386 sysfs_remove_group(&sys_dev->kobj, &mc_attr_group);
387 return 0;
388}
389
390static int mc_sysdev_resume(struct sys_device *dev)
391{
392 int cpu = dev->id;
393
394 if (!cpu_online(cpu))
395 return 0;
396
397 /* only CPU 0 will apply ucode here */
398 microcode_update_cpu(0);
399 return 0;
400}
401
402static struct sysdev_driver mc_sysdev_driver = {
403 .add = mc_sysdev_add,
404 .remove = mc_sysdev_remove,
405 .resume = mc_sysdev_resume,
406};
407
408static __cpuinit int
409mc_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
410{
411 unsigned int cpu = (unsigned long)hcpu;
412 struct sys_device *sys_dev;
413
414 sys_dev = get_cpu_sysdev(cpu);
415 switch (action) {
416 case CPU_ONLINE:
417 case CPU_ONLINE_FROZEN:
418 microcode_init_cpu(cpu);
419 case CPU_DOWN_FAILED:
420 case CPU_DOWN_FAILED_FROZEN:
421 pr_debug("microcode: CPU%d added\n", cpu);
422 if (sysfs_create_group(&sys_dev->kobj, &mc_attr_group))
423 printk(KERN_ERR "microcode: Failed to create the sysfs "
424 "group for CPU%d\n", cpu);
425 break;
426 case CPU_DOWN_PREPARE:
427 case CPU_DOWN_PREPARE_FROZEN:
428 /* Suspend is in progress, only remove the interface */
429 sysfs_remove_group(&sys_dev->kobj, &mc_attr_group);
430 pr_debug("microcode: CPU%d removed\n", cpu);
431 break;
432 case CPU_DEAD:
433 case CPU_UP_CANCELED_FROZEN:
434 /* The CPU refused to come up during a system resume */
435 microcode_fini_cpu(cpu);
436 break;
437 }
438 return NOTIFY_OK;
439}
440
441static struct notifier_block __refdata mc_cpu_notifier = {
442 .notifier_call = mc_cpu_callback,
443};
444
445static int __init microcode_init(void)
446{
447 struct cpuinfo_x86 *c = &cpu_data(0);
448 int error;
449
450 if (c->x86_vendor == X86_VENDOR_INTEL)
451 microcode_ops = init_intel_microcode();
452 else if (c->x86_vendor == X86_VENDOR_AMD)
453 microcode_ops = init_amd_microcode();
454
455 if (!microcode_ops) {
456 printk(KERN_ERR "microcode: no support for this CPU vendor\n");
457 return -ENODEV;
458 }
459
460 error = microcode_dev_init();
461 if (error)
462 return error;
463 microcode_pdev = platform_device_register_simple("microcode", -1,
464 NULL, 0);
465 if (IS_ERR(microcode_pdev)) {
466 microcode_dev_exit();
467 return PTR_ERR(microcode_pdev);
468 }
469
470 get_online_cpus();
471 error = sysdev_driver_register(&cpu_sysdev_class, &mc_sysdev_driver);
472 put_online_cpus();
473 if (error) {
474 microcode_dev_exit();
475 platform_device_unregister(microcode_pdev);
476 return error;
477 }
478
479 register_hotcpu_notifier(&mc_cpu_notifier);
480
481 printk(KERN_INFO
482 "Microcode Update Driver: v" MICROCODE_VERSION
483 " <tigran@aivazian.fsnet.co.uk>"
484 " <peter.oruba@amd.com>\n");
485
486 return 0;
487}
488
489static void __exit microcode_exit(void)
490{
491 microcode_dev_exit();
492
493 unregister_hotcpu_notifier(&mc_cpu_notifier);
494
495 get_online_cpus();
496 sysdev_driver_unregister(&cpu_sysdev_class, &mc_sysdev_driver);
497 put_online_cpus();
498
499 platform_device_unregister(microcode_pdev);
500
501 microcode_ops = NULL;
502
503 printk(KERN_INFO
504 "Microcode Update Driver: v" MICROCODE_VERSION " removed.\n");
505}
506
507module_init(microcode_init);
508module_exit(microcode_exit);
diff --git a/arch/x86/kernel/microcode_intel.c b/arch/x86/kernel/microcode_intel.c
new file mode 100644
index 000000000000..622dc4a21784
--- /dev/null
+++ b/arch/x86/kernel/microcode_intel.c
@@ -0,0 +1,480 @@
1/*
2 * Intel CPU Microcode Update Driver for Linux
3 *
4 * Copyright (C) 2000-2006 Tigran Aivazian <tigran@aivazian.fsnet.co.uk>
5 * 2006 Shaohua Li <shaohua.li@intel.com>
6 *
7 * This driver allows to upgrade microcode on Intel processors
8 * belonging to IA-32 family - PentiumPro, Pentium II,
9 * Pentium III, Xeon, Pentium 4, etc.
10 *
11 * Reference: Section 8.11 of Volume 3a, IA-32 Intel? Architecture
12 * Software Developer's Manual
13 * Order Number 253668 or free download from:
14 *
15 * http://developer.intel.com/design/pentium4/manuals/253668.htm
16 *
17 * For more information, go to http://www.urbanmyth.org/microcode
18 *
19 * This program is free software; you can redistribute it and/or
20 * modify it under the terms of the GNU General Public License
21 * as published by the Free Software Foundation; either version
22 * 2 of the License, or (at your option) any later version.
23 *
24 * 1.0 16 Feb 2000, Tigran Aivazian <tigran@sco.com>
25 * Initial release.
26 * 1.01 18 Feb 2000, Tigran Aivazian <tigran@sco.com>
27 * Added read() support + cleanups.
28 * 1.02 21 Feb 2000, Tigran Aivazian <tigran@sco.com>
29 * Added 'device trimming' support. open(O_WRONLY) zeroes
30 * and frees the saved copy of applied microcode.
31 * 1.03 29 Feb 2000, Tigran Aivazian <tigran@sco.com>
32 * Made to use devfs (/dev/cpu/microcode) + cleanups.
33 * 1.04 06 Jun 2000, Simon Trimmer <simon@veritas.com>
34 * Added misc device support (now uses both devfs and misc).
35 * Added MICROCODE_IOCFREE ioctl to clear memory.
36 * 1.05 09 Jun 2000, Simon Trimmer <simon@veritas.com>
37 * Messages for error cases (non Intel & no suitable microcode).
38 * 1.06 03 Aug 2000, Tigran Aivazian <tigran@veritas.com>
39 * Removed ->release(). Removed exclusive open and status bitmap.
40 * Added microcode_rwsem to serialize read()/write()/ioctl().
41 * Removed global kernel lock usage.
42 * 1.07 07 Sep 2000, Tigran Aivazian <tigran@veritas.com>
43 * Write 0 to 0x8B msr and then cpuid before reading revision,
44 * so that it works even if there were no update done by the
45 * BIOS. Otherwise, reading from 0x8B gives junk (which happened
46 * to be 0 on my machine which is why it worked even when I
47 * disabled update by the BIOS)
48 * Thanks to Eric W. Biederman <ebiederman@lnxi.com> for the fix.
49 * 1.08 11 Dec 2000, Richard Schaal <richard.schaal@intel.com> and
50 * Tigran Aivazian <tigran@veritas.com>
51 * Intel Pentium 4 processor support and bugfixes.
52 * 1.09 30 Oct 2001, Tigran Aivazian <tigran@veritas.com>
53 * Bugfix for HT (Hyper-Threading) enabled processors
54 * whereby processor resources are shared by all logical processors
55 * in a single CPU package.
56 * 1.10 28 Feb 2002 Asit K Mallick <asit.k.mallick@intel.com> and
57 * Tigran Aivazian <tigran@veritas.com>,
58 * Serialize updates as required on HT processors due to
59 * speculative nature of implementation.
60 * 1.11 22 Mar 2002 Tigran Aivazian <tigran@veritas.com>
61 * Fix the panic when writing zero-length microcode chunk.
62 * 1.12 29 Sep 2003 Nitin Kamble <nitin.a.kamble@intel.com>,
63 * Jun Nakajima <jun.nakajima@intel.com>
64 * Support for the microcode updates in the new format.
65 * 1.13 10 Oct 2003 Tigran Aivazian <tigran@veritas.com>
66 * Removed ->read() method and obsoleted MICROCODE_IOCFREE ioctl
67 * because we no longer hold a copy of applied microcode
68 * in kernel memory.
69 * 1.14 25 Jun 2004 Tigran Aivazian <tigran@veritas.com>
70 * Fix sigmatch() macro to handle old CPUs with pf == 0.
71 * Thanks to Stuart Swales for pointing out this bug.
72 */
73#include <linux/capability.h>
74#include <linux/kernel.h>
75#include <linux/init.h>
76#include <linux/sched.h>
77#include <linux/smp_lock.h>
78#include <linux/cpumask.h>
79#include <linux/module.h>
80#include <linux/slab.h>
81#include <linux/vmalloc.h>
82#include <linux/miscdevice.h>
83#include <linux/spinlock.h>
84#include <linux/mm.h>
85#include <linux/fs.h>
86#include <linux/mutex.h>
87#include <linux/cpu.h>
88#include <linux/firmware.h>
89#include <linux/platform_device.h>
90
91#include <asm/msr.h>
92#include <asm/uaccess.h>
93#include <asm/processor.h>
94#include <asm/microcode.h>
95
96MODULE_DESCRIPTION("Microcode Update Driver");
97MODULE_AUTHOR("Tigran Aivazian <tigran@aivazian.fsnet.co.uk>");
98MODULE_LICENSE("GPL");
99
100struct microcode_header_intel {
101 unsigned int hdrver;
102 unsigned int rev;
103 unsigned int date;
104 unsigned int sig;
105 unsigned int cksum;
106 unsigned int ldrver;
107 unsigned int pf;
108 unsigned int datasize;
109 unsigned int totalsize;
110 unsigned int reserved[3];
111};
112
113struct microcode_intel {
114 struct microcode_header_intel hdr;
115 unsigned int bits[0];
116};
117
118/* microcode format is extended from prescott processors */
119struct extended_signature {
120 unsigned int sig;
121 unsigned int pf;
122 unsigned int cksum;
123};
124
125struct extended_sigtable {
126 unsigned int count;
127 unsigned int cksum;
128 unsigned int reserved[3];
129 struct extended_signature sigs[0];
130};
131
132#define DEFAULT_UCODE_DATASIZE (2000)
133#define MC_HEADER_SIZE (sizeof(struct microcode_header_intel))
134#define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE)
135#define EXT_HEADER_SIZE (sizeof(struct extended_sigtable))
136#define EXT_SIGNATURE_SIZE (sizeof(struct extended_signature))
137#define DWSIZE (sizeof(u32))
138#define get_totalsize(mc) \
139 (((struct microcode_intel *)mc)->hdr.totalsize ? \
140 ((struct microcode_intel *)mc)->hdr.totalsize : \
141 DEFAULT_UCODE_TOTALSIZE)
142
143#define get_datasize(mc) \
144 (((struct microcode_intel *)mc)->hdr.datasize ? \
145 ((struct microcode_intel *)mc)->hdr.datasize : DEFAULT_UCODE_DATASIZE)
146
147#define sigmatch(s1, s2, p1, p2) \
148 (((s1) == (s2)) && (((p1) & (p2)) || (((p1) == 0) && ((p2) == 0))))
149
150#define exttable_size(et) ((et)->count * EXT_SIGNATURE_SIZE + EXT_HEADER_SIZE)
151
152/* serialize access to the physical write to MSR 0x79 */
153static DEFINE_SPINLOCK(microcode_update_lock);
154
155static int collect_cpu_info(int cpu_num, struct cpu_signature *csig)
156{
157 struct cpuinfo_x86 *c = &cpu_data(cpu_num);
158 unsigned int val[2];
159
160 memset(csig, 0, sizeof(*csig));
161
162 if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 < 6 ||
163 cpu_has(c, X86_FEATURE_IA64)) {
164 printk(KERN_ERR "microcode: CPU%d not a capable Intel "
165 "processor\n", cpu_num);
166 return -1;
167 }
168
169 csig->sig = cpuid_eax(0x00000001);
170
171 if ((c->x86_model >= 5) || (c->x86 > 6)) {
172 /* get processor flags from MSR 0x17 */
173 rdmsr(MSR_IA32_PLATFORM_ID, val[0], val[1]);
174 csig->pf = 1 << ((val[1] >> 18) & 7);
175 }
176
177 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
178 /* see notes above for revision 1.07. Apparent chip bug */
179 sync_core();
180 /* get the current revision from MSR 0x8B */
181 rdmsr(MSR_IA32_UCODE_REV, val[0], csig->rev);
182 pr_debug("microcode: collect_cpu_info : sig=0x%x, pf=0x%x, rev=0x%x\n",
183 csig->sig, csig->pf, csig->rev);
184
185 return 0;
186}
187
188static inline int update_match_cpu(struct cpu_signature *csig, int sig, int pf)
189{
190 return (!sigmatch(sig, csig->sig, pf, csig->pf)) ? 0 : 1;
191}
192
193static inline int
194update_match_revision(struct microcode_header_intel *mc_header, int rev)
195{
196 return (mc_header->rev <= rev) ? 0 : 1;
197}
198
199static int microcode_sanity_check(void *mc)
200{
201 struct microcode_header_intel *mc_header = mc;
202 struct extended_sigtable *ext_header = NULL;
203 struct extended_signature *ext_sig;
204 unsigned long total_size, data_size, ext_table_size;
205 int sum, orig_sum, ext_sigcount = 0, i;
206
207 total_size = get_totalsize(mc_header);
208 data_size = get_datasize(mc_header);
209 if (data_size + MC_HEADER_SIZE > total_size) {
210 printk(KERN_ERR "microcode: error! "
211 "Bad data size in microcode data file\n");
212 return -EINVAL;
213 }
214
215 if (mc_header->ldrver != 1 || mc_header->hdrver != 1) {
216 printk(KERN_ERR "microcode: error! "
217 "Unknown microcode update format\n");
218 return -EINVAL;
219 }
220 ext_table_size = total_size - (MC_HEADER_SIZE + data_size);
221 if (ext_table_size) {
222 if ((ext_table_size < EXT_HEADER_SIZE)
223 || ((ext_table_size - EXT_HEADER_SIZE) % EXT_SIGNATURE_SIZE)) {
224 printk(KERN_ERR "microcode: error! "
225 "Small exttable size in microcode data file\n");
226 return -EINVAL;
227 }
228 ext_header = mc + MC_HEADER_SIZE + data_size;
229 if (ext_table_size != exttable_size(ext_header)) {
230 printk(KERN_ERR "microcode: error! "
231 "Bad exttable size in microcode data file\n");
232 return -EFAULT;
233 }
234 ext_sigcount = ext_header->count;
235 }
236
237 /* check extended table checksum */
238 if (ext_table_size) {
239 int ext_table_sum = 0;
240 int *ext_tablep = (int *)ext_header;
241
242 i = ext_table_size / DWSIZE;
243 while (i--)
244 ext_table_sum += ext_tablep[i];
245 if (ext_table_sum) {
246 printk(KERN_WARNING "microcode: aborting, "
247 "bad extended signature table checksum\n");
248 return -EINVAL;
249 }
250 }
251
252 /* calculate the checksum */
253 orig_sum = 0;
254 i = (MC_HEADER_SIZE + data_size) / DWSIZE;
255 while (i--)
256 orig_sum += ((int *)mc)[i];
257 if (orig_sum) {
258 printk(KERN_ERR "microcode: aborting, bad checksum\n");
259 return -EINVAL;
260 }
261 if (!ext_table_size)
262 return 0;
263 /* check extended signature checksum */
264 for (i = 0; i < ext_sigcount; i++) {
265 ext_sig = (void *)ext_header + EXT_HEADER_SIZE +
266 EXT_SIGNATURE_SIZE * i;
267 sum = orig_sum
268 - (mc_header->sig + mc_header->pf + mc_header->cksum)
269 + (ext_sig->sig + ext_sig->pf + ext_sig->cksum);
270 if (sum) {
271 printk(KERN_ERR "microcode: aborting, bad checksum\n");
272 return -EINVAL;
273 }
274 }
275 return 0;
276}
277
278/*
279 * return 0 - no update found
280 * return 1 - found update
281 */
282static int
283get_matching_microcode(struct cpu_signature *cpu_sig, void *mc, int rev)
284{
285 struct microcode_header_intel *mc_header = mc;
286 struct extended_sigtable *ext_header;
287 unsigned long total_size = get_totalsize(mc_header);
288 int ext_sigcount, i;
289 struct extended_signature *ext_sig;
290
291 if (!update_match_revision(mc_header, rev))
292 return 0;
293
294 if (update_match_cpu(cpu_sig, mc_header->sig, mc_header->pf))
295 return 1;
296
297 /* Look for ext. headers: */
298 if (total_size <= get_datasize(mc_header) + MC_HEADER_SIZE)
299 return 0;
300
301 ext_header = mc + get_datasize(mc_header) + MC_HEADER_SIZE;
302 ext_sigcount = ext_header->count;
303 ext_sig = (void *)ext_header + EXT_HEADER_SIZE;
304
305 for (i = 0; i < ext_sigcount; i++) {
306 if (update_match_cpu(cpu_sig, ext_sig->sig, ext_sig->pf))
307 return 1;
308 ext_sig++;
309 }
310 return 0;
311}
312
313static void apply_microcode(int cpu)
314{
315 unsigned long flags;
316 unsigned int val[2];
317 int cpu_num = raw_smp_processor_id();
318 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
319 struct microcode_intel *mc_intel = uci->mc;
320
321 /* We should bind the task to the CPU */
322 BUG_ON(cpu_num != cpu);
323
324 if (mc_intel == NULL)
325 return;
326
327 /* serialize access to the physical write to MSR 0x79 */
328 spin_lock_irqsave(&microcode_update_lock, flags);
329
330 /* write microcode via MSR 0x79 */
331 wrmsr(MSR_IA32_UCODE_WRITE,
332 (unsigned long) mc_intel->bits,
333 (unsigned long) mc_intel->bits >> 16 >> 16);
334 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
335
336 /* see notes above for revision 1.07. Apparent chip bug */
337 sync_core();
338
339 /* get the current revision from MSR 0x8B */
340 rdmsr(MSR_IA32_UCODE_REV, val[0], val[1]);
341
342 spin_unlock_irqrestore(&microcode_update_lock, flags);
343 if (val[1] != mc_intel->hdr.rev) {
344 printk(KERN_ERR "microcode: CPU%d update from revision "
345 "0x%x to 0x%x failed\n", cpu_num, uci->cpu_sig.rev, val[1]);
346 return;
347 }
348 printk(KERN_INFO "microcode: CPU%d updated from revision "
349 "0x%x to 0x%x, date = %04x-%02x-%02x \n",
350 cpu_num, uci->cpu_sig.rev, val[1],
351 mc_intel->hdr.date & 0xffff,
352 mc_intel->hdr.date >> 24,
353 (mc_intel->hdr.date >> 16) & 0xff);
354 uci->cpu_sig.rev = val[1];
355}
356
357static int generic_load_microcode(int cpu, void *data, size_t size,
358 int (*get_ucode_data)(void *, const void *, size_t))
359{
360 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
361 u8 *ucode_ptr = data, *new_mc = NULL, *mc;
362 int new_rev = uci->cpu_sig.rev;
363 unsigned int leftover = size;
364
365 while (leftover) {
366 struct microcode_header_intel mc_header;
367 unsigned int mc_size;
368
369 if (get_ucode_data(&mc_header, ucode_ptr, sizeof(mc_header)))
370 break;
371
372 mc_size = get_totalsize(&mc_header);
373 if (!mc_size || mc_size > leftover) {
374 printk(KERN_ERR "microcode: error!"
375 "Bad data in microcode data file\n");
376 break;
377 }
378
379 mc = vmalloc(mc_size);
380 if (!mc)
381 break;
382
383 if (get_ucode_data(mc, ucode_ptr, mc_size) ||
384 microcode_sanity_check(mc) < 0) {
385 vfree(mc);
386 break;
387 }
388
389 if (get_matching_microcode(&uci->cpu_sig, mc, new_rev)) {
390 if (new_mc)
391 vfree(new_mc);
392 new_rev = mc_header.rev;
393 new_mc = mc;
394 } else
395 vfree(mc);
396
397 ucode_ptr += mc_size;
398 leftover -= mc_size;
399 }
400
401 if (new_mc) {
402 if (!leftover) {
403 if (uci->mc)
404 vfree(uci->mc);
405 uci->mc = (struct microcode_intel *)new_mc;
406 pr_debug("microcode: CPU%d found a matching microcode update with"
407 " version 0x%x (current=0x%x)\n",
408 cpu, new_rev, uci->cpu_sig.rev);
409 } else
410 vfree(new_mc);
411 }
412
413 return (int)leftover;
414}
415
416static int get_ucode_fw(void *to, const void *from, size_t n)
417{
418 memcpy(to, from, n);
419 return 0;
420}
421
422static int request_microcode_fw(int cpu, struct device *device)
423{
424 char name[30];
425 struct cpuinfo_x86 *c = &cpu_data(cpu);
426 const struct firmware *firmware;
427 int ret;
428
429 /* We should bind the task to the CPU */
430 BUG_ON(cpu != raw_smp_processor_id());
431 sprintf(name, "intel-ucode/%02x-%02x-%02x",
432 c->x86, c->x86_model, c->x86_mask);
433 ret = request_firmware(&firmware, name, device);
434 if (ret) {
435 pr_debug("microcode: data file %s load failed\n", name);
436 return ret;
437 }
438
439 ret = generic_load_microcode(cpu, (void*)firmware->data, firmware->size,
440 &get_ucode_fw);
441
442 release_firmware(firmware);
443
444 return ret;
445}
446
447static int get_ucode_user(void *to, const void *from, size_t n)
448{
449 return copy_from_user(to, from, n);
450}
451
452static int request_microcode_user(int cpu, const void __user *buf, size_t size)
453{
454 /* We should bind the task to the CPU */
455 BUG_ON(cpu != raw_smp_processor_id());
456
457 return generic_load_microcode(cpu, (void*)buf, size, &get_ucode_user);
458}
459
460static void microcode_fini_cpu(int cpu)
461{
462 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
463
464 vfree(uci->mc);
465 uci->mc = NULL;
466}
467
468struct microcode_ops microcode_intel_ops = {
469 .request_microcode_user = request_microcode_user,
470 .request_microcode_fw = request_microcode_fw,
471 .collect_cpu_info = collect_cpu_info,
472 .apply_microcode = apply_microcode,
473 .microcode_fini_cpu = microcode_fini_cpu,
474};
475
476struct microcode_ops * __init init_intel_microcode(void)
477{
478 return &microcode_intel_ops;
479}
480
diff --git a/arch/x86/kernel/mpparse.c b/arch/x86/kernel/mpparse.c
index b3fb430725cb..f98f4e1dba09 100644
--- a/arch/x86/kernel/mpparse.c
+++ b/arch/x86/kernel/mpparse.c
@@ -397,7 +397,9 @@ static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
397 generic_bigsmp_probe(); 397 generic_bigsmp_probe();
398#endif 398#endif
399 399
400#ifdef CONFIG_X86_32
400 setup_apic_routing(); 401 setup_apic_routing();
402#endif
401 if (!num_processors) 403 if (!num_processors)
402 printk(KERN_ERR "MPTABLE: no processors registered!\n"); 404 printk(KERN_ERR "MPTABLE: no processors registered!\n");
403 return num_processors; 405 return num_processors;
diff --git a/arch/x86/kernel/msr.c b/arch/x86/kernel/msr.c
index 2e2af5d18191..82a7c7ed6d45 100644
--- a/arch/x86/kernel/msr.c
+++ b/arch/x86/kernel/msr.c
@@ -163,8 +163,8 @@ static int __cpuinit msr_device_create(int cpu)
163{ 163{
164 struct device *dev; 164 struct device *dev;
165 165
166 dev = device_create_drvdata(msr_class, NULL, MKDEV(MSR_MAJOR, cpu), 166 dev = device_create(msr_class, NULL, MKDEV(MSR_MAJOR, cpu), NULL,
167 NULL, "msr%d", cpu); 167 "msr%d", cpu);
168 return IS_ERR(dev) ? PTR_ERR(dev) : 0; 168 return IS_ERR(dev) ? PTR_ERR(dev) : 0;
169} 169}
170 170
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index abb78a2cc4ad..2c97f07f1c2c 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -299,6 +299,15 @@ void acpi_nmi_disable(void)
299 on_each_cpu(__acpi_nmi_disable, NULL, 1); 299 on_each_cpu(__acpi_nmi_disable, NULL, 1);
300} 300}
301 301
302/*
303 * This function is called as soon the LAPIC NMI watchdog driver has everything
304 * in place and it's ready to check if the NMIs belong to the NMI watchdog
305 */
306void cpu_nmi_set_wd_enabled(void)
307{
308 __get_cpu_var(wd_enabled) = 1;
309}
310
302void setup_apic_nmi_watchdog(void *unused) 311void setup_apic_nmi_watchdog(void *unused)
303{ 312{
304 if (__get_cpu_var(wd_enabled)) 313 if (__get_cpu_var(wd_enabled))
@@ -311,8 +320,6 @@ void setup_apic_nmi_watchdog(void *unused)
311 320
312 switch (nmi_watchdog) { 321 switch (nmi_watchdog) {
313 case NMI_LOCAL_APIC: 322 case NMI_LOCAL_APIC:
314 /* enable it before to avoid race with handler */
315 __get_cpu_var(wd_enabled) = 1;
316 if (lapic_watchdog_init(nmi_hz) < 0) { 323 if (lapic_watchdog_init(nmi_hz) < 0) {
317 __get_cpu_var(wd_enabled) = 0; 324 __get_cpu_var(wd_enabled) = 0;
318 return; 325 return;
diff --git a/arch/x86/kernel/numaq_32.c b/arch/x86/kernel/numaq_32.c
index eecc8c18f010..4caff39078e0 100644
--- a/arch/x86/kernel/numaq_32.c
+++ b/arch/x86/kernel/numaq_32.c
@@ -229,6 +229,12 @@ static void __init smp_read_mpc_oem(struct mp_config_oemtable *oemtable,
229 } 229 }
230} 230}
231 231
232static int __init numaq_setup_ioapic_ids(void)
233{
234 /* so can skip it */
235 return 1;
236}
237
232static struct x86_quirks numaq_x86_quirks __initdata = { 238static struct x86_quirks numaq_x86_quirks __initdata = {
233 .arch_pre_time_init = numaq_pre_time_init, 239 .arch_pre_time_init = numaq_pre_time_init,
234 .arch_time_init = NULL, 240 .arch_time_init = NULL,
@@ -243,6 +249,7 @@ static struct x86_quirks numaq_x86_quirks __initdata = {
243 .mpc_oem_bus_info = mpc_oem_bus_info, 249 .mpc_oem_bus_info = mpc_oem_bus_info,
244 .mpc_oem_pci_bus = mpc_oem_pci_bus, 250 .mpc_oem_pci_bus = mpc_oem_pci_bus,
245 .smp_read_mpc_oem = smp_read_mpc_oem, 251 .smp_read_mpc_oem = smp_read_mpc_oem,
252 .setup_ioapic_ids = numaq_setup_ioapic_ids,
246}; 253};
247 254
248void numaq_mps_oem_check(struct mp_config_table *mpc, char *oem, 255void numaq_mps_oem_check(struct mp_config_table *mpc, char *oem,
diff --git a/arch/x86/kernel/olpc.c b/arch/x86/kernel/olpc.c
index 3e6672274807..7a13fac63a1f 100644
--- a/arch/x86/kernel/olpc.c
+++ b/arch/x86/kernel/olpc.c
@@ -190,12 +190,12 @@ EXPORT_SYMBOL_GPL(olpc_ec_cmd);
190static void __init platform_detect(void) 190static void __init platform_detect(void)
191{ 191{
192 size_t propsize; 192 size_t propsize;
193 u32 rev; 193 __be32 rev;
194 194
195 if (ofw("getprop", 4, 1, NULL, "board-revision-int", &rev, 4, 195 if (ofw("getprop", 4, 1, NULL, "board-revision-int", &rev, 4,
196 &propsize) || propsize != 4) { 196 &propsize) || propsize != 4) {
197 printk(KERN_ERR "ofw: getprop call failed!\n"); 197 printk(KERN_ERR "ofw: getprop call failed!\n");
198 rev = 0; 198 rev = cpu_to_be32(0);
199 } 199 }
200 olpc_platform_info.boardrev = be32_to_cpu(rev); 200 olpc_platform_info.boardrev = be32_to_cpu(rev);
201} 201}
@@ -203,7 +203,7 @@ static void __init platform_detect(void)
203static void __init platform_detect(void) 203static void __init platform_detect(void)
204{ 204{
205 /* stopgap until OFW support is added to the kernel */ 205 /* stopgap until OFW support is added to the kernel */
206 olpc_platform_info.boardrev = be32_to_cpu(0xc2); 206 olpc_platform_info.boardrev = 0xc2;
207} 207}
208#endif 208#endif
209 209
diff --git a/arch/x86/kernel/paravirt-spinlocks.c b/arch/x86/kernel/paravirt-spinlocks.c
new file mode 100644
index 000000000000..0e9f1982b1dd
--- /dev/null
+++ b/arch/x86/kernel/paravirt-spinlocks.c
@@ -0,0 +1,37 @@
1/*
2 * Split spinlock implementation out into its own file, so it can be
3 * compiled in a FTRACE-compatible way.
4 */
5#include <linux/spinlock.h>
6#include <linux/module.h>
7
8#include <asm/paravirt.h>
9
10static void default_spin_lock_flags(struct raw_spinlock *lock, unsigned long flags)
11{
12 __raw_spin_lock(lock);
13}
14
15struct pv_lock_ops pv_lock_ops = {
16#ifdef CONFIG_SMP
17 .spin_is_locked = __ticket_spin_is_locked,
18 .spin_is_contended = __ticket_spin_is_contended,
19
20 .spin_lock = __ticket_spin_lock,
21 .spin_lock_flags = default_spin_lock_flags,
22 .spin_trylock = __ticket_spin_trylock,
23 .spin_unlock = __ticket_spin_unlock,
24#endif
25};
26EXPORT_SYMBOL(pv_lock_ops);
27
28void __init paravirt_use_bytelocks(void)
29{
30#ifdef CONFIG_SMP
31 pv_lock_ops.spin_is_locked = __byte_spin_is_locked;
32 pv_lock_ops.spin_is_contended = __byte_spin_is_contended;
33 pv_lock_ops.spin_lock = __byte_spin_lock;
34 pv_lock_ops.spin_trylock = __byte_spin_trylock;
35 pv_lock_ops.spin_unlock = __byte_spin_unlock;
36#endif
37}
diff --git a/arch/x86/kernel/paravirt.c b/arch/x86/kernel/paravirt.c
index 300da17e61cb..e4c8fb608873 100644
--- a/arch/x86/kernel/paravirt.c
+++ b/arch/x86/kernel/paravirt.c
@@ -268,17 +268,6 @@ enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
268 return __get_cpu_var(paravirt_lazy_mode); 268 return __get_cpu_var(paravirt_lazy_mode);
269} 269}
270 270
271void __init paravirt_use_bytelocks(void)
272{
273#ifdef CONFIG_SMP
274 pv_lock_ops.spin_is_locked = __byte_spin_is_locked;
275 pv_lock_ops.spin_is_contended = __byte_spin_is_contended;
276 pv_lock_ops.spin_lock = __byte_spin_lock;
277 pv_lock_ops.spin_trylock = __byte_spin_trylock;
278 pv_lock_ops.spin_unlock = __byte_spin_unlock;
279#endif
280}
281
282struct pv_info pv_info = { 271struct pv_info pv_info = {
283 .name = "bare hardware", 272 .name = "bare hardware",
284 .paravirt_enabled = 0, 273 .paravirt_enabled = 0,
@@ -330,6 +319,7 @@ struct pv_cpu_ops pv_cpu_ops = {
330#endif 319#endif
331 .wbinvd = native_wbinvd, 320 .wbinvd = native_wbinvd,
332 .read_msr = native_read_msr_safe, 321 .read_msr = native_read_msr_safe,
322 .read_msr_amd = native_read_msr_amd_safe,
333 .write_msr = native_write_msr_safe, 323 .write_msr = native_write_msr_safe,
334 .read_tsc = native_read_tsc, 324 .read_tsc = native_read_tsc,
335 .read_pmc = native_read_pmc, 325 .read_pmc = native_read_pmc,
@@ -348,6 +338,10 @@ struct pv_cpu_ops pv_cpu_ops = {
348 .write_ldt_entry = native_write_ldt_entry, 338 .write_ldt_entry = native_write_ldt_entry,
349 .write_gdt_entry = native_write_gdt_entry, 339 .write_gdt_entry = native_write_gdt_entry,
350 .write_idt_entry = native_write_idt_entry, 340 .write_idt_entry = native_write_idt_entry,
341
342 .alloc_ldt = paravirt_nop,
343 .free_ldt = paravirt_nop,
344
351 .load_sp0 = native_load_sp0, 345 .load_sp0 = native_load_sp0,
352 346
353#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 347#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
@@ -373,8 +367,6 @@ struct pv_cpu_ops pv_cpu_ops = {
373 367
374struct pv_apic_ops pv_apic_ops = { 368struct pv_apic_ops pv_apic_ops = {
375#ifdef CONFIG_X86_LOCAL_APIC 369#ifdef CONFIG_X86_LOCAL_APIC
376 .apic_write = native_apic_write,
377 .apic_read = native_apic_read,
378 .setup_boot_clock = setup_boot_APIC_clock, 370 .setup_boot_clock = setup_boot_APIC_clock,
379 .setup_secondary_clock = setup_secondary_APIC_clock, 371 .setup_secondary_clock = setup_secondary_APIC_clock,
380 .startup_ipi_hook = paravirt_nop, 372 .startup_ipi_hook = paravirt_nop,
@@ -461,18 +453,6 @@ struct pv_mmu_ops pv_mmu_ops = {
461 .set_fixmap = native_set_fixmap, 453 .set_fixmap = native_set_fixmap,
462}; 454};
463 455
464struct pv_lock_ops pv_lock_ops = {
465#ifdef CONFIG_SMP
466 .spin_is_locked = __ticket_spin_is_locked,
467 .spin_is_contended = __ticket_spin_is_contended,
468
469 .spin_lock = __ticket_spin_lock,
470 .spin_trylock = __ticket_spin_trylock,
471 .spin_unlock = __ticket_spin_unlock,
472#endif
473};
474EXPORT_SYMBOL(pv_lock_ops);
475
476EXPORT_SYMBOL_GPL(pv_time_ops); 456EXPORT_SYMBOL_GPL(pv_time_ops);
477EXPORT_SYMBOL (pv_cpu_ops); 457EXPORT_SYMBOL (pv_cpu_ops);
478EXPORT_SYMBOL (pv_mmu_ops); 458EXPORT_SYMBOL (pv_mmu_ops);
diff --git a/arch/x86/kernel/paravirt_patch_32.c b/arch/x86/kernel/paravirt_patch_32.c
index 58262218781b..9fe644f4861d 100644
--- a/arch/x86/kernel/paravirt_patch_32.c
+++ b/arch/x86/kernel/paravirt_patch_32.c
@@ -23,7 +23,7 @@ unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
23 start = start_##ops##_##x; \ 23 start = start_##ops##_##x; \
24 end = end_##ops##_##x; \ 24 end = end_##ops##_##x; \
25 goto patch_site 25 goto patch_site
26 switch(type) { 26 switch (type) {
27 PATCH_SITE(pv_irq_ops, irq_disable); 27 PATCH_SITE(pv_irq_ops, irq_disable);
28 PATCH_SITE(pv_irq_ops, irq_enable); 28 PATCH_SITE(pv_irq_ops, irq_enable);
29 PATCH_SITE(pv_irq_ops, restore_fl); 29 PATCH_SITE(pv_irq_ops, restore_fl);
diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index dcdac6c826e9..e1e731d78f38 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -217,16 +217,6 @@ static inline unsigned long verify_bit_range(unsigned long* bitmap,
217 217
218#endif /* CONFIG_IOMMU_DEBUG */ 218#endif /* CONFIG_IOMMU_DEBUG */
219 219
220static inline unsigned int num_dma_pages(unsigned long dma, unsigned int dmalen)
221{
222 unsigned int npages;
223
224 npages = PAGE_ALIGN(dma + dmalen) - (dma & PAGE_MASK);
225 npages >>= PAGE_SHIFT;
226
227 return npages;
228}
229
230static inline int translation_enabled(struct iommu_table *tbl) 220static inline int translation_enabled(struct iommu_table *tbl)
231{ 221{
232 /* only PHBs with translation enabled have an IOMMU table */ 222 /* only PHBs with translation enabled have an IOMMU table */
@@ -261,7 +251,7 @@ static void iommu_range_reserve(struct iommu_table *tbl,
261 badbit, tbl, start_addr, npages); 251 badbit, tbl, start_addr, npages);
262 } 252 }
263 253
264 set_bit_string(tbl->it_map, index, npages); 254 iommu_area_reserve(tbl->it_map, index, npages);
265 255
266 spin_unlock_irqrestore(&tbl->it_lock, flags); 256 spin_unlock_irqrestore(&tbl->it_lock, flags);
267} 257}
@@ -408,7 +398,7 @@ static void calgary_unmap_sg(struct device *dev,
408 if (dmalen == 0) 398 if (dmalen == 0)
409 break; 399 break;
410 400
411 npages = num_dma_pages(dma, dmalen); 401 npages = iommu_num_pages(dma, dmalen, PAGE_SIZE);
412 iommu_free(tbl, dma, npages); 402 iommu_free(tbl, dma, npages);
413 } 403 }
414} 404}
@@ -427,7 +417,7 @@ static int calgary_map_sg(struct device *dev, struct scatterlist *sg,
427 BUG_ON(!sg_page(s)); 417 BUG_ON(!sg_page(s));
428 418
429 vaddr = (unsigned long) sg_virt(s); 419 vaddr = (unsigned long) sg_virt(s);
430 npages = num_dma_pages(vaddr, s->length); 420 npages = iommu_num_pages(vaddr, s->length, PAGE_SIZE);
431 421
432 entry = iommu_range_alloc(dev, tbl, npages); 422 entry = iommu_range_alloc(dev, tbl, npages);
433 if (entry == bad_dma_address) { 423 if (entry == bad_dma_address) {
@@ -464,7 +454,7 @@ static dma_addr_t calgary_map_single(struct device *dev, phys_addr_t paddr,
464 struct iommu_table *tbl = find_iommu_table(dev); 454 struct iommu_table *tbl = find_iommu_table(dev);
465 455
466 uaddr = (unsigned long)vaddr; 456 uaddr = (unsigned long)vaddr;
467 npages = num_dma_pages(uaddr, size); 457 npages = iommu_num_pages(uaddr, size, PAGE_SIZE);
468 458
469 return iommu_alloc(dev, tbl, vaddr, npages, direction); 459 return iommu_alloc(dev, tbl, vaddr, npages, direction);
470} 460}
@@ -475,7 +465,7 @@ static void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle,
475 struct iommu_table *tbl = find_iommu_table(dev); 465 struct iommu_table *tbl = find_iommu_table(dev);
476 unsigned int npages; 466 unsigned int npages;
477 467
478 npages = num_dma_pages(dma_handle, size); 468 npages = iommu_num_pages(dma_handle, size, PAGE_SIZE);
479 iommu_free(tbl, dma_handle, npages); 469 iommu_free(tbl, dma_handle, npages);
480} 470}
481 471
@@ -491,6 +481,8 @@ static void* calgary_alloc_coherent(struct device *dev, size_t size,
491 npages = size >> PAGE_SHIFT; 481 npages = size >> PAGE_SHIFT;
492 order = get_order(size); 482 order = get_order(size);
493 483
484 flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
485
494 /* alloc enough pages (and possibly more) */ 486 /* alloc enough pages (and possibly more) */
495 ret = (void *)__get_free_pages(flag, order); 487 ret = (void *)__get_free_pages(flag, order);
496 if (!ret) 488 if (!ret)
@@ -510,8 +502,22 @@ error:
510 return ret; 502 return ret;
511} 503}
512 504
505static void calgary_free_coherent(struct device *dev, size_t size,
506 void *vaddr, dma_addr_t dma_handle)
507{
508 unsigned int npages;
509 struct iommu_table *tbl = find_iommu_table(dev);
510
511 size = PAGE_ALIGN(size);
512 npages = size >> PAGE_SHIFT;
513
514 iommu_free(tbl, dma_handle, npages);
515 free_pages((unsigned long)vaddr, get_order(size));
516}
517
513static struct dma_mapping_ops calgary_dma_ops = { 518static struct dma_mapping_ops calgary_dma_ops = {
514 .alloc_coherent = calgary_alloc_coherent, 519 .alloc_coherent = calgary_alloc_coherent,
520 .free_coherent = calgary_free_coherent,
515 .map_single = calgary_map_single, 521 .map_single = calgary_map_single,
516 .unmap_single = calgary_unmap_single, 522 .unmap_single = calgary_unmap_single,
517 .map_sg = calgary_map_sg, 523 .map_sg = calgary_map_sg,
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index 87d4d6964ec2..192624820217 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -41,11 +41,12 @@ EXPORT_SYMBOL(bad_dma_address);
41/* Dummy device used for NULL arguments (normally ISA). Better would 41/* Dummy device used for NULL arguments (normally ISA). Better would
42 be probably a smaller DMA mask, but this is bug-to-bug compatible 42 be probably a smaller DMA mask, but this is bug-to-bug compatible
43 to older i386. */ 43 to older i386. */
44struct device fallback_dev = { 44struct device x86_dma_fallback_dev = {
45 .bus_id = "fallback device", 45 .bus_id = "fallback device",
46 .coherent_dma_mask = DMA_32BIT_MASK, 46 .coherent_dma_mask = DMA_32BIT_MASK,
47 .dma_mask = &fallback_dev.coherent_dma_mask, 47 .dma_mask = &x86_dma_fallback_dev.coherent_dma_mask,
48}; 48};
49EXPORT_SYMBOL(x86_dma_fallback_dev);
49 50
50int dma_set_mask(struct device *dev, u64 mask) 51int dma_set_mask(struct device *dev, u64 mask)
51{ 52{
@@ -82,7 +83,7 @@ void __init dma32_reserve_bootmem(void)
82 * using 512M as goal 83 * using 512M as goal
83 */ 84 */
84 align = 64ULL<<20; 85 align = 64ULL<<20;
85 size = round_up(dma32_bootmem_size, align); 86 size = roundup(dma32_bootmem_size, align);
86 dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align, 87 dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align,
87 512ULL<<20); 88 512ULL<<20);
88 if (dma32_bootmem_ptr) 89 if (dma32_bootmem_ptr)
@@ -124,15 +125,46 @@ void __init pci_iommu_alloc(void)
124 pci_swiotlb_init(); 125 pci_swiotlb_init();
125} 126}
126 127
127unsigned long iommu_num_pages(unsigned long addr, unsigned long len) 128unsigned long iommu_nr_pages(unsigned long addr, unsigned long len)
128{ 129{
129 unsigned long size = roundup((addr & ~PAGE_MASK) + len, PAGE_SIZE); 130 unsigned long size = roundup((addr & ~PAGE_MASK) + len, PAGE_SIZE);
130 131
131 return size >> PAGE_SHIFT; 132 return size >> PAGE_SHIFT;
132} 133}
133EXPORT_SYMBOL(iommu_num_pages); 134EXPORT_SYMBOL(iommu_nr_pages);
134#endif 135#endif
135 136
137void *dma_generic_alloc_coherent(struct device *dev, size_t size,
138 dma_addr_t *dma_addr, gfp_t flag)
139{
140 unsigned long dma_mask;
141 struct page *page;
142 dma_addr_t addr;
143
144 dma_mask = dma_alloc_coherent_mask(dev, flag);
145
146 flag |= __GFP_ZERO;
147again:
148 page = alloc_pages_node(dev_to_node(dev), flag, get_order(size));
149 if (!page)
150 return NULL;
151
152 addr = page_to_phys(page);
153 if (!is_buffer_dma_capable(dma_mask, addr, size)) {
154 __free_pages(page, get_order(size));
155
156 if (dma_mask < DMA_32BIT_MASK && !(flag & GFP_DMA)) {
157 flag = (flag & ~GFP_DMA32) | GFP_DMA;
158 goto again;
159 }
160
161 return NULL;
162 }
163
164 *dma_addr = addr;
165 return page_address(page);
166}
167
136/* 168/*
137 * See <Documentation/x86_64/boot-options.txt> for the iommu kernel parameter 169 * See <Documentation/x86_64/boot-options.txt> for the iommu kernel parameter
138 * documentation. 170 * documentation.
@@ -241,147 +273,6 @@ int dma_supported(struct device *dev, u64 mask)
241} 273}
242EXPORT_SYMBOL(dma_supported); 274EXPORT_SYMBOL(dma_supported);
243 275
244/* Allocate DMA memory on node near device */
245static noinline struct page *
246dma_alloc_pages(struct device *dev, gfp_t gfp, unsigned order)
247{
248 int node;
249
250 node = dev_to_node(dev);
251
252 return alloc_pages_node(node, gfp, order);
253}
254
255/*
256 * Allocate memory for a coherent mapping.
257 */
258void *
259dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
260 gfp_t gfp)
261{
262 struct dma_mapping_ops *ops = get_dma_ops(dev);
263 void *memory = NULL;
264 struct page *page;
265 unsigned long dma_mask = 0;
266 dma_addr_t bus;
267 int noretry = 0;
268
269 /* ignore region specifiers */
270 gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
271
272 if (dma_alloc_from_coherent(dev, size, dma_handle, &memory))
273 return memory;
274
275 if (!dev) {
276 dev = &fallback_dev;
277 gfp |= GFP_DMA;
278 }
279 dma_mask = dev->coherent_dma_mask;
280 if (dma_mask == 0)
281 dma_mask = (gfp & GFP_DMA) ? DMA_24BIT_MASK : DMA_32BIT_MASK;
282
283 /* Device not DMA able */
284 if (dev->dma_mask == NULL)
285 return NULL;
286
287 /* Don't invoke OOM killer or retry in lower 16MB DMA zone */
288 if (gfp & __GFP_DMA)
289 noretry = 1;
290
291#ifdef CONFIG_X86_64
292 /* Why <=? Even when the mask is smaller than 4GB it is often
293 larger than 16MB and in this case we have a chance of
294 finding fitting memory in the next higher zone first. If
295 not retry with true GFP_DMA. -AK */
296 if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA)) {
297 gfp |= GFP_DMA32;
298 if (dma_mask < DMA_32BIT_MASK)
299 noretry = 1;
300 }
301#endif
302
303 again:
304 page = dma_alloc_pages(dev,
305 noretry ? gfp | __GFP_NORETRY : gfp, get_order(size));
306 if (page == NULL)
307 return NULL;
308
309 {
310 int high, mmu;
311 bus = page_to_phys(page);
312 memory = page_address(page);
313 high = (bus + size) >= dma_mask;
314 mmu = high;
315 if (force_iommu && !(gfp & GFP_DMA))
316 mmu = 1;
317 else if (high) {
318 free_pages((unsigned long)memory,
319 get_order(size));
320
321 /* Don't use the 16MB ZONE_DMA unless absolutely
322 needed. It's better to use remapping first. */
323 if (dma_mask < DMA_32BIT_MASK && !(gfp & GFP_DMA)) {
324 gfp = (gfp & ~GFP_DMA32) | GFP_DMA;
325 goto again;
326 }
327
328 /* Let low level make its own zone decisions */
329 gfp &= ~(GFP_DMA32|GFP_DMA);
330
331 if (ops->alloc_coherent)
332 return ops->alloc_coherent(dev, size,
333 dma_handle, gfp);
334 return NULL;
335 }
336
337 memset(memory, 0, size);
338 if (!mmu) {
339 *dma_handle = bus;
340 return memory;
341 }
342 }
343
344 if (ops->alloc_coherent) {
345 free_pages((unsigned long)memory, get_order(size));
346 gfp &= ~(GFP_DMA|GFP_DMA32);
347 return ops->alloc_coherent(dev, size, dma_handle, gfp);
348 }
349
350 if (ops->map_simple) {
351 *dma_handle = ops->map_simple(dev, virt_to_phys(memory),
352 size,
353 PCI_DMA_BIDIRECTIONAL);
354 if (*dma_handle != bad_dma_address)
355 return memory;
356 }
357
358 if (panic_on_overflow)
359 panic("dma_alloc_coherent: IOMMU overflow by %lu bytes\n",
360 (unsigned long)size);
361 free_pages((unsigned long)memory, get_order(size));
362 return NULL;
363}
364EXPORT_SYMBOL(dma_alloc_coherent);
365
366/*
367 * Unmap coherent memory.
368 * The caller must ensure that the device has finished accessing the mapping.
369 */
370void dma_free_coherent(struct device *dev, size_t size,
371 void *vaddr, dma_addr_t bus)
372{
373 struct dma_mapping_ops *ops = get_dma_ops(dev);
374
375 int order = get_order(size);
376 WARN_ON(irqs_disabled()); /* for portability */
377 if (dma_release_from_coherent(dev, order, vaddr))
378 return;
379 if (ops->unmap_single)
380 ops->unmap_single(dev, bus, size, 0);
381 free_pages((unsigned long)vaddr, order);
382}
383EXPORT_SYMBOL(dma_free_coherent);
384
385static int __init pci_iommu_init(void) 276static int __init pci_iommu_init(void)
386{ 277{
387 calgary_iommu_init(); 278 calgary_iommu_init();
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 49285f8fd4d5..e3f75bbcedea 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -27,8 +27,8 @@
27#include <linux/scatterlist.h> 27#include <linux/scatterlist.h>
28#include <linux/iommu-helper.h> 28#include <linux/iommu-helper.h>
29#include <linux/sysdev.h> 29#include <linux/sysdev.h>
30#include <linux/io.h>
30#include <asm/atomic.h> 31#include <asm/atomic.h>
31#include <asm/io.h>
32#include <asm/mtrr.h> 32#include <asm/mtrr.h>
33#include <asm/pgtable.h> 33#include <asm/pgtable.h>
34#include <asm/proto.h> 34#include <asm/proto.h>
@@ -80,9 +80,10 @@ AGPEXTERN int agp_memory_reserved;
80AGPEXTERN __u32 *agp_gatt_table; 80AGPEXTERN __u32 *agp_gatt_table;
81 81
82static unsigned long next_bit; /* protected by iommu_bitmap_lock */ 82static unsigned long next_bit; /* protected by iommu_bitmap_lock */
83static int need_flush; /* global flush state. set for each gart wrap */ 83static bool need_flush; /* global flush state. set for each gart wrap */
84 84
85static unsigned long alloc_iommu(struct device *dev, int size) 85static unsigned long alloc_iommu(struct device *dev, int size,
86 unsigned long align_mask)
86{ 87{
87 unsigned long offset, flags; 88 unsigned long offset, flags;
88 unsigned long boundary_size; 89 unsigned long boundary_size;
@@ -90,26 +91,27 @@ static unsigned long alloc_iommu(struct device *dev, int size)
90 91
91 base_index = ALIGN(iommu_bus_base & dma_get_seg_boundary(dev), 92 base_index = ALIGN(iommu_bus_base & dma_get_seg_boundary(dev),
92 PAGE_SIZE) >> PAGE_SHIFT; 93 PAGE_SIZE) >> PAGE_SHIFT;
93 boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, 94 boundary_size = ALIGN((unsigned long long)dma_get_seg_boundary(dev) + 1,
94 PAGE_SIZE) >> PAGE_SHIFT; 95 PAGE_SIZE) >> PAGE_SHIFT;
95 96
96 spin_lock_irqsave(&iommu_bitmap_lock, flags); 97 spin_lock_irqsave(&iommu_bitmap_lock, flags);
97 offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, next_bit, 98 offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, next_bit,
98 size, base_index, boundary_size, 0); 99 size, base_index, boundary_size, align_mask);
99 if (offset == -1) { 100 if (offset == -1) {
100 need_flush = 1; 101 need_flush = true;
101 offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, 0, 102 offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, 0,
102 size, base_index, boundary_size, 0); 103 size, base_index, boundary_size,
104 align_mask);
103 } 105 }
104 if (offset != -1) { 106 if (offset != -1) {
105 next_bit = offset+size; 107 next_bit = offset+size;
106 if (next_bit >= iommu_pages) { 108 if (next_bit >= iommu_pages) {
107 next_bit = 0; 109 next_bit = 0;
108 need_flush = 1; 110 need_flush = true;
109 } 111 }
110 } 112 }
111 if (iommu_fullflush) 113 if (iommu_fullflush)
112 need_flush = 1; 114 need_flush = true;
113 spin_unlock_irqrestore(&iommu_bitmap_lock, flags); 115 spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
114 116
115 return offset; 117 return offset;
@@ -134,7 +136,7 @@ static void flush_gart(void)
134 spin_lock_irqsave(&iommu_bitmap_lock, flags); 136 spin_lock_irqsave(&iommu_bitmap_lock, flags);
135 if (need_flush) { 137 if (need_flush) {
136 k8_flush_garts(); 138 k8_flush_garts();
137 need_flush = 0; 139 need_flush = false;
138 } 140 }
139 spin_unlock_irqrestore(&iommu_bitmap_lock, flags); 141 spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
140} 142}
@@ -173,7 +175,8 @@ static void dump_leak(void)
173 iommu_leak_pages); 175 iommu_leak_pages);
174 for (i = 0; i < iommu_leak_pages; i += 2) { 176 for (i = 0; i < iommu_leak_pages; i += 2) {
175 printk(KERN_DEBUG "%lu: ", iommu_pages-i); 177 printk(KERN_DEBUG "%lu: ", iommu_pages-i);
176 printk_address((unsigned long) iommu_leak_tab[iommu_pages-i], 0); 178 printk_address((unsigned long) iommu_leak_tab[iommu_pages-i],
179 0);
177 printk(KERN_CONT "%c", (i+1)%2 == 0 ? '\n' : ' '); 180 printk(KERN_CONT "%c", (i+1)%2 == 0 ? '\n' : ' ');
178 } 181 }
179 printk(KERN_DEBUG "\n"); 182 printk(KERN_DEBUG "\n");
@@ -212,34 +215,24 @@ static void iommu_full(struct device *dev, size_t size, int dir)
212static inline int 215static inline int
213need_iommu(struct device *dev, unsigned long addr, size_t size) 216need_iommu(struct device *dev, unsigned long addr, size_t size)
214{ 217{
215 u64 mask = *dev->dma_mask; 218 return force_iommu ||
216 int high = addr + size > mask; 219 !is_buffer_dma_capable(*dev->dma_mask, addr, size);
217 int mmu = high;
218
219 if (force_iommu)
220 mmu = 1;
221
222 return mmu;
223} 220}
224 221
225static inline int 222static inline int
226nonforced_iommu(struct device *dev, unsigned long addr, size_t size) 223nonforced_iommu(struct device *dev, unsigned long addr, size_t size)
227{ 224{
228 u64 mask = *dev->dma_mask; 225 return !is_buffer_dma_capable(*dev->dma_mask, addr, size);
229 int high = addr + size > mask;
230 int mmu = high;
231
232 return mmu;
233} 226}
234 227
235/* Map a single continuous physical area into the IOMMU. 228/* Map a single continuous physical area into the IOMMU.
236 * Caller needs to check if the iommu is needed and flush. 229 * Caller needs to check if the iommu is needed and flush.
237 */ 230 */
238static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem, 231static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem,
239 size_t size, int dir) 232 size_t size, int dir, unsigned long align_mask)
240{ 233{
241 unsigned long npages = iommu_num_pages(phys_mem, size); 234 unsigned long npages = iommu_num_pages(phys_mem, size, PAGE_SIZE);
242 unsigned long iommu_page = alloc_iommu(dev, npages); 235 unsigned long iommu_page = alloc_iommu(dev, npages, align_mask);
243 int i; 236 int i;
244 237
245 if (iommu_page == -1) { 238 if (iommu_page == -1) {
@@ -259,16 +252,6 @@ static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem,
259 return iommu_bus_base + iommu_page*PAGE_SIZE + (phys_mem & ~PAGE_MASK); 252 return iommu_bus_base + iommu_page*PAGE_SIZE + (phys_mem & ~PAGE_MASK);
260} 253}
261 254
262static dma_addr_t
263gart_map_simple(struct device *dev, phys_addr_t paddr, size_t size, int dir)
264{
265 dma_addr_t map = dma_map_area(dev, paddr, size, dir);
266
267 flush_gart();
268
269 return map;
270}
271
272/* Map a single area into the IOMMU */ 255/* Map a single area into the IOMMU */
273static dma_addr_t 256static dma_addr_t
274gart_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir) 257gart_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir)
@@ -276,12 +259,13 @@ gart_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir)
276 unsigned long bus; 259 unsigned long bus;
277 260
278 if (!dev) 261 if (!dev)
279 dev = &fallback_dev; 262 dev = &x86_dma_fallback_dev;
280 263
281 if (!need_iommu(dev, paddr, size)) 264 if (!need_iommu(dev, paddr, size))
282 return paddr; 265 return paddr;
283 266
284 bus = gart_map_simple(dev, paddr, size, dir); 267 bus = dma_map_area(dev, paddr, size, dir, 0);
268 flush_gart();
285 269
286 return bus; 270 return bus;
287} 271}
@@ -301,7 +285,7 @@ static void gart_unmap_single(struct device *dev, dma_addr_t dma_addr,
301 return; 285 return;
302 286
303 iommu_page = (dma_addr - iommu_bus_base)>>PAGE_SHIFT; 287 iommu_page = (dma_addr - iommu_bus_base)>>PAGE_SHIFT;
304 npages = iommu_num_pages(dma_addr, size); 288 npages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
305 for (i = 0; i < npages; i++) { 289 for (i = 0; i < npages; i++) {
306 iommu_gatt_base[iommu_page + i] = gart_unmapped_entry; 290 iommu_gatt_base[iommu_page + i] = gart_unmapped_entry;
307 CLEAR_LEAK(iommu_page + i); 291 CLEAR_LEAK(iommu_page + i);
@@ -340,7 +324,7 @@ static int dma_map_sg_nonforce(struct device *dev, struct scatterlist *sg,
340 unsigned long addr = sg_phys(s); 324 unsigned long addr = sg_phys(s);
341 325
342 if (nonforced_iommu(dev, addr, s->length)) { 326 if (nonforced_iommu(dev, addr, s->length)) {
343 addr = dma_map_area(dev, addr, s->length, dir); 327 addr = dma_map_area(dev, addr, s->length, dir, 0);
344 if (addr == bad_dma_address) { 328 if (addr == bad_dma_address) {
345 if (i > 0) 329 if (i > 0)
346 gart_unmap_sg(dev, sg, i, dir); 330 gart_unmap_sg(dev, sg, i, dir);
@@ -362,7 +346,7 @@ static int __dma_map_cont(struct device *dev, struct scatterlist *start,
362 int nelems, struct scatterlist *sout, 346 int nelems, struct scatterlist *sout,
363 unsigned long pages) 347 unsigned long pages)
364{ 348{
365 unsigned long iommu_start = alloc_iommu(dev, pages); 349 unsigned long iommu_start = alloc_iommu(dev, pages, 0);
366 unsigned long iommu_page = iommu_start; 350 unsigned long iommu_page = iommu_start;
367 struct scatterlist *s; 351 struct scatterlist *s;
368 int i; 352 int i;
@@ -384,7 +368,7 @@ static int __dma_map_cont(struct device *dev, struct scatterlist *start,
384 } 368 }
385 369
386 addr = phys_addr; 370 addr = phys_addr;
387 pages = iommu_num_pages(s->offset, s->length); 371 pages = iommu_num_pages(s->offset, s->length, PAGE_SIZE);
388 while (pages--) { 372 while (pages--) {
389 iommu_gatt_base[iommu_page] = GPTE_ENCODE(addr); 373 iommu_gatt_base[iommu_page] = GPTE_ENCODE(addr);
390 SET_LEAK(iommu_page); 374 SET_LEAK(iommu_page);
@@ -427,7 +411,7 @@ gart_map_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
427 return 0; 411 return 0;
428 412
429 if (!dev) 413 if (!dev)
430 dev = &fallback_dev; 414 dev = &x86_dma_fallback_dev;
431 415
432 out = 0; 416 out = 0;
433 start = 0; 417 start = 0;
@@ -467,7 +451,7 @@ gart_map_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
467 451
468 seg_size += s->length; 452 seg_size += s->length;
469 need = nextneed; 453 need = nextneed;
470 pages += iommu_num_pages(s->offset, s->length); 454 pages += iommu_num_pages(s->offset, s->length, PAGE_SIZE);
471 ps = s; 455 ps = s;
472 } 456 }
473 if (dma_map_cont(dev, start_sg, i - start, sgmap, pages, need) < 0) 457 if (dma_map_cont(dev, start_sg, i - start, sgmap, pages, need) < 0)
@@ -499,6 +483,46 @@ error:
499 return 0; 483 return 0;
500} 484}
501 485
486/* allocate and map a coherent mapping */
487static void *
488gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
489 gfp_t flag)
490{
491 dma_addr_t paddr;
492 unsigned long align_mask;
493 struct page *page;
494
495 if (force_iommu && !(flag & GFP_DMA)) {
496 flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
497 page = alloc_pages(flag | __GFP_ZERO, get_order(size));
498 if (!page)
499 return NULL;
500
501 align_mask = (1UL << get_order(size)) - 1;
502 paddr = dma_map_area(dev, page_to_phys(page), size,
503 DMA_BIDIRECTIONAL, align_mask);
504
505 flush_gart();
506 if (paddr != bad_dma_address) {
507 *dma_addr = paddr;
508 return page_address(page);
509 }
510 __free_pages(page, get_order(size));
511 } else
512 return dma_generic_alloc_coherent(dev, size, dma_addr, flag);
513
514 return NULL;
515}
516
517/* free a coherent mapping */
518static void
519gart_free_coherent(struct device *dev, size_t size, void *vaddr,
520 dma_addr_t dma_addr)
521{
522 gart_unmap_single(dev, dma_addr, size, DMA_BIDIRECTIONAL);
523 free_pages((unsigned long)vaddr, get_order(size));
524}
525
502static int no_agp; 526static int no_agp;
503 527
504static __init unsigned long check_iommu_size(unsigned long aper, u64 aper_size) 528static __init unsigned long check_iommu_size(unsigned long aper, u64 aper_size)
@@ -626,7 +650,6 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
626 struct pci_dev *dev; 650 struct pci_dev *dev;
627 void *gatt; 651 void *gatt;
628 int i, error; 652 int i, error;
629 unsigned long start_pfn, end_pfn;
630 653
631 printk(KERN_INFO "PCI-DMA: Disabling AGP.\n"); 654 printk(KERN_INFO "PCI-DMA: Disabling AGP.\n");
632 aper_size = aper_base = info->aper_size = 0; 655 aper_size = aper_base = info->aper_size = 0;
@@ -650,13 +673,13 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
650 info->aper_size = aper_size >> 20; 673 info->aper_size = aper_size >> 20;
651 674
652 gatt_size = (aper_size >> PAGE_SHIFT) * sizeof(u32); 675 gatt_size = (aper_size >> PAGE_SHIFT) * sizeof(u32);
653 gatt = (void *)__get_free_pages(GFP_KERNEL, get_order(gatt_size)); 676 gatt = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
677 get_order(gatt_size));
654 if (!gatt) 678 if (!gatt)
655 panic("Cannot allocate GATT table"); 679 panic("Cannot allocate GATT table");
656 if (set_memory_uc((unsigned long)gatt, gatt_size >> PAGE_SHIFT)) 680 if (set_memory_uc((unsigned long)gatt, gatt_size >> PAGE_SHIFT))
657 panic("Could not set GART PTEs to uncacheable pages"); 681 panic("Could not set GART PTEs to uncacheable pages");
658 682
659 memset(gatt, 0, gatt_size);
660 agp_gatt_table = gatt; 683 agp_gatt_table = gatt;
661 684
662 enable_gart_translations(); 685 enable_gart_translations();
@@ -665,19 +688,14 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
665 if (!error) 688 if (!error)
666 error = sysdev_register(&device_gart); 689 error = sysdev_register(&device_gart);
667 if (error) 690 if (error)
668 panic("Could not register gart_sysdev -- would corrupt data on next suspend"); 691 panic("Could not register gart_sysdev -- "
692 "would corrupt data on next suspend");
669 693
670 flush_gart(); 694 flush_gart();
671 695
672 printk(KERN_INFO "PCI-DMA: aperture base @ %x size %u KB\n", 696 printk(KERN_INFO "PCI-DMA: aperture base @ %x size %u KB\n",
673 aper_base, aper_size>>10); 697 aper_base, aper_size>>10);
674 698
675 /* need to map that range */
676 end_pfn = (aper_base>>PAGE_SHIFT) + (aper_size>>PAGE_SHIFT);
677 if (end_pfn > max_low_pfn_mapped) {
678 start_pfn = (aper_base>>PAGE_SHIFT);
679 init_memory_mapping(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
680 }
681 return 0; 699 return 0;
682 700
683 nommu: 701 nommu:
@@ -687,20 +705,13 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
687 return -1; 705 return -1;
688} 706}
689 707
690extern int agp_amd64_init(void);
691
692static struct dma_mapping_ops gart_dma_ops = { 708static struct dma_mapping_ops gart_dma_ops = {
693 .map_single = gart_map_single, 709 .map_single = gart_map_single,
694 .map_simple = gart_map_simple,
695 .unmap_single = gart_unmap_single, 710 .unmap_single = gart_unmap_single,
696 .sync_single_for_cpu = NULL,
697 .sync_single_for_device = NULL,
698 .sync_single_range_for_cpu = NULL,
699 .sync_single_range_for_device = NULL,
700 .sync_sg_for_cpu = NULL,
701 .sync_sg_for_device = NULL,
702 .map_sg = gart_map_sg, 711 .map_sg = gart_map_sg,
703 .unmap_sg = gart_unmap_sg, 712 .unmap_sg = gart_unmap_sg,
713 .alloc_coherent = gart_alloc_coherent,
714 .free_coherent = gart_free_coherent,
704}; 715};
705 716
706void gart_iommu_shutdown(void) 717void gart_iommu_shutdown(void)
@@ -727,7 +738,8 @@ void __init gart_iommu_init(void)
727{ 738{
728 struct agp_kern_info info; 739 struct agp_kern_info info;
729 unsigned long iommu_start; 740 unsigned long iommu_start;
730 unsigned long aper_size; 741 unsigned long aper_base, aper_size;
742 unsigned long start_pfn, end_pfn;
731 unsigned long scratch; 743 unsigned long scratch;
732 long i; 744 long i;
733 745
@@ -759,30 +771,35 @@ void __init gart_iommu_init(void)
759 (no_agp && init_k8_gatt(&info) < 0)) { 771 (no_agp && init_k8_gatt(&info) < 0)) {
760 if (max_pfn > MAX_DMA32_PFN) { 772 if (max_pfn > MAX_DMA32_PFN) {
761 printk(KERN_WARNING "More than 4GB of memory " 773 printk(KERN_WARNING "More than 4GB of memory "
762 "but GART IOMMU not available.\n" 774 "but GART IOMMU not available.\n");
763 KERN_WARNING "falling back to iommu=soft.\n"); 775 printk(KERN_WARNING "falling back to iommu=soft.\n");
764 } 776 }
765 return; 777 return;
766 } 778 }
767 779
780 /* need to map that range */
781 aper_size = info.aper_size << 20;
782 aper_base = info.aper_base;
783 end_pfn = (aper_base>>PAGE_SHIFT) + (aper_size>>PAGE_SHIFT);
784 if (end_pfn > max_low_pfn_mapped) {
785 start_pfn = (aper_base>>PAGE_SHIFT);
786 init_memory_mapping(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
787 }
788
768 printk(KERN_INFO "PCI-DMA: using GART IOMMU.\n"); 789 printk(KERN_INFO "PCI-DMA: using GART IOMMU.\n");
769 aper_size = info.aper_size * 1024 * 1024;
770 iommu_size = check_iommu_size(info.aper_base, aper_size); 790 iommu_size = check_iommu_size(info.aper_base, aper_size);
771 iommu_pages = iommu_size >> PAGE_SHIFT; 791 iommu_pages = iommu_size >> PAGE_SHIFT;
772 792
773 iommu_gart_bitmap = (void *) __get_free_pages(GFP_KERNEL, 793 iommu_gart_bitmap = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO,
774 get_order(iommu_pages/8)); 794 get_order(iommu_pages/8));
775 if (!iommu_gart_bitmap) 795 if (!iommu_gart_bitmap)
776 panic("Cannot allocate iommu bitmap\n"); 796 panic("Cannot allocate iommu bitmap\n");
777 memset(iommu_gart_bitmap, 0, iommu_pages/8);
778 797
779#ifdef CONFIG_IOMMU_LEAK 798#ifdef CONFIG_IOMMU_LEAK
780 if (leak_trace) { 799 if (leak_trace) {
781 iommu_leak_tab = (void *)__get_free_pages(GFP_KERNEL, 800 iommu_leak_tab = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
782 get_order(iommu_pages*sizeof(void *))); 801 get_order(iommu_pages*sizeof(void *)));
783 if (iommu_leak_tab) 802 if (!iommu_leak_tab)
784 memset(iommu_leak_tab, 0, iommu_pages * 8);
785 else
786 printk(KERN_DEBUG 803 printk(KERN_DEBUG
787 "PCI-DMA: Cannot allocate leak trace area\n"); 804 "PCI-DMA: Cannot allocate leak trace area\n");
788 } 805 }
@@ -792,7 +809,7 @@ void __init gart_iommu_init(void)
792 * Out of IOMMU space handling. 809 * Out of IOMMU space handling.
793 * Reserve some invalid pages at the beginning of the GART. 810 * Reserve some invalid pages at the beginning of the GART.
794 */ 811 */
795 set_bit_string(iommu_gart_bitmap, 0, EMERGENCY_PAGES); 812 iommu_area_reserve(iommu_gart_bitmap, 0, EMERGENCY_PAGES);
796 813
797 agp_memory_reserved = iommu_size; 814 agp_memory_reserved = iommu_size;
798 printk(KERN_INFO 815 printk(KERN_INFO
@@ -850,7 +867,8 @@ void __init gart_parse_options(char *p)
850 if (!strncmp(p, "leak", 4)) { 867 if (!strncmp(p, "leak", 4)) {
851 leak_trace = 1; 868 leak_trace = 1;
852 p += 4; 869 p += 4;
853 if (*p == '=') ++p; 870 if (*p == '=')
871 ++p;
854 if (isdigit(*p) && get_option(&p, &arg)) 872 if (isdigit(*p) && get_option(&p, &arg))
855 iommu_leak_pages = arg; 873 iommu_leak_pages = arg;
856 } 874 }
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index 3f91f71cdc3e..c70ab5a5d4c8 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -14,7 +14,7 @@
14static int 14static int
15check_addr(char *name, struct device *hwdev, dma_addr_t bus, size_t size) 15check_addr(char *name, struct device *hwdev, dma_addr_t bus, size_t size)
16{ 16{
17 if (hwdev && bus + size > *hwdev->dma_mask) { 17 if (hwdev && !is_buffer_dma_capable(*hwdev->dma_mask, bus, size)) {
18 if (*hwdev->dma_mask >= DMA_32BIT_MASK) 18 if (*hwdev->dma_mask >= DMA_32BIT_MASK)
19 printk(KERN_ERR 19 printk(KERN_ERR
20 "nommu_%s: overflow %Lx+%zu of device mask %Lx\n", 20 "nommu_%s: overflow %Lx+%zu of device mask %Lx\n",
@@ -72,7 +72,15 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg,
72 return nents; 72 return nents;
73} 73}
74 74
75static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr,
76 dma_addr_t dma_addr)
77{
78 free_pages((unsigned long)vaddr, get_order(size));
79}
80
75struct dma_mapping_ops nommu_dma_ops = { 81struct dma_mapping_ops nommu_dma_ops = {
82 .alloc_coherent = dma_generic_alloc_coherent,
83 .free_coherent = nommu_free_coherent,
76 .map_single = nommu_map_single, 84 .map_single = nommu_map_single,
77 .map_sg = nommu_map_sg, 85 .map_sg = nommu_map_sg,
78 .is_phys = 1, 86 .is_phys = 1,
diff --git a/arch/x86/kernel/pcspeaker.c b/arch/x86/kernel/pcspeaker.c
index bc1f2d3ea277..a311ffcaad16 100644
--- a/arch/x86/kernel/pcspeaker.c
+++ b/arch/x86/kernel/pcspeaker.c
@@ -1,20 +1,13 @@
1#include <linux/platform_device.h> 1#include <linux/platform_device.h>
2#include <linux/errno.h> 2#include <linux/err.h>
3#include <linux/init.h> 3#include <linux/init.h>
4 4
5static __init int add_pcspkr(void) 5static __init int add_pcspkr(void)
6{ 6{
7 struct platform_device *pd; 7 struct platform_device *pd;
8 int ret;
9 8
10 pd = platform_device_alloc("pcspkr", -1); 9 pd = platform_device_register_simple("pcspkr", -1, NULL, 0);
11 if (!pd)
12 return -ENOMEM;
13 10
14 ret = platform_device_add(pd); 11 return IS_ERR(pd) ? PTR_ERR(pd) : 0;
15 if (ret)
16 platform_device_put(pd);
17
18 return ret;
19} 12}
20device_initcall(add_pcspkr); 13device_initcall(add_pcspkr);
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 7fc4d5b0a6a0..c622772744d8 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -15,7 +15,6 @@ unsigned long idle_nomwait;
15EXPORT_SYMBOL(idle_nomwait); 15EXPORT_SYMBOL(idle_nomwait);
16 16
17struct kmem_cache *task_xstate_cachep; 17struct kmem_cache *task_xstate_cachep;
18static int force_mwait __cpuinitdata;
19 18
20int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 19int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
21{ 20{
@@ -185,7 +184,8 @@ static void mwait_idle(void)
185static void poll_idle(void) 184static void poll_idle(void)
186{ 185{
187 local_irq_enable(); 186 local_irq_enable();
188 cpu_relax(); 187 while (!need_resched())
188 cpu_relax();
189} 189}
190 190
191/* 191/*
@@ -246,6 +246,14 @@ static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
246 return 1; 246 return 1;
247} 247}
248 248
249static cpumask_t c1e_mask = CPU_MASK_NONE;
250static int c1e_detected;
251
252void c1e_remove_cpu(int cpu)
253{
254 cpu_clear(cpu, c1e_mask);
255}
256
249/* 257/*
250 * C1E aware idle routine. We check for C1E active in the interrupt 258 * C1E aware idle routine. We check for C1E active in the interrupt
251 * pending message MSR. If we detect C1E, then we handle it the same 259 * pending message MSR. If we detect C1E, then we handle it the same
@@ -253,9 +261,6 @@ static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
253 */ 261 */
254static void c1e_idle(void) 262static void c1e_idle(void)
255{ 263{
256 static cpumask_t c1e_mask = CPU_MASK_NONE;
257 static int c1e_detected;
258
259 if (need_resched()) 264 if (need_resched())
260 return; 265 return;
261 266
@@ -265,8 +270,10 @@ static void c1e_idle(void)
265 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); 270 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
266 if (lo & K8_INTP_C1E_ACTIVE_MASK) { 271 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
267 c1e_detected = 1; 272 c1e_detected = 1;
268 mark_tsc_unstable("TSC halt in C1E"); 273 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
269 printk(KERN_INFO "System has C1E enabled\n"); 274 mark_tsc_unstable("TSC halt in AMD C1E");
275 printk(KERN_INFO "System has AMD C1E enabled\n");
276 set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
270 } 277 }
271 } 278 }
272 279
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 3b7a1ddcc0bc..0a1302fe6d45 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -37,6 +37,7 @@
37#include <linux/tick.h> 37#include <linux/tick.h>
38#include <linux/percpu.h> 38#include <linux/percpu.h>
39#include <linux/prctl.h> 39#include <linux/prctl.h>
40#include <linux/dmi.h>
40 41
41#include <asm/uaccess.h> 42#include <asm/uaccess.h>
42#include <asm/pgtable.h> 43#include <asm/pgtable.h>
@@ -55,6 +56,9 @@
55#include <asm/tlbflush.h> 56#include <asm/tlbflush.h>
56#include <asm/cpu.h> 57#include <asm/cpu.h>
57#include <asm/kdebug.h> 58#include <asm/kdebug.h>
59#include <asm/idle.h>
60#include <asm/syscalls.h>
61#include <asm/smp.h>
58 62
59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 63asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
60 64
@@ -72,46 +76,12 @@ unsigned long thread_saved_pc(struct task_struct *tsk)
72 return ((unsigned long *)tsk->thread.sp)[3]; 76 return ((unsigned long *)tsk->thread.sp)[3];
73} 77}
74 78
75#ifdef CONFIG_HOTPLUG_CPU 79#ifndef CONFIG_SMP
76#include <asm/nmi.h>
77
78static void cpu_exit_clear(void)
79{
80 int cpu = raw_smp_processor_id();
81
82 idle_task_exit();
83
84 cpu_uninit();
85 irq_ctx_exit(cpu);
86
87 cpu_clear(cpu, cpu_callout_map);
88 cpu_clear(cpu, cpu_callin_map);
89
90 numa_remove_cpu(cpu);
91}
92
93/* We don't actually take CPU down, just spin without interrupts. */
94static inline void play_dead(void)
95{
96 /* This must be done before dead CPU ack */
97 cpu_exit_clear();
98 mb();
99 /* Ack it */
100 __get_cpu_var(cpu_state) = CPU_DEAD;
101
102 /*
103 * With physical CPU hotplug, we should halt the cpu
104 */
105 local_irq_disable();
106 /* mask all interrupts, flush any and all caches, and halt */
107 wbinvd_halt();
108}
109#else
110static inline void play_dead(void) 80static inline void play_dead(void)
111{ 81{
112 BUG(); 82 BUG();
113} 83}
114#endif /* CONFIG_HOTPLUG_CPU */ 84#endif
115 85
116/* 86/*
117 * The idle thread. There's no useful work to be 87 * The idle thread. There's no useful work to be
@@ -153,12 +123,13 @@ void cpu_idle(void)
153 } 123 }
154} 124}
155 125
156void __show_registers(struct pt_regs *regs, int all) 126void __show_regs(struct pt_regs *regs, int all)
157{ 127{
158 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; 128 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
159 unsigned long d0, d1, d2, d3, d6, d7; 129 unsigned long d0, d1, d2, d3, d6, d7;
160 unsigned long sp; 130 unsigned long sp;
161 unsigned short ss, gs; 131 unsigned short ss, gs;
132 const char *board;
162 133
163 if (user_mode_vm(regs)) { 134 if (user_mode_vm(regs)) {
164 sp = regs->sp; 135 sp = regs->sp;
@@ -171,11 +142,15 @@ void __show_registers(struct pt_regs *regs, int all)
171 } 142 }
172 143
173 printk("\n"); 144 printk("\n");
174 printk("Pid: %d, comm: %s %s (%s %.*s)\n", 145
146 board = dmi_get_system_info(DMI_PRODUCT_NAME);
147 if (!board)
148 board = "";
149 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
175 task_pid_nr(current), current->comm, 150 task_pid_nr(current), current->comm,
176 print_tainted(), init_utsname()->release, 151 print_tainted(), init_utsname()->release,
177 (int)strcspn(init_utsname()->version, " "), 152 (int)strcspn(init_utsname()->version, " "),
178 init_utsname()->version); 153 init_utsname()->version, board);
179 154
180 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n", 155 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
181 (u16)regs->cs, regs->ip, regs->flags, 156 (u16)regs->cs, regs->ip, regs->flags,
@@ -214,7 +189,7 @@ void __show_registers(struct pt_regs *regs, int all)
214 189
215void show_regs(struct pt_regs *regs) 190void show_regs(struct pt_regs *regs)
216{ 191{
217 __show_registers(regs, 1); 192 __show_regs(regs, 1);
218 show_trace(NULL, regs, &regs->sp, regs->bp); 193 show_trace(NULL, regs, &regs->sp, regs->bp);
219} 194}
220 195
@@ -275,6 +250,14 @@ void exit_thread(void)
275 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET; 250 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
276 put_cpu(); 251 put_cpu();
277 } 252 }
253#ifdef CONFIG_X86_DS
254 /* Free any DS contexts that have not been properly released. */
255 if (unlikely(current->thread.ds_ctx)) {
256 /* we clear debugctl to make sure DS is not used. */
257 update_debugctlmsr(0);
258 ds_free(current->thread.ds_ctx);
259 }
260#endif /* CONFIG_X86_DS */
278} 261}
279 262
280void flush_thread(void) 263void flush_thread(void)
@@ -436,6 +419,35 @@ int set_tsc_mode(unsigned int val)
436 return 0; 419 return 0;
437} 420}
438 421
422#ifdef CONFIG_X86_DS
423static int update_debugctl(struct thread_struct *prev,
424 struct thread_struct *next, unsigned long debugctl)
425{
426 unsigned long ds_prev = 0;
427 unsigned long ds_next = 0;
428
429 if (prev->ds_ctx)
430 ds_prev = (unsigned long)prev->ds_ctx->ds;
431 if (next->ds_ctx)
432 ds_next = (unsigned long)next->ds_ctx->ds;
433
434 if (ds_next != ds_prev) {
435 /* we clear debugctl to make sure DS
436 * is not in use when we change it */
437 debugctl = 0;
438 update_debugctlmsr(0);
439 wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
440 }
441 return debugctl;
442}
443#else
444static int update_debugctl(struct thread_struct *prev,
445 struct thread_struct *next, unsigned long debugctl)
446{
447 return debugctl;
448}
449#endif /* CONFIG_X86_DS */
450
439static noinline void 451static noinline void
440__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, 452__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
441 struct tss_struct *tss) 453 struct tss_struct *tss)
@@ -446,14 +458,7 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
446 prev = &prev_p->thread; 458 prev = &prev_p->thread;
447 next = &next_p->thread; 459 next = &next_p->thread;
448 460
449 debugctl = prev->debugctlmsr; 461 debugctl = update_debugctl(prev, next, prev->debugctlmsr);
450 if (next->ds_area_msr != prev->ds_area_msr) {
451 /* we clear debugctl to make sure DS
452 * is not in use when we change it */
453 debugctl = 0;
454 update_debugctlmsr(0);
455 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
456 }
457 462
458 if (next->debugctlmsr != debugctl) 463 if (next->debugctlmsr != debugctl)
459 update_debugctlmsr(next->debugctlmsr); 464 update_debugctlmsr(next->debugctlmsr);
@@ -477,13 +482,13 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
477 hard_enable_TSC(); 482 hard_enable_TSC();
478 } 483 }
479 484
480#ifdef X86_BTS 485#ifdef CONFIG_X86_PTRACE_BTS
481 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) 486 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
482 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); 487 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
483 488
484 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) 489 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
485 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); 490 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
486#endif 491#endif /* CONFIG_X86_PTRACE_BTS */
487 492
488 493
489 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { 494 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 71553b664e2a..cd8c0ed02b7e 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -37,11 +37,11 @@
37#include <linux/kdebug.h> 37#include <linux/kdebug.h>
38#include <linux/tick.h> 38#include <linux/tick.h>
39#include <linux/prctl.h> 39#include <linux/prctl.h>
40#include <linux/uaccess.h>
41#include <linux/io.h>
40 42
41#include <asm/uaccess.h>
42#include <asm/pgtable.h> 43#include <asm/pgtable.h>
43#include <asm/system.h> 44#include <asm/system.h>
44#include <asm/io.h>
45#include <asm/processor.h> 45#include <asm/processor.h>
46#include <asm/i387.h> 46#include <asm/i387.h>
47#include <asm/mmu_context.h> 47#include <asm/mmu_context.h>
@@ -51,6 +51,7 @@
51#include <asm/proto.h> 51#include <asm/proto.h>
52#include <asm/ia32.h> 52#include <asm/ia32.h>
53#include <asm/idle.h> 53#include <asm/idle.h>
54#include <asm/syscalls.h>
54 55
55asmlinkage extern void ret_from_fork(void); 56asmlinkage extern void ret_from_fork(void);
56 57
@@ -85,28 +86,12 @@ void exit_idle(void)
85 __exit_idle(); 86 __exit_idle();
86} 87}
87 88
88#ifdef CONFIG_HOTPLUG_CPU 89#ifndef CONFIG_SMP
89DECLARE_PER_CPU(int, cpu_state);
90
91#include <asm/nmi.h>
92/* We halt the CPU with physical CPU hotplug */
93static inline void play_dead(void)
94{
95 idle_task_exit();
96 mb();
97 /* Ack it */
98 __get_cpu_var(cpu_state) = CPU_DEAD;
99
100 local_irq_disable();
101 /* mask all interrupts, flush any and all caches, and halt */
102 wbinvd_halt();
103}
104#else
105static inline void play_dead(void) 90static inline void play_dead(void)
106{ 91{
107 BUG(); 92 BUG();
108} 93}
109#endif /* CONFIG_HOTPLUG_CPU */ 94#endif
110 95
111/* 96/*
112 * The idle thread. There's no useful work to be 97 * The idle thread. There's no useful work to be
@@ -151,7 +136,7 @@ void cpu_idle(void)
151} 136}
152 137
153/* Prints also some state that isn't saved in the pt_regs */ 138/* Prints also some state that isn't saved in the pt_regs */
154void __show_regs(struct pt_regs * regs) 139void __show_regs(struct pt_regs *regs, int all)
155{ 140{
156 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; 141 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
157 unsigned long d0, d1, d2, d3, d6, d7; 142 unsigned long d0, d1, d2, d3, d6, d7;
@@ -160,60 +145,65 @@ void __show_regs(struct pt_regs * regs)
160 145
161 printk("\n"); 146 printk("\n");
162 print_modules(); 147 print_modules();
163 printk("Pid: %d, comm: %.20s %s %s %.*s\n", 148 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s\n",
164 current->pid, current->comm, print_tainted(), 149 current->pid, current->comm, print_tainted(),
165 init_utsname()->release, 150 init_utsname()->release,
166 (int)strcspn(init_utsname()->version, " "), 151 (int)strcspn(init_utsname()->version, " "),
167 init_utsname()->version); 152 init_utsname()->version);
168 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip); 153 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
169 printk_address(regs->ip, 1); 154 printk_address(regs->ip, 1);
170 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp, 155 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
171 regs->flags); 156 regs->sp, regs->flags);
172 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n", 157 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
173 regs->ax, regs->bx, regs->cx); 158 regs->ax, regs->bx, regs->cx);
174 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n", 159 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
175 regs->dx, regs->si, regs->di); 160 regs->dx, regs->si, regs->di);
176 printk("RBP: %016lx R08: %016lx R09: %016lx\n", 161 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
177 regs->bp, regs->r8, regs->r9); 162 regs->bp, regs->r8, regs->r9);
178 printk("R10: %016lx R11: %016lx R12: %016lx\n", 163 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
179 regs->r10, regs->r11, regs->r12); 164 regs->r10, regs->r11, regs->r12);
180 printk("R13: %016lx R14: %016lx R15: %016lx\n", 165 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
181 regs->r13, regs->r14, regs->r15); 166 regs->r13, regs->r14, regs->r15);
182 167
183 asm("movl %%ds,%0" : "=r" (ds)); 168 asm("movl %%ds,%0" : "=r" (ds));
184 asm("movl %%cs,%0" : "=r" (cs)); 169 asm("movl %%cs,%0" : "=r" (cs));
185 asm("movl %%es,%0" : "=r" (es)); 170 asm("movl %%es,%0" : "=r" (es));
186 asm("movl %%fs,%0" : "=r" (fsindex)); 171 asm("movl %%fs,%0" : "=r" (fsindex));
187 asm("movl %%gs,%0" : "=r" (gsindex)); 172 asm("movl %%gs,%0" : "=r" (gsindex));
188 173
189 rdmsrl(MSR_FS_BASE, fs); 174 rdmsrl(MSR_FS_BASE, fs);
190 rdmsrl(MSR_GS_BASE, gs); 175 rdmsrl(MSR_GS_BASE, gs);
191 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); 176 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
177
178 if (!all)
179 return;
192 180
193 cr0 = read_cr0(); 181 cr0 = read_cr0();
194 cr2 = read_cr2(); 182 cr2 = read_cr2();
195 cr3 = read_cr3(); 183 cr3 = read_cr3();
196 cr4 = read_cr4(); 184 cr4 = read_cr4();
197 185
198 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", 186 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
199 fs,fsindex,gs,gsindex,shadowgs); 187 fs, fsindex, gs, gsindex, shadowgs);
200 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0); 188 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
201 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4); 189 es, cr0);
190 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
191 cr4);
202 192
203 get_debugreg(d0, 0); 193 get_debugreg(d0, 0);
204 get_debugreg(d1, 1); 194 get_debugreg(d1, 1);
205 get_debugreg(d2, 2); 195 get_debugreg(d2, 2);
206 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2); 196 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
207 get_debugreg(d3, 3); 197 get_debugreg(d3, 3);
208 get_debugreg(d6, 6); 198 get_debugreg(d6, 6);
209 get_debugreg(d7, 7); 199 get_debugreg(d7, 7);
210 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7); 200 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
211} 201}
212 202
213void show_regs(struct pt_regs *regs) 203void show_regs(struct pt_regs *regs)
214{ 204{
215 printk("CPU %d:", smp_processor_id()); 205 printk(KERN_INFO "CPU %d:", smp_processor_id());
216 __show_regs(regs); 206 __show_regs(regs, 1);
217 show_trace(NULL, regs, (void *)(regs + 1), regs->bp); 207 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
218} 208}
219 209
@@ -238,6 +228,14 @@ void exit_thread(void)
238 t->io_bitmap_max = 0; 228 t->io_bitmap_max = 0;
239 put_cpu(); 229 put_cpu();
240 } 230 }
231#ifdef CONFIG_X86_DS
232 /* Free any DS contexts that have not been properly released. */
233 if (unlikely(t->ds_ctx)) {
234 /* we clear debugctl to make sure DS is not used. */
235 update_debugctlmsr(0);
236 ds_free(t->ds_ctx);
237 }
238#endif /* CONFIG_X86_DS */
241} 239}
242 240
243void flush_thread(void) 241void flush_thread(void)
@@ -313,10 +311,10 @@ void prepare_to_copy(struct task_struct *tsk)
313 311
314int copy_thread(int nr, unsigned long clone_flags, unsigned long sp, 312int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
315 unsigned long unused, 313 unsigned long unused,
316 struct task_struct * p, struct pt_regs * regs) 314 struct task_struct *p, struct pt_regs *regs)
317{ 315{
318 int err; 316 int err;
319 struct pt_regs * childregs; 317 struct pt_regs *childregs;
320 struct task_struct *me = current; 318 struct task_struct *me = current;
321 319
322 childregs = ((struct pt_regs *) 320 childregs = ((struct pt_regs *)
@@ -361,10 +359,10 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
361 if (test_thread_flag(TIF_IA32)) 359 if (test_thread_flag(TIF_IA32))
362 err = do_set_thread_area(p, -1, 360 err = do_set_thread_area(p, -1,
363 (struct user_desc __user *)childregs->si, 0); 361 (struct user_desc __user *)childregs->si, 0);
364 else 362 else
365#endif 363#endif
366 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); 364 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
367 if (err) 365 if (err)
368 goto out; 366 goto out;
369 } 367 }
370 err = 0; 368 err = 0;
@@ -471,13 +469,27 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
471 next = &next_p->thread; 469 next = &next_p->thread;
472 470
473 debugctl = prev->debugctlmsr; 471 debugctl = prev->debugctlmsr;
474 if (next->ds_area_msr != prev->ds_area_msr) { 472
475 /* we clear debugctl to make sure DS 473#ifdef CONFIG_X86_DS
476 * is not in use when we change it */ 474 {
477 debugctl = 0; 475 unsigned long ds_prev = 0, ds_next = 0;
478 update_debugctlmsr(0); 476
479 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr); 477 if (prev->ds_ctx)
478 ds_prev = (unsigned long)prev->ds_ctx->ds;
479 if (next->ds_ctx)
480 ds_next = (unsigned long)next->ds_ctx->ds;
481
482 if (ds_next != ds_prev) {
483 /*
484 * We clear debugctl to make sure DS
485 * is not in use when we change it:
486 */
487 debugctl = 0;
488 update_debugctlmsr(0);
489 wrmsrl(MSR_IA32_DS_AREA, ds_next);
490 }
480 } 491 }
492#endif /* CONFIG_X86_DS */
481 493
482 if (next->debugctlmsr != debugctl) 494 if (next->debugctlmsr != debugctl)
483 update_debugctlmsr(next->debugctlmsr); 495 update_debugctlmsr(next->debugctlmsr);
@@ -515,13 +527,13 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
515 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); 527 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
516 } 528 }
517 529
518#ifdef X86_BTS 530#ifdef CONFIG_X86_PTRACE_BTS
519 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) 531 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
520 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); 532 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
521 533
522 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) 534 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
523 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); 535 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
524#endif 536#endif /* CONFIG_X86_PTRACE_BTS */
525} 537}
526 538
527/* 539/*
@@ -543,7 +555,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
543 unsigned fsindex, gsindex; 555 unsigned fsindex, gsindex;
544 556
545 /* we're going to use this soon, after a few expensive things */ 557 /* we're going to use this soon, after a few expensive things */
546 if (next_p->fpu_counter>5) 558 if (next_p->fpu_counter > 5)
547 prefetch(next->xstate); 559 prefetch(next->xstate);
548 560
549 /* 561 /*
@@ -551,13 +563,13 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
551 */ 563 */
552 load_sp0(tss, next); 564 load_sp0(tss, next);
553 565
554 /* 566 /*
555 * Switch DS and ES. 567 * Switch DS and ES.
556 * This won't pick up thread selector changes, but I guess that is ok. 568 * This won't pick up thread selector changes, but I guess that is ok.
557 */ 569 */
558 savesegment(es, prev->es); 570 savesegment(es, prev->es);
559 if (unlikely(next->es | prev->es)) 571 if (unlikely(next->es | prev->es))
560 loadsegment(es, next->es); 572 loadsegment(es, next->es);
561 573
562 savesegment(ds, prev->ds); 574 savesegment(ds, prev->ds);
563 if (unlikely(next->ds | prev->ds)) 575 if (unlikely(next->ds | prev->ds))
@@ -583,7 +595,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
583 */ 595 */
584 arch_leave_lazy_cpu_mode(); 596 arch_leave_lazy_cpu_mode();
585 597
586 /* 598 /*
587 * Switch FS and GS. 599 * Switch FS and GS.
588 * 600 *
589 * Segment register != 0 always requires a reload. Also 601 * Segment register != 0 always requires a reload. Also
@@ -592,13 +604,13 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
592 */ 604 */
593 if (unlikely(fsindex | next->fsindex | prev->fs)) { 605 if (unlikely(fsindex | next->fsindex | prev->fs)) {
594 loadsegment(fs, next->fsindex); 606 loadsegment(fs, next->fsindex);
595 /* 607 /*
596 * Check if the user used a selector != 0; if yes 608 * Check if the user used a selector != 0; if yes
597 * clear 64bit base, since overloaded base is always 609 * clear 64bit base, since overloaded base is always
598 * mapped to the Null selector 610 * mapped to the Null selector
599 */ 611 */
600 if (fsindex) 612 if (fsindex)
601 prev->fs = 0; 613 prev->fs = 0;
602 } 614 }
603 /* when next process has a 64bit base use it */ 615 /* when next process has a 64bit base use it */
604 if (next->fs) 616 if (next->fs)
@@ -608,7 +620,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
608 if (unlikely(gsindex | next->gsindex | prev->gs)) { 620 if (unlikely(gsindex | next->gsindex | prev->gs)) {
609 load_gs_index(next->gsindex); 621 load_gs_index(next->gsindex);
610 if (gsindex) 622 if (gsindex)
611 prev->gs = 0; 623 prev->gs = 0;
612 } 624 }
613 if (next->gs) 625 if (next->gs)
614 wrmsrl(MSR_KERNEL_GS_BASE, next->gs); 626 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
@@ -617,12 +629,12 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
617 /* Must be after DS reload */ 629 /* Must be after DS reload */
618 unlazy_fpu(prev_p); 630 unlazy_fpu(prev_p);
619 631
620 /* 632 /*
621 * Switch the PDA and FPU contexts. 633 * Switch the PDA and FPU contexts.
622 */ 634 */
623 prev->usersp = read_pda(oldrsp); 635 prev->usersp = read_pda(oldrsp);
624 write_pda(oldrsp, next->usersp); 636 write_pda(oldrsp, next->usersp);
625 write_pda(pcurrent, next_p); 637 write_pda(pcurrent, next_p);
626 638
627 write_pda(kernelstack, 639 write_pda(kernelstack,
628 (unsigned long)task_stack_page(next_p) + 640 (unsigned long)task_stack_page(next_p) +
@@ -663,7 +675,7 @@ long sys_execve(char __user *name, char __user * __user *argv,
663 char __user * __user *envp, struct pt_regs *regs) 675 char __user * __user *envp, struct pt_regs *regs)
664{ 676{
665 long error; 677 long error;
666 char * filename; 678 char *filename;
667 679
668 filename = getname(name); 680 filename = getname(name);
669 error = PTR_ERR(filename); 681 error = PTR_ERR(filename);
@@ -721,55 +733,55 @@ asmlinkage long sys_vfork(struct pt_regs *regs)
721unsigned long get_wchan(struct task_struct *p) 733unsigned long get_wchan(struct task_struct *p)
722{ 734{
723 unsigned long stack; 735 unsigned long stack;
724 u64 fp,ip; 736 u64 fp, ip;
725 int count = 0; 737 int count = 0;
726 738
727 if (!p || p == current || p->state==TASK_RUNNING) 739 if (!p || p == current || p->state == TASK_RUNNING)
728 return 0; 740 return 0;
729 stack = (unsigned long)task_stack_page(p); 741 stack = (unsigned long)task_stack_page(p);
730 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE) 742 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
731 return 0; 743 return 0;
732 fp = *(u64 *)(p->thread.sp); 744 fp = *(u64 *)(p->thread.sp);
733 do { 745 do {
734 if (fp < (unsigned long)stack || 746 if (fp < (unsigned long)stack ||
735 fp > (unsigned long)stack+THREAD_SIZE) 747 fp >= (unsigned long)stack+THREAD_SIZE)
736 return 0; 748 return 0;
737 ip = *(u64 *)(fp+8); 749 ip = *(u64 *)(fp+8);
738 if (!in_sched_functions(ip)) 750 if (!in_sched_functions(ip))
739 return ip; 751 return ip;
740 fp = *(u64 *)fp; 752 fp = *(u64 *)fp;
741 } while (count++ < 16); 753 } while (count++ < 16);
742 return 0; 754 return 0;
743} 755}
744 756
745long do_arch_prctl(struct task_struct *task, int code, unsigned long addr) 757long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
746{ 758{
747 int ret = 0; 759 int ret = 0;
748 int doit = task == current; 760 int doit = task == current;
749 int cpu; 761 int cpu;
750 762
751 switch (code) { 763 switch (code) {
752 case ARCH_SET_GS: 764 case ARCH_SET_GS:
753 if (addr >= TASK_SIZE_OF(task)) 765 if (addr >= TASK_SIZE_OF(task))
754 return -EPERM; 766 return -EPERM;
755 cpu = get_cpu(); 767 cpu = get_cpu();
756 /* handle small bases via the GDT because that's faster to 768 /* handle small bases via the GDT because that's faster to
757 switch. */ 769 switch. */
758 if (addr <= 0xffffffff) { 770 if (addr <= 0xffffffff) {
759 set_32bit_tls(task, GS_TLS, addr); 771 set_32bit_tls(task, GS_TLS, addr);
760 if (doit) { 772 if (doit) {
761 load_TLS(&task->thread, cpu); 773 load_TLS(&task->thread, cpu);
762 load_gs_index(GS_TLS_SEL); 774 load_gs_index(GS_TLS_SEL);
763 } 775 }
764 task->thread.gsindex = GS_TLS_SEL; 776 task->thread.gsindex = GS_TLS_SEL;
765 task->thread.gs = 0; 777 task->thread.gs = 0;
766 } else { 778 } else {
767 task->thread.gsindex = 0; 779 task->thread.gsindex = 0;
768 task->thread.gs = addr; 780 task->thread.gs = addr;
769 if (doit) { 781 if (doit) {
770 load_gs_index(0); 782 load_gs_index(0);
771 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr); 783 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
772 } 784 }
773 } 785 }
774 put_cpu(); 786 put_cpu();
775 break; 787 break;
@@ -823,8 +835,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
823 rdmsrl(MSR_KERNEL_GS_BASE, base); 835 rdmsrl(MSR_KERNEL_GS_BASE, base);
824 else 836 else
825 base = task->thread.gs; 837 base = task->thread.gs;
826 } 838 } else
827 else
828 base = task->thread.gs; 839 base = task->thread.gs;
829 ret = put_user(base, (unsigned long __user *)addr); 840 ret = put_user(base, (unsigned long __user *)addr);
830 break; 841 break;
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index e37dccce85db..0a6d8c12e10d 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -14,6 +14,7 @@
14#include <linux/errno.h> 14#include <linux/errno.h>
15#include <linux/ptrace.h> 15#include <linux/ptrace.h>
16#include <linux/regset.h> 16#include <linux/regset.h>
17#include <linux/tracehook.h>
17#include <linux/user.h> 18#include <linux/user.h>
18#include <linux/elf.h> 19#include <linux/elf.h>
19#include <linux/security.h> 20#include <linux/security.h>
@@ -39,7 +40,9 @@ enum x86_regset {
39 REGSET_GENERAL, 40 REGSET_GENERAL,
40 REGSET_FP, 41 REGSET_FP,
41 REGSET_XFP, 42 REGSET_XFP,
43 REGSET_IOPERM64 = REGSET_XFP,
42 REGSET_TLS, 44 REGSET_TLS,
45 REGSET_IOPERM32,
43}; 46};
44 47
45/* 48/*
@@ -69,7 +72,7 @@ static inline bool invalid_selector(u16 value)
69 72
70#define FLAG_MASK FLAG_MASK_32 73#define FLAG_MASK FLAG_MASK_32
71 74
72static long *pt_regs_access(struct pt_regs *regs, unsigned long regno) 75static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
73{ 76{
74 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0); 77 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
75 regno >>= 2; 78 regno >>= 2;
@@ -554,45 +557,138 @@ static int ptrace_set_debugreg(struct task_struct *child,
554 return 0; 557 return 0;
555} 558}
556 559
557#ifdef X86_BTS 560/*
561 * These access the current or another (stopped) task's io permission
562 * bitmap for debugging or core dump.
563 */
564static int ioperm_active(struct task_struct *target,
565 const struct user_regset *regset)
566{
567 return target->thread.io_bitmap_max / regset->size;
568}
558 569
559static int ptrace_bts_get_size(struct task_struct *child) 570static int ioperm_get(struct task_struct *target,
571 const struct user_regset *regset,
572 unsigned int pos, unsigned int count,
573 void *kbuf, void __user *ubuf)
560{ 574{
561 if (!child->thread.ds_area_msr) 575 if (!target->thread.io_bitmap_ptr)
562 return -ENXIO; 576 return -ENXIO;
563 577
564 return ds_get_bts_index((void *)child->thread.ds_area_msr); 578 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
579 target->thread.io_bitmap_ptr,
580 0, IO_BITMAP_BYTES);
581}
582
583#ifdef CONFIG_X86_PTRACE_BTS
584/*
585 * The configuration for a particular BTS hardware implementation.
586 */
587struct bts_configuration {
588 /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */
589 unsigned char sizeof_bts;
590 /* the size of a field in the BTS record in bytes */
591 unsigned char sizeof_field;
592 /* a bitmask to enable/disable BTS in DEBUGCTL MSR */
593 unsigned long debugctl_mask;
594};
595static struct bts_configuration bts_cfg;
596
597#define BTS_MAX_RECORD_SIZE (8 * 3)
598
599
600/*
601 * Branch Trace Store (BTS) uses the following format. Different
602 * architectures vary in the size of those fields.
603 * - source linear address
604 * - destination linear address
605 * - flags
606 *
607 * Later architectures use 64bit pointers throughout, whereas earlier
608 * architectures use 32bit pointers in 32bit mode.
609 *
610 * We compute the base address for the first 8 fields based on:
611 * - the field size stored in the DS configuration
612 * - the relative field position
613 *
614 * In order to store additional information in the BTS buffer, we use
615 * a special source address to indicate that the record requires
616 * special interpretation.
617 *
618 * Netburst indicated via a bit in the flags field whether the branch
619 * was predicted; this is ignored.
620 */
621
622enum bts_field {
623 bts_from = 0,
624 bts_to,
625 bts_flags,
626
627 bts_escape = (unsigned long)-1,
628 bts_qual = bts_to,
629 bts_jiffies = bts_flags
630};
631
632static inline unsigned long bts_get(const char *base, enum bts_field field)
633{
634 base += (bts_cfg.sizeof_field * field);
635 return *(unsigned long *)base;
636}
637
638static inline void bts_set(char *base, enum bts_field field, unsigned long val)
639{
640 base += (bts_cfg.sizeof_field * field);;
641 (*(unsigned long *)base) = val;
642}
643
644/*
645 * Translate a BTS record from the raw format into the bts_struct format
646 *
647 * out (out): bts_struct interpretation
648 * raw: raw BTS record
649 */
650static void ptrace_bts_translate_record(struct bts_struct *out, const void *raw)
651{
652 memset(out, 0, sizeof(*out));
653 if (bts_get(raw, bts_from) == bts_escape) {
654 out->qualifier = bts_get(raw, bts_qual);
655 out->variant.jiffies = bts_get(raw, bts_jiffies);
656 } else {
657 out->qualifier = BTS_BRANCH;
658 out->variant.lbr.from_ip = bts_get(raw, bts_from);
659 out->variant.lbr.to_ip = bts_get(raw, bts_to);
660 }
565} 661}
566 662
567static int ptrace_bts_read_record(struct task_struct *child, 663static int ptrace_bts_read_record(struct task_struct *child, size_t index,
568 long index,
569 struct bts_struct __user *out) 664 struct bts_struct __user *out)
570{ 665{
571 struct bts_struct ret; 666 struct bts_struct ret;
572 int retval; 667 const void *bts_record;
573 int bts_end; 668 size_t bts_index, bts_end;
574 int bts_index; 669 int error;
575
576 if (!child->thread.ds_area_msr)
577 return -ENXIO;
578 670
579 if (index < 0) 671 error = ds_get_bts_end(child, &bts_end);
580 return -EINVAL; 672 if (error < 0)
673 return error;
581 674
582 bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
583 if (bts_end <= index) 675 if (bts_end <= index)
584 return -EINVAL; 676 return -EINVAL;
585 677
678 error = ds_get_bts_index(child, &bts_index);
679 if (error < 0)
680 return error;
681
586 /* translate the ptrace bts index into the ds bts index */ 682 /* translate the ptrace bts index into the ds bts index */
587 bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr); 683 bts_index += bts_end - (index + 1);
588 bts_index -= (index + 1); 684 if (bts_end <= bts_index)
589 if (bts_index < 0) 685 bts_index -= bts_end;
590 bts_index += bts_end; 686
687 error = ds_access_bts(child, bts_index, &bts_record);
688 if (error < 0)
689 return error;
591 690
592 retval = ds_read_bts((void *)child->thread.ds_area_msr, 691 ptrace_bts_translate_record(&ret, bts_record);
593 bts_index, &ret);
594 if (retval < 0)
595 return retval;
596 692
597 if (copy_to_user(out, &ret, sizeof(ret))) 693 if (copy_to_user(out, &ret, sizeof(ret)))
598 return -EFAULT; 694 return -EFAULT;
@@ -600,101 +696,106 @@ static int ptrace_bts_read_record(struct task_struct *child,
600 return sizeof(ret); 696 return sizeof(ret);
601} 697}
602 698
603static int ptrace_bts_clear(struct task_struct *child)
604{
605 if (!child->thread.ds_area_msr)
606 return -ENXIO;
607
608 return ds_clear((void *)child->thread.ds_area_msr);
609}
610
611static int ptrace_bts_drain(struct task_struct *child, 699static int ptrace_bts_drain(struct task_struct *child,
612 long size, 700 long size,
613 struct bts_struct __user *out) 701 struct bts_struct __user *out)
614{ 702{
615 int end, i; 703 struct bts_struct ret;
616 void *ds = (void *)child->thread.ds_area_msr; 704 const unsigned char *raw;
617 705 size_t end, i;
618 if (!ds) 706 int error;
619 return -ENXIO;
620 707
621 end = ds_get_bts_index(ds); 708 error = ds_get_bts_index(child, &end);
622 if (end <= 0) 709 if (error < 0)
623 return end; 710 return error;
624 711
625 if (size < (end * sizeof(struct bts_struct))) 712 if (size < (end * sizeof(struct bts_struct)))
626 return -EIO; 713 return -EIO;
627 714
628 for (i = 0; i < end; i++, out++) { 715 error = ds_access_bts(child, 0, (const void **)&raw);
629 struct bts_struct ret; 716 if (error < 0)
630 int retval; 717 return error;
631 718
632 retval = ds_read_bts(ds, i, &ret); 719 for (i = 0; i < end; i++, out++, raw += bts_cfg.sizeof_bts) {
633 if (retval < 0) 720 ptrace_bts_translate_record(&ret, raw);
634 return retval;
635 721
636 if (copy_to_user(out, &ret, sizeof(ret))) 722 if (copy_to_user(out, &ret, sizeof(ret)))
637 return -EFAULT; 723 return -EFAULT;
638 } 724 }
639 725
640 ds_clear(ds); 726 error = ds_clear_bts(child);
727 if (error < 0)
728 return error;
641 729
642 return end; 730 return end;
643} 731}
644 732
733static void ptrace_bts_ovfl(struct task_struct *child)
734{
735 send_sig(child->thread.bts_ovfl_signal, child, 0);
736}
737
645static int ptrace_bts_config(struct task_struct *child, 738static int ptrace_bts_config(struct task_struct *child,
646 long cfg_size, 739 long cfg_size,
647 const struct ptrace_bts_config __user *ucfg) 740 const struct ptrace_bts_config __user *ucfg)
648{ 741{
649 struct ptrace_bts_config cfg; 742 struct ptrace_bts_config cfg;
650 int bts_size, ret = 0; 743 int error = 0;
651 void *ds;
652 744
745 error = -EOPNOTSUPP;
746 if (!bts_cfg.sizeof_bts)
747 goto errout;
748
749 error = -EIO;
653 if (cfg_size < sizeof(cfg)) 750 if (cfg_size < sizeof(cfg))
654 return -EIO; 751 goto errout;
655 752
753 error = -EFAULT;
656 if (copy_from_user(&cfg, ucfg, sizeof(cfg))) 754 if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
657 return -EFAULT; 755 goto errout;
658 756
659 if ((int)cfg.size < 0) 757 error = -EINVAL;
660 return -EINVAL; 758 if ((cfg.flags & PTRACE_BTS_O_SIGNAL) &&
759 !(cfg.flags & PTRACE_BTS_O_ALLOC))
760 goto errout;
661 761
662 bts_size = 0; 762 if (cfg.flags & PTRACE_BTS_O_ALLOC) {
663 ds = (void *)child->thread.ds_area_msr; 763 ds_ovfl_callback_t ovfl = NULL;
664 if (ds) { 764 unsigned int sig = 0;
665 bts_size = ds_get_bts_size(ds); 765
666 if (bts_size < 0) 766 /* we ignore the error in case we were not tracing child */
667 return bts_size; 767 (void)ds_release_bts(child);
668 } 768
669 cfg.size = PAGE_ALIGN(cfg.size); 769 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
770 if (!cfg.signal)
771 goto errout;
772
773 sig = cfg.signal;
774 ovfl = ptrace_bts_ovfl;
775 }
670 776
671 if (bts_size != cfg.size) { 777 error = ds_request_bts(child, /* base = */ NULL, cfg.size, ovfl);
672 ret = ptrace_bts_realloc(child, cfg.size, 778 if (error < 0)
673 cfg.flags & PTRACE_BTS_O_CUT_SIZE);
674 if (ret < 0)
675 goto errout; 779 goto errout;
676 780
677 ds = (void *)child->thread.ds_area_msr; 781 child->thread.bts_ovfl_signal = sig;
678 } 782 }
679 783
680 if (cfg.flags & PTRACE_BTS_O_SIGNAL) 784 error = -EINVAL;
681 ret = ds_set_overflow(ds, DS_O_SIGNAL); 785 if (!child->thread.ds_ctx && cfg.flags)
682 else
683 ret = ds_set_overflow(ds, DS_O_WRAP);
684 if (ret < 0)
685 goto errout; 786 goto errout;
686 787
687 if (cfg.flags & PTRACE_BTS_O_TRACE) 788 if (cfg.flags & PTRACE_BTS_O_TRACE)
688 child->thread.debugctlmsr |= ds_debugctl_mask(); 789 child->thread.debugctlmsr |= bts_cfg.debugctl_mask;
689 else 790 else
690 child->thread.debugctlmsr &= ~ds_debugctl_mask(); 791 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
691 792
692 if (cfg.flags & PTRACE_BTS_O_SCHED) 793 if (cfg.flags & PTRACE_BTS_O_SCHED)
693 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS); 794 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
694 else 795 else
695 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); 796 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
696 797
697 ret = sizeof(cfg); 798 error = sizeof(cfg);
698 799
699out: 800out:
700 if (child->thread.debugctlmsr) 801 if (child->thread.debugctlmsr)
@@ -702,10 +803,10 @@ out:
702 else 803 else
703 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR); 804 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
704 805
705 return ret; 806 return error;
706 807
707errout: 808errout:
708 child->thread.debugctlmsr &= ~ds_debugctl_mask(); 809 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
709 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); 810 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
710 goto out; 811 goto out;
711} 812}
@@ -714,29 +815,40 @@ static int ptrace_bts_status(struct task_struct *child,
714 long cfg_size, 815 long cfg_size,
715 struct ptrace_bts_config __user *ucfg) 816 struct ptrace_bts_config __user *ucfg)
716{ 817{
717 void *ds = (void *)child->thread.ds_area_msr;
718 struct ptrace_bts_config cfg; 818 struct ptrace_bts_config cfg;
819 size_t end;
820 const void *base, *max;
821 int error;
719 822
720 if (cfg_size < sizeof(cfg)) 823 if (cfg_size < sizeof(cfg))
721 return -EIO; 824 return -EIO;
722 825
723 memset(&cfg, 0, sizeof(cfg)); 826 error = ds_get_bts_end(child, &end);
827 if (error < 0)
828 return error;
724 829
725 if (ds) { 830 error = ds_access_bts(child, /* index = */ 0, &base);
726 cfg.size = ds_get_bts_size(ds); 831 if (error < 0)
832 return error;
727 833
728 if (ds_get_overflow(ds) == DS_O_SIGNAL) 834 error = ds_access_bts(child, /* index = */ end, &max);
729 cfg.flags |= PTRACE_BTS_O_SIGNAL; 835 if (error < 0)
836 return error;
730 837
731 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) && 838 memset(&cfg, 0, sizeof(cfg));
732 child->thread.debugctlmsr & ds_debugctl_mask()) 839 cfg.size = (max - base);
733 cfg.flags |= PTRACE_BTS_O_TRACE; 840 cfg.signal = child->thread.bts_ovfl_signal;
841 cfg.bts_size = sizeof(struct bts_struct);
734 842
735 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS)) 843 if (cfg.signal)
736 cfg.flags |= PTRACE_BTS_O_SCHED; 844 cfg.flags |= PTRACE_BTS_O_SIGNAL;
737 }
738 845
739 cfg.bts_size = sizeof(struct bts_struct); 846 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
847 child->thread.debugctlmsr & bts_cfg.debugctl_mask)
848 cfg.flags |= PTRACE_BTS_O_TRACE;
849
850 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
851 cfg.flags |= PTRACE_BTS_O_SCHED;
740 852
741 if (copy_to_user(ucfg, &cfg, sizeof(cfg))) 853 if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
742 return -EFAULT; 854 return -EFAULT;
@@ -744,89 +856,38 @@ static int ptrace_bts_status(struct task_struct *child,
744 return sizeof(cfg); 856 return sizeof(cfg);
745} 857}
746 858
747
748static int ptrace_bts_write_record(struct task_struct *child, 859static int ptrace_bts_write_record(struct task_struct *child,
749 const struct bts_struct *in) 860 const struct bts_struct *in)
750{ 861{
751 int retval; 862 unsigned char bts_record[BTS_MAX_RECORD_SIZE];
752 863
753 if (!child->thread.ds_area_msr) 864 BUG_ON(BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts);
754 return -ENXIO;
755 865
756 retval = ds_write_bts((void *)child->thread.ds_area_msr, in); 866 memset(bts_record, 0, bts_cfg.sizeof_bts);
757 if (retval) 867 switch (in->qualifier) {
758 return retval; 868 case BTS_INVALID:
869 break;
759 870
760 return sizeof(*in); 871 case BTS_BRANCH:
761} 872 bts_set(bts_record, bts_from, in->variant.lbr.from_ip);
873 bts_set(bts_record, bts_to, in->variant.lbr.to_ip);
874 break;
762 875
763static int ptrace_bts_realloc(struct task_struct *child, 876 case BTS_TASK_ARRIVES:
764 int size, int reduce_size) 877 case BTS_TASK_DEPARTS:
765{ 878 bts_set(bts_record, bts_from, bts_escape);
766 unsigned long rlim, vm; 879 bts_set(bts_record, bts_qual, in->qualifier);
767 int ret, old_size; 880 bts_set(bts_record, bts_jiffies, in->variant.jiffies);
881 break;
768 882
769 if (size < 0) 883 default:
770 return -EINVAL; 884 return -EINVAL;
771
772 old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
773 if (old_size < 0)
774 return old_size;
775
776 ret = ds_free((void **)&child->thread.ds_area_msr);
777 if (ret < 0)
778 goto out;
779
780 size >>= PAGE_SHIFT;
781 old_size >>= PAGE_SHIFT;
782
783 current->mm->total_vm -= old_size;
784 current->mm->locked_vm -= old_size;
785
786 if (size == 0)
787 goto out;
788
789 rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
790 vm = current->mm->total_vm + size;
791 if (rlim < vm) {
792 ret = -ENOMEM;
793
794 if (!reduce_size)
795 goto out;
796
797 size = rlim - current->mm->total_vm;
798 if (size <= 0)
799 goto out;
800 }
801
802 rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
803 vm = current->mm->locked_vm + size;
804 if (rlim < vm) {
805 ret = -ENOMEM;
806
807 if (!reduce_size)
808 goto out;
809
810 size = rlim - current->mm->locked_vm;
811 if (size <= 0)
812 goto out;
813 } 885 }
814 886
815 ret = ds_allocate((void **)&child->thread.ds_area_msr, 887 /* The writing task will be the switched-to task on a context
816 size << PAGE_SHIFT); 888 * switch. It needs to write into the switched-from task's BTS
817 if (ret < 0) 889 * buffer. */
818 goto out; 890 return ds_unchecked_write_bts(child, bts_record, bts_cfg.sizeof_bts);
819
820 current->mm->total_vm += size;
821 current->mm->locked_vm += size;
822
823out:
824 if (child->thread.ds_area_msr)
825 set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
826 else
827 clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
828
829 return ret;
830} 891}
831 892
832void ptrace_bts_take_timestamp(struct task_struct *tsk, 893void ptrace_bts_take_timestamp(struct task_struct *tsk,
@@ -839,7 +900,66 @@ void ptrace_bts_take_timestamp(struct task_struct *tsk,
839 900
840 ptrace_bts_write_record(tsk, &rec); 901 ptrace_bts_write_record(tsk, &rec);
841} 902}
842#endif /* X86_BTS */ 903
904static const struct bts_configuration bts_cfg_netburst = {
905 .sizeof_bts = sizeof(long) * 3,
906 .sizeof_field = sizeof(long),
907 .debugctl_mask = (1<<2)|(1<<3)|(1<<5)
908};
909
910static const struct bts_configuration bts_cfg_pentium_m = {
911 .sizeof_bts = sizeof(long) * 3,
912 .sizeof_field = sizeof(long),
913 .debugctl_mask = (1<<6)|(1<<7)
914};
915
916static const struct bts_configuration bts_cfg_core2 = {
917 .sizeof_bts = 8 * 3,
918 .sizeof_field = 8,
919 .debugctl_mask = (1<<6)|(1<<7)|(1<<9)
920};
921
922static inline void bts_configure(const struct bts_configuration *cfg)
923{
924 bts_cfg = *cfg;
925}
926
927void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c)
928{
929 switch (c->x86) {
930 case 0x6:
931 switch (c->x86_model) {
932 case 0xD:
933 case 0xE: /* Pentium M */
934 bts_configure(&bts_cfg_pentium_m);
935 break;
936 case 0xF: /* Core2 */
937 case 0x1C: /* Atom */
938 bts_configure(&bts_cfg_core2);
939 break;
940 default:
941 /* sorry, don't know about them */
942 break;
943 }
944 break;
945 case 0xF:
946 switch (c->x86_model) {
947 case 0x0:
948 case 0x1:
949 case 0x2: /* Netburst */
950 bts_configure(&bts_cfg_netburst);
951 break;
952 default:
953 /* sorry, don't know about them */
954 break;
955 }
956 break;
957 default:
958 /* sorry, don't know about them */
959 break;
960 }
961}
962#endif /* CONFIG_X86_PTRACE_BTS */
843 963
844/* 964/*
845 * Called by kernel/ptrace.c when detaching.. 965 * Called by kernel/ptrace.c when detaching..
@@ -852,15 +972,15 @@ void ptrace_disable(struct task_struct *child)
852#ifdef TIF_SYSCALL_EMU 972#ifdef TIF_SYSCALL_EMU
853 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU); 973 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
854#endif 974#endif
855 if (child->thread.ds_area_msr) { 975#ifdef CONFIG_X86_PTRACE_BTS
856#ifdef X86_BTS 976 (void)ds_release_bts(child);
857 ptrace_bts_realloc(child, 0, 0); 977
858#endif 978 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
859 child->thread.debugctlmsr &= ~ds_debugctl_mask(); 979 if (!child->thread.debugctlmsr)
860 if (!child->thread.debugctlmsr) 980 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
861 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR); 981
862 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); 982 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
863 } 983#endif /* CONFIG_X86_PTRACE_BTS */
864} 984}
865 985
866#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION 986#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
@@ -980,7 +1100,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
980 /* 1100 /*
981 * These bits need more cooking - not enabled yet: 1101 * These bits need more cooking - not enabled yet:
982 */ 1102 */
983#ifdef X86_BTS 1103#ifdef CONFIG_X86_PTRACE_BTS
984 case PTRACE_BTS_CONFIG: 1104 case PTRACE_BTS_CONFIG:
985 ret = ptrace_bts_config 1105 ret = ptrace_bts_config
986 (child, data, (struct ptrace_bts_config __user *)addr); 1106 (child, data, (struct ptrace_bts_config __user *)addr);
@@ -992,7 +1112,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
992 break; 1112 break;
993 1113
994 case PTRACE_BTS_SIZE: 1114 case PTRACE_BTS_SIZE:
995 ret = ptrace_bts_get_size(child); 1115 ret = ds_get_bts_index(child, /* pos = */ NULL);
996 break; 1116 break;
997 1117
998 case PTRACE_BTS_GET: 1118 case PTRACE_BTS_GET:
@@ -1001,14 +1121,14 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1001 break; 1121 break;
1002 1122
1003 case PTRACE_BTS_CLEAR: 1123 case PTRACE_BTS_CLEAR:
1004 ret = ptrace_bts_clear(child); 1124 ret = ds_clear_bts(child);
1005 break; 1125 break;
1006 1126
1007 case PTRACE_BTS_DRAIN: 1127 case PTRACE_BTS_DRAIN:
1008 ret = ptrace_bts_drain 1128 ret = ptrace_bts_drain
1009 (child, data, (struct bts_struct __user *) addr); 1129 (child, data, (struct bts_struct __user *) addr);
1010 break; 1130 break;
1011#endif 1131#endif /* CONFIG_X86_PTRACE_BTS */
1012 1132
1013 default: 1133 default:
1014 ret = ptrace_request(child, request, addr, data); 1134 ret = ptrace_request(child, request, addr, data);
@@ -1290,6 +1410,12 @@ static const struct user_regset x86_64_regsets[] = {
1290 .size = sizeof(long), .align = sizeof(long), 1410 .size = sizeof(long), .align = sizeof(long),
1291 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set 1411 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1292 }, 1412 },
1413 [REGSET_IOPERM64] = {
1414 .core_note_type = NT_386_IOPERM,
1415 .n = IO_BITMAP_LONGS,
1416 .size = sizeof(long), .align = sizeof(long),
1417 .active = ioperm_active, .get = ioperm_get
1418 },
1293}; 1419};
1294 1420
1295static const struct user_regset_view user_x86_64_view = { 1421static const struct user_regset_view user_x86_64_view = {
@@ -1336,6 +1462,12 @@ static const struct user_regset x86_32_regsets[] = {
1336 .active = regset_tls_active, 1462 .active = regset_tls_active,
1337 .get = regset_tls_get, .set = regset_tls_set 1463 .get = regset_tls_get, .set = regset_tls_set
1338 }, 1464 },
1465 [REGSET_IOPERM32] = {
1466 .core_note_type = NT_386_IOPERM,
1467 .n = IO_BITMAP_BYTES / sizeof(u32),
1468 .size = sizeof(u32), .align = sizeof(u32),
1469 .active = ioperm_active, .get = ioperm_get
1470 },
1339}; 1471};
1340 1472
1341static const struct user_regset_view user_x86_32_view = { 1473static const struct user_regset_view user_x86_32_view = {
@@ -1357,7 +1489,8 @@ const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1357#endif 1489#endif
1358} 1490}
1359 1491
1360void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code) 1492void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1493 int error_code, int si_code)
1361{ 1494{
1362 struct siginfo info; 1495 struct siginfo info;
1363 1496
@@ -1366,7 +1499,7 @@ void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1366 1499
1367 memset(&info, 0, sizeof(info)); 1500 memset(&info, 0, sizeof(info));
1368 info.si_signo = SIGTRAP; 1501 info.si_signo = SIGTRAP;
1369 info.si_code = TRAP_BRKPT; 1502 info.si_code = si_code;
1370 1503
1371 /* User-mode ip? */ 1504 /* User-mode ip? */
1372 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL; 1505 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
@@ -1375,30 +1508,6 @@ void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1375 force_sig_info(SIGTRAP, &info, tsk); 1508 force_sig_info(SIGTRAP, &info, tsk);
1376} 1509}
1377 1510
1378static void syscall_trace(struct pt_regs *regs)
1379{
1380 if (!(current->ptrace & PT_PTRACED))
1381 return;
1382
1383#if 0
1384 printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
1385 current->comm,
1386 regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0),
1387 current_thread_info()->flags, current->ptrace);
1388#endif
1389
1390 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
1391 ? 0x80 : 0));
1392 /*
1393 * this isn't the same as continuing with a signal, but it will do
1394 * for normal use. strace only continues with a signal if the
1395 * stopping signal is not SIGTRAP. -brl
1396 */
1397 if (current->exit_code) {
1398 send_sig(current->exit_code, current, 1);
1399 current->exit_code = 0;
1400 }
1401}
1402 1511
1403#ifdef CONFIG_X86_32 1512#ifdef CONFIG_X86_32
1404# define IS_IA32 1 1513# define IS_IA32 1
@@ -1432,8 +1541,9 @@ asmregparm long syscall_trace_enter(struct pt_regs *regs)
1432 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU))) 1541 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1433 ret = -1L; 1542 ret = -1L;
1434 1543
1435 if (ret || test_thread_flag(TIF_SYSCALL_TRACE)) 1544 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1436 syscall_trace(regs); 1545 tracehook_report_syscall_entry(regs))
1546 ret = -1L;
1437 1547
1438 if (unlikely(current->audit_context)) { 1548 if (unlikely(current->audit_context)) {
1439 if (IS_IA32) 1549 if (IS_IA32)
@@ -1459,7 +1569,7 @@ asmregparm void syscall_trace_leave(struct pt_regs *regs)
1459 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax); 1569 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1460 1570
1461 if (test_thread_flag(TIF_SYSCALL_TRACE)) 1571 if (test_thread_flag(TIF_SYSCALL_TRACE))
1462 syscall_trace(regs); 1572 tracehook_report_syscall_exit(regs, 0);
1463 1573
1464 /* 1574 /*
1465 * If TIF_SYSCALL_EMU is set, we only get here because of 1575 * If TIF_SYSCALL_EMU is set, we only get here because of
@@ -1475,6 +1585,6 @@ asmregparm void syscall_trace_leave(struct pt_regs *regs)
1475 * system call instruction. 1585 * system call instruction.
1476 */ 1586 */
1477 if (test_thread_flag(TIF_SINGLESTEP) && 1587 if (test_thread_flag(TIF_SINGLESTEP) &&
1478 (current->ptrace & PT_PTRACED)) 1588 tracehook_consider_fatal_signal(current, SIGTRAP, SIG_DFL))
1479 send_sigtrap(current, regs, 0); 1589 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1480} 1590}
diff --git a/arch/x86/kernel/pvclock.c b/arch/x86/kernel/pvclock.c
index 05fbe9a0325a..4f9c55f3a7c0 100644
--- a/arch/x86/kernel/pvclock.c
+++ b/arch/x86/kernel/pvclock.c
@@ -97,6 +97,18 @@ static unsigned pvclock_get_time_values(struct pvclock_shadow_time *dst,
97 return dst->version; 97 return dst->version;
98} 98}
99 99
100unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
101{
102 u64 pv_tsc_khz = 1000000ULL << 32;
103
104 do_div(pv_tsc_khz, src->tsc_to_system_mul);
105 if (src->tsc_shift < 0)
106 pv_tsc_khz <<= -src->tsc_shift;
107 else
108 pv_tsc_khz >>= src->tsc_shift;
109 return pv_tsc_khz;
110}
111
100cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src) 112cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
101{ 113{
102 struct pvclock_shadow_time shadow; 114 struct pvclock_shadow_time shadow;
diff --git a/arch/x86/kernel/quirks.c b/arch/x86/kernel/quirks.c
index d13858818100..f6a11b9b1f98 100644
--- a/arch/x86/kernel/quirks.c
+++ b/arch/x86/kernel/quirks.c
@@ -354,9 +354,27 @@ static void ati_force_hpet_resume(void)
354 printk(KERN_DEBUG "Force enabled HPET at resume\n"); 354 printk(KERN_DEBUG "Force enabled HPET at resume\n");
355} 355}
356 356
357static u32 ati_ixp4x0_rev(struct pci_dev *dev)
358{
359 u32 d;
360 u8 b;
361
362 pci_read_config_byte(dev, 0xac, &b);
363 b &= ~(1<<5);
364 pci_write_config_byte(dev, 0xac, b);
365 pci_read_config_dword(dev, 0x70, &d);
366 d |= 1<<8;
367 pci_write_config_dword(dev, 0x70, d);
368 pci_read_config_dword(dev, 0x8, &d);
369 d &= 0xff;
370 dev_printk(KERN_DEBUG, &dev->dev, "SB4X0 revision 0x%x\n", d);
371 return d;
372}
373
357static void ati_force_enable_hpet(struct pci_dev *dev) 374static void ati_force_enable_hpet(struct pci_dev *dev)
358{ 375{
359 u32 uninitialized_var(val); 376 u32 d, val;
377 u8 b;
360 378
361 if (hpet_address || force_hpet_address) 379 if (hpet_address || force_hpet_address)
362 return; 380 return;
@@ -366,14 +384,33 @@ static void ati_force_enable_hpet(struct pci_dev *dev)
366 return; 384 return;
367 } 385 }
368 386
387 d = ati_ixp4x0_rev(dev);
388 if (d < 0x82)
389 return;
390
391 /* base address */
369 pci_write_config_dword(dev, 0x14, 0xfed00000); 392 pci_write_config_dword(dev, 0x14, 0xfed00000);
370 pci_read_config_dword(dev, 0x14, &val); 393 pci_read_config_dword(dev, 0x14, &val);
394
395 /* enable interrupt */
396 outb(0x72, 0xcd6); b = inb(0xcd7);
397 b |= 0x1;
398 outb(0x72, 0xcd6); outb(b, 0xcd7);
399 outb(0x72, 0xcd6); b = inb(0xcd7);
400 if (!(b & 0x1))
401 return;
402 pci_read_config_dword(dev, 0x64, &d);
403 d |= (1<<10);
404 pci_write_config_dword(dev, 0x64, d);
405 pci_read_config_dword(dev, 0x64, &d);
406 if (!(d & (1<<10)))
407 return;
408
371 force_hpet_address = val; 409 force_hpet_address = val;
372 force_hpet_resume_type = ATI_FORCE_HPET_RESUME; 410 force_hpet_resume_type = ATI_FORCE_HPET_RESUME;
373 dev_printk(KERN_DEBUG, &dev->dev, "Force enabled HPET at 0x%lx\n", 411 dev_printk(KERN_DEBUG, &dev->dev, "Force enabled HPET at 0x%lx\n",
374 force_hpet_address); 412 force_hpet_address);
375 cached_dev = dev; 413 cached_dev = dev;
376 return;
377} 414}
378DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_ATI, PCI_DEVICE_ID_ATI_IXP400_SMBUS, 415DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_ATI, PCI_DEVICE_ID_ATI_IXP400_SMBUS,
379 ati_force_enable_hpet); 416 ati_force_enable_hpet);
diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c
index 724adfc63cb9..f4c93f1cfc19 100644
--- a/arch/x86/kernel/reboot.c
+++ b/arch/x86/kernel/reboot.c
@@ -29,7 +29,11 @@ EXPORT_SYMBOL(pm_power_off);
29 29
30static const struct desc_ptr no_idt = {}; 30static const struct desc_ptr no_idt = {};
31static int reboot_mode; 31static int reboot_mode;
32enum reboot_type reboot_type = BOOT_KBD; 32/*
33 * Keyboard reset and triple fault may result in INIT, not RESET, which
34 * doesn't work when we're in vmx root mode. Try ACPI first.
35 */
36enum reboot_type reboot_type = BOOT_ACPI;
33int reboot_force; 37int reboot_force;
34 38
35#if defined(CONFIG_X86_32) && defined(CONFIG_SMP) 39#if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
diff --git a/arch/x86/kernel/rtc.c b/arch/x86/kernel/rtc.c
index 05191bbc68b8..0a23b5795b25 100644
--- a/arch/x86/kernel/rtc.c
+++ b/arch/x86/kernel/rtc.c
@@ -223,11 +223,25 @@ static struct platform_device rtc_device = {
223static __init int add_rtc_cmos(void) 223static __init int add_rtc_cmos(void)
224{ 224{
225#ifdef CONFIG_PNP 225#ifdef CONFIG_PNP
226 if (!pnp_platform_devices) 226 static const char *ids[] __initconst =
227 platform_device_register(&rtc_device); 227 { "PNP0b00", "PNP0b01", "PNP0b02", };
228#else 228 struct pnp_dev *dev;
229 struct pnp_id *id;
230 int i;
231
232 pnp_for_each_dev(dev) {
233 for (id = dev->id; id; id = id->next) {
234 for (i = 0; i < ARRAY_SIZE(ids); i++) {
235 if (compare_pnp_id(id, ids[i]) != 0)
236 return 0;
237 }
238 }
239 }
240#endif
241
229 platform_device_register(&rtc_device); 242 platform_device_register(&rtc_device);
230#endif /* CONFIG_PNP */ 243 dev_info(&rtc_device.dev,
244 "registered platform RTC device (no PNP device found)\n");
231 return 0; 245 return 0;
232} 246}
233device_initcall(add_rtc_cmos); 247device_initcall(add_rtc_cmos);
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 362d4e7f2d38..2255782e8d4b 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -223,6 +223,9 @@ unsigned long saved_video_mode;
223#define RAMDISK_LOAD_FLAG 0x4000 223#define RAMDISK_LOAD_FLAG 0x4000
224 224
225static char __initdata command_line[COMMAND_LINE_SIZE]; 225static char __initdata command_line[COMMAND_LINE_SIZE];
226#ifdef CONFIG_CMDLINE_BOOL
227static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
228#endif
226 229
227#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 230#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
228struct edd edd; 231struct edd edd;
@@ -299,7 +302,7 @@ static void __init relocate_initrd(void)
299 if (clen > MAX_MAP_CHUNK-slop) 302 if (clen > MAX_MAP_CHUNK-slop)
300 clen = MAX_MAP_CHUNK-slop; 303 clen = MAX_MAP_CHUNK-slop;
301 mapaddr = ramdisk_image & PAGE_MASK; 304 mapaddr = ramdisk_image & PAGE_MASK;
302 p = early_ioremap(mapaddr, clen+slop); 305 p = early_memremap(mapaddr, clen+slop);
303 memcpy(q, p+slop, clen); 306 memcpy(q, p+slop, clen);
304 early_iounmap(p, clen+slop); 307 early_iounmap(p, clen+slop);
305 q += clen; 308 q += clen;
@@ -376,7 +379,7 @@ static void __init parse_setup_data(void)
376 return; 379 return;
377 pa_data = boot_params.hdr.setup_data; 380 pa_data = boot_params.hdr.setup_data;
378 while (pa_data) { 381 while (pa_data) {
379 data = early_ioremap(pa_data, PAGE_SIZE); 382 data = early_memremap(pa_data, PAGE_SIZE);
380 switch (data->type) { 383 switch (data->type) {
381 case SETUP_E820_EXT: 384 case SETUP_E820_EXT:
382 parse_e820_ext(data, pa_data); 385 parse_e820_ext(data, pa_data);
@@ -399,7 +402,7 @@ static void __init e820_reserve_setup_data(void)
399 return; 402 return;
400 pa_data = boot_params.hdr.setup_data; 403 pa_data = boot_params.hdr.setup_data;
401 while (pa_data) { 404 while (pa_data) {
402 data = early_ioremap(pa_data, sizeof(*data)); 405 data = early_memremap(pa_data, sizeof(*data));
403 e820_update_range(pa_data, sizeof(*data)+data->len, 406 e820_update_range(pa_data, sizeof(*data)+data->len,
404 E820_RAM, E820_RESERVED_KERN); 407 E820_RAM, E820_RESERVED_KERN);
405 found = 1; 408 found = 1;
@@ -425,7 +428,7 @@ static void __init reserve_early_setup_data(void)
425 return; 428 return;
426 pa_data = boot_params.hdr.setup_data; 429 pa_data = boot_params.hdr.setup_data;
427 while (pa_data) { 430 while (pa_data) {
428 data = early_ioremap(pa_data, sizeof(*data)); 431 data = early_memremap(pa_data, sizeof(*data));
429 sprintf(buf, "setup data %x", data->type); 432 sprintf(buf, "setup data %x", data->type);
430 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf); 433 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
431 pa_data = data->next; 434 pa_data = data->next;
@@ -579,6 +582,190 @@ static struct x86_quirks default_x86_quirks __initdata;
579struct x86_quirks *x86_quirks __initdata = &default_x86_quirks; 582struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
580 583
581/* 584/*
585 * Some BIOSes seem to corrupt the low 64k of memory during events
586 * like suspend/resume and unplugging an HDMI cable. Reserve all
587 * remaining free memory in that area and fill it with a distinct
588 * pattern.
589 */
590#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
591#define MAX_SCAN_AREAS 8
592
593static int __read_mostly memory_corruption_check = -1;
594
595static unsigned __read_mostly corruption_check_size = 64*1024;
596static unsigned __read_mostly corruption_check_period = 60; /* seconds */
597
598static struct e820entry scan_areas[MAX_SCAN_AREAS];
599static int num_scan_areas;
600
601
602static int set_corruption_check(char *arg)
603{
604 char *end;
605
606 memory_corruption_check = simple_strtol(arg, &end, 10);
607
608 return (*end == 0) ? 0 : -EINVAL;
609}
610early_param("memory_corruption_check", set_corruption_check);
611
612static int set_corruption_check_period(char *arg)
613{
614 char *end;
615
616 corruption_check_period = simple_strtoul(arg, &end, 10);
617
618 return (*end == 0) ? 0 : -EINVAL;
619}
620early_param("memory_corruption_check_period", set_corruption_check_period);
621
622static int set_corruption_check_size(char *arg)
623{
624 char *end;
625 unsigned size;
626
627 size = memparse(arg, &end);
628
629 if (*end == '\0')
630 corruption_check_size = size;
631
632 return (size == corruption_check_size) ? 0 : -EINVAL;
633}
634early_param("memory_corruption_check_size", set_corruption_check_size);
635
636
637static void __init setup_bios_corruption_check(void)
638{
639 u64 addr = PAGE_SIZE; /* assume first page is reserved anyway */
640
641 if (memory_corruption_check == -1) {
642 memory_corruption_check =
643#ifdef CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
644 1
645#else
646 0
647#endif
648 ;
649 }
650
651 if (corruption_check_size == 0)
652 memory_corruption_check = 0;
653
654 if (!memory_corruption_check)
655 return;
656
657 corruption_check_size = round_up(corruption_check_size, PAGE_SIZE);
658
659 while(addr < corruption_check_size && num_scan_areas < MAX_SCAN_AREAS) {
660 u64 size;
661 addr = find_e820_area_size(addr, &size, PAGE_SIZE);
662
663 if (addr == 0)
664 break;
665
666 if ((addr + size) > corruption_check_size)
667 size = corruption_check_size - addr;
668
669 if (size == 0)
670 break;
671
672 e820_update_range(addr, size, E820_RAM, E820_RESERVED);
673 scan_areas[num_scan_areas].addr = addr;
674 scan_areas[num_scan_areas].size = size;
675 num_scan_areas++;
676
677 /* Assume we've already mapped this early memory */
678 memset(__va(addr), 0, size);
679
680 addr += size;
681 }
682
683 printk(KERN_INFO "Scanning %d areas for low memory corruption\n",
684 num_scan_areas);
685 update_e820();
686}
687
688static struct timer_list periodic_check_timer;
689
690void check_for_bios_corruption(void)
691{
692 int i;
693 int corruption = 0;
694
695 if (!memory_corruption_check)
696 return;
697
698 for(i = 0; i < num_scan_areas; i++) {
699 unsigned long *addr = __va(scan_areas[i].addr);
700 unsigned long size = scan_areas[i].size;
701
702 for(; size; addr++, size -= sizeof(unsigned long)) {
703 if (!*addr)
704 continue;
705 printk(KERN_ERR "Corrupted low memory at %p (%lx phys) = %08lx\n",
706 addr, __pa(addr), *addr);
707 corruption = 1;
708 *addr = 0;
709 }
710 }
711
712 WARN(corruption, KERN_ERR "Memory corruption detected in low memory\n");
713}
714
715static void periodic_check_for_corruption(unsigned long data)
716{
717 check_for_bios_corruption();
718 mod_timer(&periodic_check_timer, round_jiffies(jiffies + corruption_check_period*HZ));
719}
720
721void start_periodic_check_for_corruption(void)
722{
723 if (!memory_corruption_check || corruption_check_period == 0)
724 return;
725
726 printk(KERN_INFO "Scanning for low memory corruption every %d seconds\n",
727 corruption_check_period);
728
729 init_timer(&periodic_check_timer);
730 periodic_check_timer.function = &periodic_check_for_corruption;
731 periodic_check_for_corruption(0);
732}
733#endif
734
735static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
736{
737 printk(KERN_NOTICE
738 "%s detected: BIOS may corrupt low RAM, working it around.\n",
739 d->ident);
740
741 e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
742 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
743
744 return 0;
745}
746
747/* List of systems that have known low memory corruption BIOS problems */
748static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
749#ifdef CONFIG_X86_RESERVE_LOW_64K
750 {
751 .callback = dmi_low_memory_corruption,
752 .ident = "AMI BIOS",
753 .matches = {
754 DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
755 },
756 },
757 {
758 .callback = dmi_low_memory_corruption,
759 .ident = "Phoenix BIOS",
760 .matches = {
761 DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies, LTD"),
762 },
763 },
764#endif
765 {}
766};
767
768/*
582 * Determine if we were loaded by an EFI loader. If so, then we have also been 769 * Determine if we were loaded by an EFI loader. If so, then we have also been
583 * passed the efi memmap, systab, etc., so we should use these data structures 770 * passed the efi memmap, systab, etc., so we should use these data structures
584 * for initialization. Note, the efi init code path is determined by the 771 * for initialization. Note, the efi init code path is determined by the
@@ -665,11 +852,28 @@ void __init setup_arch(char **cmdline_p)
665 bss_resource.start = virt_to_phys(&__bss_start); 852 bss_resource.start = virt_to_phys(&__bss_start);
666 bss_resource.end = virt_to_phys(&__bss_stop)-1; 853 bss_resource.end = virt_to_phys(&__bss_stop)-1;
667 854
855#ifdef CONFIG_CMDLINE_BOOL
856#ifdef CONFIG_CMDLINE_OVERRIDE
857 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
858#else
859 if (builtin_cmdline[0]) {
860 /* append boot loader cmdline to builtin */
861 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
862 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
863 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
864 }
865#endif
866#endif
867
668 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 868 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
669 *cmdline_p = command_line; 869 *cmdline_p = command_line;
670 870
671 parse_early_param(); 871 parse_early_param();
672 872
873#ifdef CONFIG_X86_64
874 check_efer();
875#endif
876
673#if defined(CONFIG_VMI) && defined(CONFIG_X86_32) 877#if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
674 /* 878 /*
675 * Must be before kernel pagetables are setup 879 * Must be before kernel pagetables are setup
@@ -695,6 +899,10 @@ void __init setup_arch(char **cmdline_p)
695 899
696 finish_e820_parsing(); 900 finish_e820_parsing();
697 901
902 dmi_scan_machine();
903
904 dmi_check_system(bad_bios_dmi_table);
905
698#ifdef CONFIG_X86_32 906#ifdef CONFIG_X86_32
699 probe_roms(); 907 probe_roms();
700#endif 908#endif
@@ -738,7 +946,8 @@ void __init setup_arch(char **cmdline_p)
738#else 946#else
739 num_physpages = max_pfn; 947 num_physpages = max_pfn;
740 948
741 check_efer(); 949 if (cpu_has_x2apic)
950 check_x2apic();
742 951
743 /* How many end-of-memory variables you have, grandma! */ 952 /* How many end-of-memory variables you have, grandma! */
744 /* need this before calling reserve_initrd */ 953 /* need this before calling reserve_initrd */
@@ -750,6 +959,10 @@ void __init setup_arch(char **cmdline_p)
750 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 959 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
751#endif 960#endif
752 961
962#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
963 setup_bios_corruption_check();
964#endif
965
753 /* max_pfn_mapped is updated here */ 966 /* max_pfn_mapped is updated here */
754 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT); 967 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
755 max_pfn_mapped = max_low_pfn_mapped; 968 max_pfn_mapped = max_low_pfn_mapped;
@@ -778,8 +991,6 @@ void __init setup_arch(char **cmdline_p)
778 vsmp_init(); 991 vsmp_init();
779#endif 992#endif
780 993
781 dmi_scan_machine();
782
783 io_delay_init(); 994 io_delay_init();
784 995
785 /* 996 /*
@@ -787,6 +998,8 @@ void __init setup_arch(char **cmdline_p)
787 */ 998 */
788 acpi_boot_table_init(); 999 acpi_boot_table_init();
789 1000
1001 early_acpi_boot_init();
1002
790#ifdef CONFIG_ACPI_NUMA 1003#ifdef CONFIG_ACPI_NUMA
791 /* 1004 /*
792 * Parse SRAT to discover nodes. 1005 * Parse SRAT to discover nodes.
@@ -882,3 +1095,5 @@ void __init setup_arch(char **cmdline_p)
882#endif 1095#endif
883#endif 1096#endif
884} 1097}
1098
1099
diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
index 76e305e064f9..0e67f72d9316 100644
--- a/arch/x86/kernel/setup_percpu.c
+++ b/arch/x86/kernel/setup_percpu.c
@@ -162,9 +162,16 @@ void __init setup_per_cpu_areas(void)
162 printk(KERN_INFO 162 printk(KERN_INFO
163 "cpu %d has no node %d or node-local memory\n", 163 "cpu %d has no node %d or node-local memory\n",
164 cpu, node); 164 cpu, node);
165 if (ptr)
166 printk(KERN_DEBUG "per cpu data for cpu%d at %016lx\n",
167 cpu, __pa(ptr));
165 } 168 }
166 else 169 else {
167 ptr = alloc_bootmem_pages_node(NODE_DATA(node), size); 170 ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
171 if (ptr)
172 printk(KERN_DEBUG "per cpu data for cpu%d on node%d at %016lx\n",
173 cpu, node, __pa(ptr));
174 }
168#endif 175#endif
169 per_cpu_offset(cpu) = ptr - __per_cpu_start; 176 per_cpu_offset(cpu) = ptr - __per_cpu_start;
170 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); 177 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
diff --git a/arch/x86/kernel/sigframe.h b/arch/x86/kernel/sigframe.h
index 72bbb519d2dc..cc673aa55ce4 100644
--- a/arch/x86/kernel/sigframe.h
+++ b/arch/x86/kernel/sigframe.h
@@ -3,9 +3,18 @@ struct sigframe {
3 char __user *pretcode; 3 char __user *pretcode;
4 int sig; 4 int sig;
5 struct sigcontext sc; 5 struct sigcontext sc;
6 struct _fpstate fpstate; 6 /*
7 * fpstate is unused. fpstate is moved/allocated after
8 * retcode[] below. This movement allows to have the FP state and the
9 * future state extensions (xsave) stay together.
10 * And at the same time retaining the unused fpstate, prevents changing
11 * the offset of extramask[] in the sigframe and thus prevent any
12 * legacy application accessing/modifying it.
13 */
14 struct _fpstate fpstate_unused;
7 unsigned long extramask[_NSIG_WORDS-1]; 15 unsigned long extramask[_NSIG_WORDS-1];
8 char retcode[8]; 16 char retcode[8];
17 /* fp state follows here */
9}; 18};
10 19
11struct rt_sigframe { 20struct rt_sigframe {
@@ -15,13 +24,19 @@ struct rt_sigframe {
15 void __user *puc; 24 void __user *puc;
16 struct siginfo info; 25 struct siginfo info;
17 struct ucontext uc; 26 struct ucontext uc;
18 struct _fpstate fpstate;
19 char retcode[8]; 27 char retcode[8];
28 /* fp state follows here */
20}; 29};
21#else 30#else
22struct rt_sigframe { 31struct rt_sigframe {
23 char __user *pretcode; 32 char __user *pretcode;
24 struct ucontext uc; 33 struct ucontext uc;
25 struct siginfo info; 34 struct siginfo info;
35 /* fp state follows here */
26}; 36};
37
38int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
39 sigset_t *set, struct pt_regs *regs);
40int ia32_setup_frame(int sig, struct k_sigaction *ka,
41 sigset_t *set, struct pt_regs *regs);
27#endif 42#endif
diff --git a/arch/x86/kernel/signal_32.c b/arch/x86/kernel/signal_32.c
index 6fb5bcdd8933..d6dd057d0f22 100644
--- a/arch/x86/kernel/signal_32.c
+++ b/arch/x86/kernel/signal_32.c
@@ -17,6 +17,7 @@
17#include <linux/errno.h> 17#include <linux/errno.h>
18#include <linux/sched.h> 18#include <linux/sched.h>
19#include <linux/wait.h> 19#include <linux/wait.h>
20#include <linux/tracehook.h>
20#include <linux/elf.h> 21#include <linux/elf.h>
21#include <linux/smp.h> 22#include <linux/smp.h>
22#include <linux/mm.h> 23#include <linux/mm.h>
@@ -26,6 +27,8 @@
26#include <asm/uaccess.h> 27#include <asm/uaccess.h>
27#include <asm/i387.h> 28#include <asm/i387.h>
28#include <asm/vdso.h> 29#include <asm/vdso.h>
30#include <asm/syscall.h>
31#include <asm/syscalls.h>
29 32
30#include "sigframe.h" 33#include "sigframe.h"
31 34
@@ -110,6 +113,27 @@ asmlinkage int sys_sigaltstack(unsigned long bx)
110 return do_sigaltstack(uss, uoss, regs->sp); 113 return do_sigaltstack(uss, uoss, regs->sp);
111} 114}
112 115
116#define COPY(x) { \
117 err |= __get_user(regs->x, &sc->x); \
118}
119
120#define COPY_SEG(seg) { \
121 unsigned short tmp; \
122 err |= __get_user(tmp, &sc->seg); \
123 regs->seg = tmp; \
124}
125
126#define COPY_SEG_STRICT(seg) { \
127 unsigned short tmp; \
128 err |= __get_user(tmp, &sc->seg); \
129 regs->seg = tmp | 3; \
130}
131
132#define GET_SEG(seg) { \
133 unsigned short tmp; \
134 err |= __get_user(tmp, &sc->seg); \
135 loadsegment(seg, tmp); \
136}
113 137
114/* 138/*
115 * Do a signal return; undo the signal stack. 139 * Do a signal return; undo the signal stack.
@@ -118,28 +142,13 @@ static int
118restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc, 142restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
119 unsigned long *pax) 143 unsigned long *pax)
120{ 144{
145 void __user *buf;
146 unsigned int tmpflags;
121 unsigned int err = 0; 147 unsigned int err = 0;
122 148
123 /* Always make any pending restarted system calls return -EINTR */ 149 /* Always make any pending restarted system calls return -EINTR */
124 current_thread_info()->restart_block.fn = do_no_restart_syscall; 150 current_thread_info()->restart_block.fn = do_no_restart_syscall;
125 151
126#define COPY(x) err |= __get_user(regs->x, &sc->x)
127
128#define COPY_SEG(seg) \
129 { unsigned short tmp; \
130 err |= __get_user(tmp, &sc->seg); \
131 regs->seg = tmp; }
132
133#define COPY_SEG_STRICT(seg) \
134 { unsigned short tmp; \
135 err |= __get_user(tmp, &sc->seg); \
136 regs->seg = tmp|3; }
137
138#define GET_SEG(seg) \
139 { unsigned short tmp; \
140 err |= __get_user(tmp, &sc->seg); \
141 loadsegment(seg, tmp); }
142
143 GET_SEG(gs); 152 GET_SEG(gs);
144 COPY_SEG(fs); 153 COPY_SEG(fs);
145 COPY_SEG(es); 154 COPY_SEG(es);
@@ -149,38 +158,15 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
149 COPY_SEG_STRICT(cs); 158 COPY_SEG_STRICT(cs);
150 COPY_SEG_STRICT(ss); 159 COPY_SEG_STRICT(ss);
151 160
152 { 161 err |= __get_user(tmpflags, &sc->flags);
153 unsigned int tmpflags; 162 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
154 163 regs->orig_ax = -1; /* disable syscall checks */
155 err |= __get_user(tmpflags, &sc->flags);
156 regs->flags = (regs->flags & ~FIX_EFLAGS) |
157 (tmpflags & FIX_EFLAGS);
158 regs->orig_ax = -1; /* disable syscall checks */
159 }
160 164
161 { 165 err |= __get_user(buf, &sc->fpstate);
162 struct _fpstate __user *buf; 166 err |= restore_i387_xstate(buf);
163
164 err |= __get_user(buf, &sc->fpstate);
165 if (buf) {
166 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
167 goto badframe;
168 err |= restore_i387(buf);
169 } else {
170 struct task_struct *me = current;
171
172 if (used_math()) {
173 clear_fpu(me);
174 clear_used_math();
175 }
176 }
177 }
178 167
179 err |= __get_user(*pax, &sc->ax); 168 err |= __get_user(*pax, &sc->ax);
180 return err; 169 return err;
181
182badframe:
183 return 1;
184} 170}
185 171
186asmlinkage unsigned long sys_sigreturn(unsigned long __unused) 172asmlinkage unsigned long sys_sigreturn(unsigned long __unused)
@@ -226,9 +212,8 @@ badframe:
226 return 0; 212 return 0;
227} 213}
228 214
229asmlinkage int sys_rt_sigreturn(unsigned long __unused) 215static long do_rt_sigreturn(struct pt_regs *regs)
230{ 216{
231 struct pt_regs *regs = (struct pt_regs *)&__unused;
232 struct rt_sigframe __user *frame; 217 struct rt_sigframe __user *frame;
233 unsigned long ax; 218 unsigned long ax;
234 sigset_t set; 219 sigset_t set;
@@ -254,15 +239,22 @@ asmlinkage int sys_rt_sigreturn(unsigned long __unused)
254 return ax; 239 return ax;
255 240
256badframe: 241badframe:
257 force_sig(SIGSEGV, current); 242 signal_fault(regs, frame, "rt_sigreturn");
258 return 0; 243 return 0;
259} 244}
260 245
246asmlinkage int sys_rt_sigreturn(unsigned long __unused)
247{
248 struct pt_regs *regs = (struct pt_regs *)&__unused;
249
250 return do_rt_sigreturn(regs);
251}
252
261/* 253/*
262 * Set up a signal frame. 254 * Set up a signal frame.
263 */ 255 */
264static int 256static int
265setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate, 257setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
266 struct pt_regs *regs, unsigned long mask) 258 struct pt_regs *regs, unsigned long mask)
267{ 259{
268 int tmp, err = 0; 260 int tmp, err = 0;
@@ -289,7 +281,7 @@ setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate,
289 err |= __put_user(regs->sp, &sc->sp_at_signal); 281 err |= __put_user(regs->sp, &sc->sp_at_signal);
290 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss); 282 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss);
291 283
292 tmp = save_i387(fpstate); 284 tmp = save_i387_xstate(fpstate);
293 if (tmp < 0) 285 if (tmp < 0)
294 err = 1; 286 err = 1;
295 else 287 else
@@ -306,7 +298,8 @@ setup_sigcontext(struct sigcontext __user *sc, struct _fpstate __user *fpstate,
306 * Determine which stack to use.. 298 * Determine which stack to use..
307 */ 299 */
308static inline void __user * 300static inline void __user *
309get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size) 301get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size,
302 void **fpstate)
310{ 303{
311 unsigned long sp; 304 unsigned long sp;
312 305
@@ -332,6 +325,11 @@ get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size)
332 sp = (unsigned long) ka->sa.sa_restorer; 325 sp = (unsigned long) ka->sa.sa_restorer;
333 } 326 }
334 327
328 if (used_math()) {
329 sp = sp - sig_xstate_size;
330 *fpstate = (struct _fpstate *) sp;
331 }
332
335 sp -= frame_size; 333 sp -= frame_size;
336 /* 334 /*
337 * Align the stack pointer according to the i386 ABI, 335 * Align the stack pointer according to the i386 ABI,
@@ -343,38 +341,29 @@ get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size)
343} 341}
344 342
345static int 343static int
346setup_frame(int sig, struct k_sigaction *ka, sigset_t *set, 344__setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
347 struct pt_regs *regs) 345 struct pt_regs *regs)
348{ 346{
349 struct sigframe __user *frame; 347 struct sigframe __user *frame;
350 void __user *restorer; 348 void __user *restorer;
351 int err = 0; 349 int err = 0;
352 int usig; 350 void __user *fpstate = NULL;
353 351
354 frame = get_sigframe(ka, regs, sizeof(*frame)); 352 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
355 353
356 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 354 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
357 goto give_sigsegv; 355 return -EFAULT;
358 356
359 usig = current_thread_info()->exec_domain 357 if (__put_user(sig, &frame->sig))
360 && current_thread_info()->exec_domain->signal_invmap 358 return -EFAULT;
361 && sig < 32
362 ? current_thread_info()->exec_domain->signal_invmap[sig]
363 : sig;
364 359
365 err = __put_user(usig, &frame->sig); 360 if (setup_sigcontext(&frame->sc, fpstate, regs, set->sig[0]))
366 if (err) 361 return -EFAULT;
367 goto give_sigsegv;
368
369 err = setup_sigcontext(&frame->sc, &frame->fpstate, regs, set->sig[0]);
370 if (err)
371 goto give_sigsegv;
372 362
373 if (_NSIG_WORDS > 1) { 363 if (_NSIG_WORDS > 1) {
374 err = __copy_to_user(&frame->extramask, &set->sig[1], 364 if (__copy_to_user(&frame->extramask, &set->sig[1],
375 sizeof(frame->extramask)); 365 sizeof(frame->extramask)))
376 if (err) 366 return -EFAULT;
377 goto give_sigsegv;
378 } 367 }
379 368
380 if (current->mm->context.vdso) 369 if (current->mm->context.vdso)
@@ -399,7 +388,7 @@ setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
399 err |= __put_user(0x80cd, (short __user *)(frame->retcode+6)); 388 err |= __put_user(0x80cd, (short __user *)(frame->retcode+6));
400 389
401 if (err) 390 if (err)
402 goto give_sigsegv; 391 return -EFAULT;
403 392
404 /* Set up registers for signal handler */ 393 /* Set up registers for signal handler */
405 regs->sp = (unsigned long)frame; 394 regs->sp = (unsigned long)frame;
@@ -414,50 +403,43 @@ setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
414 regs->cs = __USER_CS; 403 regs->cs = __USER_CS;
415 404
416 return 0; 405 return 0;
417
418give_sigsegv:
419 force_sigsegv(sig, current);
420 return -EFAULT;
421} 406}
422 407
423static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 408static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
424 sigset_t *set, struct pt_regs *regs) 409 sigset_t *set, struct pt_regs *regs)
425{ 410{
426 struct rt_sigframe __user *frame; 411 struct rt_sigframe __user *frame;
427 void __user *restorer; 412 void __user *restorer;
428 int err = 0; 413 int err = 0;
429 int usig; 414 void __user *fpstate = NULL;
430 415
431 frame = get_sigframe(ka, regs, sizeof(*frame)); 416 frame = get_sigframe(ka, regs, sizeof(*frame), &fpstate);
432 417
433 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 418 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
434 goto give_sigsegv; 419 return -EFAULT;
435 420
436 usig = current_thread_info()->exec_domain 421 err |= __put_user(sig, &frame->sig);
437 && current_thread_info()->exec_domain->signal_invmap
438 && sig < 32
439 ? current_thread_info()->exec_domain->signal_invmap[sig]
440 : sig;
441
442 err |= __put_user(usig, &frame->sig);
443 err |= __put_user(&frame->info, &frame->pinfo); 422 err |= __put_user(&frame->info, &frame->pinfo);
444 err |= __put_user(&frame->uc, &frame->puc); 423 err |= __put_user(&frame->uc, &frame->puc);
445 err |= copy_siginfo_to_user(&frame->info, info); 424 err |= copy_siginfo_to_user(&frame->info, info);
446 if (err) 425 if (err)
447 goto give_sigsegv; 426 return -EFAULT;
448 427
449 /* Create the ucontext. */ 428 /* Create the ucontext. */
450 err |= __put_user(0, &frame->uc.uc_flags); 429 if (cpu_has_xsave)
430 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
431 else
432 err |= __put_user(0, &frame->uc.uc_flags);
451 err |= __put_user(0, &frame->uc.uc_link); 433 err |= __put_user(0, &frame->uc.uc_link);
452 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 434 err |= __put_user(current->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
453 err |= __put_user(sas_ss_flags(regs->sp), 435 err |= __put_user(sas_ss_flags(regs->sp),
454 &frame->uc.uc_stack.ss_flags); 436 &frame->uc.uc_stack.ss_flags);
455 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size); 437 err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size);
456 err |= setup_sigcontext(&frame->uc.uc_mcontext, &frame->fpstate, 438 err |= setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
457 regs, set->sig[0]); 439 regs, set->sig[0]);
458 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)); 440 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
459 if (err) 441 if (err)
460 goto give_sigsegv; 442 return -EFAULT;
461 443
462 /* Set up to return from userspace. */ 444 /* Set up to return from userspace. */
463 restorer = VDSO32_SYMBOL(current->mm->context.vdso, rt_sigreturn); 445 restorer = VDSO32_SYMBOL(current->mm->context.vdso, rt_sigreturn);
@@ -477,12 +459,12 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
477 err |= __put_user(0x80cd, (short __user *)(frame->retcode+5)); 459 err |= __put_user(0x80cd, (short __user *)(frame->retcode+5));
478 460
479 if (err) 461 if (err)
480 goto give_sigsegv; 462 return -EFAULT;
481 463
482 /* Set up registers for signal handler */ 464 /* Set up registers for signal handler */
483 regs->sp = (unsigned long)frame; 465 regs->sp = (unsigned long)frame;
484 regs->ip = (unsigned long)ka->sa.sa_handler; 466 regs->ip = (unsigned long)ka->sa.sa_handler;
485 regs->ax = (unsigned long)usig; 467 regs->ax = (unsigned long)sig;
486 regs->dx = (unsigned long)&frame->info; 468 regs->dx = (unsigned long)&frame->info;
487 regs->cx = (unsigned long)&frame->uc; 469 regs->cx = (unsigned long)&frame->uc;
488 470
@@ -492,15 +474,48 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
492 regs->cs = __USER_CS; 474 regs->cs = __USER_CS;
493 475
494 return 0; 476 return 0;
495
496give_sigsegv:
497 force_sigsegv(sig, current);
498 return -EFAULT;
499} 477}
500 478
501/* 479/*
502 * OK, we're invoking a handler: 480 * OK, we're invoking a handler:
503 */ 481 */
482static int signr_convert(int sig)
483{
484 struct thread_info *info = current_thread_info();
485
486 if (info->exec_domain && info->exec_domain->signal_invmap && sig < 32)
487 return info->exec_domain->signal_invmap[sig];
488 return sig;
489}
490
491#define is_ia32 1
492#define ia32_setup_frame __setup_frame
493#define ia32_setup_rt_frame __setup_rt_frame
494
495static int
496setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
497 sigset_t *set, struct pt_regs *regs)
498{
499 int usig = signr_convert(sig);
500 int ret;
501
502 /* Set up the stack frame */
503 if (is_ia32) {
504 if (ka->sa.sa_flags & SA_SIGINFO)
505 ret = ia32_setup_rt_frame(usig, ka, info, set, regs);
506 else
507 ret = ia32_setup_frame(usig, ka, set, regs);
508 } else
509 ret = __setup_rt_frame(sig, ka, info, set, regs);
510
511 if (ret) {
512 force_sigsegv(sig, current);
513 return -EFAULT;
514 }
515
516 return ret;
517}
518
504static int 519static int
505handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka, 520handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
506 sigset_t *oldset, struct pt_regs *regs) 521 sigset_t *oldset, struct pt_regs *regs)
@@ -508,9 +523,9 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
508 int ret; 523 int ret;
509 524
510 /* Are we from a system call? */ 525 /* Are we from a system call? */
511 if ((long)regs->orig_ax >= 0) { 526 if (syscall_get_nr(current, regs) >= 0) {
512 /* If so, check system call restarting.. */ 527 /* If so, check system call restarting.. */
513 switch (regs->ax) { 528 switch (syscall_get_error(current, regs)) {
514 case -ERESTART_RESTARTBLOCK: 529 case -ERESTART_RESTARTBLOCK:
515 case -ERESTARTNOHAND: 530 case -ERESTARTNOHAND:
516 regs->ax = -EINTR; 531 regs->ax = -EINTR;
@@ -537,15 +552,20 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
537 likely(test_and_clear_thread_flag(TIF_FORCED_TF))) 552 likely(test_and_clear_thread_flag(TIF_FORCED_TF)))
538 regs->flags &= ~X86_EFLAGS_TF; 553 regs->flags &= ~X86_EFLAGS_TF;
539 554
540 /* Set up the stack frame */ 555 ret = setup_rt_frame(sig, ka, info, oldset, regs);
541 if (ka->sa.sa_flags & SA_SIGINFO)
542 ret = setup_rt_frame(sig, ka, info, oldset, regs);
543 else
544 ret = setup_frame(sig, ka, oldset, regs);
545 556
546 if (ret) 557 if (ret)
547 return ret; 558 return ret;
548 559
560#ifdef CONFIG_X86_64
561 /*
562 * This has nothing to do with segment registers,
563 * despite the name. This magic affects uaccess.h
564 * macros' behavior. Reset it to the normal setting.
565 */
566 set_fs(USER_DS);
567#endif
568
549 /* 569 /*
550 * Clear the direction flag as per the ABI for function entry. 570 * Clear the direction flag as per the ABI for function entry.
551 */ 571 */
@@ -558,8 +578,6 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
558 * handler too. 578 * handler too.
559 */ 579 */
560 regs->flags &= ~X86_EFLAGS_TF; 580 regs->flags &= ~X86_EFLAGS_TF;
561 if (test_thread_flag(TIF_SINGLESTEP))
562 ptrace_notify(SIGTRAP);
563 581
564 spin_lock_irq(&current->sighand->siglock); 582 spin_lock_irq(&current->sighand->siglock);
565 sigorsets(&current->blocked, &current->blocked, &ka->sa.sa_mask); 583 sigorsets(&current->blocked, &current->blocked, &ka->sa.sa_mask);
@@ -568,9 +586,13 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
568 recalc_sigpending(); 586 recalc_sigpending();
569 spin_unlock_irq(&current->sighand->siglock); 587 spin_unlock_irq(&current->sighand->siglock);
570 588
589 tracehook_signal_handler(sig, info, ka, regs,
590 test_thread_flag(TIF_SINGLESTEP));
591
571 return 0; 592 return 0;
572} 593}
573 594
595#define NR_restart_syscall __NR_restart_syscall
574/* 596/*
575 * Note that 'init' is a special process: it doesn't get signals it doesn't 597 * Note that 'init' is a special process: it doesn't get signals it doesn't
576 * want to handle. Thus you cannot kill init even with a SIGKILL even by 598 * want to handle. Thus you cannot kill init even with a SIGKILL even by
@@ -623,9 +645,9 @@ static void do_signal(struct pt_regs *regs)
623 } 645 }
624 646
625 /* Did we come from a system call? */ 647 /* Did we come from a system call? */
626 if ((long)regs->orig_ax >= 0) { 648 if (syscall_get_nr(current, regs) >= 0) {
627 /* Restart the system call - no handlers present */ 649 /* Restart the system call - no handlers present */
628 switch (regs->ax) { 650 switch (syscall_get_error(current, regs)) {
629 case -ERESTARTNOHAND: 651 case -ERESTARTNOHAND:
630 case -ERESTARTSYS: 652 case -ERESTARTSYS:
631 case -ERESTARTNOINTR: 653 case -ERESTARTNOINTR:
@@ -634,7 +656,7 @@ static void do_signal(struct pt_regs *regs)
634 break; 656 break;
635 657
636 case -ERESTART_RESTARTBLOCK: 658 case -ERESTART_RESTARTBLOCK:
637 regs->ax = __NR_restart_syscall; 659 regs->ax = NR_restart_syscall;
638 regs->ip -= 2; 660 regs->ip -= 2;
639 break; 661 break;
640 } 662 }
@@ -657,9 +679,38 @@ static void do_signal(struct pt_regs *regs)
657void 679void
658do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags) 680do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags)
659{ 681{
682#if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE)
683 /* notify userspace of pending MCEs */
684 if (thread_info_flags & _TIF_MCE_NOTIFY)
685 mce_notify_user();
686#endif /* CONFIG_X86_64 && CONFIG_X86_MCE */
687
660 /* deal with pending signal delivery */ 688 /* deal with pending signal delivery */
661 if (thread_info_flags & _TIF_SIGPENDING) 689 if (thread_info_flags & _TIF_SIGPENDING)
662 do_signal(regs); 690 do_signal(regs);
663 691
692 if (thread_info_flags & _TIF_NOTIFY_RESUME) {
693 clear_thread_flag(TIF_NOTIFY_RESUME);
694 tracehook_notify_resume(regs);
695 }
696
697#ifdef CONFIG_X86_32
664 clear_thread_flag(TIF_IRET); 698 clear_thread_flag(TIF_IRET);
699#endif /* CONFIG_X86_32 */
700}
701
702void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
703{
704 struct task_struct *me = current;
705
706 if (show_unhandled_signals && printk_ratelimit()) {
707 printk(KERN_INFO
708 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
709 me->comm, me->pid, where, frame,
710 regs->ip, regs->sp, regs->orig_ax);
711 print_vma_addr(" in ", regs->ip);
712 printk(KERN_CONT "\n");
713 }
714
715 force_sig(SIGSEGV, me);
665} 716}
diff --git a/arch/x86/kernel/signal_64.c b/arch/x86/kernel/signal_64.c
index ca316b5b742c..a5c9627f4db9 100644
--- a/arch/x86/kernel/signal_64.c
+++ b/arch/x86/kernel/signal_64.c
@@ -15,17 +15,21 @@
15#include <linux/errno.h> 15#include <linux/errno.h>
16#include <linux/wait.h> 16#include <linux/wait.h>
17#include <linux/ptrace.h> 17#include <linux/ptrace.h>
18#include <linux/tracehook.h>
18#include <linux/unistd.h> 19#include <linux/unistd.h>
19#include <linux/stddef.h> 20#include <linux/stddef.h>
20#include <linux/personality.h> 21#include <linux/personality.h>
21#include <linux/compiler.h> 22#include <linux/compiler.h>
23#include <linux/uaccess.h>
24
22#include <asm/processor.h> 25#include <asm/processor.h>
23#include <asm/ucontext.h> 26#include <asm/ucontext.h>
24#include <asm/uaccess.h>
25#include <asm/i387.h> 27#include <asm/i387.h>
26#include <asm/proto.h> 28#include <asm/proto.h>
27#include <asm/ia32_unistd.h> 29#include <asm/ia32_unistd.h>
28#include <asm/mce.h> 30#include <asm/mce.h>
31#include <asm/syscall.h>
32#include <asm/syscalls.h>
29#include "sigframe.h" 33#include "sigframe.h"
30 34
31#define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP))) 35#define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
@@ -41,11 +45,6 @@
41# define FIX_EFLAGS __FIX_EFLAGS 45# define FIX_EFLAGS __FIX_EFLAGS
42#endif 46#endif
43 47
44int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
45 sigset_t *set, struct pt_regs * regs);
46int ia32_setup_frame(int sig, struct k_sigaction *ka,
47 sigset_t *set, struct pt_regs * regs);
48
49asmlinkage long 48asmlinkage long
50sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss, 49sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
51 struct pt_regs *regs) 50 struct pt_regs *regs)
@@ -53,67 +52,14 @@ sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
53 return do_sigaltstack(uss, uoss, regs->sp); 52 return do_sigaltstack(uss, uoss, regs->sp);
54} 53}
55 54
56/* 55#define COPY(x) { \
57 * Signal frame handlers. 56 err |= __get_user(regs->x, &sc->x); \
58 */
59
60static inline int save_i387(struct _fpstate __user *buf)
61{
62 struct task_struct *tsk = current;
63 int err = 0;
64
65 BUILD_BUG_ON(sizeof(struct user_i387_struct) !=
66 sizeof(tsk->thread.xstate->fxsave));
67
68 if ((unsigned long)buf % 16)
69 printk("save_i387: bad fpstate %p\n", buf);
70
71 if (!used_math())
72 return 0;
73 clear_used_math(); /* trigger finit */
74 if (task_thread_info(tsk)->status & TS_USEDFPU) {
75 err = save_i387_checking((struct i387_fxsave_struct __user *)
76 buf);
77 if (err)
78 return err;
79 task_thread_info(tsk)->status &= ~TS_USEDFPU;
80 stts();
81 } else {
82 if (__copy_to_user(buf, &tsk->thread.xstate->fxsave,
83 sizeof(struct i387_fxsave_struct)))
84 return -1;
85 }
86 return 1;
87} 57}
88 58
89/* 59#define COPY_SEG_STRICT(seg) { \
90 * This restores directly out of user space. Exceptions are handled. 60 unsigned short tmp; \
91 */ 61 err |= __get_user(tmp, &sc->seg); \
92static inline int restore_i387(struct _fpstate __user *buf) 62 regs->seg = tmp | 3; \
93{
94 struct task_struct *tsk = current;
95 int err;
96
97 if (!used_math()) {
98 err = init_fpu(tsk);
99 if (err)
100 return err;
101 }
102
103 if (!(task_thread_info(current)->status & TS_USEDFPU)) {
104 clts();
105 task_thread_info(current)->status |= TS_USEDFPU;
106 }
107 err = restore_fpu_checking((__force struct i387_fxsave_struct *)buf);
108 if (unlikely(err)) {
109 /*
110 * Encountered an error while doing the restore from the
111 * user buffer, clear the fpu state.
112 */
113 clear_fpu(tsk);
114 clear_used_math();
115 }
116 return err;
117} 63}
118 64
119/* 65/*
@@ -123,13 +69,13 @@ static int
123restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc, 69restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
124 unsigned long *pax) 70 unsigned long *pax)
125{ 71{
72 void __user *buf;
73 unsigned int tmpflags;
126 unsigned int err = 0; 74 unsigned int err = 0;
127 75
128 /* Always make any pending restarted system calls return -EINTR */ 76 /* Always make any pending restarted system calls return -EINTR */
129 current_thread_info()->restart_block.fn = do_no_restart_syscall; 77 current_thread_info()->restart_block.fn = do_no_restart_syscall;
130 78
131#define COPY(x) err |= __get_user(regs->x, &sc->x)
132
133 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx); 79 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx);
134 COPY(dx); COPY(cx); COPY(ip); 80 COPY(dx); COPY(cx); COPY(ip);
135 COPY(r8); 81 COPY(r8);
@@ -144,48 +90,24 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
144 /* Kernel saves and restores only the CS segment register on signals, 90 /* Kernel saves and restores only the CS segment register on signals,
145 * which is the bare minimum needed to allow mixed 32/64-bit code. 91 * which is the bare minimum needed to allow mixed 32/64-bit code.
146 * App's signal handler can save/restore other segments if needed. */ 92 * App's signal handler can save/restore other segments if needed. */
147 { 93 COPY_SEG_STRICT(cs);
148 unsigned cs;
149 err |= __get_user(cs, &sc->cs);
150 regs->cs = cs | 3; /* Force into user mode */
151 }
152 94
153 { 95 err |= __get_user(tmpflags, &sc->flags);
154 unsigned int tmpflags; 96 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
155 err |= __get_user(tmpflags, &sc->flags); 97 regs->orig_ax = -1; /* disable syscall checks */
156 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
157 regs->orig_ax = -1; /* disable syscall checks */
158 }
159 98
160 { 99 err |= __get_user(buf, &sc->fpstate);
161 struct _fpstate __user * buf; 100 err |= restore_i387_xstate(buf);
162 err |= __get_user(buf, &sc->fpstate);
163
164 if (buf) {
165 if (!access_ok(VERIFY_READ, buf, sizeof(*buf)))
166 goto badframe;
167 err |= restore_i387(buf);
168 } else {
169 struct task_struct *me = current;
170 if (used_math()) {
171 clear_fpu(me);
172 clear_used_math();
173 }
174 }
175 }
176 101
177 err |= __get_user(*pax, &sc->ax); 102 err |= __get_user(*pax, &sc->ax);
178 return err; 103 return err;
179
180badframe:
181 return 1;
182} 104}
183 105
184asmlinkage long sys_rt_sigreturn(struct pt_regs *regs) 106static long do_rt_sigreturn(struct pt_regs *regs)
185{ 107{
186 struct rt_sigframe __user *frame; 108 struct rt_sigframe __user *frame;
187 sigset_t set;
188 unsigned long ax; 109 unsigned long ax;
110 sigset_t set;
189 111
190 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long)); 112 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long));
191 if (!access_ok(VERIFY_READ, frame, sizeof(*frame))) 113 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
@@ -198,7 +120,7 @@ asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
198 current->blocked = set; 120 current->blocked = set;
199 recalc_sigpending(); 121 recalc_sigpending();
200 spin_unlock_irq(&current->sighand->siglock); 122 spin_unlock_irq(&current->sighand->siglock);
201 123
202 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &ax)) 124 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &ax))
203 goto badframe; 125 goto badframe;
204 126
@@ -208,16 +130,22 @@ asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
208 return ax; 130 return ax;
209 131
210badframe: 132badframe:
211 signal_fault(regs,frame,"sigreturn"); 133 signal_fault(regs, frame, "rt_sigreturn");
212 return 0; 134 return 0;
213} 135}
136
137asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
138{
139 return do_rt_sigreturn(regs);
140}
214 141
215/* 142/*
216 * Set up a signal frame. 143 * Set up a signal frame.
217 */ 144 */
218 145
219static inline int 146static inline int
220setup_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs, unsigned long mask, struct task_struct *me) 147setup_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs,
148 unsigned long mask, struct task_struct *me)
221{ 149{
222 int err = 0; 150 int err = 0;
223 151
@@ -269,41 +197,40 @@ get_stack(struct k_sigaction *ka, struct pt_regs *regs, unsigned long size)
269 sp = current->sas_ss_sp + current->sas_ss_size; 197 sp = current->sas_ss_sp + current->sas_ss_size;
270 } 198 }
271 199
272 return (void __user *)round_down(sp - size, 16); 200 return (void __user *)round_down(sp - size, 64);
273} 201}
274 202
275static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 203static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
276 sigset_t *set, struct pt_regs * regs) 204 sigset_t *set, struct pt_regs *regs)
277{ 205{
278 struct rt_sigframe __user *frame; 206 struct rt_sigframe __user *frame;
279 struct _fpstate __user *fp = NULL; 207 void __user *fp = NULL;
280 int err = 0; 208 int err = 0;
281 struct task_struct *me = current; 209 struct task_struct *me = current;
282 210
283 if (used_math()) { 211 if (used_math()) {
284 fp = get_stack(ka, regs, sizeof(struct _fpstate)); 212 fp = get_stack(ka, regs, sig_xstate_size);
285 frame = (void __user *)round_down( 213 frame = (void __user *)round_down(
286 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8; 214 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8;
287 215
288 if (!access_ok(VERIFY_WRITE, fp, sizeof(struct _fpstate))) 216 if (save_i387_xstate(fp) < 0)
289 goto give_sigsegv; 217 return -EFAULT;
290
291 if (save_i387(fp) < 0)
292 err |= -1;
293 } else 218 } else
294 frame = get_stack(ka, regs, sizeof(struct rt_sigframe)) - 8; 219 frame = get_stack(ka, regs, sizeof(struct rt_sigframe)) - 8;
295 220
296 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame))) 221 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
297 goto give_sigsegv; 222 return -EFAULT;
298 223
299 if (ka->sa.sa_flags & SA_SIGINFO) { 224 if (ka->sa.sa_flags & SA_SIGINFO) {
300 err |= copy_siginfo_to_user(&frame->info, info); 225 if (copy_siginfo_to_user(&frame->info, info))
301 if (err) 226 return -EFAULT;
302 goto give_sigsegv;
303 } 227 }
304 228
305 /* Create the ucontext. */ 229 /* Create the ucontext. */
306 err |= __put_user(0, &frame->uc.uc_flags); 230 if (cpu_has_xsave)
231 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
232 else
233 err |= __put_user(0, &frame->uc.uc_flags);
307 err |= __put_user(0, &frame->uc.uc_link); 234 err |= __put_user(0, &frame->uc.uc_link);
308 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp); 235 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
309 err |= __put_user(sas_ss_flags(regs->sp), 236 err |= __put_user(sas_ss_flags(regs->sp),
@@ -311,9 +238,9 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
311 err |= __put_user(me->sas_ss_size, &frame->uc.uc_stack.ss_size); 238 err |= __put_user(me->sas_ss_size, &frame->uc.uc_stack.ss_size);
312 err |= setup_sigcontext(&frame->uc.uc_mcontext, regs, set->sig[0], me); 239 err |= setup_sigcontext(&frame->uc.uc_mcontext, regs, set->sig[0], me);
313 err |= __put_user(fp, &frame->uc.uc_mcontext.fpstate); 240 err |= __put_user(fp, &frame->uc.uc_mcontext.fpstate);
314 if (sizeof(*set) == 16) { 241 if (sizeof(*set) == 16) {
315 __put_user(set->sig[0], &frame->uc.uc_sigmask.sig[0]); 242 __put_user(set->sig[0], &frame->uc.uc_sigmask.sig[0]);
316 __put_user(set->sig[1], &frame->uc.uc_sigmask.sig[1]); 243 __put_user(set->sig[1], &frame->uc.uc_sigmask.sig[1]);
317 } else 244 } else
318 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)); 245 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
319 246
@@ -324,15 +251,15 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
324 err |= __put_user(ka->sa.sa_restorer, &frame->pretcode); 251 err |= __put_user(ka->sa.sa_restorer, &frame->pretcode);
325 } else { 252 } else {
326 /* could use a vstub here */ 253 /* could use a vstub here */
327 goto give_sigsegv; 254 return -EFAULT;
328 } 255 }
329 256
330 if (err) 257 if (err)
331 goto give_sigsegv; 258 return -EFAULT;
332 259
333 /* Set up registers for signal handler */ 260 /* Set up registers for signal handler */
334 regs->di = sig; 261 regs->di = sig;
335 /* In case the signal handler was declared without prototypes */ 262 /* In case the signal handler was declared without prototypes */
336 regs->ax = 0; 263 regs->ax = 0;
337 264
338 /* This also works for non SA_SIGINFO handlers because they expect the 265 /* This also works for non SA_SIGINFO handlers because they expect the
@@ -348,44 +275,45 @@ static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
348 regs->cs = __USER_CS; 275 regs->cs = __USER_CS;
349 276
350 return 0; 277 return 0;
351
352give_sigsegv:
353 force_sigsegv(sig, current);
354 return -EFAULT;
355} 278}
356 279
357/* 280/*
358 * Return -1L or the syscall number that @regs is executing. 281 * OK, we're invoking a handler
359 */ 282 */
360static long current_syscall(struct pt_regs *regs) 283static int signr_convert(int sig)
361{ 284{
362 /* 285 return sig;
363 * We always sign-extend a -1 value being set here,
364 * so this is always either -1L or a syscall number.
365 */
366 return regs->orig_ax;
367} 286}
368 287
369/*
370 * Return a value that is -EFOO if the system call in @regs->orig_ax
371 * returned an error. This only works for @regs from @current.
372 */
373static long current_syscall_ret(struct pt_regs *regs)
374{
375#ifdef CONFIG_IA32_EMULATION 288#ifdef CONFIG_IA32_EMULATION
376 if (test_thread_flag(TIF_IA32)) 289#define is_ia32 test_thread_flag(TIF_IA32)
377 /* 290#else
378 * Sign-extend the value so (int)-EFOO becomes (long)-EFOO 291#define is_ia32 0
379 * and will match correctly in comparisons.
380 */
381 return (int) regs->ax;
382#endif 292#endif
383 return regs->ax;
384}
385 293
386/* 294static int
387 * OK, we're invoking a handler 295setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
388 */ 296 sigset_t *set, struct pt_regs *regs)
297{
298 int usig = signr_convert(sig);
299 int ret;
300
301 /* Set up the stack frame */
302 if (is_ia32) {
303 if (ka->sa.sa_flags & SA_SIGINFO)
304 ret = ia32_setup_rt_frame(usig, ka, info, set, regs);
305 else
306 ret = ia32_setup_frame(usig, ka, set, regs);
307 } else
308 ret = __setup_rt_frame(sig, ka, info, set, regs);
309
310 if (ret) {
311 force_sigsegv(sig, current);
312 return -EFAULT;
313 }
314
315 return ret;
316}
389 317
390static int 318static int
391handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka, 319handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
@@ -394,9 +322,9 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
394 int ret; 322 int ret;
395 323
396 /* Are we from a system call? */ 324 /* Are we from a system call? */
397 if (current_syscall(regs) >= 0) { 325 if (syscall_get_nr(current, regs) >= 0) {
398 /* If so, check system call restarting.. */ 326 /* If so, check system call restarting.. */
399 switch (current_syscall_ret(regs)) { 327 switch (syscall_get_error(current, regs)) {
400 case -ERESTART_RESTARTBLOCK: 328 case -ERESTART_RESTARTBLOCK:
401 case -ERESTARTNOHAND: 329 case -ERESTARTNOHAND:
402 regs->ax = -EINTR; 330 regs->ax = -EINTR;
@@ -423,50 +351,48 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
423 likely(test_and_clear_thread_flag(TIF_FORCED_TF))) 351 likely(test_and_clear_thread_flag(TIF_FORCED_TF)))
424 regs->flags &= ~X86_EFLAGS_TF; 352 regs->flags &= ~X86_EFLAGS_TF;
425 353
426#ifdef CONFIG_IA32_EMULATION
427 if (test_thread_flag(TIF_IA32)) {
428 if (ka->sa.sa_flags & SA_SIGINFO)
429 ret = ia32_setup_rt_frame(sig, ka, info, oldset, regs);
430 else
431 ret = ia32_setup_frame(sig, ka, oldset, regs);
432 } else
433#endif
434 ret = setup_rt_frame(sig, ka, info, oldset, regs); 354 ret = setup_rt_frame(sig, ka, info, oldset, regs);
435 355
436 if (ret == 0) { 356 if (ret)
437 /* 357 return ret;
438 * This has nothing to do with segment registers,
439 * despite the name. This magic affects uaccess.h
440 * macros' behavior. Reset it to the normal setting.
441 */
442 set_fs(USER_DS);
443 358
444 /* 359#ifdef CONFIG_X86_64
445 * Clear the direction flag as per the ABI for function entry. 360 /*
446 */ 361 * This has nothing to do with segment registers,
447 regs->flags &= ~X86_EFLAGS_DF; 362 * despite the name. This magic affects uaccess.h
363 * macros' behavior. Reset it to the normal setting.
364 */
365 set_fs(USER_DS);
366#endif
448 367
449 /* 368 /*
450 * Clear TF when entering the signal handler, but 369 * Clear the direction flag as per the ABI for function entry.
451 * notify any tracer that was single-stepping it. 370 */
452 * The tracer may want to single-step inside the 371 regs->flags &= ~X86_EFLAGS_DF;
453 * handler too.
454 */
455 regs->flags &= ~X86_EFLAGS_TF;
456 if (test_thread_flag(TIF_SINGLESTEP))
457 ptrace_notify(SIGTRAP);
458
459 spin_lock_irq(&current->sighand->siglock);
460 sigorsets(&current->blocked,&current->blocked,&ka->sa.sa_mask);
461 if (!(ka->sa.sa_flags & SA_NODEFER))
462 sigaddset(&current->blocked,sig);
463 recalc_sigpending();
464 spin_unlock_irq(&current->sighand->siglock);
465 }
466 372
467 return ret; 373 /*
374 * Clear TF when entering the signal handler, but
375 * notify any tracer that was single-stepping it.
376 * The tracer may want to single-step inside the
377 * handler too.
378 */
379 regs->flags &= ~X86_EFLAGS_TF;
380
381 spin_lock_irq(&current->sighand->siglock);
382 sigorsets(&current->blocked, &current->blocked, &ka->sa.sa_mask);
383 if (!(ka->sa.sa_flags & SA_NODEFER))
384 sigaddset(&current->blocked, sig);
385 recalc_sigpending();
386 spin_unlock_irq(&current->sighand->siglock);
387
388 tracehook_signal_handler(sig, info, ka, regs,
389 test_thread_flag(TIF_SINGLESTEP));
390
391 return 0;
468} 392}
469 393
394#define NR_restart_syscall \
395 test_thread_flag(TIF_IA32) ? __NR_ia32_restart_syscall : __NR_restart_syscall
470/* 396/*
471 * Note that 'init' is a special process: it doesn't get signals it doesn't 397 * Note that 'init' is a special process: it doesn't get signals it doesn't
472 * want to handle. Thus you cannot kill init even with a SIGKILL even by 398 * want to handle. Thus you cannot kill init even with a SIGKILL even by
@@ -496,7 +422,8 @@ static void do_signal(struct pt_regs *regs)
496 422
497 signr = get_signal_to_deliver(&info, &ka, regs, NULL); 423 signr = get_signal_to_deliver(&info, &ka, regs, NULL);
498 if (signr > 0) { 424 if (signr > 0) {
499 /* Re-enable any watchpoints before delivering the 425 /*
426 * Re-enable any watchpoints before delivering the
500 * signal to user space. The processor register will 427 * signal to user space. The processor register will
501 * have been cleared if the watchpoint triggered 428 * have been cleared if the watchpoint triggered
502 * inside the kernel. 429 * inside the kernel.
@@ -504,7 +431,7 @@ static void do_signal(struct pt_regs *regs)
504 if (current->thread.debugreg7) 431 if (current->thread.debugreg7)
505 set_debugreg(current->thread.debugreg7, 7); 432 set_debugreg(current->thread.debugreg7, 7);
506 433
507 /* Whee! Actually deliver the signal. */ 434 /* Whee! Actually deliver the signal. */
508 if (handle_signal(signr, &info, &ka, oldset, regs) == 0) { 435 if (handle_signal(signr, &info, &ka, oldset, regs) == 0) {
509 /* 436 /*
510 * A signal was successfully delivered; the saved 437 * A signal was successfully delivered; the saved
@@ -518,19 +445,18 @@ static void do_signal(struct pt_regs *regs)
518 } 445 }
519 446
520 /* Did we come from a system call? */ 447 /* Did we come from a system call? */
521 if (current_syscall(regs) >= 0) { 448 if (syscall_get_nr(current, regs) >= 0) {
522 /* Restart the system call - no handlers present */ 449 /* Restart the system call - no handlers present */
523 switch (current_syscall_ret(regs)) { 450 switch (syscall_get_error(current, regs)) {
524 case -ERESTARTNOHAND: 451 case -ERESTARTNOHAND:
525 case -ERESTARTSYS: 452 case -ERESTARTSYS:
526 case -ERESTARTNOINTR: 453 case -ERESTARTNOINTR:
527 regs->ax = regs->orig_ax; 454 regs->ax = regs->orig_ax;
528 regs->ip -= 2; 455 regs->ip -= 2;
529 break; 456 break;
457
530 case -ERESTART_RESTARTBLOCK: 458 case -ERESTART_RESTARTBLOCK:
531 regs->ax = test_thread_flag(TIF_IA32) ? 459 regs->ax = NR_restart_syscall;
532 __NR_ia32_restart_syscall :
533 __NR_restart_syscall;
534 regs->ip -= 2; 460 regs->ip -= 2;
535 break; 461 break;
536 } 462 }
@@ -546,29 +472,45 @@ static void do_signal(struct pt_regs *regs)
546 } 472 }
547} 473}
548 474
549void do_notify_resume(struct pt_regs *regs, void *unused, 475/*
550 __u32 thread_info_flags) 476 * notification of userspace execution resumption
477 * - triggered by the TIF_WORK_MASK flags
478 */
479void
480do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags)
551{ 481{
552#ifdef CONFIG_X86_MCE 482#if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE)
553 /* notify userspace of pending MCEs */ 483 /* notify userspace of pending MCEs */
554 if (thread_info_flags & _TIF_MCE_NOTIFY) 484 if (thread_info_flags & _TIF_MCE_NOTIFY)
555 mce_notify_user(); 485 mce_notify_user();
556#endif /* CONFIG_X86_MCE */ 486#endif /* CONFIG_X86_64 && CONFIG_X86_MCE */
557 487
558 /* deal with pending signal delivery */ 488 /* deal with pending signal delivery */
559 if (thread_info_flags & _TIF_SIGPENDING) 489 if (thread_info_flags & _TIF_SIGPENDING)
560 do_signal(regs); 490 do_signal(regs);
491
492 if (thread_info_flags & _TIF_NOTIFY_RESUME) {
493 clear_thread_flag(TIF_NOTIFY_RESUME);
494 tracehook_notify_resume(regs);
495 }
496
497#ifdef CONFIG_X86_32
498 clear_thread_flag(TIF_IRET);
499#endif /* CONFIG_X86_32 */
561} 500}
562 501
563void signal_fault(struct pt_regs *regs, void __user *frame, char *where) 502void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
564{ 503{
565 struct task_struct *me = current; 504 struct task_struct *me = current;
505
566 if (show_unhandled_signals && printk_ratelimit()) { 506 if (show_unhandled_signals && printk_ratelimit()) {
567 printk("%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx", 507 printk(KERN_INFO
568 me->comm,me->pid,where,frame,regs->ip,regs->sp,regs->orig_ax); 508 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
509 me->comm, me->pid, where, frame,
510 regs->ip, regs->sp, regs->orig_ax);
569 print_vma_addr(" in ", regs->ip); 511 print_vma_addr(" in ", regs->ip);
570 printk("\n"); 512 printk(KERN_CONT "\n");
571 } 513 }
572 514
573 force_sig(SIGSEGV, me); 515 force_sig(SIGSEGV, me);
574} 516}
diff --git a/arch/x86/kernel/smp.c b/arch/x86/kernel/smp.c
index 361b7a4c640c..18f9b19f5f8f 100644
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -214,12 +214,16 @@ void smp_call_function_single_interrupt(struct pt_regs *regs)
214struct smp_ops smp_ops = { 214struct smp_ops smp_ops = {
215 .smp_prepare_boot_cpu = native_smp_prepare_boot_cpu, 215 .smp_prepare_boot_cpu = native_smp_prepare_boot_cpu,
216 .smp_prepare_cpus = native_smp_prepare_cpus, 216 .smp_prepare_cpus = native_smp_prepare_cpus,
217 .cpu_up = native_cpu_up,
218 .smp_cpus_done = native_smp_cpus_done, 217 .smp_cpus_done = native_smp_cpus_done,
219 218
220 .smp_send_stop = native_smp_send_stop, 219 .smp_send_stop = native_smp_send_stop,
221 .smp_send_reschedule = native_smp_send_reschedule, 220 .smp_send_reschedule = native_smp_send_reschedule,
222 221
222 .cpu_up = native_cpu_up,
223 .cpu_die = native_cpu_die,
224 .cpu_disable = native_cpu_disable,
225 .play_dead = native_play_dead,
226
223 .send_call_func_ipi = native_send_call_func_ipi, 227 .send_call_func_ipi = native_send_call_func_ipi,
224 .send_call_func_single_ipi = native_send_call_func_single_ipi, 228 .send_call_func_single_ipi = native_send_call_func_single_ipi,
225}; 229};
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 7985c5b3f916..7ed9e070a6e9 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -52,6 +52,7 @@
52#include <asm/desc.h> 52#include <asm/desc.h>
53#include <asm/nmi.h> 53#include <asm/nmi.h>
54#include <asm/irq.h> 54#include <asm/irq.h>
55#include <asm/idle.h>
55#include <asm/smp.h> 56#include <asm/smp.h>
56#include <asm/trampoline.h> 57#include <asm/trampoline.h>
57#include <asm/cpu.h> 58#include <asm/cpu.h>
@@ -88,7 +89,7 @@ static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
88#define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x)) 89#define get_idle_for_cpu(x) (per_cpu(idle_thread_array, x))
89#define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p)) 90#define set_idle_for_cpu(x, p) (per_cpu(idle_thread_array, x) = (p))
90#else 91#else
91struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ; 92static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
92#define get_idle_for_cpu(x) (idle_thread_array[(x)]) 93#define get_idle_for_cpu(x) (idle_thread_array[(x)])
93#define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p)) 94#define set_idle_for_cpu(x, p) (idle_thread_array[(x)] = (p))
94#endif 95#endif
@@ -123,13 +124,12 @@ EXPORT_PER_CPU_SYMBOL(cpu_info);
123 124
124static atomic_t init_deasserted; 125static atomic_t init_deasserted;
125 126
126static int boot_cpu_logical_apicid;
127 127
128/* representing cpus for which sibling maps can be computed */ 128/* representing cpus for which sibling maps can be computed */
129static cpumask_t cpu_sibling_setup_map; 129static cpumask_t cpu_sibling_setup_map;
130 130
131/* Set if we find a B stepping CPU */ 131/* Set if we find a B stepping CPU */
132int __cpuinitdata smp_b_stepping; 132static int __cpuinitdata smp_b_stepping;
133 133
134#if defined(CONFIG_NUMA) && defined(CONFIG_X86_32) 134#if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
135 135
@@ -165,6 +165,8 @@ static void unmap_cpu_to_node(int cpu)
165#endif 165#endif
166 166
167#ifdef CONFIG_X86_32 167#ifdef CONFIG_X86_32
168static int boot_cpu_logical_apicid;
169
168u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly = 170u8 cpu_2_logical_apicid[NR_CPUS] __read_mostly =
169 { [0 ... NR_CPUS-1] = BAD_APICID }; 171 { [0 ... NR_CPUS-1] = BAD_APICID };
170 172
@@ -210,7 +212,7 @@ static void __cpuinit smp_callin(void)
210 /* 212 /*
211 * (This works even if the APIC is not enabled.) 213 * (This works even if the APIC is not enabled.)
212 */ 214 */
213 phys_id = GET_APIC_ID(read_apic_id()); 215 phys_id = read_apic_id();
214 cpuid = smp_processor_id(); 216 cpuid = smp_processor_id();
215 if (cpu_isset(cpuid, cpu_callin_map)) { 217 if (cpu_isset(cpuid, cpu_callin_map)) {
216 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__, 218 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
@@ -257,6 +259,7 @@ static void __cpuinit smp_callin(void)
257 end_local_APIC_setup(); 259 end_local_APIC_setup();
258 map_cpu_to_logical_apicid(); 260 map_cpu_to_logical_apicid();
259 261
262 notify_cpu_starting(cpuid);
260 /* 263 /*
261 * Get our bogomips. 264 * Get our bogomips.
262 * 265 *
@@ -279,6 +282,8 @@ static void __cpuinit smp_callin(void)
279 cpu_set(cpuid, cpu_callin_map); 282 cpu_set(cpuid, cpu_callin_map);
280} 283}
281 284
285static int __cpuinitdata unsafe_smp;
286
282/* 287/*
283 * Activate a secondary processor. 288 * Activate a secondary processor.
284 */ 289 */
@@ -331,14 +336,17 @@ static void __cpuinit start_secondary(void *unused)
331 * does not change while we are assigning vectors to cpus. Holding 336 * does not change while we are assigning vectors to cpus. Holding
332 * this lock ensures we don't half assign or remove an irq from a cpu. 337 * this lock ensures we don't half assign or remove an irq from a cpu.
333 */ 338 */
334 ipi_call_lock_irq(); 339 ipi_call_lock();
335 lock_vector_lock(); 340 lock_vector_lock();
336 __setup_vector_irq(smp_processor_id()); 341 __setup_vector_irq(smp_processor_id());
337 cpu_set(smp_processor_id(), cpu_online_map); 342 cpu_set(smp_processor_id(), cpu_online_map);
338 unlock_vector_lock(); 343 unlock_vector_lock();
339 ipi_call_unlock_irq(); 344 ipi_call_unlock();
340 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; 345 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
341 346
347 /* enable local interrupts */
348 local_irq_enable();
349
342 setup_secondary_clock(); 350 setup_secondary_clock();
343 351
344 wmb(); 352 wmb();
@@ -391,7 +399,7 @@ static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
391 goto valid_k7; 399 goto valid_k7;
392 400
393 /* If we get here, not a certified SMP capable AMD system. */ 401 /* If we get here, not a certified SMP capable AMD system. */
394 add_taint(TAINT_UNSAFE_SMP); 402 unsafe_smp = 1;
395 } 403 }
396 404
397valid_k7: 405valid_k7:
@@ -408,12 +416,10 @@ static void __cpuinit smp_checks(void)
408 * Don't taint if we are running SMP kernel on a single non-MP 416 * Don't taint if we are running SMP kernel on a single non-MP
409 * approved Athlon 417 * approved Athlon
410 */ 418 */
411 if (tainted & TAINT_UNSAFE_SMP) { 419 if (unsafe_smp && num_online_cpus() > 1) {
412 if (num_online_cpus()) 420 printk(KERN_INFO "WARNING: This combination of AMD"
413 printk(KERN_INFO "WARNING: This combination of AMD" 421 "processors is not suitable for SMP.\n");
414 "processors is not suitable for SMP.\n"); 422 add_taint(TAINT_UNSAFE_SMP);
415 else
416 tainted &= ~TAINT_UNSAFE_SMP;
417 } 423 }
418} 424}
419 425
@@ -550,8 +556,7 @@ static inline void __inquire_remote_apic(int apicid)
550 printk(KERN_CONT 556 printk(KERN_CONT
551 "a previous APIC delivery may have failed\n"); 557 "a previous APIC delivery may have failed\n");
552 558
553 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(apicid)); 559 apic_icr_write(APIC_DM_REMRD | regs[i], apicid);
554 apic_write(APIC_ICR, APIC_DM_REMRD | regs[i]);
555 560
556 timeout = 0; 561 timeout = 0;
557 do { 562 do {
@@ -583,11 +588,9 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
583 int maxlvt; 588 int maxlvt;
584 589
585 /* Target chip */ 590 /* Target chip */
586 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
587
588 /* Boot on the stack */ 591 /* Boot on the stack */
589 /* Kick the second */ 592 /* Kick the second */
590 apic_write(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL); 593 apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
591 594
592 pr_debug("Waiting for send to finish...\n"); 595 pr_debug("Waiting for send to finish...\n");
593 send_status = safe_apic_wait_icr_idle(); 596 send_status = safe_apic_wait_icr_idle();
@@ -596,10 +599,12 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
596 * Give the other CPU some time to accept the IPI. 599 * Give the other CPU some time to accept the IPI.
597 */ 600 */
598 udelay(200); 601 udelay(200);
599 maxlvt = lapic_get_maxlvt(); 602 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
600 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ 603 maxlvt = lapic_get_maxlvt();
601 apic_write(APIC_ESR, 0); 604 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
602 accept_status = (apic_read(APIC_ESR) & 0xEF); 605 apic_write(APIC_ESR, 0);
606 accept_status = (apic_read(APIC_ESR) & 0xEF);
607 }
603 pr_debug("NMI sent.\n"); 608 pr_debug("NMI sent.\n");
604 609
605 if (send_status) 610 if (send_status)
@@ -640,13 +645,11 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
640 /* 645 /*
641 * Turn INIT on target chip 646 * Turn INIT on target chip
642 */ 647 */
643 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
644
645 /* 648 /*
646 * Send IPI 649 * Send IPI
647 */ 650 */
648 apic_write(APIC_ICR, 651 apic_icr_write(APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT,
649 APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT); 652 phys_apicid);
650 653
651 pr_debug("Waiting for send to finish...\n"); 654 pr_debug("Waiting for send to finish...\n");
652 send_status = safe_apic_wait_icr_idle(); 655 send_status = safe_apic_wait_icr_idle();
@@ -656,10 +659,8 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
656 pr_debug("Deasserting INIT.\n"); 659 pr_debug("Deasserting INIT.\n");
657 660
658 /* Target chip */ 661 /* Target chip */
659 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
660
661 /* Send IPI */ 662 /* Send IPI */
662 apic_write(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT); 663 apic_icr_write(APIC_INT_LEVELTRIG | APIC_DM_INIT, phys_apicid);
663 664
664 pr_debug("Waiting for send to finish...\n"); 665 pr_debug("Waiting for send to finish...\n");
665 send_status = safe_apic_wait_icr_idle(); 666 send_status = safe_apic_wait_icr_idle();
@@ -702,11 +703,10 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
702 */ 703 */
703 704
704 /* Target chip */ 705 /* Target chip */
705 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
706
707 /* Boot on the stack */ 706 /* Boot on the stack */
708 /* Kick the second */ 707 /* Kick the second */
709 apic_write(APIC_ICR, APIC_DM_STARTUP | (start_eip >> 12)); 708 apic_icr_write(APIC_DM_STARTUP | (start_eip >> 12),
709 phys_apicid);
710 710
711 /* 711 /*
712 * Give the other CPU some time to accept the IPI. 712 * Give the other CPU some time to accept the IPI.
@@ -1175,10 +1175,17 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
1175 * Setup boot CPU information 1175 * Setup boot CPU information
1176 */ 1176 */
1177 smp_store_cpu_info(0); /* Final full version of the data */ 1177 smp_store_cpu_info(0); /* Final full version of the data */
1178#ifdef CONFIG_X86_32
1178 boot_cpu_logical_apicid = logical_smp_processor_id(); 1179 boot_cpu_logical_apicid = logical_smp_processor_id();
1180#endif
1179 current_thread_info()->cpu = 0; /* needed? */ 1181 current_thread_info()->cpu = 0; /* needed? */
1180 set_cpu_sibling_map(0); 1182 set_cpu_sibling_map(0);
1181 1183
1184#ifdef CONFIG_X86_64
1185 enable_IR_x2apic();
1186 setup_apic_routing();
1187#endif
1188
1182 if (smp_sanity_check(max_cpus) < 0) { 1189 if (smp_sanity_check(max_cpus) < 0) {
1183 printk(KERN_INFO "SMP disabled\n"); 1190 printk(KERN_INFO "SMP disabled\n");
1184 disable_smp(); 1191 disable_smp();
@@ -1186,9 +1193,9 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
1186 } 1193 }
1187 1194
1188 preempt_disable(); 1195 preempt_disable();
1189 if (GET_APIC_ID(read_apic_id()) != boot_cpu_physical_apicid) { 1196 if (read_apic_id() != boot_cpu_physical_apicid) {
1190 panic("Boot APIC ID in local APIC unexpected (%d vs %d)", 1197 panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1191 GET_APIC_ID(read_apic_id()), boot_cpu_physical_apicid); 1198 read_apic_id(), boot_cpu_physical_apicid);
1192 /* Or can we switch back to PIC here? */ 1199 /* Or can we switch back to PIC here? */
1193 } 1200 }
1194 preempt_enable(); 1201 preempt_enable();
@@ -1254,39 +1261,6 @@ void __init native_smp_cpus_done(unsigned int max_cpus)
1254 check_nmi_watchdog(); 1261 check_nmi_watchdog();
1255} 1262}
1256 1263
1257#ifdef CONFIG_HOTPLUG_CPU
1258
1259static void remove_siblinginfo(int cpu)
1260{
1261 int sibling;
1262 struct cpuinfo_x86 *c = &cpu_data(cpu);
1263
1264 for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1265 cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1266 /*/
1267 * last thread sibling in this cpu core going down
1268 */
1269 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1270 cpu_data(sibling).booted_cores--;
1271 }
1272
1273 for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1274 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1275 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1276 cpus_clear(per_cpu(cpu_core_map, cpu));
1277 c->phys_proc_id = 0;
1278 c->cpu_core_id = 0;
1279 cpu_clear(cpu, cpu_sibling_setup_map);
1280}
1281
1282static int additional_cpus __initdata = -1;
1283
1284static __init int setup_additional_cpus(char *s)
1285{
1286 return s && get_option(&s, &additional_cpus) ? 0 : -EINVAL;
1287}
1288early_param("additional_cpus", setup_additional_cpus);
1289
1290/* 1264/*
1291 * cpu_possible_map should be static, it cannot change as cpu's 1265 * cpu_possible_map should be static, it cannot change as cpu's
1292 * are onlined, or offlined. The reason is per-cpu data-structures 1266 * are onlined, or offlined. The reason is per-cpu data-structures
@@ -1306,24 +1280,13 @@ early_param("additional_cpus", setup_additional_cpus);
1306 */ 1280 */
1307__init void prefill_possible_map(void) 1281__init void prefill_possible_map(void)
1308{ 1282{
1309 int i; 1283 int i, possible;
1310 int possible;
1311 1284
1312 /* no processor from mptable or madt */ 1285 /* no processor from mptable or madt */
1313 if (!num_processors) 1286 if (!num_processors)
1314 num_processors = 1; 1287 num_processors = 1;
1315 1288
1316#ifdef CONFIG_HOTPLUG_CPU 1289 possible = num_processors + disabled_cpus;
1317 if (additional_cpus == -1) {
1318 if (disabled_cpus > 0)
1319 additional_cpus = disabled_cpus;
1320 else
1321 additional_cpus = 0;
1322 }
1323#else
1324 additional_cpus = 0;
1325#endif
1326 possible = num_processors + additional_cpus;
1327 if (possible > NR_CPUS) 1290 if (possible > NR_CPUS)
1328 possible = NR_CPUS; 1291 possible = NR_CPUS;
1329 1292
@@ -1336,6 +1299,31 @@ __init void prefill_possible_map(void)
1336 nr_cpu_ids = possible; 1299 nr_cpu_ids = possible;
1337} 1300}
1338 1301
1302#ifdef CONFIG_HOTPLUG_CPU
1303
1304static void remove_siblinginfo(int cpu)
1305{
1306 int sibling;
1307 struct cpuinfo_x86 *c = &cpu_data(cpu);
1308
1309 for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1310 cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1311 /*/
1312 * last thread sibling in this cpu core going down
1313 */
1314 if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
1315 cpu_data(sibling).booted_cores--;
1316 }
1317
1318 for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1319 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1320 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1321 cpus_clear(per_cpu(cpu_core_map, cpu));
1322 c->phys_proc_id = 0;
1323 c->cpu_core_id = 0;
1324 cpu_clear(cpu, cpu_sibling_setup_map);
1325}
1326
1339static void __ref remove_cpu_from_maps(int cpu) 1327static void __ref remove_cpu_from_maps(int cpu)
1340{ 1328{
1341 cpu_clear(cpu, cpu_online_map); 1329 cpu_clear(cpu, cpu_online_map);
@@ -1346,25 +1334,9 @@ static void __ref remove_cpu_from_maps(int cpu)
1346 numa_remove_cpu(cpu); 1334 numa_remove_cpu(cpu);
1347} 1335}
1348 1336
1349int __cpu_disable(void) 1337void cpu_disable_common(void)
1350{ 1338{
1351 int cpu = smp_processor_id(); 1339 int cpu = smp_processor_id();
1352
1353 /*
1354 * Perhaps use cpufreq to drop frequency, but that could go
1355 * into generic code.
1356 *
1357 * We won't take down the boot processor on i386 due to some
1358 * interrupts only being able to be serviced by the BSP.
1359 * Especially so if we're not using an IOAPIC -zwane
1360 */
1361 if (cpu == 0)
1362 return -EBUSY;
1363
1364 if (nmi_watchdog == NMI_LOCAL_APIC)
1365 stop_apic_nmi_watchdog(NULL);
1366 clear_local_APIC();
1367
1368 /* 1340 /*
1369 * HACK: 1341 * HACK:
1370 * Allow any queued timer interrupts to get serviced 1342 * Allow any queued timer interrupts to get serviced
@@ -1382,10 +1354,32 @@ int __cpu_disable(void)
1382 remove_cpu_from_maps(cpu); 1354 remove_cpu_from_maps(cpu);
1383 unlock_vector_lock(); 1355 unlock_vector_lock();
1384 fixup_irqs(cpu_online_map); 1356 fixup_irqs(cpu_online_map);
1357}
1358
1359int native_cpu_disable(void)
1360{
1361 int cpu = smp_processor_id();
1362
1363 /*
1364 * Perhaps use cpufreq to drop frequency, but that could go
1365 * into generic code.
1366 *
1367 * We won't take down the boot processor on i386 due to some
1368 * interrupts only being able to be serviced by the BSP.
1369 * Especially so if we're not using an IOAPIC -zwane
1370 */
1371 if (cpu == 0)
1372 return -EBUSY;
1373
1374 if (nmi_watchdog == NMI_LOCAL_APIC)
1375 stop_apic_nmi_watchdog(NULL);
1376 clear_local_APIC();
1377
1378 cpu_disable_common();
1385 return 0; 1379 return 0;
1386} 1380}
1387 1381
1388void __cpu_die(unsigned int cpu) 1382void native_cpu_die(unsigned int cpu)
1389{ 1383{
1390 /* We don't do anything here: idle task is faking death itself. */ 1384 /* We don't do anything here: idle task is faking death itself. */
1391 unsigned int i; 1385 unsigned int i;
@@ -1402,15 +1396,45 @@ void __cpu_die(unsigned int cpu)
1402 } 1396 }
1403 printk(KERN_ERR "CPU %u didn't die...\n", cpu); 1397 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1404} 1398}
1399
1400void play_dead_common(void)
1401{
1402 idle_task_exit();
1403 reset_lazy_tlbstate();
1404 irq_ctx_exit(raw_smp_processor_id());
1405 c1e_remove_cpu(raw_smp_processor_id());
1406
1407 mb();
1408 /* Ack it */
1409 __get_cpu_var(cpu_state) = CPU_DEAD;
1410
1411 /*
1412 * With physical CPU hotplug, we should halt the cpu
1413 */
1414 local_irq_disable();
1415}
1416
1417void native_play_dead(void)
1418{
1419 play_dead_common();
1420 wbinvd_halt();
1421}
1422
1405#else /* ... !CONFIG_HOTPLUG_CPU */ 1423#else /* ... !CONFIG_HOTPLUG_CPU */
1406int __cpu_disable(void) 1424int native_cpu_disable(void)
1407{ 1425{
1408 return -ENOSYS; 1426 return -ENOSYS;
1409} 1427}
1410 1428
1411void __cpu_die(unsigned int cpu) 1429void native_cpu_die(unsigned int cpu)
1412{ 1430{
1413 /* We said "no" in __cpu_disable */ 1431 /* We said "no" in __cpu_disable */
1414 BUG(); 1432 BUG();
1415} 1433}
1434
1435void native_play_dead(void)
1436{
1437 BUG();
1438}
1439
1416#endif 1440#endif
diff --git a/arch/x86/kernel/summit_32.c b/arch/x86/kernel/summit_32.c
index d67ce5f044ba..7b987852e876 100644
--- a/arch/x86/kernel/summit_32.c
+++ b/arch/x86/kernel/summit_32.c
@@ -30,7 +30,7 @@
30#include <linux/init.h> 30#include <linux/init.h>
31#include <asm/io.h> 31#include <asm/io.h>
32#include <asm/bios_ebda.h> 32#include <asm/bios_ebda.h>
33#include <asm/mach-summit/mach_mpparse.h> 33#include <asm/summit/mpparse.h>
34 34
35static struct rio_table_hdr *rio_table_hdr __initdata; 35static struct rio_table_hdr *rio_table_hdr __initdata;
36static struct scal_detail *scal_devs[MAX_NUMNODES] __initdata; 36static struct scal_detail *scal_devs[MAX_NUMNODES] __initdata;
diff --git a/arch/x86/kernel/sys_i386_32.c b/arch/x86/kernel/sys_i386_32.c
index 7066cb855a60..1884a8d12bfa 100644
--- a/arch/x86/kernel/sys_i386_32.c
+++ b/arch/x86/kernel/sys_i386_32.c
@@ -22,6 +22,8 @@
22#include <linux/uaccess.h> 22#include <linux/uaccess.h>
23#include <linux/unistd.h> 23#include <linux/unistd.h>
24 24
25#include <asm/syscalls.h>
26
25asmlinkage long sys_mmap2(unsigned long addr, unsigned long len, 27asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
26 unsigned long prot, unsigned long flags, 28 unsigned long prot, unsigned long flags,
27 unsigned long fd, unsigned long pgoff) 29 unsigned long fd, unsigned long pgoff)
diff --git a/arch/x86/kernel/sys_x86_64.c b/arch/x86/kernel/sys_x86_64.c
index 3b360ef33817..6bc211accf08 100644
--- a/arch/x86/kernel/sys_x86_64.c
+++ b/arch/x86/kernel/sys_x86_64.c
@@ -13,15 +13,17 @@
13#include <linux/utsname.h> 13#include <linux/utsname.h>
14#include <linux/personality.h> 14#include <linux/personality.h>
15#include <linux/random.h> 15#include <linux/random.h>
16#include <linux/uaccess.h>
16 17
17#include <asm/uaccess.h>
18#include <asm/ia32.h> 18#include <asm/ia32.h>
19#include <asm/syscalls.h>
19 20
20asmlinkage long sys_mmap(unsigned long addr, unsigned long len, unsigned long prot, unsigned long flags, 21asmlinkage long sys_mmap(unsigned long addr, unsigned long len,
21 unsigned long fd, unsigned long off) 22 unsigned long prot, unsigned long flags,
23 unsigned long fd, unsigned long off)
22{ 24{
23 long error; 25 long error;
24 struct file * file; 26 struct file *file;
25 27
26 error = -EINVAL; 28 error = -EINVAL;
27 if (off & ~PAGE_MASK) 29 if (off & ~PAGE_MASK)
@@ -56,9 +58,9 @@ static void find_start_end(unsigned long flags, unsigned long *begin,
56 unmapped base down for this case. This can give 58 unmapped base down for this case. This can give
57 conflicts with the heap, but we assume that glibc 59 conflicts with the heap, but we assume that glibc
58 malloc knows how to fall back to mmap. Give it 1GB 60 malloc knows how to fall back to mmap. Give it 1GB
59 of playground for now. -AK */ 61 of playground for now. -AK */
60 *begin = 0x40000000; 62 *begin = 0x40000000;
61 *end = 0x80000000; 63 *end = 0x80000000;
62 if (current->flags & PF_RANDOMIZE) { 64 if (current->flags & PF_RANDOMIZE) {
63 new_begin = randomize_range(*begin, *begin + 0x02000000, 0); 65 new_begin = randomize_range(*begin, *begin + 0x02000000, 0);
64 if (new_begin) 66 if (new_begin)
@@ -66,9 +68,9 @@ static void find_start_end(unsigned long flags, unsigned long *begin,
66 } 68 }
67 } else { 69 } else {
68 *begin = TASK_UNMAPPED_BASE; 70 *begin = TASK_UNMAPPED_BASE;
69 *end = TASK_SIZE; 71 *end = TASK_SIZE;
70 } 72 }
71} 73}
72 74
73unsigned long 75unsigned long
74arch_get_unmapped_area(struct file *filp, unsigned long addr, 76arch_get_unmapped_area(struct file *filp, unsigned long addr,
@@ -78,11 +80,11 @@ arch_get_unmapped_area(struct file *filp, unsigned long addr,
78 struct vm_area_struct *vma; 80 struct vm_area_struct *vma;
79 unsigned long start_addr; 81 unsigned long start_addr;
80 unsigned long begin, end; 82 unsigned long begin, end;
81 83
82 if (flags & MAP_FIXED) 84 if (flags & MAP_FIXED)
83 return addr; 85 return addr;
84 86
85 find_start_end(flags, &begin, &end); 87 find_start_end(flags, &begin, &end);
86 88
87 if (len > end) 89 if (len > end)
88 return -ENOMEM; 90 return -ENOMEM;
@@ -96,12 +98,12 @@ arch_get_unmapped_area(struct file *filp, unsigned long addr,
96 } 98 }
97 if (((flags & MAP_32BIT) || test_thread_flag(TIF_IA32)) 99 if (((flags & MAP_32BIT) || test_thread_flag(TIF_IA32))
98 && len <= mm->cached_hole_size) { 100 && len <= mm->cached_hole_size) {
99 mm->cached_hole_size = 0; 101 mm->cached_hole_size = 0;
100 mm->free_area_cache = begin; 102 mm->free_area_cache = begin;
101 } 103 }
102 addr = mm->free_area_cache; 104 addr = mm->free_area_cache;
103 if (addr < begin) 105 if (addr < begin)
104 addr = begin; 106 addr = begin;
105 start_addr = addr; 107 start_addr = addr;
106 108
107full_search: 109full_search:
@@ -127,7 +129,7 @@ full_search:
127 return addr; 129 return addr;
128 } 130 }
129 if (addr + mm->cached_hole_size < vma->vm_start) 131 if (addr + mm->cached_hole_size < vma->vm_start)
130 mm->cached_hole_size = vma->vm_start - addr; 132 mm->cached_hole_size = vma->vm_start - addr;
131 133
132 addr = vma->vm_end; 134 addr = vma->vm_end;
133 } 135 }
@@ -177,7 +179,7 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
177 vma = find_vma(mm, addr-len); 179 vma = find_vma(mm, addr-len);
178 if (!vma || addr <= vma->vm_start) 180 if (!vma || addr <= vma->vm_start)
179 /* remember the address as a hint for next time */ 181 /* remember the address as a hint for next time */
180 return (mm->free_area_cache = addr-len); 182 return mm->free_area_cache = addr-len;
181 } 183 }
182 184
183 if (mm->mmap_base < len) 185 if (mm->mmap_base < len)
@@ -194,7 +196,7 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
194 vma = find_vma(mm, addr); 196 vma = find_vma(mm, addr);
195 if (!vma || addr+len <= vma->vm_start) 197 if (!vma || addr+len <= vma->vm_start)
196 /* remember the address as a hint for next time */ 198 /* remember the address as a hint for next time */
197 return (mm->free_area_cache = addr); 199 return mm->free_area_cache = addr;
198 200
199 /* remember the largest hole we saw so far */ 201 /* remember the largest hole we saw so far */
200 if (addr + mm->cached_hole_size < vma->vm_start) 202 if (addr + mm->cached_hole_size < vma->vm_start)
@@ -224,13 +226,13 @@ bottomup:
224} 226}
225 227
226 228
227asmlinkage long sys_uname(struct new_utsname __user * name) 229asmlinkage long sys_uname(struct new_utsname __user *name)
228{ 230{
229 int err; 231 int err;
230 down_read(&uts_sem); 232 down_read(&uts_sem);
231 err = copy_to_user(name, utsname(), sizeof (*name)); 233 err = copy_to_user(name, utsname(), sizeof(*name));
232 up_read(&uts_sem); 234 up_read(&uts_sem);
233 if (personality(current->personality) == PER_LINUX32) 235 if (personality(current->personality) == PER_LINUX32)
234 err |= copy_to_user(&name->machine, "i686", 5); 236 err |= copy_to_user(&name->machine, "i686", 5);
235 return err ? -EFAULT : 0; 237 return err ? -EFAULT : 0;
236} 238}
diff --git a/arch/x86/kernel/syscall_64.c b/arch/x86/kernel/syscall_64.c
index 170d43c17487..3d1be4f0fac5 100644
--- a/arch/x86/kernel/syscall_64.c
+++ b/arch/x86/kernel/syscall_64.c
@@ -8,12 +8,12 @@
8#define __NO_STUBS 8#define __NO_STUBS
9 9
10#define __SYSCALL(nr, sym) extern asmlinkage void sym(void) ; 10#define __SYSCALL(nr, sym) extern asmlinkage void sym(void) ;
11#undef _ASM_X86_64_UNISTD_H_ 11#undef ASM_X86__UNISTD_64_H
12#include <asm/unistd_64.h> 12#include <asm/unistd_64.h>
13 13
14#undef __SYSCALL 14#undef __SYSCALL
15#define __SYSCALL(nr, sym) [nr] = sym, 15#define __SYSCALL(nr, sym) [nr] = sym,
16#undef _ASM_X86_64_UNISTD_H_ 16#undef ASM_X86__UNISTD_64_H
17 17
18typedef void (*sys_call_ptr_t)(void); 18typedef void (*sys_call_ptr_t)(void);
19 19
diff --git a/arch/x86/kernel/time_32.c b/arch/x86/kernel/time_32.c
index ffe3c664afc0..77b400f06ea2 100644
--- a/arch/x86/kernel/time_32.c
+++ b/arch/x86/kernel/time_32.c
@@ -36,6 +36,7 @@
36#include <asm/arch_hooks.h> 36#include <asm/arch_hooks.h>
37#include <asm/hpet.h> 37#include <asm/hpet.h>
38#include <asm/time.h> 38#include <asm/time.h>
39#include <asm/timer.h>
39 40
40#include "do_timer.h" 41#include "do_timer.h"
41 42
@@ -46,10 +47,9 @@ unsigned long profile_pc(struct pt_regs *regs)
46 unsigned long pc = instruction_pointer(regs); 47 unsigned long pc = instruction_pointer(regs);
47 48
48#ifdef CONFIG_SMP 49#ifdef CONFIG_SMP
49 if (!v8086_mode(regs) && SEGMENT_IS_KERNEL_CODE(regs->cs) && 50 if (!user_mode_vm(regs) && in_lock_functions(pc)) {
50 in_lock_functions(pc)) {
51#ifdef CONFIG_FRAME_POINTER 51#ifdef CONFIG_FRAME_POINTER
52 return *(unsigned long *)(regs->bp + 4); 52 return *(unsigned long *)(regs->bp + sizeof(long));
53#else 53#else
54 unsigned long *sp = (unsigned long *)&regs->sp; 54 unsigned long *sp = (unsigned long *)&regs->sp;
55 55
@@ -94,6 +94,7 @@ irqreturn_t timer_interrupt(int irq, void *dev_id)
94 94
95 do_timer_interrupt_hook(); 95 do_timer_interrupt_hook();
96 96
97#ifdef CONFIG_MCA
97 if (MCA_bus) { 98 if (MCA_bus) {
98 /* The PS/2 uses level-triggered interrupts. You can't 99 /* The PS/2 uses level-triggered interrupts. You can't
99 turn them off, nor would you want to (any attempt to 100 turn them off, nor would you want to (any attempt to
@@ -107,6 +108,7 @@ irqreturn_t timer_interrupt(int irq, void *dev_id)
107 u8 irq_v = inb_p( 0x61 ); /* read the current state */ 108 u8 irq_v = inb_p( 0x61 ); /* read the current state */
108 outb_p( irq_v|0x80, 0x61 ); /* reset the IRQ */ 109 outb_p( irq_v|0x80, 0x61 ); /* reset the IRQ */
109 } 110 }
111#endif
110 112
111 return IRQ_HANDLED; 113 return IRQ_HANDLED;
112} 114}
diff --git a/arch/x86/kernel/time_64.c b/arch/x86/kernel/time_64.c
index e3d49c553af2..cb19d650c216 100644
--- a/arch/x86/kernel/time_64.c
+++ b/arch/x86/kernel/time_64.c
@@ -16,6 +16,7 @@
16#include <linux/interrupt.h> 16#include <linux/interrupt.h>
17#include <linux/module.h> 17#include <linux/module.h>
18#include <linux/time.h> 18#include <linux/time.h>
19#include <linux/mca.h>
19 20
20#include <asm/i8253.h> 21#include <asm/i8253.h>
21#include <asm/hpet.h> 22#include <asm/hpet.h>
@@ -33,23 +34,34 @@ unsigned long profile_pc(struct pt_regs *regs)
33 /* Assume the lock function has either no stack frame or a copy 34 /* Assume the lock function has either no stack frame or a copy
34 of flags from PUSHF 35 of flags from PUSHF
35 Eflags always has bits 22 and up cleared unlike kernel addresses. */ 36 Eflags always has bits 22 and up cleared unlike kernel addresses. */
36 if (!user_mode(regs) && in_lock_functions(pc)) { 37 if (!user_mode_vm(regs) && in_lock_functions(pc)) {
38#ifdef CONFIG_FRAME_POINTER
39 return *(unsigned long *)(regs->bp + sizeof(long));
40#else
37 unsigned long *sp = (unsigned long *)regs->sp; 41 unsigned long *sp = (unsigned long *)regs->sp;
38 if (sp[0] >> 22) 42 if (sp[0] >> 22)
39 return sp[0]; 43 return sp[0];
40 if (sp[1] >> 22) 44 if (sp[1] >> 22)
41 return sp[1]; 45 return sp[1];
46#endif
42 } 47 }
43 return pc; 48 return pc;
44} 49}
45EXPORT_SYMBOL(profile_pc); 50EXPORT_SYMBOL(profile_pc);
46 51
47static irqreturn_t timer_event_interrupt(int irq, void *dev_id) 52irqreturn_t timer_interrupt(int irq, void *dev_id)
48{ 53{
49 add_pda(irq0_irqs, 1); 54 add_pda(irq0_irqs, 1);
50 55
51 global_clock_event->event_handler(global_clock_event); 56 global_clock_event->event_handler(global_clock_event);
52 57
58#ifdef CONFIG_MCA
59 if (MCA_bus) {
60 u8 irq_v = inb_p(0x61); /* read the current state */
61 outb_p(irq_v|0x80, 0x61); /* reset the IRQ */
62 }
63#endif
64
53 return IRQ_HANDLED; 65 return IRQ_HANDLED;
54} 66}
55 67
@@ -100,7 +112,7 @@ unsigned long __init calibrate_cpu(void)
100} 112}
101 113
102static struct irqaction irq0 = { 114static struct irqaction irq0 = {
103 .handler = timer_event_interrupt, 115 .handler = timer_interrupt,
104 .flags = IRQF_DISABLED | IRQF_IRQPOLL | IRQF_NOBALANCING, 116 .flags = IRQF_DISABLED | IRQF_IRQPOLL | IRQF_NOBALANCING,
105 .mask = CPU_MASK_NONE, 117 .mask = CPU_MASK_NONE,
106 .name = "timer" 118 .name = "timer"
@@ -111,16 +123,13 @@ void __init hpet_time_init(void)
111 if (!hpet_enable()) 123 if (!hpet_enable())
112 setup_pit_timer(); 124 setup_pit_timer();
113 125
126 irq0.mask = cpumask_of_cpu(0);
114 setup_irq(0, &irq0); 127 setup_irq(0, &irq0);
115} 128}
116 129
117void __init time_init(void) 130void __init time_init(void)
118{ 131{
119 tsc_init(); 132 tsc_init();
120 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
121 vgetcpu_mode = VGETCPU_RDTSCP;
122 else
123 vgetcpu_mode = VGETCPU_LSL;
124 133
125 late_time_init = choose_time_init(); 134 late_time_init = choose_time_init();
126} 135}
diff --git a/arch/x86/kernel/tlb_32.c b/arch/x86/kernel/tlb_32.c
index fec1ecedc9b7..e00534b33534 100644
--- a/arch/x86/kernel/tlb_32.c
+++ b/arch/x86/kernel/tlb_32.c
@@ -241,3 +241,11 @@ void flush_tlb_all(void)
241 on_each_cpu(do_flush_tlb_all, NULL, 1); 241 on_each_cpu(do_flush_tlb_all, NULL, 1);
242} 242}
243 243
244void reset_lazy_tlbstate(void)
245{
246 int cpu = raw_smp_processor_id();
247
248 per_cpu(cpu_tlbstate, cpu).state = 0;
249 per_cpu(cpu_tlbstate, cpu).active_mm = &init_mm;
250}
251
diff --git a/arch/x86/kernel/tls.c b/arch/x86/kernel/tls.c
index ab6bf375a307..6bb7b8579e70 100644
--- a/arch/x86/kernel/tls.c
+++ b/arch/x86/kernel/tls.c
@@ -10,6 +10,7 @@
10#include <asm/ldt.h> 10#include <asm/ldt.h>
11#include <asm/processor.h> 11#include <asm/processor.h>
12#include <asm/proto.h> 12#include <asm/proto.h>
13#include <asm/syscalls.h>
13 14
14#include "tls.h" 15#include "tls.h"
15 16
diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps.c
index 03df8e45e5a1..e062974cce34 100644
--- a/arch/x86/kernel/traps_32.c
+++ b/arch/x86/kernel/traps.c
@@ -7,13 +7,11 @@
7 */ 7 */
8 8
9/* 9/*
10 * 'Traps.c' handles hardware traps and faults after we have saved some 10 * Handle hardware traps and faults.
11 * state in 'asm.s'.
12 */ 11 */
13#include <linux/interrupt.h> 12#include <linux/interrupt.h>
14#include <linux/kallsyms.h> 13#include <linux/kallsyms.h>
15#include <linux/spinlock.h> 14#include <linux/spinlock.h>
16#include <linux/highmem.h>
17#include <linux/kprobes.h> 15#include <linux/kprobes.h>
18#include <linux/uaccess.h> 16#include <linux/uaccess.h>
19#include <linux/utsname.h> 17#include <linux/utsname.h>
@@ -32,6 +30,8 @@
32#include <linux/bug.h> 30#include <linux/bug.h>
33#include <linux/nmi.h> 31#include <linux/nmi.h>
34#include <linux/mm.h> 32#include <linux/mm.h>
33#include <linux/smp.h>
34#include <linux/io.h>
35 35
36#ifdef CONFIG_EISA 36#ifdef CONFIG_EISA
37#include <linux/ioport.h> 37#include <linux/ioport.h>
@@ -46,21 +46,31 @@
46#include <linux/edac.h> 46#include <linux/edac.h>
47#endif 47#endif
48 48
49#include <asm/arch_hooks.h>
50#include <asm/stacktrace.h> 49#include <asm/stacktrace.h>
51#include <asm/processor.h> 50#include <asm/processor.h>
52#include <asm/debugreg.h> 51#include <asm/debugreg.h>
53#include <asm/atomic.h> 52#include <asm/atomic.h>
54#include <asm/system.h> 53#include <asm/system.h>
55#include <asm/unwind.h> 54#include <asm/unwind.h>
55#include <asm/traps.h>
56#include <asm/desc.h> 56#include <asm/desc.h>
57#include <asm/i387.h> 57#include <asm/i387.h>
58
59#include <mach_traps.h>
60
61#ifdef CONFIG_X86_64
62#include <asm/pgalloc.h>
63#include <asm/proto.h>
64#include <asm/pda.h>
65#else
66#include <asm/processor-flags.h>
67#include <asm/arch_hooks.h>
58#include <asm/nmi.h> 68#include <asm/nmi.h>
59#include <asm/smp.h> 69#include <asm/smp.h>
60#include <asm/io.h> 70#include <asm/io.h>
61#include <asm/traps.h> 71#include <asm/traps.h>
62 72
63#include "mach_traps.h" 73#include "cpu/mcheck/mce.h"
64 74
65DECLARE_BITMAP(used_vectors, NR_VECTORS); 75DECLARE_BITMAP(used_vectors, NR_VECTORS);
66EXPORT_SYMBOL_GPL(used_vectors); 76EXPORT_SYMBOL_GPL(used_vectors);
@@ -77,418 +87,104 @@ char ignore_fpu_irq;
77 */ 87 */
78gate_desc idt_table[256] 88gate_desc idt_table[256]
79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, }; 89 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
80
81int panic_on_unrecovered_nmi;
82int kstack_depth_to_print = 24;
83static unsigned int code_bytes = 64;
84static int ignore_nmis;
85static int die_counter;
86
87void printk_address(unsigned long address, int reliable)
88{
89#ifdef CONFIG_KALLSYMS
90 unsigned long offset = 0;
91 unsigned long symsize;
92 const char *symname;
93 char *modname;
94 char *delim = ":";
95 char namebuf[KSYM_NAME_LEN];
96 char reliab[4] = "";
97
98 symname = kallsyms_lookup(address, &symsize, &offset,
99 &modname, namebuf);
100 if (!symname) {
101 printk(" [<%08lx>]\n", address);
102 return;
103 }
104 if (!reliable)
105 strcpy(reliab, "? ");
106
107 if (!modname)
108 modname = delim = "";
109 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
110 address, reliab, delim, modname, delim, symname, offset, symsize);
111#else
112 printk(" [<%08lx>]\n", address);
113#endif 90#endif
114}
115
116static inline int valid_stack_ptr(struct thread_info *tinfo,
117 void *p, unsigned int size)
118{
119 void *t = tinfo;
120 return p > t && p <= t + THREAD_SIZE - size;
121}
122
123/* The form of the top of the frame on the stack */
124struct stack_frame {
125 struct stack_frame *next_frame;
126 unsigned long return_address;
127};
128
129static inline unsigned long
130print_context_stack(struct thread_info *tinfo,
131 unsigned long *stack, unsigned long bp,
132 const struct stacktrace_ops *ops, void *data)
133{
134 struct stack_frame *frame = (struct stack_frame *)bp;
135
136 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
137 unsigned long addr;
138
139 addr = *stack;
140 if (__kernel_text_address(addr)) {
141 if ((unsigned long) stack == bp + 4) {
142 ops->address(data, addr, 1);
143 frame = frame->next_frame;
144 bp = (unsigned long) frame;
145 } else {
146 ops->address(data, addr, bp == 0);
147 }
148 }
149 stack++;
150 }
151 return bp;
152}
153
154void dump_trace(struct task_struct *task, struct pt_regs *regs,
155 unsigned long *stack, unsigned long bp,
156 const struct stacktrace_ops *ops, void *data)
157{
158 if (!task)
159 task = current;
160
161 if (!stack) {
162 unsigned long dummy;
163 stack = &dummy;
164 if (task != current)
165 stack = (unsigned long *)task->thread.sp;
166 }
167
168#ifdef CONFIG_FRAME_POINTER
169 if (!bp) {
170 if (task == current) {
171 /* Grab bp right from our regs */
172 asm("movl %%ebp, %0" : "=r" (bp) :);
173 } else {
174 /* bp is the last reg pushed by switch_to */
175 bp = *(unsigned long *) task->thread.sp;
176 }
177 }
178#endif
179
180 for (;;) {
181 struct thread_info *context;
182
183 context = (struct thread_info *)
184 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
185 bp = print_context_stack(context, stack, bp, ops, data);
186 /*
187 * Should be after the line below, but somewhere
188 * in early boot context comes out corrupted and we
189 * can't reference it:
190 */
191 if (ops->stack(data, "IRQ") < 0)
192 break;
193 stack = (unsigned long *)context->previous_esp;
194 if (!stack)
195 break;
196 touch_nmi_watchdog();
197 }
198}
199EXPORT_SYMBOL(dump_trace);
200
201static void
202print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
203{
204 printk(data);
205 print_symbol(msg, symbol);
206 printk("\n");
207}
208
209static void print_trace_warning(void *data, char *msg)
210{
211 printk("%s%s\n", (char *)data, msg);
212}
213 91
214static int print_trace_stack(void *data, char *name) 92static int ignore_nmis;
215{
216 return 0;
217}
218
219/*
220 * Print one address/symbol entries per line.
221 */
222static void print_trace_address(void *data, unsigned long addr, int reliable)
223{
224 printk("%s [<%08lx>] ", (char *)data, addr);
225 if (!reliable)
226 printk("? ");
227 print_symbol("%s\n", addr);
228 touch_nmi_watchdog();
229}
230
231static const struct stacktrace_ops print_trace_ops = {
232 .warning = print_trace_warning,
233 .warning_symbol = print_trace_warning_symbol,
234 .stack = print_trace_stack,
235 .address = print_trace_address,
236};
237 93
238static void 94static inline void conditional_sti(struct pt_regs *regs)
239show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
240 unsigned long *stack, unsigned long bp, char *log_lvl)
241{ 95{
242 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl); 96 if (regs->flags & X86_EFLAGS_IF)
243 printk("%s =======================\n", log_lvl); 97 local_irq_enable();
244} 98}
245 99
246void show_trace(struct task_struct *task, struct pt_regs *regs, 100static inline void preempt_conditional_sti(struct pt_regs *regs)
247 unsigned long *stack, unsigned long bp)
248{ 101{
249 show_trace_log_lvl(task, regs, stack, bp, ""); 102 inc_preempt_count();
103 if (regs->flags & X86_EFLAGS_IF)
104 local_irq_enable();
250} 105}
251 106
252static void 107static inline void preempt_conditional_cli(struct pt_regs *regs)
253show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
254 unsigned long *sp, unsigned long bp, char *log_lvl)
255{ 108{
256 unsigned long *stack; 109 if (regs->flags & X86_EFLAGS_IF)
257 int i; 110 local_irq_disable();
258 111 dec_preempt_count();
259 if (sp == NULL) {
260 if (task)
261 sp = (unsigned long *)task->thread.sp;
262 else
263 sp = (unsigned long *)&sp;
264 }
265
266 stack = sp;
267 for (i = 0; i < kstack_depth_to_print; i++) {
268 if (kstack_end(stack))
269 break;
270 if (i && ((i % 8) == 0))
271 printk("\n%s ", log_lvl);
272 printk("%08lx ", *stack++);
273 }
274 printk("\n%sCall Trace:\n", log_lvl);
275
276 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
277} 112}
278 113
279void show_stack(struct task_struct *task, unsigned long *sp) 114#ifdef CONFIG_X86_32
115static inline void
116die_if_kernel(const char *str, struct pt_regs *regs, long err)
280{ 117{
281 printk(" "); 118 if (!user_mode_vm(regs))
282 show_stack_log_lvl(task, NULL, sp, 0, ""); 119 die(str, regs, err);
283} 120}
284 121
285/* 122/*
286 * The architecture-independent dump_stack generator 123 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
124 * invalid offset set (the LAZY one) and the faulting thread has
125 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
126 * we set the offset field correctly and return 1.
287 */ 127 */
288void dump_stack(void) 128static int lazy_iobitmap_copy(void)
289{ 129{
290 unsigned long bp = 0; 130 struct thread_struct *thread;
291 unsigned long stack; 131 struct tss_struct *tss;
292 132 int cpu;
293#ifdef CONFIG_FRAME_POINTER
294 if (!bp)
295 asm("movl %%ebp, %0" : "=r" (bp):);
296#endif
297
298 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
299 current->pid, current->comm, print_tainted(),
300 init_utsname()->release,
301 (int)strcspn(init_utsname()->version, " "),
302 init_utsname()->version);
303
304 show_trace(current, NULL, &stack, bp);
305}
306
307EXPORT_SYMBOL(dump_stack);
308
309void show_registers(struct pt_regs *regs)
310{
311 int i;
312 133
313 print_modules(); 134 cpu = get_cpu();
314 __show_registers(regs, 0); 135 tss = &per_cpu(init_tss, cpu);
136 thread = &current->thread;
315 137
316 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)", 138 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
317 TASK_COMM_LEN, current->comm, task_pid_nr(current), 139 thread->io_bitmap_ptr) {
318 current_thread_info(), current, task_thread_info(current)); 140 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
319 /* 141 thread->io_bitmap_max);
320 * When in-kernel, we also print out the stack and code at the 142 /*
321 * time of the fault.. 143 * If the previously set map was extending to higher ports
322 */ 144 * than the current one, pad extra space with 0xff (no access).
323 if (!user_mode_vm(regs)) { 145 */
324 unsigned int code_prologue = code_bytes * 43 / 64; 146 if (thread->io_bitmap_max < tss->io_bitmap_max) {
325 unsigned int code_len = code_bytes; 147 memset((char *) tss->io_bitmap +
326 unsigned char c; 148 thread->io_bitmap_max, 0xff,
327 u8 *ip; 149 tss->io_bitmap_max - thread->io_bitmap_max);
328
329 printk("\n" KERN_EMERG "Stack: ");
330 show_stack_log_lvl(NULL, regs, &regs->sp, 0, KERN_EMERG);
331
332 printk(KERN_EMERG "Code: ");
333
334 ip = (u8 *)regs->ip - code_prologue;
335 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
336 /* try starting at EIP */
337 ip = (u8 *)regs->ip;
338 code_len = code_len - code_prologue + 1;
339 }
340 for (i = 0; i < code_len; i++, ip++) {
341 if (ip < (u8 *)PAGE_OFFSET ||
342 probe_kernel_address(ip, c)) {
343 printk(" Bad EIP value.");
344 break;
345 }
346 if (ip == (u8 *)regs->ip)
347 printk("<%02x> ", c);
348 else
349 printk("%02x ", c);
350 } 150 }
351 } 151 tss->io_bitmap_max = thread->io_bitmap_max;
352 printk("\n"); 152 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
353} 153 tss->io_bitmap_owner = thread;
354 154 put_cpu();
355int is_valid_bugaddr(unsigned long ip)
356{
357 unsigned short ud2;
358
359 if (ip < PAGE_OFFSET)
360 return 0;
361 if (probe_kernel_address((unsigned short *)ip, ud2))
362 return 0;
363
364 return ud2 == 0x0b0f;
365}
366
367static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
368static int die_owner = -1;
369static unsigned int die_nest_count;
370
371unsigned __kprobes long oops_begin(void)
372{
373 unsigned long flags;
374
375 oops_enter();
376
377 if (die_owner != raw_smp_processor_id()) {
378 console_verbose();
379 raw_local_irq_save(flags);
380 __raw_spin_lock(&die_lock);
381 die_owner = smp_processor_id();
382 die_nest_count = 0;
383 bust_spinlocks(1);
384 } else {
385 raw_local_irq_save(flags);
386 }
387 die_nest_count++;
388 return flags;
389}
390
391void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
392{
393 bust_spinlocks(0);
394 die_owner = -1;
395 add_taint(TAINT_DIE);
396 __raw_spin_unlock(&die_lock);
397 raw_local_irq_restore(flags);
398
399 if (!regs)
400 return;
401
402 if (kexec_should_crash(current))
403 crash_kexec(regs);
404
405 if (in_interrupt())
406 panic("Fatal exception in interrupt");
407
408 if (panic_on_oops)
409 panic("Fatal exception");
410
411 oops_exit();
412 do_exit(signr);
413}
414
415int __kprobes __die(const char *str, struct pt_regs *regs, long err)
416{
417 unsigned short ss;
418 unsigned long sp;
419 155
420 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
421#ifdef CONFIG_PREEMPT
422 printk("PREEMPT ");
423#endif
424#ifdef CONFIG_SMP
425 printk("SMP ");
426#endif
427#ifdef CONFIG_DEBUG_PAGEALLOC
428 printk("DEBUG_PAGEALLOC");
429#endif
430 printk("\n");
431 if (notify_die(DIE_OOPS, str, regs, err,
432 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
433 return 1; 156 return 1;
434
435 show_registers(regs);
436 /* Executive summary in case the oops scrolled away */
437 sp = (unsigned long) (&regs->sp);
438 savesegment(ss, ss);
439 if (user_mode(regs)) {
440 sp = regs->sp;
441 ss = regs->ss & 0xffff;
442 } 157 }
443 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip); 158 put_cpu();
444 print_symbol("%s", regs->ip);
445 printk(" SS:ESP %04x:%08lx\n", ss, sp);
446 return 0;
447}
448
449/*
450 * This is gone through when something in the kernel has done something bad
451 * and is about to be terminated:
452 */
453void die(const char *str, struct pt_regs *regs, long err)
454{
455 unsigned long flags = oops_begin();
456
457 if (die_nest_count < 3) {
458 report_bug(regs->ip, regs);
459
460 if (__die(str, regs, err))
461 regs = NULL;
462 } else {
463 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
464 }
465
466 oops_end(flags, regs, SIGSEGV);
467}
468 159
469static inline void 160 return 0;
470die_if_kernel(const char *str, struct pt_regs *regs, long err)
471{
472 if (!user_mode_vm(regs))
473 die(str, regs, err);
474} 161}
162#endif
475 163
476static void __kprobes 164static void __kprobes
477do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs, 165do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
478 long error_code, siginfo_t *info) 166 long error_code, siginfo_t *info)
479{ 167{
480 struct task_struct *tsk = current; 168 struct task_struct *tsk = current;
481 169
170#ifdef CONFIG_X86_32
482 if (regs->flags & X86_VM_MASK) { 171 if (regs->flags & X86_VM_MASK) {
483 if (vm86) 172 /*
173 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
174 * On nmi (interrupt 2), do_trap should not be called.
175 */
176 if (trapnr < 6)
484 goto vm86_trap; 177 goto vm86_trap;
485 goto trap_signal; 178 goto trap_signal;
486 } 179 }
180#endif
487 181
488 if (!user_mode(regs)) 182 if (!user_mode(regs))
489 goto kernel_trap; 183 goto kernel_trap;
490 184
185#ifdef CONFIG_X86_32
491trap_signal: 186trap_signal:
187#endif
492 /* 188 /*
493 * We want error_code and trap_no set for userspace faults and 189 * We want error_code and trap_no set for userspace faults and
494 * kernelspace faults which result in die(), but not 190 * kernelspace faults which result in die(), but not
@@ -501,6 +197,18 @@ trap_signal:
501 tsk->thread.error_code = error_code; 197 tsk->thread.error_code = error_code;
502 tsk->thread.trap_no = trapnr; 198 tsk->thread.trap_no = trapnr;
503 199
200#ifdef CONFIG_X86_64
201 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
202 printk_ratelimit()) {
203 printk(KERN_INFO
204 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
205 tsk->comm, tsk->pid, str,
206 regs->ip, regs->sp, error_code);
207 print_vma_addr(" in ", regs->ip);
208 printk("\n");
209 }
210#endif
211
504 if (info) 212 if (info)
505 force_sig_info(signr, info, tsk); 213 force_sig_info(signr, info, tsk);
506 else 214 else
@@ -515,29 +223,29 @@ kernel_trap:
515 } 223 }
516 return; 224 return;
517 225
226#ifdef CONFIG_X86_32
518vm86_trap: 227vm86_trap:
519 if (handle_vm86_trap((struct kernel_vm86_regs *) regs, 228 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
520 error_code, trapnr)) 229 error_code, trapnr))
521 goto trap_signal; 230 goto trap_signal;
522 return; 231 return;
232#endif
523} 233}
524 234
525#define DO_ERROR(trapnr, signr, str, name) \ 235#define DO_ERROR(trapnr, signr, str, name) \
526void do_##name(struct pt_regs *regs, long error_code) \ 236dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
527{ \ 237{ \
528 trace_hardirqs_fixup(); \
529 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 238 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
530 == NOTIFY_STOP) \ 239 == NOTIFY_STOP) \
531 return; \ 240 return; \
532 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \ 241 conditional_sti(regs); \
242 do_trap(trapnr, signr, str, regs, error_code, NULL); \
533} 243}
534 244
535#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \ 245#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
536void do_##name(struct pt_regs *regs, long error_code) \ 246dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
537{ \ 247{ \
538 siginfo_t info; \ 248 siginfo_t info; \
539 if (irq) \
540 local_irq_enable(); \
541 info.si_signo = signr; \ 249 info.si_signo = signr; \
542 info.si_errno = 0; \ 250 info.si_errno = 0; \
543 info.si_code = sicode; \ 251 info.si_code = sicode; \
@@ -545,90 +253,68 @@ void do_##name(struct pt_regs *regs, long error_code) \
545 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 253 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
546 == NOTIFY_STOP) \ 254 == NOTIFY_STOP) \
547 return; \ 255 return; \
548 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \ 256 conditional_sti(regs); \
257 do_trap(trapnr, signr, str, regs, error_code, &info); \
549} 258}
550 259
551#define DO_VM86_ERROR(trapnr, signr, str, name) \ 260DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
552void do_##name(struct pt_regs *regs, long error_code) \ 261DO_ERROR(4, SIGSEGV, "overflow", overflow)
553{ \ 262DO_ERROR(5, SIGSEGV, "bounds", bounds)
554 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 263DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
555 == NOTIFY_STOP) \
556 return; \
557 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
558}
559
560#define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
561void do_##name(struct pt_regs *regs, long error_code) \
562{ \
563 siginfo_t info; \
564 info.si_signo = signr; \
565 info.si_errno = 0; \
566 info.si_code = sicode; \
567 info.si_addr = (void __user *)siaddr; \
568 trace_hardirqs_fixup(); \
569 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
570 == NOTIFY_STOP) \
571 return; \
572 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
573}
574
575DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
576#ifndef CONFIG_KPROBES
577DO_VM86_ERROR(3, SIGTRAP, "int3", int3)
578#endif
579DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow)
580DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds)
581DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
582DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) 264DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
583DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) 265DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
584DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) 266DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
267#ifdef CONFIG_X86_32
585DO_ERROR(12, SIGBUS, "stack segment", stack_segment) 268DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
586DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0) 269#endif
587DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1) 270DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
271
272#ifdef CONFIG_X86_64
273/* Runs on IST stack */
274dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
275{
276 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
277 12, SIGBUS) == NOTIFY_STOP)
278 return;
279 preempt_conditional_sti(regs);
280 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
281 preempt_conditional_cli(regs);
282}
283
284dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
285{
286 static const char str[] = "double fault";
287 struct task_struct *tsk = current;
288
289 /* Return not checked because double check cannot be ignored */
290 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
588 291
589void __kprobes 292 tsk->thread.error_code = error_code;
293 tsk->thread.trap_no = 8;
294
295 /* This is always a kernel trap and never fixable (and thus must
296 never return). */
297 for (;;)
298 die(str, regs, error_code);
299}
300#endif
301
302dotraplinkage void __kprobes
590do_general_protection(struct pt_regs *regs, long error_code) 303do_general_protection(struct pt_regs *regs, long error_code)
591{ 304{
592 struct task_struct *tsk; 305 struct task_struct *tsk;
593 struct thread_struct *thread;
594 struct tss_struct *tss;
595 int cpu;
596 306
597 cpu = get_cpu(); 307 conditional_sti(regs);
598 tss = &per_cpu(init_tss, cpu);
599 thread = &current->thread;
600
601 /*
602 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
603 * invalid offset set (the LAZY one) and the faulting thread has
604 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
605 * and we set the offset field correctly. Then we let the CPU to
606 * restart the faulting instruction.
607 */
608 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
609 thread->io_bitmap_ptr) {
610 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
611 thread->io_bitmap_max);
612 /*
613 * If the previously set map was extending to higher ports
614 * than the current one, pad extra space with 0xff (no access).
615 */
616 if (thread->io_bitmap_max < tss->io_bitmap_max) {
617 memset((char *) tss->io_bitmap +
618 thread->io_bitmap_max, 0xff,
619 tss->io_bitmap_max - thread->io_bitmap_max);
620 }
621 tss->io_bitmap_max = thread->io_bitmap_max;
622 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
623 tss->io_bitmap_owner = thread;
624 put_cpu();
625 308
309#ifdef CONFIG_X86_32
310 if (lazy_iobitmap_copy()) {
311 /* restart the faulting instruction */
626 return; 312 return;
627 } 313 }
628 put_cpu();
629 314
630 if (regs->flags & X86_VM_MASK) 315 if (regs->flags & X86_VM_MASK)
631 goto gp_in_vm86; 316 goto gp_in_vm86;
317#endif
632 318
633 tsk = current; 319 tsk = current;
634 if (!user_mode(regs)) 320 if (!user_mode(regs))
@@ -650,10 +336,12 @@ do_general_protection(struct pt_regs *regs, long error_code)
650 force_sig(SIGSEGV, tsk); 336 force_sig(SIGSEGV, tsk);
651 return; 337 return;
652 338
339#ifdef CONFIG_X86_32
653gp_in_vm86: 340gp_in_vm86:
654 local_irq_enable(); 341 local_irq_enable();
655 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code); 342 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
656 return; 343 return;
344#endif
657 345
658gp_in_kernel: 346gp_in_kernel:
659 if (fixup_exception(regs)) 347 if (fixup_exception(regs))
@@ -690,7 +378,8 @@ mem_parity_error(unsigned char reason, struct pt_regs *regs)
690 printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); 378 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
691 379
692 /* Clear and disable the memory parity error line. */ 380 /* Clear and disable the memory parity error line. */
693 clear_mem_error(reason); 381 reason = (reason & 0xf) | 4;
382 outb(reason, 0x61);
694} 383}
695 384
696static notrace __kprobes void 385static notrace __kprobes void
@@ -716,7 +405,8 @@ io_check_error(unsigned char reason, struct pt_regs *regs)
716static notrace __kprobes void 405static notrace __kprobes void
717unknown_nmi_error(unsigned char reason, struct pt_regs *regs) 406unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
718{ 407{
719 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) 408 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
409 NOTIFY_STOP)
720 return; 410 return;
721#ifdef CONFIG_MCA 411#ifdef CONFIG_MCA
722 /* 412 /*
@@ -739,41 +429,6 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
739 printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); 429 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
740} 430}
741 431
742static DEFINE_SPINLOCK(nmi_print_lock);
743
744void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
745{
746 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
747 return;
748
749 spin_lock(&nmi_print_lock);
750 /*
751 * We are in trouble anyway, lets at least try
752 * to get a message out:
753 */
754 bust_spinlocks(1);
755 printk(KERN_EMERG "%s", str);
756 printk(" on CPU%d, ip %08lx, registers:\n",
757 smp_processor_id(), regs->ip);
758 show_registers(regs);
759 if (do_panic)
760 panic("Non maskable interrupt");
761 console_silent();
762 spin_unlock(&nmi_print_lock);
763 bust_spinlocks(0);
764
765 /*
766 * If we are in kernel we are probably nested up pretty bad
767 * and might aswell get out now while we still can:
768 */
769 if (!user_mode_vm(regs)) {
770 current->thread.trap_no = 2;
771 crash_kexec(regs);
772 }
773
774 do_exit(SIGSEGV);
775}
776
777static notrace __kprobes void default_do_nmi(struct pt_regs *regs) 432static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
778{ 433{
779 unsigned char reason = 0; 434 unsigned char reason = 0;
@@ -812,22 +467,25 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
812 mem_parity_error(reason, regs); 467 mem_parity_error(reason, regs);
813 if (reason & 0x40) 468 if (reason & 0x40)
814 io_check_error(reason, regs); 469 io_check_error(reason, regs);
470#ifdef CONFIG_X86_32
815 /* 471 /*
816 * Reassert NMI in case it became active meanwhile 472 * Reassert NMI in case it became active meanwhile
817 * as it's edge-triggered: 473 * as it's edge-triggered:
818 */ 474 */
819 reassert_nmi(); 475 reassert_nmi();
476#endif
820} 477}
821 478
822notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code) 479dotraplinkage notrace __kprobes void
480do_nmi(struct pt_regs *regs, long error_code)
823{ 481{
824 int cpu;
825
826 nmi_enter(); 482 nmi_enter();
827 483
828 cpu = smp_processor_id(); 484#ifdef CONFIG_X86_32
829 485 { int cpu; cpu = smp_processor_id(); ++nmi_count(cpu); }
830 ++nmi_count(cpu); 486#else
487 add_pda(__nmi_count, 1);
488#endif
831 489
832 if (!ignore_nmis) 490 if (!ignore_nmis)
833 default_do_nmi(regs); 491 default_do_nmi(regs);
@@ -847,21 +505,44 @@ void restart_nmi(void)
847 acpi_nmi_enable(); 505 acpi_nmi_enable();
848} 506}
849 507
850#ifdef CONFIG_KPROBES 508/* May run on IST stack. */
851void __kprobes do_int3(struct pt_regs *regs, long error_code) 509dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
852{ 510{
853 trace_hardirqs_fixup(); 511#ifdef CONFIG_KPROBES
854
855 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) 512 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
856 == NOTIFY_STOP) 513 == NOTIFY_STOP)
857 return; 514 return;
858 /* 515#else
859 * This is an interrupt gate, because kprobes wants interrupts 516 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
860 * disabled. Normal trap handlers don't. 517 == NOTIFY_STOP)
861 */ 518 return;
862 restore_interrupts(regs); 519#endif
520
521 preempt_conditional_sti(regs);
522 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
523 preempt_conditional_cli(regs);
524}
863 525
864 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL); 526#ifdef CONFIG_X86_64
527/* Help handler running on IST stack to switch back to user stack
528 for scheduling or signal handling. The actual stack switch is done in
529 entry.S */
530asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
531{
532 struct pt_regs *regs = eregs;
533 /* Did already sync */
534 if (eregs == (struct pt_regs *)eregs->sp)
535 ;
536 /* Exception from user space */
537 else if (user_mode(eregs))
538 regs = task_pt_regs(current);
539 /* Exception from kernel and interrupts are enabled. Move to
540 kernel process stack. */
541 else if (eregs->flags & X86_EFLAGS_IF)
542 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
543 if (eregs != regs)
544 *regs = *eregs;
545 return regs;
865} 546}
866#endif 547#endif
867 548
@@ -886,13 +567,14 @@ void __kprobes do_int3(struct pt_regs *regs, long error_code)
886 * about restoring all the debug state, and ptrace doesn't have to 567 * about restoring all the debug state, and ptrace doesn't have to
887 * find every occurrence of the TF bit that could be saved away even 568 * find every occurrence of the TF bit that could be saved away even
888 * by user code) 569 * by user code)
570 *
571 * May run on IST stack.
889 */ 572 */
890void __kprobes do_debug(struct pt_regs *regs, long error_code) 573dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
891{ 574{
892 struct task_struct *tsk = current; 575 struct task_struct *tsk = current;
893 unsigned int condition; 576 unsigned long condition;
894 577 int si_code;
895 trace_hardirqs_fixup();
896 578
897 get_debugreg(condition, 6); 579 get_debugreg(condition, 6);
898 580
@@ -905,9 +587,9 @@ void __kprobes do_debug(struct pt_regs *regs, long error_code)
905 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, 587 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
906 SIGTRAP) == NOTIFY_STOP) 588 SIGTRAP) == NOTIFY_STOP)
907 return; 589 return;
590
908 /* It's safe to allow irq's after DR6 has been saved */ 591 /* It's safe to allow irq's after DR6 has been saved */
909 if (regs->flags & X86_EFLAGS_IF) 592 preempt_conditional_sti(regs);
910 local_irq_enable();
911 593
912 /* Mask out spurious debug traps due to lazy DR7 setting */ 594 /* Mask out spurious debug traps due to lazy DR7 setting */
913 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { 595 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
@@ -915,8 +597,10 @@ void __kprobes do_debug(struct pt_regs *regs, long error_code)
915 goto clear_dr7; 597 goto clear_dr7;
916 } 598 }
917 599
600#ifdef CONFIG_X86_32
918 if (regs->flags & X86_VM_MASK) 601 if (regs->flags & X86_VM_MASK)
919 goto debug_vm86; 602 goto debug_vm86;
603#endif
920 604
921 /* Save debug status register where ptrace can see it */ 605 /* Save debug status register where ptrace can see it */
922 tsk->thread.debugreg6 = condition; 606 tsk->thread.debugreg6 = condition;
@@ -926,17 +610,13 @@ void __kprobes do_debug(struct pt_regs *regs, long error_code)
926 * kernel space (but re-enable TF when returning to user mode). 610 * kernel space (but re-enable TF when returning to user mode).
927 */ 611 */
928 if (condition & DR_STEP) { 612 if (condition & DR_STEP) {
929 /*
930 * We already checked v86 mode above, so we can
931 * check for kernel mode by just checking the CPL
932 * of CS.
933 */
934 if (!user_mode(regs)) 613 if (!user_mode(regs))
935 goto clear_TF_reenable; 614 goto clear_TF_reenable;
936 } 615 }
937 616
617 si_code = get_si_code(condition);
938 /* Ok, finally something we can handle */ 618 /* Ok, finally something we can handle */
939 send_sigtrap(tsk, regs, error_code); 619 send_sigtrap(tsk, regs, error_code, si_code);
940 620
941 /* 621 /*
942 * Disable additional traps. They'll be re-enabled when 622 * Disable additional traps. They'll be re-enabled when
@@ -944,18 +624,37 @@ void __kprobes do_debug(struct pt_regs *regs, long error_code)
944 */ 624 */
945clear_dr7: 625clear_dr7:
946 set_debugreg(0, 7); 626 set_debugreg(0, 7);
627 preempt_conditional_cli(regs);
947 return; 628 return;
948 629
630#ifdef CONFIG_X86_32
949debug_vm86: 631debug_vm86:
950 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1); 632 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
633 preempt_conditional_cli(regs);
951 return; 634 return;
635#endif
952 636
953clear_TF_reenable: 637clear_TF_reenable:
954 set_tsk_thread_flag(tsk, TIF_SINGLESTEP); 638 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
955 regs->flags &= ~X86_EFLAGS_TF; 639 regs->flags &= ~X86_EFLAGS_TF;
640 preempt_conditional_cli(regs);
956 return; 641 return;
957} 642}
958 643
644#ifdef CONFIG_X86_64
645static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
646{
647 if (fixup_exception(regs))
648 return 1;
649
650 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
651 /* Illegal floating point operation in the kernel */
652 current->thread.trap_no = trapnr;
653 die(str, regs, 0);
654 return 0;
655}
656#endif
657
959/* 658/*
960 * Note that we play around with the 'TS' bit in an attempt to get 659 * Note that we play around with the 'TS' bit in an attempt to get
961 * the correct behaviour even in the presence of the asynchronous 660 * the correct behaviour even in the presence of the asynchronous
@@ -992,7 +691,9 @@ void math_error(void __user *ip)
992 swd = get_fpu_swd(task); 691 swd = get_fpu_swd(task);
993 switch (swd & ~cwd & 0x3f) { 692 switch (swd & ~cwd & 0x3f) {
994 case 0x000: /* No unmasked exception */ 693 case 0x000: /* No unmasked exception */
694#ifdef CONFIG_X86_32
995 return; 695 return;
696#endif
996 default: /* Multiple exceptions */ 697 default: /* Multiple exceptions */
997 break; 698 break;
998 case 0x001: /* Invalid Op */ 699 case 0x001: /* Invalid Op */
@@ -1020,9 +721,18 @@ void math_error(void __user *ip)
1020 force_sig_info(SIGFPE, &info, task); 721 force_sig_info(SIGFPE, &info, task);
1021} 722}
1022 723
1023void do_coprocessor_error(struct pt_regs *regs, long error_code) 724dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
1024{ 725{
726 conditional_sti(regs);
727
728#ifdef CONFIG_X86_32
1025 ignore_fpu_irq = 1; 729 ignore_fpu_irq = 1;
730#else
731 if (!user_mode(regs) &&
732 kernel_math_error(regs, "kernel x87 math error", 16))
733 return;
734#endif
735
1026 math_error((void __user *)regs->ip); 736 math_error((void __user *)regs->ip);
1027} 737}
1028 738
@@ -1074,8 +784,12 @@ static void simd_math_error(void __user *ip)
1074 force_sig_info(SIGFPE, &info, task); 784 force_sig_info(SIGFPE, &info, task);
1075} 785}
1076 786
1077void do_simd_coprocessor_error(struct pt_regs *regs, long error_code) 787dotraplinkage void
788do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1078{ 789{
790 conditional_sti(regs);
791
792#ifdef CONFIG_X86_32
1079 if (cpu_has_xmm) { 793 if (cpu_has_xmm) {
1080 /* Handle SIMD FPU exceptions on PIII+ processors. */ 794 /* Handle SIMD FPU exceptions on PIII+ processors. */
1081 ignore_fpu_irq = 1; 795 ignore_fpu_irq = 1;
@@ -1094,16 +808,25 @@ void do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
1094 current->thread.error_code = error_code; 808 current->thread.error_code = error_code;
1095 die_if_kernel("cache flush denied", regs, error_code); 809 die_if_kernel("cache flush denied", regs, error_code);
1096 force_sig(SIGSEGV, current); 810 force_sig(SIGSEGV, current);
811#else
812 if (!user_mode(regs) &&
813 kernel_math_error(regs, "kernel simd math error", 19))
814 return;
815 simd_math_error((void __user *)regs->ip);
816#endif
1097} 817}
1098 818
1099void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code) 819dotraplinkage void
820do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1100{ 821{
822 conditional_sti(regs);
1101#if 0 823#if 0
1102 /* No need to warn about this any longer. */ 824 /* No need to warn about this any longer. */
1103 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n"); 825 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1104#endif 826#endif
1105} 827}
1106 828
829#ifdef CONFIG_X86_32
1107unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp) 830unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1108{ 831{
1109 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id()); 832 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
@@ -1122,6 +845,15 @@ unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1122 845
1123 return new_kesp; 846 return new_kesp;
1124} 847}
848#else
849asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
850{
851}
852
853asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
854{
855}
856#endif
1125 857
1126/* 858/*
1127 * 'math_state_restore()' saves the current math information in the 859 * 'math_state_restore()' saves the current math information in the
@@ -1154,14 +886,24 @@ asmlinkage void math_state_restore(void)
1154 } 886 }
1155 887
1156 clts(); /* Allow maths ops (or we recurse) */ 888 clts(); /* Allow maths ops (or we recurse) */
889#ifdef CONFIG_X86_32
1157 restore_fpu(tsk); 890 restore_fpu(tsk);
891#else
892 /*
893 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
894 */
895 if (unlikely(restore_fpu_checking(tsk))) {
896 stts();
897 force_sig(SIGSEGV, tsk);
898 return;
899 }
900#endif
1158 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */ 901 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1159 tsk->fpu_counter++; 902 tsk->fpu_counter++;
1160} 903}
1161EXPORT_SYMBOL_GPL(math_state_restore); 904EXPORT_SYMBOL_GPL(math_state_restore);
1162 905
1163#ifndef CONFIG_MATH_EMULATION 906#ifndef CONFIG_MATH_EMULATION
1164
1165asmlinkage void math_emulate(long arg) 907asmlinkage void math_emulate(long arg)
1166{ 908{
1167 printk(KERN_EMERG 909 printk(KERN_EMERG
@@ -1170,12 +912,54 @@ asmlinkage void math_emulate(long arg)
1170 force_sig(SIGFPE, current); 912 force_sig(SIGFPE, current);
1171 schedule(); 913 schedule();
1172} 914}
1173
1174#endif /* CONFIG_MATH_EMULATION */ 915#endif /* CONFIG_MATH_EMULATION */
1175 916
917dotraplinkage void __kprobes
918do_device_not_available(struct pt_regs *regs, long error)
919{
920#ifdef CONFIG_X86_32
921 if (read_cr0() & X86_CR0_EM) {
922 conditional_sti(regs);
923 math_emulate(0);
924 } else {
925 math_state_restore(); /* interrupts still off */
926 conditional_sti(regs);
927 }
928#else
929 math_state_restore();
930#endif
931}
932
933#ifdef CONFIG_X86_32
934#ifdef CONFIG_X86_MCE
935dotraplinkage void __kprobes do_machine_check(struct pt_regs *regs, long error)
936{
937 conditional_sti(regs);
938 machine_check_vector(regs, error);
939}
940#endif
941
942dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
943{
944 siginfo_t info;
945 local_irq_enable();
946
947 info.si_signo = SIGILL;
948 info.si_errno = 0;
949 info.si_code = ILL_BADSTK;
950 info.si_addr = 0;
951 if (notify_die(DIE_TRAP, "iret exception",
952 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
953 return;
954 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
955}
956#endif
957
1176void __init trap_init(void) 958void __init trap_init(void)
1177{ 959{
960#ifdef CONFIG_X86_32
1178 int i; 961 int i;
962#endif
1179 963
1180#ifdef CONFIG_EISA 964#ifdef CONFIG_EISA
1181 void __iomem *p = early_ioremap(0x0FFFD9, 4); 965 void __iomem *p = early_ioremap(0x0FFFD9, 4);
@@ -1185,29 +969,40 @@ void __init trap_init(void)
1185 early_iounmap(p, 4); 969 early_iounmap(p, 4);
1186#endif 970#endif
1187 971
1188 set_trap_gate(0, &divide_error); 972 set_intr_gate(0, &divide_error);
1189 set_intr_gate(1, &debug); 973 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1190 set_intr_gate(2, &nmi); 974 set_intr_gate_ist(2, &nmi, NMI_STACK);
1191 set_system_intr_gate(3, &int3); /* int3 can be called from all */ 975 /* int3 can be called from all */
1192 set_system_gate(4, &overflow); /* int4 can be called from all */ 976 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
1193 set_trap_gate(5, &bounds); 977 /* int4 can be called from all */
1194 set_trap_gate(6, &invalid_op); 978 set_system_intr_gate(4, &overflow);
1195 set_trap_gate(7, &device_not_available); 979 set_intr_gate(5, &bounds);
980 set_intr_gate(6, &invalid_op);
981 set_intr_gate(7, &device_not_available);
982#ifdef CONFIG_X86_32
1196 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS); 983 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1197 set_trap_gate(9, &coprocessor_segment_overrun); 984#else
1198 set_trap_gate(10, &invalid_TSS); 985 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1199 set_trap_gate(11, &segment_not_present); 986#endif
1200 set_trap_gate(12, &stack_segment); 987 set_intr_gate(9, &coprocessor_segment_overrun);
1201 set_trap_gate(13, &general_protection); 988 set_intr_gate(10, &invalid_TSS);
989 set_intr_gate(11, &segment_not_present);
990 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
991 set_intr_gate(13, &general_protection);
1202 set_intr_gate(14, &page_fault); 992 set_intr_gate(14, &page_fault);
1203 set_trap_gate(15, &spurious_interrupt_bug); 993 set_intr_gate(15, &spurious_interrupt_bug);
1204 set_trap_gate(16, &coprocessor_error); 994 set_intr_gate(16, &coprocessor_error);
1205 set_trap_gate(17, &alignment_check); 995 set_intr_gate(17, &alignment_check);
1206#ifdef CONFIG_X86_MCE 996#ifdef CONFIG_X86_MCE
1207 set_trap_gate(18, &machine_check); 997 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1208#endif 998#endif
1209 set_trap_gate(19, &simd_coprocessor_error); 999 set_intr_gate(19, &simd_coprocessor_error);
1210 1000
1001#ifdef CONFIG_IA32_EMULATION
1002 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1003#endif
1004
1005#ifdef CONFIG_X86_32
1211 if (cpu_has_fxsr) { 1006 if (cpu_has_fxsr) {
1212 printk(KERN_INFO "Enabling fast FPU save and restore... "); 1007 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1213 set_in_cr4(X86_CR4_OSFXSR); 1008 set_in_cr4(X86_CR4_OSFXSR);
@@ -1220,37 +1015,20 @@ void __init trap_init(void)
1220 printk("done.\n"); 1015 printk("done.\n");
1221 } 1016 }
1222 1017
1223 set_system_gate(SYSCALL_VECTOR, &system_call); 1018 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1224 1019
1225 /* Reserve all the builtin and the syscall vector: */ 1020 /* Reserve all the builtin and the syscall vector: */
1226 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) 1021 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1227 set_bit(i, used_vectors); 1022 set_bit(i, used_vectors);
1228 1023
1229 set_bit(SYSCALL_VECTOR, used_vectors); 1024 set_bit(SYSCALL_VECTOR, used_vectors);
1230 1025#endif
1231 init_thread_xstate();
1232 /* 1026 /*
1233 * Should be a barrier for any external CPU state: 1027 * Should be a barrier for any external CPU state:
1234 */ 1028 */
1235 cpu_init(); 1029 cpu_init();
1236 1030
1031#ifdef CONFIG_X86_32
1237 trap_init_hook(); 1032 trap_init_hook();
1033#endif
1238} 1034}
1239
1240static int __init kstack_setup(char *s)
1241{
1242 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1243
1244 return 1;
1245}
1246__setup("kstack=", kstack_setup);
1247
1248static int __init code_bytes_setup(char *s)
1249{
1250 code_bytes = simple_strtoul(s, NULL, 0);
1251 if (code_bytes > 8192)
1252 code_bytes = 8192;
1253
1254 return 1;
1255}
1256__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/traps_64.c b/arch/x86/kernel/traps_64.c
deleted file mode 100644
index 513caaca7115..000000000000
--- a/arch/x86/kernel/traps_64.c
+++ /dev/null
@@ -1,1212 +0,0 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9/*
10 * 'Traps.c' handles hardware traps and faults after we have saved some
11 * state in 'entry.S'.
12 */
13#include <linux/moduleparam.h>
14#include <linux/interrupt.h>
15#include <linux/kallsyms.h>
16#include <linux/spinlock.h>
17#include <linux/kprobes.h>
18#include <linux/uaccess.h>
19#include <linux/utsname.h>
20#include <linux/kdebug.h>
21#include <linux/kernel.h>
22#include <linux/module.h>
23#include <linux/ptrace.h>
24#include <linux/string.h>
25#include <linux/unwind.h>
26#include <linux/delay.h>
27#include <linux/errno.h>
28#include <linux/kexec.h>
29#include <linux/sched.h>
30#include <linux/timer.h>
31#include <linux/init.h>
32#include <linux/bug.h>
33#include <linux/nmi.h>
34#include <linux/mm.h>
35
36#if defined(CONFIG_EDAC)
37#include <linux/edac.h>
38#endif
39
40#include <asm/stacktrace.h>
41#include <asm/processor.h>
42#include <asm/debugreg.h>
43#include <asm/atomic.h>
44#include <asm/system.h>
45#include <asm/unwind.h>
46#include <asm/desc.h>
47#include <asm/i387.h>
48#include <asm/nmi.h>
49#include <asm/smp.h>
50#include <asm/io.h>
51#include <asm/pgalloc.h>
52#include <asm/proto.h>
53#include <asm/pda.h>
54#include <asm/traps.h>
55
56#include <mach_traps.h>
57
58int panic_on_unrecovered_nmi;
59int kstack_depth_to_print = 12;
60static unsigned int code_bytes = 64;
61static int ignore_nmis;
62static int die_counter;
63
64static inline void conditional_sti(struct pt_regs *regs)
65{
66 if (regs->flags & X86_EFLAGS_IF)
67 local_irq_enable();
68}
69
70static inline void preempt_conditional_sti(struct pt_regs *regs)
71{
72 inc_preempt_count();
73 if (regs->flags & X86_EFLAGS_IF)
74 local_irq_enable();
75}
76
77static inline void preempt_conditional_cli(struct pt_regs *regs)
78{
79 if (regs->flags & X86_EFLAGS_IF)
80 local_irq_disable();
81 /* Make sure to not schedule here because we could be running
82 on an exception stack. */
83 dec_preempt_count();
84}
85
86void printk_address(unsigned long address, int reliable)
87{
88 printk(" [<%016lx>] %s%pS\n", address, reliable ? "": "? ", (void *) address);
89}
90
91static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
92 unsigned *usedp, char **idp)
93{
94 static char ids[][8] = {
95 [DEBUG_STACK - 1] = "#DB",
96 [NMI_STACK - 1] = "NMI",
97 [DOUBLEFAULT_STACK - 1] = "#DF",
98 [STACKFAULT_STACK - 1] = "#SS",
99 [MCE_STACK - 1] = "#MC",
100#if DEBUG_STKSZ > EXCEPTION_STKSZ
101 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
102#endif
103 };
104 unsigned k;
105
106 /*
107 * Iterate over all exception stacks, and figure out whether
108 * 'stack' is in one of them:
109 */
110 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
111 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
112 /*
113 * Is 'stack' above this exception frame's end?
114 * If yes then skip to the next frame.
115 */
116 if (stack >= end)
117 continue;
118 /*
119 * Is 'stack' above this exception frame's start address?
120 * If yes then we found the right frame.
121 */
122 if (stack >= end - EXCEPTION_STKSZ) {
123 /*
124 * Make sure we only iterate through an exception
125 * stack once. If it comes up for the second time
126 * then there's something wrong going on - just
127 * break out and return NULL:
128 */
129 if (*usedp & (1U << k))
130 break;
131 *usedp |= 1U << k;
132 *idp = ids[k];
133 return (unsigned long *)end;
134 }
135 /*
136 * If this is a debug stack, and if it has a larger size than
137 * the usual exception stacks, then 'stack' might still
138 * be within the lower portion of the debug stack:
139 */
140#if DEBUG_STKSZ > EXCEPTION_STKSZ
141 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
142 unsigned j = N_EXCEPTION_STACKS - 1;
143
144 /*
145 * Black magic. A large debug stack is composed of
146 * multiple exception stack entries, which we
147 * iterate through now. Dont look:
148 */
149 do {
150 ++j;
151 end -= EXCEPTION_STKSZ;
152 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
153 } while (stack < end - EXCEPTION_STKSZ);
154 if (*usedp & (1U << j))
155 break;
156 *usedp |= 1U << j;
157 *idp = ids[j];
158 return (unsigned long *)end;
159 }
160#endif
161 }
162 return NULL;
163}
164
165/*
166 * x86-64 can have up to three kernel stacks:
167 * process stack
168 * interrupt stack
169 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
170 */
171
172static inline int valid_stack_ptr(struct thread_info *tinfo,
173 void *p, unsigned int size, void *end)
174{
175 void *t = tinfo;
176 if (end) {
177 if (p < end && p >= (end-THREAD_SIZE))
178 return 1;
179 else
180 return 0;
181 }
182 return p > t && p < t + THREAD_SIZE - size;
183}
184
185/* The form of the top of the frame on the stack */
186struct stack_frame {
187 struct stack_frame *next_frame;
188 unsigned long return_address;
189};
190
191static inline unsigned long
192print_context_stack(struct thread_info *tinfo,
193 unsigned long *stack, unsigned long bp,
194 const struct stacktrace_ops *ops, void *data,
195 unsigned long *end)
196{
197 struct stack_frame *frame = (struct stack_frame *)bp;
198
199 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
200 unsigned long addr;
201
202 addr = *stack;
203 if (__kernel_text_address(addr)) {
204 if ((unsigned long) stack == bp + 8) {
205 ops->address(data, addr, 1);
206 frame = frame->next_frame;
207 bp = (unsigned long) frame;
208 } else {
209 ops->address(data, addr, bp == 0);
210 }
211 }
212 stack++;
213 }
214 return bp;
215}
216
217void dump_trace(struct task_struct *task, struct pt_regs *regs,
218 unsigned long *stack, unsigned long bp,
219 const struct stacktrace_ops *ops, void *data)
220{
221 const unsigned cpu = get_cpu();
222 unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
223 unsigned used = 0;
224 struct thread_info *tinfo;
225
226 if (!task)
227 task = current;
228
229 if (!stack) {
230 unsigned long dummy;
231 stack = &dummy;
232 if (task && task != current)
233 stack = (unsigned long *)task->thread.sp;
234 }
235
236#ifdef CONFIG_FRAME_POINTER
237 if (!bp) {
238 if (task == current) {
239 /* Grab bp right from our regs */
240 asm("movq %%rbp, %0" : "=r" (bp) :);
241 } else {
242 /* bp is the last reg pushed by switch_to */
243 bp = *(unsigned long *) task->thread.sp;
244 }
245 }
246#endif
247
248 /*
249 * Print function call entries in all stacks, starting at the
250 * current stack address. If the stacks consist of nested
251 * exceptions
252 */
253 tinfo = task_thread_info(task);
254 for (;;) {
255 char *id;
256 unsigned long *estack_end;
257 estack_end = in_exception_stack(cpu, (unsigned long)stack,
258 &used, &id);
259
260 if (estack_end) {
261 if (ops->stack(data, id) < 0)
262 break;
263
264 bp = print_context_stack(tinfo, stack, bp, ops,
265 data, estack_end);
266 ops->stack(data, "<EOE>");
267 /*
268 * We link to the next stack via the
269 * second-to-last pointer (index -2 to end) in the
270 * exception stack:
271 */
272 stack = (unsigned long *) estack_end[-2];
273 continue;
274 }
275 if (irqstack_end) {
276 unsigned long *irqstack;
277 irqstack = irqstack_end -
278 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
279
280 if (stack >= irqstack && stack < irqstack_end) {
281 if (ops->stack(data, "IRQ") < 0)
282 break;
283 bp = print_context_stack(tinfo, stack, bp,
284 ops, data, irqstack_end);
285 /*
286 * We link to the next stack (which would be
287 * the process stack normally) the last
288 * pointer (index -1 to end) in the IRQ stack:
289 */
290 stack = (unsigned long *) (irqstack_end[-1]);
291 irqstack_end = NULL;
292 ops->stack(data, "EOI");
293 continue;
294 }
295 }
296 break;
297 }
298
299 /*
300 * This handles the process stack:
301 */
302 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL);
303 put_cpu();
304}
305EXPORT_SYMBOL(dump_trace);
306
307static void
308print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
309{
310 print_symbol(msg, symbol);
311 printk("\n");
312}
313
314static void print_trace_warning(void *data, char *msg)
315{
316 printk("%s\n", msg);
317}
318
319static int print_trace_stack(void *data, char *name)
320{
321 printk(" <%s> ", name);
322 return 0;
323}
324
325static void print_trace_address(void *data, unsigned long addr, int reliable)
326{
327 touch_nmi_watchdog();
328 printk_address(addr, reliable);
329}
330
331static const struct stacktrace_ops print_trace_ops = {
332 .warning = print_trace_warning,
333 .warning_symbol = print_trace_warning_symbol,
334 .stack = print_trace_stack,
335 .address = print_trace_address,
336};
337
338static void
339show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
340 unsigned long *stack, unsigned long bp, char *log_lvl)
341{
342 printk("\nCall Trace:\n");
343 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
344 printk("\n");
345}
346
347void show_trace(struct task_struct *task, struct pt_regs *regs,
348 unsigned long *stack, unsigned long bp)
349{
350 show_trace_log_lvl(task, regs, stack, bp, "");
351}
352
353static void
354show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
355 unsigned long *sp, unsigned long bp, char *log_lvl)
356{
357 unsigned long *stack;
358 int i;
359 const int cpu = smp_processor_id();
360 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
361 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
362
363 // debugging aid: "show_stack(NULL, NULL);" prints the
364 // back trace for this cpu.
365
366 if (sp == NULL) {
367 if (task)
368 sp = (unsigned long *)task->thread.sp;
369 else
370 sp = (unsigned long *)&sp;
371 }
372
373 stack = sp;
374 for (i = 0; i < kstack_depth_to_print; i++) {
375 if (stack >= irqstack && stack <= irqstack_end) {
376 if (stack == irqstack_end) {
377 stack = (unsigned long *) (irqstack_end[-1]);
378 printk(" <EOI> ");
379 }
380 } else {
381 if (((long) stack & (THREAD_SIZE-1)) == 0)
382 break;
383 }
384 if (i && ((i % 4) == 0))
385 printk("\n");
386 printk(" %016lx", *stack++);
387 touch_nmi_watchdog();
388 }
389 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
390}
391
392void show_stack(struct task_struct *task, unsigned long *sp)
393{
394 show_stack_log_lvl(task, NULL, sp, 0, "");
395}
396
397/*
398 * The architecture-independent dump_stack generator
399 */
400void dump_stack(void)
401{
402 unsigned long bp = 0;
403 unsigned long stack;
404
405#ifdef CONFIG_FRAME_POINTER
406 if (!bp)
407 asm("movq %%rbp, %0" : "=r" (bp):);
408#endif
409
410 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
411 current->pid, current->comm, print_tainted(),
412 init_utsname()->release,
413 (int)strcspn(init_utsname()->version, " "),
414 init_utsname()->version);
415 show_trace(NULL, NULL, &stack, bp);
416}
417
418EXPORT_SYMBOL(dump_stack);
419
420void show_registers(struct pt_regs *regs)
421{
422 int i;
423 unsigned long sp;
424 const int cpu = smp_processor_id();
425 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
426
427 sp = regs->sp;
428 printk("CPU %d ", cpu);
429 __show_regs(regs);
430 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
431 cur->comm, cur->pid, task_thread_info(cur), cur);
432
433 /*
434 * When in-kernel, we also print out the stack and code at the
435 * time of the fault..
436 */
437 if (!user_mode(regs)) {
438 unsigned int code_prologue = code_bytes * 43 / 64;
439 unsigned int code_len = code_bytes;
440 unsigned char c;
441 u8 *ip;
442
443 printk("Stack: ");
444 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
445 regs->bp, "");
446 printk("\n");
447
448 printk(KERN_EMERG "Code: ");
449
450 ip = (u8 *)regs->ip - code_prologue;
451 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
452 /* try starting at RIP */
453 ip = (u8 *)regs->ip;
454 code_len = code_len - code_prologue + 1;
455 }
456 for (i = 0; i < code_len; i++, ip++) {
457 if (ip < (u8 *)PAGE_OFFSET ||
458 probe_kernel_address(ip, c)) {
459 printk(" Bad RIP value.");
460 break;
461 }
462 if (ip == (u8 *)regs->ip)
463 printk("<%02x> ", c);
464 else
465 printk("%02x ", c);
466 }
467 }
468 printk("\n");
469}
470
471int is_valid_bugaddr(unsigned long ip)
472{
473 unsigned short ud2;
474
475 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
476 return 0;
477
478 return ud2 == 0x0b0f;
479}
480
481static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
482static int die_owner = -1;
483static unsigned int die_nest_count;
484
485unsigned __kprobes long oops_begin(void)
486{
487 int cpu;
488 unsigned long flags;
489
490 oops_enter();
491
492 /* racy, but better than risking deadlock. */
493 raw_local_irq_save(flags);
494 cpu = smp_processor_id();
495 if (!__raw_spin_trylock(&die_lock)) {
496 if (cpu == die_owner)
497 /* nested oops. should stop eventually */;
498 else
499 __raw_spin_lock(&die_lock);
500 }
501 die_nest_count++;
502 die_owner = cpu;
503 console_verbose();
504 bust_spinlocks(1);
505 return flags;
506}
507
508void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
509{
510 die_owner = -1;
511 bust_spinlocks(0);
512 die_nest_count--;
513 if (!die_nest_count)
514 /* Nest count reaches zero, release the lock. */
515 __raw_spin_unlock(&die_lock);
516 raw_local_irq_restore(flags);
517 if (!regs) {
518 oops_exit();
519 return;
520 }
521 if (panic_on_oops)
522 panic("Fatal exception");
523 oops_exit();
524 do_exit(signr);
525}
526
527int __kprobes __die(const char *str, struct pt_regs *regs, long err)
528{
529 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff, ++die_counter);
530#ifdef CONFIG_PREEMPT
531 printk("PREEMPT ");
532#endif
533#ifdef CONFIG_SMP
534 printk("SMP ");
535#endif
536#ifdef CONFIG_DEBUG_PAGEALLOC
537 printk("DEBUG_PAGEALLOC");
538#endif
539 printk("\n");
540 if (notify_die(DIE_OOPS, str, regs, err,
541 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
542 return 1;
543
544 show_registers(regs);
545 add_taint(TAINT_DIE);
546 /* Executive summary in case the oops scrolled away */
547 printk(KERN_ALERT "RIP ");
548 printk_address(regs->ip, 1);
549 printk(" RSP <%016lx>\n", regs->sp);
550 if (kexec_should_crash(current))
551 crash_kexec(regs);
552 return 0;
553}
554
555void die(const char *str, struct pt_regs *regs, long err)
556{
557 unsigned long flags = oops_begin();
558
559 if (!user_mode(regs))
560 report_bug(regs->ip, regs);
561
562 if (__die(str, regs, err))
563 regs = NULL;
564 oops_end(flags, regs, SIGSEGV);
565}
566
567notrace __kprobes void
568die_nmi(char *str, struct pt_regs *regs, int do_panic)
569{
570 unsigned long flags;
571
572 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
573 return;
574
575 flags = oops_begin();
576 /*
577 * We are in trouble anyway, lets at least try
578 * to get a message out.
579 */
580 printk(KERN_EMERG "%s", str);
581 printk(" on CPU%d, ip %08lx, registers:\n",
582 smp_processor_id(), regs->ip);
583 show_registers(regs);
584 if (kexec_should_crash(current))
585 crash_kexec(regs);
586 if (do_panic || panic_on_oops)
587 panic("Non maskable interrupt");
588 oops_end(flags, NULL, SIGBUS);
589 nmi_exit();
590 local_irq_enable();
591 do_exit(SIGBUS);
592}
593
594static void __kprobes
595do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
596 long error_code, siginfo_t *info)
597{
598 struct task_struct *tsk = current;
599
600 if (!user_mode(regs))
601 goto kernel_trap;
602
603 /*
604 * We want error_code and trap_no set for userspace faults and
605 * kernelspace faults which result in die(), but not
606 * kernelspace faults which are fixed up. die() gives the
607 * process no chance to handle the signal and notice the
608 * kernel fault information, so that won't result in polluting
609 * the information about previously queued, but not yet
610 * delivered, faults. See also do_general_protection below.
611 */
612 tsk->thread.error_code = error_code;
613 tsk->thread.trap_no = trapnr;
614
615 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
616 printk_ratelimit()) {
617 printk(KERN_INFO
618 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
619 tsk->comm, tsk->pid, str,
620 regs->ip, regs->sp, error_code);
621 print_vma_addr(" in ", regs->ip);
622 printk("\n");
623 }
624
625 if (info)
626 force_sig_info(signr, info, tsk);
627 else
628 force_sig(signr, tsk);
629 return;
630
631kernel_trap:
632 if (!fixup_exception(regs)) {
633 tsk->thread.error_code = error_code;
634 tsk->thread.trap_no = trapnr;
635 die(str, regs, error_code);
636 }
637 return;
638}
639
640#define DO_ERROR(trapnr, signr, str, name) \
641asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
642{ \
643 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
644 == NOTIFY_STOP) \
645 return; \
646 conditional_sti(regs); \
647 do_trap(trapnr, signr, str, regs, error_code, NULL); \
648}
649
650#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
651asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
652{ \
653 siginfo_t info; \
654 info.si_signo = signr; \
655 info.si_errno = 0; \
656 info.si_code = sicode; \
657 info.si_addr = (void __user *)siaddr; \
658 trace_hardirqs_fixup(); \
659 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
660 == NOTIFY_STOP) \
661 return; \
662 conditional_sti(regs); \
663 do_trap(trapnr, signr, str, regs, error_code, &info); \
664}
665
666DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
667DO_ERROR(4, SIGSEGV, "overflow", overflow)
668DO_ERROR(5, SIGSEGV, "bounds", bounds)
669DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
670DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
671DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
672DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
673DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
674
675/* Runs on IST stack */
676asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
677{
678 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
679 12, SIGBUS) == NOTIFY_STOP)
680 return;
681 preempt_conditional_sti(regs);
682 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
683 preempt_conditional_cli(regs);
684}
685
686asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
687{
688 static const char str[] = "double fault";
689 struct task_struct *tsk = current;
690
691 /* Return not checked because double check cannot be ignored */
692 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
693
694 tsk->thread.error_code = error_code;
695 tsk->thread.trap_no = 8;
696
697 /* This is always a kernel trap and never fixable (and thus must
698 never return). */
699 for (;;)
700 die(str, regs, error_code);
701}
702
703asmlinkage void __kprobes
704do_general_protection(struct pt_regs *regs, long error_code)
705{
706 struct task_struct *tsk;
707
708 conditional_sti(regs);
709
710 tsk = current;
711 if (!user_mode(regs))
712 goto gp_in_kernel;
713
714 tsk->thread.error_code = error_code;
715 tsk->thread.trap_no = 13;
716
717 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
718 printk_ratelimit()) {
719 printk(KERN_INFO
720 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
721 tsk->comm, tsk->pid,
722 regs->ip, regs->sp, error_code);
723 print_vma_addr(" in ", regs->ip);
724 printk("\n");
725 }
726
727 force_sig(SIGSEGV, tsk);
728 return;
729
730gp_in_kernel:
731 if (fixup_exception(regs))
732 return;
733
734 tsk->thread.error_code = error_code;
735 tsk->thread.trap_no = 13;
736 if (notify_die(DIE_GPF, "general protection fault", regs,
737 error_code, 13, SIGSEGV) == NOTIFY_STOP)
738 return;
739 die("general protection fault", regs, error_code);
740}
741
742static notrace __kprobes void
743mem_parity_error(unsigned char reason, struct pt_regs *regs)
744{
745 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
746 reason);
747 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
748
749#if defined(CONFIG_EDAC)
750 if (edac_handler_set()) {
751 edac_atomic_assert_error();
752 return;
753 }
754#endif
755
756 if (panic_on_unrecovered_nmi)
757 panic("NMI: Not continuing");
758
759 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
760
761 /* Clear and disable the memory parity error line. */
762 reason = (reason & 0xf) | 4;
763 outb(reason, 0x61);
764}
765
766static notrace __kprobes void
767io_check_error(unsigned char reason, struct pt_regs *regs)
768{
769 printk("NMI: IOCK error (debug interrupt?)\n");
770 show_registers(regs);
771
772 /* Re-enable the IOCK line, wait for a few seconds */
773 reason = (reason & 0xf) | 8;
774 outb(reason, 0x61);
775 mdelay(2000);
776 reason &= ~8;
777 outb(reason, 0x61);
778}
779
780static notrace __kprobes void
781unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
782{
783 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
784 return;
785 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
786 reason);
787 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
788
789 if (panic_on_unrecovered_nmi)
790 panic("NMI: Not continuing");
791
792 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
793}
794
795/* Runs on IST stack. This code must keep interrupts off all the time.
796 Nested NMIs are prevented by the CPU. */
797asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
798{
799 unsigned char reason = 0;
800 int cpu;
801
802 cpu = smp_processor_id();
803
804 /* Only the BSP gets external NMIs from the system. */
805 if (!cpu)
806 reason = get_nmi_reason();
807
808 if (!(reason & 0xc0)) {
809 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
810 == NOTIFY_STOP)
811 return;
812 /*
813 * Ok, so this is none of the documented NMI sources,
814 * so it must be the NMI watchdog.
815 */
816 if (nmi_watchdog_tick(regs, reason))
817 return;
818 if (!do_nmi_callback(regs, cpu))
819 unknown_nmi_error(reason, regs);
820
821 return;
822 }
823 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
824 return;
825
826 /* AK: following checks seem to be broken on modern chipsets. FIXME */
827 if (reason & 0x80)
828 mem_parity_error(reason, regs);
829 if (reason & 0x40)
830 io_check_error(reason, regs);
831}
832
833asmlinkage notrace __kprobes void
834do_nmi(struct pt_regs *regs, long error_code)
835{
836 nmi_enter();
837
838 add_pda(__nmi_count, 1);
839
840 if (!ignore_nmis)
841 default_do_nmi(regs);
842
843 nmi_exit();
844}
845
846void stop_nmi(void)
847{
848 acpi_nmi_disable();
849 ignore_nmis++;
850}
851
852void restart_nmi(void)
853{
854 ignore_nmis--;
855 acpi_nmi_enable();
856}
857
858/* runs on IST stack. */
859asmlinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
860{
861 trace_hardirqs_fixup();
862
863 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
864 == NOTIFY_STOP)
865 return;
866
867 preempt_conditional_sti(regs);
868 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
869 preempt_conditional_cli(regs);
870}
871
872/* Help handler running on IST stack to switch back to user stack
873 for scheduling or signal handling. The actual stack switch is done in
874 entry.S */
875asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
876{
877 struct pt_regs *regs = eregs;
878 /* Did already sync */
879 if (eregs == (struct pt_regs *)eregs->sp)
880 ;
881 /* Exception from user space */
882 else if (user_mode(eregs))
883 regs = task_pt_regs(current);
884 /* Exception from kernel and interrupts are enabled. Move to
885 kernel process stack. */
886 else if (eregs->flags & X86_EFLAGS_IF)
887 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
888 if (eregs != regs)
889 *regs = *eregs;
890 return regs;
891}
892
893/* runs on IST stack. */
894asmlinkage void __kprobes do_debug(struct pt_regs * regs,
895 unsigned long error_code)
896{
897 struct task_struct *tsk = current;
898 unsigned long condition;
899 siginfo_t info;
900
901 trace_hardirqs_fixup();
902
903 get_debugreg(condition, 6);
904
905 /*
906 * The processor cleared BTF, so don't mark that we need it set.
907 */
908 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
909 tsk->thread.debugctlmsr = 0;
910
911 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
912 SIGTRAP) == NOTIFY_STOP)
913 return;
914
915 preempt_conditional_sti(regs);
916
917 /* Mask out spurious debug traps due to lazy DR7 setting */
918 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
919 if (!tsk->thread.debugreg7)
920 goto clear_dr7;
921 }
922
923 tsk->thread.debugreg6 = condition;
924
925 /*
926 * Single-stepping through TF: make sure we ignore any events in
927 * kernel space (but re-enable TF when returning to user mode).
928 */
929 if (condition & DR_STEP) {
930 if (!user_mode(regs))
931 goto clear_TF_reenable;
932 }
933
934 /* Ok, finally something we can handle */
935 tsk->thread.trap_no = 1;
936 tsk->thread.error_code = error_code;
937 info.si_signo = SIGTRAP;
938 info.si_errno = 0;
939 info.si_code = TRAP_BRKPT;
940 info.si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
941 force_sig_info(SIGTRAP, &info, tsk);
942
943clear_dr7:
944 set_debugreg(0, 7);
945 preempt_conditional_cli(regs);
946 return;
947
948clear_TF_reenable:
949 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
950 regs->flags &= ~X86_EFLAGS_TF;
951 preempt_conditional_cli(regs);
952 return;
953}
954
955static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
956{
957 if (fixup_exception(regs))
958 return 1;
959
960 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
961 /* Illegal floating point operation in the kernel */
962 current->thread.trap_no = trapnr;
963 die(str, regs, 0);
964 return 0;
965}
966
967/*
968 * Note that we play around with the 'TS' bit in an attempt to get
969 * the correct behaviour even in the presence of the asynchronous
970 * IRQ13 behaviour
971 */
972asmlinkage void do_coprocessor_error(struct pt_regs *regs)
973{
974 void __user *ip = (void __user *)(regs->ip);
975 struct task_struct *task;
976 siginfo_t info;
977 unsigned short cwd, swd;
978
979 conditional_sti(regs);
980 if (!user_mode(regs) &&
981 kernel_math_error(regs, "kernel x87 math error", 16))
982 return;
983
984 /*
985 * Save the info for the exception handler and clear the error.
986 */
987 task = current;
988 save_init_fpu(task);
989 task->thread.trap_no = 16;
990 task->thread.error_code = 0;
991 info.si_signo = SIGFPE;
992 info.si_errno = 0;
993 info.si_code = __SI_FAULT;
994 info.si_addr = ip;
995 /*
996 * (~cwd & swd) will mask out exceptions that are not set to unmasked
997 * status. 0x3f is the exception bits in these regs, 0x200 is the
998 * C1 reg you need in case of a stack fault, 0x040 is the stack
999 * fault bit. We should only be taking one exception at a time,
1000 * so if this combination doesn't produce any single exception,
1001 * then we have a bad program that isn't synchronizing its FPU usage
1002 * and it will suffer the consequences since we won't be able to
1003 * fully reproduce the context of the exception
1004 */
1005 cwd = get_fpu_cwd(task);
1006 swd = get_fpu_swd(task);
1007 switch (swd & ~cwd & 0x3f) {
1008 case 0x000: /* No unmasked exception */
1009 default: /* Multiple exceptions */
1010 break;
1011 case 0x001: /* Invalid Op */
1012 /*
1013 * swd & 0x240 == 0x040: Stack Underflow
1014 * swd & 0x240 == 0x240: Stack Overflow
1015 * User must clear the SF bit (0x40) if set
1016 */
1017 info.si_code = FPE_FLTINV;
1018 break;
1019 case 0x002: /* Denormalize */
1020 case 0x010: /* Underflow */
1021 info.si_code = FPE_FLTUND;
1022 break;
1023 case 0x004: /* Zero Divide */
1024 info.si_code = FPE_FLTDIV;
1025 break;
1026 case 0x008: /* Overflow */
1027 info.si_code = FPE_FLTOVF;
1028 break;
1029 case 0x020: /* Precision */
1030 info.si_code = FPE_FLTRES;
1031 break;
1032 }
1033 force_sig_info(SIGFPE, &info, task);
1034}
1035
1036asmlinkage void bad_intr(void)
1037{
1038 printk("bad interrupt");
1039}
1040
1041asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1042{
1043 void __user *ip = (void __user *)(regs->ip);
1044 struct task_struct *task;
1045 siginfo_t info;
1046 unsigned short mxcsr;
1047
1048 conditional_sti(regs);
1049 if (!user_mode(regs) &&
1050 kernel_math_error(regs, "kernel simd math error", 19))
1051 return;
1052
1053 /*
1054 * Save the info for the exception handler and clear the error.
1055 */
1056 task = current;
1057 save_init_fpu(task);
1058 task->thread.trap_no = 19;
1059 task->thread.error_code = 0;
1060 info.si_signo = SIGFPE;
1061 info.si_errno = 0;
1062 info.si_code = __SI_FAULT;
1063 info.si_addr = ip;
1064 /*
1065 * The SIMD FPU exceptions are handled a little differently, as there
1066 * is only a single status/control register. Thus, to determine which
1067 * unmasked exception was caught we must mask the exception mask bits
1068 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1069 */
1070 mxcsr = get_fpu_mxcsr(task);
1071 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1072 case 0x000:
1073 default:
1074 break;
1075 case 0x001: /* Invalid Op */
1076 info.si_code = FPE_FLTINV;
1077 break;
1078 case 0x002: /* Denormalize */
1079 case 0x010: /* Underflow */
1080 info.si_code = FPE_FLTUND;
1081 break;
1082 case 0x004: /* Zero Divide */
1083 info.si_code = FPE_FLTDIV;
1084 break;
1085 case 0x008: /* Overflow */
1086 info.si_code = FPE_FLTOVF;
1087 break;
1088 case 0x020: /* Precision */
1089 info.si_code = FPE_FLTRES;
1090 break;
1091 }
1092 force_sig_info(SIGFPE, &info, task);
1093}
1094
1095asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1096{
1097}
1098
1099asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1100{
1101}
1102
1103asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1104{
1105}
1106
1107/*
1108 * 'math_state_restore()' saves the current math information in the
1109 * old math state array, and gets the new ones from the current task
1110 *
1111 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1112 * Don't touch unless you *really* know how it works.
1113 */
1114asmlinkage void math_state_restore(void)
1115{
1116 struct task_struct *me = current;
1117
1118 if (!used_math()) {
1119 local_irq_enable();
1120 /*
1121 * does a slab alloc which can sleep
1122 */
1123 if (init_fpu(me)) {
1124 /*
1125 * ran out of memory!
1126 */
1127 do_group_exit(SIGKILL);
1128 return;
1129 }
1130 local_irq_disable();
1131 }
1132
1133 clts(); /* Allow maths ops (or we recurse) */
1134 /*
1135 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
1136 */
1137 if (unlikely(restore_fpu_checking(&me->thread.xstate->fxsave))) {
1138 stts();
1139 force_sig(SIGSEGV, me);
1140 return;
1141 }
1142 task_thread_info(me)->status |= TS_USEDFPU;
1143 me->fpu_counter++;
1144}
1145EXPORT_SYMBOL_GPL(math_state_restore);
1146
1147void __init trap_init(void)
1148{
1149 set_intr_gate(0, &divide_error);
1150 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1151 set_intr_gate_ist(2, &nmi, NMI_STACK);
1152 set_system_gate_ist(3, &int3, DEBUG_STACK); /* int3 can be called from all */
1153 set_system_gate(4, &overflow); /* int4 can be called from all */
1154 set_intr_gate(5, &bounds);
1155 set_intr_gate(6, &invalid_op);
1156 set_intr_gate(7, &device_not_available);
1157 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1158 set_intr_gate(9, &coprocessor_segment_overrun);
1159 set_intr_gate(10, &invalid_TSS);
1160 set_intr_gate(11, &segment_not_present);
1161 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1162 set_intr_gate(13, &general_protection);
1163 set_intr_gate(14, &page_fault);
1164 set_intr_gate(15, &spurious_interrupt_bug);
1165 set_intr_gate(16, &coprocessor_error);
1166 set_intr_gate(17, &alignment_check);
1167#ifdef CONFIG_X86_MCE
1168 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1169#endif
1170 set_intr_gate(19, &simd_coprocessor_error);
1171
1172#ifdef CONFIG_IA32_EMULATION
1173 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1174#endif
1175 /*
1176 * initialize the per thread extended state:
1177 */
1178 init_thread_xstate();
1179 /*
1180 * Should be a barrier for any external CPU state:
1181 */
1182 cpu_init();
1183}
1184
1185static int __init oops_setup(char *s)
1186{
1187 if (!s)
1188 return -EINVAL;
1189 if (!strcmp(s, "panic"))
1190 panic_on_oops = 1;
1191 return 0;
1192}
1193early_param("oops", oops_setup);
1194
1195static int __init kstack_setup(char *s)
1196{
1197 if (!s)
1198 return -EINVAL;
1199 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1200 return 0;
1201}
1202early_param("kstack", kstack_setup);
1203
1204static int __init code_bytes_setup(char *s)
1205{
1206 code_bytes = simple_strtoul(s, NULL, 0);
1207 if (code_bytes > 8192)
1208 code_bytes = 8192;
1209
1210 return 1;
1211}
1212__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 8f98e9de1b82..161bb850fc47 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -104,7 +104,7 @@ __setup("notsc", notsc_setup);
104/* 104/*
105 * Read TSC and the reference counters. Take care of SMI disturbance 105 * Read TSC and the reference counters. Take care of SMI disturbance
106 */ 106 */
107static u64 tsc_read_refs(u64 *pm, u64 *hpet) 107static u64 tsc_read_refs(u64 *p, int hpet)
108{ 108{
109 u64 t1, t2; 109 u64 t1, t2;
110 int i; 110 int i;
@@ -112,9 +112,9 @@ static u64 tsc_read_refs(u64 *pm, u64 *hpet)
112 for (i = 0; i < MAX_RETRIES; i++) { 112 for (i = 0; i < MAX_RETRIES; i++) {
113 t1 = get_cycles(); 113 t1 = get_cycles();
114 if (hpet) 114 if (hpet)
115 *hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF; 115 *p = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF;
116 else 116 else
117 *pm = acpi_pm_read_early(); 117 *p = acpi_pm_read_early();
118 t2 = get_cycles(); 118 t2 = get_cycles();
119 if ((t2 - t1) < SMI_TRESHOLD) 119 if ((t2 - t1) < SMI_TRESHOLD)
120 return t2; 120 return t2;
@@ -123,13 +123,59 @@ static u64 tsc_read_refs(u64 *pm, u64 *hpet)
123} 123}
124 124
125/* 125/*
126 * Calculate the TSC frequency from HPET reference
127 */
128static unsigned long calc_hpet_ref(u64 deltatsc, u64 hpet1, u64 hpet2)
129{
130 u64 tmp;
131
132 if (hpet2 < hpet1)
133 hpet2 += 0x100000000ULL;
134 hpet2 -= hpet1;
135 tmp = ((u64)hpet2 * hpet_readl(HPET_PERIOD));
136 do_div(tmp, 1000000);
137 do_div(deltatsc, tmp);
138
139 return (unsigned long) deltatsc;
140}
141
142/*
143 * Calculate the TSC frequency from PMTimer reference
144 */
145static unsigned long calc_pmtimer_ref(u64 deltatsc, u64 pm1, u64 pm2)
146{
147 u64 tmp;
148
149 if (!pm1 && !pm2)
150 return ULONG_MAX;
151
152 if (pm2 < pm1)
153 pm2 += (u64)ACPI_PM_OVRRUN;
154 pm2 -= pm1;
155 tmp = pm2 * 1000000000LL;
156 do_div(tmp, PMTMR_TICKS_PER_SEC);
157 do_div(deltatsc, tmp);
158
159 return (unsigned long) deltatsc;
160}
161
162#define CAL_MS 10
163#define CAL_LATCH (CLOCK_TICK_RATE / (1000 / CAL_MS))
164#define CAL_PIT_LOOPS 1000
165
166#define CAL2_MS 50
167#define CAL2_LATCH (CLOCK_TICK_RATE / (1000 / CAL2_MS))
168#define CAL2_PIT_LOOPS 5000
169
170
171/*
126 * Try to calibrate the TSC against the Programmable 172 * Try to calibrate the TSC against the Programmable
127 * Interrupt Timer and return the frequency of the TSC 173 * Interrupt Timer and return the frequency of the TSC
128 * in kHz. 174 * in kHz.
129 * 175 *
130 * Return ULONG_MAX on failure to calibrate. 176 * Return ULONG_MAX on failure to calibrate.
131 */ 177 */
132static unsigned long pit_calibrate_tsc(void) 178static unsigned long pit_calibrate_tsc(u32 latch, unsigned long ms, int loopmin)
133{ 179{
134 u64 tsc, t1, t2, delta; 180 u64 tsc, t1, t2, delta;
135 unsigned long tscmin, tscmax; 181 unsigned long tscmin, tscmax;
@@ -144,8 +190,8 @@ static unsigned long pit_calibrate_tsc(void)
144 * (LSB then MSB) to begin countdown. 190 * (LSB then MSB) to begin countdown.
145 */ 191 */
146 outb(0xb0, 0x43); 192 outb(0xb0, 0x43);
147 outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42); 193 outb(latch & 0xff, 0x42);
148 outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42); 194 outb(latch >> 8, 0x42);
149 195
150 tsc = t1 = t2 = get_cycles(); 196 tsc = t1 = t2 = get_cycles();
151 197
@@ -166,31 +212,154 @@ static unsigned long pit_calibrate_tsc(void)
166 /* 212 /*
167 * Sanity checks: 213 * Sanity checks:
168 * 214 *
169 * If we were not able to read the PIT more than 5000 215 * If we were not able to read the PIT more than loopmin
170 * times, then we have been hit by a massive SMI 216 * times, then we have been hit by a massive SMI
171 * 217 *
172 * If the maximum is 10 times larger than the minimum, 218 * If the maximum is 10 times larger than the minimum,
173 * then we got hit by an SMI as well. 219 * then we got hit by an SMI as well.
174 */ 220 */
175 if (pitcnt < 5000 || tscmax > 10 * tscmin) 221 if (pitcnt < loopmin || tscmax > 10 * tscmin)
176 return ULONG_MAX; 222 return ULONG_MAX;
177 223
178 /* Calculate the PIT value */ 224 /* Calculate the PIT value */
179 delta = t2 - t1; 225 delta = t2 - t1;
180 do_div(delta, 50); 226 do_div(delta, ms);
181 return delta; 227 return delta;
182} 228}
183 229
230/*
231 * This reads the current MSB of the PIT counter, and
232 * checks if we are running on sufficiently fast and
233 * non-virtualized hardware.
234 *
235 * Our expectations are:
236 *
237 * - the PIT is running at roughly 1.19MHz
238 *
239 * - each IO is going to take about 1us on real hardware,
240 * but we allow it to be much faster (by a factor of 10) or
241 * _slightly_ slower (ie we allow up to a 2us read+counter
242 * update - anything else implies a unacceptably slow CPU
243 * or PIT for the fast calibration to work.
244 *
245 * - with 256 PIT ticks to read the value, we have 214us to
246 * see the same MSB (and overhead like doing a single TSC
247 * read per MSB value etc).
248 *
249 * - We're doing 2 reads per loop (LSB, MSB), and we expect
250 * them each to take about a microsecond on real hardware.
251 * So we expect a count value of around 100. But we'll be
252 * generous, and accept anything over 50.
253 *
254 * - if the PIT is stuck, and we see *many* more reads, we
255 * return early (and the next caller of pit_expect_msb()
256 * then consider it a failure when they don't see the
257 * next expected value).
258 *
259 * These expectations mean that we know that we have seen the
260 * transition from one expected value to another with a fairly
261 * high accuracy, and we didn't miss any events. We can thus
262 * use the TSC value at the transitions to calculate a pretty
263 * good value for the TSC frequencty.
264 */
265static inline int pit_expect_msb(unsigned char val)
266{
267 int count = 0;
268
269 for (count = 0; count < 50000; count++) {
270 /* Ignore LSB */
271 inb(0x42);
272 if (inb(0x42) != val)
273 break;
274 }
275 return count > 50;
276}
277
278/*
279 * How many MSB values do we want to see? We aim for a
280 * 15ms calibration, which assuming a 2us counter read
281 * error should give us roughly 150 ppm precision for
282 * the calibration.
283 */
284#define QUICK_PIT_MS 15
285#define QUICK_PIT_ITERATIONS (QUICK_PIT_MS * PIT_TICK_RATE / 1000 / 256)
286
287static unsigned long quick_pit_calibrate(void)
288{
289 /* Set the Gate high, disable speaker */
290 outb((inb(0x61) & ~0x02) | 0x01, 0x61);
291
292 /*
293 * Counter 2, mode 0 (one-shot), binary count
294 *
295 * NOTE! Mode 2 decrements by two (and then the
296 * output is flipped each time, giving the same
297 * final output frequency as a decrement-by-one),
298 * so mode 0 is much better when looking at the
299 * individual counts.
300 */
301 outb(0xb0, 0x43);
302
303 /* Start at 0xffff */
304 outb(0xff, 0x42);
305 outb(0xff, 0x42);
306
307 if (pit_expect_msb(0xff)) {
308 int i;
309 u64 t1, t2, delta;
310 unsigned char expect = 0xfe;
311
312 t1 = get_cycles();
313 for (i = 0; i < QUICK_PIT_ITERATIONS; i++, expect--) {
314 if (!pit_expect_msb(expect))
315 goto failed;
316 }
317 t2 = get_cycles();
318
319 /*
320 * Make sure we can rely on the second TSC timestamp:
321 */
322 if (!pit_expect_msb(expect))
323 goto failed;
324
325 /*
326 * Ok, if we get here, then we've seen the
327 * MSB of the PIT decrement QUICK_PIT_ITERATIONS
328 * times, and each MSB had many hits, so we never
329 * had any sudden jumps.
330 *
331 * As a result, we can depend on there not being
332 * any odd delays anywhere, and the TSC reads are
333 * reliable.
334 *
335 * kHz = ticks / time-in-seconds / 1000;
336 * kHz = (t2 - t1) / (QPI * 256 / PIT_TICK_RATE) / 1000
337 * kHz = ((t2 - t1) * PIT_TICK_RATE) / (QPI * 256 * 1000)
338 */
339 delta = (t2 - t1)*PIT_TICK_RATE;
340 do_div(delta, QUICK_PIT_ITERATIONS*256*1000);
341 printk("Fast TSC calibration using PIT\n");
342 return delta;
343 }
344failed:
345 return 0;
346}
184 347
185/** 348/**
186 * native_calibrate_tsc - calibrate the tsc on boot 349 * native_calibrate_tsc - calibrate the tsc on boot
187 */ 350 */
188unsigned long native_calibrate_tsc(void) 351unsigned long native_calibrate_tsc(void)
189{ 352{
190 u64 tsc1, tsc2, delta, pm1, pm2, hpet1, hpet2; 353 u64 tsc1, tsc2, delta, ref1, ref2;
191 unsigned long tsc_pit_min = ULONG_MAX, tsc_ref_min = ULONG_MAX; 354 unsigned long tsc_pit_min = ULONG_MAX, tsc_ref_min = ULONG_MAX;
192 unsigned long flags; 355 unsigned long flags, latch, ms, fast_calibrate;
193 int hpet = is_hpet_enabled(), i; 356 int hpet = is_hpet_enabled(), i, loopmin;
357
358 local_irq_save(flags);
359 fast_calibrate = quick_pit_calibrate();
360 local_irq_restore(flags);
361 if (fast_calibrate)
362 return fast_calibrate;
194 363
195 /* 364 /*
196 * Run 5 calibration loops to get the lowest frequency value 365 * Run 5 calibration loops to get the lowest frequency value
@@ -216,7 +385,13 @@ unsigned long native_calibrate_tsc(void)
216 * calibration delay loop as we have to wait for a certain 385 * calibration delay loop as we have to wait for a certain
217 * amount of time anyway. 386 * amount of time anyway.
218 */ 387 */
219 for (i = 0; i < 5; i++) { 388
389 /* Preset PIT loop values */
390 latch = CAL_LATCH;
391 ms = CAL_MS;
392 loopmin = CAL_PIT_LOOPS;
393
394 for (i = 0; i < 3; i++) {
220 unsigned long tsc_pit_khz; 395 unsigned long tsc_pit_khz;
221 396
222 /* 397 /*
@@ -226,16 +401,16 @@ unsigned long native_calibrate_tsc(void)
226 * read the end value. 401 * read the end value.
227 */ 402 */
228 local_irq_save(flags); 403 local_irq_save(flags);
229 tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL); 404 tsc1 = tsc_read_refs(&ref1, hpet);
230 tsc_pit_khz = pit_calibrate_tsc(); 405 tsc_pit_khz = pit_calibrate_tsc(latch, ms, loopmin);
231 tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL); 406 tsc2 = tsc_read_refs(&ref2, hpet);
232 local_irq_restore(flags); 407 local_irq_restore(flags);
233 408
234 /* Pick the lowest PIT TSC calibration so far */ 409 /* Pick the lowest PIT TSC calibration so far */
235 tsc_pit_min = min(tsc_pit_min, tsc_pit_khz); 410 tsc_pit_min = min(tsc_pit_min, tsc_pit_khz);
236 411
237 /* hpet or pmtimer available ? */ 412 /* hpet or pmtimer available ? */
238 if (!hpet && !pm1 && !pm2) 413 if (!hpet && !ref1 && !ref2)
239 continue; 414 continue;
240 415
241 /* Check, whether the sampling was disturbed by an SMI */ 416 /* Check, whether the sampling was disturbed by an SMI */
@@ -243,23 +418,41 @@ unsigned long native_calibrate_tsc(void)
243 continue; 418 continue;
244 419
245 tsc2 = (tsc2 - tsc1) * 1000000LL; 420 tsc2 = (tsc2 - tsc1) * 1000000LL;
421 if (hpet)
422 tsc2 = calc_hpet_ref(tsc2, ref1, ref2);
423 else
424 tsc2 = calc_pmtimer_ref(tsc2, ref1, ref2);
246 425
247 if (hpet) { 426 tsc_ref_min = min(tsc_ref_min, (unsigned long) tsc2);
248 if (hpet2 < hpet1) 427
249 hpet2 += 0x100000000ULL; 428 /* Check the reference deviation */
250 hpet2 -= hpet1; 429 delta = ((u64) tsc_pit_min) * 100;
251 tsc1 = ((u64)hpet2 * hpet_readl(HPET_PERIOD)); 430 do_div(delta, tsc_ref_min);
252 do_div(tsc1, 1000000); 431
253 } else { 432 /*
254 if (pm2 < pm1) 433 * If both calibration results are inside a 10% window
255 pm2 += (u64)ACPI_PM_OVRRUN; 434 * then we can be sure, that the calibration
256 pm2 -= pm1; 435 * succeeded. We break out of the loop right away. We
257 tsc1 = pm2 * 1000000000LL; 436 * use the reference value, as it is more precise.
258 do_div(tsc1, PMTMR_TICKS_PER_SEC); 437 */
438 if (delta >= 90 && delta <= 110) {
439 printk(KERN_INFO
440 "TSC: PIT calibration matches %s. %d loops\n",
441 hpet ? "HPET" : "PMTIMER", i + 1);
442 return tsc_ref_min;
259 } 443 }
260 444
261 do_div(tsc2, tsc1); 445 /*
262 tsc_ref_min = min(tsc_ref_min, (unsigned long) tsc2); 446 * Check whether PIT failed more than once. This
447 * happens in virtualized environments. We need to
448 * give the virtual PC a slightly longer timeframe for
449 * the HPET/PMTIMER to make the result precise.
450 */
451 if (i == 1 && tsc_pit_min == ULONG_MAX) {
452 latch = CAL2_LATCH;
453 ms = CAL2_MS;
454 loopmin = CAL2_PIT_LOOPS;
455 }
263 } 456 }
264 457
265 /* 458 /*
@@ -270,7 +463,7 @@ unsigned long native_calibrate_tsc(void)
270 printk(KERN_WARNING "TSC: Unable to calibrate against PIT\n"); 463 printk(KERN_WARNING "TSC: Unable to calibrate against PIT\n");
271 464
272 /* We don't have an alternative source, disable TSC */ 465 /* We don't have an alternative source, disable TSC */
273 if (!hpet && !pm1 && !pm2) { 466 if (!hpet && !ref1 && !ref2) {
274 printk("TSC: No reference (HPET/PMTIMER) available\n"); 467 printk("TSC: No reference (HPET/PMTIMER) available\n");
275 return 0; 468 return 0;
276 } 469 }
@@ -278,7 +471,7 @@ unsigned long native_calibrate_tsc(void)
278 /* The alternative source failed as well, disable TSC */ 471 /* The alternative source failed as well, disable TSC */
279 if (tsc_ref_min == ULONG_MAX) { 472 if (tsc_ref_min == ULONG_MAX) {
280 printk(KERN_WARNING "TSC: HPET/PMTIMER calibration " 473 printk(KERN_WARNING "TSC: HPET/PMTIMER calibration "
281 "failed due to SMI disturbance.\n"); 474 "failed.\n");
282 return 0; 475 return 0;
283 } 476 }
284 477
@@ -290,44 +483,25 @@ unsigned long native_calibrate_tsc(void)
290 } 483 }
291 484
292 /* We don't have an alternative source, use the PIT calibration value */ 485 /* We don't have an alternative source, use the PIT calibration value */
293 if (!hpet && !pm1 && !pm2) { 486 if (!hpet && !ref1 && !ref2) {
294 printk(KERN_INFO "TSC: Using PIT calibration value\n"); 487 printk(KERN_INFO "TSC: Using PIT calibration value\n");
295 return tsc_pit_min; 488 return tsc_pit_min;
296 } 489 }
297 490
298 /* The alternative source failed, use the PIT calibration value */ 491 /* The alternative source failed, use the PIT calibration value */
299 if (tsc_ref_min == ULONG_MAX) { 492 if (tsc_ref_min == ULONG_MAX) {
300 printk(KERN_WARNING "TSC: HPET/PMTIMER calibration failed due " 493 printk(KERN_WARNING "TSC: HPET/PMTIMER calibration failed. "
301 "to SMI disturbance. Using PIT calibration\n"); 494 "Using PIT calibration\n");
302 return tsc_pit_min; 495 return tsc_pit_min;
303 } 496 }
304 497
305 /* Check the reference deviation */
306 delta = ((u64) tsc_pit_min) * 100;
307 do_div(delta, tsc_ref_min);
308
309 /*
310 * If both calibration results are inside a 5% window, the we
311 * use the lower frequency of those as it is probably the
312 * closest estimate.
313 */
314 if (delta >= 95 && delta <= 105) {
315 printk(KERN_INFO "TSC: PIT calibration confirmed by %s.\n",
316 hpet ? "HPET" : "PMTIMER");
317 printk(KERN_INFO "TSC: using %s calibration value\n",
318 tsc_pit_min <= tsc_ref_min ? "PIT" :
319 hpet ? "HPET" : "PMTIMER");
320 return tsc_pit_min <= tsc_ref_min ? tsc_pit_min : tsc_ref_min;
321 }
322
323 printk(KERN_WARNING "TSC: PIT calibration deviates from %s: %lu %lu.\n",
324 hpet ? "HPET" : "PMTIMER", tsc_pit_min, tsc_ref_min);
325
326 /* 498 /*
327 * The calibration values differ too much. In doubt, we use 499 * The calibration values differ too much. In doubt, we use
328 * the PIT value as we know that there are PMTIMERs around 500 * the PIT value as we know that there are PMTIMERs around
329 * running at double speed. 501 * running at double speed. At least we let the user know:
330 */ 502 */
503 printk(KERN_WARNING "TSC: PIT calibration deviates from %s: %lu %lu.\n",
504 hpet ? "HPET" : "PMTIMER", tsc_pit_min, tsc_ref_min);
331 printk(KERN_INFO "TSC: Using PIT calibration value\n"); 505 printk(KERN_INFO "TSC: Using PIT calibration value\n");
332 return tsc_pit_min; 506 return tsc_pit_min;
333} 507}
diff --git a/arch/x86/kernel/visws_quirks.c b/arch/x86/kernel/visws_quirks.c
index 594ef47f0a63..61a97e616f70 100644
--- a/arch/x86/kernel/visws_quirks.c
+++ b/arch/x86/kernel/visws_quirks.c
@@ -25,45 +25,31 @@
25#include <asm/visws/cobalt.h> 25#include <asm/visws/cobalt.h>
26#include <asm/visws/piix4.h> 26#include <asm/visws/piix4.h>
27#include <asm/arch_hooks.h> 27#include <asm/arch_hooks.h>
28#include <asm/io_apic.h>
28#include <asm/fixmap.h> 29#include <asm/fixmap.h>
29#include <asm/reboot.h> 30#include <asm/reboot.h>
30#include <asm/setup.h> 31#include <asm/setup.h>
31#include <asm/e820.h> 32#include <asm/e820.h>
32#include <asm/smp.h>
33#include <asm/io.h> 33#include <asm/io.h>
34 34
35#include <mach_ipi.h> 35#include <mach_ipi.h>
36 36
37#include "mach_apic.h" 37#include "mach_apic.h"
38 38
39#include <linux/init.h>
40#include <linux/smp.h>
41
42#include <linux/kernel_stat.h> 39#include <linux/kernel_stat.h>
43#include <linux/interrupt.h>
44#include <linux/init.h>
45 40
46#include <asm/io.h>
47#include <asm/apic.h>
48#include <asm/i8259.h> 41#include <asm/i8259.h>
49#include <asm/irq_vectors.h> 42#include <asm/irq_vectors.h>
50#include <asm/visws/cobalt.h>
51#include <asm/visws/lithium.h> 43#include <asm/visws/lithium.h>
52#include <asm/visws/piix4.h>
53 44
54#include <linux/sched.h> 45#include <linux/sched.h>
55#include <linux/kernel.h> 46#include <linux/kernel.h>
56#include <linux/init.h>
57#include <linux/pci.h> 47#include <linux/pci.h>
58#include <linux/pci_ids.h> 48#include <linux/pci_ids.h>
59 49
60extern int no_broadcast; 50extern int no_broadcast;
61 51
62#include <asm/io.h>
63#include <asm/apic.h> 52#include <asm/apic.h>
64#include <asm/arch_hooks.h>
65#include <asm/visws/cobalt.h>
66#include <asm/visws/lithium.h>
67 53
68char visws_board_type = -1; 54char visws_board_type = -1;
69char visws_board_rev = -1; 55char visws_board_rev = -1;
diff --git a/arch/x86/kernel/vm86_32.c b/arch/x86/kernel/vm86_32.c
index 38f566fa27d2..4eeb5cf9720d 100644
--- a/arch/x86/kernel/vm86_32.c
+++ b/arch/x86/kernel/vm86_32.c
@@ -46,6 +46,7 @@
46#include <asm/io.h> 46#include <asm/io.h>
47#include <asm/tlbflush.h> 47#include <asm/tlbflush.h>
48#include <asm/irq.h> 48#include <asm/irq.h>
49#include <asm/syscalls.h>
49 50
50/* 51/*
51 * Known problems: 52 * Known problems:
diff --git a/arch/x86/kernel/vmi_32.c b/arch/x86/kernel/vmi_32.c
index 6ca515d6db54..8b6c393ab9fd 100644
--- a/arch/x86/kernel/vmi_32.c
+++ b/arch/x86/kernel/vmi_32.c
@@ -235,7 +235,7 @@ static void vmi_write_ldt_entry(struct desc_struct *dt, int entry,
235 const void *desc) 235 const void *desc)
236{ 236{
237 u32 *ldt_entry = (u32 *)desc; 237 u32 *ldt_entry = (u32 *)desc;
238 vmi_ops.write_idt_entry(dt, entry, ldt_entry[0], ldt_entry[1]); 238 vmi_ops.write_ldt_entry(dt, entry, ldt_entry[0], ldt_entry[1]);
239} 239}
240 240
241static void vmi_load_sp0(struct tss_struct *tss, 241static void vmi_load_sp0(struct tss_struct *tss,
@@ -393,13 +393,13 @@ static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
393} 393}
394#endif 394#endif
395 395
396static void vmi_allocate_pte(struct mm_struct *mm, u32 pfn) 396static void vmi_allocate_pte(struct mm_struct *mm, unsigned long pfn)
397{ 397{
398 vmi_set_page_type(pfn, VMI_PAGE_L1); 398 vmi_set_page_type(pfn, VMI_PAGE_L1);
399 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0); 399 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
400} 400}
401 401
402static void vmi_allocate_pmd(struct mm_struct *mm, u32 pfn) 402static void vmi_allocate_pmd(struct mm_struct *mm, unsigned long pfn)
403{ 403{
404 /* 404 /*
405 * This call comes in very early, before mem_map is setup. 405 * This call comes in very early, before mem_map is setup.
@@ -410,20 +410,20 @@ static void vmi_allocate_pmd(struct mm_struct *mm, u32 pfn)
410 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0); 410 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
411} 411}
412 412
413static void vmi_allocate_pmd_clone(u32 pfn, u32 clonepfn, u32 start, u32 count) 413static void vmi_allocate_pmd_clone(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count)
414{ 414{
415 vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE); 415 vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE);
416 vmi_check_page_type(clonepfn, VMI_PAGE_L2); 416 vmi_check_page_type(clonepfn, VMI_PAGE_L2);
417 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count); 417 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
418} 418}
419 419
420static void vmi_release_pte(u32 pfn) 420static void vmi_release_pte(unsigned long pfn)
421{ 421{
422 vmi_ops.release_page(pfn, VMI_PAGE_L1); 422 vmi_ops.release_page(pfn, VMI_PAGE_L1);
423 vmi_set_page_type(pfn, VMI_PAGE_NORMAL); 423 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
424} 424}
425 425
426static void vmi_release_pmd(u32 pfn) 426static void vmi_release_pmd(unsigned long pfn)
427{ 427{
428 vmi_ops.release_page(pfn, VMI_PAGE_L2); 428 vmi_ops.release_page(pfn, VMI_PAGE_L2);
429 vmi_set_page_type(pfn, VMI_PAGE_NORMAL); 429 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
@@ -905,8 +905,8 @@ static inline int __init activate_vmi(void)
905#endif 905#endif
906 906
907#ifdef CONFIG_X86_LOCAL_APIC 907#ifdef CONFIG_X86_LOCAL_APIC
908 para_fill(pv_apic_ops.apic_read, APICRead); 908 para_fill(apic_ops->read, APICRead);
909 para_fill(pv_apic_ops.apic_write, APICWrite); 909 para_fill(apic_ops->write, APICWrite);
910#endif 910#endif
911 911
912 /* 912 /*
diff --git a/arch/x86/kernel/vmlinux_32.lds.S b/arch/x86/kernel/vmlinux_32.lds.S
index af5bdad84604..a9b8560adbc2 100644
--- a/arch/x86/kernel/vmlinux_32.lds.S
+++ b/arch/x86/kernel/vmlinux_32.lds.S
@@ -140,10 +140,10 @@ SECTIONS
140 *(.con_initcall.init) 140 *(.con_initcall.init)
141 __con_initcall_end = .; 141 __con_initcall_end = .;
142 } 142 }
143 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 143 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
144 __x86cpuvendor_start = .; 144 __x86_cpu_dev_start = .;
145 *(.x86cpuvendor.init) 145 *(.x86_cpu_dev.init)
146 __x86cpuvendor_end = .; 146 __x86_cpu_dev_end = .;
147 } 147 }
148 SECURITY_INIT 148 SECURITY_INIT
149 . = ALIGN(4); 149 . = ALIGN(4);
@@ -180,6 +180,7 @@ SECTIONS
180 . = ALIGN(PAGE_SIZE); 180 . = ALIGN(PAGE_SIZE);
181 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) { 181 .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) {
182 __per_cpu_start = .; 182 __per_cpu_start = .;
183 *(.data.percpu.page_aligned)
183 *(.data.percpu) 184 *(.data.percpu)
184 *(.data.percpu.shared_aligned) 185 *(.data.percpu.shared_aligned)
185 __per_cpu_end = .; 186 __per_cpu_end = .;
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index 63e5c1a22e88..46e05447405b 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -168,12 +168,11 @@ SECTIONS
168 *(.con_initcall.init) 168 *(.con_initcall.init)
169 } 169 }
170 __con_initcall_end = .; 170 __con_initcall_end = .;
171 . = ALIGN(16); 171 __x86_cpu_dev_start = .;
172 __x86cpuvendor_start = .; 172 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
173 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 173 *(.x86_cpu_dev.init)
174 *(.x86cpuvendor.init)
175 } 174 }
176 __x86cpuvendor_end = .; 175 __x86_cpu_dev_end = .;
177 SECURITY_INIT 176 SECURITY_INIT
178 177
179 . = ALIGN(8); 178 . = ALIGN(8);
diff --git a/arch/x86/kernel/vsmp_64.c b/arch/x86/kernel/vsmp_64.c
index 0c029e8959c7..7766d36983fc 100644
--- a/arch/x86/kernel/vsmp_64.c
+++ b/arch/x86/kernel/vsmp_64.c
@@ -61,7 +61,7 @@ static void vsmp_irq_enable(void)
61 native_restore_fl((flags | X86_EFLAGS_IF) & (~X86_EFLAGS_AC)); 61 native_restore_fl((flags | X86_EFLAGS_IF) & (~X86_EFLAGS_AC));
62} 62}
63 63
64static unsigned __init vsmp_patch(u8 type, u16 clobbers, void *ibuf, 64static unsigned __init_or_module vsmp_patch(u8 type, u16 clobbers, void *ibuf,
65 unsigned long addr, unsigned len) 65 unsigned long addr, unsigned len)
66{ 66{
67 switch (type) { 67 switch (type) {
diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c
new file mode 100644
index 000000000000..9abac8a9d823
--- /dev/null
+++ b/arch/x86/kernel/xsave.c
@@ -0,0 +1,345 @@
1/*
2 * xsave/xrstor support.
3 *
4 * Author: Suresh Siddha <suresh.b.siddha@intel.com>
5 */
6#include <linux/bootmem.h>
7#include <linux/compat.h>
8#include <asm/i387.h>
9#ifdef CONFIG_IA32_EMULATION
10#include <asm/sigcontext32.h>
11#endif
12#include <asm/xcr.h>
13
14/*
15 * Supported feature mask by the CPU and the kernel.
16 */
17u64 pcntxt_mask;
18
19struct _fpx_sw_bytes fx_sw_reserved;
20#ifdef CONFIG_IA32_EMULATION
21struct _fpx_sw_bytes fx_sw_reserved_ia32;
22#endif
23
24/*
25 * Check for the presence of extended state information in the
26 * user fpstate pointer in the sigcontext.
27 */
28int check_for_xstate(struct i387_fxsave_struct __user *buf,
29 void __user *fpstate,
30 struct _fpx_sw_bytes *fx_sw_user)
31{
32 int min_xstate_size = sizeof(struct i387_fxsave_struct) +
33 sizeof(struct xsave_hdr_struct);
34 unsigned int magic2;
35 int err;
36
37 err = __copy_from_user(fx_sw_user, &buf->sw_reserved[0],
38 sizeof(struct _fpx_sw_bytes));
39
40 if (err)
41 return err;
42
43 /*
44 * First Magic check failed.
45 */
46 if (fx_sw_user->magic1 != FP_XSTATE_MAGIC1)
47 return -1;
48
49 /*
50 * Check for error scenarios.
51 */
52 if (fx_sw_user->xstate_size < min_xstate_size ||
53 fx_sw_user->xstate_size > xstate_size ||
54 fx_sw_user->xstate_size > fx_sw_user->extended_size)
55 return -1;
56
57 err = __get_user(magic2, (__u32 *) (((void *)fpstate) +
58 fx_sw_user->extended_size -
59 FP_XSTATE_MAGIC2_SIZE));
60 /*
61 * Check for the presence of second magic word at the end of memory
62 * layout. This detects the case where the user just copied the legacy
63 * fpstate layout with out copying the extended state information
64 * in the memory layout.
65 */
66 if (err || magic2 != FP_XSTATE_MAGIC2)
67 return -1;
68
69 return 0;
70}
71
72#ifdef CONFIG_X86_64
73/*
74 * Signal frame handlers.
75 */
76
77int save_i387_xstate(void __user *buf)
78{
79 struct task_struct *tsk = current;
80 int err = 0;
81
82 if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size))
83 return -EACCES;
84
85 BUG_ON(sig_xstate_size < xstate_size);
86
87 if ((unsigned long)buf % 64)
88 printk("save_i387_xstate: bad fpstate %p\n", buf);
89
90 if (!used_math())
91 return 0;
92 clear_used_math(); /* trigger finit */
93 if (task_thread_info(tsk)->status & TS_USEDFPU) {
94 /*
95 * Start with clearing the user buffer. This will present a
96 * clean context for the bytes not touched by the fxsave/xsave.
97 */
98 err = __clear_user(buf, sig_xstate_size);
99 if (err)
100 return err;
101
102 if (task_thread_info(tsk)->status & TS_XSAVE)
103 err = xsave_user(buf);
104 else
105 err = fxsave_user(buf);
106
107 if (err)
108 return err;
109 task_thread_info(tsk)->status &= ~TS_USEDFPU;
110 stts();
111 } else {
112 if (__copy_to_user(buf, &tsk->thread.xstate->fxsave,
113 xstate_size))
114 return -1;
115 }
116
117 if (task_thread_info(tsk)->status & TS_XSAVE) {
118 struct _fpstate __user *fx = buf;
119 struct _xstate __user *x = buf;
120 u64 xstate_bv;
121
122 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved,
123 sizeof(struct _fpx_sw_bytes));
124
125 err |= __put_user(FP_XSTATE_MAGIC2,
126 (__u32 __user *) (buf + sig_xstate_size
127 - FP_XSTATE_MAGIC2_SIZE));
128
129 /*
130 * Read the xstate_bv which we copied (directly from the cpu or
131 * from the state in task struct) to the user buffers and
132 * set the FP/SSE bits.
133 */
134 err |= __get_user(xstate_bv, &x->xstate_hdr.xstate_bv);
135
136 /*
137 * For legacy compatible, we always set FP/SSE bits in the bit
138 * vector while saving the state to the user context. This will
139 * enable us capturing any changes(during sigreturn) to
140 * the FP/SSE bits by the legacy applications which don't touch
141 * xstate_bv in the xsave header.
142 *
143 * xsave aware apps can change the xstate_bv in the xsave
144 * header as well as change any contents in the memory layout.
145 * xrestore as part of sigreturn will capture all the changes.
146 */
147 xstate_bv |= XSTATE_FPSSE;
148
149 err |= __put_user(xstate_bv, &x->xstate_hdr.xstate_bv);
150
151 if (err)
152 return err;
153 }
154
155 return 1;
156}
157
158/*
159 * Restore the extended state if present. Otherwise, restore the FP/SSE
160 * state.
161 */
162int restore_user_xstate(void __user *buf)
163{
164 struct _fpx_sw_bytes fx_sw_user;
165 u64 mask;
166 int err;
167
168 if (((unsigned long)buf % 64) ||
169 check_for_xstate(buf, buf, &fx_sw_user))
170 goto fx_only;
171
172 mask = fx_sw_user.xstate_bv;
173
174 /*
175 * restore the state passed by the user.
176 */
177 err = xrestore_user(buf, mask);
178 if (err)
179 return err;
180
181 /*
182 * init the state skipped by the user.
183 */
184 mask = pcntxt_mask & ~mask;
185
186 xrstor_state(init_xstate_buf, mask);
187
188 return 0;
189
190fx_only:
191 /*
192 * couldn't find the extended state information in the
193 * memory layout. Restore just the FP/SSE and init all
194 * the other extended state.
195 */
196 xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE);
197 return fxrstor_checking((__force struct i387_fxsave_struct *)buf);
198}
199
200/*
201 * This restores directly out of user space. Exceptions are handled.
202 */
203int restore_i387_xstate(void __user *buf)
204{
205 struct task_struct *tsk = current;
206 int err = 0;
207
208 if (!buf) {
209 if (used_math())
210 goto clear;
211 return 0;
212 } else
213 if (!access_ok(VERIFY_READ, buf, sig_xstate_size))
214 return -EACCES;
215
216 if (!used_math()) {
217 err = init_fpu(tsk);
218 if (err)
219 return err;
220 }
221
222 if (!(task_thread_info(current)->status & TS_USEDFPU)) {
223 clts();
224 task_thread_info(current)->status |= TS_USEDFPU;
225 }
226 if (task_thread_info(tsk)->status & TS_XSAVE)
227 err = restore_user_xstate(buf);
228 else
229 err = fxrstor_checking((__force struct i387_fxsave_struct *)
230 buf);
231 if (unlikely(err)) {
232 /*
233 * Encountered an error while doing the restore from the
234 * user buffer, clear the fpu state.
235 */
236clear:
237 clear_fpu(tsk);
238 clear_used_math();
239 }
240 return err;
241}
242#endif
243
244/*
245 * Prepare the SW reserved portion of the fxsave memory layout, indicating
246 * the presence of the extended state information in the memory layout
247 * pointed by the fpstate pointer in the sigcontext.
248 * This will be saved when ever the FP and extended state context is
249 * saved on the user stack during the signal handler delivery to the user.
250 */
251void prepare_fx_sw_frame(void)
252{
253 int size_extended = (xstate_size - sizeof(struct i387_fxsave_struct)) +
254 FP_XSTATE_MAGIC2_SIZE;
255
256 sig_xstate_size = sizeof(struct _fpstate) + size_extended;
257
258#ifdef CONFIG_IA32_EMULATION
259 sig_xstate_ia32_size = sizeof(struct _fpstate_ia32) + size_extended;
260#endif
261
262 memset(&fx_sw_reserved, 0, sizeof(fx_sw_reserved));
263
264 fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
265 fx_sw_reserved.extended_size = sig_xstate_size;
266 fx_sw_reserved.xstate_bv = pcntxt_mask;
267 fx_sw_reserved.xstate_size = xstate_size;
268#ifdef CONFIG_IA32_EMULATION
269 memcpy(&fx_sw_reserved_ia32, &fx_sw_reserved,
270 sizeof(struct _fpx_sw_bytes));
271 fx_sw_reserved_ia32.extended_size = sig_xstate_ia32_size;
272#endif
273}
274
275/*
276 * Represents init state for the supported extended state.
277 */
278struct xsave_struct *init_xstate_buf;
279
280#ifdef CONFIG_X86_64
281unsigned int sig_xstate_size = sizeof(struct _fpstate);
282#endif
283
284/*
285 * Enable the extended processor state save/restore feature
286 */
287void __cpuinit xsave_init(void)
288{
289 if (!cpu_has_xsave)
290 return;
291
292 set_in_cr4(X86_CR4_OSXSAVE);
293
294 /*
295 * Enable all the features that the HW is capable of
296 * and the Linux kernel is aware of.
297 */
298 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
299}
300
301/*
302 * setup the xstate image representing the init state
303 */
304static void __init setup_xstate_init(void)
305{
306 init_xstate_buf = alloc_bootmem(xstate_size);
307 init_xstate_buf->i387.mxcsr = MXCSR_DEFAULT;
308}
309
310/*
311 * Enable and initialize the xsave feature.
312 */
313void __init xsave_cntxt_init(void)
314{
315 unsigned int eax, ebx, ecx, edx;
316
317 cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
318 pcntxt_mask = eax + ((u64)edx << 32);
319
320 if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
321 printk(KERN_ERR "FP/SSE not shown under xsave features 0x%llx\n",
322 pcntxt_mask);
323 BUG();
324 }
325
326 /*
327 * for now OS knows only about FP/SSE
328 */
329 pcntxt_mask = pcntxt_mask & XCNTXT_MASK;
330 xsave_init();
331
332 /*
333 * Recompute the context size for enabled features
334 */
335 cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
336 xstate_size = ebx;
337
338 prepare_fx_sw_frame();
339
340 setup_xstate_init();
341
342 printk(KERN_INFO "xsave/xrstor: enabled xstate_bv 0x%llx, "
343 "cntxt size 0x%x\n",
344 pcntxt_mask, xstate_size);
345}
diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
index d0e940bb6f40..c02343594b4d 100644
--- a/arch/x86/kvm/Makefile
+++ b/arch/x86/kvm/Makefile
@@ -3,10 +3,13 @@
3# 3#
4 4
5common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \ 5common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \
6 coalesced_mmio.o) 6 coalesced_mmio.o irq_comm.o)
7ifeq ($(CONFIG_KVM_TRACE),y) 7ifeq ($(CONFIG_KVM_TRACE),y)
8common-objs += $(addprefix ../../../virt/kvm/, kvm_trace.o) 8common-objs += $(addprefix ../../../virt/kvm/, kvm_trace.o)
9endif 9endif
10ifeq ($(CONFIG_DMAR),y)
11common-objs += $(addprefix ../../../virt/kvm/, vtd.o)
12endif
10 13
11EXTRA_CFLAGS += -Ivirt/kvm -Iarch/x86/kvm 14EXTRA_CFLAGS += -Ivirt/kvm -Iarch/x86/kvm
12 15
diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
index 1bf8f57a3041..11c6725fb798 100644
--- a/arch/x86/kvm/i8254.c
+++ b/arch/x86/kvm/i8254.c
@@ -200,13 +200,14 @@ static int __pit_timer_fn(struct kvm_kpit_state *ps)
200 200
201 if (!atomic_inc_and_test(&pt->pending)) 201 if (!atomic_inc_and_test(&pt->pending))
202 set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests); 202 set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
203 if (vcpu0 && waitqueue_active(&vcpu0->wq)) { 203
204 vcpu0->arch.mp_state = KVM_MP_STATE_RUNNABLE; 204 if (vcpu0 && waitqueue_active(&vcpu0->wq))
205 wake_up_interruptible(&vcpu0->wq); 205 wake_up_interruptible(&vcpu0->wq);
206 }
207 206
208 hrtimer_add_expires_ns(&pt->timer, pt->period); 207 hrtimer_add_expires_ns(&pt->timer, pt->period);
209 pt->scheduled = ktime_to_ns(hrtimer_get_expires(&pt->timer)); 208 pt->scheduled = hrtimer_get_expires_ns(&pt->timer);
209 if (pt->period)
210 ps->channels[0].count_load_time = hrtimer_get_expires(&pt->timer);
210 211
211 return (pt->period == 0 ? 0 : 1); 212 return (pt->period == 0 ? 0 : 1);
212} 213}
@@ -215,12 +216,22 @@ int pit_has_pending_timer(struct kvm_vcpu *vcpu)
215{ 216{
216 struct kvm_pit *pit = vcpu->kvm->arch.vpit; 217 struct kvm_pit *pit = vcpu->kvm->arch.vpit;
217 218
218 if (pit && vcpu->vcpu_id == 0 && pit->pit_state.inject_pending) 219 if (pit && vcpu->vcpu_id == 0 && pit->pit_state.irq_ack)
219 return atomic_read(&pit->pit_state.pit_timer.pending); 220 return atomic_read(&pit->pit_state.pit_timer.pending);
220
221 return 0; 221 return 0;
222} 222}
223 223
224static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
225{
226 struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
227 irq_ack_notifier);
228 spin_lock(&ps->inject_lock);
229 if (atomic_dec_return(&ps->pit_timer.pending) < 0)
230 atomic_inc(&ps->pit_timer.pending);
231 ps->irq_ack = 1;
232 spin_unlock(&ps->inject_lock);
233}
234
224static enum hrtimer_restart pit_timer_fn(struct hrtimer *data) 235static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
225{ 236{
226 struct kvm_kpit_state *ps; 237 struct kvm_kpit_state *ps;
@@ -255,8 +266,9 @@ static void destroy_pit_timer(struct kvm_kpit_timer *pt)
255 hrtimer_cancel(&pt->timer); 266 hrtimer_cancel(&pt->timer);
256} 267}
257 268
258static void create_pit_timer(struct kvm_kpit_timer *pt, u32 val, int is_period) 269static void create_pit_timer(struct kvm_kpit_state *ps, u32 val, int is_period)
259{ 270{
271 struct kvm_kpit_timer *pt = &ps->pit_timer;
260 s64 interval; 272 s64 interval;
261 273
262 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ); 274 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
@@ -268,6 +280,7 @@ static void create_pit_timer(struct kvm_kpit_timer *pt, u32 val, int is_period)
268 pt->period = (is_period == 0) ? 0 : interval; 280 pt->period = (is_period == 0) ? 0 : interval;
269 pt->timer.function = pit_timer_fn; 281 pt->timer.function = pit_timer_fn;
270 atomic_set(&pt->pending, 0); 282 atomic_set(&pt->pending, 0);
283 ps->irq_ack = 1;
271 284
272 hrtimer_start(&pt->timer, ktime_add_ns(ktime_get(), interval), 285 hrtimer_start(&pt->timer, ktime_add_ns(ktime_get(), interval),
273 HRTIMER_MODE_ABS); 286 HRTIMER_MODE_ABS);
@@ -302,11 +315,11 @@ static void pit_load_count(struct kvm *kvm, int channel, u32 val)
302 case 1: 315 case 1:
303 /* FIXME: enhance mode 4 precision */ 316 /* FIXME: enhance mode 4 precision */
304 case 4: 317 case 4:
305 create_pit_timer(&ps->pit_timer, val, 0); 318 create_pit_timer(ps, val, 0);
306 break; 319 break;
307 case 2: 320 case 2:
308 case 3: 321 case 3:
309 create_pit_timer(&ps->pit_timer, val, 1); 322 create_pit_timer(ps, val, 1);
310 break; 323 break;
311 default: 324 default:
312 destroy_pit_timer(&ps->pit_timer); 325 destroy_pit_timer(&ps->pit_timer);
@@ -520,7 +533,7 @@ void kvm_pit_reset(struct kvm_pit *pit)
520 mutex_unlock(&pit->pit_state.lock); 533 mutex_unlock(&pit->pit_state.lock);
521 534
522 atomic_set(&pit->pit_state.pit_timer.pending, 0); 535 atomic_set(&pit->pit_state.pit_timer.pending, 0);
523 pit->pit_state.inject_pending = 1; 536 pit->pit_state.irq_ack = 1;
524} 537}
525 538
526struct kvm_pit *kvm_create_pit(struct kvm *kvm) 539struct kvm_pit *kvm_create_pit(struct kvm *kvm)
@@ -534,6 +547,7 @@ struct kvm_pit *kvm_create_pit(struct kvm *kvm)
534 547
535 mutex_init(&pit->pit_state.lock); 548 mutex_init(&pit->pit_state.lock);
536 mutex_lock(&pit->pit_state.lock); 549 mutex_lock(&pit->pit_state.lock);
550 spin_lock_init(&pit->pit_state.inject_lock);
537 551
538 /* Initialize PIO device */ 552 /* Initialize PIO device */
539 pit->dev.read = pit_ioport_read; 553 pit->dev.read = pit_ioport_read;
@@ -555,6 +569,9 @@ struct kvm_pit *kvm_create_pit(struct kvm *kvm)
555 pit_state->pit = pit; 569 pit_state->pit = pit;
556 hrtimer_init(&pit_state->pit_timer.timer, 570 hrtimer_init(&pit_state->pit_timer.timer,
557 CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 571 CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
572 pit_state->irq_ack_notifier.gsi = 0;
573 pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
574 kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
558 mutex_unlock(&pit->pit_state.lock); 575 mutex_unlock(&pit->pit_state.lock);
559 576
560 kvm_pit_reset(pit); 577 kvm_pit_reset(pit);
@@ -578,10 +595,8 @@ void kvm_free_pit(struct kvm *kvm)
578static void __inject_pit_timer_intr(struct kvm *kvm) 595static void __inject_pit_timer_intr(struct kvm *kvm)
579{ 596{
580 mutex_lock(&kvm->lock); 597 mutex_lock(&kvm->lock);
581 kvm_ioapic_set_irq(kvm->arch.vioapic, 0, 1); 598 kvm_set_irq(kvm, 0, 1);
582 kvm_ioapic_set_irq(kvm->arch.vioapic, 0, 0); 599 kvm_set_irq(kvm, 0, 0);
583 kvm_pic_set_irq(pic_irqchip(kvm), 0, 1);
584 kvm_pic_set_irq(pic_irqchip(kvm), 0, 0);
585 mutex_unlock(&kvm->lock); 600 mutex_unlock(&kvm->lock);
586} 601}
587 602
@@ -592,37 +607,19 @@ void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
592 struct kvm_kpit_state *ps; 607 struct kvm_kpit_state *ps;
593 608
594 if (vcpu && pit) { 609 if (vcpu && pit) {
610 int inject = 0;
595 ps = &pit->pit_state; 611 ps = &pit->pit_state;
596 612
597 /* Try to inject pending interrupts when: 613 /* Try to inject pending interrupts when
598 * 1. Pending exists 614 * last one has been acked.
599 * 2. Last interrupt was accepted or waited for too long time*/ 615 */
600 if (atomic_read(&ps->pit_timer.pending) && 616 spin_lock(&ps->inject_lock);
601 (ps->inject_pending || 617 if (atomic_read(&ps->pit_timer.pending) && ps->irq_ack) {
602 (jiffies - ps->last_injected_time 618 ps->irq_ack = 0;
603 >= KVM_MAX_PIT_INTR_INTERVAL))) { 619 inject = 1;
604 ps->inject_pending = 0;
605 __inject_pit_timer_intr(kvm);
606 ps->last_injected_time = jiffies;
607 }
608 }
609}
610
611void kvm_pit_timer_intr_post(struct kvm_vcpu *vcpu, int vec)
612{
613 struct kvm_arch *arch = &vcpu->kvm->arch;
614 struct kvm_kpit_state *ps;
615
616 if (vcpu && arch->vpit) {
617 ps = &arch->vpit->pit_state;
618 if (atomic_read(&ps->pit_timer.pending) &&
619 (((arch->vpic->pics[0].imr & 1) == 0 &&
620 arch->vpic->pics[0].irq_base == vec) ||
621 (arch->vioapic->redirtbl[0].fields.vector == vec &&
622 arch->vioapic->redirtbl[0].fields.mask != 1))) {
623 ps->inject_pending = 1;
624 atomic_dec(&ps->pit_timer.pending);
625 ps->channels[0].count_load_time = ktime_get();
626 } 620 }
621 spin_unlock(&ps->inject_lock);
622 if (inject)
623 __inject_pit_timer_intr(kvm);
627 } 624 }
628} 625}
diff --git a/arch/x86/kvm/i8254.h b/arch/x86/kvm/i8254.h
index db25c2a6c8c4..e436d4983aa1 100644
--- a/arch/x86/kvm/i8254.h
+++ b/arch/x86/kvm/i8254.h
@@ -8,7 +8,6 @@ struct kvm_kpit_timer {
8 int irq; 8 int irq;
9 s64 period; /* unit: ns */ 9 s64 period; /* unit: ns */
10 s64 scheduled; 10 s64 scheduled;
11 ktime_t last_update;
12 atomic_t pending; 11 atomic_t pending;
13}; 12};
14 13
@@ -34,8 +33,9 @@ struct kvm_kpit_state {
34 u32 speaker_data_on; 33 u32 speaker_data_on;
35 struct mutex lock; 34 struct mutex lock;
36 struct kvm_pit *pit; 35 struct kvm_pit *pit;
37 bool inject_pending; /* if inject pending interrupts */ 36 spinlock_t inject_lock;
38 unsigned long last_injected_time; 37 unsigned long irq_ack;
38 struct kvm_irq_ack_notifier irq_ack_notifier;
39}; 39};
40 40
41struct kvm_pit { 41struct kvm_pit {
@@ -54,7 +54,6 @@ struct kvm_pit {
54#define KVM_PIT_CHANNEL_MASK 0x3 54#define KVM_PIT_CHANNEL_MASK 0x3
55 55
56void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu); 56void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu);
57void kvm_pit_timer_intr_post(struct kvm_vcpu *vcpu, int vec);
58void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val); 57void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val);
59struct kvm_pit *kvm_create_pit(struct kvm *kvm); 58struct kvm_pit *kvm_create_pit(struct kvm *kvm);
60void kvm_free_pit(struct kvm *kvm); 59void kvm_free_pit(struct kvm *kvm);
diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
index c31164e8aa46..17e41e165f1a 100644
--- a/arch/x86/kvm/i8259.c
+++ b/arch/x86/kvm/i8259.c
@@ -30,6 +30,19 @@
30 30
31#include <linux/kvm_host.h> 31#include <linux/kvm_host.h>
32 32
33static void pic_clear_isr(struct kvm_kpic_state *s, int irq)
34{
35 s->isr &= ~(1 << irq);
36 s->isr_ack |= (1 << irq);
37}
38
39void kvm_pic_clear_isr_ack(struct kvm *kvm)
40{
41 struct kvm_pic *s = pic_irqchip(kvm);
42 s->pics[0].isr_ack = 0xff;
43 s->pics[1].isr_ack = 0xff;
44}
45
33/* 46/*
34 * set irq level. If an edge is detected, then the IRR is set to 1 47 * set irq level. If an edge is detected, then the IRR is set to 1
35 */ 48 */
@@ -141,11 +154,12 @@ void kvm_pic_set_irq(void *opaque, int irq, int level)
141 */ 154 */
142static inline void pic_intack(struct kvm_kpic_state *s, int irq) 155static inline void pic_intack(struct kvm_kpic_state *s, int irq)
143{ 156{
157 s->isr |= 1 << irq;
144 if (s->auto_eoi) { 158 if (s->auto_eoi) {
145 if (s->rotate_on_auto_eoi) 159 if (s->rotate_on_auto_eoi)
146 s->priority_add = (irq + 1) & 7; 160 s->priority_add = (irq + 1) & 7;
147 } else 161 pic_clear_isr(s, irq);
148 s->isr |= (1 << irq); 162 }
149 /* 163 /*
150 * We don't clear a level sensitive interrupt here 164 * We don't clear a level sensitive interrupt here
151 */ 165 */
@@ -153,9 +167,10 @@ static inline void pic_intack(struct kvm_kpic_state *s, int irq)
153 s->irr &= ~(1 << irq); 167 s->irr &= ~(1 << irq);
154} 168}
155 169
156int kvm_pic_read_irq(struct kvm_pic *s) 170int kvm_pic_read_irq(struct kvm *kvm)
157{ 171{
158 int irq, irq2, intno; 172 int irq, irq2, intno;
173 struct kvm_pic *s = pic_irqchip(kvm);
159 174
160 irq = pic_get_irq(&s->pics[0]); 175 irq = pic_get_irq(&s->pics[0]);
161 if (irq >= 0) { 176 if (irq >= 0) {
@@ -181,16 +196,32 @@ int kvm_pic_read_irq(struct kvm_pic *s)
181 intno = s->pics[0].irq_base + irq; 196 intno = s->pics[0].irq_base + irq;
182 } 197 }
183 pic_update_irq(s); 198 pic_update_irq(s);
199 kvm_notify_acked_irq(kvm, irq);
184 200
185 return intno; 201 return intno;
186} 202}
187 203
188void kvm_pic_reset(struct kvm_kpic_state *s) 204void kvm_pic_reset(struct kvm_kpic_state *s)
189{ 205{
206 int irq, irqbase;
207 struct kvm *kvm = s->pics_state->irq_request_opaque;
208 struct kvm_vcpu *vcpu0 = kvm->vcpus[0];
209
210 if (s == &s->pics_state->pics[0])
211 irqbase = 0;
212 else
213 irqbase = 8;
214
215 for (irq = 0; irq < PIC_NUM_PINS/2; irq++) {
216 if (vcpu0 && kvm_apic_accept_pic_intr(vcpu0))
217 if (s->irr & (1 << irq) || s->isr & (1 << irq))
218 kvm_notify_acked_irq(kvm, irq+irqbase);
219 }
190 s->last_irr = 0; 220 s->last_irr = 0;
191 s->irr = 0; 221 s->irr = 0;
192 s->imr = 0; 222 s->imr = 0;
193 s->isr = 0; 223 s->isr = 0;
224 s->isr_ack = 0xff;
194 s->priority_add = 0; 225 s->priority_add = 0;
195 s->irq_base = 0; 226 s->irq_base = 0;
196 s->read_reg_select = 0; 227 s->read_reg_select = 0;
@@ -243,7 +274,7 @@ static void pic_ioport_write(void *opaque, u32 addr, u32 val)
243 priority = get_priority(s, s->isr); 274 priority = get_priority(s, s->isr);
244 if (priority != 8) { 275 if (priority != 8) {
245 irq = (priority + s->priority_add) & 7; 276 irq = (priority + s->priority_add) & 7;
246 s->isr &= ~(1 << irq); 277 pic_clear_isr(s, irq);
247 if (cmd == 5) 278 if (cmd == 5)
248 s->priority_add = (irq + 1) & 7; 279 s->priority_add = (irq + 1) & 7;
249 pic_update_irq(s->pics_state); 280 pic_update_irq(s->pics_state);
@@ -251,7 +282,7 @@ static void pic_ioport_write(void *opaque, u32 addr, u32 val)
251 break; 282 break;
252 case 3: 283 case 3:
253 irq = val & 7; 284 irq = val & 7;
254 s->isr &= ~(1 << irq); 285 pic_clear_isr(s, irq);
255 pic_update_irq(s->pics_state); 286 pic_update_irq(s->pics_state);
256 break; 287 break;
257 case 6: 288 case 6:
@@ -260,8 +291,8 @@ static void pic_ioport_write(void *opaque, u32 addr, u32 val)
260 break; 291 break;
261 case 7: 292 case 7:
262 irq = val & 7; 293 irq = val & 7;
263 s->isr &= ~(1 << irq);
264 s->priority_add = (irq + 1) & 7; 294 s->priority_add = (irq + 1) & 7;
295 pic_clear_isr(s, irq);
265 pic_update_irq(s->pics_state); 296 pic_update_irq(s->pics_state);
266 break; 297 break;
267 default: 298 default:
@@ -303,7 +334,7 @@ static u32 pic_poll_read(struct kvm_kpic_state *s, u32 addr1)
303 s->pics_state->pics[0].irr &= ~(1 << 2); 334 s->pics_state->pics[0].irr &= ~(1 << 2);
304 } 335 }
305 s->irr &= ~(1 << ret); 336 s->irr &= ~(1 << ret);
306 s->isr &= ~(1 << ret); 337 pic_clear_isr(s, ret);
307 if (addr1 >> 7 || ret != 2) 338 if (addr1 >> 7 || ret != 2)
308 pic_update_irq(s->pics_state); 339 pic_update_irq(s->pics_state);
309 } else { 340 } else {
@@ -422,10 +453,14 @@ static void pic_irq_request(void *opaque, int level)
422{ 453{
423 struct kvm *kvm = opaque; 454 struct kvm *kvm = opaque;
424 struct kvm_vcpu *vcpu = kvm->vcpus[0]; 455 struct kvm_vcpu *vcpu = kvm->vcpus[0];
456 struct kvm_pic *s = pic_irqchip(kvm);
457 int irq = pic_get_irq(&s->pics[0]);
425 458
426 pic_irqchip(kvm)->output = level; 459 s->output = level;
427 if (vcpu) 460 if (vcpu && level && (s->pics[0].isr_ack & (1 << irq))) {
461 s->pics[0].isr_ack &= ~(1 << irq);
428 kvm_vcpu_kick(vcpu); 462 kvm_vcpu_kick(vcpu);
463 }
429} 464}
430 465
431struct kvm_pic *kvm_create_pic(struct kvm *kvm) 466struct kvm_pic *kvm_create_pic(struct kvm *kvm)
diff --git a/arch/x86/kvm/irq.c b/arch/x86/kvm/irq.c
index 76d736b5f664..c019b8edcdb7 100644
--- a/arch/x86/kvm/irq.c
+++ b/arch/x86/kvm/irq.c
@@ -72,7 +72,7 @@ int kvm_cpu_get_interrupt(struct kvm_vcpu *v)
72 if (kvm_apic_accept_pic_intr(v)) { 72 if (kvm_apic_accept_pic_intr(v)) {
73 s = pic_irqchip(v->kvm); 73 s = pic_irqchip(v->kvm);
74 s->output = 0; /* PIC */ 74 s->output = 0; /* PIC */
75 vector = kvm_pic_read_irq(s); 75 vector = kvm_pic_read_irq(v->kvm);
76 } 76 }
77 } 77 }
78 return vector; 78 return vector;
@@ -90,7 +90,6 @@ EXPORT_SYMBOL_GPL(kvm_inject_pending_timer_irqs);
90void kvm_timer_intr_post(struct kvm_vcpu *vcpu, int vec) 90void kvm_timer_intr_post(struct kvm_vcpu *vcpu, int vec)
91{ 91{
92 kvm_apic_timer_intr_post(vcpu, vec); 92 kvm_apic_timer_intr_post(vcpu, vec);
93 kvm_pit_timer_intr_post(vcpu, vec);
94 /* TODO: PIT, RTC etc. */ 93 /* TODO: PIT, RTC etc. */
95} 94}
96EXPORT_SYMBOL_GPL(kvm_timer_intr_post); 95EXPORT_SYMBOL_GPL(kvm_timer_intr_post);
diff --git a/arch/x86/kvm/irq.h b/arch/x86/kvm/irq.h
index 7ca47cbb48bb..f17c8f5bbf31 100644
--- a/arch/x86/kvm/irq.h
+++ b/arch/x86/kvm/irq.h
@@ -42,6 +42,7 @@ struct kvm_kpic_state {
42 u8 irr; /* interrupt request register */ 42 u8 irr; /* interrupt request register */
43 u8 imr; /* interrupt mask register */ 43 u8 imr; /* interrupt mask register */
44 u8 isr; /* interrupt service register */ 44 u8 isr; /* interrupt service register */
45 u8 isr_ack; /* interrupt ack detection */
45 u8 priority_add; /* highest irq priority */ 46 u8 priority_add; /* highest irq priority */
46 u8 irq_base; 47 u8 irq_base;
47 u8 read_reg_select; 48 u8 read_reg_select;
@@ -63,12 +64,13 @@ struct kvm_pic {
63 void *irq_request_opaque; 64 void *irq_request_opaque;
64 int output; /* intr from master PIC */ 65 int output; /* intr from master PIC */
65 struct kvm_io_device dev; 66 struct kvm_io_device dev;
67 void (*ack_notifier)(void *opaque, int irq);
66}; 68};
67 69
68struct kvm_pic *kvm_create_pic(struct kvm *kvm); 70struct kvm_pic *kvm_create_pic(struct kvm *kvm);
69void kvm_pic_set_irq(void *opaque, int irq, int level); 71int kvm_pic_read_irq(struct kvm *kvm);
70int kvm_pic_read_irq(struct kvm_pic *s);
71void kvm_pic_update_irq(struct kvm_pic *s); 72void kvm_pic_update_irq(struct kvm_pic *s);
73void kvm_pic_clear_isr_ack(struct kvm *kvm);
72 74
73static inline struct kvm_pic *pic_irqchip(struct kvm *kvm) 75static inline struct kvm_pic *pic_irqchip(struct kvm *kvm)
74{ 76{
diff --git a/arch/x86/kvm/kvm_cache_regs.h b/arch/x86/kvm/kvm_cache_regs.h
new file mode 100644
index 000000000000..1ff819dce7d3
--- /dev/null
+++ b/arch/x86/kvm/kvm_cache_regs.h
@@ -0,0 +1,32 @@
1#ifndef ASM_KVM_CACHE_REGS_H
2#define ASM_KVM_CACHE_REGS_H
3
4static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu,
5 enum kvm_reg reg)
6{
7 if (!test_bit(reg, (unsigned long *)&vcpu->arch.regs_avail))
8 kvm_x86_ops->cache_reg(vcpu, reg);
9
10 return vcpu->arch.regs[reg];
11}
12
13static inline void kvm_register_write(struct kvm_vcpu *vcpu,
14 enum kvm_reg reg,
15 unsigned long val)
16{
17 vcpu->arch.regs[reg] = val;
18 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_dirty);
19 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
20}
21
22static inline unsigned long kvm_rip_read(struct kvm_vcpu *vcpu)
23{
24 return kvm_register_read(vcpu, VCPU_REGS_RIP);
25}
26
27static inline void kvm_rip_write(struct kvm_vcpu *vcpu, unsigned long val)
28{
29 kvm_register_write(vcpu, VCPU_REGS_RIP, val);
30}
31
32#endif
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index a5b61de6adf1..0fc3cab48943 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -32,6 +32,7 @@
32#include <asm/current.h> 32#include <asm/current.h>
33#include <asm/apicdef.h> 33#include <asm/apicdef.h>
34#include <asm/atomic.h> 34#include <asm/atomic.h>
35#include "kvm_cache_regs.h"
35#include "irq.h" 36#include "irq.h"
36 37
37#define PRId64 "d" 38#define PRId64 "d"
@@ -338,13 +339,7 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
338 } else 339 } else
339 apic_clear_vector(vector, apic->regs + APIC_TMR); 340 apic_clear_vector(vector, apic->regs + APIC_TMR);
340 341
341 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE) 342 kvm_vcpu_kick(vcpu);
342 kvm_vcpu_kick(vcpu);
343 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
344 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
345 if (waitqueue_active(&vcpu->wq))
346 wake_up_interruptible(&vcpu->wq);
347 }
348 343
349 result = (orig_irr == 0); 344 result = (orig_irr == 0);
350 break; 345 break;
@@ -370,21 +365,18 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
370 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; 365 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
371 kvm_vcpu_kick(vcpu); 366 kvm_vcpu_kick(vcpu);
372 } else { 367 } else {
373 printk(KERN_DEBUG 368 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
374 "Ignoring de-assert INIT to vcpu %d\n", 369 vcpu->vcpu_id);
375 vcpu->vcpu_id);
376 } 370 }
377
378 break; 371 break;
379 372
380 case APIC_DM_STARTUP: 373 case APIC_DM_STARTUP:
381 printk(KERN_DEBUG "SIPI to vcpu %d vector 0x%02x\n", 374 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
382 vcpu->vcpu_id, vector); 375 vcpu->vcpu_id, vector);
383 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { 376 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
384 vcpu->arch.sipi_vector = vector; 377 vcpu->arch.sipi_vector = vector;
385 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED; 378 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
386 if (waitqueue_active(&vcpu->wq)) 379 kvm_vcpu_kick(vcpu);
387 wake_up_interruptible(&vcpu->wq);
388 } 380 }
389 break; 381 break;
390 382
@@ -438,7 +430,7 @@ struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
438static void apic_set_eoi(struct kvm_lapic *apic) 430static void apic_set_eoi(struct kvm_lapic *apic)
439{ 431{
440 int vector = apic_find_highest_isr(apic); 432 int vector = apic_find_highest_isr(apic);
441 433 int trigger_mode;
442 /* 434 /*
443 * Not every write EOI will has corresponding ISR, 435 * Not every write EOI will has corresponding ISR,
444 * one example is when Kernel check timer on setup_IO_APIC 436 * one example is when Kernel check timer on setup_IO_APIC
@@ -450,7 +442,10 @@ static void apic_set_eoi(struct kvm_lapic *apic)
450 apic_update_ppr(apic); 442 apic_update_ppr(apic);
451 443
452 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR)) 444 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
453 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector); 445 trigger_mode = IOAPIC_LEVEL_TRIG;
446 else
447 trigger_mode = IOAPIC_EDGE_TRIG;
448 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
454} 449}
455 450
456static void apic_send_ipi(struct kvm_lapic *apic) 451static void apic_send_ipi(struct kvm_lapic *apic)
@@ -558,8 +553,7 @@ static void __report_tpr_access(struct kvm_lapic *apic, bool write)
558 struct kvm_run *run = vcpu->run; 553 struct kvm_run *run = vcpu->run;
559 554
560 set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests); 555 set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
561 kvm_x86_ops->cache_regs(vcpu); 556 run->tpr_access.rip = kvm_rip_read(vcpu);
562 run->tpr_access.rip = vcpu->arch.rip;
563 run->tpr_access.is_write = write; 557 run->tpr_access.is_write = write;
564} 558}
565 559
@@ -683,9 +677,9 @@ static void apic_mmio_write(struct kvm_io_device *this,
683 * Refer SDM 8.4.1 677 * Refer SDM 8.4.1
684 */ 678 */
685 if (len != 4 || alignment) { 679 if (len != 4 || alignment) {
686 if (printk_ratelimit()) 680 /* Don't shout loud, $infamous_os would cause only noise. */
687 printk(KERN_ERR "apic write: bad size=%d %lx\n", 681 apic_debug("apic write: bad size=%d %lx\n",
688 len, (long)address); 682 len, (long)address);
689 return; 683 return;
690 } 684 }
691 685
@@ -947,10 +941,9 @@ static int __apic_timer_fn(struct kvm_lapic *apic)
947 941
948 if(!atomic_inc_and_test(&apic->timer.pending)) 942 if(!atomic_inc_and_test(&apic->timer.pending))
949 set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests); 943 set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
950 if (waitqueue_active(q)) { 944 if (waitqueue_active(q))
951 apic->vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
952 wake_up_interruptible(q); 945 wake_up_interruptible(q);
953 } 946
954 if (apic_lvtt_period(apic)) { 947 if (apic_lvtt_period(apic)) {
955 result = 1; 948 result = 1;
956 hrtimer_add_expires_ns(&apic->timer.dev, apic->timer.period); 949 hrtimer_add_expires_ns(&apic->timer.dev, apic->timer.period);
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 0bfe2bd305eb..99c239c5c0ac 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -70,6 +70,9 @@ static int dbg = 0;
70module_param(dbg, bool, 0644); 70module_param(dbg, bool, 0644);
71#endif 71#endif
72 72
73static int oos_shadow = 1;
74module_param(oos_shadow, bool, 0644);
75
73#ifndef MMU_DEBUG 76#ifndef MMU_DEBUG
74#define ASSERT(x) do { } while (0) 77#define ASSERT(x) do { } while (0)
75#else 78#else
@@ -135,18 +138,24 @@ module_param(dbg, bool, 0644);
135#define ACC_USER_MASK PT_USER_MASK 138#define ACC_USER_MASK PT_USER_MASK
136#define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) 139#define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
137 140
138struct kvm_pv_mmu_op_buffer { 141#define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
139 void *ptr;
140 unsigned len;
141 unsigned processed;
142 char buf[512] __aligned(sizeof(long));
143};
144 142
145struct kvm_rmap_desc { 143struct kvm_rmap_desc {
146 u64 *shadow_ptes[RMAP_EXT]; 144 u64 *shadow_ptes[RMAP_EXT];
147 struct kvm_rmap_desc *more; 145 struct kvm_rmap_desc *more;
148}; 146};
149 147
148struct kvm_shadow_walk {
149 int (*entry)(struct kvm_shadow_walk *walk, struct kvm_vcpu *vcpu,
150 u64 addr, u64 *spte, int level);
151};
152
153struct kvm_unsync_walk {
154 int (*entry) (struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk);
155};
156
157typedef int (*mmu_parent_walk_fn) (struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp);
158
150static struct kmem_cache *pte_chain_cache; 159static struct kmem_cache *pte_chain_cache;
151static struct kmem_cache *rmap_desc_cache; 160static struct kmem_cache *rmap_desc_cache;
152static struct kmem_cache *mmu_page_header_cache; 161static struct kmem_cache *mmu_page_header_cache;
@@ -405,16 +414,19 @@ static int host_largepage_backed(struct kvm *kvm, gfn_t gfn)
405{ 414{
406 struct vm_area_struct *vma; 415 struct vm_area_struct *vma;
407 unsigned long addr; 416 unsigned long addr;
417 int ret = 0;
408 418
409 addr = gfn_to_hva(kvm, gfn); 419 addr = gfn_to_hva(kvm, gfn);
410 if (kvm_is_error_hva(addr)) 420 if (kvm_is_error_hva(addr))
411 return 0; 421 return ret;
412 422
423 down_read(&current->mm->mmap_sem);
413 vma = find_vma(current->mm, addr); 424 vma = find_vma(current->mm, addr);
414 if (vma && is_vm_hugetlb_page(vma)) 425 if (vma && is_vm_hugetlb_page(vma))
415 return 1; 426 ret = 1;
427 up_read(&current->mm->mmap_sem);
416 428
417 return 0; 429 return ret;
418} 430}
419 431
420static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn) 432static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
@@ -649,8 +661,6 @@ static void rmap_write_protect(struct kvm *kvm, u64 gfn)
649 661
650 if (write_protected) 662 if (write_protected)
651 kvm_flush_remote_tlbs(kvm); 663 kvm_flush_remote_tlbs(kvm);
652
653 account_shadowed(kvm, gfn);
654} 664}
655 665
656static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp) 666static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
@@ -711,6 +721,10 @@ static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
711 u64 *spte; 721 u64 *spte;
712 int young = 0; 722 int young = 0;
713 723
724 /* always return old for EPT */
725 if (!shadow_accessed_mask)
726 return 0;
727
714 spte = rmap_next(kvm, rmapp, NULL); 728 spte = rmap_next(kvm, rmapp, NULL);
715 while (spte) { 729 while (spte) {
716 int _young; 730 int _young;
@@ -855,6 +869,77 @@ static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
855 BUG(); 869 BUG();
856} 870}
857 871
872
873static void mmu_parent_walk(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
874 mmu_parent_walk_fn fn)
875{
876 struct kvm_pte_chain *pte_chain;
877 struct hlist_node *node;
878 struct kvm_mmu_page *parent_sp;
879 int i;
880
881 if (!sp->multimapped && sp->parent_pte) {
882 parent_sp = page_header(__pa(sp->parent_pte));
883 fn(vcpu, parent_sp);
884 mmu_parent_walk(vcpu, parent_sp, fn);
885 return;
886 }
887 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
888 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
889 if (!pte_chain->parent_ptes[i])
890 break;
891 parent_sp = page_header(__pa(pte_chain->parent_ptes[i]));
892 fn(vcpu, parent_sp);
893 mmu_parent_walk(vcpu, parent_sp, fn);
894 }
895}
896
897static void kvm_mmu_update_unsync_bitmap(u64 *spte)
898{
899 unsigned int index;
900 struct kvm_mmu_page *sp = page_header(__pa(spte));
901
902 index = spte - sp->spt;
903 __set_bit(index, sp->unsync_child_bitmap);
904 sp->unsync_children = 1;
905}
906
907static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp)
908{
909 struct kvm_pte_chain *pte_chain;
910 struct hlist_node *node;
911 int i;
912
913 if (!sp->parent_pte)
914 return;
915
916 if (!sp->multimapped) {
917 kvm_mmu_update_unsync_bitmap(sp->parent_pte);
918 return;
919 }
920
921 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
922 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
923 if (!pte_chain->parent_ptes[i])
924 break;
925 kvm_mmu_update_unsync_bitmap(pte_chain->parent_ptes[i]);
926 }
927}
928
929static int unsync_walk_fn(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
930{
931 sp->unsync_children = 1;
932 kvm_mmu_update_parents_unsync(sp);
933 return 1;
934}
935
936static void kvm_mmu_mark_parents_unsync(struct kvm_vcpu *vcpu,
937 struct kvm_mmu_page *sp)
938{
939 mmu_parent_walk(vcpu, sp, unsync_walk_fn);
940 kvm_mmu_update_parents_unsync(sp);
941}
942
858static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, 943static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
859 struct kvm_mmu_page *sp) 944 struct kvm_mmu_page *sp)
860{ 945{
@@ -864,6 +949,58 @@ static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
864 sp->spt[i] = shadow_trap_nonpresent_pte; 949 sp->spt[i] = shadow_trap_nonpresent_pte;
865} 950}
866 951
952static int nonpaging_sync_page(struct kvm_vcpu *vcpu,
953 struct kvm_mmu_page *sp)
954{
955 return 1;
956}
957
958static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
959{
960}
961
962#define for_each_unsync_children(bitmap, idx) \
963 for (idx = find_first_bit(bitmap, 512); \
964 idx < 512; \
965 idx = find_next_bit(bitmap, 512, idx+1))
966
967static int mmu_unsync_walk(struct kvm_mmu_page *sp,
968 struct kvm_unsync_walk *walker)
969{
970 int i, ret;
971
972 if (!sp->unsync_children)
973 return 0;
974
975 for_each_unsync_children(sp->unsync_child_bitmap, i) {
976 u64 ent = sp->spt[i];
977
978 if (is_shadow_present_pte(ent)) {
979 struct kvm_mmu_page *child;
980 child = page_header(ent & PT64_BASE_ADDR_MASK);
981
982 if (child->unsync_children) {
983 ret = mmu_unsync_walk(child, walker);
984 if (ret)
985 return ret;
986 __clear_bit(i, sp->unsync_child_bitmap);
987 }
988
989 if (child->unsync) {
990 ret = walker->entry(child, walker);
991 __clear_bit(i, sp->unsync_child_bitmap);
992 if (ret)
993 return ret;
994 }
995 }
996 }
997
998 if (find_first_bit(sp->unsync_child_bitmap, 512) == 512)
999 sp->unsync_children = 0;
1000
1001 return 0;
1002}
1003
867static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn) 1004static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
868{ 1005{
869 unsigned index; 1006 unsigned index;
@@ -884,6 +1021,59 @@ static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
884 return NULL; 1021 return NULL;
885} 1022}
886 1023
1024static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1025{
1026 WARN_ON(!sp->unsync);
1027 sp->unsync = 0;
1028 --kvm->stat.mmu_unsync;
1029}
1030
1031static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp);
1032
1033static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1034{
1035 if (sp->role.glevels != vcpu->arch.mmu.root_level) {
1036 kvm_mmu_zap_page(vcpu->kvm, sp);
1037 return 1;
1038 }
1039
1040 rmap_write_protect(vcpu->kvm, sp->gfn);
1041 if (vcpu->arch.mmu.sync_page(vcpu, sp)) {
1042 kvm_mmu_zap_page(vcpu->kvm, sp);
1043 return 1;
1044 }
1045
1046 kvm_mmu_flush_tlb(vcpu);
1047 kvm_unlink_unsync_page(vcpu->kvm, sp);
1048 return 0;
1049}
1050
1051struct sync_walker {
1052 struct kvm_vcpu *vcpu;
1053 struct kvm_unsync_walk walker;
1054};
1055
1056static int mmu_sync_fn(struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk)
1057{
1058 struct sync_walker *sync_walk = container_of(walk, struct sync_walker,
1059 walker);
1060 struct kvm_vcpu *vcpu = sync_walk->vcpu;
1061
1062 kvm_sync_page(vcpu, sp);
1063 return (need_resched() || spin_needbreak(&vcpu->kvm->mmu_lock));
1064}
1065
1066static void mmu_sync_children(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1067{
1068 struct sync_walker walker = {
1069 .walker = { .entry = mmu_sync_fn, },
1070 .vcpu = vcpu,
1071 };
1072
1073 while (mmu_unsync_walk(sp, &walker.walker))
1074 cond_resched_lock(&vcpu->kvm->mmu_lock);
1075}
1076
887static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, 1077static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
888 gfn_t gfn, 1078 gfn_t gfn,
889 gva_t gaddr, 1079 gva_t gaddr,
@@ -897,7 +1087,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
897 unsigned quadrant; 1087 unsigned quadrant;
898 struct hlist_head *bucket; 1088 struct hlist_head *bucket;
899 struct kvm_mmu_page *sp; 1089 struct kvm_mmu_page *sp;
900 struct hlist_node *node; 1090 struct hlist_node *node, *tmp;
901 1091
902 role.word = 0; 1092 role.word = 0;
903 role.glevels = vcpu->arch.mmu.root_level; 1093 role.glevels = vcpu->arch.mmu.root_level;
@@ -913,9 +1103,20 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
913 gfn, role.word); 1103 gfn, role.word);
914 index = kvm_page_table_hashfn(gfn); 1104 index = kvm_page_table_hashfn(gfn);
915 bucket = &vcpu->kvm->arch.mmu_page_hash[index]; 1105 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
916 hlist_for_each_entry(sp, node, bucket, hash_link) 1106 hlist_for_each_entry_safe(sp, node, tmp, bucket, hash_link)
917 if (sp->gfn == gfn && sp->role.word == role.word) { 1107 if (sp->gfn == gfn) {
1108 if (sp->unsync)
1109 if (kvm_sync_page(vcpu, sp))
1110 continue;
1111
1112 if (sp->role.word != role.word)
1113 continue;
1114
918 mmu_page_add_parent_pte(vcpu, sp, parent_pte); 1115 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
1116 if (sp->unsync_children) {
1117 set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests);
1118 kvm_mmu_mark_parents_unsync(vcpu, sp);
1119 }
919 pgprintk("%s: found\n", __func__); 1120 pgprintk("%s: found\n", __func__);
920 return sp; 1121 return sp;
921 } 1122 }
@@ -927,8 +1128,10 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
927 sp->gfn = gfn; 1128 sp->gfn = gfn;
928 sp->role = role; 1129 sp->role = role;
929 hlist_add_head(&sp->hash_link, bucket); 1130 hlist_add_head(&sp->hash_link, bucket);
930 if (!metaphysical) 1131 if (!metaphysical) {
931 rmap_write_protect(vcpu->kvm, gfn); 1132 rmap_write_protect(vcpu->kvm, gfn);
1133 account_shadowed(vcpu->kvm, gfn);
1134 }
932 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte) 1135 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
933 vcpu->arch.mmu.prefetch_page(vcpu, sp); 1136 vcpu->arch.mmu.prefetch_page(vcpu, sp);
934 else 1137 else
@@ -936,6 +1139,35 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
936 return sp; 1139 return sp;
937} 1140}
938 1141
1142static int walk_shadow(struct kvm_shadow_walk *walker,
1143 struct kvm_vcpu *vcpu, u64 addr)
1144{
1145 hpa_t shadow_addr;
1146 int level;
1147 int r;
1148 u64 *sptep;
1149 unsigned index;
1150
1151 shadow_addr = vcpu->arch.mmu.root_hpa;
1152 level = vcpu->arch.mmu.shadow_root_level;
1153 if (level == PT32E_ROOT_LEVEL) {
1154 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
1155 shadow_addr &= PT64_BASE_ADDR_MASK;
1156 --level;
1157 }
1158
1159 while (level >= PT_PAGE_TABLE_LEVEL) {
1160 index = SHADOW_PT_INDEX(addr, level);
1161 sptep = ((u64 *)__va(shadow_addr)) + index;
1162 r = walker->entry(walker, vcpu, addr, sptep, level);
1163 if (r)
1164 return r;
1165 shadow_addr = *sptep & PT64_BASE_ADDR_MASK;
1166 --level;
1167 }
1168 return 0;
1169}
1170
939static void kvm_mmu_page_unlink_children(struct kvm *kvm, 1171static void kvm_mmu_page_unlink_children(struct kvm *kvm,
940 struct kvm_mmu_page *sp) 1172 struct kvm_mmu_page *sp)
941{ 1173{
@@ -951,7 +1183,6 @@ static void kvm_mmu_page_unlink_children(struct kvm *kvm,
951 rmap_remove(kvm, &pt[i]); 1183 rmap_remove(kvm, &pt[i]);
952 pt[i] = shadow_trap_nonpresent_pte; 1184 pt[i] = shadow_trap_nonpresent_pte;
953 } 1185 }
954 kvm_flush_remote_tlbs(kvm);
955 return; 1186 return;
956 } 1187 }
957 1188
@@ -970,7 +1201,6 @@ static void kvm_mmu_page_unlink_children(struct kvm *kvm,
970 } 1201 }
971 pt[i] = shadow_trap_nonpresent_pte; 1202 pt[i] = shadow_trap_nonpresent_pte;
972 } 1203 }
973 kvm_flush_remote_tlbs(kvm);
974} 1204}
975 1205
976static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) 1206static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
@@ -987,11 +1217,10 @@ static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
987 kvm->vcpus[i]->arch.last_pte_updated = NULL; 1217 kvm->vcpus[i]->arch.last_pte_updated = NULL;
988} 1218}
989 1219
990static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp) 1220static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
991{ 1221{
992 u64 *parent_pte; 1222 u64 *parent_pte;
993 1223
994 ++kvm->stat.mmu_shadow_zapped;
995 while (sp->multimapped || sp->parent_pte) { 1224 while (sp->multimapped || sp->parent_pte) {
996 if (!sp->multimapped) 1225 if (!sp->multimapped)
997 parent_pte = sp->parent_pte; 1226 parent_pte = sp->parent_pte;
@@ -1006,21 +1235,59 @@ static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1006 kvm_mmu_put_page(sp, parent_pte); 1235 kvm_mmu_put_page(sp, parent_pte);
1007 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); 1236 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
1008 } 1237 }
1238}
1239
1240struct zap_walker {
1241 struct kvm_unsync_walk walker;
1242 struct kvm *kvm;
1243 int zapped;
1244};
1245
1246static int mmu_zap_fn(struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk)
1247{
1248 struct zap_walker *zap_walk = container_of(walk, struct zap_walker,
1249 walker);
1250 kvm_mmu_zap_page(zap_walk->kvm, sp);
1251 zap_walk->zapped = 1;
1252 return 0;
1253}
1254
1255static int mmu_zap_unsync_children(struct kvm *kvm, struct kvm_mmu_page *sp)
1256{
1257 struct zap_walker walker = {
1258 .walker = { .entry = mmu_zap_fn, },
1259 .kvm = kvm,
1260 .zapped = 0,
1261 };
1262
1263 if (sp->role.level == PT_PAGE_TABLE_LEVEL)
1264 return 0;
1265 mmu_unsync_walk(sp, &walker.walker);
1266 return walker.zapped;
1267}
1268
1269static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1270{
1271 int ret;
1272 ++kvm->stat.mmu_shadow_zapped;
1273 ret = mmu_zap_unsync_children(kvm, sp);
1009 kvm_mmu_page_unlink_children(kvm, sp); 1274 kvm_mmu_page_unlink_children(kvm, sp);
1275 kvm_mmu_unlink_parents(kvm, sp);
1276 kvm_flush_remote_tlbs(kvm);
1277 if (!sp->role.invalid && !sp->role.metaphysical)
1278 unaccount_shadowed(kvm, sp->gfn);
1279 if (sp->unsync)
1280 kvm_unlink_unsync_page(kvm, sp);
1010 if (!sp->root_count) { 1281 if (!sp->root_count) {
1011 if (!sp->role.metaphysical && !sp->role.invalid)
1012 unaccount_shadowed(kvm, sp->gfn);
1013 hlist_del(&sp->hash_link); 1282 hlist_del(&sp->hash_link);
1014 kvm_mmu_free_page(kvm, sp); 1283 kvm_mmu_free_page(kvm, sp);
1015 } else { 1284 } else {
1016 int invalid = sp->role.invalid;
1017 list_move(&sp->link, &kvm->arch.active_mmu_pages);
1018 sp->role.invalid = 1; 1285 sp->role.invalid = 1;
1286 list_move(&sp->link, &kvm->arch.active_mmu_pages);
1019 kvm_reload_remote_mmus(kvm); 1287 kvm_reload_remote_mmus(kvm);
1020 if (!sp->role.metaphysical && !invalid)
1021 unaccount_shadowed(kvm, sp->gfn);
1022 } 1288 }
1023 kvm_mmu_reset_last_pte_updated(kvm); 1289 kvm_mmu_reset_last_pte_updated(kvm);
1290 return ret;
1024} 1291}
1025 1292
1026/* 1293/*
@@ -1073,8 +1340,9 @@ static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1073 if (sp->gfn == gfn && !sp->role.metaphysical) { 1340 if (sp->gfn == gfn && !sp->role.metaphysical) {
1074 pgprintk("%s: gfn %lx role %x\n", __func__, gfn, 1341 pgprintk("%s: gfn %lx role %x\n", __func__, gfn,
1075 sp->role.word); 1342 sp->role.word);
1076 kvm_mmu_zap_page(kvm, sp);
1077 r = 1; 1343 r = 1;
1344 if (kvm_mmu_zap_page(kvm, sp))
1345 n = bucket->first;
1078 } 1346 }
1079 return r; 1347 return r;
1080} 1348}
@@ -1097,6 +1365,20 @@ static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
1097 __set_bit(slot, &sp->slot_bitmap); 1365 __set_bit(slot, &sp->slot_bitmap);
1098} 1366}
1099 1367
1368static void mmu_convert_notrap(struct kvm_mmu_page *sp)
1369{
1370 int i;
1371 u64 *pt = sp->spt;
1372
1373 if (shadow_trap_nonpresent_pte == shadow_notrap_nonpresent_pte)
1374 return;
1375
1376 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1377 if (pt[i] == shadow_notrap_nonpresent_pte)
1378 set_shadow_pte(&pt[i], shadow_trap_nonpresent_pte);
1379 }
1380}
1381
1100struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) 1382struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1101{ 1383{
1102 struct page *page; 1384 struct page *page;
@@ -1106,51 +1388,60 @@ struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1106 if (gpa == UNMAPPED_GVA) 1388 if (gpa == UNMAPPED_GVA)
1107 return NULL; 1389 return NULL;
1108 1390
1109 down_read(&current->mm->mmap_sem);
1110 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); 1391 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1111 up_read(&current->mm->mmap_sem);
1112 1392
1113 return page; 1393 return page;
1114} 1394}
1115 1395
1116static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, 1396static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1117 unsigned pt_access, unsigned pte_access,
1118 int user_fault, int write_fault, int dirty,
1119 int *ptwrite, int largepage, gfn_t gfn,
1120 pfn_t pfn, bool speculative)
1121{ 1397{
1122 u64 spte; 1398 unsigned index;
1123 int was_rmapped = 0; 1399 struct hlist_head *bucket;
1124 int was_writeble = is_writeble_pte(*shadow_pte); 1400 struct kvm_mmu_page *s;
1401 struct hlist_node *node, *n;
1125 1402
1126 pgprintk("%s: spte %llx access %x write_fault %d" 1403 index = kvm_page_table_hashfn(sp->gfn);
1127 " user_fault %d gfn %lx\n", 1404 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1128 __func__, *shadow_pte, pt_access, 1405 /* don't unsync if pagetable is shadowed with multiple roles */
1129 write_fault, user_fault, gfn); 1406 hlist_for_each_entry_safe(s, node, n, bucket, hash_link) {
1407 if (s->gfn != sp->gfn || s->role.metaphysical)
1408 continue;
1409 if (s->role.word != sp->role.word)
1410 return 1;
1411 }
1412 kvm_mmu_mark_parents_unsync(vcpu, sp);
1413 ++vcpu->kvm->stat.mmu_unsync;
1414 sp->unsync = 1;
1415 mmu_convert_notrap(sp);
1416 return 0;
1417}
1130 1418
1131 if (is_rmap_pte(*shadow_pte)) { 1419static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
1132 /* 1420 bool can_unsync)
1133 * If we overwrite a PTE page pointer with a 2MB PMD, unlink 1421{
1134 * the parent of the now unreachable PTE. 1422 struct kvm_mmu_page *shadow;
1135 */
1136 if (largepage && !is_large_pte(*shadow_pte)) {
1137 struct kvm_mmu_page *child;
1138 u64 pte = *shadow_pte;
1139 1423
1140 child = page_header(pte & PT64_BASE_ADDR_MASK); 1424 shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn);
1141 mmu_page_remove_parent_pte(child, shadow_pte); 1425 if (shadow) {
1142 } else if (pfn != spte_to_pfn(*shadow_pte)) { 1426 if (shadow->role.level != PT_PAGE_TABLE_LEVEL)
1143 pgprintk("hfn old %lx new %lx\n", 1427 return 1;
1144 spte_to_pfn(*shadow_pte), pfn); 1428 if (shadow->unsync)
1145 rmap_remove(vcpu->kvm, shadow_pte); 1429 return 0;
1146 } else { 1430 if (can_unsync && oos_shadow)
1147 if (largepage) 1431 return kvm_unsync_page(vcpu, shadow);
1148 was_rmapped = is_large_pte(*shadow_pte); 1432 return 1;
1149 else
1150 was_rmapped = 1;
1151 }
1152 } 1433 }
1434 return 0;
1435}
1153 1436
1437static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1438 unsigned pte_access, int user_fault,
1439 int write_fault, int dirty, int largepage,
1440 gfn_t gfn, pfn_t pfn, bool speculative,
1441 bool can_unsync)
1442{
1443 u64 spte;
1444 int ret = 0;
1154 /* 1445 /*
1155 * We don't set the accessed bit, since we sometimes want to see 1446 * We don't set the accessed bit, since we sometimes want to see
1156 * whether the guest actually used the pte (in order to detect 1447 * whether the guest actually used the pte (in order to detect
@@ -1158,7 +1449,7 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1158 */ 1449 */
1159 spte = shadow_base_present_pte | shadow_dirty_mask; 1450 spte = shadow_base_present_pte | shadow_dirty_mask;
1160 if (!speculative) 1451 if (!speculative)
1161 pte_access |= PT_ACCESSED_MASK; 1452 spte |= shadow_accessed_mask;
1162 if (!dirty) 1453 if (!dirty)
1163 pte_access &= ~ACC_WRITE_MASK; 1454 pte_access &= ~ACC_WRITE_MASK;
1164 if (pte_access & ACC_EXEC_MASK) 1455 if (pte_access & ACC_EXEC_MASK)
@@ -1174,35 +1465,82 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1174 1465
1175 if ((pte_access & ACC_WRITE_MASK) 1466 if ((pte_access & ACC_WRITE_MASK)
1176 || (write_fault && !is_write_protection(vcpu) && !user_fault)) { 1467 || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
1177 struct kvm_mmu_page *shadow; 1468
1469 if (largepage && has_wrprotected_page(vcpu->kvm, gfn)) {
1470 ret = 1;
1471 spte = shadow_trap_nonpresent_pte;
1472 goto set_pte;
1473 }
1178 1474
1179 spte |= PT_WRITABLE_MASK; 1475 spte |= PT_WRITABLE_MASK;
1180 1476
1181 shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn); 1477 if (mmu_need_write_protect(vcpu, gfn, can_unsync)) {
1182 if (shadow ||
1183 (largepage && has_wrprotected_page(vcpu->kvm, gfn))) {
1184 pgprintk("%s: found shadow page for %lx, marking ro\n", 1478 pgprintk("%s: found shadow page for %lx, marking ro\n",
1185 __func__, gfn); 1479 __func__, gfn);
1480 ret = 1;
1186 pte_access &= ~ACC_WRITE_MASK; 1481 pte_access &= ~ACC_WRITE_MASK;
1187 if (is_writeble_pte(spte)) { 1482 if (is_writeble_pte(spte))
1188 spte &= ~PT_WRITABLE_MASK; 1483 spte &= ~PT_WRITABLE_MASK;
1189 kvm_x86_ops->tlb_flush(vcpu);
1190 }
1191 if (write_fault)
1192 *ptwrite = 1;
1193 } 1484 }
1194 } 1485 }
1195 1486
1196 if (pte_access & ACC_WRITE_MASK) 1487 if (pte_access & ACC_WRITE_MASK)
1197 mark_page_dirty(vcpu->kvm, gfn); 1488 mark_page_dirty(vcpu->kvm, gfn);
1198 1489
1199 pgprintk("%s: setting spte %llx\n", __func__, spte); 1490set_pte:
1200 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1201 (spte&PT_PAGE_SIZE_MASK)? "2MB" : "4kB",
1202 (spte&PT_WRITABLE_MASK)?"RW":"R", gfn, spte, shadow_pte);
1203 set_shadow_pte(shadow_pte, spte); 1491 set_shadow_pte(shadow_pte, spte);
1204 if (!was_rmapped && (spte & PT_PAGE_SIZE_MASK) 1492 return ret;
1205 && (spte & PT_PRESENT_MASK)) 1493}
1494
1495static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1496 unsigned pt_access, unsigned pte_access,
1497 int user_fault, int write_fault, int dirty,
1498 int *ptwrite, int largepage, gfn_t gfn,
1499 pfn_t pfn, bool speculative)
1500{
1501 int was_rmapped = 0;
1502 int was_writeble = is_writeble_pte(*shadow_pte);
1503
1504 pgprintk("%s: spte %llx access %x write_fault %d"
1505 " user_fault %d gfn %lx\n",
1506 __func__, *shadow_pte, pt_access,
1507 write_fault, user_fault, gfn);
1508
1509 if (is_rmap_pte(*shadow_pte)) {
1510 /*
1511 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1512 * the parent of the now unreachable PTE.
1513 */
1514 if (largepage && !is_large_pte(*shadow_pte)) {
1515 struct kvm_mmu_page *child;
1516 u64 pte = *shadow_pte;
1517
1518 child = page_header(pte & PT64_BASE_ADDR_MASK);
1519 mmu_page_remove_parent_pte(child, shadow_pte);
1520 } else if (pfn != spte_to_pfn(*shadow_pte)) {
1521 pgprintk("hfn old %lx new %lx\n",
1522 spte_to_pfn(*shadow_pte), pfn);
1523 rmap_remove(vcpu->kvm, shadow_pte);
1524 } else {
1525 if (largepage)
1526 was_rmapped = is_large_pte(*shadow_pte);
1527 else
1528 was_rmapped = 1;
1529 }
1530 }
1531 if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault,
1532 dirty, largepage, gfn, pfn, speculative, true)) {
1533 if (write_fault)
1534 *ptwrite = 1;
1535 kvm_x86_ops->tlb_flush(vcpu);
1536 }
1537
1538 pgprintk("%s: setting spte %llx\n", __func__, *shadow_pte);
1539 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1540 is_large_pte(*shadow_pte)? "2MB" : "4kB",
1541 is_present_pte(*shadow_pte)?"RW":"R", gfn,
1542 *shadow_pte, shadow_pte);
1543 if (!was_rmapped && is_large_pte(*shadow_pte))
1206 ++vcpu->kvm->stat.lpages; 1544 ++vcpu->kvm->stat.lpages;
1207 1545
1208 page_header_update_slot(vcpu->kvm, shadow_pte, gfn); 1546 page_header_update_slot(vcpu->kvm, shadow_pte, gfn);
@@ -1226,54 +1564,67 @@ static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
1226{ 1564{
1227} 1565}
1228 1566
1229static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, 1567struct direct_shadow_walk {
1230 int largepage, gfn_t gfn, pfn_t pfn, 1568 struct kvm_shadow_walk walker;
1231 int level) 1569 pfn_t pfn;
1232{ 1570 int write;
1233 hpa_t table_addr = vcpu->arch.mmu.root_hpa; 1571 int largepage;
1234 int pt_write = 0; 1572 int pt_write;
1235 1573};
1236 for (; ; level--) {
1237 u32 index = PT64_INDEX(v, level);
1238 u64 *table;
1239
1240 ASSERT(VALID_PAGE(table_addr));
1241 table = __va(table_addr);
1242 1574
1243 if (level == 1) { 1575static int direct_map_entry(struct kvm_shadow_walk *_walk,
1244 mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL, 1576 struct kvm_vcpu *vcpu,
1245 0, write, 1, &pt_write, 0, gfn, pfn, false); 1577 u64 addr, u64 *sptep, int level)
1246 return pt_write; 1578{
1247 } 1579 struct direct_shadow_walk *walk =
1580 container_of(_walk, struct direct_shadow_walk, walker);
1581 struct kvm_mmu_page *sp;
1582 gfn_t pseudo_gfn;
1583 gfn_t gfn = addr >> PAGE_SHIFT;
1584
1585 if (level == PT_PAGE_TABLE_LEVEL
1586 || (walk->largepage && level == PT_DIRECTORY_LEVEL)) {
1587 mmu_set_spte(vcpu, sptep, ACC_ALL, ACC_ALL,
1588 0, walk->write, 1, &walk->pt_write,
1589 walk->largepage, gfn, walk->pfn, false);
1590 ++vcpu->stat.pf_fixed;
1591 return 1;
1592 }
1248 1593
1249 if (largepage && level == 2) { 1594 if (*sptep == shadow_trap_nonpresent_pte) {
1250 mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL, 1595 pseudo_gfn = (addr & PT64_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
1251 0, write, 1, &pt_write, 1, gfn, pfn, false); 1596 sp = kvm_mmu_get_page(vcpu, pseudo_gfn, (gva_t)addr, level - 1,
1252 return pt_write; 1597 1, ACC_ALL, sptep);
1598 if (!sp) {
1599 pgprintk("nonpaging_map: ENOMEM\n");
1600 kvm_release_pfn_clean(walk->pfn);
1601 return -ENOMEM;
1253 } 1602 }
1254 1603
1255 if (table[index] == shadow_trap_nonpresent_pte) { 1604 set_shadow_pte(sptep,
1256 struct kvm_mmu_page *new_table; 1605 __pa(sp->spt)
1257 gfn_t pseudo_gfn; 1606 | PT_PRESENT_MASK | PT_WRITABLE_MASK
1258 1607 | shadow_user_mask | shadow_x_mask);
1259 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
1260 >> PAGE_SHIFT;
1261 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
1262 v, level - 1,
1263 1, ACC_ALL, &table[index]);
1264 if (!new_table) {
1265 pgprintk("nonpaging_map: ENOMEM\n");
1266 kvm_release_pfn_clean(pfn);
1267 return -ENOMEM;
1268 }
1269
1270 set_shadow_pte(&table[index],
1271 __pa(new_table->spt)
1272 | PT_PRESENT_MASK | PT_WRITABLE_MASK
1273 | shadow_user_mask | shadow_x_mask);
1274 }
1275 table_addr = table[index] & PT64_BASE_ADDR_MASK;
1276 } 1608 }
1609 return 0;
1610}
1611
1612static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
1613 int largepage, gfn_t gfn, pfn_t pfn)
1614{
1615 int r;
1616 struct direct_shadow_walk walker = {
1617 .walker = { .entry = direct_map_entry, },
1618 .pfn = pfn,
1619 .largepage = largepage,
1620 .write = write,
1621 .pt_write = 0,
1622 };
1623
1624 r = walk_shadow(&walker.walker, vcpu, gfn << PAGE_SHIFT);
1625 if (r < 0)
1626 return r;
1627 return walker.pt_write;
1277} 1628}
1278 1629
1279static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn) 1630static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
@@ -1283,16 +1634,14 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
1283 pfn_t pfn; 1634 pfn_t pfn;
1284 unsigned long mmu_seq; 1635 unsigned long mmu_seq;
1285 1636
1286 down_read(&current->mm->mmap_sem);
1287 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { 1637 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1288 gfn &= ~(KVM_PAGES_PER_HPAGE-1); 1638 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1289 largepage = 1; 1639 largepage = 1;
1290 } 1640 }
1291 1641
1292 mmu_seq = vcpu->kvm->mmu_notifier_seq; 1642 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1293 /* implicit mb(), we'll read before PT lock is unlocked */ 1643 smp_rmb();
1294 pfn = gfn_to_pfn(vcpu->kvm, gfn); 1644 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1295 up_read(&current->mm->mmap_sem);
1296 1645
1297 /* mmio */ 1646 /* mmio */
1298 if (is_error_pfn(pfn)) { 1647 if (is_error_pfn(pfn)) {
@@ -1304,8 +1653,7 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
1304 if (mmu_notifier_retry(vcpu, mmu_seq)) 1653 if (mmu_notifier_retry(vcpu, mmu_seq))
1305 goto out_unlock; 1654 goto out_unlock;
1306 kvm_mmu_free_some_pages(vcpu); 1655 kvm_mmu_free_some_pages(vcpu);
1307 r = __direct_map(vcpu, v, write, largepage, gfn, pfn, 1656 r = __direct_map(vcpu, v, write, largepage, gfn, pfn);
1308 PT32E_ROOT_LEVEL);
1309 spin_unlock(&vcpu->kvm->mmu_lock); 1657 spin_unlock(&vcpu->kvm->mmu_lock);
1310 1658
1311 1659
@@ -1401,6 +1749,37 @@ static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
1401 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); 1749 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
1402} 1750}
1403 1751
1752static void mmu_sync_roots(struct kvm_vcpu *vcpu)
1753{
1754 int i;
1755 struct kvm_mmu_page *sp;
1756
1757 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1758 return;
1759 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1760 hpa_t root = vcpu->arch.mmu.root_hpa;
1761 sp = page_header(root);
1762 mmu_sync_children(vcpu, sp);
1763 return;
1764 }
1765 for (i = 0; i < 4; ++i) {
1766 hpa_t root = vcpu->arch.mmu.pae_root[i];
1767
1768 if (root) {
1769 root &= PT64_BASE_ADDR_MASK;
1770 sp = page_header(root);
1771 mmu_sync_children(vcpu, sp);
1772 }
1773 }
1774}
1775
1776void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
1777{
1778 spin_lock(&vcpu->kvm->mmu_lock);
1779 mmu_sync_roots(vcpu);
1780 spin_unlock(&vcpu->kvm->mmu_lock);
1781}
1782
1404static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) 1783static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1405{ 1784{
1406 return vaddr; 1785 return vaddr;
@@ -1442,15 +1821,13 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
1442 if (r) 1821 if (r)
1443 return r; 1822 return r;
1444 1823
1445 down_read(&current->mm->mmap_sem);
1446 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { 1824 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1447 gfn &= ~(KVM_PAGES_PER_HPAGE-1); 1825 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1448 largepage = 1; 1826 largepage = 1;
1449 } 1827 }
1450 mmu_seq = vcpu->kvm->mmu_notifier_seq; 1828 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1451 /* implicit mb(), we'll read before PT lock is unlocked */ 1829 smp_rmb();
1452 pfn = gfn_to_pfn(vcpu->kvm, gfn); 1830 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1453 up_read(&current->mm->mmap_sem);
1454 if (is_error_pfn(pfn)) { 1831 if (is_error_pfn(pfn)) {
1455 kvm_release_pfn_clean(pfn); 1832 kvm_release_pfn_clean(pfn);
1456 return 1; 1833 return 1;
@@ -1460,7 +1837,7 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
1460 goto out_unlock; 1837 goto out_unlock;
1461 kvm_mmu_free_some_pages(vcpu); 1838 kvm_mmu_free_some_pages(vcpu);
1462 r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK, 1839 r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
1463 largepage, gfn, pfn, kvm_x86_ops->get_tdp_level()); 1840 largepage, gfn, pfn);
1464 spin_unlock(&vcpu->kvm->mmu_lock); 1841 spin_unlock(&vcpu->kvm->mmu_lock);
1465 1842
1466 return r; 1843 return r;
@@ -1485,6 +1862,8 @@ static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1485 context->gva_to_gpa = nonpaging_gva_to_gpa; 1862 context->gva_to_gpa = nonpaging_gva_to_gpa;
1486 context->free = nonpaging_free; 1863 context->free = nonpaging_free;
1487 context->prefetch_page = nonpaging_prefetch_page; 1864 context->prefetch_page = nonpaging_prefetch_page;
1865 context->sync_page = nonpaging_sync_page;
1866 context->invlpg = nonpaging_invlpg;
1488 context->root_level = 0; 1867 context->root_level = 0;
1489 context->shadow_root_level = PT32E_ROOT_LEVEL; 1868 context->shadow_root_level = PT32E_ROOT_LEVEL;
1490 context->root_hpa = INVALID_PAGE; 1869 context->root_hpa = INVALID_PAGE;
@@ -1532,6 +1911,8 @@ static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1532 context->page_fault = paging64_page_fault; 1911 context->page_fault = paging64_page_fault;
1533 context->gva_to_gpa = paging64_gva_to_gpa; 1912 context->gva_to_gpa = paging64_gva_to_gpa;
1534 context->prefetch_page = paging64_prefetch_page; 1913 context->prefetch_page = paging64_prefetch_page;
1914 context->sync_page = paging64_sync_page;
1915 context->invlpg = paging64_invlpg;
1535 context->free = paging_free; 1916 context->free = paging_free;
1536 context->root_level = level; 1917 context->root_level = level;
1537 context->shadow_root_level = level; 1918 context->shadow_root_level = level;
@@ -1553,6 +1934,8 @@ static int paging32_init_context(struct kvm_vcpu *vcpu)
1553 context->gva_to_gpa = paging32_gva_to_gpa; 1934 context->gva_to_gpa = paging32_gva_to_gpa;
1554 context->free = paging_free; 1935 context->free = paging_free;
1555 context->prefetch_page = paging32_prefetch_page; 1936 context->prefetch_page = paging32_prefetch_page;
1937 context->sync_page = paging32_sync_page;
1938 context->invlpg = paging32_invlpg;
1556 context->root_level = PT32_ROOT_LEVEL; 1939 context->root_level = PT32_ROOT_LEVEL;
1557 context->shadow_root_level = PT32E_ROOT_LEVEL; 1940 context->shadow_root_level = PT32E_ROOT_LEVEL;
1558 context->root_hpa = INVALID_PAGE; 1941 context->root_hpa = INVALID_PAGE;
@@ -1572,6 +1955,8 @@ static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
1572 context->page_fault = tdp_page_fault; 1955 context->page_fault = tdp_page_fault;
1573 context->free = nonpaging_free; 1956 context->free = nonpaging_free;
1574 context->prefetch_page = nonpaging_prefetch_page; 1957 context->prefetch_page = nonpaging_prefetch_page;
1958 context->sync_page = nonpaging_sync_page;
1959 context->invlpg = nonpaging_invlpg;
1575 context->shadow_root_level = kvm_x86_ops->get_tdp_level(); 1960 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
1576 context->root_hpa = INVALID_PAGE; 1961 context->root_hpa = INVALID_PAGE;
1577 1962
@@ -1643,6 +2028,7 @@ int kvm_mmu_load(struct kvm_vcpu *vcpu)
1643 spin_lock(&vcpu->kvm->mmu_lock); 2028 spin_lock(&vcpu->kvm->mmu_lock);
1644 kvm_mmu_free_some_pages(vcpu); 2029 kvm_mmu_free_some_pages(vcpu);
1645 mmu_alloc_roots(vcpu); 2030 mmu_alloc_roots(vcpu);
2031 mmu_sync_roots(vcpu);
1646 spin_unlock(&vcpu->kvm->mmu_lock); 2032 spin_unlock(&vcpu->kvm->mmu_lock);
1647 kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa); 2033 kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
1648 kvm_mmu_flush_tlb(vcpu); 2034 kvm_mmu_flush_tlb(vcpu);
@@ -1763,15 +2149,13 @@ static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1763 return; 2149 return;
1764 gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; 2150 gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
1765 2151
1766 down_read(&current->mm->mmap_sem);
1767 if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) { 2152 if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) {
1768 gfn &= ~(KVM_PAGES_PER_HPAGE-1); 2153 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1769 vcpu->arch.update_pte.largepage = 1; 2154 vcpu->arch.update_pte.largepage = 1;
1770 } 2155 }
1771 vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq; 2156 vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
1772 /* implicit mb(), we'll read before PT lock is unlocked */ 2157 smp_rmb();
1773 pfn = gfn_to_pfn(vcpu->kvm, gfn); 2158 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1774 up_read(&current->mm->mmap_sem);
1775 2159
1776 if (is_error_pfn(pfn)) { 2160 if (is_error_pfn(pfn)) {
1777 kvm_release_pfn_clean(pfn); 2161 kvm_release_pfn_clean(pfn);
@@ -1833,7 +2217,7 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1833 index = kvm_page_table_hashfn(gfn); 2217 index = kvm_page_table_hashfn(gfn);
1834 bucket = &vcpu->kvm->arch.mmu_page_hash[index]; 2218 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1835 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) { 2219 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
1836 if (sp->gfn != gfn || sp->role.metaphysical) 2220 if (sp->gfn != gfn || sp->role.metaphysical || sp->role.invalid)
1837 continue; 2221 continue;
1838 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; 2222 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1839 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); 2223 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
@@ -1851,7 +2235,8 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1851 */ 2235 */
1852 pgprintk("misaligned: gpa %llx bytes %d role %x\n", 2236 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1853 gpa, bytes, sp->role.word); 2237 gpa, bytes, sp->role.word);
1854 kvm_mmu_zap_page(vcpu->kvm, sp); 2238 if (kvm_mmu_zap_page(vcpu->kvm, sp))
2239 n = bucket->first;
1855 ++vcpu->kvm->stat.mmu_flooded; 2240 ++vcpu->kvm->stat.mmu_flooded;
1856 continue; 2241 continue;
1857 } 2242 }
@@ -1965,6 +2350,16 @@ out:
1965} 2350}
1966EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); 2351EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1967 2352
2353void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
2354{
2355 spin_lock(&vcpu->kvm->mmu_lock);
2356 vcpu->arch.mmu.invlpg(vcpu, gva);
2357 spin_unlock(&vcpu->kvm->mmu_lock);
2358 kvm_mmu_flush_tlb(vcpu);
2359 ++vcpu->stat.invlpg;
2360}
2361EXPORT_SYMBOL_GPL(kvm_mmu_invlpg);
2362
1968void kvm_enable_tdp(void) 2363void kvm_enable_tdp(void)
1969{ 2364{
1970 tdp_enabled = true; 2365 tdp_enabled = true;
@@ -2051,6 +2446,7 @@ void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2051{ 2446{
2052 struct kvm_mmu_page *sp; 2447 struct kvm_mmu_page *sp;
2053 2448
2449 spin_lock(&kvm->mmu_lock);
2054 list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) { 2450 list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
2055 int i; 2451 int i;
2056 u64 *pt; 2452 u64 *pt;
@@ -2064,6 +2460,8 @@ void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2064 if (pt[i] & PT_WRITABLE_MASK) 2460 if (pt[i] & PT_WRITABLE_MASK)
2065 pt[i] &= ~PT_WRITABLE_MASK; 2461 pt[i] &= ~PT_WRITABLE_MASK;
2066 } 2462 }
2463 kvm_flush_remote_tlbs(kvm);
2464 spin_unlock(&kvm->mmu_lock);
2067} 2465}
2068 2466
2069void kvm_mmu_zap_all(struct kvm *kvm) 2467void kvm_mmu_zap_all(struct kvm *kvm)
@@ -2072,7 +2470,9 @@ void kvm_mmu_zap_all(struct kvm *kvm)
2072 2470
2073 spin_lock(&kvm->mmu_lock); 2471 spin_lock(&kvm->mmu_lock);
2074 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) 2472 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link)
2075 kvm_mmu_zap_page(kvm, sp); 2473 if (kvm_mmu_zap_page(kvm, sp))
2474 node = container_of(kvm->arch.active_mmu_pages.next,
2475 struct kvm_mmu_page, link);
2076 spin_unlock(&kvm->mmu_lock); 2476 spin_unlock(&kvm->mmu_lock);
2077 2477
2078 kvm_flush_remote_tlbs(kvm); 2478 kvm_flush_remote_tlbs(kvm);
@@ -2287,18 +2687,18 @@ int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
2287 gpa_t addr, unsigned long *ret) 2687 gpa_t addr, unsigned long *ret)
2288{ 2688{
2289 int r; 2689 int r;
2290 struct kvm_pv_mmu_op_buffer buffer; 2690 struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer;
2291 2691
2292 buffer.ptr = buffer.buf; 2692 buffer->ptr = buffer->buf;
2293 buffer.len = min_t(unsigned long, bytes, sizeof buffer.buf); 2693 buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf);
2294 buffer.processed = 0; 2694 buffer->processed = 0;
2295 2695
2296 r = kvm_read_guest(vcpu->kvm, addr, buffer.buf, buffer.len); 2696 r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len);
2297 if (r) 2697 if (r)
2298 goto out; 2698 goto out;
2299 2699
2300 while (buffer.len) { 2700 while (buffer->len) {
2301 r = kvm_pv_mmu_op_one(vcpu, &buffer); 2701 r = kvm_pv_mmu_op_one(vcpu, buffer);
2302 if (r < 0) 2702 if (r < 0)
2303 goto out; 2703 goto out;
2304 if (r == 0) 2704 if (r == 0)
@@ -2307,7 +2707,7 @@ int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
2307 2707
2308 r = 1; 2708 r = 1;
2309out: 2709out:
2310 *ret = buffer.processed; 2710 *ret = buffer->processed;
2311 return r; 2711 return r;
2312} 2712}
2313 2713
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 4a814bff21f2..613ec9aa674a 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -25,11 +25,11 @@
25#if PTTYPE == 64 25#if PTTYPE == 64
26 #define pt_element_t u64 26 #define pt_element_t u64
27 #define guest_walker guest_walker64 27 #define guest_walker guest_walker64
28 #define shadow_walker shadow_walker64
28 #define FNAME(name) paging##64_##name 29 #define FNAME(name) paging##64_##name
29 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK 30 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
30 #define PT_DIR_BASE_ADDR_MASK PT64_DIR_BASE_ADDR_MASK 31 #define PT_DIR_BASE_ADDR_MASK PT64_DIR_BASE_ADDR_MASK
31 #define PT_INDEX(addr, level) PT64_INDEX(addr, level) 32 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
32 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
33 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level) 33 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
34 #define PT_LEVEL_BITS PT64_LEVEL_BITS 34 #define PT_LEVEL_BITS PT64_LEVEL_BITS
35 #ifdef CONFIG_X86_64 35 #ifdef CONFIG_X86_64
@@ -42,11 +42,11 @@
42#elif PTTYPE == 32 42#elif PTTYPE == 32
43 #define pt_element_t u32 43 #define pt_element_t u32
44 #define guest_walker guest_walker32 44 #define guest_walker guest_walker32
45 #define shadow_walker shadow_walker32
45 #define FNAME(name) paging##32_##name 46 #define FNAME(name) paging##32_##name
46 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK 47 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
47 #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK 48 #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK
48 #define PT_INDEX(addr, level) PT32_INDEX(addr, level) 49 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
49 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
50 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level) 50 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
51 #define PT_LEVEL_BITS PT32_LEVEL_BITS 51 #define PT_LEVEL_BITS PT32_LEVEL_BITS
52 #define PT_MAX_FULL_LEVELS 2 52 #define PT_MAX_FULL_LEVELS 2
@@ -73,6 +73,17 @@ struct guest_walker {
73 u32 error_code; 73 u32 error_code;
74}; 74};
75 75
76struct shadow_walker {
77 struct kvm_shadow_walk walker;
78 struct guest_walker *guest_walker;
79 int user_fault;
80 int write_fault;
81 int largepage;
82 int *ptwrite;
83 pfn_t pfn;
84 u64 *sptep;
85};
86
76static gfn_t gpte_to_gfn(pt_element_t gpte) 87static gfn_t gpte_to_gfn(pt_element_t gpte)
77{ 88{
78 return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT; 89 return (gpte & PT_BASE_ADDR_MASK) >> PAGE_SHIFT;
@@ -91,14 +102,10 @@ static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
91 pt_element_t *table; 102 pt_element_t *table;
92 struct page *page; 103 struct page *page;
93 104
94 down_read(&current->mm->mmap_sem);
95 page = gfn_to_page(kvm, table_gfn); 105 page = gfn_to_page(kvm, table_gfn);
96 up_read(&current->mm->mmap_sem);
97 106
98 table = kmap_atomic(page, KM_USER0); 107 table = kmap_atomic(page, KM_USER0);
99
100 ret = CMPXCHG(&table[index], orig_pte, new_pte); 108 ret = CMPXCHG(&table[index], orig_pte, new_pte);
101
102 kunmap_atomic(table, KM_USER0); 109 kunmap_atomic(table, KM_USER0);
103 110
104 kvm_release_page_dirty(page); 111 kvm_release_page_dirty(page);
@@ -274,86 +281,89 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
274/* 281/*
275 * Fetch a shadow pte for a specific level in the paging hierarchy. 282 * Fetch a shadow pte for a specific level in the paging hierarchy.
276 */ 283 */
277static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, 284static int FNAME(shadow_walk_entry)(struct kvm_shadow_walk *_sw,
278 struct guest_walker *walker, 285 struct kvm_vcpu *vcpu, u64 addr,
279 int user_fault, int write_fault, int largepage, 286 u64 *sptep, int level)
280 int *ptwrite, pfn_t pfn)
281{ 287{
282 hpa_t shadow_addr; 288 struct shadow_walker *sw =
283 int level; 289 container_of(_sw, struct shadow_walker, walker);
284 u64 *shadow_ent; 290 struct guest_walker *gw = sw->guest_walker;
285 unsigned access = walker->pt_access; 291 unsigned access = gw->pt_access;
286 292 struct kvm_mmu_page *shadow_page;
287 if (!is_present_pte(walker->ptes[walker->level - 1])) 293 u64 spte;
288 return NULL; 294 int metaphysical;
289 295 gfn_t table_gfn;
290 shadow_addr = vcpu->arch.mmu.root_hpa; 296 int r;
291 level = vcpu->arch.mmu.shadow_root_level; 297 pt_element_t curr_pte;
292 if (level == PT32E_ROOT_LEVEL) { 298
293 shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; 299 if (level == PT_PAGE_TABLE_LEVEL
294 shadow_addr &= PT64_BASE_ADDR_MASK; 300 || (sw->largepage && level == PT_DIRECTORY_LEVEL)) {
295 --level; 301 mmu_set_spte(vcpu, sptep, access, gw->pte_access & access,
302 sw->user_fault, sw->write_fault,
303 gw->ptes[gw->level-1] & PT_DIRTY_MASK,
304 sw->ptwrite, sw->largepage, gw->gfn, sw->pfn,
305 false);
306 sw->sptep = sptep;
307 return 1;
296 } 308 }
297 309
298 for (; ; level--) { 310 if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep))
299 u32 index = SHADOW_PT_INDEX(addr, level); 311 return 0;
300 struct kvm_mmu_page *shadow_page;
301 u64 shadow_pte;
302 int metaphysical;
303 gfn_t table_gfn;
304
305 shadow_ent = ((u64 *)__va(shadow_addr)) + index;
306 if (level == PT_PAGE_TABLE_LEVEL)
307 break;
308
309 if (largepage && level == PT_DIRECTORY_LEVEL)
310 break;
311 312
312 if (is_shadow_present_pte(*shadow_ent) 313 if (is_large_pte(*sptep)) {
313 && !is_large_pte(*shadow_ent)) { 314 set_shadow_pte(sptep, shadow_trap_nonpresent_pte);
314 shadow_addr = *shadow_ent & PT64_BASE_ADDR_MASK; 315 kvm_flush_remote_tlbs(vcpu->kvm);
315 continue; 316 rmap_remove(vcpu->kvm, sptep);
316 } 317 }
317 318
318 if (is_large_pte(*shadow_ent)) 319 if (level == PT_DIRECTORY_LEVEL && gw->level == PT_DIRECTORY_LEVEL) {
319 rmap_remove(vcpu->kvm, shadow_ent); 320 metaphysical = 1;
320 321 if (!is_dirty_pte(gw->ptes[level - 1]))
321 if (level - 1 == PT_PAGE_TABLE_LEVEL 322 access &= ~ACC_WRITE_MASK;
322 && walker->level == PT_DIRECTORY_LEVEL) { 323 table_gfn = gpte_to_gfn(gw->ptes[level - 1]);
323 metaphysical = 1; 324 } else {
324 if (!is_dirty_pte(walker->ptes[level - 1])) 325 metaphysical = 0;
325 access &= ~ACC_WRITE_MASK; 326 table_gfn = gw->table_gfn[level - 2];
326 table_gfn = gpte_to_gfn(walker->ptes[level - 1]); 327 }
327 } else { 328 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, (gva_t)addr, level-1,
328 metaphysical = 0; 329 metaphysical, access, sptep);
329 table_gfn = walker->table_gfn[level - 2]; 330 if (!metaphysical) {
330 } 331 r = kvm_read_guest_atomic(vcpu->kvm, gw->pte_gpa[level - 2],
331 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1, 332 &curr_pte, sizeof(curr_pte));
332 metaphysical, access, 333 if (r || curr_pte != gw->ptes[level - 2]) {
333 shadow_ent); 334 kvm_release_pfn_clean(sw->pfn);
334 if (!metaphysical) { 335 sw->sptep = NULL;
335 int r; 336 return 1;
336 pt_element_t curr_pte;
337 r = kvm_read_guest_atomic(vcpu->kvm,
338 walker->pte_gpa[level - 2],
339 &curr_pte, sizeof(curr_pte));
340 if (r || curr_pte != walker->ptes[level - 2]) {
341 kvm_release_pfn_clean(pfn);
342 return NULL;
343 }
344 } 337 }
345 shadow_addr = __pa(shadow_page->spt);
346 shadow_pte = shadow_addr | PT_PRESENT_MASK | PT_ACCESSED_MASK
347 | PT_WRITABLE_MASK | PT_USER_MASK;
348 set_shadow_pte(shadow_ent, shadow_pte);
349 } 338 }
350 339
351 mmu_set_spte(vcpu, shadow_ent, access, walker->pte_access & access, 340 spte = __pa(shadow_page->spt) | PT_PRESENT_MASK | PT_ACCESSED_MASK
352 user_fault, write_fault, 341 | PT_WRITABLE_MASK | PT_USER_MASK;
353 walker->ptes[walker->level-1] & PT_DIRTY_MASK, 342 *sptep = spte;
354 ptwrite, largepage, walker->gfn, pfn, false); 343 return 0;
344}
345
346static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
347 struct guest_walker *guest_walker,
348 int user_fault, int write_fault, int largepage,
349 int *ptwrite, pfn_t pfn)
350{
351 struct shadow_walker walker = {
352 .walker = { .entry = FNAME(shadow_walk_entry), },
353 .guest_walker = guest_walker,
354 .user_fault = user_fault,
355 .write_fault = write_fault,
356 .largepage = largepage,
357 .ptwrite = ptwrite,
358 .pfn = pfn,
359 };
360
361 if (!is_present_pte(guest_walker->ptes[guest_walker->level - 1]))
362 return NULL;
363
364 walk_shadow(&walker.walker, vcpu, addr);
355 365
356 return shadow_ent; 366 return walker.sptep;
357} 367}
358 368
359/* 369/*
@@ -407,7 +417,6 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
407 return 0; 417 return 0;
408 } 418 }
409 419
410 down_read(&current->mm->mmap_sem);
411 if (walker.level == PT_DIRECTORY_LEVEL) { 420 if (walker.level == PT_DIRECTORY_LEVEL) {
412 gfn_t large_gfn; 421 gfn_t large_gfn;
413 large_gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE-1); 422 large_gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE-1);
@@ -417,9 +426,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
417 } 426 }
418 } 427 }
419 mmu_seq = vcpu->kvm->mmu_notifier_seq; 428 mmu_seq = vcpu->kvm->mmu_notifier_seq;
420 /* implicit mb(), we'll read before PT lock is unlocked */ 429 smp_rmb();
421 pfn = gfn_to_pfn(vcpu->kvm, walker.gfn); 430 pfn = gfn_to_pfn(vcpu->kvm, walker.gfn);
422 up_read(&current->mm->mmap_sem);
423 431
424 /* mmio */ 432 /* mmio */
425 if (is_error_pfn(pfn)) { 433 if (is_error_pfn(pfn)) {
@@ -453,6 +461,31 @@ out_unlock:
453 return 0; 461 return 0;
454} 462}
455 463
464static int FNAME(shadow_invlpg_entry)(struct kvm_shadow_walk *_sw,
465 struct kvm_vcpu *vcpu, u64 addr,
466 u64 *sptep, int level)
467{
468
469 if (level == PT_PAGE_TABLE_LEVEL) {
470 if (is_shadow_present_pte(*sptep))
471 rmap_remove(vcpu->kvm, sptep);
472 set_shadow_pte(sptep, shadow_trap_nonpresent_pte);
473 return 1;
474 }
475 if (!is_shadow_present_pte(*sptep))
476 return 1;
477 return 0;
478}
479
480static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
481{
482 struct shadow_walker walker = {
483 .walker = { .entry = FNAME(shadow_invlpg_entry), },
484 };
485
486 walk_shadow(&walker.walker, vcpu, gva);
487}
488
456static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr) 489static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
457{ 490{
458 struct guest_walker walker; 491 struct guest_walker walker;
@@ -499,12 +532,66 @@ static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
499 } 532 }
500} 533}
501 534
535/*
536 * Using the cached information from sp->gfns is safe because:
537 * - The spte has a reference to the struct page, so the pfn for a given gfn
538 * can't change unless all sptes pointing to it are nuked first.
539 * - Alias changes zap the entire shadow cache.
540 */
541static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
542{
543 int i, offset, nr_present;
544
545 offset = nr_present = 0;
546
547 if (PTTYPE == 32)
548 offset = sp->role.quadrant << PT64_LEVEL_BITS;
549
550 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
551 unsigned pte_access;
552 pt_element_t gpte;
553 gpa_t pte_gpa;
554 gfn_t gfn = sp->gfns[i];
555
556 if (!is_shadow_present_pte(sp->spt[i]))
557 continue;
558
559 pte_gpa = gfn_to_gpa(sp->gfn);
560 pte_gpa += (i+offset) * sizeof(pt_element_t);
561
562 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
563 sizeof(pt_element_t)))
564 return -EINVAL;
565
566 if (gpte_to_gfn(gpte) != gfn || !is_present_pte(gpte) ||
567 !(gpte & PT_ACCESSED_MASK)) {
568 u64 nonpresent;
569
570 rmap_remove(vcpu->kvm, &sp->spt[i]);
571 if (is_present_pte(gpte))
572 nonpresent = shadow_trap_nonpresent_pte;
573 else
574 nonpresent = shadow_notrap_nonpresent_pte;
575 set_shadow_pte(&sp->spt[i], nonpresent);
576 continue;
577 }
578
579 nr_present++;
580 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
581 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
582 is_dirty_pte(gpte), 0, gfn,
583 spte_to_pfn(sp->spt[i]), true, false);
584 }
585
586 return !nr_present;
587}
588
502#undef pt_element_t 589#undef pt_element_t
503#undef guest_walker 590#undef guest_walker
591#undef shadow_walker
504#undef FNAME 592#undef FNAME
505#undef PT_BASE_ADDR_MASK 593#undef PT_BASE_ADDR_MASK
506#undef PT_INDEX 594#undef PT_INDEX
507#undef SHADOW_PT_INDEX
508#undef PT_LEVEL_MASK 595#undef PT_LEVEL_MASK
509#undef PT_DIR_BASE_ADDR_MASK 596#undef PT_DIR_BASE_ADDR_MASK
510#undef PT_LEVEL_BITS 597#undef PT_LEVEL_BITS
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index e2ee264740c7..9c4ce657d963 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -18,6 +18,7 @@
18#include "kvm_svm.h" 18#include "kvm_svm.h"
19#include "irq.h" 19#include "irq.h"
20#include "mmu.h" 20#include "mmu.h"
21#include "kvm_cache_regs.h"
21 22
22#include <linux/module.h> 23#include <linux/module.h>
23#include <linux/kernel.h> 24#include <linux/kernel.h>
@@ -35,10 +36,6 @@ MODULE_LICENSE("GPL");
35#define IOPM_ALLOC_ORDER 2 36#define IOPM_ALLOC_ORDER 2
36#define MSRPM_ALLOC_ORDER 1 37#define MSRPM_ALLOC_ORDER 1
37 38
38#define DB_VECTOR 1
39#define UD_VECTOR 6
40#define GP_VECTOR 13
41
42#define DR7_GD_MASK (1 << 13) 39#define DR7_GD_MASK (1 << 13)
43#define DR6_BD_MASK (1 << 13) 40#define DR6_BD_MASK (1 << 13)
44 41
@@ -47,7 +44,7 @@ MODULE_LICENSE("GPL");
47 44
48#define SVM_FEATURE_NPT (1 << 0) 45#define SVM_FEATURE_NPT (1 << 0)
49#define SVM_FEATURE_LBRV (1 << 1) 46#define SVM_FEATURE_LBRV (1 << 1)
50#define SVM_DEATURE_SVML (1 << 2) 47#define SVM_FEATURE_SVML (1 << 2)
51 48
52#define DEBUGCTL_RESERVED_BITS (~(0x3fULL)) 49#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
53 50
@@ -62,6 +59,7 @@ static int npt = 1;
62module_param(npt, int, S_IRUGO); 59module_param(npt, int, S_IRUGO);
63 60
64static void kvm_reput_irq(struct vcpu_svm *svm); 61static void kvm_reput_irq(struct vcpu_svm *svm);
62static void svm_flush_tlb(struct kvm_vcpu *vcpu);
65 63
66static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) 64static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
67{ 65{
@@ -235,13 +233,11 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
235 printk(KERN_DEBUG "%s: NOP\n", __func__); 233 printk(KERN_DEBUG "%s: NOP\n", __func__);
236 return; 234 return;
237 } 235 }
238 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE) 236 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
239 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n", 237 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
240 __func__, 238 __func__, kvm_rip_read(vcpu), svm->next_rip);
241 svm->vmcb->save.rip,
242 svm->next_rip);
243 239
244 vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip; 240 kvm_rip_write(vcpu, svm->next_rip);
245 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK; 241 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
246 242
247 vcpu->arch.interrupt_window_open = 1; 243 vcpu->arch.interrupt_window_open = 1;
@@ -529,6 +525,7 @@ static void init_vmcb(struct vcpu_svm *svm)
529 (1ULL << INTERCEPT_CPUID) | 525 (1ULL << INTERCEPT_CPUID) |
530 (1ULL << INTERCEPT_INVD) | 526 (1ULL << INTERCEPT_INVD) |
531 (1ULL << INTERCEPT_HLT) | 527 (1ULL << INTERCEPT_HLT) |
528 (1ULL << INTERCEPT_INVLPG) |
532 (1ULL << INTERCEPT_INVLPGA) | 529 (1ULL << INTERCEPT_INVLPGA) |
533 (1ULL << INTERCEPT_IOIO_PROT) | 530 (1ULL << INTERCEPT_IOIO_PROT) |
534 (1ULL << INTERCEPT_MSR_PROT) | 531 (1ULL << INTERCEPT_MSR_PROT) |
@@ -580,6 +577,7 @@ static void init_vmcb(struct vcpu_svm *svm)
580 save->dr7 = 0x400; 577 save->dr7 = 0x400;
581 save->rflags = 2; 578 save->rflags = 2;
582 save->rip = 0x0000fff0; 579 save->rip = 0x0000fff0;
580 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
583 581
584 /* 582 /*
585 * cr0 val on cpu init should be 0x60000010, we enable cpu 583 * cr0 val on cpu init should be 0x60000010, we enable cpu
@@ -592,7 +590,8 @@ static void init_vmcb(struct vcpu_svm *svm)
592 if (npt_enabled) { 590 if (npt_enabled) {
593 /* Setup VMCB for Nested Paging */ 591 /* Setup VMCB for Nested Paging */
594 control->nested_ctl = 1; 592 control->nested_ctl = 1;
595 control->intercept &= ~(1ULL << INTERCEPT_TASK_SWITCH); 593 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
594 (1ULL << INTERCEPT_INVLPG));
596 control->intercept_exceptions &= ~(1 << PF_VECTOR); 595 control->intercept_exceptions &= ~(1 << PF_VECTOR);
597 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK| 596 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
598 INTERCEPT_CR3_MASK); 597 INTERCEPT_CR3_MASK);
@@ -614,10 +613,12 @@ static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
614 init_vmcb(svm); 613 init_vmcb(svm);
615 614
616 if (vcpu->vcpu_id != 0) { 615 if (vcpu->vcpu_id != 0) {
617 svm->vmcb->save.rip = 0; 616 kvm_rip_write(vcpu, 0);
618 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12; 617 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
619 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8; 618 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
620 } 619 }
620 vcpu->arch.regs_avail = ~0;
621 vcpu->arch.regs_dirty = ~0;
621 622
622 return 0; 623 return 0;
623} 624}
@@ -720,23 +721,6 @@ static void svm_vcpu_put(struct kvm_vcpu *vcpu)
720 rdtscll(vcpu->arch.host_tsc); 721 rdtscll(vcpu->arch.host_tsc);
721} 722}
722 723
723static void svm_cache_regs(struct kvm_vcpu *vcpu)
724{
725 struct vcpu_svm *svm = to_svm(vcpu);
726
727 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
728 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
729 vcpu->arch.rip = svm->vmcb->save.rip;
730}
731
732static void svm_decache_regs(struct kvm_vcpu *vcpu)
733{
734 struct vcpu_svm *svm = to_svm(vcpu);
735 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
736 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
737 svm->vmcb->save.rip = vcpu->arch.rip;
738}
739
740static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu) 724static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
741{ 725{
742 return to_svm(vcpu)->vmcb->save.rflags; 726 return to_svm(vcpu)->vmcb->save.rflags;
@@ -878,6 +862,10 @@ set:
878static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) 862static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
879{ 863{
880 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE; 864 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
865 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
866
867 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
868 force_new_asid(vcpu);
881 869
882 vcpu->arch.cr4 = cr4; 870 vcpu->arch.cr4 = cr4;
883 if (!npt_enabled) 871 if (!npt_enabled)
@@ -1027,8 +1015,15 @@ static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1027 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code, 1015 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1028 (u32)fault_address, (u32)(fault_address >> 32), 1016 (u32)fault_address, (u32)(fault_address >> 32),
1029 handler); 1017 handler);
1018 /*
1019 * FIXME: Tis shouldn't be necessary here, but there is a flush
1020 * missing in the MMU code. Until we find this bug, flush the
1021 * complete TLB here on an NPF
1022 */
1023 if (npt_enabled)
1024 svm_flush_tlb(&svm->vcpu);
1030 1025
1031 if (event_injection) 1026 if (!npt_enabled && event_injection)
1032 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address); 1027 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1033 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code); 1028 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1034} 1029}
@@ -1127,14 +1122,14 @@ static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1127 1122
1128static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run) 1123static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1129{ 1124{
1130 svm->next_rip = svm->vmcb->save.rip + 1; 1125 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1131 skip_emulated_instruction(&svm->vcpu); 1126 skip_emulated_instruction(&svm->vcpu);
1132 return kvm_emulate_halt(&svm->vcpu); 1127 return kvm_emulate_halt(&svm->vcpu);
1133} 1128}
1134 1129
1135static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run) 1130static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1136{ 1131{
1137 svm->next_rip = svm->vmcb->save.rip + 3; 1132 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1138 skip_emulated_instruction(&svm->vcpu); 1133 skip_emulated_instruction(&svm->vcpu);
1139 kvm_emulate_hypercall(&svm->vcpu); 1134 kvm_emulate_hypercall(&svm->vcpu);
1140 return 1; 1135 return 1;
@@ -1166,11 +1161,18 @@ static int task_switch_interception(struct vcpu_svm *svm,
1166 1161
1167static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run) 1162static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1168{ 1163{
1169 svm->next_rip = svm->vmcb->save.rip + 2; 1164 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1170 kvm_emulate_cpuid(&svm->vcpu); 1165 kvm_emulate_cpuid(&svm->vcpu);
1171 return 1; 1166 return 1;
1172} 1167}
1173 1168
1169static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1170{
1171 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
1172 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1173 return 1;
1174}
1175
1174static int emulate_on_interception(struct vcpu_svm *svm, 1176static int emulate_on_interception(struct vcpu_svm *svm,
1175 struct kvm_run *kvm_run) 1177 struct kvm_run *kvm_run)
1176{ 1178{
@@ -1261,9 +1263,9 @@ static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1261 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data, 1263 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1262 (u32)(data >> 32), handler); 1264 (u32)(data >> 32), handler);
1263 1265
1264 svm->vmcb->save.rax = data & 0xffffffff; 1266 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
1265 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32; 1267 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1266 svm->next_rip = svm->vmcb->save.rip + 2; 1268 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1267 skip_emulated_instruction(&svm->vcpu); 1269 skip_emulated_instruction(&svm->vcpu);
1268 } 1270 }
1269 return 1; 1271 return 1;
@@ -1347,13 +1349,13 @@ static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1347static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run) 1349static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1348{ 1350{
1349 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX]; 1351 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1350 u64 data = (svm->vmcb->save.rax & -1u) 1352 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
1351 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32); 1353 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1352 1354
1353 KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32), 1355 KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
1354 handler); 1356 handler);
1355 1357
1356 svm->next_rip = svm->vmcb->save.rip + 2; 1358 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1357 if (svm_set_msr(&svm->vcpu, ecx, data)) 1359 if (svm_set_msr(&svm->vcpu, ecx, data))
1358 kvm_inject_gp(&svm->vcpu, 0); 1360 kvm_inject_gp(&svm->vcpu, 0);
1359 else 1361 else
@@ -1424,7 +1426,7 @@ static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1424 [SVM_EXIT_CPUID] = cpuid_interception, 1426 [SVM_EXIT_CPUID] = cpuid_interception,
1425 [SVM_EXIT_INVD] = emulate_on_interception, 1427 [SVM_EXIT_INVD] = emulate_on_interception,
1426 [SVM_EXIT_HLT] = halt_interception, 1428 [SVM_EXIT_HLT] = halt_interception,
1427 [SVM_EXIT_INVLPG] = emulate_on_interception, 1429 [SVM_EXIT_INVLPG] = invlpg_interception,
1428 [SVM_EXIT_INVLPGA] = invalid_op_interception, 1430 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1429 [SVM_EXIT_IOIO] = io_interception, 1431 [SVM_EXIT_IOIO] = io_interception,
1430 [SVM_EXIT_MSR] = msr_interception, 1432 [SVM_EXIT_MSR] = msr_interception,
@@ -1526,6 +1528,7 @@ static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1526 1528
1527 KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler); 1529 KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
1528 1530
1531 ++svm->vcpu.stat.irq_injections;
1529 control = &svm->vmcb->control; 1532 control = &svm->vmcb->control;
1530 control->int_vector = irq; 1533 control->int_vector = irq;
1531 control->int_ctl &= ~V_INTR_PRIO_MASK; 1534 control->int_ctl &= ~V_INTR_PRIO_MASK;
@@ -1704,6 +1707,12 @@ static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
1704 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK; 1707 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
1705} 1708}
1706 1709
1710#ifdef CONFIG_X86_64
1711#define R "r"
1712#else
1713#define R "e"
1714#endif
1715
1707static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 1716static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1708{ 1717{
1709 struct vcpu_svm *svm = to_svm(vcpu); 1718 struct vcpu_svm *svm = to_svm(vcpu);
@@ -1711,6 +1720,10 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1711 u16 gs_selector; 1720 u16 gs_selector;
1712 u16 ldt_selector; 1721 u16 ldt_selector;
1713 1722
1723 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
1724 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
1725 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
1726
1714 pre_svm_run(svm); 1727 pre_svm_run(svm);
1715 1728
1716 sync_lapic_to_cr8(vcpu); 1729 sync_lapic_to_cr8(vcpu);
@@ -1738,19 +1751,14 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1738 local_irq_enable(); 1751 local_irq_enable();
1739 1752
1740 asm volatile ( 1753 asm volatile (
1754 "push %%"R"bp; \n\t"
1755 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
1756 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
1757 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
1758 "mov %c[rsi](%[svm]), %%"R"si \n\t"
1759 "mov %c[rdi](%[svm]), %%"R"di \n\t"
1760 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
1741#ifdef CONFIG_X86_64 1761#ifdef CONFIG_X86_64
1742 "push %%rbp; \n\t"
1743#else
1744 "push %%ebp; \n\t"
1745#endif
1746
1747#ifdef CONFIG_X86_64
1748 "mov %c[rbx](%[svm]), %%rbx \n\t"
1749 "mov %c[rcx](%[svm]), %%rcx \n\t"
1750 "mov %c[rdx](%[svm]), %%rdx \n\t"
1751 "mov %c[rsi](%[svm]), %%rsi \n\t"
1752 "mov %c[rdi](%[svm]), %%rdi \n\t"
1753 "mov %c[rbp](%[svm]), %%rbp \n\t"
1754 "mov %c[r8](%[svm]), %%r8 \n\t" 1762 "mov %c[r8](%[svm]), %%r8 \n\t"
1755 "mov %c[r9](%[svm]), %%r9 \n\t" 1763 "mov %c[r9](%[svm]), %%r9 \n\t"
1756 "mov %c[r10](%[svm]), %%r10 \n\t" 1764 "mov %c[r10](%[svm]), %%r10 \n\t"
@@ -1759,41 +1767,24 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1759 "mov %c[r13](%[svm]), %%r13 \n\t" 1767 "mov %c[r13](%[svm]), %%r13 \n\t"
1760 "mov %c[r14](%[svm]), %%r14 \n\t" 1768 "mov %c[r14](%[svm]), %%r14 \n\t"
1761 "mov %c[r15](%[svm]), %%r15 \n\t" 1769 "mov %c[r15](%[svm]), %%r15 \n\t"
1762#else
1763 "mov %c[rbx](%[svm]), %%ebx \n\t"
1764 "mov %c[rcx](%[svm]), %%ecx \n\t"
1765 "mov %c[rdx](%[svm]), %%edx \n\t"
1766 "mov %c[rsi](%[svm]), %%esi \n\t"
1767 "mov %c[rdi](%[svm]), %%edi \n\t"
1768 "mov %c[rbp](%[svm]), %%ebp \n\t"
1769#endif 1770#endif
1770 1771
1771#ifdef CONFIG_X86_64
1772 /* Enter guest mode */ 1772 /* Enter guest mode */
1773 "push %%rax \n\t" 1773 "push %%"R"ax \n\t"
1774 "mov %c[vmcb](%[svm]), %%rax \n\t" 1774 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
1775 __ex(SVM_VMLOAD) "\n\t" 1775 __ex(SVM_VMLOAD) "\n\t"
1776 __ex(SVM_VMRUN) "\n\t" 1776 __ex(SVM_VMRUN) "\n\t"
1777 __ex(SVM_VMSAVE) "\n\t" 1777 __ex(SVM_VMSAVE) "\n\t"
1778 "pop %%rax \n\t" 1778 "pop %%"R"ax \n\t"
1779#else
1780 /* Enter guest mode */
1781 "push %%eax \n\t"
1782 "mov %c[vmcb](%[svm]), %%eax \n\t"
1783 __ex(SVM_VMLOAD) "\n\t"
1784 __ex(SVM_VMRUN) "\n\t"
1785 __ex(SVM_VMSAVE) "\n\t"
1786 "pop %%eax \n\t"
1787#endif
1788 1779
1789 /* Save guest registers, load host registers */ 1780 /* Save guest registers, load host registers */
1781 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
1782 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
1783 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
1784 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
1785 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
1786 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
1790#ifdef CONFIG_X86_64 1787#ifdef CONFIG_X86_64
1791 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1792 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1793 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1794 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1795 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1796 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1797 "mov %%r8, %c[r8](%[svm]) \n\t" 1788 "mov %%r8, %c[r8](%[svm]) \n\t"
1798 "mov %%r9, %c[r9](%[svm]) \n\t" 1789 "mov %%r9, %c[r9](%[svm]) \n\t"
1799 "mov %%r10, %c[r10](%[svm]) \n\t" 1790 "mov %%r10, %c[r10](%[svm]) \n\t"
@@ -1802,18 +1793,8 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1802 "mov %%r13, %c[r13](%[svm]) \n\t" 1793 "mov %%r13, %c[r13](%[svm]) \n\t"
1803 "mov %%r14, %c[r14](%[svm]) \n\t" 1794 "mov %%r14, %c[r14](%[svm]) \n\t"
1804 "mov %%r15, %c[r15](%[svm]) \n\t" 1795 "mov %%r15, %c[r15](%[svm]) \n\t"
1805
1806 "pop %%rbp; \n\t"
1807#else
1808 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1809 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1810 "mov %%edx, %c[rdx](%[svm]) \n\t"
1811 "mov %%esi, %c[rsi](%[svm]) \n\t"
1812 "mov %%edi, %c[rdi](%[svm]) \n\t"
1813 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1814
1815 "pop %%ebp; \n\t"
1816#endif 1796#endif
1797 "pop %%"R"bp"
1817 : 1798 :
1818 : [svm]"a"(svm), 1799 : [svm]"a"(svm),
1819 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)), 1800 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
@@ -1834,11 +1815,9 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1834 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15])) 1815 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1835#endif 1816#endif
1836 : "cc", "memory" 1817 : "cc", "memory"
1818 , R"bx", R"cx", R"dx", R"si", R"di"
1837#ifdef CONFIG_X86_64 1819#ifdef CONFIG_X86_64
1838 , "rbx", "rcx", "rdx", "rsi", "rdi"
1839 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15" 1820 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1840#else
1841 , "ebx", "ecx", "edx" , "esi", "edi"
1842#endif 1821#endif
1843 ); 1822 );
1844 1823
@@ -1846,6 +1825,9 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1846 load_db_regs(svm->host_db_regs); 1825 load_db_regs(svm->host_db_regs);
1847 1826
1848 vcpu->arch.cr2 = svm->vmcb->save.cr2; 1827 vcpu->arch.cr2 = svm->vmcb->save.cr2;
1828 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
1829 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
1830 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
1849 1831
1850 write_dr6(svm->host_dr6); 1832 write_dr6(svm->host_dr6);
1851 write_dr7(svm->host_dr7); 1833 write_dr7(svm->host_dr7);
@@ -1867,6 +1849,8 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1867 svm->next_rip = 0; 1849 svm->next_rip = 0;
1868} 1850}
1869 1851
1852#undef R
1853
1870static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root) 1854static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1871{ 1855{
1872 struct vcpu_svm *svm = to_svm(vcpu); 1856 struct vcpu_svm *svm = to_svm(vcpu);
@@ -1965,8 +1949,6 @@ static struct kvm_x86_ops svm_x86_ops = {
1965 .set_gdt = svm_set_gdt, 1949 .set_gdt = svm_set_gdt,
1966 .get_dr = svm_get_dr, 1950 .get_dr = svm_get_dr,
1967 .set_dr = svm_set_dr, 1951 .set_dr = svm_set_dr,
1968 .cache_regs = svm_cache_regs,
1969 .decache_regs = svm_decache_regs,
1970 .get_rflags = svm_get_rflags, 1952 .get_rflags = svm_get_rflags,
1971 .set_rflags = svm_set_rflags, 1953 .set_rflags = svm_set_rflags,
1972 1954
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 2a69773e3b26..2643b430d83a 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -26,6 +26,8 @@
26#include <linux/highmem.h> 26#include <linux/highmem.h>
27#include <linux/sched.h> 27#include <linux/sched.h>
28#include <linux/moduleparam.h> 28#include <linux/moduleparam.h>
29#include "kvm_cache_regs.h"
30#include "x86.h"
29 31
30#include <asm/io.h> 32#include <asm/io.h>
31#include <asm/desc.h> 33#include <asm/desc.h>
@@ -47,6 +49,9 @@ module_param(flexpriority_enabled, bool, 0);
47static int enable_ept = 1; 49static int enable_ept = 1;
48module_param(enable_ept, bool, 0); 50module_param(enable_ept, bool, 0);
49 51
52static int emulate_invalid_guest_state = 0;
53module_param(emulate_invalid_guest_state, bool, 0);
54
50struct vmcs { 55struct vmcs {
51 u32 revision_id; 56 u32 revision_id;
52 u32 abort; 57 u32 abort;
@@ -56,6 +61,7 @@ struct vmcs {
56struct vcpu_vmx { 61struct vcpu_vmx {
57 struct kvm_vcpu vcpu; 62 struct kvm_vcpu vcpu;
58 struct list_head local_vcpus_link; 63 struct list_head local_vcpus_link;
64 unsigned long host_rsp;
59 int launched; 65 int launched;
60 u8 fail; 66 u8 fail;
61 u32 idt_vectoring_info; 67 u32 idt_vectoring_info;
@@ -83,6 +89,7 @@ struct vcpu_vmx {
83 } irq; 89 } irq;
84 } rmode; 90 } rmode;
85 int vpid; 91 int vpid;
92 bool emulation_required;
86}; 93};
87 94
88static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu) 95static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
@@ -468,7 +475,7 @@ static void update_exception_bitmap(struct kvm_vcpu *vcpu)
468 if (!vcpu->fpu_active) 475 if (!vcpu->fpu_active)
469 eb |= 1u << NM_VECTOR; 476 eb |= 1u << NM_VECTOR;
470 if (vcpu->guest_debug.enabled) 477 if (vcpu->guest_debug.enabled)
471 eb |= 1u << 1; 478 eb |= 1u << DB_VECTOR;
472 if (vcpu->arch.rmode.active) 479 if (vcpu->arch.rmode.active)
473 eb = ~0; 480 eb = ~0;
474 if (vm_need_ept()) 481 if (vm_need_ept())
@@ -715,9 +722,9 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
715 unsigned long rip; 722 unsigned long rip;
716 u32 interruptibility; 723 u32 interruptibility;
717 724
718 rip = vmcs_readl(GUEST_RIP); 725 rip = kvm_rip_read(vcpu);
719 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN); 726 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
720 vmcs_writel(GUEST_RIP, rip); 727 kvm_rip_write(vcpu, rip);
721 728
722 /* 729 /*
723 * We emulated an instruction, so temporary interrupt blocking 730 * We emulated an instruction, so temporary interrupt blocking
@@ -733,19 +740,35 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
733static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr, 740static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
734 bool has_error_code, u32 error_code) 741 bool has_error_code, u32 error_code)
735{ 742{
743 struct vcpu_vmx *vmx = to_vmx(vcpu);
744
745 if (has_error_code)
746 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
747
748 if (vcpu->arch.rmode.active) {
749 vmx->rmode.irq.pending = true;
750 vmx->rmode.irq.vector = nr;
751 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
752 if (nr == BP_VECTOR)
753 vmx->rmode.irq.rip++;
754 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
755 nr | INTR_TYPE_SOFT_INTR
756 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
757 | INTR_INFO_VALID_MASK);
758 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
759 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
760 return;
761 }
762
736 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 763 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
737 nr | INTR_TYPE_EXCEPTION 764 nr | INTR_TYPE_EXCEPTION
738 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0) 765 | (has_error_code ? INTR_INFO_DELIVER_CODE_MASK : 0)
739 | INTR_INFO_VALID_MASK); 766 | INTR_INFO_VALID_MASK);
740 if (has_error_code)
741 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
742} 767}
743 768
744static bool vmx_exception_injected(struct kvm_vcpu *vcpu) 769static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
745{ 770{
746 struct vcpu_vmx *vmx = to_vmx(vcpu); 771 return false;
747
748 return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
749} 772}
750 773
751/* 774/*
@@ -947,24 +970,19 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
947 return ret; 970 return ret;
948} 971}
949 972
950/* 973static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
951 * Sync the rsp and rip registers into the vcpu structure. This allows
952 * registers to be accessed by indexing vcpu->arch.regs.
953 */
954static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
955{
956 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
957 vcpu->arch.rip = vmcs_readl(GUEST_RIP);
958}
959
960/*
961 * Syncs rsp and rip back into the vmcs. Should be called after possible
962 * modification.
963 */
964static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
965{ 974{
966 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]); 975 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
967 vmcs_writel(GUEST_RIP, vcpu->arch.rip); 976 switch (reg) {
977 case VCPU_REGS_RSP:
978 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
979 break;
980 case VCPU_REGS_RIP:
981 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
982 break;
983 default:
984 break;
985 }
968} 986}
969 987
970static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg) 988static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
@@ -1007,17 +1025,9 @@ static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
1007 1025
1008static int vmx_get_irq(struct kvm_vcpu *vcpu) 1026static int vmx_get_irq(struct kvm_vcpu *vcpu)
1009{ 1027{
1010 struct vcpu_vmx *vmx = to_vmx(vcpu); 1028 if (!vcpu->arch.interrupt.pending)
1011 u32 idtv_info_field; 1029 return -1;
1012 1030 return vcpu->arch.interrupt.nr;
1013 idtv_info_field = vmx->idt_vectoring_info;
1014 if (idtv_info_field & INTR_INFO_VALID_MASK) {
1015 if (is_external_interrupt(idtv_info_field))
1016 return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
1017 else
1018 printk(KERN_DEBUG "pending exception: not handled yet\n");
1019 }
1020 return -1;
1021} 1031}
1022 1032
1023static __init int cpu_has_kvm_support(void) 1033static __init int cpu_has_kvm_support(void)
@@ -1031,9 +1041,9 @@ static __init int vmx_disabled_by_bios(void)
1031 u64 msr; 1041 u64 msr;
1032 1042
1033 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr); 1043 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1034 return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED | 1044 return (msr & (FEATURE_CONTROL_LOCKED |
1035 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED)) 1045 FEATURE_CONTROL_VMXON_ENABLED))
1036 == MSR_IA32_FEATURE_CONTROL_LOCKED; 1046 == FEATURE_CONTROL_LOCKED;
1037 /* locked but not enabled */ 1047 /* locked but not enabled */
1038} 1048}
1039 1049
@@ -1045,14 +1055,14 @@ static void hardware_enable(void *garbage)
1045 1055
1046 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu)); 1056 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1047 rdmsrl(MSR_IA32_FEATURE_CONTROL, old); 1057 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1048 if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED | 1058 if ((old & (FEATURE_CONTROL_LOCKED |
1049 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED)) 1059 FEATURE_CONTROL_VMXON_ENABLED))
1050 != (MSR_IA32_FEATURE_CONTROL_LOCKED | 1060 != (FEATURE_CONTROL_LOCKED |
1051 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED)) 1061 FEATURE_CONTROL_VMXON_ENABLED))
1052 /* enable and lock */ 1062 /* enable and lock */
1053 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 1063 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1054 MSR_IA32_FEATURE_CONTROL_LOCKED | 1064 FEATURE_CONTROL_LOCKED |
1055 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED); 1065 FEATURE_CONTROL_VMXON_ENABLED);
1056 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */ 1066 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1057 asm volatile (ASM_VMX_VMXON_RAX 1067 asm volatile (ASM_VMX_VMXON_RAX
1058 : : "a"(&phys_addr), "m"(phys_addr) 1068 : : "a"(&phys_addr), "m"(phys_addr)
@@ -1120,7 +1130,8 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1120 CPU_BASED_CR3_STORE_EXITING | 1130 CPU_BASED_CR3_STORE_EXITING |
1121 CPU_BASED_USE_IO_BITMAPS | 1131 CPU_BASED_USE_IO_BITMAPS |
1122 CPU_BASED_MOV_DR_EXITING | 1132 CPU_BASED_MOV_DR_EXITING |
1123 CPU_BASED_USE_TSC_OFFSETING; 1133 CPU_BASED_USE_TSC_OFFSETING |
1134 CPU_BASED_INVLPG_EXITING;
1124 opt = CPU_BASED_TPR_SHADOW | 1135 opt = CPU_BASED_TPR_SHADOW |
1125 CPU_BASED_USE_MSR_BITMAPS | 1136 CPU_BASED_USE_MSR_BITMAPS |
1126 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; 1137 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
@@ -1149,9 +1160,11 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1149 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW; 1160 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1150#endif 1161#endif
1151 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) { 1162 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1152 /* CR3 accesses don't need to cause VM Exits when EPT enabled */ 1163 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1164 enabled */
1153 min &= ~(CPU_BASED_CR3_LOAD_EXITING | 1165 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1154 CPU_BASED_CR3_STORE_EXITING); 1166 CPU_BASED_CR3_STORE_EXITING |
1167 CPU_BASED_INVLPG_EXITING);
1155 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS, 1168 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1156 &_cpu_based_exec_control) < 0) 1169 &_cpu_based_exec_control) < 0)
1157 return -EIO; 1170 return -EIO;
@@ -1288,7 +1301,9 @@ static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1288static void enter_pmode(struct kvm_vcpu *vcpu) 1301static void enter_pmode(struct kvm_vcpu *vcpu)
1289{ 1302{
1290 unsigned long flags; 1303 unsigned long flags;
1304 struct vcpu_vmx *vmx = to_vmx(vcpu);
1291 1305
1306 vmx->emulation_required = 1;
1292 vcpu->arch.rmode.active = 0; 1307 vcpu->arch.rmode.active = 0;
1293 1308
1294 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base); 1309 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
@@ -1305,6 +1320,9 @@ static void enter_pmode(struct kvm_vcpu *vcpu)
1305 1320
1306 update_exception_bitmap(vcpu); 1321 update_exception_bitmap(vcpu);
1307 1322
1323 if (emulate_invalid_guest_state)
1324 return;
1325
1308 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es); 1326 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1309 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds); 1327 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1310 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs); 1328 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
@@ -1345,7 +1363,9 @@ static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1345static void enter_rmode(struct kvm_vcpu *vcpu) 1363static void enter_rmode(struct kvm_vcpu *vcpu)
1346{ 1364{
1347 unsigned long flags; 1365 unsigned long flags;
1366 struct vcpu_vmx *vmx = to_vmx(vcpu);
1348 1367
1368 vmx->emulation_required = 1;
1349 vcpu->arch.rmode.active = 1; 1369 vcpu->arch.rmode.active = 1;
1350 1370
1351 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE); 1371 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
@@ -1367,6 +1387,9 @@ static void enter_rmode(struct kvm_vcpu *vcpu)
1367 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME); 1387 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1368 update_exception_bitmap(vcpu); 1388 update_exception_bitmap(vcpu);
1369 1389
1390 if (emulate_invalid_guest_state)
1391 goto continue_rmode;
1392
1370 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4); 1393 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1371 vmcs_write32(GUEST_SS_LIMIT, 0xffff); 1394 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1372 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3); 1395 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
@@ -1382,6 +1405,7 @@ static void enter_rmode(struct kvm_vcpu *vcpu)
1382 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs); 1405 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1383 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs); 1406 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1384 1407
1408continue_rmode:
1385 kvm_mmu_reset_context(vcpu); 1409 kvm_mmu_reset_context(vcpu);
1386 init_rmode(vcpu->kvm); 1410 init_rmode(vcpu->kvm);
1387} 1411}
@@ -1715,6 +1739,186 @@ static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1715 vmcs_writel(GUEST_GDTR_BASE, dt->base); 1739 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1716} 1740}
1717 1741
1742static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1743{
1744 struct kvm_segment var;
1745 u32 ar;
1746
1747 vmx_get_segment(vcpu, &var, seg);
1748 ar = vmx_segment_access_rights(&var);
1749
1750 if (var.base != (var.selector << 4))
1751 return false;
1752 if (var.limit != 0xffff)
1753 return false;
1754 if (ar != 0xf3)
1755 return false;
1756
1757 return true;
1758}
1759
1760static bool code_segment_valid(struct kvm_vcpu *vcpu)
1761{
1762 struct kvm_segment cs;
1763 unsigned int cs_rpl;
1764
1765 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1766 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1767
1768 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1769 return false;
1770 if (!cs.s)
1771 return false;
1772 if (!(~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK))) {
1773 if (cs.dpl > cs_rpl)
1774 return false;
1775 } else if (cs.type & AR_TYPE_CODE_MASK) {
1776 if (cs.dpl != cs_rpl)
1777 return false;
1778 }
1779 if (!cs.present)
1780 return false;
1781
1782 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1783 return true;
1784}
1785
1786static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1787{
1788 struct kvm_segment ss;
1789 unsigned int ss_rpl;
1790
1791 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1792 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1793
1794 if ((ss.type != 3) || (ss.type != 7))
1795 return false;
1796 if (!ss.s)
1797 return false;
1798 if (ss.dpl != ss_rpl) /* DPL != RPL */
1799 return false;
1800 if (!ss.present)
1801 return false;
1802
1803 return true;
1804}
1805
1806static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1807{
1808 struct kvm_segment var;
1809 unsigned int rpl;
1810
1811 vmx_get_segment(vcpu, &var, seg);
1812 rpl = var.selector & SELECTOR_RPL_MASK;
1813
1814 if (!var.s)
1815 return false;
1816 if (!var.present)
1817 return false;
1818 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1819 if (var.dpl < rpl) /* DPL < RPL */
1820 return false;
1821 }
1822
1823 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1824 * rights flags
1825 */
1826 return true;
1827}
1828
1829static bool tr_valid(struct kvm_vcpu *vcpu)
1830{
1831 struct kvm_segment tr;
1832
1833 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1834
1835 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1836 return false;
1837 if ((tr.type != 3) || (tr.type != 11)) /* TODO: Check if guest is in IA32e mode */
1838 return false;
1839 if (!tr.present)
1840 return false;
1841
1842 return true;
1843}
1844
1845static bool ldtr_valid(struct kvm_vcpu *vcpu)
1846{
1847 struct kvm_segment ldtr;
1848
1849 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1850
1851 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1852 return false;
1853 if (ldtr.type != 2)
1854 return false;
1855 if (!ldtr.present)
1856 return false;
1857
1858 return true;
1859}
1860
1861static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
1862{
1863 struct kvm_segment cs, ss;
1864
1865 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1866 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1867
1868 return ((cs.selector & SELECTOR_RPL_MASK) ==
1869 (ss.selector & SELECTOR_RPL_MASK));
1870}
1871
1872/*
1873 * Check if guest state is valid. Returns true if valid, false if
1874 * not.
1875 * We assume that registers are always usable
1876 */
1877static bool guest_state_valid(struct kvm_vcpu *vcpu)
1878{
1879 /* real mode guest state checks */
1880 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
1881 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
1882 return false;
1883 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
1884 return false;
1885 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
1886 return false;
1887 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
1888 return false;
1889 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
1890 return false;
1891 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
1892 return false;
1893 } else {
1894 /* protected mode guest state checks */
1895 if (!cs_ss_rpl_check(vcpu))
1896 return false;
1897 if (!code_segment_valid(vcpu))
1898 return false;
1899 if (!stack_segment_valid(vcpu))
1900 return false;
1901 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
1902 return false;
1903 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
1904 return false;
1905 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
1906 return false;
1907 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
1908 return false;
1909 if (!tr_valid(vcpu))
1910 return false;
1911 if (!ldtr_valid(vcpu))
1912 return false;
1913 }
1914 /* TODO:
1915 * - Add checks on RIP
1916 * - Add checks on RFLAGS
1917 */
1918
1919 return true;
1920}
1921
1718static int init_rmode_tss(struct kvm *kvm) 1922static int init_rmode_tss(struct kvm *kvm)
1719{ 1923{
1720 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT; 1924 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
@@ -1726,7 +1930,8 @@ static int init_rmode_tss(struct kvm *kvm)
1726 if (r < 0) 1930 if (r < 0)
1727 goto out; 1931 goto out;
1728 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE; 1932 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1729 r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16)); 1933 r = kvm_write_guest_page(kvm, fn++, &data,
1934 TSS_IOPB_BASE_OFFSET, sizeof(u16));
1730 if (r < 0) 1935 if (r < 0)
1731 goto out; 1936 goto out;
1732 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE); 1937 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
@@ -1789,7 +1994,7 @@ static void seg_setup(int seg)
1789 vmcs_write16(sf->selector, 0); 1994 vmcs_write16(sf->selector, 0);
1790 vmcs_writel(sf->base, 0); 1995 vmcs_writel(sf->base, 0);
1791 vmcs_write32(sf->limit, 0xffff); 1996 vmcs_write32(sf->limit, 0xffff);
1792 vmcs_write32(sf->ar_bytes, 0x93); 1997 vmcs_write32(sf->ar_bytes, 0xf3);
1793} 1998}
1794 1999
1795static int alloc_apic_access_page(struct kvm *kvm) 2000static int alloc_apic_access_page(struct kvm *kvm)
@@ -1808,9 +2013,7 @@ static int alloc_apic_access_page(struct kvm *kvm)
1808 if (r) 2013 if (r)
1809 goto out; 2014 goto out;
1810 2015
1811 down_read(&current->mm->mmap_sem);
1812 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00); 2016 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
1813 up_read(&current->mm->mmap_sem);
1814out: 2017out:
1815 up_write(&kvm->slots_lock); 2018 up_write(&kvm->slots_lock);
1816 return r; 2019 return r;
@@ -1832,10 +2035,8 @@ static int alloc_identity_pagetable(struct kvm *kvm)
1832 if (r) 2035 if (r)
1833 goto out; 2036 goto out;
1834 2037
1835 down_read(&current->mm->mmap_sem);
1836 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm, 2038 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
1837 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT); 2039 VMX_EPT_IDENTITY_PAGETABLE_ADDR >> PAGE_SHIFT);
1838 up_read(&current->mm->mmap_sem);
1839out: 2040out:
1840 up_write(&kvm->slots_lock); 2041 up_write(&kvm->slots_lock);
1841 return r; 2042 return r;
@@ -1917,7 +2118,8 @@ static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1917 } 2118 }
1918 if (!vm_need_ept()) 2119 if (!vm_need_ept())
1919 exec_control |= CPU_BASED_CR3_STORE_EXITING | 2120 exec_control |= CPU_BASED_CR3_STORE_EXITING |
1920 CPU_BASED_CR3_LOAD_EXITING; 2121 CPU_BASED_CR3_LOAD_EXITING |
2122 CPU_BASED_INVLPG_EXITING;
1921 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control); 2123 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1922 2124
1923 if (cpu_has_secondary_exec_ctrls()) { 2125 if (cpu_has_secondary_exec_ctrls()) {
@@ -2019,6 +2221,7 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2019 u64 msr; 2221 u64 msr;
2020 int ret; 2222 int ret;
2021 2223
2224 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2022 down_read(&vcpu->kvm->slots_lock); 2225 down_read(&vcpu->kvm->slots_lock);
2023 if (!init_rmode(vmx->vcpu.kvm)) { 2226 if (!init_rmode(vmx->vcpu.kvm)) {
2024 ret = -ENOMEM; 2227 ret = -ENOMEM;
@@ -2036,6 +2239,7 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2036 2239
2037 fx_init(&vmx->vcpu); 2240 fx_init(&vmx->vcpu);
2038 2241
2242 seg_setup(VCPU_SREG_CS);
2039 /* 2243 /*
2040 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode 2244 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2041 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh. 2245 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
@@ -2047,8 +2251,6 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2047 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8); 2251 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2048 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12); 2252 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2049 } 2253 }
2050 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2051 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2052 2254
2053 seg_setup(VCPU_SREG_DS); 2255 seg_setup(VCPU_SREG_DS);
2054 seg_setup(VCPU_SREG_ES); 2256 seg_setup(VCPU_SREG_ES);
@@ -2072,10 +2274,10 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2072 2274
2073 vmcs_writel(GUEST_RFLAGS, 0x02); 2275 vmcs_writel(GUEST_RFLAGS, 0x02);
2074 if (vmx->vcpu.vcpu_id == 0) 2276 if (vmx->vcpu.vcpu_id == 0)
2075 vmcs_writel(GUEST_RIP, 0xfff0); 2277 kvm_rip_write(vcpu, 0xfff0);
2076 else 2278 else
2077 vmcs_writel(GUEST_RIP, 0); 2279 kvm_rip_write(vcpu, 0);
2078 vmcs_writel(GUEST_RSP, 0); 2280 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2079 2281
2080 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */ 2282 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
2081 vmcs_writel(GUEST_DR7, 0x400); 2283 vmcs_writel(GUEST_DR7, 0x400);
@@ -2125,6 +2327,9 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2125 2327
2126 ret = 0; 2328 ret = 0;
2127 2329
2330 /* HACK: Don't enable emulation on guest boot/reset */
2331 vmx->emulation_required = 0;
2332
2128out: 2333out:
2129 up_read(&vcpu->kvm->slots_lock); 2334 up_read(&vcpu->kvm->slots_lock);
2130 return ret; 2335 return ret;
@@ -2136,14 +2341,15 @@ static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
2136 2341
2137 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler); 2342 KVMTRACE_1D(INJ_VIRQ, vcpu, (u32)irq, handler);
2138 2343
2344 ++vcpu->stat.irq_injections;
2139 if (vcpu->arch.rmode.active) { 2345 if (vcpu->arch.rmode.active) {
2140 vmx->rmode.irq.pending = true; 2346 vmx->rmode.irq.pending = true;
2141 vmx->rmode.irq.vector = irq; 2347 vmx->rmode.irq.vector = irq;
2142 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP); 2348 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2143 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 2349 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2144 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK); 2350 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2145 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1); 2351 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2146 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1); 2352 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2147 return; 2353 return;
2148 } 2354 }
2149 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 2355 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
@@ -2154,7 +2360,6 @@ static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2154{ 2360{
2155 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 2361 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2156 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR); 2362 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2157 vcpu->arch.nmi_pending = 0;
2158} 2363}
2159 2364
2160static void kvm_do_inject_irq(struct kvm_vcpu *vcpu) 2365static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
@@ -2166,7 +2371,7 @@ static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
2166 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]); 2371 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2167 if (!vcpu->arch.irq_pending[word_index]) 2372 if (!vcpu->arch.irq_pending[word_index])
2168 clear_bit(word_index, &vcpu->arch.irq_summary); 2373 clear_bit(word_index, &vcpu->arch.irq_summary);
2169 vmx_inject_irq(vcpu, irq); 2374 kvm_queue_interrupt(vcpu, irq);
2170} 2375}
2171 2376
2172 2377
@@ -2180,13 +2385,12 @@ static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2180 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0); 2385 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2181 2386
2182 if (vcpu->arch.interrupt_window_open && 2387 if (vcpu->arch.interrupt_window_open &&
2183 vcpu->arch.irq_summary && 2388 vcpu->arch.irq_summary && !vcpu->arch.interrupt.pending)
2184 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
2185 /*
2186 * If interrupts enabled, and not blocked by sti or mov ss. Good.
2187 */
2188 kvm_do_inject_irq(vcpu); 2389 kvm_do_inject_irq(vcpu);
2189 2390
2391 if (vcpu->arch.interrupt_window_open && vcpu->arch.interrupt.pending)
2392 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
2393
2190 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); 2394 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2191 if (!vcpu->arch.interrupt_window_open && 2395 if (!vcpu->arch.interrupt_window_open &&
2192 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window)) 2396 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
@@ -2237,9 +2441,6 @@ static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
2237static int handle_rmode_exception(struct kvm_vcpu *vcpu, 2441static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2238 int vec, u32 err_code) 2442 int vec, u32 err_code)
2239{ 2443{
2240 if (!vcpu->arch.rmode.active)
2241 return 0;
2242
2243 /* 2444 /*
2244 * Instruction with address size override prefix opcode 0x67 2445 * Instruction with address size override prefix opcode 0x67
2245 * Cause the #SS fault with 0 error code in VM86 mode. 2446 * Cause the #SS fault with 0 error code in VM86 mode.
@@ -2247,6 +2448,25 @@ static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2247 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) 2448 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2248 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE) 2449 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2249 return 1; 2450 return 1;
2451 /*
2452 * Forward all other exceptions that are valid in real mode.
2453 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2454 * the required debugging infrastructure rework.
2455 */
2456 switch (vec) {
2457 case DE_VECTOR:
2458 case DB_VECTOR:
2459 case BP_VECTOR:
2460 case OF_VECTOR:
2461 case BR_VECTOR:
2462 case UD_VECTOR:
2463 case DF_VECTOR:
2464 case SS_VECTOR:
2465 case GP_VECTOR:
2466 case MF_VECTOR:
2467 kvm_queue_exception(vcpu, vec);
2468 return 1;
2469 }
2250 return 0; 2470 return 0;
2251} 2471}
2252 2472
@@ -2288,7 +2508,7 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2288 } 2508 }
2289 2509
2290 error_code = 0; 2510 error_code = 0;
2291 rip = vmcs_readl(GUEST_RIP); 2511 rip = kvm_rip_read(vcpu);
2292 if (intr_info & INTR_INFO_DELIVER_CODE_MASK) 2512 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2293 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE); 2513 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2294 if (is_page_fault(intr_info)) { 2514 if (is_page_fault(intr_info)) {
@@ -2298,7 +2518,7 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2298 cr2 = vmcs_readl(EXIT_QUALIFICATION); 2518 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2299 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2, 2519 KVMTRACE_3D(PAGE_FAULT, vcpu, error_code, (u32)cr2,
2300 (u32)((u64)cr2 >> 32), handler); 2520 (u32)((u64)cr2 >> 32), handler);
2301 if (vect_info & VECTORING_INFO_VALID_MASK) 2521 if (vcpu->arch.interrupt.pending || vcpu->arch.exception.pending)
2302 kvm_mmu_unprotect_page_virt(vcpu, cr2); 2522 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2303 return kvm_mmu_page_fault(vcpu, cr2, error_code); 2523 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2304 } 2524 }
@@ -2386,27 +2606,25 @@ static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2386 reg = (exit_qualification >> 8) & 15; 2606 reg = (exit_qualification >> 8) & 15;
2387 switch ((exit_qualification >> 4) & 3) { 2607 switch ((exit_qualification >> 4) & 3) {
2388 case 0: /* mov to cr */ 2608 case 0: /* mov to cr */
2389 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)vcpu->arch.regs[reg], 2609 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr,
2390 (u32)((u64)vcpu->arch.regs[reg] >> 32), handler); 2610 (u32)kvm_register_read(vcpu, reg),
2611 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2612 handler);
2391 switch (cr) { 2613 switch (cr) {
2392 case 0: 2614 case 0:
2393 vcpu_load_rsp_rip(vcpu); 2615 kvm_set_cr0(vcpu, kvm_register_read(vcpu, reg));
2394 kvm_set_cr0(vcpu, vcpu->arch.regs[reg]);
2395 skip_emulated_instruction(vcpu); 2616 skip_emulated_instruction(vcpu);
2396 return 1; 2617 return 1;
2397 case 3: 2618 case 3:
2398 vcpu_load_rsp_rip(vcpu); 2619 kvm_set_cr3(vcpu, kvm_register_read(vcpu, reg));
2399 kvm_set_cr3(vcpu, vcpu->arch.regs[reg]);
2400 skip_emulated_instruction(vcpu); 2620 skip_emulated_instruction(vcpu);
2401 return 1; 2621 return 1;
2402 case 4: 2622 case 4:
2403 vcpu_load_rsp_rip(vcpu); 2623 kvm_set_cr4(vcpu, kvm_register_read(vcpu, reg));
2404 kvm_set_cr4(vcpu, vcpu->arch.regs[reg]);
2405 skip_emulated_instruction(vcpu); 2624 skip_emulated_instruction(vcpu);
2406 return 1; 2625 return 1;
2407 case 8: 2626 case 8:
2408 vcpu_load_rsp_rip(vcpu); 2627 kvm_set_cr8(vcpu, kvm_register_read(vcpu, reg));
2409 kvm_set_cr8(vcpu, vcpu->arch.regs[reg]);
2410 skip_emulated_instruction(vcpu); 2628 skip_emulated_instruction(vcpu);
2411 if (irqchip_in_kernel(vcpu->kvm)) 2629 if (irqchip_in_kernel(vcpu->kvm))
2412 return 1; 2630 return 1;
@@ -2415,7 +2633,6 @@ static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2415 }; 2633 };
2416 break; 2634 break;
2417 case 2: /* clts */ 2635 case 2: /* clts */
2418 vcpu_load_rsp_rip(vcpu);
2419 vmx_fpu_deactivate(vcpu); 2636 vmx_fpu_deactivate(vcpu);
2420 vcpu->arch.cr0 &= ~X86_CR0_TS; 2637 vcpu->arch.cr0 &= ~X86_CR0_TS;
2421 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0); 2638 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
@@ -2426,21 +2643,17 @@ static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2426 case 1: /*mov from cr*/ 2643 case 1: /*mov from cr*/
2427 switch (cr) { 2644 switch (cr) {
2428 case 3: 2645 case 3:
2429 vcpu_load_rsp_rip(vcpu); 2646 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2430 vcpu->arch.regs[reg] = vcpu->arch.cr3;
2431 vcpu_put_rsp_rip(vcpu);
2432 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, 2647 KVMTRACE_3D(CR_READ, vcpu, (u32)cr,
2433 (u32)vcpu->arch.regs[reg], 2648 (u32)kvm_register_read(vcpu, reg),
2434 (u32)((u64)vcpu->arch.regs[reg] >> 32), 2649 (u32)((u64)kvm_register_read(vcpu, reg) >> 32),
2435 handler); 2650 handler);
2436 skip_emulated_instruction(vcpu); 2651 skip_emulated_instruction(vcpu);
2437 return 1; 2652 return 1;
2438 case 8: 2653 case 8:
2439 vcpu_load_rsp_rip(vcpu); 2654 kvm_register_write(vcpu, reg, kvm_get_cr8(vcpu));
2440 vcpu->arch.regs[reg] = kvm_get_cr8(vcpu);
2441 vcpu_put_rsp_rip(vcpu);
2442 KVMTRACE_2D(CR_READ, vcpu, (u32)cr, 2655 KVMTRACE_2D(CR_READ, vcpu, (u32)cr,
2443 (u32)vcpu->arch.regs[reg], handler); 2656 (u32)kvm_register_read(vcpu, reg), handler);
2444 skip_emulated_instruction(vcpu); 2657 skip_emulated_instruction(vcpu);
2445 return 1; 2658 return 1;
2446 } 2659 }
@@ -2472,7 +2685,6 @@ static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2472 exit_qualification = vmcs_readl(EXIT_QUALIFICATION); 2685 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2473 dr = exit_qualification & 7; 2686 dr = exit_qualification & 7;
2474 reg = (exit_qualification >> 8) & 15; 2687 reg = (exit_qualification >> 8) & 15;
2475 vcpu_load_rsp_rip(vcpu);
2476 if (exit_qualification & 16) { 2688 if (exit_qualification & 16) {
2477 /* mov from dr */ 2689 /* mov from dr */
2478 switch (dr) { 2690 switch (dr) {
@@ -2485,12 +2697,11 @@ static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2485 default: 2697 default:
2486 val = 0; 2698 val = 0;
2487 } 2699 }
2488 vcpu->arch.regs[reg] = val; 2700 kvm_register_write(vcpu, reg, val);
2489 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler); 2701 KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
2490 } else { 2702 } else {
2491 /* mov to dr */ 2703 /* mov to dr */
2492 } 2704 }
2493 vcpu_put_rsp_rip(vcpu);
2494 skip_emulated_instruction(vcpu); 2705 skip_emulated_instruction(vcpu);
2495 return 1; 2706 return 1;
2496} 2707}
@@ -2583,6 +2794,15 @@ static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2583 return 1; 2794 return 1;
2584} 2795}
2585 2796
2797static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2798{
2799 u64 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2800
2801 kvm_mmu_invlpg(vcpu, exit_qualification);
2802 skip_emulated_instruction(vcpu);
2803 return 1;
2804}
2805
2586static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 2806static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2587{ 2807{
2588 skip_emulated_instruction(vcpu); 2808 skip_emulated_instruction(vcpu);
@@ -2695,6 +2915,43 @@ static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2695 return 1; 2915 return 1;
2696} 2916}
2697 2917
2918static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
2919 struct kvm_run *kvm_run)
2920{
2921 struct vcpu_vmx *vmx = to_vmx(vcpu);
2922 int err;
2923
2924 preempt_enable();
2925 local_irq_enable();
2926
2927 while (!guest_state_valid(vcpu)) {
2928 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2929
2930 switch (err) {
2931 case EMULATE_DONE:
2932 break;
2933 case EMULATE_DO_MMIO:
2934 kvm_report_emulation_failure(vcpu, "mmio");
2935 /* TODO: Handle MMIO */
2936 return;
2937 default:
2938 kvm_report_emulation_failure(vcpu, "emulation failure");
2939 return;
2940 }
2941
2942 if (signal_pending(current))
2943 break;
2944 if (need_resched())
2945 schedule();
2946 }
2947
2948 local_irq_disable();
2949 preempt_disable();
2950
2951 /* Guest state should be valid now, no more emulation should be needed */
2952 vmx->emulation_required = 0;
2953}
2954
2698/* 2955/*
2699 * The exit handlers return 1 if the exit was handled fully and guest execution 2956 * The exit handlers return 1 if the exit was handled fully and guest execution
2700 * may resume. Otherwise they set the kvm_run parameter to indicate what needs 2957 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
@@ -2714,6 +2971,7 @@ static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2714 [EXIT_REASON_MSR_WRITE] = handle_wrmsr, 2971 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
2715 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window, 2972 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
2716 [EXIT_REASON_HLT] = handle_halt, 2973 [EXIT_REASON_HLT] = handle_halt,
2974 [EXIT_REASON_INVLPG] = handle_invlpg,
2717 [EXIT_REASON_VMCALL] = handle_vmcall, 2975 [EXIT_REASON_VMCALL] = handle_vmcall,
2718 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold, 2976 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
2719 [EXIT_REASON_APIC_ACCESS] = handle_apic_access, 2977 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
@@ -2735,8 +2993,8 @@ static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2735 struct vcpu_vmx *vmx = to_vmx(vcpu); 2993 struct vcpu_vmx *vmx = to_vmx(vcpu);
2736 u32 vectoring_info = vmx->idt_vectoring_info; 2994 u32 vectoring_info = vmx->idt_vectoring_info;
2737 2995
2738 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)vmcs_readl(GUEST_RIP), 2996 KVMTRACE_3D(VMEXIT, vcpu, exit_reason, (u32)kvm_rip_read(vcpu),
2739 (u32)((u64)vmcs_readl(GUEST_RIP) >> 32), entryexit); 2997 (u32)((u64)kvm_rip_read(vcpu) >> 32), entryexit);
2740 2998
2741 /* Access CR3 don't cause VMExit in paging mode, so we need 2999 /* Access CR3 don't cause VMExit in paging mode, so we need
2742 * to sync with guest real CR3. */ 3000 * to sync with guest real CR3. */
@@ -2829,88 +3087,92 @@ static void enable_intr_window(struct kvm_vcpu *vcpu)
2829 enable_irq_window(vcpu); 3087 enable_irq_window(vcpu);
2830} 3088}
2831 3089
2832static void vmx_intr_assist(struct kvm_vcpu *vcpu) 3090static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
2833{ 3091{
2834 struct vcpu_vmx *vmx = to_vmx(vcpu); 3092 u32 exit_intr_info;
2835 u32 idtv_info_field, intr_info_field, exit_intr_info_field; 3093 u32 idt_vectoring_info;
2836 int vector; 3094 bool unblock_nmi;
3095 u8 vector;
3096 int type;
3097 bool idtv_info_valid;
3098 u32 error;
2837 3099
2838 update_tpr_threshold(vcpu); 3100 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2839 3101 if (cpu_has_virtual_nmis()) {
2840 intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD); 3102 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
2841 exit_intr_info_field = vmcs_read32(VM_EXIT_INTR_INFO); 3103 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
2842 idtv_info_field = vmx->idt_vectoring_info; 3104 /*
2843 if (intr_info_field & INTR_INFO_VALID_MASK) { 3105 * SDM 3: 25.7.1.2
2844 if (idtv_info_field & INTR_INFO_VALID_MASK) { 3106 * Re-set bit "block by NMI" before VM entry if vmexit caused by
2845 /* TODO: fault when IDT_Vectoring */ 3107 * a guest IRET fault.
2846 if (printk_ratelimit()) 3108 */
2847 printk(KERN_ERR "Fault when IDT_Vectoring\n"); 3109 if (unblock_nmi && vector != DF_VECTOR)
2848 } 3110 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2849 enable_intr_window(vcpu); 3111 GUEST_INTR_STATE_NMI);
2850 return;
2851 } 3112 }
2852 if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2853 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2854 == INTR_TYPE_EXT_INTR
2855 && vcpu->arch.rmode.active) {
2856 u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2857
2858 vmx_inject_irq(vcpu, vect);
2859 enable_intr_window(vcpu);
2860 return;
2861 }
2862
2863 KVMTRACE_1D(REDELIVER_EVT, vcpu, idtv_info_field, handler);
2864 3113
3114 idt_vectoring_info = vmx->idt_vectoring_info;
3115 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3116 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3117 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3118 if (vmx->vcpu.arch.nmi_injected) {
2865 /* 3119 /*
2866 * SDM 3: 25.7.1.2 3120 * SDM 3: 25.7.1.2
2867 * Clear bit "block by NMI" before VM entry if a NMI delivery 3121 * Clear bit "block by NMI" before VM entry if a NMI delivery
2868 * faulted. 3122 * faulted.
2869 */ 3123 */
2870 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK) 3124 if (idtv_info_valid && type == INTR_TYPE_NMI_INTR)
2871 == INTR_TYPE_NMI_INTR && cpu_has_virtual_nmis()) 3125 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2872 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 3126 GUEST_INTR_STATE_NMI);
2873 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3127 else
2874 ~GUEST_INTR_STATE_NMI); 3128 vmx->vcpu.arch.nmi_injected = false;
2875 3129 }
2876 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field 3130 kvm_clear_exception_queue(&vmx->vcpu);
2877 & ~INTR_INFO_RESVD_BITS_MASK); 3131 if (idtv_info_valid && type == INTR_TYPE_EXCEPTION) {
2878 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 3132 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
2879 vmcs_read32(VM_EXIT_INSTRUCTION_LEN)); 3133 error = vmcs_read32(IDT_VECTORING_ERROR_CODE);
2880 3134 kvm_queue_exception_e(&vmx->vcpu, vector, error);
2881 if (unlikely(idtv_info_field & INTR_INFO_DELIVER_CODE_MASK)) 3135 } else
2882 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 3136 kvm_queue_exception(&vmx->vcpu, vector);
2883 vmcs_read32(IDT_VECTORING_ERROR_CODE)); 3137 vmx->idt_vectoring_info = 0;
2884 enable_intr_window(vcpu);
2885 return;
2886 } 3138 }
3139 kvm_clear_interrupt_queue(&vmx->vcpu);
3140 if (idtv_info_valid && type == INTR_TYPE_EXT_INTR) {
3141 kvm_queue_interrupt(&vmx->vcpu, vector);
3142 vmx->idt_vectoring_info = 0;
3143 }
3144}
3145
3146static void vmx_intr_assist(struct kvm_vcpu *vcpu)
3147{
3148 update_tpr_threshold(vcpu);
3149
2887 if (cpu_has_virtual_nmis()) { 3150 if (cpu_has_virtual_nmis()) {
2888 /* 3151 if (vcpu->arch.nmi_pending && !vcpu->arch.nmi_injected) {
2889 * SDM 3: 25.7.1.2 3152 if (vmx_nmi_enabled(vcpu)) {
2890 * Re-set bit "block by NMI" before VM entry if vmexit caused by 3153 vcpu->arch.nmi_pending = false;
2891 * a guest IRET fault. 3154 vcpu->arch.nmi_injected = true;
2892 */ 3155 } else {
2893 if ((exit_intr_info_field & INTR_INFO_UNBLOCK_NMI) && 3156 enable_intr_window(vcpu);
2894 (exit_intr_info_field & INTR_INFO_VECTOR_MASK) != 8) 3157 return;
2895 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 3158 }
2896 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) | 3159 }
2897 GUEST_INTR_STATE_NMI); 3160 if (vcpu->arch.nmi_injected) {
2898 else if (vcpu->arch.nmi_pending) { 3161 vmx_inject_nmi(vcpu);
2899 if (vmx_nmi_enabled(vcpu))
2900 vmx_inject_nmi(vcpu);
2901 enable_intr_window(vcpu); 3162 enable_intr_window(vcpu);
2902 return; 3163 return;
2903 } 3164 }
2904
2905 } 3165 }
2906 if (!kvm_cpu_has_interrupt(vcpu)) 3166 if (!vcpu->arch.interrupt.pending && kvm_cpu_has_interrupt(vcpu)) {
2907 return; 3167 if (vmx_irq_enabled(vcpu))
2908 if (vmx_irq_enabled(vcpu)) { 3168 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
2909 vector = kvm_cpu_get_interrupt(vcpu); 3169 else
2910 vmx_inject_irq(vcpu, vector); 3170 enable_irq_window(vcpu);
2911 kvm_timer_intr_post(vcpu, vector); 3171 }
2912 } else 3172 if (vcpu->arch.interrupt.pending) {
2913 enable_irq_window(vcpu); 3173 vmx_inject_irq(vcpu, vcpu->arch.interrupt.nr);
3174 kvm_timer_intr_post(vcpu, vcpu->arch.interrupt.nr);
3175 }
2914} 3176}
2915 3177
2916/* 3178/*
@@ -2922,9 +3184,9 @@ static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2922static void fixup_rmode_irq(struct vcpu_vmx *vmx) 3184static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2923{ 3185{
2924 vmx->rmode.irq.pending = 0; 3186 vmx->rmode.irq.pending = 0;
2925 if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip) 3187 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
2926 return; 3188 return;
2927 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip); 3189 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
2928 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) { 3190 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2929 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK; 3191 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2930 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR; 3192 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
@@ -2936,11 +3198,30 @@ static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2936 | vmx->rmode.irq.vector; 3198 | vmx->rmode.irq.vector;
2937} 3199}
2938 3200
3201#ifdef CONFIG_X86_64
3202#define R "r"
3203#define Q "q"
3204#else
3205#define R "e"
3206#define Q "l"
3207#endif
3208
2939static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 3209static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2940{ 3210{
2941 struct vcpu_vmx *vmx = to_vmx(vcpu); 3211 struct vcpu_vmx *vmx = to_vmx(vcpu);
2942 u32 intr_info; 3212 u32 intr_info;
2943 3213
3214 /* Handle invalid guest state instead of entering VMX */
3215 if (vmx->emulation_required && emulate_invalid_guest_state) {
3216 handle_invalid_guest_state(vcpu, kvm_run);
3217 return;
3218 }
3219
3220 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3221 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3222 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3223 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3224
2944 /* 3225 /*
2945 * Loading guest fpu may have cleared host cr0.ts 3226 * Loading guest fpu may have cleared host cr0.ts
2946 */ 3227 */
@@ -2948,26 +3229,25 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2948 3229
2949 asm( 3230 asm(
2950 /* Store host registers */ 3231 /* Store host registers */
2951#ifdef CONFIG_X86_64 3232 "push %%"R"dx; push %%"R"bp;"
2952 "push %%rdx; push %%rbp;" 3233 "push %%"R"cx \n\t"
2953 "push %%rcx \n\t" 3234 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
2954#else 3235 "je 1f \n\t"
2955 "push %%edx; push %%ebp;" 3236 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
2956 "push %%ecx \n\t"
2957#endif
2958 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t" 3237 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3238 "1: \n\t"
2959 /* Check if vmlaunch of vmresume is needed */ 3239 /* Check if vmlaunch of vmresume is needed */
2960 "cmpl $0, %c[launched](%0) \n\t" 3240 "cmpl $0, %c[launched](%0) \n\t"
2961 /* Load guest registers. Don't clobber flags. */ 3241 /* Load guest registers. Don't clobber flags. */
3242 "mov %c[cr2](%0), %%"R"ax \n\t"
3243 "mov %%"R"ax, %%cr2 \n\t"
3244 "mov %c[rax](%0), %%"R"ax \n\t"
3245 "mov %c[rbx](%0), %%"R"bx \n\t"
3246 "mov %c[rdx](%0), %%"R"dx \n\t"
3247 "mov %c[rsi](%0), %%"R"si \n\t"
3248 "mov %c[rdi](%0), %%"R"di \n\t"
3249 "mov %c[rbp](%0), %%"R"bp \n\t"
2962#ifdef CONFIG_X86_64 3250#ifdef CONFIG_X86_64
2963 "mov %c[cr2](%0), %%rax \n\t"
2964 "mov %%rax, %%cr2 \n\t"
2965 "mov %c[rax](%0), %%rax \n\t"
2966 "mov %c[rbx](%0), %%rbx \n\t"
2967 "mov %c[rdx](%0), %%rdx \n\t"
2968 "mov %c[rsi](%0), %%rsi \n\t"
2969 "mov %c[rdi](%0), %%rdi \n\t"
2970 "mov %c[rbp](%0), %%rbp \n\t"
2971 "mov %c[r8](%0), %%r8 \n\t" 3251 "mov %c[r8](%0), %%r8 \n\t"
2972 "mov %c[r9](%0), %%r9 \n\t" 3252 "mov %c[r9](%0), %%r9 \n\t"
2973 "mov %c[r10](%0), %%r10 \n\t" 3253 "mov %c[r10](%0), %%r10 \n\t"
@@ -2976,18 +3256,9 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2976 "mov %c[r13](%0), %%r13 \n\t" 3256 "mov %c[r13](%0), %%r13 \n\t"
2977 "mov %c[r14](%0), %%r14 \n\t" 3257 "mov %c[r14](%0), %%r14 \n\t"
2978 "mov %c[r15](%0), %%r15 \n\t" 3258 "mov %c[r15](%0), %%r15 \n\t"
2979 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
2980#else
2981 "mov %c[cr2](%0), %%eax \n\t"
2982 "mov %%eax, %%cr2 \n\t"
2983 "mov %c[rax](%0), %%eax \n\t"
2984 "mov %c[rbx](%0), %%ebx \n\t"
2985 "mov %c[rdx](%0), %%edx \n\t"
2986 "mov %c[rsi](%0), %%esi \n\t"
2987 "mov %c[rdi](%0), %%edi \n\t"
2988 "mov %c[rbp](%0), %%ebp \n\t"
2989 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
2990#endif 3259#endif
3260 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3261
2991 /* Enter guest mode */ 3262 /* Enter guest mode */
2992 "jne .Llaunched \n\t" 3263 "jne .Llaunched \n\t"
2993 __ex(ASM_VMX_VMLAUNCH) "\n\t" 3264 __ex(ASM_VMX_VMLAUNCH) "\n\t"
@@ -2995,15 +3266,15 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2995 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t" 3266 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
2996 ".Lkvm_vmx_return: " 3267 ".Lkvm_vmx_return: "
2997 /* Save guest registers, load host registers, keep flags */ 3268 /* Save guest registers, load host registers, keep flags */
3269 "xchg %0, (%%"R"sp) \n\t"
3270 "mov %%"R"ax, %c[rax](%0) \n\t"
3271 "mov %%"R"bx, %c[rbx](%0) \n\t"
3272 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3273 "mov %%"R"dx, %c[rdx](%0) \n\t"
3274 "mov %%"R"si, %c[rsi](%0) \n\t"
3275 "mov %%"R"di, %c[rdi](%0) \n\t"
3276 "mov %%"R"bp, %c[rbp](%0) \n\t"
2998#ifdef CONFIG_X86_64 3277#ifdef CONFIG_X86_64
2999 "xchg %0, (%%rsp) \n\t"
3000 "mov %%rax, %c[rax](%0) \n\t"
3001 "mov %%rbx, %c[rbx](%0) \n\t"
3002 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
3003 "mov %%rdx, %c[rdx](%0) \n\t"
3004 "mov %%rsi, %c[rsi](%0) \n\t"
3005 "mov %%rdi, %c[rdi](%0) \n\t"
3006 "mov %%rbp, %c[rbp](%0) \n\t"
3007 "mov %%r8, %c[r8](%0) \n\t" 3278 "mov %%r8, %c[r8](%0) \n\t"
3008 "mov %%r9, %c[r9](%0) \n\t" 3279 "mov %%r9, %c[r9](%0) \n\t"
3009 "mov %%r10, %c[r10](%0) \n\t" 3280 "mov %%r10, %c[r10](%0) \n\t"
@@ -3012,28 +3283,16 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3012 "mov %%r13, %c[r13](%0) \n\t" 3283 "mov %%r13, %c[r13](%0) \n\t"
3013 "mov %%r14, %c[r14](%0) \n\t" 3284 "mov %%r14, %c[r14](%0) \n\t"
3014 "mov %%r15, %c[r15](%0) \n\t" 3285 "mov %%r15, %c[r15](%0) \n\t"
3015 "mov %%cr2, %%rax \n\t"
3016 "mov %%rax, %c[cr2](%0) \n\t"
3017
3018 "pop %%rbp; pop %%rbp; pop %%rdx \n\t"
3019#else
3020 "xchg %0, (%%esp) \n\t"
3021 "mov %%eax, %c[rax](%0) \n\t"
3022 "mov %%ebx, %c[rbx](%0) \n\t"
3023 "pushl (%%esp); popl %c[rcx](%0) \n\t"
3024 "mov %%edx, %c[rdx](%0) \n\t"
3025 "mov %%esi, %c[rsi](%0) \n\t"
3026 "mov %%edi, %c[rdi](%0) \n\t"
3027 "mov %%ebp, %c[rbp](%0) \n\t"
3028 "mov %%cr2, %%eax \n\t"
3029 "mov %%eax, %c[cr2](%0) \n\t"
3030
3031 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
3032#endif 3286#endif
3287 "mov %%cr2, %%"R"ax \n\t"
3288 "mov %%"R"ax, %c[cr2](%0) \n\t"
3289
3290 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3033 "setbe %c[fail](%0) \n\t" 3291 "setbe %c[fail](%0) \n\t"
3034 : : "c"(vmx), "d"((unsigned long)HOST_RSP), 3292 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3035 [launched]"i"(offsetof(struct vcpu_vmx, launched)), 3293 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3036 [fail]"i"(offsetof(struct vcpu_vmx, fail)), 3294 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3295 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3037 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])), 3296 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3038 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])), 3297 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3039 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])), 3298 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
@@ -3053,14 +3312,15 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3053#endif 3312#endif
3054 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)) 3313 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3055 : "cc", "memory" 3314 : "cc", "memory"
3315 , R"bx", R"di", R"si"
3056#ifdef CONFIG_X86_64 3316#ifdef CONFIG_X86_64
3057 , "rbx", "rdi", "rsi"
3058 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15" 3317 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3059#else
3060 , "ebx", "edi", "rsi"
3061#endif 3318#endif
3062 ); 3319 );
3063 3320
3321 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3322 vcpu->arch.regs_dirty = 0;
3323
3064 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD); 3324 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3065 if (vmx->rmode.irq.pending) 3325 if (vmx->rmode.irq.pending)
3066 fixup_rmode_irq(vmx); 3326 fixup_rmode_irq(vmx);
@@ -3080,8 +3340,13 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3080 KVMTRACE_0D(NMI, vcpu, handler); 3340 KVMTRACE_0D(NMI, vcpu, handler);
3081 asm("int $2"); 3341 asm("int $2");
3082 } 3342 }
3343
3344 vmx_complete_interrupts(vmx);
3083} 3345}
3084 3346
3347#undef R
3348#undef Q
3349
3085static void vmx_free_vmcs(struct kvm_vcpu *vcpu) 3350static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3086{ 3351{
3087 struct vcpu_vmx *vmx = to_vmx(vcpu); 3352 struct vcpu_vmx *vmx = to_vmx(vcpu);
@@ -3224,8 +3489,7 @@ static struct kvm_x86_ops vmx_x86_ops = {
3224 .set_idt = vmx_set_idt, 3489 .set_idt = vmx_set_idt,
3225 .get_gdt = vmx_get_gdt, 3490 .get_gdt = vmx_get_gdt,
3226 .set_gdt = vmx_set_gdt, 3491 .set_gdt = vmx_set_gdt,
3227 .cache_regs = vcpu_load_rsp_rip, 3492 .cache_reg = vmx_cache_reg,
3228 .decache_regs = vcpu_put_rsp_rip,
3229 .get_rflags = vmx_get_rflags, 3493 .get_rflags = vmx_get_rflags,
3230 .set_rflags = vmx_set_rflags, 3494 .set_rflags = vmx_set_rflags,
3231 3495
@@ -3301,8 +3565,7 @@ static int __init vmx_init(void)
3301 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK | 3565 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
3302 VMX_EPT_WRITABLE_MASK | 3566 VMX_EPT_WRITABLE_MASK |
3303 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT); 3567 VMX_EPT_DEFAULT_MT << VMX_EPT_MT_EPTE_SHIFT);
3304 kvm_mmu_set_mask_ptes(0ull, VMX_EPT_FAKE_ACCESSED_MASK, 3568 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
3305 VMX_EPT_FAKE_DIRTY_MASK, 0ull,
3306 VMX_EPT_EXECUTABLE_MASK); 3569 VMX_EPT_EXECUTABLE_MASK);
3307 kvm_enable_tdp(); 3570 kvm_enable_tdp();
3308 } else 3571 } else
diff --git a/arch/x86/kvm/vmx.h b/arch/x86/kvm/vmx.h
index 425a13436b3f..3e010d21fdd7 100644
--- a/arch/x86/kvm/vmx.h
+++ b/arch/x86/kvm/vmx.h
@@ -331,24 +331,6 @@ enum vmcs_field {
331 331
332#define AR_RESERVD_MASK 0xfffe0f00 332#define AR_RESERVD_MASK 0xfffe0f00
333 333
334#define MSR_IA32_VMX_BASIC 0x480
335#define MSR_IA32_VMX_PINBASED_CTLS 0x481
336#define MSR_IA32_VMX_PROCBASED_CTLS 0x482
337#define MSR_IA32_VMX_EXIT_CTLS 0x483
338#define MSR_IA32_VMX_ENTRY_CTLS 0x484
339#define MSR_IA32_VMX_MISC 0x485
340#define MSR_IA32_VMX_CR0_FIXED0 0x486
341#define MSR_IA32_VMX_CR0_FIXED1 0x487
342#define MSR_IA32_VMX_CR4_FIXED0 0x488
343#define MSR_IA32_VMX_CR4_FIXED1 0x489
344#define MSR_IA32_VMX_VMCS_ENUM 0x48a
345#define MSR_IA32_VMX_PROCBASED_CTLS2 0x48b
346#define MSR_IA32_VMX_EPT_VPID_CAP 0x48c
347
348#define MSR_IA32_FEATURE_CONTROL 0x3a
349#define MSR_IA32_FEATURE_CONTROL_LOCKED 0x1
350#define MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED 0x4
351
352#define APIC_ACCESS_PAGE_PRIVATE_MEMSLOT 9 334#define APIC_ACCESS_PAGE_PRIVATE_MEMSLOT 9
353#define IDENTITY_PAGETABLE_PRIVATE_MEMSLOT 10 335#define IDENTITY_PAGETABLE_PRIVATE_MEMSLOT 10
354 336
@@ -370,8 +352,6 @@ enum vmcs_field {
370#define VMX_EPT_READABLE_MASK 0x1ull 352#define VMX_EPT_READABLE_MASK 0x1ull
371#define VMX_EPT_WRITABLE_MASK 0x2ull 353#define VMX_EPT_WRITABLE_MASK 0x2ull
372#define VMX_EPT_EXECUTABLE_MASK 0x4ull 354#define VMX_EPT_EXECUTABLE_MASK 0x4ull
373#define VMX_EPT_FAKE_ACCESSED_MASK (1ull << 62)
374#define VMX_EPT_FAKE_DIRTY_MASK (1ull << 63)
375 355
376#define VMX_EPT_IDENTITY_PAGETABLE_ADDR 0xfffbc000ul 356#define VMX_EPT_IDENTITY_PAGETABLE_ADDR 0xfffbc000ul
377 357
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 0d682fc6aeb3..4f0677d1eae8 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -4,10 +4,14 @@
4 * derived from drivers/kvm/kvm_main.c 4 * derived from drivers/kvm/kvm_main.c
5 * 5 *
6 * Copyright (C) 2006 Qumranet, Inc. 6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
7 * 9 *
8 * Authors: 10 * Authors:
9 * Avi Kivity <avi@qumranet.com> 11 * Avi Kivity <avi@qumranet.com>
10 * Yaniv Kamay <yaniv@qumranet.com> 12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
11 * 15 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See 16 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory. 17 * the COPYING file in the top-level directory.
@@ -19,14 +23,18 @@
19#include "mmu.h" 23#include "mmu.h"
20#include "i8254.h" 24#include "i8254.h"
21#include "tss.h" 25#include "tss.h"
26#include "kvm_cache_regs.h"
27#include "x86.h"
22 28
23#include <linux/clocksource.h> 29#include <linux/clocksource.h>
30#include <linux/interrupt.h>
24#include <linux/kvm.h> 31#include <linux/kvm.h>
25#include <linux/fs.h> 32#include <linux/fs.h>
26#include <linux/vmalloc.h> 33#include <linux/vmalloc.h>
27#include <linux/module.h> 34#include <linux/module.h>
28#include <linux/mman.h> 35#include <linux/mman.h>
29#include <linux/highmem.h> 36#include <linux/highmem.h>
37#include <linux/intel-iommu.h>
30 38
31#include <asm/uaccess.h> 39#include <asm/uaccess.h>
32#include <asm/msr.h> 40#include <asm/msr.h>
@@ -61,6 +69,7 @@ static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
61 struct kvm_cpuid_entry2 __user *entries); 69 struct kvm_cpuid_entry2 __user *entries);
62 70
63struct kvm_x86_ops *kvm_x86_ops; 71struct kvm_x86_ops *kvm_x86_ops;
72EXPORT_SYMBOL_GPL(kvm_x86_ops);
64 73
65struct kvm_stats_debugfs_item debugfs_entries[] = { 74struct kvm_stats_debugfs_item debugfs_entries[] = {
66 { "pf_fixed", VCPU_STAT(pf_fixed) }, 75 { "pf_fixed", VCPU_STAT(pf_fixed) },
@@ -83,6 +92,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
83 { "fpu_reload", VCPU_STAT(fpu_reload) }, 92 { "fpu_reload", VCPU_STAT(fpu_reload) },
84 { "insn_emulation", VCPU_STAT(insn_emulation) }, 93 { "insn_emulation", VCPU_STAT(insn_emulation) },
85 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, 94 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
95 { "irq_injections", VCPU_STAT(irq_injections) },
86 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, 96 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
87 { "mmu_pte_write", VM_STAT(mmu_pte_write) }, 97 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
88 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, 98 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
@@ -90,12 +100,12 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
90 { "mmu_flooded", VM_STAT(mmu_flooded) }, 100 { "mmu_flooded", VM_STAT(mmu_flooded) },
91 { "mmu_recycled", VM_STAT(mmu_recycled) }, 101 { "mmu_recycled", VM_STAT(mmu_recycled) },
92 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, 102 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
103 { "mmu_unsync", VM_STAT(mmu_unsync) },
93 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, 104 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
94 { "largepages", VM_STAT(lpages) }, 105 { "largepages", VM_STAT(lpages) },
95 { NULL } 106 { NULL }
96}; 107};
97 108
98
99unsigned long segment_base(u16 selector) 109unsigned long segment_base(u16 selector)
100{ 110{
101 struct descriptor_table gdt; 111 struct descriptor_table gdt;
@@ -352,6 +362,7 @@ EXPORT_SYMBOL_GPL(kvm_set_cr4);
352void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) 362void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
353{ 363{
354 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) { 364 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
365 kvm_mmu_sync_roots(vcpu);
355 kvm_mmu_flush_tlb(vcpu); 366 kvm_mmu_flush_tlb(vcpu);
356 return; 367 return;
357 } 368 }
@@ -564,7 +575,7 @@ static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *
564 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32); 575 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
565 576
566 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n", 577 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
567 __FUNCTION__, tsc_khz, hv_clock->tsc_shift, 578 __func__, tsc_khz, hv_clock->tsc_shift,
568 hv_clock->tsc_to_system_mul); 579 hv_clock->tsc_to_system_mul);
569} 580}
570 581
@@ -662,6 +673,18 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
662 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n", 673 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
663 __func__, data); 674 __func__, data);
664 break; 675 break;
676 case MSR_IA32_DEBUGCTLMSR:
677 if (!data) {
678 /* We support the non-activated case already */
679 break;
680 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
681 /* Values other than LBR and BTF are vendor-specific,
682 thus reserved and should throw a #GP */
683 return 1;
684 }
685 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
686 __func__, data);
687 break;
665 case MSR_IA32_UCODE_REV: 688 case MSR_IA32_UCODE_REV:
666 case MSR_IA32_UCODE_WRITE: 689 case MSR_IA32_UCODE_WRITE:
667 break; 690 break;
@@ -692,10 +715,8 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
692 /* ...but clean it before doing the actual write */ 715 /* ...but clean it before doing the actual write */
693 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1); 716 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
694 717
695 down_read(&current->mm->mmap_sem);
696 vcpu->arch.time_page = 718 vcpu->arch.time_page =
697 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT); 719 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
698 up_read(&current->mm->mmap_sem);
699 720
700 if (is_error_page(vcpu->arch.time_page)) { 721 if (is_error_page(vcpu->arch.time_page)) {
701 kvm_release_page_clean(vcpu->arch.time_page); 722 kvm_release_page_clean(vcpu->arch.time_page);
@@ -752,8 +773,14 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
752 case MSR_IA32_MC0_MISC+8: 773 case MSR_IA32_MC0_MISC+8:
753 case MSR_IA32_MC0_MISC+12: 774 case MSR_IA32_MC0_MISC+12:
754 case MSR_IA32_MC0_MISC+16: 775 case MSR_IA32_MC0_MISC+16:
776 case MSR_IA32_MC0_MISC+20:
755 case MSR_IA32_UCODE_REV: 777 case MSR_IA32_UCODE_REV:
756 case MSR_IA32_EBL_CR_POWERON: 778 case MSR_IA32_EBL_CR_POWERON:
779 case MSR_IA32_DEBUGCTLMSR:
780 case MSR_IA32_LASTBRANCHFROMIP:
781 case MSR_IA32_LASTBRANCHTOIP:
782 case MSR_IA32_LASTINTFROMIP:
783 case MSR_IA32_LASTINTTOIP:
757 data = 0; 784 data = 0;
758 break; 785 break;
759 case MSR_MTRRcap: 786 case MSR_MTRRcap:
@@ -901,6 +928,9 @@ int kvm_dev_ioctl_check_extension(long ext)
901 case KVM_CAP_PV_MMU: 928 case KVM_CAP_PV_MMU:
902 r = !tdp_enabled; 929 r = !tdp_enabled;
903 break; 930 break;
931 case KVM_CAP_IOMMU:
932 r = intel_iommu_found();
933 break;
904 default: 934 default:
905 r = 0; 935 r = 0;
906 break; 936 break;
@@ -1303,28 +1333,33 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
1303 struct kvm_vcpu *vcpu = filp->private_data; 1333 struct kvm_vcpu *vcpu = filp->private_data;
1304 void __user *argp = (void __user *)arg; 1334 void __user *argp = (void __user *)arg;
1305 int r; 1335 int r;
1336 struct kvm_lapic_state *lapic = NULL;
1306 1337
1307 switch (ioctl) { 1338 switch (ioctl) {
1308 case KVM_GET_LAPIC: { 1339 case KVM_GET_LAPIC: {
1309 struct kvm_lapic_state lapic; 1340 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1310 1341
1311 memset(&lapic, 0, sizeof lapic); 1342 r = -ENOMEM;
1312 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic); 1343 if (!lapic)
1344 goto out;
1345 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1313 if (r) 1346 if (r)
1314 goto out; 1347 goto out;
1315 r = -EFAULT; 1348 r = -EFAULT;
1316 if (copy_to_user(argp, &lapic, sizeof lapic)) 1349 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1317 goto out; 1350 goto out;
1318 r = 0; 1351 r = 0;
1319 break; 1352 break;
1320 } 1353 }
1321 case KVM_SET_LAPIC: { 1354 case KVM_SET_LAPIC: {
1322 struct kvm_lapic_state lapic; 1355 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1323 1356 r = -ENOMEM;
1357 if (!lapic)
1358 goto out;
1324 r = -EFAULT; 1359 r = -EFAULT;
1325 if (copy_from_user(&lapic, argp, sizeof lapic)) 1360 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1326 goto out; 1361 goto out;
1327 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);; 1362 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1328 if (r) 1363 if (r)
1329 goto out; 1364 goto out;
1330 r = 0; 1365 r = 0;
@@ -1422,6 +1457,8 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
1422 r = -EINVAL; 1457 r = -EINVAL;
1423 } 1458 }
1424out: 1459out:
1460 if (lapic)
1461 kfree(lapic);
1425 return r; 1462 return r;
1426} 1463}
1427 1464
@@ -1630,6 +1667,15 @@ long kvm_arch_vm_ioctl(struct file *filp,
1630 struct kvm *kvm = filp->private_data; 1667 struct kvm *kvm = filp->private_data;
1631 void __user *argp = (void __user *)arg; 1668 void __user *argp = (void __user *)arg;
1632 int r = -EINVAL; 1669 int r = -EINVAL;
1670 /*
1671 * This union makes it completely explicit to gcc-3.x
1672 * that these two variables' stack usage should be
1673 * combined, not added together.
1674 */
1675 union {
1676 struct kvm_pit_state ps;
1677 struct kvm_memory_alias alias;
1678 } u;
1633 1679
1634 switch (ioctl) { 1680 switch (ioctl) {
1635 case KVM_SET_TSS_ADDR: 1681 case KVM_SET_TSS_ADDR:
@@ -1661,17 +1707,14 @@ long kvm_arch_vm_ioctl(struct file *filp,
1661 case KVM_GET_NR_MMU_PAGES: 1707 case KVM_GET_NR_MMU_PAGES:
1662 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); 1708 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1663 break; 1709 break;
1664 case KVM_SET_MEMORY_ALIAS: { 1710 case KVM_SET_MEMORY_ALIAS:
1665 struct kvm_memory_alias alias;
1666
1667 r = -EFAULT; 1711 r = -EFAULT;
1668 if (copy_from_user(&alias, argp, sizeof alias)) 1712 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1669 goto out; 1713 goto out;
1670 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias); 1714 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1671 if (r) 1715 if (r)
1672 goto out; 1716 goto out;
1673 break; 1717 break;
1674 }
1675 case KVM_CREATE_IRQCHIP: 1718 case KVM_CREATE_IRQCHIP:
1676 r = -ENOMEM; 1719 r = -ENOMEM;
1677 kvm->arch.vpic = kvm_create_pic(kvm); 1720 kvm->arch.vpic = kvm_create_pic(kvm);
@@ -1699,13 +1742,7 @@ long kvm_arch_vm_ioctl(struct file *filp,
1699 goto out; 1742 goto out;
1700 if (irqchip_in_kernel(kvm)) { 1743 if (irqchip_in_kernel(kvm)) {
1701 mutex_lock(&kvm->lock); 1744 mutex_lock(&kvm->lock);
1702 if (irq_event.irq < 16) 1745 kvm_set_irq(kvm, irq_event.irq, irq_event.level);
1703 kvm_pic_set_irq(pic_irqchip(kvm),
1704 irq_event.irq,
1705 irq_event.level);
1706 kvm_ioapic_set_irq(kvm->arch.vioapic,
1707 irq_event.irq,
1708 irq_event.level);
1709 mutex_unlock(&kvm->lock); 1746 mutex_unlock(&kvm->lock);
1710 r = 0; 1747 r = 0;
1711 } 1748 }
@@ -1713,65 +1750,77 @@ long kvm_arch_vm_ioctl(struct file *filp,
1713 } 1750 }
1714 case KVM_GET_IRQCHIP: { 1751 case KVM_GET_IRQCHIP: {
1715 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ 1752 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1716 struct kvm_irqchip chip; 1753 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1717 1754
1718 r = -EFAULT; 1755 r = -ENOMEM;
1719 if (copy_from_user(&chip, argp, sizeof chip)) 1756 if (!chip)
1720 goto out; 1757 goto out;
1758 r = -EFAULT;
1759 if (copy_from_user(chip, argp, sizeof *chip))
1760 goto get_irqchip_out;
1721 r = -ENXIO; 1761 r = -ENXIO;
1722 if (!irqchip_in_kernel(kvm)) 1762 if (!irqchip_in_kernel(kvm))
1723 goto out; 1763 goto get_irqchip_out;
1724 r = kvm_vm_ioctl_get_irqchip(kvm, &chip); 1764 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1725 if (r) 1765 if (r)
1726 goto out; 1766 goto get_irqchip_out;
1727 r = -EFAULT; 1767 r = -EFAULT;
1728 if (copy_to_user(argp, &chip, sizeof chip)) 1768 if (copy_to_user(argp, chip, sizeof *chip))
1729 goto out; 1769 goto get_irqchip_out;
1730 r = 0; 1770 r = 0;
1771 get_irqchip_out:
1772 kfree(chip);
1773 if (r)
1774 goto out;
1731 break; 1775 break;
1732 } 1776 }
1733 case KVM_SET_IRQCHIP: { 1777 case KVM_SET_IRQCHIP: {
1734 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ 1778 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1735 struct kvm_irqchip chip; 1779 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1736 1780
1737 r = -EFAULT; 1781 r = -ENOMEM;
1738 if (copy_from_user(&chip, argp, sizeof chip)) 1782 if (!chip)
1739 goto out; 1783 goto out;
1784 r = -EFAULT;
1785 if (copy_from_user(chip, argp, sizeof *chip))
1786 goto set_irqchip_out;
1740 r = -ENXIO; 1787 r = -ENXIO;
1741 if (!irqchip_in_kernel(kvm)) 1788 if (!irqchip_in_kernel(kvm))
1742 goto out; 1789 goto set_irqchip_out;
1743 r = kvm_vm_ioctl_set_irqchip(kvm, &chip); 1790 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1744 if (r) 1791 if (r)
1745 goto out; 1792 goto set_irqchip_out;
1746 r = 0; 1793 r = 0;
1794 set_irqchip_out:
1795 kfree(chip);
1796 if (r)
1797 goto out;
1747 break; 1798 break;
1748 } 1799 }
1749 case KVM_GET_PIT: { 1800 case KVM_GET_PIT: {
1750 struct kvm_pit_state ps;
1751 r = -EFAULT; 1801 r = -EFAULT;
1752 if (copy_from_user(&ps, argp, sizeof ps)) 1802 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1753 goto out; 1803 goto out;
1754 r = -ENXIO; 1804 r = -ENXIO;
1755 if (!kvm->arch.vpit) 1805 if (!kvm->arch.vpit)
1756 goto out; 1806 goto out;
1757 r = kvm_vm_ioctl_get_pit(kvm, &ps); 1807 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1758 if (r) 1808 if (r)
1759 goto out; 1809 goto out;
1760 r = -EFAULT; 1810 r = -EFAULT;
1761 if (copy_to_user(argp, &ps, sizeof ps)) 1811 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1762 goto out; 1812 goto out;
1763 r = 0; 1813 r = 0;
1764 break; 1814 break;
1765 } 1815 }
1766 case KVM_SET_PIT: { 1816 case KVM_SET_PIT: {
1767 struct kvm_pit_state ps;
1768 r = -EFAULT; 1817 r = -EFAULT;
1769 if (copy_from_user(&ps, argp, sizeof ps)) 1818 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1770 goto out; 1819 goto out;
1771 r = -ENXIO; 1820 r = -ENXIO;
1772 if (!kvm->arch.vpit) 1821 if (!kvm->arch.vpit)
1773 goto out; 1822 goto out;
1774 r = kvm_vm_ioctl_set_pit(kvm, &ps); 1823 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1775 if (r) 1824 if (r)
1776 goto out; 1825 goto out;
1777 r = 0; 1826 r = 0;
@@ -2018,9 +2067,7 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
2018 2067
2019 val = *(u64 *)new; 2068 val = *(u64 *)new;
2020 2069
2021 down_read(&current->mm->mmap_sem);
2022 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); 2070 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2023 up_read(&current->mm->mmap_sem);
2024 2071
2025 kaddr = kmap_atomic(page, KM_USER0); 2072 kaddr = kmap_atomic(page, KM_USER0);
2026 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val); 2073 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
@@ -2040,6 +2087,7 @@ static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2040 2087
2041int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address) 2088int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2042{ 2089{
2090 kvm_mmu_invlpg(vcpu, address);
2043 return X86EMUL_CONTINUE; 2091 return X86EMUL_CONTINUE;
2044} 2092}
2045 2093
@@ -2080,7 +2128,7 @@ int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2080void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context) 2128void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2081{ 2129{
2082 u8 opcodes[4]; 2130 u8 opcodes[4];
2083 unsigned long rip = vcpu->arch.rip; 2131 unsigned long rip = kvm_rip_read(vcpu);
2084 unsigned long rip_linear; 2132 unsigned long rip_linear;
2085 2133
2086 if (!printk_ratelimit()) 2134 if (!printk_ratelimit())
@@ -2102,6 +2150,14 @@ static struct x86_emulate_ops emulate_ops = {
2102 .cmpxchg_emulated = emulator_cmpxchg_emulated, 2150 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2103}; 2151};
2104 2152
2153static void cache_all_regs(struct kvm_vcpu *vcpu)
2154{
2155 kvm_register_read(vcpu, VCPU_REGS_RAX);
2156 kvm_register_read(vcpu, VCPU_REGS_RSP);
2157 kvm_register_read(vcpu, VCPU_REGS_RIP);
2158 vcpu->arch.regs_dirty = ~0;
2159}
2160
2105int emulate_instruction(struct kvm_vcpu *vcpu, 2161int emulate_instruction(struct kvm_vcpu *vcpu,
2106 struct kvm_run *run, 2162 struct kvm_run *run,
2107 unsigned long cr2, 2163 unsigned long cr2,
@@ -2111,8 +2167,15 @@ int emulate_instruction(struct kvm_vcpu *vcpu,
2111 int r; 2167 int r;
2112 struct decode_cache *c; 2168 struct decode_cache *c;
2113 2169
2170 kvm_clear_exception_queue(vcpu);
2114 vcpu->arch.mmio_fault_cr2 = cr2; 2171 vcpu->arch.mmio_fault_cr2 = cr2;
2115 kvm_x86_ops->cache_regs(vcpu); 2172 /*
2173 * TODO: fix x86_emulate.c to use guest_read/write_register
2174 * instead of direct ->regs accesses, can save hundred cycles
2175 * on Intel for instructions that don't read/change RSP, for
2176 * for example.
2177 */
2178 cache_all_regs(vcpu);
2116 2179
2117 vcpu->mmio_is_write = 0; 2180 vcpu->mmio_is_write = 0;
2118 vcpu->arch.pio.string = 0; 2181 vcpu->arch.pio.string = 0;
@@ -2172,7 +2235,6 @@ int emulate_instruction(struct kvm_vcpu *vcpu,
2172 return EMULATE_DO_MMIO; 2235 return EMULATE_DO_MMIO;
2173 } 2236 }
2174 2237
2175 kvm_x86_ops->decache_regs(vcpu);
2176 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags); 2238 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2177 2239
2178 if (vcpu->mmio_is_write) { 2240 if (vcpu->mmio_is_write) {
@@ -2225,20 +2287,19 @@ int complete_pio(struct kvm_vcpu *vcpu)
2225 struct kvm_pio_request *io = &vcpu->arch.pio; 2287 struct kvm_pio_request *io = &vcpu->arch.pio;
2226 long delta; 2288 long delta;
2227 int r; 2289 int r;
2228 2290 unsigned long val;
2229 kvm_x86_ops->cache_regs(vcpu);
2230 2291
2231 if (!io->string) { 2292 if (!io->string) {
2232 if (io->in) 2293 if (io->in) {
2233 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data, 2294 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2234 io->size); 2295 memcpy(&val, vcpu->arch.pio_data, io->size);
2296 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2297 }
2235 } else { 2298 } else {
2236 if (io->in) { 2299 if (io->in) {
2237 r = pio_copy_data(vcpu); 2300 r = pio_copy_data(vcpu);
2238 if (r) { 2301 if (r)
2239 kvm_x86_ops->cache_regs(vcpu);
2240 return r; 2302 return r;
2241 }
2242 } 2303 }
2243 2304
2244 delta = 1; 2305 delta = 1;
@@ -2248,19 +2309,24 @@ int complete_pio(struct kvm_vcpu *vcpu)
2248 * The size of the register should really depend on 2309 * The size of the register should really depend on
2249 * current address size. 2310 * current address size.
2250 */ 2311 */
2251 vcpu->arch.regs[VCPU_REGS_RCX] -= delta; 2312 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2313 val -= delta;
2314 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2252 } 2315 }
2253 if (io->down) 2316 if (io->down)
2254 delta = -delta; 2317 delta = -delta;
2255 delta *= io->size; 2318 delta *= io->size;
2256 if (io->in) 2319 if (io->in) {
2257 vcpu->arch.regs[VCPU_REGS_RDI] += delta; 2320 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2258 else 2321 val += delta;
2259 vcpu->arch.regs[VCPU_REGS_RSI] += delta; 2322 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2323 } else {
2324 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2325 val += delta;
2326 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2327 }
2260 } 2328 }
2261 2329
2262 kvm_x86_ops->decache_regs(vcpu);
2263
2264 io->count -= io->cur_count; 2330 io->count -= io->cur_count;
2265 io->cur_count = 0; 2331 io->cur_count = 0;
2266 2332
@@ -2313,6 +2379,7 @@ int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2313 int size, unsigned port) 2379 int size, unsigned port)
2314{ 2380{
2315 struct kvm_io_device *pio_dev; 2381 struct kvm_io_device *pio_dev;
2382 unsigned long val;
2316 2383
2317 vcpu->run->exit_reason = KVM_EXIT_IO; 2384 vcpu->run->exit_reason = KVM_EXIT_IO;
2318 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; 2385 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
@@ -2333,8 +2400,8 @@ int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2333 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size, 2400 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2334 handler); 2401 handler);
2335 2402
2336 kvm_x86_ops->cache_regs(vcpu); 2403 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2337 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4); 2404 memcpy(vcpu->arch.pio_data, &val, 4);
2338 2405
2339 kvm_x86_ops->skip_emulated_instruction(vcpu); 2406 kvm_x86_ops->skip_emulated_instruction(vcpu);
2340 2407
@@ -2492,11 +2559,6 @@ int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2492 KVMTRACE_0D(HLT, vcpu, handler); 2559 KVMTRACE_0D(HLT, vcpu, handler);
2493 if (irqchip_in_kernel(vcpu->kvm)) { 2560 if (irqchip_in_kernel(vcpu->kvm)) {
2494 vcpu->arch.mp_state = KVM_MP_STATE_HALTED; 2561 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2495 up_read(&vcpu->kvm->slots_lock);
2496 kvm_vcpu_block(vcpu);
2497 down_read(&vcpu->kvm->slots_lock);
2498 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2499 return -EINTR;
2500 return 1; 2562 return 1;
2501 } else { 2563 } else {
2502 vcpu->run->exit_reason = KVM_EXIT_HLT; 2564 vcpu->run->exit_reason = KVM_EXIT_HLT;
@@ -2519,13 +2581,11 @@ int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2519 unsigned long nr, a0, a1, a2, a3, ret; 2581 unsigned long nr, a0, a1, a2, a3, ret;
2520 int r = 1; 2582 int r = 1;
2521 2583
2522 kvm_x86_ops->cache_regs(vcpu); 2584 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2523 2585 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2524 nr = vcpu->arch.regs[VCPU_REGS_RAX]; 2586 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2525 a0 = vcpu->arch.regs[VCPU_REGS_RBX]; 2587 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2526 a1 = vcpu->arch.regs[VCPU_REGS_RCX]; 2588 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2527 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2528 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2529 2589
2530 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler); 2590 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2531 2591
@@ -2548,8 +2608,7 @@ int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2548 ret = -KVM_ENOSYS; 2608 ret = -KVM_ENOSYS;
2549 break; 2609 break;
2550 } 2610 }
2551 vcpu->arch.regs[VCPU_REGS_RAX] = ret; 2611 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2552 kvm_x86_ops->decache_regs(vcpu);
2553 ++vcpu->stat.hypercalls; 2612 ++vcpu->stat.hypercalls;
2554 return r; 2613 return r;
2555} 2614}
@@ -2559,6 +2618,7 @@ int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2559{ 2618{
2560 char instruction[3]; 2619 char instruction[3];
2561 int ret = 0; 2620 int ret = 0;
2621 unsigned long rip = kvm_rip_read(vcpu);
2562 2622
2563 2623
2564 /* 2624 /*
@@ -2568,9 +2628,8 @@ int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2568 */ 2628 */
2569 kvm_mmu_zap_all(vcpu->kvm); 2629 kvm_mmu_zap_all(vcpu->kvm);
2570 2630
2571 kvm_x86_ops->cache_regs(vcpu);
2572 kvm_x86_ops->patch_hypercall(vcpu, instruction); 2631 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2573 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu) 2632 if (emulator_write_emulated(rip, instruction, 3, vcpu)
2574 != X86EMUL_CONTINUE) 2633 != X86EMUL_CONTINUE)
2575 ret = -EFAULT; 2634 ret = -EFAULT;
2576 2635
@@ -2700,13 +2759,12 @@ void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2700 u32 function, index; 2759 u32 function, index;
2701 struct kvm_cpuid_entry2 *e, *best; 2760 struct kvm_cpuid_entry2 *e, *best;
2702 2761
2703 kvm_x86_ops->cache_regs(vcpu); 2762 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2704 function = vcpu->arch.regs[VCPU_REGS_RAX]; 2763 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2705 index = vcpu->arch.regs[VCPU_REGS_RCX]; 2764 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2706 vcpu->arch.regs[VCPU_REGS_RAX] = 0; 2765 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2707 vcpu->arch.regs[VCPU_REGS_RBX] = 0; 2766 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2708 vcpu->arch.regs[VCPU_REGS_RCX] = 0; 2767 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2709 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2710 best = NULL; 2768 best = NULL;
2711 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 2769 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2712 e = &vcpu->arch.cpuid_entries[i]; 2770 e = &vcpu->arch.cpuid_entries[i];
@@ -2724,18 +2782,17 @@ void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2724 best = e; 2782 best = e;
2725 } 2783 }
2726 if (best) { 2784 if (best) {
2727 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax; 2785 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2728 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx; 2786 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2729 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx; 2787 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2730 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx; 2788 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2731 } 2789 }
2732 kvm_x86_ops->decache_regs(vcpu);
2733 kvm_x86_ops->skip_emulated_instruction(vcpu); 2790 kvm_x86_ops->skip_emulated_instruction(vcpu);
2734 KVMTRACE_5D(CPUID, vcpu, function, 2791 KVMTRACE_5D(CPUID, vcpu, function,
2735 (u32)vcpu->arch.regs[VCPU_REGS_RAX], 2792 (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2736 (u32)vcpu->arch.regs[VCPU_REGS_RBX], 2793 (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2737 (u32)vcpu->arch.regs[VCPU_REGS_RCX], 2794 (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2738 (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler); 2795 (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2739} 2796}
2740EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); 2797EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2741 2798
@@ -2776,9 +2833,7 @@ static void vapic_enter(struct kvm_vcpu *vcpu)
2776 if (!apic || !apic->vapic_addr) 2833 if (!apic || !apic->vapic_addr)
2777 return; 2834 return;
2778 2835
2779 down_read(&current->mm->mmap_sem);
2780 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT); 2836 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2781 up_read(&current->mm->mmap_sem);
2782 2837
2783 vcpu->arch.apic->vapic_page = page; 2838 vcpu->arch.apic->vapic_page = page;
2784} 2839}
@@ -2796,28 +2851,10 @@ static void vapic_exit(struct kvm_vcpu *vcpu)
2796 up_read(&vcpu->kvm->slots_lock); 2851 up_read(&vcpu->kvm->slots_lock);
2797} 2852}
2798 2853
2799static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 2854static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2800{ 2855{
2801 int r; 2856 int r;
2802 2857
2803 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2804 pr_debug("vcpu %d received sipi with vector # %x\n",
2805 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2806 kvm_lapic_reset(vcpu);
2807 r = kvm_x86_ops->vcpu_reset(vcpu);
2808 if (r)
2809 return r;
2810 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2811 }
2812
2813 down_read(&vcpu->kvm->slots_lock);
2814 vapic_enter(vcpu);
2815
2816preempted:
2817 if (vcpu->guest_debug.enabled)
2818 kvm_x86_ops->guest_debug_pre(vcpu);
2819
2820again:
2821 if (vcpu->requests) 2858 if (vcpu->requests)
2822 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) 2859 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2823 kvm_mmu_unload(vcpu); 2860 kvm_mmu_unload(vcpu);
@@ -2829,6 +2866,8 @@ again:
2829 if (vcpu->requests) { 2866 if (vcpu->requests) {
2830 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests)) 2867 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2831 __kvm_migrate_timers(vcpu); 2868 __kvm_migrate_timers(vcpu);
2869 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2870 kvm_mmu_sync_roots(vcpu);
2832 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests)) 2871 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2833 kvm_x86_ops->tlb_flush(vcpu); 2872 kvm_x86_ops->tlb_flush(vcpu);
2834 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS, 2873 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
@@ -2854,21 +2893,15 @@ again:
2854 2893
2855 local_irq_disable(); 2894 local_irq_disable();
2856 2895
2857 if (vcpu->requests || need_resched()) { 2896 if (vcpu->requests || need_resched() || signal_pending(current)) {
2858 local_irq_enable(); 2897 local_irq_enable();
2859 preempt_enable(); 2898 preempt_enable();
2860 r = 1; 2899 r = 1;
2861 goto out; 2900 goto out;
2862 } 2901 }
2863 2902
2864 if (signal_pending(current)) { 2903 if (vcpu->guest_debug.enabled)
2865 local_irq_enable(); 2904 kvm_x86_ops->guest_debug_pre(vcpu);
2866 preempt_enable();
2867 r = -EINTR;
2868 kvm_run->exit_reason = KVM_EXIT_INTR;
2869 ++vcpu->stat.signal_exits;
2870 goto out;
2871 }
2872 2905
2873 vcpu->guest_mode = 1; 2906 vcpu->guest_mode = 1;
2874 /* 2907 /*
@@ -2917,8 +2950,8 @@ again:
2917 * Profile KVM exit RIPs: 2950 * Profile KVM exit RIPs:
2918 */ 2951 */
2919 if (unlikely(prof_on == KVM_PROFILING)) { 2952 if (unlikely(prof_on == KVM_PROFILING)) {
2920 kvm_x86_ops->cache_regs(vcpu); 2953 unsigned long rip = kvm_rip_read(vcpu);
2921 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip); 2954 profile_hit(KVM_PROFILING, (void *)rip);
2922 } 2955 }
2923 2956
2924 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu)) 2957 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
@@ -2927,26 +2960,63 @@ again:
2927 kvm_lapic_sync_from_vapic(vcpu); 2960 kvm_lapic_sync_from_vapic(vcpu);
2928 2961
2929 r = kvm_x86_ops->handle_exit(kvm_run, vcpu); 2962 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2963out:
2964 return r;
2965}
2930 2966
2931 if (r > 0) { 2967static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2932 if (dm_request_for_irq_injection(vcpu, kvm_run)) { 2968{
2933 r = -EINTR; 2969 int r;
2934 kvm_run->exit_reason = KVM_EXIT_INTR; 2970
2935 ++vcpu->stat.request_irq_exits; 2971 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2936 goto out; 2972 pr_debug("vcpu %d received sipi with vector # %x\n",
2937 } 2973 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2938 if (!need_resched()) 2974 kvm_lapic_reset(vcpu);
2939 goto again; 2975 r = kvm_x86_ops->vcpu_reset(vcpu);
2976 if (r)
2977 return r;
2978 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2940 } 2979 }
2941 2980
2942out: 2981 down_read(&vcpu->kvm->slots_lock);
2943 up_read(&vcpu->kvm->slots_lock); 2982 vapic_enter(vcpu);
2944 if (r > 0) { 2983
2945 kvm_resched(vcpu); 2984 r = 1;
2946 down_read(&vcpu->kvm->slots_lock); 2985 while (r > 0) {
2947 goto preempted; 2986 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
2987 r = vcpu_enter_guest(vcpu, kvm_run);
2988 else {
2989 up_read(&vcpu->kvm->slots_lock);
2990 kvm_vcpu_block(vcpu);
2991 down_read(&vcpu->kvm->slots_lock);
2992 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
2993 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
2994 vcpu->arch.mp_state =
2995 KVM_MP_STATE_RUNNABLE;
2996 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2997 r = -EINTR;
2998 }
2999
3000 if (r > 0) {
3001 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3002 r = -EINTR;
3003 kvm_run->exit_reason = KVM_EXIT_INTR;
3004 ++vcpu->stat.request_irq_exits;
3005 }
3006 if (signal_pending(current)) {
3007 r = -EINTR;
3008 kvm_run->exit_reason = KVM_EXIT_INTR;
3009 ++vcpu->stat.signal_exits;
3010 }
3011 if (need_resched()) {
3012 up_read(&vcpu->kvm->slots_lock);
3013 kvm_resched(vcpu);
3014 down_read(&vcpu->kvm->slots_lock);
3015 }
3016 }
2948 } 3017 }
2949 3018
3019 up_read(&vcpu->kvm->slots_lock);
2950 post_kvm_run_save(vcpu, kvm_run); 3020 post_kvm_run_save(vcpu, kvm_run);
2951 3021
2952 vapic_exit(vcpu); 3022 vapic_exit(vcpu);
@@ -2966,6 +3036,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2966 3036
2967 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { 3037 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2968 kvm_vcpu_block(vcpu); 3038 kvm_vcpu_block(vcpu);
3039 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2969 r = -EAGAIN; 3040 r = -EAGAIN;
2970 goto out; 3041 goto out;
2971 } 3042 }
@@ -2999,11 +3070,9 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2999 } 3070 }
3000 } 3071 }
3001#endif 3072#endif
3002 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) { 3073 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3003 kvm_x86_ops->cache_regs(vcpu); 3074 kvm_register_write(vcpu, VCPU_REGS_RAX,
3004 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret; 3075 kvm_run->hypercall.ret);
3005 kvm_x86_ops->decache_regs(vcpu);
3006 }
3007 3076
3008 r = __vcpu_run(vcpu, kvm_run); 3077 r = __vcpu_run(vcpu, kvm_run);
3009 3078
@@ -3019,28 +3088,26 @@ int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3019{ 3088{
3020 vcpu_load(vcpu); 3089 vcpu_load(vcpu);
3021 3090
3022 kvm_x86_ops->cache_regs(vcpu); 3091 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3023 3092 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3024 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX]; 3093 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3025 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX]; 3094 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3026 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX]; 3095 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3027 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX]; 3096 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3028 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI]; 3097 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3029 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI]; 3098 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3030 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3031 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3032#ifdef CONFIG_X86_64 3099#ifdef CONFIG_X86_64
3033 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8]; 3100 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3034 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9]; 3101 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3035 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10]; 3102 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3036 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11]; 3103 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3037 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12]; 3104 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3038 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13]; 3105 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3039 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14]; 3106 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3040 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15]; 3107 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3041#endif 3108#endif
3042 3109
3043 regs->rip = vcpu->arch.rip; 3110 regs->rip = kvm_rip_read(vcpu);
3044 regs->rflags = kvm_x86_ops->get_rflags(vcpu); 3111 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3045 3112
3046 /* 3113 /*
@@ -3058,29 +3125,29 @@ int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3058{ 3125{
3059 vcpu_load(vcpu); 3126 vcpu_load(vcpu);
3060 3127
3061 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax; 3128 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3062 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx; 3129 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3063 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx; 3130 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3064 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx; 3131 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3065 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi; 3132 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3066 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi; 3133 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3067 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp; 3134 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3068 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp; 3135 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3069#ifdef CONFIG_X86_64 3136#ifdef CONFIG_X86_64
3070 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8; 3137 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3071 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9; 3138 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3072 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10; 3139 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3073 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11; 3140 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3074 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12; 3141 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3075 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13; 3142 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3076 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14; 3143 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3077 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15; 3144 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3145
3078#endif 3146#endif
3079 3147
3080 vcpu->arch.rip = regs->rip; 3148 kvm_rip_write(vcpu, regs->rip);
3081 kvm_x86_ops->set_rflags(vcpu, regs->rflags); 3149 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3082 3150
3083 kvm_x86_ops->decache_regs(vcpu);
3084 3151
3085 vcpu->arch.exception.pending = false; 3152 vcpu->arch.exception.pending = false;
3086 3153
@@ -3294,11 +3361,33 @@ static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3294 return 0; 3361 return 0;
3295} 3362}
3296 3363
3364static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3365{
3366 struct kvm_segment segvar = {
3367 .base = selector << 4,
3368 .limit = 0xffff,
3369 .selector = selector,
3370 .type = 3,
3371 .present = 1,
3372 .dpl = 3,
3373 .db = 0,
3374 .s = 1,
3375 .l = 0,
3376 .g = 0,
3377 .avl = 0,
3378 .unusable = 0,
3379 };
3380 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3381 return 0;
3382}
3383
3297int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, 3384int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3298 int type_bits, int seg) 3385 int type_bits, int seg)
3299{ 3386{
3300 struct kvm_segment kvm_seg; 3387 struct kvm_segment kvm_seg;
3301 3388
3389 if (!(vcpu->arch.cr0 & X86_CR0_PE))
3390 return kvm_load_realmode_segment(vcpu, selector, seg);
3302 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg)) 3391 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3303 return 1; 3392 return 1;
3304 kvm_seg.type |= type_bits; 3393 kvm_seg.type |= type_bits;
@@ -3316,17 +3405,16 @@ static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3316 struct tss_segment_32 *tss) 3405 struct tss_segment_32 *tss)
3317{ 3406{
3318 tss->cr3 = vcpu->arch.cr3; 3407 tss->cr3 = vcpu->arch.cr3;
3319 tss->eip = vcpu->arch.rip; 3408 tss->eip = kvm_rip_read(vcpu);
3320 tss->eflags = kvm_x86_ops->get_rflags(vcpu); 3409 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3321 tss->eax = vcpu->arch.regs[VCPU_REGS_RAX]; 3410 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3322 tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX]; 3411 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3323 tss->edx = vcpu->arch.regs[VCPU_REGS_RDX]; 3412 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3324 tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX]; 3413 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3325 tss->esp = vcpu->arch.regs[VCPU_REGS_RSP]; 3414 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3326 tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP]; 3415 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3327 tss->esi = vcpu->arch.regs[VCPU_REGS_RSI]; 3416 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3328 tss->edi = vcpu->arch.regs[VCPU_REGS_RDI]; 3417 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3329
3330 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES); 3418 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3331 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS); 3419 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3332 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS); 3420 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
@@ -3342,17 +3430,17 @@ static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3342{ 3430{
3343 kvm_set_cr3(vcpu, tss->cr3); 3431 kvm_set_cr3(vcpu, tss->cr3);
3344 3432
3345 vcpu->arch.rip = tss->eip; 3433 kvm_rip_write(vcpu, tss->eip);
3346 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2); 3434 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3347 3435
3348 vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax; 3436 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3349 vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx; 3437 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3350 vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx; 3438 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3351 vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx; 3439 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3352 vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp; 3440 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3353 vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp; 3441 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3354 vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi; 3442 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3355 vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi; 3443 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3356 3444
3357 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR)) 3445 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3358 return 1; 3446 return 1;
@@ -3380,16 +3468,16 @@ static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3380static void save_state_to_tss16(struct kvm_vcpu *vcpu, 3468static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3381 struct tss_segment_16 *tss) 3469 struct tss_segment_16 *tss)
3382{ 3470{
3383 tss->ip = vcpu->arch.rip; 3471 tss->ip = kvm_rip_read(vcpu);
3384 tss->flag = kvm_x86_ops->get_rflags(vcpu); 3472 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3385 tss->ax = vcpu->arch.regs[VCPU_REGS_RAX]; 3473 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3386 tss->cx = vcpu->arch.regs[VCPU_REGS_RCX]; 3474 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3387 tss->dx = vcpu->arch.regs[VCPU_REGS_RDX]; 3475 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3388 tss->bx = vcpu->arch.regs[VCPU_REGS_RBX]; 3476 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3389 tss->sp = vcpu->arch.regs[VCPU_REGS_RSP]; 3477 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3390 tss->bp = vcpu->arch.regs[VCPU_REGS_RBP]; 3478 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3391 tss->si = vcpu->arch.regs[VCPU_REGS_RSI]; 3479 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3392 tss->di = vcpu->arch.regs[VCPU_REGS_RDI]; 3480 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3393 3481
3394 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES); 3482 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3395 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS); 3483 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
@@ -3402,16 +3490,16 @@ static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3402static int load_state_from_tss16(struct kvm_vcpu *vcpu, 3490static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3403 struct tss_segment_16 *tss) 3491 struct tss_segment_16 *tss)
3404{ 3492{
3405 vcpu->arch.rip = tss->ip; 3493 kvm_rip_write(vcpu, tss->ip);
3406 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2); 3494 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3407 vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax; 3495 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3408 vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx; 3496 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3409 vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx; 3497 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3410 vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx; 3498 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3411 vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp; 3499 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3412 vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp; 3500 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3413 vcpu->arch.regs[VCPU_REGS_RSI] = tss->si; 3501 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3414 vcpu->arch.regs[VCPU_REGS_RDI] = tss->di; 3502 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3415 3503
3416 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR)) 3504 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3417 return 1; 3505 return 1;
@@ -3534,7 +3622,6 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3534 } 3622 }
3535 3623
3536 kvm_x86_ops->skip_emulated_instruction(vcpu); 3624 kvm_x86_ops->skip_emulated_instruction(vcpu);
3537 kvm_x86_ops->cache_regs(vcpu);
3538 3625
3539 if (nseg_desc.type & 8) 3626 if (nseg_desc.type & 8)
3540 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base, 3627 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
@@ -3559,7 +3646,6 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3559 tr_seg.type = 11; 3646 tr_seg.type = 11;
3560 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR); 3647 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3561out: 3648out:
3562 kvm_x86_ops->decache_regs(vcpu);
3563 return ret; 3649 return ret;
3564} 3650}
3565EXPORT_SYMBOL_GPL(kvm_task_switch); 3651EXPORT_SYMBOL_GPL(kvm_task_switch);
@@ -3622,6 +3708,7 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3622 pr_debug("Set back pending irq %d\n", 3708 pr_debug("Set back pending irq %d\n",
3623 pending_vec); 3709 pending_vec);
3624 } 3710 }
3711 kvm_pic_clear_isr_ack(vcpu->kvm);
3625 } 3712 }
3626 3713
3627 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); 3714 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
@@ -3634,6 +3721,12 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3634 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); 3721 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3635 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); 3722 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3636 3723
3724 /* Older userspace won't unhalt the vcpu on reset. */
3725 if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3726 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3727 !(vcpu->arch.cr0 & X86_CR0_PE))
3728 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3729
3637 vcpu_put(vcpu); 3730 vcpu_put(vcpu);
3638 3731
3639 return 0; 3732 return 0;
@@ -3918,6 +4011,7 @@ struct kvm *kvm_arch_create_vm(void)
3918 return ERR_PTR(-ENOMEM); 4011 return ERR_PTR(-ENOMEM);
3919 4012
3920 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); 4013 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4014 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
3921 4015
3922 return kvm; 4016 return kvm;
3923} 4017}
@@ -3950,6 +4044,8 @@ static void kvm_free_vcpus(struct kvm *kvm)
3950 4044
3951void kvm_arch_destroy_vm(struct kvm *kvm) 4045void kvm_arch_destroy_vm(struct kvm *kvm)
3952{ 4046{
4047 kvm_iommu_unmap_guest(kvm);
4048 kvm_free_all_assigned_devices(kvm);
3953 kvm_free_pit(kvm); 4049 kvm_free_pit(kvm);
3954 kfree(kvm->arch.vpic); 4050 kfree(kvm->arch.vpic);
3955 kfree(kvm->arch.vioapic); 4051 kfree(kvm->arch.vioapic);
@@ -3981,7 +4077,7 @@ int kvm_arch_set_memory_region(struct kvm *kvm,
3981 userspace_addr = do_mmap(NULL, 0, 4077 userspace_addr = do_mmap(NULL, 0,
3982 npages * PAGE_SIZE, 4078 npages * PAGE_SIZE,
3983 PROT_READ | PROT_WRITE, 4079 PROT_READ | PROT_WRITE,
3984 MAP_SHARED | MAP_ANONYMOUS, 4080 MAP_PRIVATE | MAP_ANONYMOUS,
3985 0); 4081 0);
3986 up_write(&current->mm->mmap_sem); 4082 up_write(&current->mm->mmap_sem);
3987 4083
diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
new file mode 100644
index 000000000000..6a4be78a7384
--- /dev/null
+++ b/arch/x86/kvm/x86.h
@@ -0,0 +1,22 @@
1#ifndef ARCH_X86_KVM_X86_H
2#define ARCH_X86_KVM_X86_H
3
4#include <linux/kvm_host.h>
5
6static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
7{
8 vcpu->arch.exception.pending = false;
9}
10
11static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector)
12{
13 vcpu->arch.interrupt.pending = true;
14 vcpu->arch.interrupt.nr = vector;
15}
16
17static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
18{
19 vcpu->arch.interrupt.pending = false;
20}
21
22#endif
diff --git a/arch/x86/kvm/x86_emulate.c b/arch/x86/kvm/x86_emulate.c
index f2f90468f8b1..ea051173b0da 100644
--- a/arch/x86/kvm/x86_emulate.c
+++ b/arch/x86/kvm/x86_emulate.c
@@ -26,6 +26,7 @@
26#define DPRINTF(_f, _a ...) printf(_f , ## _a) 26#define DPRINTF(_f, _a ...) printf(_f , ## _a)
27#else 27#else
28#include <linux/kvm_host.h> 28#include <linux/kvm_host.h>
29#include "kvm_cache_regs.h"
29#define DPRINTF(x...) do {} while (0) 30#define DPRINTF(x...) do {} while (0)
30#endif 31#endif
31#include <linux/module.h> 32#include <linux/module.h>
@@ -46,25 +47,26 @@
46#define ImplicitOps (1<<1) /* Implicit in opcode. No generic decode. */ 47#define ImplicitOps (1<<1) /* Implicit in opcode. No generic decode. */
47#define DstReg (2<<1) /* Register operand. */ 48#define DstReg (2<<1) /* Register operand. */
48#define DstMem (3<<1) /* Memory operand. */ 49#define DstMem (3<<1) /* Memory operand. */
49#define DstMask (3<<1) 50#define DstAcc (4<<1) /* Destination Accumulator */
51#define DstMask (7<<1)
50/* Source operand type. */ 52/* Source operand type. */
51#define SrcNone (0<<3) /* No source operand. */ 53#define SrcNone (0<<4) /* No source operand. */
52#define SrcImplicit (0<<3) /* Source operand is implicit in the opcode. */ 54#define SrcImplicit (0<<4) /* Source operand is implicit in the opcode. */
53#define SrcReg (1<<3) /* Register operand. */ 55#define SrcReg (1<<4) /* Register operand. */
54#define SrcMem (2<<3) /* Memory operand. */ 56#define SrcMem (2<<4) /* Memory operand. */
55#define SrcMem16 (3<<3) /* Memory operand (16-bit). */ 57#define SrcMem16 (3<<4) /* Memory operand (16-bit). */
56#define SrcMem32 (4<<3) /* Memory operand (32-bit). */ 58#define SrcMem32 (4<<4) /* Memory operand (32-bit). */
57#define SrcImm (5<<3) /* Immediate operand. */ 59#define SrcImm (5<<4) /* Immediate operand. */
58#define SrcImmByte (6<<3) /* 8-bit sign-extended immediate operand. */ 60#define SrcImmByte (6<<4) /* 8-bit sign-extended immediate operand. */
59#define SrcMask (7<<3) 61#define SrcMask (7<<4)
60/* Generic ModRM decode. */ 62/* Generic ModRM decode. */
61#define ModRM (1<<6) 63#define ModRM (1<<7)
62/* Destination is only written; never read. */ 64/* Destination is only written; never read. */
63#define Mov (1<<7) 65#define Mov (1<<8)
64#define BitOp (1<<8) 66#define BitOp (1<<9)
65#define MemAbs (1<<9) /* Memory operand is absolute displacement */ 67#define MemAbs (1<<10) /* Memory operand is absolute displacement */
66#define String (1<<10) /* String instruction (rep capable) */ 68#define String (1<<12) /* String instruction (rep capable) */
67#define Stack (1<<11) /* Stack instruction (push/pop) */ 69#define Stack (1<<13) /* Stack instruction (push/pop) */
68#define Group (1<<14) /* Bits 3:5 of modrm byte extend opcode */ 70#define Group (1<<14) /* Bits 3:5 of modrm byte extend opcode */
69#define GroupDual (1<<15) /* Alternate decoding of mod == 3 */ 71#define GroupDual (1<<15) /* Alternate decoding of mod == 3 */
70#define GroupMask 0xff /* Group number stored in bits 0:7 */ 72#define GroupMask 0xff /* Group number stored in bits 0:7 */
@@ -94,7 +96,7 @@ static u16 opcode_table[256] = {
94 /* 0x20 - 0x27 */ 96 /* 0x20 - 0x27 */
95 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 97 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM,
96 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM, 98 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM,
97 SrcImmByte, SrcImm, 0, 0, 99 DstAcc | SrcImmByte, DstAcc | SrcImm, 0, 0,
98 /* 0x28 - 0x2F */ 100 /* 0x28 - 0x2F */
99 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 101 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM,
100 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM, 102 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM,
@@ -106,7 +108,8 @@ static u16 opcode_table[256] = {
106 /* 0x38 - 0x3F */ 108 /* 0x38 - 0x3F */
107 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM, 109 ByteOp | DstMem | SrcReg | ModRM, DstMem | SrcReg | ModRM,
108 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM, 110 ByteOp | DstReg | SrcMem | ModRM, DstReg | SrcMem | ModRM,
109 0, 0, 0, 0, 111 ByteOp | DstAcc | SrcImm, DstAcc | SrcImm,
112 0, 0,
110 /* 0x40 - 0x47 */ 113 /* 0x40 - 0x47 */
111 DstReg, DstReg, DstReg, DstReg, DstReg, DstReg, DstReg, DstReg, 114 DstReg, DstReg, DstReg, DstReg, DstReg, DstReg, DstReg, DstReg,
112 /* 0x48 - 0x4F */ 115 /* 0x48 - 0x4F */
@@ -153,9 +156,16 @@ static u16 opcode_table[256] = {
153 0, 0, ByteOp | ImplicitOps | Mov | String, ImplicitOps | Mov | String, 156 0, 0, ByteOp | ImplicitOps | Mov | String, ImplicitOps | Mov | String,
154 ByteOp | ImplicitOps | Mov | String, ImplicitOps | Mov | String, 157 ByteOp | ImplicitOps | Mov | String, ImplicitOps | Mov | String,
155 ByteOp | ImplicitOps | String, ImplicitOps | String, 158 ByteOp | ImplicitOps | String, ImplicitOps | String,
156 /* 0xB0 - 0xBF */ 159 /* 0xB0 - 0xB7 */
157 0, 0, 0, 0, 0, 0, 0, 0, 160 ByteOp | DstReg | SrcImm | Mov, ByteOp | DstReg | SrcImm | Mov,
158 DstReg | SrcImm | Mov, 0, 0, 0, 0, 0, 0, 0, 161 ByteOp | DstReg | SrcImm | Mov, ByteOp | DstReg | SrcImm | Mov,
162 ByteOp | DstReg | SrcImm | Mov, ByteOp | DstReg | SrcImm | Mov,
163 ByteOp | DstReg | SrcImm | Mov, ByteOp | DstReg | SrcImm | Mov,
164 /* 0xB8 - 0xBF */
165 DstReg | SrcImm | Mov, DstReg | SrcImm | Mov,
166 DstReg | SrcImm | Mov, DstReg | SrcImm | Mov,
167 DstReg | SrcImm | Mov, DstReg | SrcImm | Mov,
168 DstReg | SrcImm | Mov, DstReg | SrcImm | Mov,
159 /* 0xC0 - 0xC7 */ 169 /* 0xC0 - 0xC7 */
160 ByteOp | DstMem | SrcImm | ModRM, DstMem | SrcImmByte | ModRM, 170 ByteOp | DstMem | SrcImm | ModRM, DstMem | SrcImmByte | ModRM,
161 0, ImplicitOps | Stack, 0, 0, 171 0, ImplicitOps | Stack, 0, 0,
@@ -169,17 +179,20 @@ static u16 opcode_table[256] = {
169 /* 0xD8 - 0xDF */ 179 /* 0xD8 - 0xDF */
170 0, 0, 0, 0, 0, 0, 0, 0, 180 0, 0, 0, 0, 0, 0, 0, 0,
171 /* 0xE0 - 0xE7 */ 181 /* 0xE0 - 0xE7 */
172 0, 0, 0, 0, 0, 0, 0, 0, 182 0, 0, 0, 0,
183 SrcNone | ByteOp | ImplicitOps, SrcNone | ImplicitOps,
184 SrcNone | ByteOp | ImplicitOps, SrcNone | ImplicitOps,
173 /* 0xE8 - 0xEF */ 185 /* 0xE8 - 0xEF */
174 ImplicitOps | Stack, SrcImm | ImplicitOps, 186 ImplicitOps | Stack, SrcImm | ImplicitOps,
175 ImplicitOps, SrcImmByte | ImplicitOps, 187 ImplicitOps, SrcImmByte | ImplicitOps,
176 0, 0, 0, 0, 188 SrcNone | ByteOp | ImplicitOps, SrcNone | ImplicitOps,
189 SrcNone | ByteOp | ImplicitOps, SrcNone | ImplicitOps,
177 /* 0xF0 - 0xF7 */ 190 /* 0xF0 - 0xF7 */
178 0, 0, 0, 0, 191 0, 0, 0, 0,
179 ImplicitOps, ImplicitOps, Group | Group3_Byte, Group | Group3, 192 ImplicitOps, ImplicitOps, Group | Group3_Byte, Group | Group3,
180 /* 0xF8 - 0xFF */ 193 /* 0xF8 - 0xFF */
181 ImplicitOps, 0, ImplicitOps, ImplicitOps, 194 ImplicitOps, 0, ImplicitOps, ImplicitOps,
182 0, 0, Group | Group4, Group | Group5, 195 ImplicitOps, ImplicitOps, Group | Group4, Group | Group5,
183}; 196};
184 197
185static u16 twobyte_table[256] = { 198static u16 twobyte_table[256] = {
@@ -268,15 +281,16 @@ static u16 group_table[] = {
268 ByteOp | DstMem | SrcNone | ModRM, ByteOp | DstMem | SrcNone | ModRM, 281 ByteOp | DstMem | SrcNone | ModRM, ByteOp | DstMem | SrcNone | ModRM,
269 0, 0, 0, 0, 282 0, 0, 0, 0,
270 [Group3*8] = 283 [Group3*8] =
271 DstMem | SrcImm | ModRM | SrcImm, 0, 284 DstMem | SrcImm | ModRM, 0,
272 DstMem | SrcNone | ModRM, ByteOp | DstMem | SrcNone | ModRM, 285 DstMem | SrcNone | ModRM, DstMem | SrcNone | ModRM,
273 0, 0, 0, 0, 286 0, 0, 0, 0,
274 [Group4*8] = 287 [Group4*8] =
275 ByteOp | DstMem | SrcNone | ModRM, ByteOp | DstMem | SrcNone | ModRM, 288 ByteOp | DstMem | SrcNone | ModRM, ByteOp | DstMem | SrcNone | ModRM,
276 0, 0, 0, 0, 0, 0, 289 0, 0, 0, 0, 0, 0,
277 [Group5*8] = 290 [Group5*8] =
278 DstMem | SrcNone | ModRM, DstMem | SrcNone | ModRM, 0, 0, 291 DstMem | SrcNone | ModRM, DstMem | SrcNone | ModRM,
279 SrcMem | ModRM, 0, SrcMem | ModRM | Stack, 0, 292 SrcMem | ModRM | Stack, 0,
293 SrcMem | ModRM | Stack, 0, SrcMem | ModRM | Stack, 0,
280 [Group7*8] = 294 [Group7*8] =
281 0, 0, ModRM | SrcMem, ModRM | SrcMem, 295 0, 0, ModRM | SrcMem, ModRM | SrcMem,
282 SrcNone | ModRM | DstMem | Mov, 0, 296 SrcNone | ModRM | DstMem | Mov, 0,
@@ -839,7 +853,7 @@ x86_decode_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
839 /* Shadow copy of register state. Committed on successful emulation. */ 853 /* Shadow copy of register state. Committed on successful emulation. */
840 854
841 memset(c, 0, sizeof(struct decode_cache)); 855 memset(c, 0, sizeof(struct decode_cache));
842 c->eip = ctxt->vcpu->arch.rip; 856 c->eip = kvm_rip_read(ctxt->vcpu);
843 ctxt->cs_base = seg_base(ctxt, VCPU_SREG_CS); 857 ctxt->cs_base = seg_base(ctxt, VCPU_SREG_CS);
844 memcpy(c->regs, ctxt->vcpu->arch.regs, sizeof c->regs); 858 memcpy(c->regs, ctxt->vcpu->arch.regs, sizeof c->regs);
845 859
@@ -1048,6 +1062,23 @@ done_prefixes:
1048 } 1062 }
1049 c->dst.type = OP_MEM; 1063 c->dst.type = OP_MEM;
1050 break; 1064 break;
1065 case DstAcc:
1066 c->dst.type = OP_REG;
1067 c->dst.bytes = c->op_bytes;
1068 c->dst.ptr = &c->regs[VCPU_REGS_RAX];
1069 switch (c->op_bytes) {
1070 case 1:
1071 c->dst.val = *(u8 *)c->dst.ptr;
1072 break;
1073 case 2:
1074 c->dst.val = *(u16 *)c->dst.ptr;
1075 break;
1076 case 4:
1077 c->dst.val = *(u32 *)c->dst.ptr;
1078 break;
1079 }
1080 c->dst.orig_val = c->dst.val;
1081 break;
1051 } 1082 }
1052 1083
1053 if (c->rip_relative) 1084 if (c->rip_relative)
@@ -1151,6 +1182,14 @@ static inline int emulate_grp45(struct x86_emulate_ctxt *ctxt,
1151 case 1: /* dec */ 1182 case 1: /* dec */
1152 emulate_1op("dec", c->dst, ctxt->eflags); 1183 emulate_1op("dec", c->dst, ctxt->eflags);
1153 break; 1184 break;
1185 case 2: /* call near abs */ {
1186 long int old_eip;
1187 old_eip = c->eip;
1188 c->eip = c->src.val;
1189 c->src.val = old_eip;
1190 emulate_push(ctxt);
1191 break;
1192 }
1154 case 4: /* jmp abs */ 1193 case 4: /* jmp abs */
1155 c->eip = c->src.val; 1194 c->eip = c->src.val;
1156 break; 1195 break;
@@ -1251,6 +1290,8 @@ x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
1251 u64 msr_data; 1290 u64 msr_data;
1252 unsigned long saved_eip = 0; 1291 unsigned long saved_eip = 0;
1253 struct decode_cache *c = &ctxt->decode; 1292 struct decode_cache *c = &ctxt->decode;
1293 unsigned int port;
1294 int io_dir_in;
1254 int rc = 0; 1295 int rc = 0;
1255 1296
1256 /* Shadow copy of register state. Committed on successful emulation. 1297 /* Shadow copy of register state. Committed on successful emulation.
@@ -1267,7 +1308,7 @@ x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
1267 if (c->rep_prefix && (c->d & String)) { 1308 if (c->rep_prefix && (c->d & String)) {
1268 /* All REP prefixes have the same first termination condition */ 1309 /* All REP prefixes have the same first termination condition */
1269 if (c->regs[VCPU_REGS_RCX] == 0) { 1310 if (c->regs[VCPU_REGS_RCX] == 0) {
1270 ctxt->vcpu->arch.rip = c->eip; 1311 kvm_rip_write(ctxt->vcpu, c->eip);
1271 goto done; 1312 goto done;
1272 } 1313 }
1273 /* The second termination condition only applies for REPE 1314 /* The second termination condition only applies for REPE
@@ -1281,17 +1322,17 @@ x86_emulate_insn(struct x86_emulate_ctxt *ctxt, struct x86_emulate_ops *ops)
1281 (c->b == 0xae) || (c->b == 0xaf)) { 1322 (c->b == 0xae) || (c->b == 0xaf)) {
1282 if ((c->rep_prefix == REPE_PREFIX) && 1323 if ((c->rep_prefix == REPE_PREFIX) &&
1283 ((ctxt->eflags & EFLG_ZF) == 0)) { 1324 ((ctxt->eflags & EFLG_ZF) == 0)) {
1284 ctxt->vcpu->arch.rip = c->eip; 1325 kvm_rip_write(ctxt->vcpu, c->eip);
1285 goto done; 1326 goto done;
1286 } 1327 }
1287 if ((c->rep_prefix == REPNE_PREFIX) && 1328 if ((c->rep_prefix == REPNE_PREFIX) &&
1288 ((ctxt->eflags & EFLG_ZF) == EFLG_ZF)) { 1329 ((ctxt->eflags & EFLG_ZF) == EFLG_ZF)) {
1289 ctxt->vcpu->arch.rip = c->eip; 1330 kvm_rip_write(ctxt->vcpu, c->eip);
1290 goto done; 1331 goto done;
1291 } 1332 }
1292 } 1333 }
1293 c->regs[VCPU_REGS_RCX]--; 1334 c->regs[VCPU_REGS_RCX]--;
1294 c->eip = ctxt->vcpu->arch.rip; 1335 c->eip = kvm_rip_read(ctxt->vcpu);
1295 } 1336 }
1296 1337
1297 if (c->src.type == OP_MEM) { 1338 if (c->src.type == OP_MEM) {
@@ -1351,27 +1392,10 @@ special_insn:
1351 sbb: /* sbb */ 1392 sbb: /* sbb */
1352 emulate_2op_SrcV("sbb", c->src, c->dst, ctxt->eflags); 1393 emulate_2op_SrcV("sbb", c->src, c->dst, ctxt->eflags);
1353 break; 1394 break;
1354 case 0x20 ... 0x23: 1395 case 0x20 ... 0x25:
1355 and: /* and */ 1396 and: /* and */
1356 emulate_2op_SrcV("and", c->src, c->dst, ctxt->eflags); 1397 emulate_2op_SrcV("and", c->src, c->dst, ctxt->eflags);
1357 break; 1398 break;
1358 case 0x24: /* and al imm8 */
1359 c->dst.type = OP_REG;
1360 c->dst.ptr = &c->regs[VCPU_REGS_RAX];
1361 c->dst.val = *(u8 *)c->dst.ptr;
1362 c->dst.bytes = 1;
1363 c->dst.orig_val = c->dst.val;
1364 goto and;
1365 case 0x25: /* and ax imm16, or eax imm32 */
1366 c->dst.type = OP_REG;
1367 c->dst.bytes = c->op_bytes;
1368 c->dst.ptr = &c->regs[VCPU_REGS_RAX];
1369 if (c->op_bytes == 2)
1370 c->dst.val = *(u16 *)c->dst.ptr;
1371 else
1372 c->dst.val = *(u32 *)c->dst.ptr;
1373 c->dst.orig_val = c->dst.val;
1374 goto and;
1375 case 0x28 ... 0x2d: 1399 case 0x28 ... 0x2d:
1376 sub: /* sub */ 1400 sub: /* sub */
1377 emulate_2op_SrcV("sub", c->src, c->dst, ctxt->eflags); 1401 emulate_2op_SrcV("sub", c->src, c->dst, ctxt->eflags);
@@ -1659,7 +1683,7 @@ special_insn:
1659 case 0xae ... 0xaf: /* scas */ 1683 case 0xae ... 0xaf: /* scas */
1660 DPRINTF("Urk! I don't handle SCAS.\n"); 1684 DPRINTF("Urk! I don't handle SCAS.\n");
1661 goto cannot_emulate; 1685 goto cannot_emulate;
1662 case 0xb8: /* mov r, imm */ 1686 case 0xb0 ... 0xbf: /* mov r, imm */
1663 goto mov; 1687 goto mov;
1664 case 0xc0 ... 0xc1: 1688 case 0xc0 ... 0xc1:
1665 emulate_grp2(ctxt); 1689 emulate_grp2(ctxt);
@@ -1679,6 +1703,16 @@ special_insn:
1679 c->src.val = c->regs[VCPU_REGS_RCX]; 1703 c->src.val = c->regs[VCPU_REGS_RCX];
1680 emulate_grp2(ctxt); 1704 emulate_grp2(ctxt);
1681 break; 1705 break;
1706 case 0xe4: /* inb */
1707 case 0xe5: /* in */
1708 port = insn_fetch(u8, 1, c->eip);
1709 io_dir_in = 1;
1710 goto do_io;
1711 case 0xe6: /* outb */
1712 case 0xe7: /* out */
1713 port = insn_fetch(u8, 1, c->eip);
1714 io_dir_in = 0;
1715 goto do_io;
1682 case 0xe8: /* call (near) */ { 1716 case 0xe8: /* call (near) */ {
1683 long int rel; 1717 long int rel;
1684 switch (c->op_bytes) { 1718 switch (c->op_bytes) {
@@ -1729,6 +1763,22 @@ special_insn:
1729 jmp_rel(c, c->src.val); 1763 jmp_rel(c, c->src.val);
1730 c->dst.type = OP_NONE; /* Disable writeback. */ 1764 c->dst.type = OP_NONE; /* Disable writeback. */
1731 break; 1765 break;
1766 case 0xec: /* in al,dx */
1767 case 0xed: /* in (e/r)ax,dx */
1768 port = c->regs[VCPU_REGS_RDX];
1769 io_dir_in = 1;
1770 goto do_io;
1771 case 0xee: /* out al,dx */
1772 case 0xef: /* out (e/r)ax,dx */
1773 port = c->regs[VCPU_REGS_RDX];
1774 io_dir_in = 0;
1775 do_io: if (kvm_emulate_pio(ctxt->vcpu, NULL, io_dir_in,
1776 (c->d & ByteOp) ? 1 : c->op_bytes,
1777 port) != 0) {
1778 c->eip = saved_eip;
1779 goto cannot_emulate;
1780 }
1781 return 0;
1732 case 0xf4: /* hlt */ 1782 case 0xf4: /* hlt */
1733 ctxt->vcpu->arch.halt_request = 1; 1783 ctxt->vcpu->arch.halt_request = 1;
1734 break; 1784 break;
@@ -1754,6 +1804,14 @@ special_insn:
1754 ctxt->eflags |= X86_EFLAGS_IF; 1804 ctxt->eflags |= X86_EFLAGS_IF;
1755 c->dst.type = OP_NONE; /* Disable writeback. */ 1805 c->dst.type = OP_NONE; /* Disable writeback. */
1756 break; 1806 break;
1807 case 0xfc: /* cld */
1808 ctxt->eflags &= ~EFLG_DF;
1809 c->dst.type = OP_NONE; /* Disable writeback. */
1810 break;
1811 case 0xfd: /* std */
1812 ctxt->eflags |= EFLG_DF;
1813 c->dst.type = OP_NONE; /* Disable writeback. */
1814 break;
1757 case 0xfe ... 0xff: /* Grp4/Grp5 */ 1815 case 0xfe ... 0xff: /* Grp4/Grp5 */
1758 rc = emulate_grp45(ctxt, ops); 1816 rc = emulate_grp45(ctxt, ops);
1759 if (rc != 0) 1817 if (rc != 0)
@@ -1768,7 +1826,7 @@ writeback:
1768 1826
1769 /* Commit shadow register state. */ 1827 /* Commit shadow register state. */
1770 memcpy(ctxt->vcpu->arch.regs, c->regs, sizeof c->regs); 1828 memcpy(ctxt->vcpu->arch.regs, c->regs, sizeof c->regs);
1771 ctxt->vcpu->arch.rip = c->eip; 1829 kvm_rip_write(ctxt->vcpu, c->eip);
1772 1830
1773done: 1831done:
1774 if (rc == X86EMUL_UNHANDLEABLE) { 1832 if (rc == X86EMUL_UNHANDLEABLE) {
@@ -1793,7 +1851,7 @@ twobyte_insn:
1793 goto done; 1851 goto done;
1794 1852
1795 /* Let the processor re-execute the fixed hypercall */ 1853 /* Let the processor re-execute the fixed hypercall */
1796 c->eip = ctxt->vcpu->arch.rip; 1854 c->eip = kvm_rip_read(ctxt->vcpu);
1797 /* Disable writeback. */ 1855 /* Disable writeback. */
1798 c->dst.type = OP_NONE; 1856 c->dst.type = OP_NONE;
1799 break; 1857 break;
@@ -1889,7 +1947,7 @@ twobyte_insn:
1889 rc = kvm_set_msr(ctxt->vcpu, c->regs[VCPU_REGS_RCX], msr_data); 1947 rc = kvm_set_msr(ctxt->vcpu, c->regs[VCPU_REGS_RCX], msr_data);
1890 if (rc) { 1948 if (rc) {
1891 kvm_inject_gp(ctxt->vcpu, 0); 1949 kvm_inject_gp(ctxt->vcpu, 0);
1892 c->eip = ctxt->vcpu->arch.rip; 1950 c->eip = kvm_rip_read(ctxt->vcpu);
1893 } 1951 }
1894 rc = X86EMUL_CONTINUE; 1952 rc = X86EMUL_CONTINUE;
1895 c->dst.type = OP_NONE; 1953 c->dst.type = OP_NONE;
@@ -1899,7 +1957,7 @@ twobyte_insn:
1899 rc = kvm_get_msr(ctxt->vcpu, c->regs[VCPU_REGS_RCX], &msr_data); 1957 rc = kvm_get_msr(ctxt->vcpu, c->regs[VCPU_REGS_RCX], &msr_data);
1900 if (rc) { 1958 if (rc) {
1901 kvm_inject_gp(ctxt->vcpu, 0); 1959 kvm_inject_gp(ctxt->vcpu, 0);
1902 c->eip = ctxt->vcpu->arch.rip; 1960 c->eip = kvm_rip_read(ctxt->vcpu);
1903 } else { 1961 } else {
1904 c->regs[VCPU_REGS_RAX] = (u32)msr_data; 1962 c->regs[VCPU_REGS_RAX] = (u32)msr_data;
1905 c->regs[VCPU_REGS_RDX] = msr_data >> 32; 1963 c->regs[VCPU_REGS_RDX] = msr_data >> 32;
diff --git a/arch/x86/lguest/boot.c b/arch/x86/lguest/boot.c
index d9249a882aa5..65f0b8a47bed 100644
--- a/arch/x86/lguest/boot.c
+++ b/arch/x86/lguest/boot.c
@@ -55,6 +55,7 @@
55#include <linux/lguest_launcher.h> 55#include <linux/lguest_launcher.h>
56#include <linux/virtio_console.h> 56#include <linux/virtio_console.h>
57#include <linux/pm.h> 57#include <linux/pm.h>
58#include <asm/apic.h>
58#include <asm/lguest.h> 59#include <asm/lguest.h>
59#include <asm/paravirt.h> 60#include <asm/paravirt.h>
60#include <asm/param.h> 61#include <asm/param.h>
@@ -783,14 +784,44 @@ static void lguest_wbinvd(void)
783 * code qualifies for Advanced. It will also never interrupt anything. It 784 * code qualifies for Advanced. It will also never interrupt anything. It
784 * does, however, allow us to get through the Linux boot code. */ 785 * does, however, allow us to get through the Linux boot code. */
785#ifdef CONFIG_X86_LOCAL_APIC 786#ifdef CONFIG_X86_LOCAL_APIC
786static void lguest_apic_write(unsigned long reg, u32 v) 787static void lguest_apic_write(u32 reg, u32 v)
787{ 788{
788} 789}
789 790
790static u32 lguest_apic_read(unsigned long reg) 791static u32 lguest_apic_read(u32 reg)
791{ 792{
792 return 0; 793 return 0;
793} 794}
795
796static u64 lguest_apic_icr_read(void)
797{
798 return 0;
799}
800
801static void lguest_apic_icr_write(u32 low, u32 id)
802{
803 /* Warn to see if there's any stray references */
804 WARN_ON(1);
805}
806
807static void lguest_apic_wait_icr_idle(void)
808{
809 return;
810}
811
812static u32 lguest_apic_safe_wait_icr_idle(void)
813{
814 return 0;
815}
816
817static struct apic_ops lguest_basic_apic_ops = {
818 .read = lguest_apic_read,
819 .write = lguest_apic_write,
820 .icr_read = lguest_apic_icr_read,
821 .icr_write = lguest_apic_icr_write,
822 .wait_icr_idle = lguest_apic_wait_icr_idle,
823 .safe_wait_icr_idle = lguest_apic_safe_wait_icr_idle,
824};
794#endif 825#endif
795 826
796/* STOP! Until an interrupt comes in. */ 827/* STOP! Until an interrupt comes in. */
@@ -990,8 +1021,7 @@ __init void lguest_init(void)
990 1021
991#ifdef CONFIG_X86_LOCAL_APIC 1022#ifdef CONFIG_X86_LOCAL_APIC
992 /* apic read/write intercepts */ 1023 /* apic read/write intercepts */
993 pv_apic_ops.apic_write = lguest_apic_write; 1024 apic_ops = &lguest_basic_apic_ops;
994 pv_apic_ops.apic_read = lguest_apic_read;
995#endif 1025#endif
996 1026
997 /* time operations */ 1027 /* time operations */
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index aa3fa4119424..55e11aa6d66c 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -17,9 +17,6 @@ ifeq ($(CONFIG_X86_32),y)
17 lib-$(CONFIG_X86_USE_3DNOW) += mmx_32.o 17 lib-$(CONFIG_X86_USE_3DNOW) += mmx_32.o
18else 18else
19 obj-y += io_64.o iomap_copy_64.o 19 obj-y += io_64.o iomap_copy_64.o
20
21 CFLAGS_csum-partial_64.o := -funroll-loops
22
23 lib-y += csum-partial_64.o csum-copy_64.o csum-wrappers_64.o 20 lib-y += csum-partial_64.o csum-copy_64.o csum-wrappers_64.o
24 lib-y += thunk_64.o clear_page_64.o copy_page_64.o 21 lib-y += thunk_64.o clear_page_64.o copy_page_64.o
25 lib-y += memmove_64.o memset_64.o 22 lib-y += memmove_64.o memset_64.o
diff --git a/arch/x86/lib/msr-on-cpu.c b/arch/x86/lib/msr-on-cpu.c
index 01b868ba82f8..321cf720dbb6 100644
--- a/arch/x86/lib/msr-on-cpu.c
+++ b/arch/x86/lib/msr-on-cpu.c
@@ -16,37 +16,46 @@ static void __rdmsr_on_cpu(void *info)
16 rdmsr(rv->msr_no, rv->l, rv->h); 16 rdmsr(rv->msr_no, rv->l, rv->h);
17} 17}
18 18
19static void __rdmsr_safe_on_cpu(void *info) 19static void __wrmsr_on_cpu(void *info)
20{ 20{
21 struct msr_info *rv = info; 21 struct msr_info *rv = info;
22 22
23 rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h); 23 wrmsr(rv->msr_no, rv->l, rv->h);
24} 24}
25 25
26static int _rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h, int safe) 26int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
27{ 27{
28 int err = 0; 28 int err;
29 struct msr_info rv; 29 struct msr_info rv;
30 30
31 rv.msr_no = msr_no; 31 rv.msr_no = msr_no;
32 if (safe) { 32 err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
33 err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu,
34 &rv, 1);
35 err = err ? err : rv.err;
36 } else {
37 err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
38 }
39 *l = rv.l; 33 *l = rv.l;
40 *h = rv.h; 34 *h = rv.h;
41 35
42 return err; 36 return err;
43} 37}
44 38
45static void __wrmsr_on_cpu(void *info) 39int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
40{
41 int err;
42 struct msr_info rv;
43
44 rv.msr_no = msr_no;
45 rv.l = l;
46 rv.h = h;
47 err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
48
49 return err;
50}
51
52/* These "safe" variants are slower and should be used when the target MSR
53 may not actually exist. */
54static void __rdmsr_safe_on_cpu(void *info)
46{ 55{
47 struct msr_info *rv = info; 56 struct msr_info *rv = info;
48 57
49 wrmsr(rv->msr_no, rv->l, rv->h); 58 rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
50} 59}
51 60
52static void __wrmsr_safe_on_cpu(void *info) 61static void __wrmsr_safe_on_cpu(void *info)
@@ -56,45 +65,30 @@ static void __wrmsr_safe_on_cpu(void *info)
56 rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h); 65 rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
57} 66}
58 67
59static int _wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h, int safe) 68int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
60{ 69{
61 int err = 0; 70 int err;
62 struct msr_info rv; 71 struct msr_info rv;
63 72
64 rv.msr_no = msr_no; 73 rv.msr_no = msr_no;
65 rv.l = l; 74 err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
66 rv.h = h; 75 *l = rv.l;
67 if (safe) { 76 *h = rv.h;
68 err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu,
69 &rv, 1);
70 err = err ? err : rv.err;
71 } else {
72 err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
73 }
74
75 return err;
76}
77 77
78int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h) 78 return err ? err : rv.err;
79{
80 return _wrmsr_on_cpu(cpu, msr_no, l, h, 0);
81} 79}
82 80
83int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
84{
85 return _rdmsr_on_cpu(cpu, msr_no, l, h, 0);
86}
87
88/* These "safe" variants are slower and should be used when the target MSR
89 may not actually exist. */
90int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h) 81int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
91{ 82{
92 return _wrmsr_on_cpu(cpu, msr_no, l, h, 1); 83 int err;
93} 84 struct msr_info rv;
94 85
95int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h) 86 rv.msr_no = msr_no;
96{ 87 rv.l = l;
97 return _rdmsr_on_cpu(cpu, msr_no, l, h, 1); 88 rv.h = h;
89 err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
90
91 return err ? err : rv.err;
98} 92}
99 93
100EXPORT_SYMBOL(rdmsr_on_cpu); 94EXPORT_SYMBOL(rdmsr_on_cpu);
diff --git a/arch/x86/lib/string_32.c b/arch/x86/lib/string_32.c
index 94972e7c094d..82004d2bf05e 100644
--- a/arch/x86/lib/string_32.c
+++ b/arch/x86/lib/string_32.c
@@ -22,7 +22,7 @@ char *strcpy(char *dest, const char *src)
22 "testb %%al,%%al\n\t" 22 "testb %%al,%%al\n\t"
23 "jne 1b" 23 "jne 1b"
24 : "=&S" (d0), "=&D" (d1), "=&a" (d2) 24 : "=&S" (d0), "=&D" (d1), "=&a" (d2)
25 :"0" (src), "1" (dest) : "memory"); 25 : "0" (src), "1" (dest) : "memory");
26 return dest; 26 return dest;
27} 27}
28EXPORT_SYMBOL(strcpy); 28EXPORT_SYMBOL(strcpy);
@@ -42,7 +42,7 @@ char *strncpy(char *dest, const char *src, size_t count)
42 "stosb\n" 42 "stosb\n"
43 "2:" 43 "2:"
44 : "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3) 44 : "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3)
45 :"0" (src), "1" (dest), "2" (count) : "memory"); 45 : "0" (src), "1" (dest), "2" (count) : "memory");
46 return dest; 46 return dest;
47} 47}
48EXPORT_SYMBOL(strncpy); 48EXPORT_SYMBOL(strncpy);
@@ -60,7 +60,7 @@ char *strcat(char *dest, const char *src)
60 "testb %%al,%%al\n\t" 60 "testb %%al,%%al\n\t"
61 "jne 1b" 61 "jne 1b"
62 : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3) 62 : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
63 : "0" (src), "1" (dest), "2" (0), "3" (0xffffffffu): "memory"); 63 : "0" (src), "1" (dest), "2" (0), "3" (0xffffffffu) : "memory");
64 return dest; 64 return dest;
65} 65}
66EXPORT_SYMBOL(strcat); 66EXPORT_SYMBOL(strcat);
@@ -105,9 +105,9 @@ int strcmp(const char *cs, const char *ct)
105 "2:\tsbbl %%eax,%%eax\n\t" 105 "2:\tsbbl %%eax,%%eax\n\t"
106 "orb $1,%%al\n" 106 "orb $1,%%al\n"
107 "3:" 107 "3:"
108 :"=a" (res), "=&S" (d0), "=&D" (d1) 108 : "=a" (res), "=&S" (d0), "=&D" (d1)
109 :"1" (cs), "2" (ct) 109 : "1" (cs), "2" (ct)
110 :"memory"); 110 : "memory");
111 return res; 111 return res;
112} 112}
113EXPORT_SYMBOL(strcmp); 113EXPORT_SYMBOL(strcmp);
@@ -130,9 +130,9 @@ int strncmp(const char *cs, const char *ct, size_t count)
130 "3:\tsbbl %%eax,%%eax\n\t" 130 "3:\tsbbl %%eax,%%eax\n\t"
131 "orb $1,%%al\n" 131 "orb $1,%%al\n"
132 "4:" 132 "4:"
133 :"=a" (res), "=&S" (d0), "=&D" (d1), "=&c" (d2) 133 : "=a" (res), "=&S" (d0), "=&D" (d1), "=&c" (d2)
134 :"1" (cs), "2" (ct), "3" (count) 134 : "1" (cs), "2" (ct), "3" (count)
135 :"memory"); 135 : "memory");
136 return res; 136 return res;
137} 137}
138EXPORT_SYMBOL(strncmp); 138EXPORT_SYMBOL(strncmp);
@@ -152,9 +152,9 @@ char *strchr(const char *s, int c)
152 "movl $1,%1\n" 152 "movl $1,%1\n"
153 "2:\tmovl %1,%0\n\t" 153 "2:\tmovl %1,%0\n\t"
154 "decl %0" 154 "decl %0"
155 :"=a" (res), "=&S" (d0) 155 : "=a" (res), "=&S" (d0)
156 :"1" (s), "0" (c) 156 : "1" (s), "0" (c)
157 :"memory"); 157 : "memory");
158 return res; 158 return res;
159} 159}
160EXPORT_SYMBOL(strchr); 160EXPORT_SYMBOL(strchr);
@@ -169,9 +169,9 @@ size_t strlen(const char *s)
169 "scasb\n\t" 169 "scasb\n\t"
170 "notl %0\n\t" 170 "notl %0\n\t"
171 "decl %0" 171 "decl %0"
172 :"=c" (res), "=&D" (d0) 172 : "=c" (res), "=&D" (d0)
173 :"1" (s), "a" (0), "0" (0xffffffffu) 173 : "1" (s), "a" (0), "0" (0xffffffffu)
174 :"memory"); 174 : "memory");
175 return res; 175 return res;
176} 176}
177EXPORT_SYMBOL(strlen); 177EXPORT_SYMBOL(strlen);
@@ -189,9 +189,9 @@ void *memchr(const void *cs, int c, size_t count)
189 "je 1f\n\t" 189 "je 1f\n\t"
190 "movl $1,%0\n" 190 "movl $1,%0\n"
191 "1:\tdecl %0" 191 "1:\tdecl %0"
192 :"=D" (res), "=&c" (d0) 192 : "=D" (res), "=&c" (d0)
193 :"a" (c), "0" (cs), "1" (count) 193 : "a" (c), "0" (cs), "1" (count)
194 :"memory"); 194 : "memory");
195 return res; 195 return res;
196} 196}
197EXPORT_SYMBOL(memchr); 197EXPORT_SYMBOL(memchr);
@@ -228,9 +228,9 @@ size_t strnlen(const char *s, size_t count)
228 "cmpl $-1,%1\n\t" 228 "cmpl $-1,%1\n\t"
229 "jne 1b\n" 229 "jne 1b\n"
230 "3:\tsubl %2,%0" 230 "3:\tsubl %2,%0"
231 :"=a" (res), "=&d" (d0) 231 : "=a" (res), "=&d" (d0)
232 :"c" (s), "1" (count) 232 : "c" (s), "1" (count)
233 :"memory"); 233 : "memory");
234 return res; 234 return res;
235} 235}
236EXPORT_SYMBOL(strnlen); 236EXPORT_SYMBOL(strnlen);
diff --git a/arch/x86/lib/strstr_32.c b/arch/x86/lib/strstr_32.c
index 42e8a50303f3..8e2d55f754bf 100644
--- a/arch/x86/lib/strstr_32.c
+++ b/arch/x86/lib/strstr_32.c
@@ -23,9 +23,9 @@ __asm__ __volatile__(
23 "jne 1b\n\t" 23 "jne 1b\n\t"
24 "xorl %%eax,%%eax\n\t" 24 "xorl %%eax,%%eax\n\t"
25 "2:" 25 "2:"
26 :"=a" (__res), "=&c" (d0), "=&S" (d1) 26 : "=a" (__res), "=&c" (d0), "=&S" (d1)
27 :"0" (0), "1" (0xffffffff), "2" (cs), "g" (ct) 27 : "0" (0), "1" (0xffffffff), "2" (cs), "g" (ct)
28 :"dx", "di"); 28 : "dx", "di");
29return __res; 29return __res;
30} 30}
31 31
diff --git a/arch/x86/lib/usercopy_32.c b/arch/x86/lib/usercopy_32.c
index 24e60944971a..9e68075544f6 100644
--- a/arch/x86/lib/usercopy_32.c
+++ b/arch/x86/lib/usercopy_32.c
@@ -14,6 +14,13 @@
14#include <asm/uaccess.h> 14#include <asm/uaccess.h>
15#include <asm/mmx.h> 15#include <asm/mmx.h>
16 16
17#ifdef CONFIG_X86_INTEL_USERCOPY
18/*
19 * Alignment at which movsl is preferred for bulk memory copies.
20 */
21struct movsl_mask movsl_mask __read_mostly;
22#endif
23
17static inline int __movsl_is_ok(unsigned long a1, unsigned long a2, unsigned long n) 24static inline int __movsl_is_ok(unsigned long a1, unsigned long a2, unsigned long n)
18{ 25{
19#ifdef CONFIG_X86_INTEL_USERCOPY 26#ifdef CONFIG_X86_INTEL_USERCOPY
diff --git a/arch/x86/mach-default/setup.c b/arch/x86/mach-default/setup.c
index 3d317836be9e..37b9ae4d44c5 100644
--- a/arch/x86/mach-default/setup.c
+++ b/arch/x86/mach-default/setup.c
@@ -10,13 +10,15 @@
10#include <asm/e820.h> 10#include <asm/e820.h>
11#include <asm/setup.h> 11#include <asm/setup.h>
12 12
13#include <mach_ipi.h>
14
13#ifdef CONFIG_HOTPLUG_CPU 15#ifdef CONFIG_HOTPLUG_CPU
14#define DEFAULT_SEND_IPI (1) 16#define DEFAULT_SEND_IPI (1)
15#else 17#else
16#define DEFAULT_SEND_IPI (0) 18#define DEFAULT_SEND_IPI (0)
17#endif 19#endif
18 20
19int no_broadcast=DEFAULT_SEND_IPI; 21int no_broadcast = DEFAULT_SEND_IPI;
20 22
21/** 23/**
22 * pre_intr_init_hook - initialisation prior to setting up interrupt vectors 24 * pre_intr_init_hook - initialisation prior to setting up interrupt vectors
@@ -36,15 +38,6 @@ void __init pre_intr_init_hook(void)
36 init_ISA_irqs(); 38 init_ISA_irqs();
37} 39}
38 40
39/*
40 * IRQ2 is cascade interrupt to second interrupt controller
41 */
42static struct irqaction irq2 = {
43 .handler = no_action,
44 .mask = CPU_MASK_NONE,
45 .name = "cascade",
46};
47
48/** 41/**
49 * intr_init_hook - post gate setup interrupt initialisation 42 * intr_init_hook - post gate setup interrupt initialisation
50 * 43 *
@@ -60,12 +53,6 @@ void __init intr_init_hook(void)
60 if (x86_quirks->arch_intr_init()) 53 if (x86_quirks->arch_intr_init())
61 return; 54 return;
62 } 55 }
63#ifdef CONFIG_X86_LOCAL_APIC
64 apic_intr_init();
65#endif
66
67 if (!acpi_ioapic)
68 setup_irq(2, &irq2);
69} 56}
70 57
71/** 58/**
diff --git a/arch/x86/mach-es7000/Makefile b/arch/x86/mach-es7000/Makefile
deleted file mode 100644
index 3ef8b43b62fc..000000000000
--- a/arch/x86/mach-es7000/Makefile
+++ /dev/null
@@ -1,5 +0,0 @@
1#
2# Makefile for the linux kernel.
3#
4
5obj-$(CONFIG_X86_ES7000) := es7000plat.o
diff --git a/arch/x86/mach-es7000/es7000.h b/arch/x86/mach-es7000/es7000.h
deleted file mode 100644
index c8d5aa132fa0..000000000000
--- a/arch/x86/mach-es7000/es7000.h
+++ /dev/null
@@ -1,114 +0,0 @@
1/*
2 * Written by: Garry Forsgren, Unisys Corporation
3 * Natalie Protasevich, Unisys Corporation
4 * This file contains the code to configure and interface
5 * with Unisys ES7000 series hardware system manager.
6 *
7 * Copyright (c) 2003 Unisys Corporation. All Rights Reserved.
8 *
9 * This program is free software; you can redistribute it and/or modify it
10 * under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it would be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, write the Free Software Foundation, Inc., 59
19 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
20 *
21 * Contact information: Unisys Corporation, Township Line & Union Meeting
22 * Roads-A, Unisys Way, Blue Bell, Pennsylvania, 19424, or:
23 *
24 * http://www.unisys.com
25 */
26
27/*
28 * ES7000 chipsets
29 */
30
31#define NON_UNISYS 0
32#define ES7000_CLASSIC 1
33#define ES7000_ZORRO 2
34
35
36#define MIP_REG 1
37#define MIP_PSAI_REG 4
38
39#define MIP_BUSY 1
40#define MIP_SPIN 0xf0000
41#define MIP_VALID 0x0100000000000000ULL
42#define MIP_PORT(VALUE) ((VALUE >> 32) & 0xffff)
43
44#define MIP_RD_LO(VALUE) (VALUE & 0xffffffff)
45
46struct mip_reg_info {
47 unsigned long long mip_info;
48 unsigned long long delivery_info;
49 unsigned long long host_reg;
50 unsigned long long mip_reg;
51};
52
53struct part_info {
54 unsigned char type;
55 unsigned char length;
56 unsigned char part_id;
57 unsigned char apic_mode;
58 unsigned long snum;
59 char ptype[16];
60 char sname[64];
61 char pname[64];
62};
63
64struct psai {
65 unsigned long long entry_type;
66 unsigned long long addr;
67 unsigned long long bep_addr;
68};
69
70struct es7000_mem_info {
71 unsigned char type;
72 unsigned char length;
73 unsigned char resv[6];
74 unsigned long long start;
75 unsigned long long size;
76};
77
78struct es7000_oem_table {
79 unsigned long long hdr;
80 struct mip_reg_info mip;
81 struct part_info pif;
82 struct es7000_mem_info shm;
83 struct psai psai;
84};
85
86#ifdef CONFIG_ACPI
87
88struct oem_table {
89 struct acpi_table_header Header;
90 u32 OEMTableAddr;
91 u32 OEMTableSize;
92};
93
94extern int find_unisys_acpi_oem_table(unsigned long *oem_addr);
95#endif
96
97struct mip_reg {
98 unsigned long long off_0;
99 unsigned long long off_8;
100 unsigned long long off_10;
101 unsigned long long off_18;
102 unsigned long long off_20;
103 unsigned long long off_28;
104 unsigned long long off_30;
105 unsigned long long off_38;
106};
107
108#define MIP_SW_APIC 0x1020b
109#define MIP_FUNC(VALUE) (VALUE & 0xff)
110
111extern int parse_unisys_oem (char *oemptr);
112extern void setup_unisys(void);
113extern int es7000_start_cpu(int cpu, unsigned long eip);
114extern void es7000_sw_apic(void);
diff --git a/arch/x86/mach-generic/Makefile b/arch/x86/mach-generic/Makefile
index 0dbd7803a1d5..6730f4e7c744 100644
--- a/arch/x86/mach-generic/Makefile
+++ b/arch/x86/mach-generic/Makefile
@@ -9,4 +9,3 @@ obj-$(CONFIG_X86_NUMAQ) += numaq.o
9obj-$(CONFIG_X86_SUMMIT) += summit.o 9obj-$(CONFIG_X86_SUMMIT) += summit.o
10obj-$(CONFIG_X86_BIGSMP) += bigsmp.o 10obj-$(CONFIG_X86_BIGSMP) += bigsmp.o
11obj-$(CONFIG_X86_ES7000) += es7000.o 11obj-$(CONFIG_X86_ES7000) += es7000.o
12obj-$(CONFIG_X86_ES7000) += ../../x86/mach-es7000/
diff --git a/arch/x86/mach-generic/bigsmp.c b/arch/x86/mach-generic/bigsmp.c
index 59d771714559..df37fc9d6a26 100644
--- a/arch/x86/mach-generic/bigsmp.c
+++ b/arch/x86/mach-generic/bigsmp.c
@@ -5,18 +5,17 @@
5#define APIC_DEFINITION 1 5#define APIC_DEFINITION 1
6#include <linux/threads.h> 6#include <linux/threads.h>
7#include <linux/cpumask.h> 7#include <linux/cpumask.h>
8#include <asm/smp.h>
9#include <asm/mpspec.h> 8#include <asm/mpspec.h>
10#include <asm/genapic.h> 9#include <asm/genapic.h>
11#include <asm/fixmap.h> 10#include <asm/fixmap.h>
12#include <asm/apicdef.h> 11#include <asm/apicdef.h>
13#include <linux/kernel.h> 12#include <linux/kernel.h>
14#include <linux/smp.h>
15#include <linux/init.h> 13#include <linux/init.h>
16#include <linux/dmi.h> 14#include <linux/dmi.h>
17#include <asm/mach-bigsmp/mach_apic.h> 15#include <asm/bigsmp/apicdef.h>
18#include <asm/mach-bigsmp/mach_apicdef.h> 16#include <linux/smp.h>
19#include <asm/mach-bigsmp/mach_ipi.h> 17#include <asm/bigsmp/apic.h>
18#include <asm/bigsmp/ipi.h>
20#include <asm/mach-default/mach_mpparse.h> 19#include <asm/mach-default/mach_mpparse.h>
21 20
22static int dmi_bigsmp; /* can be set by dmi scanners */ 21static int dmi_bigsmp; /* can be set by dmi scanners */
diff --git a/arch/x86/mach-generic/es7000.c b/arch/x86/mach-generic/es7000.c
index 4742626f08c4..6513d41ea21e 100644
--- a/arch/x86/mach-generic/es7000.c
+++ b/arch/x86/mach-generic/es7000.c
@@ -4,20 +4,19 @@
4#define APIC_DEFINITION 1 4#define APIC_DEFINITION 1
5#include <linux/threads.h> 5#include <linux/threads.h>
6#include <linux/cpumask.h> 6#include <linux/cpumask.h>
7#include <asm/smp.h>
8#include <asm/mpspec.h> 7#include <asm/mpspec.h>
9#include <asm/genapic.h> 8#include <asm/genapic.h>
10#include <asm/fixmap.h> 9#include <asm/fixmap.h>
11#include <asm/apicdef.h> 10#include <asm/apicdef.h>
12#include <linux/kernel.h> 11#include <linux/kernel.h>
13#include <linux/string.h> 12#include <linux/string.h>
14#include <linux/smp.h>
15#include <linux/init.h> 13#include <linux/init.h>
16#include <asm/mach-es7000/mach_apicdef.h> 14#include <asm/es7000/apicdef.h>
17#include <asm/mach-es7000/mach_apic.h> 15#include <linux/smp.h>
18#include <asm/mach-es7000/mach_ipi.h> 16#include <asm/es7000/apic.h>
19#include <asm/mach-es7000/mach_mpparse.h> 17#include <asm/es7000/ipi.h>
20#include <asm/mach-es7000/mach_wakecpu.h> 18#include <asm/es7000/mpparse.h>
19#include <asm/es7000/wakecpu.h>
21 20
22static int probe_es7000(void) 21static int probe_es7000(void)
23{ 22{
@@ -48,16 +47,26 @@ static __init int mps_oem_check(struct mp_config_table *mpc, char *oem,
48/* Hook from generic ACPI tables.c */ 47/* Hook from generic ACPI tables.c */
49static int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id) 48static int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id)
50{ 49{
51 unsigned long oem_addr; 50 unsigned long oem_addr = 0;
51 int check_dsdt;
52 int ret = 0;
53
54 /* check dsdt at first to avoid clear fix_map for oem_addr */
55 check_dsdt = es7000_check_dsdt();
56
52 if (!find_unisys_acpi_oem_table(&oem_addr)) { 57 if (!find_unisys_acpi_oem_table(&oem_addr)) {
53 if (es7000_check_dsdt()) 58 if (check_dsdt)
54 return parse_unisys_oem((char *)oem_addr); 59 ret = parse_unisys_oem((char *)oem_addr);
55 else { 60 else {
56 setup_unisys(); 61 setup_unisys();
57 return 1; 62 ret = 1;
58 } 63 }
64 /*
65 * we need to unmap it
66 */
67 unmap_unisys_acpi_oem_table(oem_addr);
59 } 68 }
60 return 0; 69 return ret;
61} 70}
62#else 71#else
63static int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id) 72static int __init acpi_madt_oem_check(char *oem_id, char *oem_table_id)
diff --git a/arch/x86/mach-generic/numaq.c b/arch/x86/mach-generic/numaq.c
index 8091e68764c4..8cf58394975e 100644
--- a/arch/x86/mach-generic/numaq.c
+++ b/arch/x86/mach-generic/numaq.c
@@ -4,7 +4,6 @@
4#define APIC_DEFINITION 1 4#define APIC_DEFINITION 1
5#include <linux/threads.h> 5#include <linux/threads.h>
6#include <linux/cpumask.h> 6#include <linux/cpumask.h>
7#include <linux/smp.h>
8#include <asm/mpspec.h> 7#include <asm/mpspec.h>
9#include <asm/genapic.h> 8#include <asm/genapic.h>
10#include <asm/fixmap.h> 9#include <asm/fixmap.h>
@@ -12,11 +11,12 @@
12#include <linux/kernel.h> 11#include <linux/kernel.h>
13#include <linux/string.h> 12#include <linux/string.h>
14#include <linux/init.h> 13#include <linux/init.h>
15#include <asm/mach-numaq/mach_apic.h> 14#include <asm/numaq/apicdef.h>
16#include <asm/mach-numaq/mach_apicdef.h> 15#include <linux/smp.h>
17#include <asm/mach-numaq/mach_ipi.h> 16#include <asm/numaq/apic.h>
18#include <asm/mach-numaq/mach_mpparse.h> 17#include <asm/numaq/ipi.h>
19#include <asm/mach-numaq/mach_wakecpu.h> 18#include <asm/numaq/mpparse.h>
19#include <asm/numaq/wakecpu.h>
20#include <asm/numaq.h> 20#include <asm/numaq.h>
21 21
22static int mps_oem_check(struct mp_config_table *mpc, char *oem, 22static int mps_oem_check(struct mp_config_table *mpc, char *oem,
diff --git a/arch/x86/mach-generic/summit.c b/arch/x86/mach-generic/summit.c
index a97ea0f35b1e..6ad6b67a723d 100644
--- a/arch/x86/mach-generic/summit.c
+++ b/arch/x86/mach-generic/summit.c
@@ -4,19 +4,18 @@
4#define APIC_DEFINITION 1 4#define APIC_DEFINITION 1
5#include <linux/threads.h> 5#include <linux/threads.h>
6#include <linux/cpumask.h> 6#include <linux/cpumask.h>
7#include <asm/smp.h>
8#include <asm/mpspec.h> 7#include <asm/mpspec.h>
9#include <asm/genapic.h> 8#include <asm/genapic.h>
10#include <asm/fixmap.h> 9#include <asm/fixmap.h>
11#include <asm/apicdef.h> 10#include <asm/apicdef.h>
12#include <linux/kernel.h> 11#include <linux/kernel.h>
13#include <linux/string.h> 12#include <linux/string.h>
14#include <linux/smp.h>
15#include <linux/init.h> 13#include <linux/init.h>
16#include <asm/mach-summit/mach_apic.h> 14#include <asm/summit/apicdef.h>
17#include <asm/mach-summit/mach_apicdef.h> 15#include <linux/smp.h>
18#include <asm/mach-summit/mach_ipi.h> 16#include <asm/summit/apic.h>
19#include <asm/mach-summit/mach_mpparse.h> 17#include <asm/summit/ipi.h>
18#include <asm/summit/mpparse.h>
20 19
21static int probe_summit(void) 20static int probe_summit(void)
22{ 21{
diff --git a/arch/x86/mach-voyager/voyager_smp.c b/arch/x86/mach-voyager/voyager_smp.c
index ee0fba092157..199a5f4a873c 100644
--- a/arch/x86/mach-voyager/voyager_smp.c
+++ b/arch/x86/mach-voyager/voyager_smp.c
@@ -448,6 +448,8 @@ static void __init start_secondary(void *unused)
448 448
449 VDEBUG(("VOYAGER SMP: CPU%d, stack at about %p\n", cpuid, &cpuid)); 449 VDEBUG(("VOYAGER SMP: CPU%d, stack at about %p\n", cpuid, &cpuid));
450 450
451 notify_cpu_starting(cpuid);
452
451 /* enable interrupts */ 453 /* enable interrupts */
452 local_irq_enable(); 454 local_irq_enable();
453 455
diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile
index dfb932dcf136..59f89b434b45 100644
--- a/arch/x86/mm/Makefile
+++ b/arch/x86/mm/Makefile
@@ -13,12 +13,8 @@ obj-$(CONFIG_MMIOTRACE) += mmiotrace.o
13mmiotrace-y := pf_in.o mmio-mod.o 13mmiotrace-y := pf_in.o mmio-mod.o
14obj-$(CONFIG_MMIOTRACE_TEST) += testmmiotrace.o 14obj-$(CONFIG_MMIOTRACE_TEST) += testmmiotrace.o
15 15
16ifeq ($(CONFIG_X86_32),y) 16obj-$(CONFIG_NUMA) += numa_$(BITS).o
17obj-$(CONFIG_NUMA) += discontig_32.o
18else
19obj-$(CONFIG_NUMA) += numa_64.o
20obj-$(CONFIG_K8_NUMA) += k8topology_64.o 17obj-$(CONFIG_K8_NUMA) += k8topology_64.o
21endif
22obj-$(CONFIG_ACPI_NUMA) += srat_$(BITS).o 18obj-$(CONFIG_ACPI_NUMA) += srat_$(BITS).o
23 19
24obj-$(CONFIG_MEMTEST) += memtest.o 20obj-$(CONFIG_MEMTEST) += memtest.o
diff --git a/arch/x86/mm/dump_pagetables.c b/arch/x86/mm/dump_pagetables.c
index a20d1fa64b4e..e7277cbcfb40 100644
--- a/arch/x86/mm/dump_pagetables.c
+++ b/arch/x86/mm/dump_pagetables.c
@@ -148,8 +148,8 @@ static void note_page(struct seq_file *m, struct pg_state *st,
148 * we have now. "break" is either changing perms, levels or 148 * we have now. "break" is either changing perms, levels or
149 * address space marker. 149 * address space marker.
150 */ 150 */
151 prot = pgprot_val(new_prot) & ~(PTE_PFN_MASK); 151 prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
152 cur = pgprot_val(st->current_prot) & ~(PTE_PFN_MASK); 152 cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
153 153
154 if (!st->level) { 154 if (!st->level) {
155 /* First entry */ 155 /* First entry */
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index 455f3fe67b42..31e8730fa246 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -35,6 +35,7 @@
35#include <asm/tlbflush.h> 35#include <asm/tlbflush.h>
36#include <asm/proto.h> 36#include <asm/proto.h>
37#include <asm-generic/sections.h> 37#include <asm-generic/sections.h>
38#include <asm/traps.h>
38 39
39/* 40/*
40 * Page fault error code bits 41 * Page fault error code bits
@@ -357,8 +358,6 @@ static int is_errata100(struct pt_regs *regs, unsigned long address)
357 return 0; 358 return 0;
358} 359}
359 360
360void do_invalid_op(struct pt_regs *, unsigned long);
361
362static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 361static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
363{ 362{
364#ifdef CONFIG_X86_F00F_BUG 363#ifdef CONFIG_X86_F00F_BUG
@@ -593,11 +592,6 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
593 unsigned long flags; 592 unsigned long flags;
594#endif 593#endif
595 594
596 /*
597 * We can fault from pretty much anywhere, with unknown IRQ state.
598 */
599 trace_hardirqs_fixup();
600
601 tsk = current; 595 tsk = current;
602 mm = tsk->mm; 596 mm = tsk->mm;
603 prefetchw(&mm->mmap_sem); 597 prefetchw(&mm->mmap_sem);
@@ -646,24 +640,23 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
646 } 640 }
647 641
648 642
649#ifdef CONFIG_X86_32
650 /* It's safe to allow irq's after cr2 has been saved and the vmalloc
651 fault has been handled. */
652 if (regs->flags & (X86_EFLAGS_IF | X86_VM_MASK))
653 local_irq_enable();
654
655 /* 643 /*
656 * If we're in an interrupt, have no user context or are running in an 644 * It's safe to allow irq's after cr2 has been saved and the
657 * atomic region then we must not take the fault. 645 * vmalloc fault has been handled.
646 *
647 * User-mode registers count as a user access even for any
648 * potential system fault or CPU buglet.
658 */ 649 */
659 if (in_atomic() || !mm) 650 if (user_mode_vm(regs)) {
660 goto bad_area_nosemaphore; 651 local_irq_enable();
661#else /* CONFIG_X86_64 */ 652 error_code |= PF_USER;
662 if (likely(regs->flags & X86_EFLAGS_IF)) 653 } else if (regs->flags & X86_EFLAGS_IF)
663 local_irq_enable(); 654 local_irq_enable();
664 655
656#ifdef CONFIG_X86_64
665 if (unlikely(error_code & PF_RSVD)) 657 if (unlikely(error_code & PF_RSVD))
666 pgtable_bad(address, regs, error_code); 658 pgtable_bad(address, regs, error_code);
659#endif
667 660
668 /* 661 /*
669 * If we're in an interrupt, have no user context or are running in an 662 * If we're in an interrupt, have no user context or are running in an
@@ -672,15 +665,9 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
672 if (unlikely(in_atomic() || !mm)) 665 if (unlikely(in_atomic() || !mm))
673 goto bad_area_nosemaphore; 666 goto bad_area_nosemaphore;
674 667
675 /*
676 * User-mode registers count as a user access even for any
677 * potential system fault or CPU buglet.
678 */
679 if (user_mode_vm(regs))
680 error_code |= PF_USER;
681again: 668again:
682#endif 669 /*
683 /* When running in the kernel we expect faults to occur only to 670 * When running in the kernel we expect faults to occur only to
684 * addresses in user space. All other faults represent errors in the 671 * addresses in user space. All other faults represent errors in the
685 * kernel and should generate an OOPS. Unfortunately, in the case of an 672 * kernel and should generate an OOPS. Unfortunately, in the case of an
686 * erroneous fault occurring in a code path which already holds mmap_sem 673 * erroneous fault occurring in a code path which already holds mmap_sem
@@ -743,9 +730,6 @@ good_area:
743 goto bad_area; 730 goto bad_area;
744 } 731 }
745 732
746#ifdef CONFIG_X86_32
747survive:
748#endif
749 /* 733 /*
750 * If for any reason at all we couldn't handle the fault, 734 * If for any reason at all we couldn't handle the fault,
751 * make sure we exit gracefully rather than endlessly redo 735 * make sure we exit gracefully rather than endlessly redo
@@ -880,12 +864,11 @@ out_of_memory:
880 up_read(&mm->mmap_sem); 864 up_read(&mm->mmap_sem);
881 if (is_global_init(tsk)) { 865 if (is_global_init(tsk)) {
882 yield(); 866 yield();
883#ifdef CONFIG_X86_32 867 /*
884 down_read(&mm->mmap_sem); 868 * Re-lookup the vma - in theory the vma tree might
885 goto survive; 869 * have changed:
886#else 870 */
887 goto again; 871 goto again;
888#endif
889 } 872 }
890 873
891 printk("VM: killing process %s\n", tsk->comm); 874 printk("VM: killing process %s\n", tsk->comm);
@@ -915,15 +898,15 @@ LIST_HEAD(pgd_list);
915 898
916void vmalloc_sync_all(void) 899void vmalloc_sync_all(void)
917{ 900{
918#ifdef CONFIG_X86_32
919 unsigned long start = VMALLOC_START & PGDIR_MASK;
920 unsigned long address; 901 unsigned long address;
921 902
903#ifdef CONFIG_X86_32
922 if (SHARED_KERNEL_PMD) 904 if (SHARED_KERNEL_PMD)
923 return; 905 return;
924 906
925 BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK); 907 for (address = VMALLOC_START & PMD_MASK;
926 for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) { 908 address >= TASK_SIZE && address < FIXADDR_TOP;
909 address += PMD_SIZE) {
927 unsigned long flags; 910 unsigned long flags;
928 struct page *page; 911 struct page *page;
929 912
@@ -936,10 +919,8 @@ void vmalloc_sync_all(void)
936 spin_unlock_irqrestore(&pgd_lock, flags); 919 spin_unlock_irqrestore(&pgd_lock, flags);
937 } 920 }
938#else /* CONFIG_X86_64 */ 921#else /* CONFIG_X86_64 */
939 unsigned long start = VMALLOC_START & PGDIR_MASK; 922 for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END;
940 unsigned long address; 923 address += PGDIR_SIZE) {
941
942 for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) {
943 const pgd_t *pgd_ref = pgd_offset_k(address); 924 const pgd_t *pgd_ref = pgd_offset_k(address);
944 unsigned long flags; 925 unsigned long flags;
945 struct page *page; 926 struct page *page;
diff --git a/arch/x86/mm/gup.c b/arch/x86/mm/gup.c
index 007bb06c7504..4ba373c5b8c8 100644
--- a/arch/x86/mm/gup.c
+++ b/arch/x86/mm/gup.c
@@ -82,7 +82,7 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
82 pte_t pte = gup_get_pte(ptep); 82 pte_t pte = gup_get_pte(ptep);
83 struct page *page; 83 struct page *page;
84 84
85 if ((pte_val(pte) & (mask | _PAGE_SPECIAL)) != mask) { 85 if ((pte_flags(pte) & (mask | _PAGE_SPECIAL)) != mask) {
86 pte_unmap(ptep); 86 pte_unmap(ptep);
87 return 0; 87 return 0;
88 } 88 }
@@ -116,10 +116,10 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
116 mask = _PAGE_PRESENT|_PAGE_USER; 116 mask = _PAGE_PRESENT|_PAGE_USER;
117 if (write) 117 if (write)
118 mask |= _PAGE_RW; 118 mask |= _PAGE_RW;
119 if ((pte_val(pte) & mask) != mask) 119 if ((pte_flags(pte) & mask) != mask)
120 return 0; 120 return 0;
121 /* hugepages are never "special" */ 121 /* hugepages are never "special" */
122 VM_BUG_ON(pte_val(pte) & _PAGE_SPECIAL); 122 VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
123 VM_BUG_ON(!pfn_valid(pte_pfn(pte))); 123 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
124 124
125 refs = 0; 125 refs = 0;
@@ -173,10 +173,10 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
173 mask = _PAGE_PRESENT|_PAGE_USER; 173 mask = _PAGE_PRESENT|_PAGE_USER;
174 if (write) 174 if (write)
175 mask |= _PAGE_RW; 175 mask |= _PAGE_RW;
176 if ((pte_val(pte) & mask) != mask) 176 if ((pte_flags(pte) & mask) != mask)
177 return 0; 177 return 0;
178 /* hugepages are never "special" */ 178 /* hugepages are never "special" */
179 VM_BUG_ON(pte_val(pte) & _PAGE_SPECIAL); 179 VM_BUG_ON(pte_flags(pte) & _PAGE_SPECIAL);
180 VM_BUG_ON(!pfn_valid(pte_pfn(pte))); 180 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
181 181
182 refs = 0; 182 refs = 0;
diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index d37f29376b0c..8396868e82c5 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -31,6 +31,7 @@
31#include <linux/cpumask.h> 31#include <linux/cpumask.h>
32 32
33#include <asm/asm.h> 33#include <asm/asm.h>
34#include <asm/bios_ebda.h>
34#include <asm/processor.h> 35#include <asm/processor.h>
35#include <asm/system.h> 36#include <asm/system.h>
36#include <asm/uaccess.h> 37#include <asm/uaccess.h>
@@ -47,6 +48,7 @@
47#include <asm/paravirt.h> 48#include <asm/paravirt.h>
48#include <asm/setup.h> 49#include <asm/setup.h>
49#include <asm/cacheflush.h> 50#include <asm/cacheflush.h>
51#include <asm/smp.h>
50 52
51unsigned int __VMALLOC_RESERVE = 128 << 20; 53unsigned int __VMALLOC_RESERVE = 128 << 20;
52 54
@@ -194,11 +196,30 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
194 pgd_t *pgd; 196 pgd_t *pgd;
195 pmd_t *pmd; 197 pmd_t *pmd;
196 pte_t *pte; 198 pte_t *pte;
197 unsigned pages_2m = 0, pages_4k = 0; 199 unsigned pages_2m, pages_4k;
200 int mapping_iter;
201
202 /*
203 * First iteration will setup identity mapping using large/small pages
204 * based on use_pse, with other attributes same as set by
205 * the early code in head_32.S
206 *
207 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
208 * as desired for the kernel identity mapping.
209 *
210 * This two pass mechanism conforms to the TLB app note which says:
211 *
212 * "Software should not write to a paging-structure entry in a way
213 * that would change, for any linear address, both the page size
214 * and either the page frame or attributes."
215 */
216 mapping_iter = 1;
198 217
199 if (!cpu_has_pse) 218 if (!cpu_has_pse)
200 use_pse = 0; 219 use_pse = 0;
201 220
221repeat:
222 pages_2m = pages_4k = 0;
202 pfn = start_pfn; 223 pfn = start_pfn;
203 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 224 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
204 pgd = pgd_base + pgd_idx; 225 pgd = pgd_base + pgd_idx;
@@ -224,6 +245,13 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
224 if (use_pse) { 245 if (use_pse) {
225 unsigned int addr2; 246 unsigned int addr2;
226 pgprot_t prot = PAGE_KERNEL_LARGE; 247 pgprot_t prot = PAGE_KERNEL_LARGE;
248 /*
249 * first pass will use the same initial
250 * identity mapping attribute + _PAGE_PSE.
251 */
252 pgprot_t init_prot =
253 __pgprot(PTE_IDENT_ATTR |
254 _PAGE_PSE);
227 255
228 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 256 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
229 PAGE_OFFSET + PAGE_SIZE-1; 257 PAGE_OFFSET + PAGE_SIZE-1;
@@ -233,7 +261,10 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
233 prot = PAGE_KERNEL_LARGE_EXEC; 261 prot = PAGE_KERNEL_LARGE_EXEC;
234 262
235 pages_2m++; 263 pages_2m++;
236 set_pmd(pmd, pfn_pmd(pfn, prot)); 264 if (mapping_iter == 1)
265 set_pmd(pmd, pfn_pmd(pfn, init_prot));
266 else
267 set_pmd(pmd, pfn_pmd(pfn, prot));
237 268
238 pfn += PTRS_PER_PTE; 269 pfn += PTRS_PER_PTE;
239 continue; 270 continue;
@@ -245,17 +276,43 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
245 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 276 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
246 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 277 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
247 pgprot_t prot = PAGE_KERNEL; 278 pgprot_t prot = PAGE_KERNEL;
279 /*
280 * first pass will use the same initial
281 * identity mapping attribute.
282 */
283 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
248 284
249 if (is_kernel_text(addr)) 285 if (is_kernel_text(addr))
250 prot = PAGE_KERNEL_EXEC; 286 prot = PAGE_KERNEL_EXEC;
251 287
252 pages_4k++; 288 pages_4k++;
253 set_pte(pte, pfn_pte(pfn, prot)); 289 if (mapping_iter == 1)
290 set_pte(pte, pfn_pte(pfn, init_prot));
291 else
292 set_pte(pte, pfn_pte(pfn, prot));
254 } 293 }
255 } 294 }
256 } 295 }
257 update_page_count(PG_LEVEL_2M, pages_2m); 296 if (mapping_iter == 1) {
258 update_page_count(PG_LEVEL_4K, pages_4k); 297 /*
298 * update direct mapping page count only in the first
299 * iteration.
300 */
301 update_page_count(PG_LEVEL_2M, pages_2m);
302 update_page_count(PG_LEVEL_4K, pages_4k);
303
304 /*
305 * local global flush tlb, which will flush the previous
306 * mappings present in both small and large page TLB's.
307 */
308 __flush_tlb_all();
309
310 /*
311 * Second iteration will set the actual desired PTE attributes.
312 */
313 mapping_iter = 2;
314 goto repeat;
315 }
259} 316}
260 317
261/* 318/*
@@ -458,11 +515,7 @@ static void __init pagetable_init(void)
458{ 515{
459 pgd_t *pgd_base = swapper_pg_dir; 516 pgd_t *pgd_base = swapper_pg_dir;
460 517
461 paravirt_pagetable_setup_start(pgd_base);
462
463 permanent_kmaps_init(pgd_base); 518 permanent_kmaps_init(pgd_base);
464
465 paravirt_pagetable_setup_done(pgd_base);
466} 519}
467 520
468#ifdef CONFIG_ACPI_SLEEP 521#ifdef CONFIG_ACPI_SLEEP
@@ -505,7 +558,7 @@ void zap_low_mappings(void)
505 558
506int nx_enabled; 559int nx_enabled;
507 560
508pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL); 561pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
509EXPORT_SYMBOL_GPL(__supported_pte_mask); 562EXPORT_SYMBOL_GPL(__supported_pte_mask);
510 563
511#ifdef CONFIG_X86_PAE 564#ifdef CONFIG_X86_PAE
@@ -722,7 +775,7 @@ void __init setup_bootmem_allocator(void)
722 after_init_bootmem = 1; 775 after_init_bootmem = 1;
723} 776}
724 777
725static void __init find_early_table_space(unsigned long end) 778static void __init find_early_table_space(unsigned long end, int use_pse)
726{ 779{
727 unsigned long puds, pmds, ptes, tables, start; 780 unsigned long puds, pmds, ptes, tables, start;
728 781
@@ -732,7 +785,7 @@ static void __init find_early_table_space(unsigned long end)
732 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; 785 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
733 tables += PAGE_ALIGN(pmds * sizeof(pmd_t)); 786 tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
734 787
735 if (cpu_has_pse) { 788 if (use_pse) {
736 unsigned long extra; 789 unsigned long extra;
737 790
738 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); 791 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
@@ -772,12 +825,22 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
772 pgd_t *pgd_base = swapper_pg_dir; 825 pgd_t *pgd_base = swapper_pg_dir;
773 unsigned long start_pfn, end_pfn; 826 unsigned long start_pfn, end_pfn;
774 unsigned long big_page_start; 827 unsigned long big_page_start;
828#ifdef CONFIG_DEBUG_PAGEALLOC
829 /*
830 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
831 * This will simplify cpa(), which otherwise needs to support splitting
832 * large pages into small in interrupt context, etc.
833 */
834 int use_pse = 0;
835#else
836 int use_pse = cpu_has_pse;
837#endif
775 838
776 /* 839 /*
777 * Find space for the kernel direct mapping tables. 840 * Find space for the kernel direct mapping tables.
778 */ 841 */
779 if (!after_init_bootmem) 842 if (!after_init_bootmem)
780 find_early_table_space(end); 843 find_early_table_space(end, use_pse);
781 844
782#ifdef CONFIG_X86_PAE 845#ifdef CONFIG_X86_PAE
783 set_nx(); 846 set_nx();
@@ -823,7 +886,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
823 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT); 886 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
824 if (start_pfn < end_pfn) 887 if (start_pfn < end_pfn)
825 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 888 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
826 cpu_has_pse); 889 use_pse);
827 890
828 /* tail is not big page alignment ? */ 891 /* tail is not big page alignment ? */
829 start_pfn = end_pfn; 892 start_pfn = end_pfn;
@@ -907,6 +970,8 @@ void __init mem_init(void)
907 int codesize, reservedpages, datasize, initsize; 970 int codesize, reservedpages, datasize, initsize;
908 int tmp; 971 int tmp;
909 972
973 start_periodic_check_for_corruption();
974
910#ifdef CONFIG_FLATMEM 975#ifdef CONFIG_FLATMEM
911 BUG_ON(!mem_map); 976 BUG_ON(!mem_map);
912#endif 977#endif
@@ -986,7 +1051,6 @@ void __init mem_init(void)
986 if (boot_cpu_data.wp_works_ok < 0) 1051 if (boot_cpu_data.wp_works_ok < 0)
987 test_wp_bit(); 1052 test_wp_bit();
988 1053
989 cpa_init();
990 save_pg_dir(); 1054 save_pg_dir();
991 zap_low_mappings(); 1055 zap_low_mappings();
992} 1056}
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index d3746efb060d..b8e461d49412 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -31,6 +31,7 @@
31#include <linux/nmi.h> 31#include <linux/nmi.h>
32 32
33#include <asm/processor.h> 33#include <asm/processor.h>
34#include <asm/bios_ebda.h>
34#include <asm/system.h> 35#include <asm/system.h>
35#include <asm/uaccess.h> 36#include <asm/uaccess.h>
36#include <asm/pgtable.h> 37#include <asm/pgtable.h>
@@ -88,6 +89,62 @@ early_param("gbpages", parse_direct_gbpages_on);
88 89
89int after_bootmem; 90int after_bootmem;
90 91
92pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
93EXPORT_SYMBOL_GPL(__supported_pte_mask);
94
95static int do_not_nx __cpuinitdata;
96
97/*
98 * noexec=on|off
99 * Control non-executable mappings for 64-bit processes.
100 *
101 * on Enable (default)
102 * off Disable
103 */
104static int __init nonx_setup(char *str)
105{
106 if (!str)
107 return -EINVAL;
108 if (!strncmp(str, "on", 2)) {
109 __supported_pte_mask |= _PAGE_NX;
110 do_not_nx = 0;
111 } else if (!strncmp(str, "off", 3)) {
112 do_not_nx = 1;
113 __supported_pte_mask &= ~_PAGE_NX;
114 }
115 return 0;
116}
117early_param("noexec", nonx_setup);
118
119void __cpuinit check_efer(void)
120{
121 unsigned long efer;
122
123 rdmsrl(MSR_EFER, efer);
124 if (!(efer & EFER_NX) || do_not_nx)
125 __supported_pte_mask &= ~_PAGE_NX;
126}
127
128int force_personality32;
129
130/*
131 * noexec32=on|off
132 * Control non executable heap for 32bit processes.
133 * To control the stack too use noexec=off
134 *
135 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
136 * off PROT_READ implies PROT_EXEC
137 */
138static int __init nonx32_setup(char *str)
139{
140 if (!strcmp(str, "on"))
141 force_personality32 &= ~READ_IMPLIES_EXEC;
142 else if (!strcmp(str, "off"))
143 force_personality32 |= READ_IMPLIES_EXEC;
144 return 1;
145}
146__setup("noexec32=", nonx32_setup);
147
91/* 148/*
92 * NOTE: This function is marked __ref because it calls __init function 149 * NOTE: This function is marked __ref because it calls __init function
93 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. 150 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
@@ -139,9 +196,6 @@ set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
139 } 196 }
140 197
141 pte = pte_offset_kernel(pmd, vaddr); 198 pte = pte_offset_kernel(pmd, vaddr);
142 if (!pte_none(*pte) && pte_val(new_pte) &&
143 pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
144 pte_ERROR(*pte);
145 set_pte(pte, new_pte); 199 set_pte(pte, new_pte);
146 200
147 /* 201 /*
@@ -225,7 +279,7 @@ void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
225void __init cleanup_highmap(void) 279void __init cleanup_highmap(void)
226{ 280{
227 unsigned long vaddr = __START_KERNEL_map; 281 unsigned long vaddr = __START_KERNEL_map;
228 unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1; 282 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
229 pmd_t *pmd = level2_kernel_pgt; 283 pmd_t *pmd = level2_kernel_pgt;
230 pmd_t *last_pmd = pmd + PTRS_PER_PMD; 284 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
231 285
@@ -256,7 +310,7 @@ static __ref void *alloc_low_page(unsigned long *phys)
256 if (pfn >= table_top) 310 if (pfn >= table_top)
257 panic("alloc_low_page: ran out of memory"); 311 panic("alloc_low_page: ran out of memory");
258 312
259 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); 313 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
260 memset(adr, 0, PAGE_SIZE); 314 memset(adr, 0, PAGE_SIZE);
261 *phys = pfn * PAGE_SIZE; 315 *phys = pfn * PAGE_SIZE;
262 return adr; 316 return adr;
@@ -271,7 +325,8 @@ static __ref void unmap_low_page(void *adr)
271} 325}
272 326
273static unsigned long __meminit 327static unsigned long __meminit
274phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end) 328phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
329 pgprot_t prot)
275{ 330{
276 unsigned pages = 0; 331 unsigned pages = 0;
277 unsigned long last_map_addr = end; 332 unsigned long last_map_addr = end;
@@ -289,36 +344,43 @@ phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end)
289 break; 344 break;
290 } 345 }
291 346
347 /*
348 * We will re-use the existing mapping.
349 * Xen for example has some special requirements, like mapping
350 * pagetable pages as RO. So assume someone who pre-setup
351 * these mappings are more intelligent.
352 */
292 if (pte_val(*pte)) 353 if (pte_val(*pte))
293 continue; 354 continue;
294 355
295 if (0) 356 if (0)
296 printk(" pte=%p addr=%lx pte=%016lx\n", 357 printk(" pte=%p addr=%lx pte=%016lx\n",
297 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); 358 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
298 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL));
299 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
300 pages++; 359 pages++;
360 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
361 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
301 } 362 }
363
302 update_page_count(PG_LEVEL_4K, pages); 364 update_page_count(PG_LEVEL_4K, pages);
303 365
304 return last_map_addr; 366 return last_map_addr;
305} 367}
306 368
307static unsigned long __meminit 369static unsigned long __meminit
308phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end) 370phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
371 pgprot_t prot)
309{ 372{
310 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd); 373 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
311 374
312 return phys_pte_init(pte, address, end); 375 return phys_pte_init(pte, address, end, prot);
313} 376}
314 377
315static unsigned long __meminit 378static unsigned long __meminit
316phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, 379phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
317 unsigned long page_size_mask) 380 unsigned long page_size_mask, pgprot_t prot)
318{ 381{
319 unsigned long pages = 0; 382 unsigned long pages = 0;
320 unsigned long last_map_addr = end; 383 unsigned long last_map_addr = end;
321 unsigned long start = address;
322 384
323 int i = pmd_index(address); 385 int i = pmd_index(address);
324 386
@@ -326,6 +388,7 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
326 unsigned long pte_phys; 388 unsigned long pte_phys;
327 pmd_t *pmd = pmd_page + pmd_index(address); 389 pmd_t *pmd = pmd_page + pmd_index(address);
328 pte_t *pte; 390 pte_t *pte;
391 pgprot_t new_prot = prot;
329 392
330 if (address >= end) { 393 if (address >= end) {
331 if (!after_bootmem) { 394 if (!after_bootmem) {
@@ -339,27 +402,40 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
339 if (!pmd_large(*pmd)) { 402 if (!pmd_large(*pmd)) {
340 spin_lock(&init_mm.page_table_lock); 403 spin_lock(&init_mm.page_table_lock);
341 last_map_addr = phys_pte_update(pmd, address, 404 last_map_addr = phys_pte_update(pmd, address,
342 end); 405 end, prot);
343 spin_unlock(&init_mm.page_table_lock); 406 spin_unlock(&init_mm.page_table_lock);
407 continue;
344 } 408 }
345 /* Count entries we're using from level2_ident_pgt */ 409 /*
346 if (start == 0) 410 * If we are ok with PG_LEVEL_2M mapping, then we will
347 pages++; 411 * use the existing mapping,
348 continue; 412 *
413 * Otherwise, we will split the large page mapping but
414 * use the same existing protection bits except for
415 * large page, so that we don't violate Intel's TLB
416 * Application note (317080) which says, while changing
417 * the page sizes, new and old translations should
418 * not differ with respect to page frame and
419 * attributes.
420 */
421 if (page_size_mask & (1 << PG_LEVEL_2M))
422 continue;
423 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
349 } 424 }
350 425
351 if (page_size_mask & (1<<PG_LEVEL_2M)) { 426 if (page_size_mask & (1<<PG_LEVEL_2M)) {
352 pages++; 427 pages++;
353 spin_lock(&init_mm.page_table_lock); 428 spin_lock(&init_mm.page_table_lock);
354 set_pte((pte_t *)pmd, 429 set_pte((pte_t *)pmd,
355 pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); 430 pfn_pte(address >> PAGE_SHIFT,
431 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
356 spin_unlock(&init_mm.page_table_lock); 432 spin_unlock(&init_mm.page_table_lock);
357 last_map_addr = (address & PMD_MASK) + PMD_SIZE; 433 last_map_addr = (address & PMD_MASK) + PMD_SIZE;
358 continue; 434 continue;
359 } 435 }
360 436
361 pte = alloc_low_page(&pte_phys); 437 pte = alloc_low_page(&pte_phys);
362 last_map_addr = phys_pte_init(pte, address, end); 438 last_map_addr = phys_pte_init(pte, address, end, new_prot);
363 unmap_low_page(pte); 439 unmap_low_page(pte);
364 440
365 spin_lock(&init_mm.page_table_lock); 441 spin_lock(&init_mm.page_table_lock);
@@ -372,12 +448,12 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
372 448
373static unsigned long __meminit 449static unsigned long __meminit
374phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end, 450phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
375 unsigned long page_size_mask) 451 unsigned long page_size_mask, pgprot_t prot)
376{ 452{
377 pmd_t *pmd = pmd_offset(pud, 0); 453 pmd_t *pmd = pmd_offset(pud, 0);
378 unsigned long last_map_addr; 454 unsigned long last_map_addr;
379 455
380 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask); 456 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
381 __flush_tlb_all(); 457 __flush_tlb_all();
382 return last_map_addr; 458 return last_map_addr;
383} 459}
@@ -394,6 +470,7 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
394 unsigned long pmd_phys; 470 unsigned long pmd_phys;
395 pud_t *pud = pud_page + pud_index(addr); 471 pud_t *pud = pud_page + pud_index(addr);
396 pmd_t *pmd; 472 pmd_t *pmd;
473 pgprot_t prot = PAGE_KERNEL;
397 474
398 if (addr >= end) 475 if (addr >= end)
399 break; 476 break;
@@ -405,10 +482,26 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
405 } 482 }
406 483
407 if (pud_val(*pud)) { 484 if (pud_val(*pud)) {
408 if (!pud_large(*pud)) 485 if (!pud_large(*pud)) {
409 last_map_addr = phys_pmd_update(pud, addr, end, 486 last_map_addr = phys_pmd_update(pud, addr, end,
410 page_size_mask); 487 page_size_mask, prot);
411 continue; 488 continue;
489 }
490 /*
491 * If we are ok with PG_LEVEL_1G mapping, then we will
492 * use the existing mapping.
493 *
494 * Otherwise, we will split the gbpage mapping but use
495 * the same existing protection bits except for large
496 * page, so that we don't violate Intel's TLB
497 * Application note (317080) which says, while changing
498 * the page sizes, new and old translations should
499 * not differ with respect to page frame and
500 * attributes.
501 */
502 if (page_size_mask & (1 << PG_LEVEL_1G))
503 continue;
504 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
412 } 505 }
413 506
414 if (page_size_mask & (1<<PG_LEVEL_1G)) { 507 if (page_size_mask & (1<<PG_LEVEL_1G)) {
@@ -422,7 +515,8 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
422 } 515 }
423 516
424 pmd = alloc_low_page(&pmd_phys); 517 pmd = alloc_low_page(&pmd_phys);
425 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask); 518 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
519 prot);
426 unmap_low_page(pmd); 520 unmap_low_page(pmd);
427 521
428 spin_lock(&init_mm.page_table_lock); 522 spin_lock(&init_mm.page_table_lock);
@@ -430,6 +524,7 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
430 spin_unlock(&init_mm.page_table_lock); 524 spin_unlock(&init_mm.page_table_lock);
431 } 525 }
432 __flush_tlb_all(); 526 __flush_tlb_all();
527
433 update_page_count(PG_LEVEL_1G, pages); 528 update_page_count(PG_LEVEL_1G, pages);
434 529
435 return last_map_addr; 530 return last_map_addr;
@@ -446,27 +541,28 @@ phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
446 return phys_pud_init(pud, addr, end, page_size_mask); 541 return phys_pud_init(pud, addr, end, page_size_mask);
447} 542}
448 543
449static void __init find_early_table_space(unsigned long end) 544static void __init find_early_table_space(unsigned long end, int use_pse,
545 int use_gbpages)
450{ 546{
451 unsigned long puds, pmds, ptes, tables, start; 547 unsigned long puds, pmds, ptes, tables, start;
452 548
453 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; 549 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
454 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE); 550 tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
455 if (direct_gbpages) { 551 if (use_gbpages) {
456 unsigned long extra; 552 unsigned long extra;
457 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT); 553 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
458 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT; 554 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
459 } else 555 } else
460 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; 556 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
461 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE); 557 tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
462 558
463 if (cpu_has_pse) { 559 if (use_pse) {
464 unsigned long extra; 560 unsigned long extra;
465 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); 561 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
466 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT; 562 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
467 } else 563 } else
468 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; 564 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
469 tables += round_up(ptes * sizeof(pte_t), PAGE_SIZE); 565 tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
470 566
471 /* 567 /*
472 * RED-PEN putting page tables only on node 0 could 568 * RED-PEN putting page tables only on node 0 could
@@ -528,6 +624,7 @@ static unsigned long __init kernel_physical_mapping_init(unsigned long start,
528 pgd_populate(&init_mm, pgd, __va(pud_phys)); 624 pgd_populate(&init_mm, pgd, __va(pud_phys));
529 spin_unlock(&init_mm.page_table_lock); 625 spin_unlock(&init_mm.page_table_lock);
530 } 626 }
627 __flush_tlb_all();
531 628
532 return last_map_addr; 629 return last_map_addr;
533} 630}
@@ -571,6 +668,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
571 668
572 struct map_range mr[NR_RANGE_MR]; 669 struct map_range mr[NR_RANGE_MR];
573 int nr_range, i; 670 int nr_range, i;
671 int use_pse, use_gbpages;
574 672
575 printk(KERN_INFO "init_memory_mapping\n"); 673 printk(KERN_INFO "init_memory_mapping\n");
576 674
@@ -584,9 +682,21 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
584 if (!after_bootmem) 682 if (!after_bootmem)
585 init_gbpages(); 683 init_gbpages();
586 684
587 if (direct_gbpages) 685#ifdef CONFIG_DEBUG_PAGEALLOC
686 /*
687 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
688 * This will simplify cpa(), which otherwise needs to support splitting
689 * large pages into small in interrupt context, etc.
690 */
691 use_pse = use_gbpages = 0;
692#else
693 use_pse = cpu_has_pse;
694 use_gbpages = direct_gbpages;
695#endif
696
697 if (use_gbpages)
588 page_size_mask |= 1 << PG_LEVEL_1G; 698 page_size_mask |= 1 << PG_LEVEL_1G;
589 if (cpu_has_pse) 699 if (use_pse)
590 page_size_mask |= 1 << PG_LEVEL_2M; 700 page_size_mask |= 1 << PG_LEVEL_2M;
591 701
592 memset(mr, 0, sizeof(mr)); 702 memset(mr, 0, sizeof(mr));
@@ -636,7 +746,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
636 old_start = mr[i].start; 746 old_start = mr[i].start;
637 memmove(&mr[i], &mr[i+1], 747 memmove(&mr[i], &mr[i+1],
638 (nr_range - 1 - i) * sizeof (struct map_range)); 748 (nr_range - 1 - i) * sizeof (struct map_range));
639 mr[i].start = old_start; 749 mr[i--].start = old_start;
640 nr_range--; 750 nr_range--;
641 } 751 }
642 752
@@ -647,7 +757,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
647 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k")); 757 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
648 758
649 if (!after_bootmem) 759 if (!after_bootmem)
650 find_early_table_space(end); 760 find_early_table_space(end, use_pse, use_gbpages);
651 761
652 for (i = 0; i < nr_range; i++) 762 for (i = 0; i < nr_range; i++)
653 last_map_addr = kernel_physical_mapping_init( 763 last_map_addr = kernel_physical_mapping_init(
@@ -769,6 +879,8 @@ void __init mem_init(void)
769{ 879{
770 long codesize, reservedpages, datasize, initsize; 880 long codesize, reservedpages, datasize, initsize;
771 881
882 start_periodic_check_for_corruption();
883
772 pci_iommu_alloc(); 884 pci_iommu_alloc();
773 885
774 /* clear_bss() already clear the empty_zero_page */ 886 /* clear_bss() already clear the empty_zero_page */
@@ -806,8 +918,6 @@ void __init mem_init(void)
806 reservedpages << (PAGE_SHIFT-10), 918 reservedpages << (PAGE_SHIFT-10),
807 datasize >> 10, 919 datasize >> 10,
808 initsize >> 10); 920 initsize >> 10);
809
810 cpa_init();
811} 921}
812 922
813void free_init_pages(char *what, unsigned long begin, unsigned long end) 923void free_init_pages(char *what, unsigned long begin, unsigned long end)
diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
index d4b6e6a29ae3..ae71e11eb3e5 100644
--- a/arch/x86/mm/ioremap.c
+++ b/arch/x86/mm/ioremap.c
@@ -24,18 +24,47 @@
24 24
25#ifdef CONFIG_X86_64 25#ifdef CONFIG_X86_64
26 26
27static inline int phys_addr_valid(unsigned long addr)
28{
29 return addr < (1UL << boot_cpu_data.x86_phys_bits);
30}
31
27unsigned long __phys_addr(unsigned long x) 32unsigned long __phys_addr(unsigned long x)
28{ 33{
29 if (x >= __START_KERNEL_map) 34 if (x >= __START_KERNEL_map) {
30 return x - __START_KERNEL_map + phys_base; 35 x -= __START_KERNEL_map;
31 return x - PAGE_OFFSET; 36 VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE);
37 x += phys_base;
38 } else {
39 VIRTUAL_BUG_ON(x < PAGE_OFFSET);
40 x -= PAGE_OFFSET;
41 VIRTUAL_BUG_ON(system_state == SYSTEM_BOOTING ? x > MAXMEM :
42 !phys_addr_valid(x));
43 }
44 return x;
32} 45}
33EXPORT_SYMBOL(__phys_addr); 46EXPORT_SYMBOL(__phys_addr);
34 47
35static inline int phys_addr_valid(unsigned long addr) 48bool __virt_addr_valid(unsigned long x)
36{ 49{
37 return addr < (1UL << boot_cpu_data.x86_phys_bits); 50 if (x >= __START_KERNEL_map) {
51 x -= __START_KERNEL_map;
52 if (x >= KERNEL_IMAGE_SIZE)
53 return false;
54 x += phys_base;
55 } else {
56 if (x < PAGE_OFFSET)
57 return false;
58 x -= PAGE_OFFSET;
59 if (system_state == SYSTEM_BOOTING ?
60 x > MAXMEM : !phys_addr_valid(x)) {
61 return false;
62 }
63 }
64
65 return pfn_valid(x >> PAGE_SHIFT);
38} 66}
67EXPORT_SYMBOL(__virt_addr_valid);
39 68
40#else 69#else
41 70
@@ -44,6 +73,28 @@ static inline int phys_addr_valid(unsigned long addr)
44 return 1; 73 return 1;
45} 74}
46 75
76#ifdef CONFIG_DEBUG_VIRTUAL
77unsigned long __phys_addr(unsigned long x)
78{
79 /* VMALLOC_* aren't constants; not available at the boot time */
80 VIRTUAL_BUG_ON(x < PAGE_OFFSET);
81 VIRTUAL_BUG_ON(system_state != SYSTEM_BOOTING &&
82 is_vmalloc_addr((void *) x));
83 return x - PAGE_OFFSET;
84}
85EXPORT_SYMBOL(__phys_addr);
86#endif
87
88bool __virt_addr_valid(unsigned long x)
89{
90 if (x < PAGE_OFFSET)
91 return false;
92 if (system_state != SYSTEM_BOOTING && is_vmalloc_addr((void *) x))
93 return false;
94 return pfn_valid((x - PAGE_OFFSET) >> PAGE_SHIFT);
95}
96EXPORT_SYMBOL(__virt_addr_valid);
97
47#endif 98#endif
48 99
49int page_is_ram(unsigned long pagenr) 100int page_is_ram(unsigned long pagenr)
@@ -83,6 +134,25 @@ int page_is_ram(unsigned long pagenr)
83 return 0; 134 return 0;
84} 135}
85 136
137int pagerange_is_ram(unsigned long start, unsigned long end)
138{
139 int ram_page = 0, not_rampage = 0;
140 unsigned long page_nr;
141
142 for (page_nr = (start >> PAGE_SHIFT); page_nr < (end >> PAGE_SHIFT);
143 ++page_nr) {
144 if (page_is_ram(page_nr))
145 ram_page = 1;
146 else
147 not_rampage = 1;
148
149 if (ram_page == not_rampage)
150 return -1;
151 }
152
153 return ram_page;
154}
155
86/* 156/*
87 * Fix up the linear direct mapping of the kernel to avoid cache attribute 157 * Fix up the linear direct mapping of the kernel to avoid cache attribute
88 * conflicts. 158 * conflicts.
@@ -150,6 +220,12 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr,
150 return (__force void __iomem *)phys_to_virt(phys_addr); 220 return (__force void __iomem *)phys_to_virt(phys_addr);
151 221
152 /* 222 /*
223 * Check if the request spans more than any BAR in the iomem resource
224 * tree.
225 */
226 WARN_ON(iomem_map_sanity_check(phys_addr, size));
227
228 /*
153 * Don't allow anybody to remap normal RAM that we're using.. 229 * Don't allow anybody to remap normal RAM that we're using..
154 */ 230 */
155 for (pfn = phys_addr >> PAGE_SHIFT; 231 for (pfn = phys_addr >> PAGE_SHIFT;
@@ -204,16 +280,16 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr,
204 switch (prot_val) { 280 switch (prot_val) {
205 case _PAGE_CACHE_UC: 281 case _PAGE_CACHE_UC:
206 default: 282 default:
207 prot = PAGE_KERNEL_NOCACHE; 283 prot = PAGE_KERNEL_IO_NOCACHE;
208 break; 284 break;
209 case _PAGE_CACHE_UC_MINUS: 285 case _PAGE_CACHE_UC_MINUS:
210 prot = PAGE_KERNEL_UC_MINUS; 286 prot = PAGE_KERNEL_IO_UC_MINUS;
211 break; 287 break;
212 case _PAGE_CACHE_WC: 288 case _PAGE_CACHE_WC:
213 prot = PAGE_KERNEL_WC; 289 prot = PAGE_KERNEL_IO_WC;
214 break; 290 break;
215 case _PAGE_CACHE_WB: 291 case _PAGE_CACHE_WB:
216 prot = PAGE_KERNEL; 292 prot = PAGE_KERNEL_IO;
217 break; 293 break;
218 } 294 }
219 295
@@ -421,7 +497,7 @@ void unxlate_dev_mem_ptr(unsigned long phys, void *addr)
421 return; 497 return;
422} 498}
423 499
424int __initdata early_ioremap_debug; 500static int __initdata early_ioremap_debug;
425 501
426static int __init early_ioremap_debug_setup(char *str) 502static int __init early_ioremap_debug_setup(char *str)
427{ 503{
@@ -530,12 +606,12 @@ static void __init __early_set_fixmap(enum fixed_addresses idx,
530} 606}
531 607
532static inline void __init early_set_fixmap(enum fixed_addresses idx, 608static inline void __init early_set_fixmap(enum fixed_addresses idx,
533 unsigned long phys) 609 unsigned long phys, pgprot_t prot)
534{ 610{
535 if (after_paging_init) 611 if (after_paging_init)
536 set_fixmap(idx, phys); 612 __set_fixmap(idx, phys, prot);
537 else 613 else
538 __early_set_fixmap(idx, phys, PAGE_KERNEL); 614 __early_set_fixmap(idx, phys, prot);
539} 615}
540 616
541static inline void __init early_clear_fixmap(enum fixed_addresses idx) 617static inline void __init early_clear_fixmap(enum fixed_addresses idx)
@@ -546,16 +622,22 @@ static inline void __init early_clear_fixmap(enum fixed_addresses idx)
546 __early_set_fixmap(idx, 0, __pgprot(0)); 622 __early_set_fixmap(idx, 0, __pgprot(0));
547} 623}
548 624
549 625static void *prev_map[FIX_BTMAPS_SLOTS] __initdata;
550int __initdata early_ioremap_nested; 626static unsigned long prev_size[FIX_BTMAPS_SLOTS] __initdata;
551
552static int __init check_early_ioremap_leak(void) 627static int __init check_early_ioremap_leak(void)
553{ 628{
554 if (!early_ioremap_nested) 629 int count = 0;
630 int i;
631
632 for (i = 0; i < FIX_BTMAPS_SLOTS; i++)
633 if (prev_map[i])
634 count++;
635
636 if (!count)
555 return 0; 637 return 0;
556 WARN(1, KERN_WARNING 638 WARN(1, KERN_WARNING
557 "Debug warning: early ioremap leak of %d areas detected.\n", 639 "Debug warning: early ioremap leak of %d areas detected.\n",
558 early_ioremap_nested); 640 count);
559 printk(KERN_WARNING 641 printk(KERN_WARNING
560 "please boot with early_ioremap_debug and report the dmesg.\n"); 642 "please boot with early_ioremap_debug and report the dmesg.\n");
561 643
@@ -563,18 +645,33 @@ static int __init check_early_ioremap_leak(void)
563} 645}
564late_initcall(check_early_ioremap_leak); 646late_initcall(check_early_ioremap_leak);
565 647
566void __init *early_ioremap(unsigned long phys_addr, unsigned long size) 648static void __init *__early_ioremap(unsigned long phys_addr, unsigned long size, pgprot_t prot)
567{ 649{
568 unsigned long offset, last_addr; 650 unsigned long offset, last_addr;
569 unsigned int nrpages, nesting; 651 unsigned int nrpages;
570 enum fixed_addresses idx0, idx; 652 enum fixed_addresses idx0, idx;
653 int i, slot;
571 654
572 WARN_ON(system_state != SYSTEM_BOOTING); 655 WARN_ON(system_state != SYSTEM_BOOTING);
573 656
574 nesting = early_ioremap_nested; 657 slot = -1;
658 for (i = 0; i < FIX_BTMAPS_SLOTS; i++) {
659 if (!prev_map[i]) {
660 slot = i;
661 break;
662 }
663 }
664
665 if (slot < 0) {
666 printk(KERN_INFO "early_iomap(%08lx, %08lx) not found slot\n",
667 phys_addr, size);
668 WARN_ON(1);
669 return NULL;
670 }
671
575 if (early_ioremap_debug) { 672 if (early_ioremap_debug) {
576 printk(KERN_INFO "early_ioremap(%08lx, %08lx) [%d] => ", 673 printk(KERN_INFO "early_ioremap(%08lx, %08lx) [%d] => ",
577 phys_addr, size, nesting); 674 phys_addr, size, slot);
578 dump_stack(); 675 dump_stack();
579 } 676 }
580 677
@@ -585,17 +682,13 @@ void __init *early_ioremap(unsigned long phys_addr, unsigned long size)
585 return NULL; 682 return NULL;
586 } 683 }
587 684
588 if (nesting >= FIX_BTMAPS_NESTING) { 685 prev_size[slot] = size;
589 WARN_ON(1);
590 return NULL;
591 }
592 early_ioremap_nested++;
593 /* 686 /*
594 * Mappings have to be page-aligned 687 * Mappings have to be page-aligned
595 */ 688 */
596 offset = phys_addr & ~PAGE_MASK; 689 offset = phys_addr & ~PAGE_MASK;
597 phys_addr &= PAGE_MASK; 690 phys_addr &= PAGE_MASK;
598 size = PAGE_ALIGN(last_addr) - phys_addr; 691 size = PAGE_ALIGN(last_addr + 1) - phys_addr;
599 692
600 /* 693 /*
601 * Mappings have to fit in the FIX_BTMAP area. 694 * Mappings have to fit in the FIX_BTMAP area.
@@ -609,10 +702,10 @@ void __init *early_ioremap(unsigned long phys_addr, unsigned long size)
609 /* 702 /*
610 * Ok, go for it.. 703 * Ok, go for it..
611 */ 704 */
612 idx0 = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*nesting; 705 idx0 = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*slot;
613 idx = idx0; 706 idx = idx0;
614 while (nrpages > 0) { 707 while (nrpages > 0) {
615 early_set_fixmap(idx, phys_addr); 708 early_set_fixmap(idx, phys_addr, prot);
616 phys_addr += PAGE_SIZE; 709 phys_addr += PAGE_SIZE;
617 --idx; 710 --idx;
618 --nrpages; 711 --nrpages;
@@ -620,7 +713,20 @@ void __init *early_ioremap(unsigned long phys_addr, unsigned long size)
620 if (early_ioremap_debug) 713 if (early_ioremap_debug)
621 printk(KERN_CONT "%08lx + %08lx\n", offset, fix_to_virt(idx0)); 714 printk(KERN_CONT "%08lx + %08lx\n", offset, fix_to_virt(idx0));
622 715
623 return (void *) (offset + fix_to_virt(idx0)); 716 prev_map[slot] = (void *) (offset + fix_to_virt(idx0));
717 return prev_map[slot];
718}
719
720/* Remap an IO device */
721void __init *early_ioremap(unsigned long phys_addr, unsigned long size)
722{
723 return __early_ioremap(phys_addr, size, PAGE_KERNEL_IO);
724}
725
726/* Remap memory */
727void __init *early_memremap(unsigned long phys_addr, unsigned long size)
728{
729 return __early_ioremap(phys_addr, size, PAGE_KERNEL);
624} 730}
625 731
626void __init early_iounmap(void *addr, unsigned long size) 732void __init early_iounmap(void *addr, unsigned long size)
@@ -629,15 +735,33 @@ void __init early_iounmap(void *addr, unsigned long size)
629 unsigned long offset; 735 unsigned long offset;
630 unsigned int nrpages; 736 unsigned int nrpages;
631 enum fixed_addresses idx; 737 enum fixed_addresses idx;
632 int nesting; 738 int i, slot;
739
740 slot = -1;
741 for (i = 0; i < FIX_BTMAPS_SLOTS; i++) {
742 if (prev_map[i] == addr) {
743 slot = i;
744 break;
745 }
746 }
747
748 if (slot < 0) {
749 printk(KERN_INFO "early_iounmap(%p, %08lx) not found slot\n",
750 addr, size);
751 WARN_ON(1);
752 return;
753 }
633 754
634 nesting = --early_ioremap_nested; 755 if (prev_size[slot] != size) {
635 if (WARN_ON(nesting < 0)) 756 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d] size not consistent %08lx\n",
757 addr, size, slot, prev_size[slot]);
758 WARN_ON(1);
636 return; 759 return;
760 }
637 761
638 if (early_ioremap_debug) { 762 if (early_ioremap_debug) {
639 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d]\n", addr, 763 printk(KERN_INFO "early_iounmap(%p, %08lx) [%d]\n", addr,
640 size, nesting); 764 size, slot);
641 dump_stack(); 765 dump_stack();
642 } 766 }
643 767
@@ -649,12 +773,13 @@ void __init early_iounmap(void *addr, unsigned long size)
649 offset = virt_addr & ~PAGE_MASK; 773 offset = virt_addr & ~PAGE_MASK;
650 nrpages = PAGE_ALIGN(offset + size - 1) >> PAGE_SHIFT; 774 nrpages = PAGE_ALIGN(offset + size - 1) >> PAGE_SHIFT;
651 775
652 idx = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*nesting; 776 idx = FIX_BTMAP_BEGIN - NR_FIX_BTMAPS*slot;
653 while (nrpages > 0) { 777 while (nrpages > 0) {
654 early_clear_fixmap(idx); 778 early_clear_fixmap(idx);
655 --idx; 779 --idx;
656 --nrpages; 780 --nrpages;
657 } 781 }
782 prev_map[slot] = 0;
658} 783}
659 784
660void __this_fixmap_does_not_exist(void) 785void __this_fixmap_does_not_exist(void)
diff --git a/arch/x86/mm/discontig_32.c b/arch/x86/mm/numa_32.c
index 62fa440678d8..847c164725f4 100644
--- a/arch/x86/mm/discontig_32.c
+++ b/arch/x86/mm/numa_32.c
@@ -328,7 +328,7 @@ void __init initmem_init(unsigned long start_pfn,
328 328
329 get_memcfg_numa(); 329 get_memcfg_numa();
330 330
331 kva_pages = round_up(calculate_numa_remap_pages(), PTRS_PER_PTE); 331 kva_pages = roundup(calculate_numa_remap_pages(), PTRS_PER_PTE);
332 332
333 kva_target_pfn = round_down(max_low_pfn - kva_pages, PTRS_PER_PTE); 333 kva_target_pfn = round_down(max_low_pfn - kva_pages, PTRS_PER_PTE);
334 do { 334 do {
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
index a4dd793d6003..cebcbf152d46 100644
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -79,7 +79,7 @@ static int __init allocate_cachealigned_memnodemap(void)
79 return 0; 79 return 0;
80 80
81 addr = 0x8000; 81 addr = 0x8000;
82 nodemap_size = round_up(sizeof(s16) * memnodemapsize, L1_CACHE_BYTES); 82 nodemap_size = roundup(sizeof(s16) * memnodemapsize, L1_CACHE_BYTES);
83 nodemap_addr = find_e820_area(addr, max_pfn<<PAGE_SHIFT, 83 nodemap_addr = find_e820_area(addr, max_pfn<<PAGE_SHIFT,
84 nodemap_size, L1_CACHE_BYTES); 84 nodemap_size, L1_CACHE_BYTES);
85 if (nodemap_addr == -1UL) { 85 if (nodemap_addr == -1UL) {
@@ -176,10 +176,10 @@ void __init setup_node_bootmem(int nodeid, unsigned long start,
176 unsigned long start_pfn, last_pfn, bootmap_pages, bootmap_size; 176 unsigned long start_pfn, last_pfn, bootmap_pages, bootmap_size;
177 unsigned long bootmap_start, nodedata_phys; 177 unsigned long bootmap_start, nodedata_phys;
178 void *bootmap; 178 void *bootmap;
179 const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE); 179 const int pgdat_size = roundup(sizeof(pg_data_t), PAGE_SIZE);
180 int nid; 180 int nid;
181 181
182 start = round_up(start, ZONE_ALIGN); 182 start = roundup(start, ZONE_ALIGN);
183 183
184 printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, 184 printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid,
185 start, end); 185 start, end);
@@ -210,9 +210,9 @@ void __init setup_node_bootmem(int nodeid, unsigned long start,
210 bootmap_pages = bootmem_bootmap_pages(last_pfn - start_pfn); 210 bootmap_pages = bootmem_bootmap_pages(last_pfn - start_pfn);
211 nid = phys_to_nid(nodedata_phys); 211 nid = phys_to_nid(nodedata_phys);
212 if (nid == nodeid) 212 if (nid == nodeid)
213 bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE); 213 bootmap_start = roundup(nodedata_phys + pgdat_size, PAGE_SIZE);
214 else 214 else
215 bootmap_start = round_up(start, PAGE_SIZE); 215 bootmap_start = roundup(start, PAGE_SIZE);
216 /* 216 /*
217 * SMP_CACHE_BYTES could be enough, but init_bootmem_node like 217 * SMP_CACHE_BYTES could be enough, but init_bootmem_node like
218 * to use that to align to PAGE_SIZE 218 * to use that to align to PAGE_SIZE
diff --git a/arch/x86/mm/pageattr-test.c b/arch/x86/mm/pageattr-test.c
index d4aa503caaa2..e1d106909218 100644
--- a/arch/x86/mm/pageattr-test.c
+++ b/arch/x86/mm/pageattr-test.c
@@ -32,7 +32,7 @@ enum {
32 GPS = (1<<30) 32 GPS = (1<<30)
33}; 33};
34 34
35#define PAGE_TESTBIT __pgprot(_PAGE_UNUSED1) 35#define PAGE_CPA_TEST __pgprot(_PAGE_CPA_TEST)
36 36
37static int pte_testbit(pte_t pte) 37static int pte_testbit(pte_t pte)
38{ 38{
@@ -118,6 +118,7 @@ static int pageattr_test(void)
118 unsigned int level; 118 unsigned int level;
119 int i, k; 119 int i, k;
120 int err; 120 int err;
121 unsigned long test_addr;
121 122
122 if (print) 123 if (print)
123 printk(KERN_INFO "CPA self-test:\n"); 124 printk(KERN_INFO "CPA self-test:\n");
@@ -172,7 +173,8 @@ static int pageattr_test(void)
172 continue; 173 continue;
173 } 174 }
174 175
175 err = change_page_attr_set(addr[i], len[i], PAGE_TESTBIT); 176 test_addr = addr[i];
177 err = change_page_attr_set(&test_addr, len[i], PAGE_CPA_TEST, 0);
176 if (err < 0) { 178 if (err < 0) {
177 printk(KERN_ERR "CPA %d failed %d\n", i, err); 179 printk(KERN_ERR "CPA %d failed %d\n", i, err);
178 failed++; 180 failed++;
@@ -204,7 +206,8 @@ static int pageattr_test(void)
204 failed++; 206 failed++;
205 continue; 207 continue;
206 } 208 }
207 err = change_page_attr_clear(addr[i], len[i], PAGE_TESTBIT); 209 test_addr = addr[i];
210 err = change_page_attr_clear(&test_addr, len[i], PAGE_CPA_TEST, 0);
208 if (err < 0) { 211 if (err < 0) {
209 printk(KERN_ERR "CPA reverting failed: %d\n", err); 212 printk(KERN_ERR "CPA reverting failed: %d\n", err);
210 failed++; 213 failed++;
diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c
index 43e2f8483e4f..a9ec89c3fbca 100644
--- a/arch/x86/mm/pageattr.c
+++ b/arch/x86/mm/pageattr.c
@@ -25,15 +25,27 @@
25 * The current flushing context - we pass it instead of 5 arguments: 25 * The current flushing context - we pass it instead of 5 arguments:
26 */ 26 */
27struct cpa_data { 27struct cpa_data {
28 unsigned long vaddr; 28 unsigned long *vaddr;
29 pgprot_t mask_set; 29 pgprot_t mask_set;
30 pgprot_t mask_clr; 30 pgprot_t mask_clr;
31 int numpages; 31 int numpages;
32 int flushtlb; 32 int flags;
33 unsigned long pfn; 33 unsigned long pfn;
34 unsigned force_split : 1; 34 unsigned force_split : 1;
35 int curpage;
35}; 36};
36 37
38/*
39 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
40 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
41 * entries change the page attribute in parallel to some other cpu
42 * splitting a large page entry along with changing the attribute.
43 */
44static DEFINE_SPINLOCK(cpa_lock);
45
46#define CPA_FLUSHTLB 1
47#define CPA_ARRAY 2
48
37#ifdef CONFIG_PROC_FS 49#ifdef CONFIG_PROC_FS
38static unsigned long direct_pages_count[PG_LEVEL_NUM]; 50static unsigned long direct_pages_count[PG_LEVEL_NUM];
39 51
@@ -84,7 +96,7 @@ static inline unsigned long highmap_start_pfn(void)
84 96
85static inline unsigned long highmap_end_pfn(void) 97static inline unsigned long highmap_end_pfn(void)
86{ 98{
87 return __pa(round_up((unsigned long)_end, PMD_SIZE)) >> PAGE_SHIFT; 99 return __pa(roundup((unsigned long)_end, PMD_SIZE)) >> PAGE_SHIFT;
88} 100}
89 101
90#endif 102#endif
@@ -190,6 +202,41 @@ static void cpa_flush_range(unsigned long start, int numpages, int cache)
190 } 202 }
191} 203}
192 204
205static void cpa_flush_array(unsigned long *start, int numpages, int cache)
206{
207 unsigned int i, level;
208 unsigned long *addr;
209
210 BUG_ON(irqs_disabled());
211
212 on_each_cpu(__cpa_flush_range, NULL, 1);
213
214 if (!cache)
215 return;
216
217 /* 4M threshold */
218 if (numpages >= 1024) {
219 if (boot_cpu_data.x86_model >= 4)
220 wbinvd();
221 return;
222 }
223 /*
224 * We only need to flush on one CPU,
225 * clflush is a MESI-coherent instruction that
226 * will cause all other CPUs to flush the same
227 * cachelines:
228 */
229 for (i = 0, addr = start; i < numpages; i++, addr++) {
230 pte_t *pte = lookup_address(*addr, &level);
231
232 /*
233 * Only flush present addresses:
234 */
235 if (pte && (pte_val(*pte) & _PAGE_PRESENT))
236 clflush_cache_range((void *) *addr, PAGE_SIZE);
237 }
238}
239
193/* 240/*
194 * Certain areas of memory on x86 require very specific protection flags, 241 * Certain areas of memory on x86 require very specific protection flags,
195 * for example the BIOS area or kernel text. Callers don't always get this 242 * for example the BIOS area or kernel text. Callers don't always get this
@@ -398,7 +445,7 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
398 */ 445 */
399 new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot)); 446 new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot));
400 __set_pmd_pte(kpte, address, new_pte); 447 __set_pmd_pte(kpte, address, new_pte);
401 cpa->flushtlb = 1; 448 cpa->flags |= CPA_FLUSHTLB;
402 do_split = 0; 449 do_split = 0;
403 } 450 }
404 451
@@ -408,84 +455,6 @@ out_unlock:
408 return do_split; 455 return do_split;
409} 456}
410 457
411static LIST_HEAD(page_pool);
412static unsigned long pool_size, pool_pages, pool_low;
413static unsigned long pool_used, pool_failed;
414
415static void cpa_fill_pool(struct page **ret)
416{
417 gfp_t gfp = GFP_KERNEL;
418 unsigned long flags;
419 struct page *p;
420
421 /*
422 * Avoid recursion (on debug-pagealloc) and also signal
423 * our priority to get to these pagetables:
424 */
425 if (current->flags & PF_MEMALLOC)
426 return;
427 current->flags |= PF_MEMALLOC;
428
429 /*
430 * Allocate atomically from atomic contexts:
431 */
432 if (in_atomic() || irqs_disabled() || debug_pagealloc)
433 gfp = GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
434
435 while (pool_pages < pool_size || (ret && !*ret)) {
436 p = alloc_pages(gfp, 0);
437 if (!p) {
438 pool_failed++;
439 break;
440 }
441 /*
442 * If the call site needs a page right now, provide it:
443 */
444 if (ret && !*ret) {
445 *ret = p;
446 continue;
447 }
448 spin_lock_irqsave(&pgd_lock, flags);
449 list_add(&p->lru, &page_pool);
450 pool_pages++;
451 spin_unlock_irqrestore(&pgd_lock, flags);
452 }
453
454 current->flags &= ~PF_MEMALLOC;
455}
456
457#define SHIFT_MB (20 - PAGE_SHIFT)
458#define ROUND_MB_GB ((1 << 10) - 1)
459#define SHIFT_MB_GB 10
460#define POOL_PAGES_PER_GB 16
461
462void __init cpa_init(void)
463{
464 struct sysinfo si;
465 unsigned long gb;
466
467 si_meminfo(&si);
468 /*
469 * Calculate the number of pool pages:
470 *
471 * Convert totalram (nr of pages) to MiB and round to the next
472 * GiB. Shift MiB to Gib and multiply the result by
473 * POOL_PAGES_PER_GB:
474 */
475 if (debug_pagealloc) {
476 gb = ((si.totalram >> SHIFT_MB) + ROUND_MB_GB) >> SHIFT_MB_GB;
477 pool_size = POOL_PAGES_PER_GB * gb;
478 } else {
479 pool_size = 1;
480 }
481 pool_low = pool_size;
482
483 cpa_fill_pool(NULL);
484 printk(KERN_DEBUG
485 "CPA: page pool initialized %lu of %lu pages preallocated\n",
486 pool_pages, pool_size);
487}
488
489static int split_large_page(pte_t *kpte, unsigned long address) 458static int split_large_page(pte_t *kpte, unsigned long address)
490{ 459{
491 unsigned long flags, pfn, pfninc = 1; 460 unsigned long flags, pfn, pfninc = 1;
@@ -494,28 +463,15 @@ static int split_large_page(pte_t *kpte, unsigned long address)
494 pgprot_t ref_prot; 463 pgprot_t ref_prot;
495 struct page *base; 464 struct page *base;
496 465
497 /* 466 if (!debug_pagealloc)
498 * Get a page from the pool. The pool list is protected by the 467 spin_unlock(&cpa_lock);
499 * pgd_lock, which we have to take anyway for the split 468 base = alloc_pages(GFP_KERNEL, 0);
500 * operation: 469 if (!debug_pagealloc)
501 */ 470 spin_lock(&cpa_lock);
502 spin_lock_irqsave(&pgd_lock, flags); 471 if (!base)
503 if (list_empty(&page_pool)) { 472 return -ENOMEM;
504 spin_unlock_irqrestore(&pgd_lock, flags);
505 base = NULL;
506 cpa_fill_pool(&base);
507 if (!base)
508 return -ENOMEM;
509 spin_lock_irqsave(&pgd_lock, flags);
510 } else {
511 base = list_first_entry(&page_pool, struct page, lru);
512 list_del(&base->lru);
513 pool_pages--;
514
515 if (pool_pages < pool_low)
516 pool_low = pool_pages;
517 }
518 473
474 spin_lock_irqsave(&pgd_lock, flags);
519 /* 475 /*
520 * Check for races, another CPU might have split this page 476 * Check for races, another CPU might have split this page
521 * up for us already: 477 * up for us already:
@@ -572,11 +528,8 @@ out_unlock:
572 * If we dropped out via the lookup_address check under 528 * If we dropped out via the lookup_address check under
573 * pgd_lock then stick the page back into the pool: 529 * pgd_lock then stick the page back into the pool:
574 */ 530 */
575 if (base) { 531 if (base)
576 list_add(&base->lru, &page_pool); 532 __free_page(base);
577 pool_pages++;
578 } else
579 pool_used++;
580 spin_unlock_irqrestore(&pgd_lock, flags); 533 spin_unlock_irqrestore(&pgd_lock, flags);
581 534
582 return 0; 535 return 0;
@@ -584,11 +537,16 @@ out_unlock:
584 537
585static int __change_page_attr(struct cpa_data *cpa, int primary) 538static int __change_page_attr(struct cpa_data *cpa, int primary)
586{ 539{
587 unsigned long address = cpa->vaddr; 540 unsigned long address;
588 int do_split, err; 541 int do_split, err;
589 unsigned int level; 542 unsigned int level;
590 pte_t *kpte, old_pte; 543 pte_t *kpte, old_pte;
591 544
545 if (cpa->flags & CPA_ARRAY)
546 address = cpa->vaddr[cpa->curpage];
547 else
548 address = *cpa->vaddr;
549
592repeat: 550repeat:
593 kpte = lookup_address(address, &level); 551 kpte = lookup_address(address, &level);
594 if (!kpte) 552 if (!kpte)
@@ -600,7 +558,7 @@ repeat:
600 return 0; 558 return 0;
601 WARN(1, KERN_WARNING "CPA: called for zero pte. " 559 WARN(1, KERN_WARNING "CPA: called for zero pte. "
602 "vaddr = %lx cpa->vaddr = %lx\n", address, 560 "vaddr = %lx cpa->vaddr = %lx\n", address,
603 cpa->vaddr); 561 *cpa->vaddr);
604 return -EINVAL; 562 return -EINVAL;
605 } 563 }
606 564
@@ -626,7 +584,7 @@ repeat:
626 */ 584 */
627 if (pte_val(old_pte) != pte_val(new_pte)) { 585 if (pte_val(old_pte) != pte_val(new_pte)) {
628 set_pte_atomic(kpte, new_pte); 586 set_pte_atomic(kpte, new_pte);
629 cpa->flushtlb = 1; 587 cpa->flags |= CPA_FLUSHTLB;
630 } 588 }
631 cpa->numpages = 1; 589 cpa->numpages = 1;
632 return 0; 590 return 0;
@@ -650,7 +608,25 @@ repeat:
650 */ 608 */
651 err = split_large_page(kpte, address); 609 err = split_large_page(kpte, address);
652 if (!err) { 610 if (!err) {
653 cpa->flushtlb = 1; 611 /*
612 * Do a global flush tlb after splitting the large page
613 * and before we do the actual change page attribute in the PTE.
614 *
615 * With out this, we violate the TLB application note, that says
616 * "The TLBs may contain both ordinary and large-page
617 * translations for a 4-KByte range of linear addresses. This
618 * may occur if software modifies the paging structures so that
619 * the page size used for the address range changes. If the two
620 * translations differ with respect to page frame or attributes
621 * (e.g., permissions), processor behavior is undefined and may
622 * be implementation-specific."
623 *
624 * We do this global tlb flush inside the cpa_lock, so that we
625 * don't allow any other cpu, with stale tlb entries change the
626 * page attribute in parallel, that also falls into the
627 * just split large page entry.
628 */
629 flush_tlb_all();
654 goto repeat; 630 goto repeat;
655 } 631 }
656 632
@@ -663,6 +639,7 @@ static int cpa_process_alias(struct cpa_data *cpa)
663{ 639{
664 struct cpa_data alias_cpa; 640 struct cpa_data alias_cpa;
665 int ret = 0; 641 int ret = 0;
642 unsigned long temp_cpa_vaddr, vaddr;
666 643
667 if (cpa->pfn >= max_pfn_mapped) 644 if (cpa->pfn >= max_pfn_mapped)
668 return 0; 645 return 0;
@@ -675,16 +652,24 @@ static int cpa_process_alias(struct cpa_data *cpa)
675 * No need to redo, when the primary call touched the direct 652 * No need to redo, when the primary call touched the direct
676 * mapping already: 653 * mapping already:
677 */ 654 */
678 if (!(within(cpa->vaddr, PAGE_OFFSET, 655 if (cpa->flags & CPA_ARRAY)
656 vaddr = cpa->vaddr[cpa->curpage];
657 else
658 vaddr = *cpa->vaddr;
659
660 if (!(within(vaddr, PAGE_OFFSET,
679 PAGE_OFFSET + (max_low_pfn_mapped << PAGE_SHIFT)) 661 PAGE_OFFSET + (max_low_pfn_mapped << PAGE_SHIFT))
680#ifdef CONFIG_X86_64 662#ifdef CONFIG_X86_64
681 || within(cpa->vaddr, PAGE_OFFSET + (1UL<<32), 663 || within(vaddr, PAGE_OFFSET + (1UL<<32),
682 PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)) 664 PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))
683#endif 665#endif
684 )) { 666 )) {
685 667
686 alias_cpa = *cpa; 668 alias_cpa = *cpa;
687 alias_cpa.vaddr = (unsigned long) __va(cpa->pfn << PAGE_SHIFT); 669 temp_cpa_vaddr = (unsigned long) __va(cpa->pfn << PAGE_SHIFT);
670 alias_cpa.vaddr = &temp_cpa_vaddr;
671 alias_cpa.flags &= ~CPA_ARRAY;
672
688 673
689 ret = __change_page_attr_set_clr(&alias_cpa, 0); 674 ret = __change_page_attr_set_clr(&alias_cpa, 0);
690 } 675 }
@@ -696,7 +681,7 @@ static int cpa_process_alias(struct cpa_data *cpa)
696 * No need to redo, when the primary call touched the high 681 * No need to redo, when the primary call touched the high
697 * mapping already: 682 * mapping already:
698 */ 683 */
699 if (within(cpa->vaddr, (unsigned long) _text, (unsigned long) _end)) 684 if (within(vaddr, (unsigned long) _text, (unsigned long) _end))
700 return 0; 685 return 0;
701 686
702 /* 687 /*
@@ -707,8 +692,9 @@ static int cpa_process_alias(struct cpa_data *cpa)
707 return 0; 692 return 0;
708 693
709 alias_cpa = *cpa; 694 alias_cpa = *cpa;
710 alias_cpa.vaddr = 695 temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) + __START_KERNEL_map - phys_base;
711 (cpa->pfn << PAGE_SHIFT) + __START_KERNEL_map - phys_base; 696 alias_cpa.vaddr = &temp_cpa_vaddr;
697 alias_cpa.flags &= ~CPA_ARRAY;
712 698
713 /* 699 /*
714 * The high mapping range is imprecise, so ignore the return value. 700 * The high mapping range is imprecise, so ignore the return value.
@@ -728,8 +714,15 @@ static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
728 * preservation check. 714 * preservation check.
729 */ 715 */
730 cpa->numpages = numpages; 716 cpa->numpages = numpages;
717 /* for array changes, we can't use large page */
718 if (cpa->flags & CPA_ARRAY)
719 cpa->numpages = 1;
731 720
721 if (!debug_pagealloc)
722 spin_lock(&cpa_lock);
732 ret = __change_page_attr(cpa, checkalias); 723 ret = __change_page_attr(cpa, checkalias);
724 if (!debug_pagealloc)
725 spin_unlock(&cpa_lock);
733 if (ret) 726 if (ret)
734 return ret; 727 return ret;
735 728
@@ -746,7 +739,11 @@ static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
746 */ 739 */
747 BUG_ON(cpa->numpages > numpages); 740 BUG_ON(cpa->numpages > numpages);
748 numpages -= cpa->numpages; 741 numpages -= cpa->numpages;
749 cpa->vaddr += cpa->numpages * PAGE_SIZE; 742 if (cpa->flags & CPA_ARRAY)
743 cpa->curpage++;
744 else
745 *cpa->vaddr += cpa->numpages * PAGE_SIZE;
746
750 } 747 }
751 return 0; 748 return 0;
752} 749}
@@ -757,9 +754,9 @@ static inline int cache_attr(pgprot_t attr)
757 (_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD); 754 (_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD);
758} 755}
759 756
760static int change_page_attr_set_clr(unsigned long addr, int numpages, 757static int change_page_attr_set_clr(unsigned long *addr, int numpages,
761 pgprot_t mask_set, pgprot_t mask_clr, 758 pgprot_t mask_set, pgprot_t mask_clr,
762 int force_split) 759 int force_split, int array)
763{ 760{
764 struct cpa_data cpa; 761 struct cpa_data cpa;
765 int ret, cache, checkalias; 762 int ret, cache, checkalias;
@@ -774,21 +771,38 @@ static int change_page_attr_set_clr(unsigned long addr, int numpages,
774 return 0; 771 return 0;
775 772
776 /* Ensure we are PAGE_SIZE aligned */ 773 /* Ensure we are PAGE_SIZE aligned */
777 if (addr & ~PAGE_MASK) { 774 if (!array) {
778 addr &= PAGE_MASK; 775 if (*addr & ~PAGE_MASK) {
779 /* 776 *addr &= PAGE_MASK;
780 * People should not be passing in unaligned addresses: 777 /*
781 */ 778 * People should not be passing in unaligned addresses:
782 WARN_ON_ONCE(1); 779 */
780 WARN_ON_ONCE(1);
781 }
782 } else {
783 int i;
784 for (i = 0; i < numpages; i++) {
785 if (addr[i] & ~PAGE_MASK) {
786 addr[i] &= PAGE_MASK;
787 WARN_ON_ONCE(1);
788 }
789 }
783 } 790 }
784 791
792 /* Must avoid aliasing mappings in the highmem code */
793 kmap_flush_unused();
794
785 cpa.vaddr = addr; 795 cpa.vaddr = addr;
786 cpa.numpages = numpages; 796 cpa.numpages = numpages;
787 cpa.mask_set = mask_set; 797 cpa.mask_set = mask_set;
788 cpa.mask_clr = mask_clr; 798 cpa.mask_clr = mask_clr;
789 cpa.flushtlb = 0; 799 cpa.flags = 0;
800 cpa.curpage = 0;
790 cpa.force_split = force_split; 801 cpa.force_split = force_split;
791 802
803 if (array)
804 cpa.flags |= CPA_ARRAY;
805
792 /* No alias checking for _NX bit modifications */ 806 /* No alias checking for _NX bit modifications */
793 checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX; 807 checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;
794 808
@@ -797,7 +811,7 @@ static int change_page_attr_set_clr(unsigned long addr, int numpages,
797 /* 811 /*
798 * Check whether we really changed something: 812 * Check whether we really changed something:
799 */ 813 */
800 if (!cpa.flushtlb) 814 if (!(cpa.flags & CPA_FLUSHTLB))
801 goto out; 815 goto out;
802 816
803 /* 817 /*
@@ -812,27 +826,30 @@ static int change_page_attr_set_clr(unsigned long addr, int numpages,
812 * error case we fall back to cpa_flush_all (which uses 826 * error case we fall back to cpa_flush_all (which uses
813 * wbindv): 827 * wbindv):
814 */ 828 */
815 if (!ret && cpu_has_clflush) 829 if (!ret && cpu_has_clflush) {
816 cpa_flush_range(addr, numpages, cache); 830 if (cpa.flags & CPA_ARRAY)
817 else 831 cpa_flush_array(addr, numpages, cache);
832 else
833 cpa_flush_range(*addr, numpages, cache);
834 } else
818 cpa_flush_all(cache); 835 cpa_flush_all(cache);
819 836
820out: 837out:
821 cpa_fill_pool(NULL);
822
823 return ret; 838 return ret;
824} 839}
825 840
826static inline int change_page_attr_set(unsigned long addr, int numpages, 841static inline int change_page_attr_set(unsigned long *addr, int numpages,
827 pgprot_t mask) 842 pgprot_t mask, int array)
828{ 843{
829 return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0); 844 return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
845 array);
830} 846}
831 847
832static inline int change_page_attr_clear(unsigned long addr, int numpages, 848static inline int change_page_attr_clear(unsigned long *addr, int numpages,
833 pgprot_t mask) 849 pgprot_t mask, int array)
834{ 850{
835 return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0); 851 return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
852 array);
836} 853}
837 854
838int _set_memory_uc(unsigned long addr, int numpages) 855int _set_memory_uc(unsigned long addr, int numpages)
@@ -840,8 +857,8 @@ int _set_memory_uc(unsigned long addr, int numpages)
840 /* 857 /*
841 * for now UC MINUS. see comments in ioremap_nocache() 858 * for now UC MINUS. see comments in ioremap_nocache()
842 */ 859 */
843 return change_page_attr_set(addr, numpages, 860 return change_page_attr_set(&addr, numpages,
844 __pgprot(_PAGE_CACHE_UC_MINUS)); 861 __pgprot(_PAGE_CACHE_UC_MINUS), 0);
845} 862}
846 863
847int set_memory_uc(unsigned long addr, int numpages) 864int set_memory_uc(unsigned long addr, int numpages)
@@ -857,10 +874,48 @@ int set_memory_uc(unsigned long addr, int numpages)
857} 874}
858EXPORT_SYMBOL(set_memory_uc); 875EXPORT_SYMBOL(set_memory_uc);
859 876
877int set_memory_array_uc(unsigned long *addr, int addrinarray)
878{
879 unsigned long start;
880 unsigned long end;
881 int i;
882 /*
883 * for now UC MINUS. see comments in ioremap_nocache()
884 */
885 for (i = 0; i < addrinarray; i++) {
886 start = __pa(addr[i]);
887 for (end = start + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
888 if (end != __pa(addr[i + 1]))
889 break;
890 i++;
891 }
892 if (reserve_memtype(start, end, _PAGE_CACHE_UC_MINUS, NULL))
893 goto out;
894 }
895
896 return change_page_attr_set(addr, addrinarray,
897 __pgprot(_PAGE_CACHE_UC_MINUS), 1);
898out:
899 for (i = 0; i < addrinarray; i++) {
900 unsigned long tmp = __pa(addr[i]);
901
902 if (tmp == start)
903 break;
904 for (end = tmp + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
905 if (end != __pa(addr[i + 1]))
906 break;
907 i++;
908 }
909 free_memtype(tmp, end);
910 }
911 return -EINVAL;
912}
913EXPORT_SYMBOL(set_memory_array_uc);
914
860int _set_memory_wc(unsigned long addr, int numpages) 915int _set_memory_wc(unsigned long addr, int numpages)
861{ 916{
862 return change_page_attr_set(addr, numpages, 917 return change_page_attr_set(&addr, numpages,
863 __pgprot(_PAGE_CACHE_WC)); 918 __pgprot(_PAGE_CACHE_WC), 0);
864} 919}
865 920
866int set_memory_wc(unsigned long addr, int numpages) 921int set_memory_wc(unsigned long addr, int numpages)
@@ -878,8 +933,8 @@ EXPORT_SYMBOL(set_memory_wc);
878 933
879int _set_memory_wb(unsigned long addr, int numpages) 934int _set_memory_wb(unsigned long addr, int numpages)
880{ 935{
881 return change_page_attr_clear(addr, numpages, 936 return change_page_attr_clear(&addr, numpages,
882 __pgprot(_PAGE_CACHE_MASK)); 937 __pgprot(_PAGE_CACHE_MASK), 0);
883} 938}
884 939
885int set_memory_wb(unsigned long addr, int numpages) 940int set_memory_wb(unsigned long addr, int numpages)
@@ -890,37 +945,59 @@ int set_memory_wb(unsigned long addr, int numpages)
890} 945}
891EXPORT_SYMBOL(set_memory_wb); 946EXPORT_SYMBOL(set_memory_wb);
892 947
948int set_memory_array_wb(unsigned long *addr, int addrinarray)
949{
950 int i;
951
952 for (i = 0; i < addrinarray; i++) {
953 unsigned long start = __pa(addr[i]);
954 unsigned long end;
955
956 for (end = start + PAGE_SIZE; i < addrinarray - 1; end += PAGE_SIZE) {
957 if (end != __pa(addr[i + 1]))
958 break;
959 i++;
960 }
961 free_memtype(start, end);
962 }
963 return change_page_attr_clear(addr, addrinarray,
964 __pgprot(_PAGE_CACHE_MASK), 1);
965}
966EXPORT_SYMBOL(set_memory_array_wb);
967
893int set_memory_x(unsigned long addr, int numpages) 968int set_memory_x(unsigned long addr, int numpages)
894{ 969{
895 return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX)); 970 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
896} 971}
897EXPORT_SYMBOL(set_memory_x); 972EXPORT_SYMBOL(set_memory_x);
898 973
899int set_memory_nx(unsigned long addr, int numpages) 974int set_memory_nx(unsigned long addr, int numpages)
900{ 975{
901 return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX)); 976 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
902} 977}
903EXPORT_SYMBOL(set_memory_nx); 978EXPORT_SYMBOL(set_memory_nx);
904 979
905int set_memory_ro(unsigned long addr, int numpages) 980int set_memory_ro(unsigned long addr, int numpages)
906{ 981{
907 return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW)); 982 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
908} 983}
984EXPORT_SYMBOL_GPL(set_memory_ro);
909 985
910int set_memory_rw(unsigned long addr, int numpages) 986int set_memory_rw(unsigned long addr, int numpages)
911{ 987{
912 return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW)); 988 return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
913} 989}
990EXPORT_SYMBOL_GPL(set_memory_rw);
914 991
915int set_memory_np(unsigned long addr, int numpages) 992int set_memory_np(unsigned long addr, int numpages)
916{ 993{
917 return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT)); 994 return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
918} 995}
919 996
920int set_memory_4k(unsigned long addr, int numpages) 997int set_memory_4k(unsigned long addr, int numpages)
921{ 998{
922 return change_page_attr_set_clr(addr, numpages, __pgprot(0), 999 return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
923 __pgprot(0), 1); 1000 __pgprot(0), 1, 0);
924} 1001}
925 1002
926int set_pages_uc(struct page *page, int numpages) 1003int set_pages_uc(struct page *page, int numpages)
@@ -973,22 +1050,38 @@ int set_pages_rw(struct page *page, int numpages)
973 1050
974static int __set_pages_p(struct page *page, int numpages) 1051static int __set_pages_p(struct page *page, int numpages)
975{ 1052{
976 struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page), 1053 unsigned long tempaddr = (unsigned long) page_address(page);
1054 struct cpa_data cpa = { .vaddr = &tempaddr,
977 .numpages = numpages, 1055 .numpages = numpages,
978 .mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW), 1056 .mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
979 .mask_clr = __pgprot(0)}; 1057 .mask_clr = __pgprot(0),
1058 .flags = 0};
980 1059
981 return __change_page_attr_set_clr(&cpa, 1); 1060 /*
1061 * No alias checking needed for setting present flag. otherwise,
1062 * we may need to break large pages for 64-bit kernel text
1063 * mappings (this adds to complexity if we want to do this from
1064 * atomic context especially). Let's keep it simple!
1065 */
1066 return __change_page_attr_set_clr(&cpa, 0);
982} 1067}
983 1068
984static int __set_pages_np(struct page *page, int numpages) 1069static int __set_pages_np(struct page *page, int numpages)
985{ 1070{
986 struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page), 1071 unsigned long tempaddr = (unsigned long) page_address(page);
1072 struct cpa_data cpa = { .vaddr = &tempaddr,
987 .numpages = numpages, 1073 .numpages = numpages,
988 .mask_set = __pgprot(0), 1074 .mask_set = __pgprot(0),
989 .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW)}; 1075 .mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
1076 .flags = 0};
990 1077
991 return __change_page_attr_set_clr(&cpa, 1); 1078 /*
1079 * No alias checking needed for setting not present flag. otherwise,
1080 * we may need to break large pages for 64-bit kernel text
1081 * mappings (this adds to complexity if we want to do this from
1082 * atomic context especially). Let's keep it simple!
1083 */
1084 return __change_page_attr_set_clr(&cpa, 0);
992} 1085}
993 1086
994void kernel_map_pages(struct page *page, int numpages, int enable) 1087void kernel_map_pages(struct page *page, int numpages, int enable)
@@ -1008,11 +1101,8 @@ void kernel_map_pages(struct page *page, int numpages, int enable)
1008 1101
1009 /* 1102 /*
1010 * The return value is ignored as the calls cannot fail. 1103 * The return value is ignored as the calls cannot fail.
1011 * Large pages are kept enabled at boot time, and are 1104 * Large pages for identity mappings are not used at boot time
1012 * split up quickly with DEBUG_PAGEALLOC. If a splitup 1105 * and hence no memory allocations during large page split.
1013 * fails here (due to temporary memory shortage) no damage
1014 * is done because we just keep the largepage intact up
1015 * to the next attempt when it will likely be split up:
1016 */ 1106 */
1017 if (enable) 1107 if (enable)
1018 __set_pages_p(page, numpages); 1108 __set_pages_p(page, numpages);
@@ -1024,53 +1114,8 @@ void kernel_map_pages(struct page *page, int numpages, int enable)
1024 * but that can deadlock->flush only current cpu: 1114 * but that can deadlock->flush only current cpu:
1025 */ 1115 */
1026 __flush_tlb_all(); 1116 __flush_tlb_all();
1027
1028 /*
1029 * Try to refill the page pool here. We can do this only after
1030 * the tlb flush.
1031 */
1032 cpa_fill_pool(NULL);
1033} 1117}
1034 1118
1035#ifdef CONFIG_DEBUG_FS
1036static int dpa_show(struct seq_file *m, void *v)
1037{
1038 seq_puts(m, "DEBUG_PAGEALLOC\n");
1039 seq_printf(m, "pool_size : %lu\n", pool_size);
1040 seq_printf(m, "pool_pages : %lu\n", pool_pages);
1041 seq_printf(m, "pool_low : %lu\n", pool_low);
1042 seq_printf(m, "pool_used : %lu\n", pool_used);
1043 seq_printf(m, "pool_failed : %lu\n", pool_failed);
1044
1045 return 0;
1046}
1047
1048static int dpa_open(struct inode *inode, struct file *filp)
1049{
1050 return single_open(filp, dpa_show, NULL);
1051}
1052
1053static const struct file_operations dpa_fops = {
1054 .open = dpa_open,
1055 .read = seq_read,
1056 .llseek = seq_lseek,
1057 .release = single_release,
1058};
1059
1060static int __init debug_pagealloc_proc_init(void)
1061{
1062 struct dentry *de;
1063
1064 de = debugfs_create_file("debug_pagealloc", 0600, NULL, NULL,
1065 &dpa_fops);
1066 if (!de)
1067 return -ENOMEM;
1068
1069 return 0;
1070}
1071__initcall(debug_pagealloc_proc_init);
1072#endif
1073
1074#ifdef CONFIG_HIBERNATION 1119#ifdef CONFIG_HIBERNATION
1075 1120
1076bool kernel_page_present(struct page *page) 1121bool kernel_page_present(struct page *page)
diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c
index 2a50e0fa64a5..738fd0f24958 100644
--- a/arch/x86/mm/pat.c
+++ b/arch/x86/mm/pat.c
@@ -7,24 +7,24 @@
7 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen. 7 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
8 */ 8 */
9 9
10#include <linux/mm.h> 10#include <linux/seq_file.h>
11#include <linux/bootmem.h>
12#include <linux/debugfs.h>
11#include <linux/kernel.h> 13#include <linux/kernel.h>
12#include <linux/gfp.h> 14#include <linux/gfp.h>
15#include <linux/mm.h>
13#include <linux/fs.h> 16#include <linux/fs.h>
14#include <linux/bootmem.h>
15#include <linux/debugfs.h>
16#include <linux/seq_file.h>
17 17
18#include <asm/msr.h> 18#include <asm/cacheflush.h>
19#include <asm/tlbflush.h>
20#include <asm/processor.h> 19#include <asm/processor.h>
21#include <asm/page.h> 20#include <asm/tlbflush.h>
22#include <asm/pgtable.h> 21#include <asm/pgtable.h>
23#include <asm/pat.h>
24#include <asm/e820.h>
25#include <asm/cacheflush.h>
26#include <asm/fcntl.h> 22#include <asm/fcntl.h>
23#include <asm/e820.h>
27#include <asm/mtrr.h> 24#include <asm/mtrr.h>
25#include <asm/page.h>
26#include <asm/msr.h>
27#include <asm/pat.h>
28#include <asm/io.h> 28#include <asm/io.h>
29 29
30#ifdef CONFIG_X86_PAT 30#ifdef CONFIG_X86_PAT
@@ -46,6 +46,7 @@ early_param("nopat", nopat);
46 46
47 47
48static int debug_enable; 48static int debug_enable;
49
49static int __init pat_debug_setup(char *str) 50static int __init pat_debug_setup(char *str)
50{ 51{
51 debug_enable = 1; 52 debug_enable = 1;
@@ -145,14 +146,14 @@ static char *cattr_name(unsigned long flags)
145 */ 146 */
146 147
147struct memtype { 148struct memtype {
148 u64 start; 149 u64 start;
149 u64 end; 150 u64 end;
150 unsigned long type; 151 unsigned long type;
151 struct list_head nd; 152 struct list_head nd;
152}; 153};
153 154
154static LIST_HEAD(memtype_list); 155static LIST_HEAD(memtype_list);
155static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */ 156static DEFINE_SPINLOCK(memtype_lock); /* protects memtype list */
156 157
157/* 158/*
158 * Does intersection of PAT memory type and MTRR memory type and returns 159 * Does intersection of PAT memory type and MTRR memory type and returns
@@ -180,8 +181,8 @@ static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type)
180 return req_type; 181 return req_type;
181} 182}
182 183
183static int chk_conflict(struct memtype *new, struct memtype *entry, 184static int
184 unsigned long *type) 185chk_conflict(struct memtype *new, struct memtype *entry, unsigned long *type)
185{ 186{
186 if (new->type != entry->type) { 187 if (new->type != entry->type) {
187 if (type) { 188 if (type) {
@@ -211,6 +212,66 @@ static struct memtype *cached_entry;
211static u64 cached_start; 212static u64 cached_start;
212 213
213/* 214/*
215 * For RAM pages, mark the pages as non WB memory type using
216 * PageNonWB (PG_arch_1). We allow only one set_memory_uc() or
217 * set_memory_wc() on a RAM page at a time before marking it as WB again.
218 * This is ok, because only one driver will be owning the page and
219 * doing set_memory_*() calls.
220 *
221 * For now, we use PageNonWB to track that the RAM page is being mapped
222 * as non WB. In future, we will have to use one more flag
223 * (or some other mechanism in page_struct) to distinguish between
224 * UC and WC mapping.
225 */
226static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type,
227 unsigned long *new_type)
228{
229 struct page *page;
230 u64 pfn, end_pfn;
231
232 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
233 page = pfn_to_page(pfn);
234 if (page_mapped(page) || PageNonWB(page))
235 goto out;
236
237 SetPageNonWB(page);
238 }
239 return 0;
240
241out:
242 end_pfn = pfn;
243 for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) {
244 page = pfn_to_page(pfn);
245 ClearPageNonWB(page);
246 }
247
248 return -EINVAL;
249}
250
251static int free_ram_pages_type(u64 start, u64 end)
252{
253 struct page *page;
254 u64 pfn, end_pfn;
255
256 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
257 page = pfn_to_page(pfn);
258 if (page_mapped(page) || !PageNonWB(page))
259 goto out;
260
261 ClearPageNonWB(page);
262 }
263 return 0;
264
265out:
266 end_pfn = pfn;
267 for (pfn = (start >> PAGE_SHIFT); pfn < end_pfn; ++pfn) {
268 page = pfn_to_page(pfn);
269 SetPageNonWB(page);
270 }
271 return -EINVAL;
272}
273
274/*
214 * req_type typically has one of the: 275 * req_type typically has one of the:
215 * - _PAGE_CACHE_WB 276 * - _PAGE_CACHE_WB
216 * - _PAGE_CACHE_WC 277 * - _PAGE_CACHE_WC
@@ -226,14 +287,15 @@ static u64 cached_start;
226 * it will return a negative return value. 287 * it will return a negative return value.
227 */ 288 */
228int reserve_memtype(u64 start, u64 end, unsigned long req_type, 289int reserve_memtype(u64 start, u64 end, unsigned long req_type,
229 unsigned long *new_type) 290 unsigned long *new_type)
230{ 291{
231 struct memtype *new, *entry; 292 struct memtype *new, *entry;
232 unsigned long actual_type; 293 unsigned long actual_type;
233 struct list_head *where; 294 struct list_head *where;
295 int is_range_ram;
234 int err = 0; 296 int err = 0;
235 297
236 BUG_ON(start >= end); /* end is exclusive */ 298 BUG_ON(start >= end); /* end is exclusive */
237 299
238 if (!pat_enabled) { 300 if (!pat_enabled) {
239 /* This is identical to page table setting without PAT */ 301 /* This is identical to page table setting without PAT */
@@ -266,17 +328,24 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type,
266 actual_type = _PAGE_CACHE_WB; 328 actual_type = _PAGE_CACHE_WB;
267 else 329 else
268 actual_type = _PAGE_CACHE_UC_MINUS; 330 actual_type = _PAGE_CACHE_UC_MINUS;
269 } else 331 } else {
270 actual_type = pat_x_mtrr_type(start, end, 332 actual_type = pat_x_mtrr_type(start, end,
271 req_type & _PAGE_CACHE_MASK); 333 req_type & _PAGE_CACHE_MASK);
334 }
335
336 is_range_ram = pagerange_is_ram(start, end);
337 if (is_range_ram == 1)
338 return reserve_ram_pages_type(start, end, req_type, new_type);
339 else if (is_range_ram < 0)
340 return -EINVAL;
272 341
273 new = kmalloc(sizeof(struct memtype), GFP_KERNEL); 342 new = kmalloc(sizeof(struct memtype), GFP_KERNEL);
274 if (!new) 343 if (!new)
275 return -ENOMEM; 344 return -ENOMEM;
276 345
277 new->start = start; 346 new->start = start;
278 new->end = end; 347 new->end = end;
279 new->type = actual_type; 348 new->type = actual_type;
280 349
281 if (new_type) 350 if (new_type)
282 *new_type = actual_type; 351 *new_type = actual_type;
@@ -335,6 +404,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type,
335 start, end, cattr_name(new->type), cattr_name(req_type)); 404 start, end, cattr_name(new->type), cattr_name(req_type));
336 kfree(new); 405 kfree(new);
337 spin_unlock(&memtype_lock); 406 spin_unlock(&memtype_lock);
407
338 return err; 408 return err;
339 } 409 }
340 410
@@ -358,6 +428,7 @@ int free_memtype(u64 start, u64 end)
358{ 428{
359 struct memtype *entry; 429 struct memtype *entry;
360 int err = -EINVAL; 430 int err = -EINVAL;
431 int is_range_ram;
361 432
362 if (!pat_enabled) 433 if (!pat_enabled)
363 return 0; 434 return 0;
@@ -366,6 +437,12 @@ int free_memtype(u64 start, u64 end)
366 if (is_ISA_range(start, end - 1)) 437 if (is_ISA_range(start, end - 1))
367 return 0; 438 return 0;
368 439
440 is_range_ram = pagerange_is_ram(start, end);
441 if (is_range_ram == 1)
442 return free_ram_pages_type(start, end);
443 else if (is_range_ram < 0)
444 return -EINVAL;
445
369 spin_lock(&memtype_lock); 446 spin_lock(&memtype_lock);
370 list_for_each_entry(entry, &memtype_list, nd) { 447 list_for_each_entry(entry, &memtype_list, nd) {
371 if (entry->start == start && entry->end == end) { 448 if (entry->start == start && entry->end == end) {
@@ -386,6 +463,7 @@ int free_memtype(u64 start, u64 end)
386 } 463 }
387 464
388 dprintk("free_memtype request 0x%Lx-0x%Lx\n", start, end); 465 dprintk("free_memtype request 0x%Lx-0x%Lx\n", start, end);
466
389 return err; 467 return err;
390} 468}
391 469
@@ -492,9 +570,9 @@ int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
492 570
493void map_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot) 571void map_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot)
494{ 572{
573 unsigned long want_flags = (pgprot_val(vma_prot) & _PAGE_CACHE_MASK);
495 u64 addr = (u64)pfn << PAGE_SHIFT; 574 u64 addr = (u64)pfn << PAGE_SHIFT;
496 unsigned long flags; 575 unsigned long flags;
497 unsigned long want_flags = (pgprot_val(vma_prot) & _PAGE_CACHE_MASK);
498 576
499 reserve_memtype(addr, addr + size, want_flags, &flags); 577 reserve_memtype(addr, addr + size, want_flags, &flags);
500 if (flags != want_flags) { 578 if (flags != want_flags) {
@@ -514,7 +592,7 @@ void unmap_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot)
514 free_memtype(addr, addr + size); 592 free_memtype(addr, addr + size);
515} 593}
516 594
517#if defined(CONFIG_DEBUG_FS) 595#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
518 596
519/* get Nth element of the linked list */ 597/* get Nth element of the linked list */
520static struct memtype *memtype_get_idx(loff_t pos) 598static struct memtype *memtype_get_idx(loff_t pos)
@@ -537,6 +615,7 @@ static struct memtype *memtype_get_idx(loff_t pos)
537 } 615 }
538 spin_unlock(&memtype_lock); 616 spin_unlock(&memtype_lock);
539 kfree(print_entry); 617 kfree(print_entry);
618
540 return NULL; 619 return NULL;
541} 620}
542 621
@@ -567,6 +646,7 @@ static int memtype_seq_show(struct seq_file *seq, void *v)
567 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type), 646 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
568 print_entry->start, print_entry->end); 647 print_entry->start, print_entry->end);
569 kfree(print_entry); 648 kfree(print_entry);
649
570 return 0; 650 return 0;
571} 651}
572 652
@@ -598,4 +678,4 @@ static int __init pat_memtype_list_init(void)
598 678
599late_initcall(pat_memtype_list_init); 679late_initcall(pat_memtype_list_init);
600 680
601#endif /* CONFIG_DEBUG_FS */ 681#endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */
diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
index d50302774fe2..86f2ffc43c3d 100644
--- a/arch/x86/mm/pgtable.c
+++ b/arch/x86/mm/pgtable.c
@@ -63,10 +63,8 @@ static inline void pgd_list_del(pgd_t *pgd)
63#define UNSHARED_PTRS_PER_PGD \ 63#define UNSHARED_PTRS_PER_PGD \
64 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD) 64 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
65 65
66static void pgd_ctor(void *p) 66static void pgd_ctor(pgd_t *pgd)
67{ 67{
68 pgd_t *pgd = p;
69
70 /* If the pgd points to a shared pagetable level (either the 68 /* If the pgd points to a shared pagetable level (either the
71 ptes in non-PAE, or shared PMD in PAE), then just copy the 69 ptes in non-PAE, or shared PMD in PAE), then just copy the
72 references from swapper_pg_dir. */ 70 references from swapper_pg_dir. */
@@ -87,7 +85,7 @@ static void pgd_ctor(void *p)
87 pgd_list_add(pgd); 85 pgd_list_add(pgd);
88} 86}
89 87
90static void pgd_dtor(void *pgd) 88static void pgd_dtor(pgd_t *pgd)
91{ 89{
92 unsigned long flags; /* can be called from interrupt context */ 90 unsigned long flags; /* can be called from interrupt context */
93 91
diff --git a/arch/x86/mm/pgtable_32.c b/arch/x86/mm/pgtable_32.c
index cab0abbd1ebe..0951db9ee519 100644
--- a/arch/x86/mm/pgtable_32.c
+++ b/arch/x86/mm/pgtable_32.c
@@ -123,7 +123,8 @@ static int __init parse_vmalloc(char *arg)
123 if (!arg) 123 if (!arg)
124 return -EINVAL; 124 return -EINVAL;
125 125
126 __VMALLOC_RESERVE = memparse(arg, &arg); 126 /* Add VMALLOC_OFFSET to the parsed value due to vm area guard hole*/
127 __VMALLOC_RESERVE = memparse(arg, &arg) + VMALLOC_OFFSET;
127 return 0; 128 return 0;
128} 129}
129early_param("vmalloc", parse_vmalloc); 130early_param("vmalloc", parse_vmalloc);
diff --git a/arch/x86/mm/srat_64.c b/arch/x86/mm/srat_64.c
index 1b4763e26ea9..51c0a2fc14fe 100644
--- a/arch/x86/mm/srat_64.c
+++ b/arch/x86/mm/srat_64.c
@@ -138,7 +138,7 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa)
138 return; 138 return;
139 } 139 }
140 140
141 if (is_uv_system()) 141 if (get_uv_system_type() >= UV_X2APIC)
142 apic_id = (pa->apic_id << 8) | pa->local_sapic_eid; 142 apic_id = (pa->apic_id << 8) | pa->local_sapic_eid;
143 else 143 else
144 apic_id = pa->apic_id; 144 apic_id = pa->apic_id;
diff --git a/arch/x86/oprofile/Makefile b/arch/x86/oprofile/Makefile
index 30f3eb366667..446902b2a6b6 100644
--- a/arch/x86/oprofile/Makefile
+++ b/arch/x86/oprofile/Makefile
@@ -7,6 +7,6 @@ DRIVER_OBJS = $(addprefix ../../../drivers/oprofile/, \
7 timer_int.o ) 7 timer_int.o )
8 8
9oprofile-y := $(DRIVER_OBJS) init.o backtrace.o 9oprofile-y := $(DRIVER_OBJS) init.o backtrace.o
10oprofile-$(CONFIG_X86_LOCAL_APIC) += nmi_int.o op_model_athlon.o \ 10oprofile-$(CONFIG_X86_LOCAL_APIC) += nmi_int.o op_model_amd.o \
11 op_model_ppro.o op_model_p4.o 11 op_model_ppro.o op_model_p4.o
12oprofile-$(CONFIG_X86_IO_APIC) += nmi_timer_int.o 12oprofile-$(CONFIG_X86_IO_APIC) += nmi_timer_int.o
diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c
index 0227694f7dab..57f6c9088081 100644
--- a/arch/x86/oprofile/nmi_int.c
+++ b/arch/x86/oprofile/nmi_int.c
@@ -1,10 +1,11 @@
1/** 1/**
2 * @file nmi_int.c 2 * @file nmi_int.c
3 * 3 *
4 * @remark Copyright 2002 OProfile authors 4 * @remark Copyright 2002-2008 OProfile authors
5 * @remark Read the file COPYING 5 * @remark Read the file COPYING
6 * 6 *
7 * @author John Levon <levon@movementarian.org> 7 * @author John Levon <levon@movementarian.org>
8 * @author Robert Richter <robert.richter@amd.com>
8 */ 9 */
9 10
10#include <linux/init.h> 11#include <linux/init.h>
@@ -295,10 +296,12 @@ static void nmi_cpu_shutdown(void *dummy)
295 296
296static void nmi_shutdown(void) 297static void nmi_shutdown(void)
297{ 298{
298 struct op_msrs *msrs = &get_cpu_var(cpu_msrs); 299 struct op_msrs *msrs;
300
299 nmi_enabled = 0; 301 nmi_enabled = 0;
300 on_each_cpu(nmi_cpu_shutdown, NULL, 1); 302 on_each_cpu(nmi_cpu_shutdown, NULL, 1);
301 unregister_die_notifier(&profile_exceptions_nb); 303 unregister_die_notifier(&profile_exceptions_nb);
304 msrs = &get_cpu_var(cpu_msrs);
302 model->shutdown(msrs); 305 model->shutdown(msrs);
303 free_msrs(); 306 free_msrs();
304 put_cpu_var(cpu_msrs); 307 put_cpu_var(cpu_msrs);
@@ -437,6 +440,7 @@ int __init op_nmi_init(struct oprofile_operations *ops)
437 __u8 vendor = boot_cpu_data.x86_vendor; 440 __u8 vendor = boot_cpu_data.x86_vendor;
438 __u8 family = boot_cpu_data.x86; 441 __u8 family = boot_cpu_data.x86;
439 char *cpu_type; 442 char *cpu_type;
443 int ret = 0;
440 444
441 if (!cpu_has_apic) 445 if (!cpu_has_apic)
442 return -ENODEV; 446 return -ENODEV;
@@ -449,19 +453,23 @@ int __init op_nmi_init(struct oprofile_operations *ops)
449 default: 453 default:
450 return -ENODEV; 454 return -ENODEV;
451 case 6: 455 case 6:
452 model = &op_athlon_spec; 456 model = &op_amd_spec;
453 cpu_type = "i386/athlon"; 457 cpu_type = "i386/athlon";
454 break; 458 break;
455 case 0xf: 459 case 0xf:
456 model = &op_athlon_spec; 460 model = &op_amd_spec;
457 /* Actually it could be i386/hammer too, but give 461 /* Actually it could be i386/hammer too, but give
458 user space an consistent name. */ 462 user space an consistent name. */
459 cpu_type = "x86-64/hammer"; 463 cpu_type = "x86-64/hammer";
460 break; 464 break;
461 case 0x10: 465 case 0x10:
462 model = &op_athlon_spec; 466 model = &op_amd_spec;
463 cpu_type = "x86-64/family10"; 467 cpu_type = "x86-64/family10";
464 break; 468 break;
469 case 0x11:
470 model = &op_amd_spec;
471 cpu_type = "x86-64/family11h";
472 break;
465 } 473 }
466 break; 474 break;
467 475
@@ -488,17 +496,24 @@ int __init op_nmi_init(struct oprofile_operations *ops)
488 return -ENODEV; 496 return -ENODEV;
489 } 497 }
490 498
491 init_sysfs();
492#ifdef CONFIG_SMP 499#ifdef CONFIG_SMP
493 register_cpu_notifier(&oprofile_cpu_nb); 500 register_cpu_notifier(&oprofile_cpu_nb);
494#endif 501#endif
495 using_nmi = 1; 502 /* default values, can be overwritten by model */
496 ops->create_files = nmi_create_files; 503 ops->create_files = nmi_create_files;
497 ops->setup = nmi_setup; 504 ops->setup = nmi_setup;
498 ops->shutdown = nmi_shutdown; 505 ops->shutdown = nmi_shutdown;
499 ops->start = nmi_start; 506 ops->start = nmi_start;
500 ops->stop = nmi_stop; 507 ops->stop = nmi_stop;
501 ops->cpu_type = cpu_type; 508 ops->cpu_type = cpu_type;
509
510 if (model->init)
511 ret = model->init(ops);
512 if (ret)
513 return ret;
514
515 init_sysfs();
516 using_nmi = 1;
502 printk(KERN_INFO "oprofile: using NMI interrupt.\n"); 517 printk(KERN_INFO "oprofile: using NMI interrupt.\n");
503 return 0; 518 return 0;
504} 519}
@@ -511,4 +526,6 @@ void op_nmi_exit(void)
511 unregister_cpu_notifier(&oprofile_cpu_nb); 526 unregister_cpu_notifier(&oprofile_cpu_nb);
512#endif 527#endif
513 } 528 }
529 if (model->exit)
530 model->exit();
514} 531}
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c
new file mode 100644
index 000000000000..d9faf607b3a6
--- /dev/null
+++ b/arch/x86/oprofile/op_model_amd.c
@@ -0,0 +1,543 @@
1/*
2 * @file op_model_amd.c
3 * athlon / K7 / K8 / Family 10h model-specific MSR operations
4 *
5 * @remark Copyright 2002-2008 OProfile authors
6 * @remark Read the file COPYING
7 *
8 * @author John Levon
9 * @author Philippe Elie
10 * @author Graydon Hoare
11 * @author Robert Richter <robert.richter@amd.com>
12 * @author Barry Kasindorf
13*/
14
15#include <linux/oprofile.h>
16#include <linux/device.h>
17#include <linux/pci.h>
18
19#include <asm/ptrace.h>
20#include <asm/msr.h>
21#include <asm/nmi.h>
22
23#include "op_x86_model.h"
24#include "op_counter.h"
25
26#define NUM_COUNTERS 4
27#define NUM_CONTROLS 4
28
29#define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0)
30#define CTR_READ(l, h, msrs, c) do {rdmsr(msrs->counters[(c)].addr, (l), (h)); } while (0)
31#define CTR_WRITE(l, msrs, c) do {wrmsr(msrs->counters[(c)].addr, -(unsigned int)(l), -1); } while (0)
32#define CTR_OVERFLOWED(n) (!((n) & (1U<<31)))
33
34#define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0)
35#define CTRL_READ(l, h, msrs, c) do {rdmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
36#define CTRL_WRITE(l, h, msrs, c) do {wrmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
37#define CTRL_SET_ACTIVE(n) (n |= (1<<22))
38#define CTRL_SET_INACTIVE(n) (n &= ~(1<<22))
39#define CTRL_CLEAR_LO(x) (x &= (1<<21))
40#define CTRL_CLEAR_HI(x) (x &= 0xfffffcf0)
41#define CTRL_SET_ENABLE(val) (val |= 1<<20)
42#define CTRL_SET_USR(val, u) (val |= ((u & 1) << 16))
43#define CTRL_SET_KERN(val, k) (val |= ((k & 1) << 17))
44#define CTRL_SET_UM(val, m) (val |= (m << 8))
45#define CTRL_SET_EVENT_LOW(val, e) (val |= (e & 0xff))
46#define CTRL_SET_EVENT_HIGH(val, e) (val |= ((e >> 8) & 0xf))
47#define CTRL_SET_HOST_ONLY(val, h) (val |= ((h & 1) << 9))
48#define CTRL_SET_GUEST_ONLY(val, h) (val |= ((h & 1) << 8))
49
50static unsigned long reset_value[NUM_COUNTERS];
51
52#ifdef CONFIG_OPROFILE_IBS
53
54/* IbsFetchCtl bits/masks */
55#define IBS_FETCH_HIGH_VALID_BIT (1UL << 17) /* bit 49 */
56#define IBS_FETCH_HIGH_ENABLE (1UL << 16) /* bit 48 */
57#define IBS_FETCH_LOW_MAX_CNT_MASK 0x0000FFFFUL /* MaxCnt mask */
58
59/*IbsOpCtl bits */
60#define IBS_OP_LOW_VALID_BIT (1ULL<<18) /* bit 18 */
61#define IBS_OP_LOW_ENABLE (1ULL<<17) /* bit 17 */
62
63/* Codes used in cpu_buffer.c */
64/* This produces duplicate code, need to be fixed */
65#define IBS_FETCH_BEGIN 3
66#define IBS_OP_BEGIN 4
67
68/* The function interface needs to be fixed, something like add
69 data. Should then be added to linux/oprofile.h. */
70extern void oprofile_add_ibs_sample(struct pt_regs *const regs,
71 unsigned int * const ibs_sample, u8 code);
72
73struct ibs_fetch_sample {
74 /* MSRC001_1031 IBS Fetch Linear Address Register */
75 unsigned int ibs_fetch_lin_addr_low;
76 unsigned int ibs_fetch_lin_addr_high;
77 /* MSRC001_1030 IBS Fetch Control Register */
78 unsigned int ibs_fetch_ctl_low;
79 unsigned int ibs_fetch_ctl_high;
80 /* MSRC001_1032 IBS Fetch Physical Address Register */
81 unsigned int ibs_fetch_phys_addr_low;
82 unsigned int ibs_fetch_phys_addr_high;
83};
84
85struct ibs_op_sample {
86 /* MSRC001_1034 IBS Op Logical Address Register (IbsRIP) */
87 unsigned int ibs_op_rip_low;
88 unsigned int ibs_op_rip_high;
89 /* MSRC001_1035 IBS Op Data Register */
90 unsigned int ibs_op_data1_low;
91 unsigned int ibs_op_data1_high;
92 /* MSRC001_1036 IBS Op Data 2 Register */
93 unsigned int ibs_op_data2_low;
94 unsigned int ibs_op_data2_high;
95 /* MSRC001_1037 IBS Op Data 3 Register */
96 unsigned int ibs_op_data3_low;
97 unsigned int ibs_op_data3_high;
98 /* MSRC001_1038 IBS DC Linear Address Register (IbsDcLinAd) */
99 unsigned int ibs_dc_linear_low;
100 unsigned int ibs_dc_linear_high;
101 /* MSRC001_1039 IBS DC Physical Address Register (IbsDcPhysAd) */
102 unsigned int ibs_dc_phys_low;
103 unsigned int ibs_dc_phys_high;
104};
105
106/*
107 * unitialize the APIC for the IBS interrupts if needed on AMD Family10h+
108*/
109static void clear_ibs_nmi(void);
110
111static int ibs_allowed; /* AMD Family10h and later */
112
113struct op_ibs_config {
114 unsigned long op_enabled;
115 unsigned long fetch_enabled;
116 unsigned long max_cnt_fetch;
117 unsigned long max_cnt_op;
118 unsigned long rand_en;
119 unsigned long dispatched_ops;
120};
121
122static struct op_ibs_config ibs_config;
123
124#endif
125
126/* functions for op_amd_spec */
127
128static void op_amd_fill_in_addresses(struct op_msrs * const msrs)
129{
130 int i;
131
132 for (i = 0; i < NUM_COUNTERS; i++) {
133 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
134 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
135 else
136 msrs->counters[i].addr = 0;
137 }
138
139 for (i = 0; i < NUM_CONTROLS; i++) {
140 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i))
141 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
142 else
143 msrs->controls[i].addr = 0;
144 }
145}
146
147
148static void op_amd_setup_ctrs(struct op_msrs const * const msrs)
149{
150 unsigned int low, high;
151 int i;
152
153 /* clear all counters */
154 for (i = 0 ; i < NUM_CONTROLS; ++i) {
155 if (unlikely(!CTRL_IS_RESERVED(msrs, i)))
156 continue;
157 CTRL_READ(low, high, msrs, i);
158 CTRL_CLEAR_LO(low);
159 CTRL_CLEAR_HI(high);
160 CTRL_WRITE(low, high, msrs, i);
161 }
162
163 /* avoid a false detection of ctr overflows in NMI handler */
164 for (i = 0; i < NUM_COUNTERS; ++i) {
165 if (unlikely(!CTR_IS_RESERVED(msrs, i)))
166 continue;
167 CTR_WRITE(1, msrs, i);
168 }
169
170 /* enable active counters */
171 for (i = 0; i < NUM_COUNTERS; ++i) {
172 if ((counter_config[i].enabled) && (CTR_IS_RESERVED(msrs, i))) {
173 reset_value[i] = counter_config[i].count;
174
175 CTR_WRITE(counter_config[i].count, msrs, i);
176
177 CTRL_READ(low, high, msrs, i);
178 CTRL_CLEAR_LO(low);
179 CTRL_CLEAR_HI(high);
180 CTRL_SET_ENABLE(low);
181 CTRL_SET_USR(low, counter_config[i].user);
182 CTRL_SET_KERN(low, counter_config[i].kernel);
183 CTRL_SET_UM(low, counter_config[i].unit_mask);
184 CTRL_SET_EVENT_LOW(low, counter_config[i].event);
185 CTRL_SET_EVENT_HIGH(high, counter_config[i].event);
186 CTRL_SET_HOST_ONLY(high, 0);
187 CTRL_SET_GUEST_ONLY(high, 0);
188
189 CTRL_WRITE(low, high, msrs, i);
190 } else {
191 reset_value[i] = 0;
192 }
193 }
194}
195
196#ifdef CONFIG_OPROFILE_IBS
197
198static inline int
199op_amd_handle_ibs(struct pt_regs * const regs,
200 struct op_msrs const * const msrs)
201{
202 unsigned int low, high;
203 struct ibs_fetch_sample ibs_fetch;
204 struct ibs_op_sample ibs_op;
205
206 if (!ibs_allowed)
207 return 1;
208
209 if (ibs_config.fetch_enabled) {
210 rdmsr(MSR_AMD64_IBSFETCHCTL, low, high);
211 if (high & IBS_FETCH_HIGH_VALID_BIT) {
212 ibs_fetch.ibs_fetch_ctl_high = high;
213 ibs_fetch.ibs_fetch_ctl_low = low;
214 rdmsr(MSR_AMD64_IBSFETCHLINAD, low, high);
215 ibs_fetch.ibs_fetch_lin_addr_high = high;
216 ibs_fetch.ibs_fetch_lin_addr_low = low;
217 rdmsr(MSR_AMD64_IBSFETCHPHYSAD, low, high);
218 ibs_fetch.ibs_fetch_phys_addr_high = high;
219 ibs_fetch.ibs_fetch_phys_addr_low = low;
220
221 oprofile_add_ibs_sample(regs,
222 (unsigned int *)&ibs_fetch,
223 IBS_FETCH_BEGIN);
224
225 /*reenable the IRQ */
226 rdmsr(MSR_AMD64_IBSFETCHCTL, low, high);
227 high &= ~IBS_FETCH_HIGH_VALID_BIT;
228 high |= IBS_FETCH_HIGH_ENABLE;
229 low &= IBS_FETCH_LOW_MAX_CNT_MASK;
230 wrmsr(MSR_AMD64_IBSFETCHCTL, low, high);
231 }
232 }
233
234 if (ibs_config.op_enabled) {
235 rdmsr(MSR_AMD64_IBSOPCTL, low, high);
236 if (low & IBS_OP_LOW_VALID_BIT) {
237 rdmsr(MSR_AMD64_IBSOPRIP, low, high);
238 ibs_op.ibs_op_rip_low = low;
239 ibs_op.ibs_op_rip_high = high;
240 rdmsr(MSR_AMD64_IBSOPDATA, low, high);
241 ibs_op.ibs_op_data1_low = low;
242 ibs_op.ibs_op_data1_high = high;
243 rdmsr(MSR_AMD64_IBSOPDATA2, low, high);
244 ibs_op.ibs_op_data2_low = low;
245 ibs_op.ibs_op_data2_high = high;
246 rdmsr(MSR_AMD64_IBSOPDATA3, low, high);
247 ibs_op.ibs_op_data3_low = low;
248 ibs_op.ibs_op_data3_high = high;
249 rdmsr(MSR_AMD64_IBSDCLINAD, low, high);
250 ibs_op.ibs_dc_linear_low = low;
251 ibs_op.ibs_dc_linear_high = high;
252 rdmsr(MSR_AMD64_IBSDCPHYSAD, low, high);
253 ibs_op.ibs_dc_phys_low = low;
254 ibs_op.ibs_dc_phys_high = high;
255
256 /* reenable the IRQ */
257 oprofile_add_ibs_sample(regs,
258 (unsigned int *)&ibs_op,
259 IBS_OP_BEGIN);
260 rdmsr(MSR_AMD64_IBSOPCTL, low, high);
261 high = 0;
262 low &= ~IBS_OP_LOW_VALID_BIT;
263 low |= IBS_OP_LOW_ENABLE;
264 wrmsr(MSR_AMD64_IBSOPCTL, low, high);
265 }
266 }
267
268 return 1;
269}
270
271#endif
272
273static int op_amd_check_ctrs(struct pt_regs * const regs,
274 struct op_msrs const * const msrs)
275{
276 unsigned int low, high;
277 int i;
278
279 for (i = 0 ; i < NUM_COUNTERS; ++i) {
280 if (!reset_value[i])
281 continue;
282 CTR_READ(low, high, msrs, i);
283 if (CTR_OVERFLOWED(low)) {
284 oprofile_add_sample(regs, i);
285 CTR_WRITE(reset_value[i], msrs, i);
286 }
287 }
288
289#ifdef CONFIG_OPROFILE_IBS
290 op_amd_handle_ibs(regs, msrs);
291#endif
292
293 /* See op_model_ppro.c */
294 return 1;
295}
296
297static void op_amd_start(struct op_msrs const * const msrs)
298{
299 unsigned int low, high;
300 int i;
301 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
302 if (reset_value[i]) {
303 CTRL_READ(low, high, msrs, i);
304 CTRL_SET_ACTIVE(low);
305 CTRL_WRITE(low, high, msrs, i);
306 }
307 }
308
309#ifdef CONFIG_OPROFILE_IBS
310 if (ibs_allowed && ibs_config.fetch_enabled) {
311 low = (ibs_config.max_cnt_fetch >> 4) & 0xFFFF;
312 high = IBS_FETCH_HIGH_ENABLE;
313 wrmsr(MSR_AMD64_IBSFETCHCTL, low, high);
314 }
315
316 if (ibs_allowed && ibs_config.op_enabled) {
317 low = ((ibs_config.max_cnt_op >> 4) & 0xFFFF) + IBS_OP_LOW_ENABLE;
318 high = 0;
319 wrmsr(MSR_AMD64_IBSOPCTL, low, high);
320 }
321#endif
322}
323
324
325static void op_amd_stop(struct op_msrs const * const msrs)
326{
327 unsigned int low, high;
328 int i;
329
330 /* Subtle: stop on all counters to avoid race with
331 * setting our pm callback */
332 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
333 if (!reset_value[i])
334 continue;
335 CTRL_READ(low, high, msrs, i);
336 CTRL_SET_INACTIVE(low);
337 CTRL_WRITE(low, high, msrs, i);
338 }
339
340#ifdef CONFIG_OPROFILE_IBS
341 if (ibs_allowed && ibs_config.fetch_enabled) {
342 low = 0; /* clear max count and enable */
343 high = 0;
344 wrmsr(MSR_AMD64_IBSFETCHCTL, low, high);
345 }
346
347 if (ibs_allowed && ibs_config.op_enabled) {
348 low = 0; /* clear max count and enable */
349 high = 0;
350 wrmsr(MSR_AMD64_IBSOPCTL, low, high);
351 }
352#endif
353}
354
355static void op_amd_shutdown(struct op_msrs const * const msrs)
356{
357 int i;
358
359 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
360 if (CTR_IS_RESERVED(msrs, i))
361 release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
362 }
363 for (i = 0 ; i < NUM_CONTROLS ; ++i) {
364 if (CTRL_IS_RESERVED(msrs, i))
365 release_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
366 }
367}
368
369#ifndef CONFIG_OPROFILE_IBS
370
371/* no IBS support */
372
373static int op_amd_init(struct oprofile_operations *ops)
374{
375 return 0;
376}
377
378static void op_amd_exit(void) {}
379
380#else
381
382static u8 ibs_eilvt_off;
383
384static inline void apic_init_ibs_nmi_per_cpu(void *arg)
385{
386 ibs_eilvt_off = setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_NMI, 0);
387}
388
389static inline void apic_clear_ibs_nmi_per_cpu(void *arg)
390{
391 setup_APIC_eilvt_ibs(0, APIC_EILVT_MSG_FIX, 1);
392}
393
394static int pfm_amd64_setup_eilvt(void)
395{
396#define IBSCTL_LVTOFFSETVAL (1 << 8)
397#define IBSCTL 0x1cc
398 struct pci_dev *cpu_cfg;
399 int nodes;
400 u32 value = 0;
401
402 /* per CPU setup */
403 on_each_cpu(apic_init_ibs_nmi_per_cpu, NULL, 1);
404
405 nodes = 0;
406 cpu_cfg = NULL;
407 do {
408 cpu_cfg = pci_get_device(PCI_VENDOR_ID_AMD,
409 PCI_DEVICE_ID_AMD_10H_NB_MISC,
410 cpu_cfg);
411 if (!cpu_cfg)
412 break;
413 ++nodes;
414 pci_write_config_dword(cpu_cfg, IBSCTL, ibs_eilvt_off
415 | IBSCTL_LVTOFFSETVAL);
416 pci_read_config_dword(cpu_cfg, IBSCTL, &value);
417 if (value != (ibs_eilvt_off | IBSCTL_LVTOFFSETVAL)) {
418 printk(KERN_DEBUG "Failed to setup IBS LVT offset, "
419 "IBSCTL = 0x%08x", value);
420 return 1;
421 }
422 } while (1);
423
424 if (!nodes) {
425 printk(KERN_DEBUG "No CPU node configured for IBS");
426 return 1;
427 }
428
429#ifdef CONFIG_NUMA
430 /* Sanity check */
431 /* Works only for 64bit with proper numa implementation. */
432 if (nodes != num_possible_nodes()) {
433 printk(KERN_DEBUG "Failed to setup CPU node(s) for IBS, "
434 "found: %d, expected %d",
435 nodes, num_possible_nodes());
436 return 1;
437 }
438#endif
439 return 0;
440}
441
442/*
443 * initialize the APIC for the IBS interrupts
444 * if available (AMD Family10h rev B0 and later)
445 */
446static void setup_ibs(void)
447{
448 ibs_allowed = boot_cpu_has(X86_FEATURE_IBS);
449
450 if (!ibs_allowed)
451 return;
452
453 if (pfm_amd64_setup_eilvt()) {
454 ibs_allowed = 0;
455 return;
456 }
457
458 printk(KERN_INFO "oprofile: AMD IBS detected\n");
459}
460
461
462/*
463 * unitialize the APIC for the IBS interrupts if needed on AMD Family10h
464 * rev B0 and later */
465static void clear_ibs_nmi(void)
466{
467 if (ibs_allowed)
468 on_each_cpu(apic_clear_ibs_nmi_per_cpu, NULL, 1);
469}
470
471static int (*create_arch_files)(struct super_block * sb, struct dentry * root);
472
473static int setup_ibs_files(struct super_block * sb, struct dentry * root)
474{
475 char buf[12];
476 struct dentry *dir;
477 int ret = 0;
478
479 /* architecture specific files */
480 if (create_arch_files)
481 ret = create_arch_files(sb, root);
482
483 if (ret)
484 return ret;
485
486 if (!ibs_allowed)
487 return ret;
488
489 /* model specific files */
490
491 /* setup some reasonable defaults */
492 ibs_config.max_cnt_fetch = 250000;
493 ibs_config.fetch_enabled = 0;
494 ibs_config.max_cnt_op = 250000;
495 ibs_config.op_enabled = 0;
496 ibs_config.dispatched_ops = 1;
497 snprintf(buf, sizeof(buf), "ibs_fetch");
498 dir = oprofilefs_mkdir(sb, root, buf);
499 oprofilefs_create_ulong(sb, dir, "rand_enable",
500 &ibs_config.rand_en);
501 oprofilefs_create_ulong(sb, dir, "enable",
502 &ibs_config.fetch_enabled);
503 oprofilefs_create_ulong(sb, dir, "max_count",
504 &ibs_config.max_cnt_fetch);
505 snprintf(buf, sizeof(buf), "ibs_uops");
506 dir = oprofilefs_mkdir(sb, root, buf);
507 oprofilefs_create_ulong(sb, dir, "enable",
508 &ibs_config.op_enabled);
509 oprofilefs_create_ulong(sb, dir, "max_count",
510 &ibs_config.max_cnt_op);
511 oprofilefs_create_ulong(sb, dir, "dispatched_ops",
512 &ibs_config.dispatched_ops);
513
514 return 0;
515}
516
517static int op_amd_init(struct oprofile_operations *ops)
518{
519 setup_ibs();
520 create_arch_files = ops->create_files;
521 ops->create_files = setup_ibs_files;
522 return 0;
523}
524
525static void op_amd_exit(void)
526{
527 clear_ibs_nmi();
528}
529
530#endif
531
532struct op_x86_model_spec const op_amd_spec = {
533 .init = op_amd_init,
534 .exit = op_amd_exit,
535 .num_counters = NUM_COUNTERS,
536 .num_controls = NUM_CONTROLS,
537 .fill_in_addresses = &op_amd_fill_in_addresses,
538 .setup_ctrs = &op_amd_setup_ctrs,
539 .check_ctrs = &op_amd_check_ctrs,
540 .start = &op_amd_start,
541 .stop = &op_amd_stop,
542 .shutdown = &op_amd_shutdown
543};
diff --git a/arch/x86/oprofile/op_model_athlon.c b/arch/x86/oprofile/op_model_athlon.c
deleted file mode 100644
index 3d534879a9dc..000000000000
--- a/arch/x86/oprofile/op_model_athlon.c
+++ /dev/null
@@ -1,190 +0,0 @@
1/*
2 * @file op_model_athlon.h
3 * athlon / K7 / K8 / Family 10h model-specific MSR operations
4 *
5 * @remark Copyright 2002 OProfile authors
6 * @remark Read the file COPYING
7 *
8 * @author John Levon
9 * @author Philippe Elie
10 * @author Graydon Hoare
11 */
12
13#include <linux/oprofile.h>
14#include <asm/ptrace.h>
15#include <asm/msr.h>
16#include <asm/nmi.h>
17
18#include "op_x86_model.h"
19#include "op_counter.h"
20
21#define NUM_COUNTERS 4
22#define NUM_CONTROLS 4
23
24#define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0)
25#define CTR_READ(l, h, msrs, c) do {rdmsr(msrs->counters[(c)].addr, (l), (h)); } while (0)
26#define CTR_WRITE(l, msrs, c) do {wrmsr(msrs->counters[(c)].addr, -(unsigned int)(l), -1); } while (0)
27#define CTR_OVERFLOWED(n) (!((n) & (1U<<31)))
28
29#define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0)
30#define CTRL_READ(l, h, msrs, c) do {rdmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
31#define CTRL_WRITE(l, h, msrs, c) do {wrmsr(msrs->controls[(c)].addr, (l), (h)); } while (0)
32#define CTRL_SET_ACTIVE(n) (n |= (1<<22))
33#define CTRL_SET_INACTIVE(n) (n &= ~(1<<22))
34#define CTRL_CLEAR_LO(x) (x &= (1<<21))
35#define CTRL_CLEAR_HI(x) (x &= 0xfffffcf0)
36#define CTRL_SET_ENABLE(val) (val |= 1<<20)
37#define CTRL_SET_USR(val, u) (val |= ((u & 1) << 16))
38#define CTRL_SET_KERN(val, k) (val |= ((k & 1) << 17))
39#define CTRL_SET_UM(val, m) (val |= (m << 8))
40#define CTRL_SET_EVENT_LOW(val, e) (val |= (e & 0xff))
41#define CTRL_SET_EVENT_HIGH(val, e) (val |= ((e >> 8) & 0xf))
42#define CTRL_SET_HOST_ONLY(val, h) (val |= ((h & 1) << 9))
43#define CTRL_SET_GUEST_ONLY(val, h) (val |= ((h & 1) << 8))
44
45static unsigned long reset_value[NUM_COUNTERS];
46
47static void athlon_fill_in_addresses(struct op_msrs * const msrs)
48{
49 int i;
50
51 for (i = 0; i < NUM_COUNTERS; i++) {
52 if (reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
53 msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
54 else
55 msrs->counters[i].addr = 0;
56 }
57
58 for (i = 0; i < NUM_CONTROLS; i++) {
59 if (reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i))
60 msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
61 else
62 msrs->controls[i].addr = 0;
63 }
64}
65
66
67static void athlon_setup_ctrs(struct op_msrs const * const msrs)
68{
69 unsigned int low, high;
70 int i;
71
72 /* clear all counters */
73 for (i = 0 ; i < NUM_CONTROLS; ++i) {
74 if (unlikely(!CTRL_IS_RESERVED(msrs, i)))
75 continue;
76 CTRL_READ(low, high, msrs, i);
77 CTRL_CLEAR_LO(low);
78 CTRL_CLEAR_HI(high);
79 CTRL_WRITE(low, high, msrs, i);
80 }
81
82 /* avoid a false detection of ctr overflows in NMI handler */
83 for (i = 0; i < NUM_COUNTERS; ++i) {
84 if (unlikely(!CTR_IS_RESERVED(msrs, i)))
85 continue;
86 CTR_WRITE(1, msrs, i);
87 }
88
89 /* enable active counters */
90 for (i = 0; i < NUM_COUNTERS; ++i) {
91 if ((counter_config[i].enabled) && (CTR_IS_RESERVED(msrs, i))) {
92 reset_value[i] = counter_config[i].count;
93
94 CTR_WRITE(counter_config[i].count, msrs, i);
95
96 CTRL_READ(low, high, msrs, i);
97 CTRL_CLEAR_LO(low);
98 CTRL_CLEAR_HI(high);
99 CTRL_SET_ENABLE(low);
100 CTRL_SET_USR(low, counter_config[i].user);
101 CTRL_SET_KERN(low, counter_config[i].kernel);
102 CTRL_SET_UM(low, counter_config[i].unit_mask);
103 CTRL_SET_EVENT_LOW(low, counter_config[i].event);
104 CTRL_SET_EVENT_HIGH(high, counter_config[i].event);
105 CTRL_SET_HOST_ONLY(high, 0);
106 CTRL_SET_GUEST_ONLY(high, 0);
107
108 CTRL_WRITE(low, high, msrs, i);
109 } else {
110 reset_value[i] = 0;
111 }
112 }
113}
114
115
116static int athlon_check_ctrs(struct pt_regs * const regs,
117 struct op_msrs const * const msrs)
118{
119 unsigned int low, high;
120 int i;
121
122 for (i = 0 ; i < NUM_COUNTERS; ++i) {
123 if (!reset_value[i])
124 continue;
125 CTR_READ(low, high, msrs, i);
126 if (CTR_OVERFLOWED(low)) {
127 oprofile_add_sample(regs, i);
128 CTR_WRITE(reset_value[i], msrs, i);
129 }
130 }
131
132 /* See op_model_ppro.c */
133 return 1;
134}
135
136
137static void athlon_start(struct op_msrs const * const msrs)
138{
139 unsigned int low, high;
140 int i;
141 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
142 if (reset_value[i]) {
143 CTRL_READ(low, high, msrs, i);
144 CTRL_SET_ACTIVE(low);
145 CTRL_WRITE(low, high, msrs, i);
146 }
147 }
148}
149
150
151static void athlon_stop(struct op_msrs const * const msrs)
152{
153 unsigned int low, high;
154 int i;
155
156 /* Subtle: stop on all counters to avoid race with
157 * setting our pm callback */
158 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
159 if (!reset_value[i])
160 continue;
161 CTRL_READ(low, high, msrs, i);
162 CTRL_SET_INACTIVE(low);
163 CTRL_WRITE(low, high, msrs, i);
164 }
165}
166
167static void athlon_shutdown(struct op_msrs const * const msrs)
168{
169 int i;
170
171 for (i = 0 ; i < NUM_COUNTERS ; ++i) {
172 if (CTR_IS_RESERVED(msrs, i))
173 release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
174 }
175 for (i = 0 ; i < NUM_CONTROLS ; ++i) {
176 if (CTRL_IS_RESERVED(msrs, i))
177 release_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
178 }
179}
180
181struct op_x86_model_spec const op_athlon_spec = {
182 .num_counters = NUM_COUNTERS,
183 .num_controls = NUM_CONTROLS,
184 .fill_in_addresses = &athlon_fill_in_addresses,
185 .setup_ctrs = &athlon_setup_ctrs,
186 .check_ctrs = &athlon_check_ctrs,
187 .start = &athlon_start,
188 .stop = &athlon_stop,
189 .shutdown = &athlon_shutdown
190};
diff --git a/arch/x86/oprofile/op_model_p4.c b/arch/x86/oprofile/op_model_p4.c
index 56b4757a1f47..43ac5af338d8 100644
--- a/arch/x86/oprofile/op_model_p4.c
+++ b/arch/x86/oprofile/op_model_p4.c
@@ -10,11 +10,12 @@
10 10
11#include <linux/oprofile.h> 11#include <linux/oprofile.h>
12#include <linux/smp.h> 12#include <linux/smp.h>
13#include <linux/ptrace.h>
14#include <linux/nmi.h>
13#include <asm/msr.h> 15#include <asm/msr.h>
14#include <asm/ptrace.h>
15#include <asm/fixmap.h> 16#include <asm/fixmap.h>
16#include <asm/apic.h> 17#include <asm/apic.h>
17#include <asm/nmi.h> 18
18 19
19#include "op_x86_model.h" 20#include "op_x86_model.h"
20#include "op_counter.h" 21#include "op_counter.h"
@@ -40,7 +41,7 @@ static unsigned int num_controls = NUM_CONTROLS_NON_HT;
40static inline void setup_num_counters(void) 41static inline void setup_num_counters(void)
41{ 42{
42#ifdef CONFIG_SMP 43#ifdef CONFIG_SMP
43 if (smp_num_siblings == 2){ 44 if (smp_num_siblings == 2) {
44 num_counters = NUM_COUNTERS_HT2; 45 num_counters = NUM_COUNTERS_HT2;
45 num_controls = NUM_CONTROLS_HT2; 46 num_controls = NUM_CONTROLS_HT2;
46 } 47 }
@@ -86,7 +87,7 @@ struct p4_event_binding {
86#define CTR_FLAME_2 (1 << 6) 87#define CTR_FLAME_2 (1 << 6)
87#define CTR_IQ_5 (1 << 7) 88#define CTR_IQ_5 (1 << 7)
88 89
89static struct p4_counter_binding p4_counters [NUM_COUNTERS_NON_HT] = { 90static struct p4_counter_binding p4_counters[NUM_COUNTERS_NON_HT] = {
90 { CTR_BPU_0, MSR_P4_BPU_PERFCTR0, MSR_P4_BPU_CCCR0 }, 91 { CTR_BPU_0, MSR_P4_BPU_PERFCTR0, MSR_P4_BPU_CCCR0 },
91 { CTR_MS_0, MSR_P4_MS_PERFCTR0, MSR_P4_MS_CCCR0 }, 92 { CTR_MS_0, MSR_P4_MS_PERFCTR0, MSR_P4_MS_CCCR0 },
92 { CTR_FLAME_0, MSR_P4_FLAME_PERFCTR0, MSR_P4_FLAME_CCCR0 }, 93 { CTR_FLAME_0, MSR_P4_FLAME_PERFCTR0, MSR_P4_FLAME_CCCR0 },
@@ -97,32 +98,32 @@ static struct p4_counter_binding p4_counters [NUM_COUNTERS_NON_HT] = {
97 { CTR_IQ_5, MSR_P4_IQ_PERFCTR5, MSR_P4_IQ_CCCR5 } 98 { CTR_IQ_5, MSR_P4_IQ_PERFCTR5, MSR_P4_IQ_CCCR5 }
98}; 99};
99 100
100#define NUM_UNUSED_CCCRS NUM_CCCRS_NON_HT - NUM_COUNTERS_NON_HT 101#define NUM_UNUSED_CCCRS (NUM_CCCRS_NON_HT - NUM_COUNTERS_NON_HT)
101 102
102/* p4 event codes in libop/op_event.h are indices into this table. */ 103/* p4 event codes in libop/op_event.h are indices into this table. */
103 104
104static struct p4_event_binding p4_events[NUM_EVENTS] = { 105static struct p4_event_binding p4_events[NUM_EVENTS] = {
105 106
106 { /* BRANCH_RETIRED */ 107 { /* BRANCH_RETIRED */
107 0x05, 0x06, 108 0x05, 0x06,
108 { {CTR_IQ_4, MSR_P4_CRU_ESCR2}, 109 { {CTR_IQ_4, MSR_P4_CRU_ESCR2},
109 {CTR_IQ_5, MSR_P4_CRU_ESCR3} } 110 {CTR_IQ_5, MSR_P4_CRU_ESCR3} }
110 }, 111 },
111 112
112 { /* MISPRED_BRANCH_RETIRED */ 113 { /* MISPRED_BRANCH_RETIRED */
113 0x04, 0x03, 114 0x04, 0x03,
114 { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, 115 { { CTR_IQ_4, MSR_P4_CRU_ESCR0},
115 { CTR_IQ_5, MSR_P4_CRU_ESCR1} } 116 { CTR_IQ_5, MSR_P4_CRU_ESCR1} }
116 }, 117 },
117 118
118 { /* TC_DELIVER_MODE */ 119 { /* TC_DELIVER_MODE */
119 0x01, 0x01, 120 0x01, 0x01,
120 { { CTR_MS_0, MSR_P4_TC_ESCR0}, 121 { { CTR_MS_0, MSR_P4_TC_ESCR0},
121 { CTR_MS_2, MSR_P4_TC_ESCR1} } 122 { CTR_MS_2, MSR_P4_TC_ESCR1} }
122 }, 123 },
123 124
124 { /* BPU_FETCH_REQUEST */ 125 { /* BPU_FETCH_REQUEST */
125 0x00, 0x03, 126 0x00, 0x03,
126 { { CTR_BPU_0, MSR_P4_BPU_ESCR0}, 127 { { CTR_BPU_0, MSR_P4_BPU_ESCR0},
127 { CTR_BPU_2, MSR_P4_BPU_ESCR1} } 128 { CTR_BPU_2, MSR_P4_BPU_ESCR1} }
128 }, 129 },
@@ -146,7 +147,7 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
146 }, 147 },
147 148
148 { /* LOAD_PORT_REPLAY */ 149 { /* LOAD_PORT_REPLAY */
149 0x02, 0x04, 150 0x02, 0x04,
150 { { CTR_FLAME_0, MSR_P4_SAAT_ESCR0}, 151 { { CTR_FLAME_0, MSR_P4_SAAT_ESCR0},
151 { CTR_FLAME_2, MSR_P4_SAAT_ESCR1} } 152 { CTR_FLAME_2, MSR_P4_SAAT_ESCR1} }
152 }, 153 },
@@ -170,43 +171,43 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
170 }, 171 },
171 172
172 { /* BSQ_CACHE_REFERENCE */ 173 { /* BSQ_CACHE_REFERENCE */
173 0x07, 0x0c, 174 0x07, 0x0c,
174 { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, 175 { { CTR_BPU_0, MSR_P4_BSU_ESCR0},
175 { CTR_BPU_2, MSR_P4_BSU_ESCR1} } 176 { CTR_BPU_2, MSR_P4_BSU_ESCR1} }
176 }, 177 },
177 178
178 { /* IOQ_ALLOCATION */ 179 { /* IOQ_ALLOCATION */
179 0x06, 0x03, 180 0x06, 0x03,
180 { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, 181 { { CTR_BPU_0, MSR_P4_FSB_ESCR0},
181 { 0, 0 } } 182 { 0, 0 } }
182 }, 183 },
183 184
184 { /* IOQ_ACTIVE_ENTRIES */ 185 { /* IOQ_ACTIVE_ENTRIES */
185 0x06, 0x1a, 186 0x06, 0x1a,
186 { { CTR_BPU_2, MSR_P4_FSB_ESCR1}, 187 { { CTR_BPU_2, MSR_P4_FSB_ESCR1},
187 { 0, 0 } } 188 { 0, 0 } }
188 }, 189 },
189 190
190 { /* FSB_DATA_ACTIVITY */ 191 { /* FSB_DATA_ACTIVITY */
191 0x06, 0x17, 192 0x06, 0x17,
192 { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, 193 { { CTR_BPU_0, MSR_P4_FSB_ESCR0},
193 { CTR_BPU_2, MSR_P4_FSB_ESCR1} } 194 { CTR_BPU_2, MSR_P4_FSB_ESCR1} }
194 }, 195 },
195 196
196 { /* BSQ_ALLOCATION */ 197 { /* BSQ_ALLOCATION */
197 0x07, 0x05, 198 0x07, 0x05,
198 { { CTR_BPU_0, MSR_P4_BSU_ESCR0}, 199 { { CTR_BPU_0, MSR_P4_BSU_ESCR0},
199 { 0, 0 } } 200 { 0, 0 } }
200 }, 201 },
201 202
202 { /* BSQ_ACTIVE_ENTRIES */ 203 { /* BSQ_ACTIVE_ENTRIES */
203 0x07, 0x06, 204 0x07, 0x06,
204 { { CTR_BPU_2, MSR_P4_BSU_ESCR1 /* guess */}, 205 { { CTR_BPU_2, MSR_P4_BSU_ESCR1 /* guess */},
205 { 0, 0 } } 206 { 0, 0 } }
206 }, 207 },
207 208
208 { /* X87_ASSIST */ 209 { /* X87_ASSIST */
209 0x05, 0x03, 210 0x05, 0x03,
210 { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, 211 { { CTR_IQ_4, MSR_P4_CRU_ESCR2},
211 { CTR_IQ_5, MSR_P4_CRU_ESCR3} } 212 { CTR_IQ_5, MSR_P4_CRU_ESCR3} }
212 }, 213 },
@@ -216,21 +217,21 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
216 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 217 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
217 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 218 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
218 }, 219 },
219 220
220 { /* PACKED_SP_UOP */ 221 { /* PACKED_SP_UOP */
221 0x01, 0x08, 222 0x01, 0x08,
222 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 223 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
223 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 224 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
224 }, 225 },
225 226
226 { /* PACKED_DP_UOP */ 227 { /* PACKED_DP_UOP */
227 0x01, 0x0c, 228 0x01, 0x0c,
228 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 229 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
229 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 230 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
230 }, 231 },
231 232
232 { /* SCALAR_SP_UOP */ 233 { /* SCALAR_SP_UOP */
233 0x01, 0x0a, 234 0x01, 0x0a,
234 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 235 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
235 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 236 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
236 }, 237 },
@@ -242,31 +243,31 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
242 }, 243 },
243 244
244 { /* 64BIT_MMX_UOP */ 245 { /* 64BIT_MMX_UOP */
245 0x01, 0x02, 246 0x01, 0x02,
246 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 247 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
247 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 248 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
248 }, 249 },
249 250
250 { /* 128BIT_MMX_UOP */ 251 { /* 128BIT_MMX_UOP */
251 0x01, 0x1a, 252 0x01, 0x1a,
252 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 253 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
253 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 254 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
254 }, 255 },
255 256
256 { /* X87_FP_UOP */ 257 { /* X87_FP_UOP */
257 0x01, 0x04, 258 0x01, 0x04,
258 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 259 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
259 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 260 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
260 }, 261 },
261 262
262 { /* X87_SIMD_MOVES_UOP */ 263 { /* X87_SIMD_MOVES_UOP */
263 0x01, 0x2e, 264 0x01, 0x2e,
264 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0}, 265 { { CTR_FLAME_0, MSR_P4_FIRM_ESCR0},
265 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} } 266 { CTR_FLAME_2, MSR_P4_FIRM_ESCR1} }
266 }, 267 },
267 268
268 { /* MACHINE_CLEAR */ 269 { /* MACHINE_CLEAR */
269 0x05, 0x02, 270 0x05, 0x02,
270 { { CTR_IQ_4, MSR_P4_CRU_ESCR2}, 271 { { CTR_IQ_4, MSR_P4_CRU_ESCR2},
271 { CTR_IQ_5, MSR_P4_CRU_ESCR3} } 272 { CTR_IQ_5, MSR_P4_CRU_ESCR3} }
272 }, 273 },
@@ -276,9 +277,9 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
276 { { CTR_BPU_0, MSR_P4_FSB_ESCR0}, 277 { { CTR_BPU_0, MSR_P4_FSB_ESCR0},
277 { CTR_BPU_2, MSR_P4_FSB_ESCR1} } 278 { CTR_BPU_2, MSR_P4_FSB_ESCR1} }
278 }, 279 },
279 280
280 { /* TC_MS_XFER */ 281 { /* TC_MS_XFER */
281 0x00, 0x05, 282 0x00, 0x05,
282 { { CTR_MS_0, MSR_P4_MS_ESCR0}, 283 { { CTR_MS_0, MSR_P4_MS_ESCR0},
283 { CTR_MS_2, MSR_P4_MS_ESCR1} } 284 { CTR_MS_2, MSR_P4_MS_ESCR1} }
284 }, 285 },
@@ -308,7 +309,7 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
308 }, 309 },
309 310
310 { /* INSTR_RETIRED */ 311 { /* INSTR_RETIRED */
311 0x04, 0x02, 312 0x04, 0x02,
312 { { CTR_IQ_4, MSR_P4_CRU_ESCR0}, 313 { { CTR_IQ_4, MSR_P4_CRU_ESCR0},
313 { CTR_IQ_5, MSR_P4_CRU_ESCR1} } 314 { CTR_IQ_5, MSR_P4_CRU_ESCR1} }
314 }, 315 },
@@ -319,14 +320,14 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
319 { CTR_IQ_5, MSR_P4_CRU_ESCR1} } 320 { CTR_IQ_5, MSR_P4_CRU_ESCR1} }
320 }, 321 },
321 322
322 { /* UOP_TYPE */ 323 { /* UOP_TYPE */
323 0x02, 0x02, 324 0x02, 0x02,
324 { { CTR_IQ_4, MSR_P4_RAT_ESCR0}, 325 { { CTR_IQ_4, MSR_P4_RAT_ESCR0},
325 { CTR_IQ_5, MSR_P4_RAT_ESCR1} } 326 { CTR_IQ_5, MSR_P4_RAT_ESCR1} }
326 }, 327 },
327 328
328 { /* RETIRED_MISPRED_BRANCH_TYPE */ 329 { /* RETIRED_MISPRED_BRANCH_TYPE */
329 0x02, 0x05, 330 0x02, 0x05,
330 { { CTR_MS_0, MSR_P4_TBPU_ESCR0}, 331 { { CTR_MS_0, MSR_P4_TBPU_ESCR0},
331 { CTR_MS_2, MSR_P4_TBPU_ESCR1} } 332 { CTR_MS_2, MSR_P4_TBPU_ESCR1} }
332 }, 333 },
@@ -349,8 +350,8 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
349#define ESCR_SET_OS_1(escr, os) ((escr) |= (((os) & 1) << 1)) 350#define ESCR_SET_OS_1(escr, os) ((escr) |= (((os) & 1) << 1))
350#define ESCR_SET_EVENT_SELECT(escr, sel) ((escr) |= (((sel) & 0x3f) << 25)) 351#define ESCR_SET_EVENT_SELECT(escr, sel) ((escr) |= (((sel) & 0x3f) << 25))
351#define ESCR_SET_EVENT_MASK(escr, mask) ((escr) |= (((mask) & 0xffff) << 9)) 352#define ESCR_SET_EVENT_MASK(escr, mask) ((escr) |= (((mask) & 0xffff) << 9))
352#define ESCR_READ(escr,high,ev,i) do {rdmsr(ev->bindings[(i)].escr_address, (escr), (high));} while (0) 353#define ESCR_READ(escr, high, ev, i) do {rdmsr(ev->bindings[(i)].escr_address, (escr), (high)); } while (0)
353#define ESCR_WRITE(escr,high,ev,i) do {wrmsr(ev->bindings[(i)].escr_address, (escr), (high));} while (0) 354#define ESCR_WRITE(escr, high, ev, i) do {wrmsr(ev->bindings[(i)].escr_address, (escr), (high)); } while (0)
354 355
355#define CCCR_RESERVED_BITS 0x38030FFF 356#define CCCR_RESERVED_BITS 0x38030FFF
356#define CCCR_CLEAR(cccr) ((cccr) &= CCCR_RESERVED_BITS) 357#define CCCR_CLEAR(cccr) ((cccr) &= CCCR_RESERVED_BITS)
@@ -360,15 +361,15 @@ static struct p4_event_binding p4_events[NUM_EVENTS] = {
360#define CCCR_SET_PMI_OVF_1(cccr) ((cccr) |= (1<<27)) 361#define CCCR_SET_PMI_OVF_1(cccr) ((cccr) |= (1<<27))
361#define CCCR_SET_ENABLE(cccr) ((cccr) |= (1<<12)) 362#define CCCR_SET_ENABLE(cccr) ((cccr) |= (1<<12))
362#define CCCR_SET_DISABLE(cccr) ((cccr) &= ~(1<<12)) 363#define CCCR_SET_DISABLE(cccr) ((cccr) &= ~(1<<12))
363#define CCCR_READ(low, high, i) do {rdmsr(p4_counters[(i)].cccr_address, (low), (high));} while (0) 364#define CCCR_READ(low, high, i) do {rdmsr(p4_counters[(i)].cccr_address, (low), (high)); } while (0)
364#define CCCR_WRITE(low, high, i) do {wrmsr(p4_counters[(i)].cccr_address, (low), (high));} while (0) 365#define CCCR_WRITE(low, high, i) do {wrmsr(p4_counters[(i)].cccr_address, (low), (high)); } while (0)
365#define CCCR_OVF_P(cccr) ((cccr) & (1U<<31)) 366#define CCCR_OVF_P(cccr) ((cccr) & (1U<<31))
366#define CCCR_CLEAR_OVF(cccr) ((cccr) &= (~(1U<<31))) 367#define CCCR_CLEAR_OVF(cccr) ((cccr) &= (~(1U<<31)))
367 368
368#define CTRL_IS_RESERVED(msrs,c) (msrs->controls[(c)].addr ? 1 : 0) 369#define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0)
369#define CTR_IS_RESERVED(msrs,c) (msrs->counters[(c)].addr ? 1 : 0) 370#define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0)
370#define CTR_READ(l,h,i) do {rdmsr(p4_counters[(i)].counter_address, (l), (h));} while (0) 371#define CTR_READ(l, h, i) do {rdmsr(p4_counters[(i)].counter_address, (l), (h)); } while (0)
371#define CTR_WRITE(l,i) do {wrmsr(p4_counters[(i)].counter_address, -(u32)(l), -1);} while (0) 372#define CTR_WRITE(l, i) do {wrmsr(p4_counters[(i)].counter_address, -(u32)(l), -1); } while (0)
372#define CTR_OVERFLOW_P(ctr) (!((ctr) & 0x80000000)) 373#define CTR_OVERFLOW_P(ctr) (!((ctr) & 0x80000000))
373 374
374 375
@@ -380,7 +381,7 @@ static unsigned int get_stagger(void)
380#ifdef CONFIG_SMP 381#ifdef CONFIG_SMP
381 int cpu = smp_processor_id(); 382 int cpu = smp_processor_id();
382 return (cpu != first_cpu(per_cpu(cpu_sibling_map, cpu))); 383 return (cpu != first_cpu(per_cpu(cpu_sibling_map, cpu)));
383#endif 384#endif
384 return 0; 385 return 0;
385} 386}
386 387
@@ -395,25 +396,23 @@ static unsigned long reset_value[NUM_COUNTERS_NON_HT];
395 396
396static void p4_fill_in_addresses(struct op_msrs * const msrs) 397static void p4_fill_in_addresses(struct op_msrs * const msrs)
397{ 398{
398 unsigned int i; 399 unsigned int i;
399 unsigned int addr, cccraddr, stag; 400 unsigned int addr, cccraddr, stag;
400 401
401 setup_num_counters(); 402 setup_num_counters();
402 stag = get_stagger(); 403 stag = get_stagger();
403 404
404 /* initialize some registers */ 405 /* initialize some registers */
405 for (i = 0; i < num_counters; ++i) { 406 for (i = 0; i < num_counters; ++i)
406 msrs->counters[i].addr = 0; 407 msrs->counters[i].addr = 0;
407 } 408 for (i = 0; i < num_controls; ++i)
408 for (i = 0; i < num_controls; ++i) {
409 msrs->controls[i].addr = 0; 409 msrs->controls[i].addr = 0;
410 } 410
411
412 /* the counter & cccr registers we pay attention to */ 411 /* the counter & cccr registers we pay attention to */
413 for (i = 0; i < num_counters; ++i) { 412 for (i = 0; i < num_counters; ++i) {
414 addr = p4_counters[VIRT_CTR(stag, i)].counter_address; 413 addr = p4_counters[VIRT_CTR(stag, i)].counter_address;
415 cccraddr = p4_counters[VIRT_CTR(stag, i)].cccr_address; 414 cccraddr = p4_counters[VIRT_CTR(stag, i)].cccr_address;
416 if (reserve_perfctr_nmi(addr)){ 415 if (reserve_perfctr_nmi(addr)) {
417 msrs->counters[i].addr = addr; 416 msrs->counters[i].addr = addr;
418 msrs->controls[i].addr = cccraddr; 417 msrs->controls[i].addr = cccraddr;
419 } 418 }
@@ -447,22 +446,22 @@ static void p4_fill_in_addresses(struct op_msrs * const msrs)
447 if (reserve_evntsel_nmi(addr)) 446 if (reserve_evntsel_nmi(addr))
448 msrs->controls[i].addr = addr; 447 msrs->controls[i].addr = addr;
449 } 448 }
450 449
451 for (addr = MSR_P4_MS_ESCR0 + stag; 450 for (addr = MSR_P4_MS_ESCR0 + stag;
452 addr <= MSR_P4_TC_ESCR1; ++i, addr += addr_increment()) { 451 addr <= MSR_P4_TC_ESCR1; ++i, addr += addr_increment()) {
453 if (reserve_evntsel_nmi(addr)) 452 if (reserve_evntsel_nmi(addr))
454 msrs->controls[i].addr = addr; 453 msrs->controls[i].addr = addr;
455 } 454 }
456 455
457 for (addr = MSR_P4_IX_ESCR0 + stag; 456 for (addr = MSR_P4_IX_ESCR0 + stag;
458 addr <= MSR_P4_CRU_ESCR3; ++i, addr += addr_increment()) { 457 addr <= MSR_P4_CRU_ESCR3; ++i, addr += addr_increment()) {
459 if (reserve_evntsel_nmi(addr)) 458 if (reserve_evntsel_nmi(addr))
460 msrs->controls[i].addr = addr; 459 msrs->controls[i].addr = addr;
461 } 460 }
462 461
463 /* there are 2 remaining non-contiguously located ESCRs */ 462 /* there are 2 remaining non-contiguously located ESCRs */
464 463
465 if (num_counters == NUM_COUNTERS_NON_HT) { 464 if (num_counters == NUM_COUNTERS_NON_HT) {
466 /* standard non-HT CPUs handle both remaining ESCRs*/ 465 /* standard non-HT CPUs handle both remaining ESCRs*/
467 if (reserve_evntsel_nmi(MSR_P4_CRU_ESCR5)) 466 if (reserve_evntsel_nmi(MSR_P4_CRU_ESCR5))
468 msrs->controls[i++].addr = MSR_P4_CRU_ESCR5; 467 msrs->controls[i++].addr = MSR_P4_CRU_ESCR5;
@@ -498,20 +497,20 @@ static void pmc_setup_one_p4_counter(unsigned int ctr)
498 unsigned int stag; 497 unsigned int stag;
499 498
500 stag = get_stagger(); 499 stag = get_stagger();
501 500
502 /* convert from counter *number* to counter *bit* */ 501 /* convert from counter *number* to counter *bit* */
503 counter_bit = 1 << VIRT_CTR(stag, ctr); 502 counter_bit = 1 << VIRT_CTR(stag, ctr);
504 503
505 /* find our event binding structure. */ 504 /* find our event binding structure. */
506 if (counter_config[ctr].event <= 0 || counter_config[ctr].event > NUM_EVENTS) { 505 if (counter_config[ctr].event <= 0 || counter_config[ctr].event > NUM_EVENTS) {
507 printk(KERN_ERR 506 printk(KERN_ERR
508 "oprofile: P4 event code 0x%lx out of range\n", 507 "oprofile: P4 event code 0x%lx out of range\n",
509 counter_config[ctr].event); 508 counter_config[ctr].event);
510 return; 509 return;
511 } 510 }
512 511
513 ev = &(p4_events[counter_config[ctr].event - 1]); 512 ev = &(p4_events[counter_config[ctr].event - 1]);
514 513
515 for (i = 0; i < maxbind; i++) { 514 for (i = 0; i < maxbind; i++) {
516 if (ev->bindings[i].virt_counter & counter_bit) { 515 if (ev->bindings[i].virt_counter & counter_bit) {
517 516
@@ -526,25 +525,24 @@ static void pmc_setup_one_p4_counter(unsigned int ctr)
526 ESCR_SET_OS_1(escr, counter_config[ctr].kernel); 525 ESCR_SET_OS_1(escr, counter_config[ctr].kernel);
527 } 526 }
528 ESCR_SET_EVENT_SELECT(escr, ev->event_select); 527 ESCR_SET_EVENT_SELECT(escr, ev->event_select);
529 ESCR_SET_EVENT_MASK(escr, counter_config[ctr].unit_mask); 528 ESCR_SET_EVENT_MASK(escr, counter_config[ctr].unit_mask);
530 ESCR_WRITE(escr, high, ev, i); 529 ESCR_WRITE(escr, high, ev, i);
531 530
532 /* modify CCCR */ 531 /* modify CCCR */
533 CCCR_READ(cccr, high, VIRT_CTR(stag, ctr)); 532 CCCR_READ(cccr, high, VIRT_CTR(stag, ctr));
534 CCCR_CLEAR(cccr); 533 CCCR_CLEAR(cccr);
535 CCCR_SET_REQUIRED_BITS(cccr); 534 CCCR_SET_REQUIRED_BITS(cccr);
536 CCCR_SET_ESCR_SELECT(cccr, ev->escr_select); 535 CCCR_SET_ESCR_SELECT(cccr, ev->escr_select);
537 if (stag == 0) { 536 if (stag == 0)
538 CCCR_SET_PMI_OVF_0(cccr); 537 CCCR_SET_PMI_OVF_0(cccr);
539 } else { 538 else
540 CCCR_SET_PMI_OVF_1(cccr); 539 CCCR_SET_PMI_OVF_1(cccr);
541 }
542 CCCR_WRITE(cccr, high, VIRT_CTR(stag, ctr)); 540 CCCR_WRITE(cccr, high, VIRT_CTR(stag, ctr));
543 return; 541 return;
544 } 542 }
545 } 543 }
546 544
547 printk(KERN_ERR 545 printk(KERN_ERR
548 "oprofile: P4 event code 0x%lx no binding, stag %d ctr %d\n", 546 "oprofile: P4 event code 0x%lx no binding, stag %d ctr %d\n",
549 counter_config[ctr].event, stag, ctr); 547 counter_config[ctr].event, stag, ctr);
550} 548}
@@ -559,14 +557,14 @@ static void p4_setup_ctrs(struct op_msrs const * const msrs)
559 stag = get_stagger(); 557 stag = get_stagger();
560 558
561 rdmsr(MSR_IA32_MISC_ENABLE, low, high); 559 rdmsr(MSR_IA32_MISC_ENABLE, low, high);
562 if (! MISC_PMC_ENABLED_P(low)) { 560 if (!MISC_PMC_ENABLED_P(low)) {
563 printk(KERN_ERR "oprofile: P4 PMC not available\n"); 561 printk(KERN_ERR "oprofile: P4 PMC not available\n");
564 return; 562 return;
565 } 563 }
566 564
567 /* clear the cccrs we will use */ 565 /* clear the cccrs we will use */
568 for (i = 0 ; i < num_counters ; i++) { 566 for (i = 0 ; i < num_counters ; i++) {
569 if (unlikely(!CTRL_IS_RESERVED(msrs,i))) 567 if (unlikely(!CTRL_IS_RESERVED(msrs, i)))
570 continue; 568 continue;
571 rdmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high); 569 rdmsr(p4_counters[VIRT_CTR(stag, i)].cccr_address, low, high);
572 CCCR_CLEAR(low); 570 CCCR_CLEAR(low);
@@ -576,14 +574,14 @@ static void p4_setup_ctrs(struct op_msrs const * const msrs)
576 574
577 /* clear all escrs (including those outside our concern) */ 575 /* clear all escrs (including those outside our concern) */
578 for (i = num_counters; i < num_controls; i++) { 576 for (i = num_counters; i < num_controls; i++) {
579 if (unlikely(!CTRL_IS_RESERVED(msrs,i))) 577 if (unlikely(!CTRL_IS_RESERVED(msrs, i)))
580 continue; 578 continue;
581 wrmsr(msrs->controls[i].addr, 0, 0); 579 wrmsr(msrs->controls[i].addr, 0, 0);
582 } 580 }
583 581
584 /* setup all counters */ 582 /* setup all counters */
585 for (i = 0 ; i < num_counters ; ++i) { 583 for (i = 0 ; i < num_counters ; ++i) {
586 if ((counter_config[i].enabled) && (CTRL_IS_RESERVED(msrs,i))) { 584 if ((counter_config[i].enabled) && (CTRL_IS_RESERVED(msrs, i))) {
587 reset_value[i] = counter_config[i].count; 585 reset_value[i] = counter_config[i].count;
588 pmc_setup_one_p4_counter(i); 586 pmc_setup_one_p4_counter(i);
589 CTR_WRITE(counter_config[i].count, VIRT_CTR(stag, i)); 587 CTR_WRITE(counter_config[i].count, VIRT_CTR(stag, i));
@@ -603,11 +601,11 @@ static int p4_check_ctrs(struct pt_regs * const regs,
603 stag = get_stagger(); 601 stag = get_stagger();
604 602
605 for (i = 0; i < num_counters; ++i) { 603 for (i = 0; i < num_counters; ++i) {
606 604
607 if (!reset_value[i]) 605 if (!reset_value[i])
608 continue; 606 continue;
609 607
610 /* 608 /*
611 * there is some eccentricity in the hardware which 609 * there is some eccentricity in the hardware which
612 * requires that we perform 2 extra corrections: 610 * requires that we perform 2 extra corrections:
613 * 611 *
@@ -616,24 +614,24 @@ static int p4_check_ctrs(struct pt_regs * const regs,
616 * 614 *
617 * - write the counter back twice to ensure it gets 615 * - write the counter back twice to ensure it gets
618 * updated properly. 616 * updated properly.
619 * 617 *
620 * the former seems to be related to extra NMIs happening 618 * the former seems to be related to extra NMIs happening
621 * during the current NMI; the latter is reported as errata 619 * during the current NMI; the latter is reported as errata
622 * N15 in intel doc 249199-029, pentium 4 specification 620 * N15 in intel doc 249199-029, pentium 4 specification
623 * update, though their suggested work-around does not 621 * update, though their suggested work-around does not
624 * appear to solve the problem. 622 * appear to solve the problem.
625 */ 623 */
626 624
627 real = VIRT_CTR(stag, i); 625 real = VIRT_CTR(stag, i);
628 626
629 CCCR_READ(low, high, real); 627 CCCR_READ(low, high, real);
630 CTR_READ(ctr, high, real); 628 CTR_READ(ctr, high, real);
631 if (CCCR_OVF_P(low) || CTR_OVERFLOW_P(ctr)) { 629 if (CCCR_OVF_P(low) || CTR_OVERFLOW_P(ctr)) {
632 oprofile_add_sample(regs, i); 630 oprofile_add_sample(regs, i);
633 CTR_WRITE(reset_value[i], real); 631 CTR_WRITE(reset_value[i], real);
634 CCCR_CLEAR_OVF(low); 632 CCCR_CLEAR_OVF(low);
635 CCCR_WRITE(low, high, real); 633 CCCR_WRITE(low, high, real);
636 CTR_WRITE(reset_value[i], real); 634 CTR_WRITE(reset_value[i], real);
637 } 635 }
638 } 636 }
639 637
@@ -683,15 +681,16 @@ static void p4_shutdown(struct op_msrs const * const msrs)
683 int i; 681 int i;
684 682
685 for (i = 0 ; i < num_counters ; ++i) { 683 for (i = 0 ; i < num_counters ; ++i) {
686 if (CTR_IS_RESERVED(msrs,i)) 684 if (CTR_IS_RESERVED(msrs, i))
687 release_perfctr_nmi(msrs->counters[i].addr); 685 release_perfctr_nmi(msrs->counters[i].addr);
688 } 686 }
689 /* some of the control registers are specially reserved in 687 /*
688 * some of the control registers are specially reserved in
690 * conjunction with the counter registers (hence the starting offset). 689 * conjunction with the counter registers (hence the starting offset).
691 * This saves a few bits. 690 * This saves a few bits.
692 */ 691 */
693 for (i = num_counters ; i < num_controls ; ++i) { 692 for (i = num_counters ; i < num_controls ; ++i) {
694 if (CTRL_IS_RESERVED(msrs,i)) 693 if (CTRL_IS_RESERVED(msrs, i))
695 release_evntsel_nmi(msrs->controls[i].addr); 694 release_evntsel_nmi(msrs->controls[i].addr);
696 } 695 }
697} 696}
diff --git a/arch/x86/oprofile/op_x86_model.h b/arch/x86/oprofile/op_x86_model.h
index 45b605fa71d0..05a0261ba0c3 100644
--- a/arch/x86/oprofile/op_x86_model.h
+++ b/arch/x86/oprofile/op_x86_model.h
@@ -32,6 +32,8 @@ struct pt_regs;
32 * various x86 CPU models' perfctr support. 32 * various x86 CPU models' perfctr support.
33 */ 33 */
34struct op_x86_model_spec { 34struct op_x86_model_spec {
35 int (*init)(struct oprofile_operations *ops);
36 void (*exit)(void);
35 unsigned int const num_counters; 37 unsigned int const num_counters;
36 unsigned int const num_controls; 38 unsigned int const num_controls;
37 void (*fill_in_addresses)(struct op_msrs * const msrs); 39 void (*fill_in_addresses)(struct op_msrs * const msrs);
@@ -46,6 +48,6 @@ struct op_x86_model_spec {
46extern struct op_x86_model_spec const op_ppro_spec; 48extern struct op_x86_model_spec const op_ppro_spec;
47extern struct op_x86_model_spec const op_p4_spec; 49extern struct op_x86_model_spec const op_p4_spec;
48extern struct op_x86_model_spec const op_p4_ht2_spec; 50extern struct op_x86_model_spec const op_p4_ht2_spec;
49extern struct op_x86_model_spec const op_athlon_spec; 51extern struct op_x86_model_spec const op_amd_spec;
50 52
51#endif /* OP_X86_MODEL_H */ 53#endif /* OP_X86_MODEL_H */
diff --git a/arch/x86/pci/acpi.c b/arch/x86/pci/acpi.c
index 19af06927fbc..1d88d2b39771 100644
--- a/arch/x86/pci/acpi.c
+++ b/arch/x86/pci/acpi.c
@@ -250,10 +250,5 @@ int __init pci_acpi_init(void)
250 acpi_pci_irq_enable(dev); 250 acpi_pci_irq_enable(dev);
251 } 251 }
252 252
253#ifdef CONFIG_X86_IO_APIC
254 if (acpi_ioapic)
255 print_IO_APIC();
256#endif
257
258 return 0; 253 return 0;
259} 254}
diff --git a/arch/x86/pci/amd_bus.c b/arch/x86/pci/amd_bus.c
index 6a0fca78c362..22e057665e55 100644
--- a/arch/x86/pci/amd_bus.c
+++ b/arch/x86/pci/amd_bus.c
@@ -580,7 +580,7 @@ static int __cpuinit amd_cpu_notify(struct notifier_block *self,
580 unsigned long action, void *hcpu) 580 unsigned long action, void *hcpu)
581{ 581{
582 int cpu = (long)hcpu; 582 int cpu = (long)hcpu;
583 switch(action) { 583 switch (action) {
584 case CPU_ONLINE: 584 case CPU_ONLINE:
585 case CPU_ONLINE_FROZEN: 585 case CPU_ONLINE_FROZEN:
586 smp_call_function_single(cpu, enable_pci_io_ecs, NULL, 0); 586 smp_call_function_single(cpu, enable_pci_io_ecs, NULL, 0);
diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
index 4bdaa590375d..3c27a809393b 100644
--- a/arch/x86/pci/fixup.c
+++ b/arch/x86/pci/fixup.c
@@ -511,3 +511,31 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1201, fam10h_pci_cfg_space_size);
511DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1202, fam10h_pci_cfg_space_size); 511DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1202, fam10h_pci_cfg_space_size);
512DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1203, fam10h_pci_cfg_space_size); 512DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1203, fam10h_pci_cfg_space_size);
513DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1204, fam10h_pci_cfg_space_size); 513DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_AMD, 0x1204, fam10h_pci_cfg_space_size);
514
515/*
516 * SB600: Disable BAR1 on device 14.0 to avoid HPET resources from
517 * confusing the PCI engine:
518 */
519static void sb600_disable_hpet_bar(struct pci_dev *dev)
520{
521 u8 val;
522
523 /*
524 * The SB600 and SB700 both share the same device
525 * ID, but the PM register 0x55 does something different
526 * for the SB700, so make sure we are dealing with the
527 * SB600 before touching the bit:
528 */
529
530 pci_read_config_byte(dev, 0x08, &val);
531
532 if (val < 0x2F) {
533 outb(0x55, 0xCD6);
534 val = inb(0xCD7);
535
536 /* Set bit 7 in PM register 0x55 */
537 outb(0x55, 0xCD6);
538 outb(val | 0x80, 0xCD7);
539 }
540}
541DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_ATI, 0x4385, sb600_disable_hpet_bar);
diff --git a/arch/x86/pci/i386.c b/arch/x86/pci/i386.c
index 8791fc55e715..844df0cbbd3e 100644
--- a/arch/x86/pci/i386.c
+++ b/arch/x86/pci/i386.c
@@ -33,6 +33,7 @@
33#include <linux/bootmem.h> 33#include <linux/bootmem.h>
34 34
35#include <asm/pat.h> 35#include <asm/pat.h>
36#include <asm/e820.h>
36 37
37#include "pci.h" 38#include "pci.h"
38 39
@@ -227,6 +228,8 @@ void __init pcibios_resource_survey(void)
227 pcibios_allocate_bus_resources(&pci_root_buses); 228 pcibios_allocate_bus_resources(&pci_root_buses);
228 pcibios_allocate_resources(0); 229 pcibios_allocate_resources(0);
229 pcibios_allocate_resources(1); 230 pcibios_allocate_resources(1);
231
232 e820_reserve_resources_late();
230} 233}
231 234
232/** 235/**
diff --git a/arch/x86/pci/irq.c b/arch/x86/pci/irq.c
index 8e077185e185..006599db0dc7 100644
--- a/arch/x86/pci/irq.c
+++ b/arch/x86/pci/irq.c
@@ -1043,35 +1043,44 @@ static void __init pcibios_fixup_irqs(void)
1043 if (io_apic_assign_pci_irqs) { 1043 if (io_apic_assign_pci_irqs) {
1044 int irq; 1044 int irq;
1045 1045
1046 if (pin) { 1046 if (!pin)
1047 /* 1047 continue;
1048 * interrupt pins are numbered starting 1048
1049 * from 1 1049 /*
1050 */ 1050 * interrupt pins are numbered starting from 1
1051 pin--; 1051 */
1052 irq = IO_APIC_get_PCI_irq_vector(dev->bus->number, 1052 pin--;
1053 PCI_SLOT(dev->devfn), pin); 1053 irq = IO_APIC_get_PCI_irq_vector(dev->bus->number,
1054 /* 1054 PCI_SLOT(dev->devfn), pin);
1055 * Busses behind bridges are typically not listed in the MP-table. 1055 /*
1056 * In this case we have to look up the IRQ based on the parent bus, 1056 * Busses behind bridges are typically not listed in the
1057 * parent slot, and pin number. The SMP code detects such bridged 1057 * MP-table. In this case we have to look up the IRQ
1058 * busses itself so we should get into this branch reliably. 1058 * based on the parent bus, parent slot, and pin number.
1059 */ 1059 * The SMP code detects such bridged busses itself so we
1060 if (irq < 0 && dev->bus->parent) { /* go back to the bridge */ 1060 * should get into this branch reliably.
1061 struct pci_dev *bridge = dev->bus->self; 1061 */
1062 1062 if (irq < 0 && dev->bus->parent) {
1063 pin = (pin + PCI_SLOT(dev->devfn)) % 4; 1063 /* go back to the bridge */
1064 irq = IO_APIC_get_PCI_irq_vector(bridge->bus->number, 1064 struct pci_dev *bridge = dev->bus->self;
1065 PCI_SLOT(bridge->devfn), pin); 1065 int bus;
1066 if (irq >= 0) 1066
1067 dev_warn(&dev->dev, "using bridge %s INT %c to get IRQ %d\n", 1067 pin = (pin + PCI_SLOT(dev->devfn)) % 4;
1068 pci_name(bridge), 1068 bus = bridge->bus->number;
1069 'A' + pin, irq); 1069 irq = IO_APIC_get_PCI_irq_vector(bus,
1070 } 1070 PCI_SLOT(bridge->devfn), pin);
1071 if (irq >= 0) { 1071 if (irq >= 0)
1072 dev_info(&dev->dev, "PCI->APIC IRQ transform: INT %c -> IRQ %d\n", 'A' + pin, irq); 1072 dev_warn(&dev->dev,
1073 dev->irq = irq; 1073 "using bridge %s INT %c to "
1074 } 1074 "get IRQ %d\n",
1075 pci_name(bridge),
1076 'A' + pin, irq);
1077 }
1078 if (irq >= 0) {
1079 dev_info(&dev->dev,
1080 "PCI->APIC IRQ transform: INT %c "
1081 "-> IRQ %d\n",
1082 'A' + pin, irq);
1083 dev->irq = irq;
1075 } 1084 }
1076 } 1085 }
1077#endif 1086#endif
diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
index d9635764ce3d..654a2234f8f3 100644
--- a/arch/x86/pci/mmconfig-shared.c
+++ b/arch/x86/pci/mmconfig-shared.c
@@ -209,7 +209,7 @@ static int __init pci_mmcfg_check_hostbridge(void)
209 return name != NULL; 209 return name != NULL;
210} 210}
211 211
212static void __init pci_mmcfg_insert_resources(unsigned long resource_flags) 212static void __init pci_mmcfg_insert_resources(void)
213{ 213{
214#define PCI_MMCFG_RESOURCE_NAME_LEN 19 214#define PCI_MMCFG_RESOURCE_NAME_LEN 19
215 int i; 215 int i;
@@ -233,7 +233,7 @@ static void __init pci_mmcfg_insert_resources(unsigned long resource_flags)
233 cfg->pci_segment); 233 cfg->pci_segment);
234 res->start = cfg->address; 234 res->start = cfg->address;
235 res->end = res->start + (num_buses << 20) - 1; 235 res->end = res->start + (num_buses << 20) - 1;
236 res->flags = IORESOURCE_MEM | resource_flags; 236 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
237 insert_resource(&iomem_resource, res); 237 insert_resource(&iomem_resource, res);
238 names += PCI_MMCFG_RESOURCE_NAME_LEN; 238 names += PCI_MMCFG_RESOURCE_NAME_LEN;
239 } 239 }
@@ -434,11 +434,9 @@ static void __init __pci_mmcfg_init(int early)
434 (pci_mmcfg_config[0].address == 0)) 434 (pci_mmcfg_config[0].address == 0))
435 return; 435 return;
436 436
437 if (pci_mmcfg_arch_init()) { 437 if (pci_mmcfg_arch_init())
438 if (known_bridge)
439 pci_mmcfg_insert_resources(IORESOURCE_BUSY);
440 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF; 438 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
441 } else { 439 else {
442 /* 440 /*
443 * Signal not to attempt to insert mmcfg resources because 441 * Signal not to attempt to insert mmcfg resources because
444 * the architecture mmcfg setup could not initialize. 442 * the architecture mmcfg setup could not initialize.
@@ -475,7 +473,7 @@ static int __init pci_mmcfg_late_insert_resources(void)
475 * marked so it won't cause request errors when __request_region is 473 * marked so it won't cause request errors when __request_region is
476 * called. 474 * called.
477 */ 475 */
478 pci_mmcfg_insert_resources(0); 476 pci_mmcfg_insert_resources();
479 477
480 return 0; 478 return 0;
481} 479}
diff --git a/arch/x86/power/cpu_32.c b/arch/x86/power/cpu_32.c
index d3e083dea720..274d06082f48 100644
--- a/arch/x86/power/cpu_32.c
+++ b/arch/x86/power/cpu_32.c
@@ -11,6 +11,7 @@
11#include <linux/suspend.h> 11#include <linux/suspend.h>
12#include <asm/mtrr.h> 12#include <asm/mtrr.h>
13#include <asm/mce.h> 13#include <asm/mce.h>
14#include <asm/xcr.h>
14 15
15static struct saved_context saved_context; 16static struct saved_context saved_context;
16 17
@@ -126,6 +127,12 @@ static void __restore_processor_state(struct saved_context *ctxt)
126 if (boot_cpu_has(X86_FEATURE_SEP)) 127 if (boot_cpu_has(X86_FEATURE_SEP))
127 enable_sep_cpu(); 128 enable_sep_cpu();
128 129
130 /*
131 * restore XCR0 for xsave capable cpu's.
132 */
133 if (cpu_has_xsave)
134 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
135
129 fix_processor_context(); 136 fix_processor_context();
130 do_fpu_end(); 137 do_fpu_end();
131 mtrr_ap_init(); 138 mtrr_ap_init();
diff --git a/arch/x86/power/cpu_64.c b/arch/x86/power/cpu_64.c
index 66bdfb591fd8..e3b6cf70d62c 100644
--- a/arch/x86/power/cpu_64.c
+++ b/arch/x86/power/cpu_64.c
@@ -14,6 +14,7 @@
14#include <asm/page.h> 14#include <asm/page.h>
15#include <asm/pgtable.h> 15#include <asm/pgtable.h>
16#include <asm/mtrr.h> 16#include <asm/mtrr.h>
17#include <asm/xcr.h>
17 18
18static void fix_processor_context(void); 19static void fix_processor_context(void);
19 20
@@ -122,6 +123,12 @@ static void __restore_processor_state(struct saved_context *ctxt)
122 wrmsrl(MSR_GS_BASE, ctxt->gs_base); 123 wrmsrl(MSR_GS_BASE, ctxt->gs_base);
123 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base); 124 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
124 125
126 /*
127 * restore XCR0 for xsave capable cpu's.
128 */
129 if (cpu_has_xsave)
130 xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
131
125 fix_processor_context(); 132 fix_processor_context();
126 133
127 do_fpu_end(); 134 do_fpu_end();
diff --git a/arch/x86/power/hibernate_asm_32.S b/arch/x86/power/hibernate_asm_32.S
index 4fc7e872c85e..d1e9b53f9d33 100644
--- a/arch/x86/power/hibernate_asm_32.S
+++ b/arch/x86/power/hibernate_asm_32.S
@@ -1,5 +1,3 @@
1.text
2
3/* 1/*
4 * This may not use any stack, nor any variable that is not "NoSave": 2 * This may not use any stack, nor any variable that is not "NoSave":
5 * 3 *
@@ -12,17 +10,18 @@
12#include <asm/segment.h> 10#include <asm/segment.h>
13#include <asm/page.h> 11#include <asm/page.h>
14#include <asm/asm-offsets.h> 12#include <asm/asm-offsets.h>
13#include <asm/processor-flags.h>
15 14
16 .text 15.text
17 16
18ENTRY(swsusp_arch_suspend) 17ENTRY(swsusp_arch_suspend)
19
20 movl %esp, saved_context_esp 18 movl %esp, saved_context_esp
21 movl %ebx, saved_context_ebx 19 movl %ebx, saved_context_ebx
22 movl %ebp, saved_context_ebp 20 movl %ebp, saved_context_ebp
23 movl %esi, saved_context_esi 21 movl %esi, saved_context_esi
24 movl %edi, saved_context_edi 22 movl %edi, saved_context_edi
25 pushfl ; popl saved_context_eflags 23 pushfl
24 popl saved_context_eflags
26 25
27 call swsusp_save 26 call swsusp_save
28 ret 27 ret
@@ -59,7 +58,7 @@ done:
59 movl mmu_cr4_features, %ecx 58 movl mmu_cr4_features, %ecx
60 jecxz 1f # cr4 Pentium and higher, skip if zero 59 jecxz 1f # cr4 Pentium and higher, skip if zero
61 movl %ecx, %edx 60 movl %ecx, %edx
62 andl $~(1<<7), %edx; # PGE 61 andl $~(X86_CR4_PGE), %edx
63 movl %edx, %cr4; # turn off PGE 62 movl %edx, %cr4; # turn off PGE
641: 631:
65 movl %cr3, %eax; # flush TLB 64 movl %cr3, %eax; # flush TLB
@@ -74,7 +73,8 @@ done:
74 movl saved_context_esi, %esi 73 movl saved_context_esi, %esi
75 movl saved_context_edi, %edi 74 movl saved_context_edi, %edi
76 75
77 pushl saved_context_eflags ; popfl 76 pushl saved_context_eflags
77 popfl
78 78
79 xorl %eax, %eax 79 xorl %eax, %eax
80 80
diff --git a/arch/x86/xen/Kconfig b/arch/x86/xen/Kconfig
index 3815e425f470..87b9ab166423 100644
--- a/arch/x86/xen/Kconfig
+++ b/arch/x86/xen/Kconfig
@@ -26,5 +26,13 @@ config XEN_MAX_DOMAIN_MEMORY
26 26
27config XEN_SAVE_RESTORE 27config XEN_SAVE_RESTORE
28 bool 28 bool
29 depends on PM 29 depends on XEN && PM
30 default y \ No newline at end of file 30 default y
31
32config XEN_DEBUG_FS
33 bool "Enable Xen debug and tuning parameters in debugfs"
34 depends on XEN && DEBUG_FS
35 default n
36 help
37 Enable statistics output and various tuning options in debugfs.
38 Enabling this option may incur a significant performance overhead.
diff --git a/arch/x86/xen/Makefile b/arch/x86/xen/Makefile
index 59c1e539aed2..313947940a1a 100644
--- a/arch/x86/xen/Makefile
+++ b/arch/x86/xen/Makefile
@@ -1,4 +1,12 @@
1obj-y := enlighten.o setup.o multicalls.o mmu.o \ 1ifdef CONFIG_FTRACE
2# Do not profile debug and lowlevel utilities
3CFLAGS_REMOVE_spinlock.o = -pg
4CFLAGS_REMOVE_time.o = -pg
5CFLAGS_REMOVE_irq.o = -pg
6endif
7
8obj-y := enlighten.o setup.o multicalls.o mmu.o irq.o \
2 time.o xen-asm_$(BITS).o grant-table.o suspend.o 9 time.o xen-asm_$(BITS).o grant-table.o suspend.o
3 10
4obj-$(CONFIG_SMP) += smp.o 11obj-$(CONFIG_SMP) += smp.o spinlock.o
12obj-$(CONFIG_XEN_DEBUG_FS) += debugfs.o \ No newline at end of file
diff --git a/arch/x86/xen/debugfs.c b/arch/x86/xen/debugfs.c
new file mode 100644
index 000000000000..b53225d2cac3
--- /dev/null
+++ b/arch/x86/xen/debugfs.c
@@ -0,0 +1,123 @@
1#include <linux/init.h>
2#include <linux/debugfs.h>
3#include <linux/module.h>
4
5#include "debugfs.h"
6
7static struct dentry *d_xen_debug;
8
9struct dentry * __init xen_init_debugfs(void)
10{
11 if (!d_xen_debug) {
12 d_xen_debug = debugfs_create_dir("xen", NULL);
13
14 if (!d_xen_debug)
15 pr_warning("Could not create 'xen' debugfs directory\n");
16 }
17
18 return d_xen_debug;
19}
20
21struct array_data
22{
23 void *array;
24 unsigned elements;
25};
26
27static int u32_array_open(struct inode *inode, struct file *file)
28{
29 file->private_data = NULL;
30 return nonseekable_open(inode, file);
31}
32
33static size_t format_array(char *buf, size_t bufsize, const char *fmt,
34 u32 *array, unsigned array_size)
35{
36 size_t ret = 0;
37 unsigned i;
38
39 for(i = 0; i < array_size; i++) {
40 size_t len;
41
42 len = snprintf(buf, bufsize, fmt, array[i]);
43 len++; /* ' ' or '\n' */
44 ret += len;
45
46 if (buf) {
47 buf += len;
48 bufsize -= len;
49 buf[-1] = (i == array_size-1) ? '\n' : ' ';
50 }
51 }
52
53 ret++; /* \0 */
54 if (buf)
55 *buf = '\0';
56
57 return ret;
58}
59
60static char *format_array_alloc(const char *fmt, u32 *array, unsigned array_size)
61{
62 size_t len = format_array(NULL, 0, fmt, array, array_size);
63 char *ret;
64
65 ret = kmalloc(len, GFP_KERNEL);
66 if (ret == NULL)
67 return NULL;
68
69 format_array(ret, len, fmt, array, array_size);
70 return ret;
71}
72
73static ssize_t u32_array_read(struct file *file, char __user *buf, size_t len,
74 loff_t *ppos)
75{
76 struct inode *inode = file->f_path.dentry->d_inode;
77 struct array_data *data = inode->i_private;
78 size_t size;
79
80 if (*ppos == 0) {
81 if (file->private_data) {
82 kfree(file->private_data);
83 file->private_data = NULL;
84 }
85
86 file->private_data = format_array_alloc("%u", data->array, data->elements);
87 }
88
89 size = 0;
90 if (file->private_data)
91 size = strlen(file->private_data);
92
93 return simple_read_from_buffer(buf, len, ppos, file->private_data, size);
94}
95
96static int xen_array_release(struct inode *inode, struct file *file)
97{
98 kfree(file->private_data);
99
100 return 0;
101}
102
103static struct file_operations u32_array_fops = {
104 .owner = THIS_MODULE,
105 .open = u32_array_open,
106 .release= xen_array_release,
107 .read = u32_array_read,
108};
109
110struct dentry *xen_debugfs_create_u32_array(const char *name, mode_t mode,
111 struct dentry *parent,
112 u32 *array, unsigned elements)
113{
114 struct array_data *data = kmalloc(sizeof(*data), GFP_KERNEL);
115
116 if (data == NULL)
117 return NULL;
118
119 data->array = array;
120 data->elements = elements;
121
122 return debugfs_create_file(name, mode, parent, data, &u32_array_fops);
123}
diff --git a/arch/x86/xen/debugfs.h b/arch/x86/xen/debugfs.h
new file mode 100644
index 000000000000..e28132084832
--- /dev/null
+++ b/arch/x86/xen/debugfs.h
@@ -0,0 +1,10 @@
1#ifndef _XEN_DEBUGFS_H
2#define _XEN_DEBUGFS_H
3
4struct dentry * __init xen_init_debugfs(void);
5
6struct dentry *xen_debugfs_create_u32_array(const char *name, mode_t mode,
7 struct dentry *parent,
8 u32 *array, unsigned elements);
9
10#endif /* _XEN_DEBUGFS_H */
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 9ff6e3cbf08f..0013a729b41d 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -30,12 +30,12 @@
30#include <xen/interface/xen.h> 30#include <xen/interface/xen.h>
31#include <xen/interface/physdev.h> 31#include <xen/interface/physdev.h>
32#include <xen/interface/vcpu.h> 32#include <xen/interface/vcpu.h>
33#include <xen/interface/sched.h>
34#include <xen/features.h> 33#include <xen/features.h>
35#include <xen/page.h> 34#include <xen/page.h>
36#include <xen/hvc-console.h> 35#include <xen/hvc-console.h>
37 36
38#include <asm/paravirt.h> 37#include <asm/paravirt.h>
38#include <asm/apic.h>
39#include <asm/page.h> 39#include <asm/page.h>
40#include <asm/xen/hypercall.h> 40#include <asm/xen/hypercall.h>
41#include <asm/xen/hypervisor.h> 41#include <asm/xen/hypervisor.h>
@@ -57,6 +57,9 @@ EXPORT_SYMBOL_GPL(hypercall_page);
57DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); 57DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
58DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); 58DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
59 59
60enum xen_domain_type xen_domain_type = XEN_NATIVE;
61EXPORT_SYMBOL_GPL(xen_domain_type);
62
60/* 63/*
61 * Identity map, in addition to plain kernel map. This needs to be 64 * Identity map, in addition to plain kernel map. This needs to be
62 * large enough to allocate page table pages to allocate the rest. 65 * large enough to allocate page table pages to allocate the rest.
@@ -110,7 +113,14 @@ struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
110 * 113 *
111 * 0: not available, 1: available 114 * 0: not available, 1: available
112 */ 115 */
113static int have_vcpu_info_placement = 1; 116static int have_vcpu_info_placement =
117#ifdef CONFIG_X86_32
118 1
119#else
120 0
121#endif
122 ;
123
114 124
115static void xen_vcpu_setup(int cpu) 125static void xen_vcpu_setup(int cpu)
116{ 126{
@@ -226,103 +236,68 @@ static unsigned long xen_get_debugreg(int reg)
226 return HYPERVISOR_get_debugreg(reg); 236 return HYPERVISOR_get_debugreg(reg);
227} 237}
228 238
229static unsigned long xen_save_fl(void) 239static void xen_leave_lazy(void)
230{ 240{
231 struct vcpu_info *vcpu; 241 paravirt_leave_lazy(paravirt_get_lazy_mode());
232 unsigned long flags; 242 xen_mc_flush();
233
234 vcpu = x86_read_percpu(xen_vcpu);
235
236 /* flag has opposite sense of mask */
237 flags = !vcpu->evtchn_upcall_mask;
238
239 /* convert to IF type flag
240 -0 -> 0x00000000
241 -1 -> 0xffffffff
242 */
243 return (-flags) & X86_EFLAGS_IF;
244} 243}
245 244
246static void xen_restore_fl(unsigned long flags) 245static unsigned long xen_store_tr(void)
247{ 246{
248 struct vcpu_info *vcpu; 247 return 0;
249
250 /* convert from IF type flag */
251 flags = !(flags & X86_EFLAGS_IF);
252
253 /* There's a one instruction preempt window here. We need to
254 make sure we're don't switch CPUs between getting the vcpu
255 pointer and updating the mask. */
256 preempt_disable();
257 vcpu = x86_read_percpu(xen_vcpu);
258 vcpu->evtchn_upcall_mask = flags;
259 preempt_enable_no_resched();
260
261 /* Doesn't matter if we get preempted here, because any
262 pending event will get dealt with anyway. */
263
264 if (flags == 0) {
265 preempt_check_resched();
266 barrier(); /* unmask then check (avoid races) */
267 if (unlikely(vcpu->evtchn_upcall_pending))
268 force_evtchn_callback();
269 }
270} 248}
271 249
272static void xen_irq_disable(void) 250/*
251 * Set the page permissions for a particular virtual address. If the
252 * address is a vmalloc mapping (or other non-linear mapping), then
253 * find the linear mapping of the page and also set its protections to
254 * match.
255 */
256static void set_aliased_prot(void *v, pgprot_t prot)
273{ 257{
274 /* There's a one instruction preempt window here. We need to 258 int level;
275 make sure we're don't switch CPUs between getting the vcpu 259 pte_t *ptep;
276 pointer and updating the mask. */ 260 pte_t pte;
277 preempt_disable(); 261 unsigned long pfn;
278 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1; 262 struct page *page;
279 preempt_enable_no_resched();
280}
281 263
282static void xen_irq_enable(void) 264 ptep = lookup_address((unsigned long)v, &level);
283{ 265 BUG_ON(ptep == NULL);
284 struct vcpu_info *vcpu;
285 266
286 /* We don't need to worry about being preempted here, since 267 pfn = pte_pfn(*ptep);
287 either a) interrupts are disabled, so no preemption, or b) 268 page = pfn_to_page(pfn);
288 the caller is confused and is trying to re-enable interrupts
289 on an indeterminate processor. */
290 269
291 vcpu = x86_read_percpu(xen_vcpu); 270 pte = pfn_pte(pfn, prot);
292 vcpu->evtchn_upcall_mask = 0;
293 271
294 /* Doesn't matter if we get preempted here, because any 272 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
295 pending event will get dealt with anyway. */ 273 BUG();
296 274
297 barrier(); /* unmask then check (avoid races) */ 275 if (!PageHighMem(page)) {
298 if (unlikely(vcpu->evtchn_upcall_pending)) 276 void *av = __va(PFN_PHYS(pfn));
299 force_evtchn_callback();
300}
301 277
302static void xen_safe_halt(void) 278 if (av != v)
303{ 279 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
304 /* Blocking includes an implicit local_irq_enable(). */ 280 BUG();
305 if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0) 281 } else
306 BUG(); 282 kmap_flush_unused();
307} 283}
308 284
309static void xen_halt(void) 285static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
310{ 286{
311 if (irqs_disabled()) 287 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
312 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); 288 int i;
313 else
314 xen_safe_halt();
315}
316 289
317static void xen_leave_lazy(void) 290 for(i = 0; i < entries; i += entries_per_page)
318{ 291 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
319 paravirt_leave_lazy(paravirt_get_lazy_mode());
320 xen_mc_flush();
321} 292}
322 293
323static unsigned long xen_store_tr(void) 294static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
324{ 295{
325 return 0; 296 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
297 int i;
298
299 for(i = 0; i < entries; i += entries_per_page)
300 set_aliased_prot(ldt + i, PAGE_KERNEL);
326} 301}
327 302
328static void xen_set_ldt(const void *addr, unsigned entries) 303static void xen_set_ldt(const void *addr, unsigned entries)
@@ -425,8 +400,7 @@ static void xen_load_gs_index(unsigned int idx)
425static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, 400static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
426 const void *ptr) 401 const void *ptr)
427{ 402{
428 unsigned long lp = (unsigned long)&dt[entrynum]; 403 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
429 xmaddr_t mach_lp = virt_to_machine(lp);
430 u64 entry = *(u64 *)ptr; 404 u64 entry = *(u64 *)ptr;
431 405
432 preempt_disable(); 406 preempt_disable();
@@ -559,7 +533,7 @@ static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
559} 533}
560 534
561static void xen_load_sp0(struct tss_struct *tss, 535static void xen_load_sp0(struct tss_struct *tss,
562 struct thread_struct *thread) 536 struct thread_struct *thread)
563{ 537{
564 struct multicall_space mcs = xen_mc_entry(0); 538 struct multicall_space mcs = xen_mc_entry(0);
565 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); 539 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
@@ -580,16 +554,47 @@ static void xen_io_delay(void)
580} 554}
581 555
582#ifdef CONFIG_X86_LOCAL_APIC 556#ifdef CONFIG_X86_LOCAL_APIC
583static u32 xen_apic_read(unsigned long reg) 557static u32 xen_apic_read(u32 reg)
584{ 558{
585 return 0; 559 return 0;
586} 560}
587 561
588static void xen_apic_write(unsigned long reg, u32 val) 562static void xen_apic_write(u32 reg, u32 val)
589{ 563{
590 /* Warn to see if there's any stray references */ 564 /* Warn to see if there's any stray references */
591 WARN_ON(1); 565 WARN_ON(1);
592} 566}
567
568static u64 xen_apic_icr_read(void)
569{
570 return 0;
571}
572
573static void xen_apic_icr_write(u32 low, u32 id)
574{
575 /* Warn to see if there's any stray references */
576 WARN_ON(1);
577}
578
579static void xen_apic_wait_icr_idle(void)
580{
581 return;
582}
583
584static u32 xen_safe_apic_wait_icr_idle(void)
585{
586 return 0;
587}
588
589static struct apic_ops xen_basic_apic_ops = {
590 .read = xen_apic_read,
591 .write = xen_apic_write,
592 .icr_read = xen_apic_icr_read,
593 .icr_write = xen_apic_icr_write,
594 .wait_icr_idle = xen_apic_wait_icr_idle,
595 .safe_wait_icr_idle = xen_safe_apic_wait_icr_idle,
596};
597
593#endif 598#endif
594 599
595static void xen_flush_tlb(void) 600static void xen_flush_tlb(void)
@@ -803,6 +808,19 @@ static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
803 ret = -EFAULT; 808 ret = -EFAULT;
804 break; 809 break;
805#endif 810#endif
811
812 case MSR_STAR:
813 case MSR_CSTAR:
814 case MSR_LSTAR:
815 case MSR_SYSCALL_MASK:
816 case MSR_IA32_SYSENTER_CS:
817 case MSR_IA32_SYSENTER_ESP:
818 case MSR_IA32_SYSENTER_EIP:
819 /* Fast syscall setup is all done in hypercalls, so
820 these are all ignored. Stub them out here to stop
821 Xen console noise. */
822 break;
823
806 default: 824 default:
807 ret = native_write_msr_safe(msr, low, high); 825 ret = native_write_msr_safe(msr, low, high);
808 } 826 }
@@ -812,7 +830,7 @@ static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
812 830
813/* Early in boot, while setting up the initial pagetable, assume 831/* Early in boot, while setting up the initial pagetable, assume
814 everything is pinned. */ 832 everything is pinned. */
815static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn) 833static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
816{ 834{
817#ifdef CONFIG_FLATMEM 835#ifdef CONFIG_FLATMEM
818 BUG_ON(mem_map); /* should only be used early */ 836 BUG_ON(mem_map); /* should only be used early */
@@ -822,7 +840,7 @@ static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
822 840
823/* Early release_pte assumes that all pts are pinned, since there's 841/* Early release_pte assumes that all pts are pinned, since there's
824 only init_mm and anything attached to that is pinned. */ 842 only init_mm and anything attached to that is pinned. */
825static void xen_release_pte_init(u32 pfn) 843static void xen_release_pte_init(unsigned long pfn)
826{ 844{
827 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); 845 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
828} 846}
@@ -838,7 +856,7 @@ static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
838 856
839/* This needs to make sure the new pte page is pinned iff its being 857/* This needs to make sure the new pte page is pinned iff its being
840 attached to a pinned pagetable. */ 858 attached to a pinned pagetable. */
841static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level) 859static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
842{ 860{
843 struct page *page = pfn_to_page(pfn); 861 struct page *page = pfn_to_page(pfn);
844 862
@@ -846,8 +864,8 @@ static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
846 SetPagePinned(page); 864 SetPagePinned(page);
847 865
848 if (!PageHighMem(page)) { 866 if (!PageHighMem(page)) {
849 make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); 867 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
850 if (level == PT_PTE) 868 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
851 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); 869 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
852 } else 870 } else
853 /* make sure there are no stray mappings of 871 /* make sure there are no stray mappings of
@@ -856,12 +874,12 @@ static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
856 } 874 }
857} 875}
858 876
859static void xen_alloc_pte(struct mm_struct *mm, u32 pfn) 877static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
860{ 878{
861 xen_alloc_ptpage(mm, pfn, PT_PTE); 879 xen_alloc_ptpage(mm, pfn, PT_PTE);
862} 880}
863 881
864static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn) 882static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
865{ 883{
866 xen_alloc_ptpage(mm, pfn, PT_PMD); 884 xen_alloc_ptpage(mm, pfn, PT_PMD);
867} 885}
@@ -909,13 +927,13 @@ static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
909} 927}
910 928
911/* This should never happen until we're OK to use struct page */ 929/* This should never happen until we're OK to use struct page */
912static void xen_release_ptpage(u32 pfn, unsigned level) 930static void xen_release_ptpage(unsigned long pfn, unsigned level)
913{ 931{
914 struct page *page = pfn_to_page(pfn); 932 struct page *page = pfn_to_page(pfn);
915 933
916 if (PagePinned(page)) { 934 if (PagePinned(page)) {
917 if (!PageHighMem(page)) { 935 if (!PageHighMem(page)) {
918 if (level == PT_PTE) 936 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
919 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); 937 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
920 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); 938 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
921 } 939 }
@@ -923,23 +941,23 @@ static void xen_release_ptpage(u32 pfn, unsigned level)
923 } 941 }
924} 942}
925 943
926static void xen_release_pte(u32 pfn) 944static void xen_release_pte(unsigned long pfn)
927{ 945{
928 xen_release_ptpage(pfn, PT_PTE); 946 xen_release_ptpage(pfn, PT_PTE);
929} 947}
930 948
931static void xen_release_pmd(u32 pfn) 949static void xen_release_pmd(unsigned long pfn)
932{ 950{
933 xen_release_ptpage(pfn, PT_PMD); 951 xen_release_ptpage(pfn, PT_PMD);
934} 952}
935 953
936#if PAGETABLE_LEVELS == 4 954#if PAGETABLE_LEVELS == 4
937static void xen_alloc_pud(struct mm_struct *mm, u32 pfn) 955static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
938{ 956{
939 xen_alloc_ptpage(mm, pfn, PT_PUD); 957 xen_alloc_ptpage(mm, pfn, PT_PUD);
940} 958}
941 959
942static void xen_release_pud(u32 pfn) 960static void xen_release_pud(unsigned long pfn)
943{ 961{
944 xen_release_ptpage(pfn, PT_PUD); 962 xen_release_ptpage(pfn, PT_PUD);
945} 963}
@@ -962,6 +980,7 @@ static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
962} 980}
963#endif 981#endif
964 982
983#ifdef CONFIG_X86_32
965static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) 984static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
966{ 985{
967 /* If there's an existing pte, then don't allow _PAGE_RW to be set */ 986 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
@@ -980,6 +999,7 @@ static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
980 999
981 xen_set_pte(ptep, pte); 1000 xen_set_pte(ptep, pte);
982} 1001}
1002#endif
983 1003
984static __init void xen_pagetable_setup_start(pgd_t *base) 1004static __init void xen_pagetable_setup_start(pgd_t *base)
985{ 1005{
@@ -1046,7 +1066,6 @@ void xen_setup_vcpu_info_placement(void)
1046 1066
1047 /* xen_vcpu_setup managed to place the vcpu_info within the 1067 /* xen_vcpu_setup managed to place the vcpu_info within the
1048 percpu area for all cpus, so make use of it */ 1068 percpu area for all cpus, so make use of it */
1049#ifdef CONFIG_X86_32
1050 if (have_vcpu_info_placement) { 1069 if (have_vcpu_info_placement) {
1051 printk(KERN_INFO "Xen: using vcpu_info placement\n"); 1070 printk(KERN_INFO "Xen: using vcpu_info placement\n");
1052 1071
@@ -1056,7 +1075,6 @@ void xen_setup_vcpu_info_placement(void)
1056 pv_irq_ops.irq_enable = xen_irq_enable_direct; 1075 pv_irq_ops.irq_enable = xen_irq_enable_direct;
1057 pv_mmu_ops.read_cr2 = xen_read_cr2_direct; 1076 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1058 } 1077 }
1059#endif
1060} 1078}
1061 1079
1062static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, 1080static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
@@ -1077,12 +1095,10 @@ static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1077 goto patch_site 1095 goto patch_site
1078 1096
1079 switch (type) { 1097 switch (type) {
1080#ifdef CONFIG_X86_32
1081 SITE(pv_irq_ops, irq_enable); 1098 SITE(pv_irq_ops, irq_enable);
1082 SITE(pv_irq_ops, irq_disable); 1099 SITE(pv_irq_ops, irq_disable);
1083 SITE(pv_irq_ops, save_fl); 1100 SITE(pv_irq_ops, save_fl);
1084 SITE(pv_irq_ops, restore_fl); 1101 SITE(pv_irq_ops, restore_fl);
1085#endif /* CONFIG_X86_32 */
1086#undef SITE 1102#undef SITE
1087 1103
1088 patch_site: 1104 patch_site:
@@ -1220,6 +1236,9 @@ static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1220 .load_gs_index = xen_load_gs_index, 1236 .load_gs_index = xen_load_gs_index,
1221#endif 1237#endif
1222 1238
1239 .alloc_ldt = xen_alloc_ldt,
1240 .free_ldt = xen_free_ldt,
1241
1223 .store_gdt = native_store_gdt, 1242 .store_gdt = native_store_gdt,
1224 .store_idt = native_store_idt, 1243 .store_idt = native_store_idt,
1225 .store_tr = xen_store_tr, 1244 .store_tr = xen_store_tr,
@@ -1241,40 +1260,8 @@ static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1241 }, 1260 },
1242}; 1261};
1243 1262
1244static void __init __xen_init_IRQ(void)
1245{
1246#ifdef CONFIG_X86_64
1247 int i;
1248
1249 /* Create identity vector->irq map */
1250 for(i = 0; i < NR_VECTORS; i++) {
1251 int cpu;
1252
1253 for_each_possible_cpu(cpu)
1254 per_cpu(vector_irq, cpu)[i] = i;
1255 }
1256#endif /* CONFIG_X86_64 */
1257
1258 xen_init_IRQ();
1259}
1260
1261static const struct pv_irq_ops xen_irq_ops __initdata = {
1262 .init_IRQ = __xen_init_IRQ,
1263 .save_fl = xen_save_fl,
1264 .restore_fl = xen_restore_fl,
1265 .irq_disable = xen_irq_disable,
1266 .irq_enable = xen_irq_enable,
1267 .safe_halt = xen_safe_halt,
1268 .halt = xen_halt,
1269#ifdef CONFIG_X86_64
1270 .adjust_exception_frame = xen_adjust_exception_frame,
1271#endif
1272};
1273
1274static const struct pv_apic_ops xen_apic_ops __initdata = { 1263static const struct pv_apic_ops xen_apic_ops __initdata = {
1275#ifdef CONFIG_X86_LOCAL_APIC 1264#ifdef CONFIG_X86_LOCAL_APIC
1276 .apic_write = xen_apic_write,
1277 .apic_read = xen_apic_read,
1278 .setup_boot_clock = paravirt_nop, 1265 .setup_boot_clock = paravirt_nop,
1279 .setup_secondary_clock = paravirt_nop, 1266 .setup_secondary_clock = paravirt_nop,
1280 .startup_ipi_hook = paravirt_nop, 1267 .startup_ipi_hook = paravirt_nop,
@@ -1324,7 +1311,7 @@ static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1324 .ptep_modify_prot_commit = __ptep_modify_prot_commit, 1311 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1325 1312
1326 .pte_val = xen_pte_val, 1313 .pte_val = xen_pte_val,
1327 .pte_flags = native_pte_val, 1314 .pte_flags = native_pte_flags,
1328 .pgd_val = xen_pgd_val, 1315 .pgd_val = xen_pgd_val,
1329 1316
1330 .make_pte = xen_make_pte, 1317 .make_pte = xen_make_pte,
@@ -1413,7 +1400,7 @@ static void __init xen_reserve_top(void)
1413 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) 1400 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1414 top = pp.virt_start; 1401 top = pp.virt_start;
1415 1402
1416 reserve_top_address(-top + 2 * PAGE_SIZE); 1403 reserve_top_address(-top);
1417#endif /* CONFIG_X86_32 */ 1404#endif /* CONFIG_X86_32 */
1418} 1405}
1419 1406
@@ -1447,48 +1434,11 @@ static void *m2v(phys_addr_t maddr)
1447 return __ka(m2p(maddr)); 1434 return __ka(m2p(maddr));
1448} 1435}
1449 1436
1450#ifdef CONFIG_X86_64
1451static void walk(pgd_t *pgd, unsigned long addr)
1452{
1453 unsigned l4idx = pgd_index(addr);
1454 unsigned l3idx = pud_index(addr);
1455 unsigned l2idx = pmd_index(addr);
1456 unsigned l1idx = pte_index(addr);
1457 pgd_t l4;
1458 pud_t l3;
1459 pmd_t l2;
1460 pte_t l1;
1461
1462 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1463 pgd, addr, l4idx, l3idx, l2idx, l1idx);
1464
1465 l4 = pgd[l4idx];
1466 xen_raw_printk(" l4: %016lx\n", l4.pgd);
1467 xen_raw_printk(" %016lx\n", pgd_val(l4));
1468
1469 l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1470 xen_raw_printk(" l3: %016lx\n", l3.pud);
1471 xen_raw_printk(" %016lx\n", pud_val(l3));
1472
1473 l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1474 xen_raw_printk(" l2: %016lx\n", l2.pmd);
1475 xen_raw_printk(" %016lx\n", pmd_val(l2));
1476
1477 l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1478 xen_raw_printk(" l1: %016lx\n", l1.pte);
1479 xen_raw_printk(" %016lx\n", pte_val(l1));
1480}
1481#endif
1482
1483static void set_page_prot(void *addr, pgprot_t prot) 1437static void set_page_prot(void *addr, pgprot_t prot)
1484{ 1438{
1485 unsigned long pfn = __pa(addr) >> PAGE_SHIFT; 1439 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1486 pte_t pte = pfn_pte(pfn, prot); 1440 pte_t pte = pfn_pte(pfn, prot);
1487 1441
1488 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1489 addr, pfn, get_phys_to_machine(pfn),
1490 pgprot_val(prot), pte.pte);
1491
1492 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) 1442 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1493 BUG(); 1443 BUG();
1494} 1444}
@@ -1664,6 +1614,8 @@ asmlinkage void __init xen_start_kernel(void)
1664 if (!xen_start_info) 1614 if (!xen_start_info)
1665 return; 1615 return;
1666 1616
1617 xen_domain_type = XEN_PV_DOMAIN;
1618
1667 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); 1619 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1668 1620
1669 xen_setup_features(); 1621 xen_setup_features();
@@ -1673,10 +1625,18 @@ asmlinkage void __init xen_start_kernel(void)
1673 pv_init_ops = xen_init_ops; 1625 pv_init_ops = xen_init_ops;
1674 pv_time_ops = xen_time_ops; 1626 pv_time_ops = xen_time_ops;
1675 pv_cpu_ops = xen_cpu_ops; 1627 pv_cpu_ops = xen_cpu_ops;
1676 pv_irq_ops = xen_irq_ops;
1677 pv_apic_ops = xen_apic_ops; 1628 pv_apic_ops = xen_apic_ops;
1678 pv_mmu_ops = xen_mmu_ops; 1629 pv_mmu_ops = xen_mmu_ops;
1679 1630
1631 xen_init_irq_ops();
1632
1633#ifdef CONFIG_X86_LOCAL_APIC
1634 /*
1635 * set up the basic apic ops.
1636 */
1637 apic_ops = &xen_basic_apic_ops;
1638#endif
1639
1680 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { 1640 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1681 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; 1641 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1682 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; 1642 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
@@ -1700,7 +1660,7 @@ asmlinkage void __init xen_start_kernel(void)
1700 1660
1701 /* Prevent unwanted bits from being set in PTEs. */ 1661 /* Prevent unwanted bits from being set in PTEs. */
1702 __supported_pte_mask &= ~_PAGE_GLOBAL; 1662 __supported_pte_mask &= ~_PAGE_GLOBAL;
1703 if (!is_initial_xendomain()) 1663 if (!xen_initial_domain())
1704 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); 1664 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1705 1665
1706 /* Don't do the full vcpu_info placement stuff until we have a 1666 /* Don't do the full vcpu_info placement stuff until we have a
@@ -1735,7 +1695,7 @@ asmlinkage void __init xen_start_kernel(void)
1735 boot_params.hdr.ramdisk_size = xen_start_info->mod_len; 1695 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1736 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); 1696 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1737 1697
1738 if (!is_initial_xendomain()) { 1698 if (!xen_initial_domain()) {
1739 add_preferred_console("xenboot", 0, NULL); 1699 add_preferred_console("xenboot", 0, NULL);
1740 add_preferred_console("tty", 0, NULL); 1700 add_preferred_console("tty", 0, NULL);
1741 add_preferred_console("hvc", 0, NULL); 1701 add_preferred_console("hvc", 0, NULL);
@@ -1743,15 +1703,6 @@ asmlinkage void __init xen_start_kernel(void)
1743 1703
1744 xen_raw_console_write("about to get started...\n"); 1704 xen_raw_console_write("about to get started...\n");
1745 1705
1746#if 0
1747 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1748 &boot_params, __pa_symbol(&boot_params),
1749 __va(__pa_symbol(&boot_params)));
1750
1751 walk(pgd, &boot_params);
1752 walk(pgd, __va(__pa(&boot_params)));
1753#endif
1754
1755 /* Start the world */ 1706 /* Start the world */
1756#ifdef CONFIG_X86_32 1707#ifdef CONFIG_X86_32
1757 i386_start_kernel(); 1708 i386_start_kernel();
diff --git a/arch/x86/xen/irq.c b/arch/x86/xen/irq.c
new file mode 100644
index 000000000000..28b85ab8422e
--- /dev/null
+++ b/arch/x86/xen/irq.c
@@ -0,0 +1,143 @@
1#include <linux/hardirq.h>
2
3#include <xen/interface/xen.h>
4#include <xen/interface/sched.h>
5#include <xen/interface/vcpu.h>
6
7#include <asm/xen/hypercall.h>
8#include <asm/xen/hypervisor.h>
9
10#include "xen-ops.h"
11
12/*
13 * Force a proper event-channel callback from Xen after clearing the
14 * callback mask. We do this in a very simple manner, by making a call
15 * down into Xen. The pending flag will be checked by Xen on return.
16 */
17void xen_force_evtchn_callback(void)
18{
19 (void)HYPERVISOR_xen_version(0, NULL);
20}
21
22static void __init __xen_init_IRQ(void)
23{
24#ifdef CONFIG_X86_64
25 int i;
26
27 /* Create identity vector->irq map */
28 for(i = 0; i < NR_VECTORS; i++) {
29 int cpu;
30
31 for_each_possible_cpu(cpu)
32 per_cpu(vector_irq, cpu)[i] = i;
33 }
34#endif /* CONFIG_X86_64 */
35
36 xen_init_IRQ();
37}
38
39static unsigned long xen_save_fl(void)
40{
41 struct vcpu_info *vcpu;
42 unsigned long flags;
43
44 vcpu = x86_read_percpu(xen_vcpu);
45
46 /* flag has opposite sense of mask */
47 flags = !vcpu->evtchn_upcall_mask;
48
49 /* convert to IF type flag
50 -0 -> 0x00000000
51 -1 -> 0xffffffff
52 */
53 return (-flags) & X86_EFLAGS_IF;
54}
55
56static void xen_restore_fl(unsigned long flags)
57{
58 struct vcpu_info *vcpu;
59
60 /* convert from IF type flag */
61 flags = !(flags & X86_EFLAGS_IF);
62
63 /* There's a one instruction preempt window here. We need to
64 make sure we're don't switch CPUs between getting the vcpu
65 pointer and updating the mask. */
66 preempt_disable();
67 vcpu = x86_read_percpu(xen_vcpu);
68 vcpu->evtchn_upcall_mask = flags;
69 preempt_enable_no_resched();
70
71 /* Doesn't matter if we get preempted here, because any
72 pending event will get dealt with anyway. */
73
74 if (flags == 0) {
75 preempt_check_resched();
76 barrier(); /* unmask then check (avoid races) */
77 if (unlikely(vcpu->evtchn_upcall_pending))
78 xen_force_evtchn_callback();
79 }
80}
81
82static void xen_irq_disable(void)
83{
84 /* There's a one instruction preempt window here. We need to
85 make sure we're don't switch CPUs between getting the vcpu
86 pointer and updating the mask. */
87 preempt_disable();
88 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
89 preempt_enable_no_resched();
90}
91
92static void xen_irq_enable(void)
93{
94 struct vcpu_info *vcpu;
95
96 /* We don't need to worry about being preempted here, since
97 either a) interrupts are disabled, so no preemption, or b)
98 the caller is confused and is trying to re-enable interrupts
99 on an indeterminate processor. */
100
101 vcpu = x86_read_percpu(xen_vcpu);
102 vcpu->evtchn_upcall_mask = 0;
103
104 /* Doesn't matter if we get preempted here, because any
105 pending event will get dealt with anyway. */
106
107 barrier(); /* unmask then check (avoid races) */
108 if (unlikely(vcpu->evtchn_upcall_pending))
109 xen_force_evtchn_callback();
110}
111
112static void xen_safe_halt(void)
113{
114 /* Blocking includes an implicit local_irq_enable(). */
115 if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0)
116 BUG();
117}
118
119static void xen_halt(void)
120{
121 if (irqs_disabled())
122 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
123 else
124 xen_safe_halt();
125}
126
127static const struct pv_irq_ops xen_irq_ops __initdata = {
128 .init_IRQ = __xen_init_IRQ,
129 .save_fl = xen_save_fl,
130 .restore_fl = xen_restore_fl,
131 .irq_disable = xen_irq_disable,
132 .irq_enable = xen_irq_enable,
133 .safe_halt = xen_safe_halt,
134 .halt = xen_halt,
135#ifdef CONFIG_X86_64
136 .adjust_exception_frame = xen_adjust_exception_frame,
137#endif
138};
139
140void __init xen_init_irq_ops()
141{
142 pv_irq_ops = xen_irq_ops;
143}
diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
index aa37469da696..ae173f6edd8b 100644
--- a/arch/x86/xen/mmu.c
+++ b/arch/x86/xen/mmu.c
@@ -40,6 +40,7 @@
40 */ 40 */
41#include <linux/sched.h> 41#include <linux/sched.h>
42#include <linux/highmem.h> 42#include <linux/highmem.h>
43#include <linux/debugfs.h>
43#include <linux/bug.h> 44#include <linux/bug.h>
44 45
45#include <asm/pgtable.h> 46#include <asm/pgtable.h>
@@ -57,6 +58,61 @@
57 58
58#include "multicalls.h" 59#include "multicalls.h"
59#include "mmu.h" 60#include "mmu.h"
61#include "debugfs.h"
62
63#define MMU_UPDATE_HISTO 30
64
65#ifdef CONFIG_XEN_DEBUG_FS
66
67static struct {
68 u32 pgd_update;
69 u32 pgd_update_pinned;
70 u32 pgd_update_batched;
71
72 u32 pud_update;
73 u32 pud_update_pinned;
74 u32 pud_update_batched;
75
76 u32 pmd_update;
77 u32 pmd_update_pinned;
78 u32 pmd_update_batched;
79
80 u32 pte_update;
81 u32 pte_update_pinned;
82 u32 pte_update_batched;
83
84 u32 mmu_update;
85 u32 mmu_update_extended;
86 u32 mmu_update_histo[MMU_UPDATE_HISTO];
87
88 u32 prot_commit;
89 u32 prot_commit_batched;
90
91 u32 set_pte_at;
92 u32 set_pte_at_batched;
93 u32 set_pte_at_pinned;
94 u32 set_pte_at_current;
95 u32 set_pte_at_kernel;
96} mmu_stats;
97
98static u8 zero_stats;
99
100static inline void check_zero(void)
101{
102 if (unlikely(zero_stats)) {
103 memset(&mmu_stats, 0, sizeof(mmu_stats));
104 zero_stats = 0;
105 }
106}
107
108#define ADD_STATS(elem, val) \
109 do { check_zero(); mmu_stats.elem += (val); } while(0)
110
111#else /* !CONFIG_XEN_DEBUG_FS */
112
113#define ADD_STATS(elem, val) do { (void)(val); } while(0)
114
115#endif /* CONFIG_XEN_DEBUG_FS */
60 116
61/* 117/*
62 * Just beyond the highest usermode address. STACK_TOP_MAX has a 118 * Just beyond the highest usermode address. STACK_TOP_MAX has a
@@ -229,25 +285,35 @@ void make_lowmem_page_readwrite(void *vaddr)
229} 285}
230 286
231 287
232static bool page_pinned(void *ptr) 288static bool xen_page_pinned(void *ptr)
233{ 289{
234 struct page *page = virt_to_page(ptr); 290 struct page *page = virt_to_page(ptr);
235 291
236 return PagePinned(page); 292 return PagePinned(page);
237} 293}
238 294
239static void extend_mmu_update(const struct mmu_update *update) 295static void xen_extend_mmu_update(const struct mmu_update *update)
240{ 296{
241 struct multicall_space mcs; 297 struct multicall_space mcs;
242 struct mmu_update *u; 298 struct mmu_update *u;
243 299
244 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); 300 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
245 301
246 if (mcs.mc != NULL) 302 if (mcs.mc != NULL) {
303 ADD_STATS(mmu_update_extended, 1);
304 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
305
247 mcs.mc->args[1]++; 306 mcs.mc->args[1]++;
248 else { 307
308 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
309 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
310 else
311 ADD_STATS(mmu_update_histo[0], 1);
312 } else {
313 ADD_STATS(mmu_update, 1);
249 mcs = __xen_mc_entry(sizeof(*u)); 314 mcs = __xen_mc_entry(sizeof(*u));
250 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); 315 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
316 ADD_STATS(mmu_update_histo[1], 1);
251 } 317 }
252 318
253 u = mcs.args; 319 u = mcs.args;
@@ -265,7 +331,9 @@ void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
265 /* ptr may be ioremapped for 64-bit pagetable setup */ 331 /* ptr may be ioremapped for 64-bit pagetable setup */
266 u.ptr = arbitrary_virt_to_machine(ptr).maddr; 332 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
267 u.val = pmd_val_ma(val); 333 u.val = pmd_val_ma(val);
268 extend_mmu_update(&u); 334 xen_extend_mmu_update(&u);
335
336 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
269 337
270 xen_mc_issue(PARAVIRT_LAZY_MMU); 338 xen_mc_issue(PARAVIRT_LAZY_MMU);
271 339
@@ -274,13 +342,17 @@ void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
274 342
275void xen_set_pmd(pmd_t *ptr, pmd_t val) 343void xen_set_pmd(pmd_t *ptr, pmd_t val)
276{ 344{
345 ADD_STATS(pmd_update, 1);
346
277 /* If page is not pinned, we can just update the entry 347 /* If page is not pinned, we can just update the entry
278 directly */ 348 directly */
279 if (!page_pinned(ptr)) { 349 if (!xen_page_pinned(ptr)) {
280 *ptr = val; 350 *ptr = val;
281 return; 351 return;
282 } 352 }
283 353
354 ADD_STATS(pmd_update_pinned, 1);
355
284 xen_set_pmd_hyper(ptr, val); 356 xen_set_pmd_hyper(ptr, val);
285} 357}
286 358
@@ -300,12 +372,18 @@ void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
300 if (mm == &init_mm) 372 if (mm == &init_mm)
301 preempt_disable(); 373 preempt_disable();
302 374
375 ADD_STATS(set_pte_at, 1);
376// ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
377 ADD_STATS(set_pte_at_current, mm == current->mm);
378 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
379
303 if (mm == current->mm || mm == &init_mm) { 380 if (mm == current->mm || mm == &init_mm) {
304 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { 381 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
305 struct multicall_space mcs; 382 struct multicall_space mcs;
306 mcs = xen_mc_entry(0); 383 mcs = xen_mc_entry(0);
307 384
308 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); 385 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
386 ADD_STATS(set_pte_at_batched, 1);
309 xen_mc_issue(PARAVIRT_LAZY_MMU); 387 xen_mc_issue(PARAVIRT_LAZY_MMU);
310 goto out; 388 goto out;
311 } else 389 } else
@@ -334,7 +412,10 @@ void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
334 412
335 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; 413 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
336 u.val = pte_val_ma(pte); 414 u.val = pte_val_ma(pte);
337 extend_mmu_update(&u); 415 xen_extend_mmu_update(&u);
416
417 ADD_STATS(prot_commit, 1);
418 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
338 419
339 xen_mc_issue(PARAVIRT_LAZY_MMU); 420 xen_mc_issue(PARAVIRT_LAZY_MMU);
340} 421}
@@ -400,7 +481,9 @@ void xen_set_pud_hyper(pud_t *ptr, pud_t val)
400 /* ptr may be ioremapped for 64-bit pagetable setup */ 481 /* ptr may be ioremapped for 64-bit pagetable setup */
401 u.ptr = arbitrary_virt_to_machine(ptr).maddr; 482 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
402 u.val = pud_val_ma(val); 483 u.val = pud_val_ma(val);
403 extend_mmu_update(&u); 484 xen_extend_mmu_update(&u);
485
486 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
404 487
405 xen_mc_issue(PARAVIRT_LAZY_MMU); 488 xen_mc_issue(PARAVIRT_LAZY_MMU);
406 489
@@ -409,18 +492,26 @@ void xen_set_pud_hyper(pud_t *ptr, pud_t val)
409 492
410void xen_set_pud(pud_t *ptr, pud_t val) 493void xen_set_pud(pud_t *ptr, pud_t val)
411{ 494{
495 ADD_STATS(pud_update, 1);
496
412 /* If page is not pinned, we can just update the entry 497 /* If page is not pinned, we can just update the entry
413 directly */ 498 directly */
414 if (!page_pinned(ptr)) { 499 if (!xen_page_pinned(ptr)) {
415 *ptr = val; 500 *ptr = val;
416 return; 501 return;
417 } 502 }
418 503
504 ADD_STATS(pud_update_pinned, 1);
505
419 xen_set_pud_hyper(ptr, val); 506 xen_set_pud_hyper(ptr, val);
420} 507}
421 508
422void xen_set_pte(pte_t *ptep, pte_t pte) 509void xen_set_pte(pte_t *ptep, pte_t pte)
423{ 510{
511 ADD_STATS(pte_update, 1);
512// ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
513 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
514
424#ifdef CONFIG_X86_PAE 515#ifdef CONFIG_X86_PAE
425 ptep->pte_high = pte.pte_high; 516 ptep->pte_high = pte.pte_high;
426 smp_wmb(); 517 smp_wmb();
@@ -490,7 +581,7 @@ static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
490 581
491 u.ptr = virt_to_machine(ptr).maddr; 582 u.ptr = virt_to_machine(ptr).maddr;
492 u.val = pgd_val_ma(val); 583 u.val = pgd_val_ma(val);
493 extend_mmu_update(&u); 584 xen_extend_mmu_update(&u);
494} 585}
495 586
496/* 587/*
@@ -517,17 +608,22 @@ void xen_set_pgd(pgd_t *ptr, pgd_t val)
517{ 608{
518 pgd_t *user_ptr = xen_get_user_pgd(ptr); 609 pgd_t *user_ptr = xen_get_user_pgd(ptr);
519 610
611 ADD_STATS(pgd_update, 1);
612
520 /* If page is not pinned, we can just update the entry 613 /* If page is not pinned, we can just update the entry
521 directly */ 614 directly */
522 if (!page_pinned(ptr)) { 615 if (!xen_page_pinned(ptr)) {
523 *ptr = val; 616 *ptr = val;
524 if (user_ptr) { 617 if (user_ptr) {
525 WARN_ON(page_pinned(user_ptr)); 618 WARN_ON(xen_page_pinned(user_ptr));
526 *user_ptr = val; 619 *user_ptr = val;
527 } 620 }
528 return; 621 return;
529 } 622 }
530 623
624 ADD_STATS(pgd_update_pinned, 1);
625 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
626
531 /* If it's pinned, then we can at least batch the kernel and 627 /* If it's pinned, then we can at least batch the kernel and
532 user updates together. */ 628 user updates together. */
533 xen_mc_batch(); 629 xen_mc_batch();
@@ -555,9 +651,12 @@ void xen_set_pgd(pgd_t *ptr, pgd_t val)
555 * For 64-bit, we must skip the Xen hole in the middle of the address 651 * For 64-bit, we must skip the Xen hole in the middle of the address
556 * space, just after the big x86-64 virtual hole. 652 * space, just after the big x86-64 virtual hole.
557 */ 653 */
558static int pgd_walk(pgd_t *pgd, int (*func)(struct page *, enum pt_level), 654static int xen_pgd_walk(struct mm_struct *mm,
559 unsigned long limit) 655 int (*func)(struct mm_struct *mm, struct page *,
656 enum pt_level),
657 unsigned long limit)
560{ 658{
659 pgd_t *pgd = mm->pgd;
561 int flush = 0; 660 int flush = 0;
562 unsigned hole_low, hole_high; 661 unsigned hole_low, hole_high;
563 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; 662 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
@@ -590,8 +689,6 @@ static int pgd_walk(pgd_t *pgd, int (*func)(struct page *, enum pt_level),
590 pmdidx_limit = 0; 689 pmdidx_limit = 0;
591#endif 690#endif
592 691
593 flush |= (*func)(virt_to_page(pgd), PT_PGD);
594
595 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { 692 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
596 pud_t *pud; 693 pud_t *pud;
597 694
@@ -604,7 +701,7 @@ static int pgd_walk(pgd_t *pgd, int (*func)(struct page *, enum pt_level),
604 pud = pud_offset(&pgd[pgdidx], 0); 701 pud = pud_offset(&pgd[pgdidx], 0);
605 702
606 if (PTRS_PER_PUD > 1) /* not folded */ 703 if (PTRS_PER_PUD > 1) /* not folded */
607 flush |= (*func)(virt_to_page(pud), PT_PUD); 704 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
608 705
609 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { 706 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
610 pmd_t *pmd; 707 pmd_t *pmd;
@@ -619,7 +716,7 @@ static int pgd_walk(pgd_t *pgd, int (*func)(struct page *, enum pt_level),
619 pmd = pmd_offset(&pud[pudidx], 0); 716 pmd = pmd_offset(&pud[pudidx], 0);
620 717
621 if (PTRS_PER_PMD > 1) /* not folded */ 718 if (PTRS_PER_PMD > 1) /* not folded */
622 flush |= (*func)(virt_to_page(pmd), PT_PMD); 719 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
623 720
624 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { 721 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
625 struct page *pte; 722 struct page *pte;
@@ -633,28 +730,34 @@ static int pgd_walk(pgd_t *pgd, int (*func)(struct page *, enum pt_level),
633 continue; 730 continue;
634 731
635 pte = pmd_page(pmd[pmdidx]); 732 pte = pmd_page(pmd[pmdidx]);
636 flush |= (*func)(pte, PT_PTE); 733 flush |= (*func)(mm, pte, PT_PTE);
637 } 734 }
638 } 735 }
639 } 736 }
737
640out: 738out:
739 /* Do the top level last, so that the callbacks can use it as
740 a cue to do final things like tlb flushes. */
741 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
641 742
642 return flush; 743 return flush;
643} 744}
644 745
645static spinlock_t *lock_pte(struct page *page) 746/* If we're using split pte locks, then take the page's lock and
747 return a pointer to it. Otherwise return NULL. */
748static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
646{ 749{
647 spinlock_t *ptl = NULL; 750 spinlock_t *ptl = NULL;
648 751
649#if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS 752#if USE_SPLIT_PTLOCKS
650 ptl = __pte_lockptr(page); 753 ptl = __pte_lockptr(page);
651 spin_lock(ptl); 754 spin_lock_nest_lock(ptl, &mm->page_table_lock);
652#endif 755#endif
653 756
654 return ptl; 757 return ptl;
655} 758}
656 759
657static void do_unlock(void *v) 760static void xen_pte_unlock(void *v)
658{ 761{
659 spinlock_t *ptl = v; 762 spinlock_t *ptl = v;
660 spin_unlock(ptl); 763 spin_unlock(ptl);
@@ -672,7 +775,8 @@ static void xen_do_pin(unsigned level, unsigned long pfn)
672 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); 775 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
673} 776}
674 777
675static int pin_page(struct page *page, enum pt_level level) 778static int xen_pin_page(struct mm_struct *mm, struct page *page,
779 enum pt_level level)
676{ 780{
677 unsigned pgfl = TestSetPagePinned(page); 781 unsigned pgfl = TestSetPagePinned(page);
678 int flush; 782 int flush;
@@ -691,21 +795,40 @@ static int pin_page(struct page *page, enum pt_level level)
691 795
692 flush = 0; 796 flush = 0;
693 797
798 /*
799 * We need to hold the pagetable lock between the time
800 * we make the pagetable RO and when we actually pin
801 * it. If we don't, then other users may come in and
802 * attempt to update the pagetable by writing it,
803 * which will fail because the memory is RO but not
804 * pinned, so Xen won't do the trap'n'emulate.
805 *
806 * If we're using split pte locks, we can't hold the
807 * entire pagetable's worth of locks during the
808 * traverse, because we may wrap the preempt count (8
809 * bits). The solution is to mark RO and pin each PTE
810 * page while holding the lock. This means the number
811 * of locks we end up holding is never more than a
812 * batch size (~32 entries, at present).
813 *
814 * If we're not using split pte locks, we needn't pin
815 * the PTE pages independently, because we're
816 * protected by the overall pagetable lock.
817 */
694 ptl = NULL; 818 ptl = NULL;
695 if (level == PT_PTE) 819 if (level == PT_PTE)
696 ptl = lock_pte(page); 820 ptl = xen_pte_lock(page, mm);
697 821
698 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, 822 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
699 pfn_pte(pfn, PAGE_KERNEL_RO), 823 pfn_pte(pfn, PAGE_KERNEL_RO),
700 level == PT_PGD ? UVMF_TLB_FLUSH : 0); 824 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
701 825
702 if (level == PT_PTE) 826 if (ptl) {
703 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); 827 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
704 828
705 if (ptl) {
706 /* Queue a deferred unlock for when this batch 829 /* Queue a deferred unlock for when this batch
707 is completed. */ 830 is completed. */
708 xen_mc_callback(do_unlock, ptl); 831 xen_mc_callback(xen_pte_unlock, ptl);
709 } 832 }
710 } 833 }
711 834
@@ -715,11 +838,11 @@ static int pin_page(struct page *page, enum pt_level level)
715/* This is called just after a mm has been created, but it has not 838/* This is called just after a mm has been created, but it has not
716 been used yet. We need to make sure that its pagetable is all 839 been used yet. We need to make sure that its pagetable is all
717 read-only, and can be pinned. */ 840 read-only, and can be pinned. */
718void xen_pgd_pin(pgd_t *pgd) 841static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
719{ 842{
720 xen_mc_batch(); 843 xen_mc_batch();
721 844
722 if (pgd_walk(pgd, pin_page, USER_LIMIT)) { 845 if (xen_pgd_walk(mm, xen_pin_page, USER_LIMIT)) {
723 /* re-enable interrupts for kmap_flush_unused */ 846 /* re-enable interrupts for kmap_flush_unused */
724 xen_mc_issue(0); 847 xen_mc_issue(0);
725 kmap_flush_unused(); 848 kmap_flush_unused();
@@ -733,25 +856,35 @@ void xen_pgd_pin(pgd_t *pgd)
733 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); 856 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
734 857
735 if (user_pgd) { 858 if (user_pgd) {
736 pin_page(virt_to_page(user_pgd), PT_PGD); 859 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
737 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(user_pgd))); 860 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(user_pgd)));
738 } 861 }
739 } 862 }
740#else /* CONFIG_X86_32 */ 863#else /* CONFIG_X86_32 */
741#ifdef CONFIG_X86_PAE 864#ifdef CONFIG_X86_PAE
742 /* Need to make sure unshared kernel PMD is pinnable */ 865 /* Need to make sure unshared kernel PMD is pinnable */
743 pin_page(virt_to_page(pgd_page(pgd[pgd_index(TASK_SIZE)])), PT_PMD); 866 xen_pin_page(mm, virt_to_page(pgd_page(pgd[pgd_index(TASK_SIZE)])),
867 PT_PMD);
744#endif 868#endif
745 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); 869 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
746#endif /* CONFIG_X86_64 */ 870#endif /* CONFIG_X86_64 */
747 xen_mc_issue(0); 871 xen_mc_issue(0);
748} 872}
749 873
874static void xen_pgd_pin(struct mm_struct *mm)
875{
876 __xen_pgd_pin(mm, mm->pgd);
877}
878
750/* 879/*
751 * On save, we need to pin all pagetables to make sure they get their 880 * On save, we need to pin all pagetables to make sure they get their
752 * mfns turned into pfns. Search the list for any unpinned pgds and pin 881 * mfns turned into pfns. Search the list for any unpinned pgds and pin
753 * them (unpinned pgds are not currently in use, probably because the 882 * them (unpinned pgds are not currently in use, probably because the
754 * process is under construction or destruction). 883 * process is under construction or destruction).
884 *
885 * Expected to be called in stop_machine() ("equivalent to taking
886 * every spinlock in the system"), so the locking doesn't really
887 * matter all that much.
755 */ 888 */
756void xen_mm_pin_all(void) 889void xen_mm_pin_all(void)
757{ 890{
@@ -762,7 +895,7 @@ void xen_mm_pin_all(void)
762 895
763 list_for_each_entry(page, &pgd_list, lru) { 896 list_for_each_entry(page, &pgd_list, lru) {
764 if (!PagePinned(page)) { 897 if (!PagePinned(page)) {
765 xen_pgd_pin((pgd_t *)page_address(page)); 898 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
766 SetPageSavePinned(page); 899 SetPageSavePinned(page);
767 } 900 }
768 } 901 }
@@ -775,7 +908,8 @@ void xen_mm_pin_all(void)
775 * that's before we have page structures to store the bits. So do all 908 * that's before we have page structures to store the bits. So do all
776 * the book-keeping now. 909 * the book-keeping now.
777 */ 910 */
778static __init int mark_pinned(struct page *page, enum pt_level level) 911static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
912 enum pt_level level)
779{ 913{
780 SetPagePinned(page); 914 SetPagePinned(page);
781 return 0; 915 return 0;
@@ -783,10 +917,11 @@ static __init int mark_pinned(struct page *page, enum pt_level level)
783 917
784void __init xen_mark_init_mm_pinned(void) 918void __init xen_mark_init_mm_pinned(void)
785{ 919{
786 pgd_walk(init_mm.pgd, mark_pinned, FIXADDR_TOP); 920 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
787} 921}
788 922
789static int unpin_page(struct page *page, enum pt_level level) 923static int xen_unpin_page(struct mm_struct *mm, struct page *page,
924 enum pt_level level)
790{ 925{
791 unsigned pgfl = TestClearPagePinned(page); 926 unsigned pgfl = TestClearPagePinned(page);
792 927
@@ -796,10 +931,18 @@ static int unpin_page(struct page *page, enum pt_level level)
796 spinlock_t *ptl = NULL; 931 spinlock_t *ptl = NULL;
797 struct multicall_space mcs; 932 struct multicall_space mcs;
798 933
934 /*
935 * Do the converse to pin_page. If we're using split
936 * pte locks, we must be holding the lock for while
937 * the pte page is unpinned but still RO to prevent
938 * concurrent updates from seeing it in this
939 * partially-pinned state.
940 */
799 if (level == PT_PTE) { 941 if (level == PT_PTE) {
800 ptl = lock_pte(page); 942 ptl = xen_pte_lock(page, mm);
801 943
802 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); 944 if (ptl)
945 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
803 } 946 }
804 947
805 mcs = __xen_mc_entry(0); 948 mcs = __xen_mc_entry(0);
@@ -810,7 +953,7 @@ static int unpin_page(struct page *page, enum pt_level level)
810 953
811 if (ptl) { 954 if (ptl) {
812 /* unlock when batch completed */ 955 /* unlock when batch completed */
813 xen_mc_callback(do_unlock, ptl); 956 xen_mc_callback(xen_pte_unlock, ptl);
814 } 957 }
815 } 958 }
816 959
@@ -818,7 +961,7 @@ static int unpin_page(struct page *page, enum pt_level level)
818} 961}
819 962
820/* Release a pagetables pages back as normal RW */ 963/* Release a pagetables pages back as normal RW */
821static void xen_pgd_unpin(pgd_t *pgd) 964static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
822{ 965{
823 xen_mc_batch(); 966 xen_mc_batch();
824 967
@@ -830,21 +973,27 @@ static void xen_pgd_unpin(pgd_t *pgd)
830 973
831 if (user_pgd) { 974 if (user_pgd) {
832 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(user_pgd))); 975 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(user_pgd)));
833 unpin_page(virt_to_page(user_pgd), PT_PGD); 976 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
834 } 977 }
835 } 978 }
836#endif 979#endif
837 980
838#ifdef CONFIG_X86_PAE 981#ifdef CONFIG_X86_PAE
839 /* Need to make sure unshared kernel PMD is unpinned */ 982 /* Need to make sure unshared kernel PMD is unpinned */
840 pin_page(virt_to_page(pgd_page(pgd[pgd_index(TASK_SIZE)])), PT_PMD); 983 xen_unpin_page(mm, virt_to_page(pgd_page(pgd[pgd_index(TASK_SIZE)])),
984 PT_PMD);
841#endif 985#endif
842 986
843 pgd_walk(pgd, unpin_page, USER_LIMIT); 987 xen_pgd_walk(mm, xen_unpin_page, USER_LIMIT);
844 988
845 xen_mc_issue(0); 989 xen_mc_issue(0);
846} 990}
847 991
992static void xen_pgd_unpin(struct mm_struct *mm)
993{
994 __xen_pgd_unpin(mm, mm->pgd);
995}
996
848/* 997/*
849 * On resume, undo any pinning done at save, so that the rest of the 998 * On resume, undo any pinning done at save, so that the rest of the
850 * kernel doesn't see any unexpected pinned pagetables. 999 * kernel doesn't see any unexpected pinned pagetables.
@@ -859,7 +1008,7 @@ void xen_mm_unpin_all(void)
859 list_for_each_entry(page, &pgd_list, lru) { 1008 list_for_each_entry(page, &pgd_list, lru) {
860 if (PageSavePinned(page)) { 1009 if (PageSavePinned(page)) {
861 BUG_ON(!PagePinned(page)); 1010 BUG_ON(!PagePinned(page));
862 xen_pgd_unpin((pgd_t *)page_address(page)); 1011 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
863 ClearPageSavePinned(page); 1012 ClearPageSavePinned(page);
864 } 1013 }
865 } 1014 }
@@ -870,14 +1019,14 @@ void xen_mm_unpin_all(void)
870void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) 1019void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
871{ 1020{
872 spin_lock(&next->page_table_lock); 1021 spin_lock(&next->page_table_lock);
873 xen_pgd_pin(next->pgd); 1022 xen_pgd_pin(next);
874 spin_unlock(&next->page_table_lock); 1023 spin_unlock(&next->page_table_lock);
875} 1024}
876 1025
877void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) 1026void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
878{ 1027{
879 spin_lock(&mm->page_table_lock); 1028 spin_lock(&mm->page_table_lock);
880 xen_pgd_pin(mm->pgd); 1029 xen_pgd_pin(mm);
881 spin_unlock(&mm->page_table_lock); 1030 spin_unlock(&mm->page_table_lock);
882} 1031}
883 1032
@@ -907,7 +1056,7 @@ static void drop_other_mm_ref(void *info)
907 } 1056 }
908} 1057}
909 1058
910static void drop_mm_ref(struct mm_struct *mm) 1059static void xen_drop_mm_ref(struct mm_struct *mm)
911{ 1060{
912 cpumask_t mask; 1061 cpumask_t mask;
913 unsigned cpu; 1062 unsigned cpu;
@@ -937,7 +1086,7 @@ static void drop_mm_ref(struct mm_struct *mm)
937 smp_call_function_mask(mask, drop_other_mm_ref, mm, 1); 1086 smp_call_function_mask(mask, drop_other_mm_ref, mm, 1);
938} 1087}
939#else 1088#else
940static void drop_mm_ref(struct mm_struct *mm) 1089static void xen_drop_mm_ref(struct mm_struct *mm)
941{ 1090{
942 if (current->active_mm == mm) 1091 if (current->active_mm == mm)
943 load_cr3(swapper_pg_dir); 1092 load_cr3(swapper_pg_dir);
@@ -961,14 +1110,77 @@ static void drop_mm_ref(struct mm_struct *mm)
961void xen_exit_mmap(struct mm_struct *mm) 1110void xen_exit_mmap(struct mm_struct *mm)
962{ 1111{
963 get_cpu(); /* make sure we don't move around */ 1112 get_cpu(); /* make sure we don't move around */
964 drop_mm_ref(mm); 1113 xen_drop_mm_ref(mm);
965 put_cpu(); 1114 put_cpu();
966 1115
967 spin_lock(&mm->page_table_lock); 1116 spin_lock(&mm->page_table_lock);
968 1117
969 /* pgd may not be pinned in the error exit path of execve */ 1118 /* pgd may not be pinned in the error exit path of execve */
970 if (page_pinned(mm->pgd)) 1119 if (xen_page_pinned(mm->pgd))
971 xen_pgd_unpin(mm->pgd); 1120 xen_pgd_unpin(mm);
972 1121
973 spin_unlock(&mm->page_table_lock); 1122 spin_unlock(&mm->page_table_lock);
974} 1123}
1124
1125#ifdef CONFIG_XEN_DEBUG_FS
1126
1127static struct dentry *d_mmu_debug;
1128
1129static int __init xen_mmu_debugfs(void)
1130{
1131 struct dentry *d_xen = xen_init_debugfs();
1132
1133 if (d_xen == NULL)
1134 return -ENOMEM;
1135
1136 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1137
1138 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
1139
1140 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
1141 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
1142 &mmu_stats.pgd_update_pinned);
1143 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
1144 &mmu_stats.pgd_update_pinned);
1145
1146 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
1147 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
1148 &mmu_stats.pud_update_pinned);
1149 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
1150 &mmu_stats.pud_update_pinned);
1151
1152 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
1153 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
1154 &mmu_stats.pmd_update_pinned);
1155 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
1156 &mmu_stats.pmd_update_pinned);
1157
1158 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
1159// debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
1160// &mmu_stats.pte_update_pinned);
1161 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
1162 &mmu_stats.pte_update_pinned);
1163
1164 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
1165 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
1166 &mmu_stats.mmu_update_extended);
1167 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
1168 mmu_stats.mmu_update_histo, 20);
1169
1170 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
1171 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
1172 &mmu_stats.set_pte_at_batched);
1173 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
1174 &mmu_stats.set_pte_at_current);
1175 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
1176 &mmu_stats.set_pte_at_kernel);
1177
1178 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
1179 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
1180 &mmu_stats.prot_commit_batched);
1181
1182 return 0;
1183}
1184fs_initcall(xen_mmu_debugfs);
1185
1186#endif /* CONFIG_XEN_DEBUG_FS */
diff --git a/arch/x86/xen/mmu.h b/arch/x86/xen/mmu.h
index 0f59bd03f9e3..98d71659da5a 100644
--- a/arch/x86/xen/mmu.h
+++ b/arch/x86/xen/mmu.h
@@ -18,9 +18,6 @@ void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next);
18void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm); 18void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm);
19void xen_exit_mmap(struct mm_struct *mm); 19void xen_exit_mmap(struct mm_struct *mm);
20 20
21void xen_pgd_pin(pgd_t *pgd);
22//void xen_pgd_unpin(pgd_t *pgd);
23
24pteval_t xen_pte_val(pte_t); 21pteval_t xen_pte_val(pte_t);
25pmdval_t xen_pmd_val(pmd_t); 22pmdval_t xen_pmd_val(pmd_t);
26pgdval_t xen_pgd_val(pgd_t); 23pgdval_t xen_pgd_val(pgd_t);
diff --git a/arch/x86/xen/multicalls.c b/arch/x86/xen/multicalls.c
index 9efd1c6c9776..8ea8a0d0b0de 100644
--- a/arch/x86/xen/multicalls.c
+++ b/arch/x86/xen/multicalls.c
@@ -21,16 +21,20 @@
21 */ 21 */
22#include <linux/percpu.h> 22#include <linux/percpu.h>
23#include <linux/hardirq.h> 23#include <linux/hardirq.h>
24#include <linux/debugfs.h>
24 25
25#include <asm/xen/hypercall.h> 26#include <asm/xen/hypercall.h>
26 27
27#include "multicalls.h" 28#include "multicalls.h"
29#include "debugfs.h"
30
31#define MC_BATCH 32
28 32
29#define MC_DEBUG 1 33#define MC_DEBUG 1
30 34
31#define MC_BATCH 32
32#define MC_ARGS (MC_BATCH * 16) 35#define MC_ARGS (MC_BATCH * 16)
33 36
37
34struct mc_buffer { 38struct mc_buffer {
35 struct multicall_entry entries[MC_BATCH]; 39 struct multicall_entry entries[MC_BATCH];
36#if MC_DEBUG 40#if MC_DEBUG
@@ -47,6 +51,76 @@ struct mc_buffer {
47static DEFINE_PER_CPU(struct mc_buffer, mc_buffer); 51static DEFINE_PER_CPU(struct mc_buffer, mc_buffer);
48DEFINE_PER_CPU(unsigned long, xen_mc_irq_flags); 52DEFINE_PER_CPU(unsigned long, xen_mc_irq_flags);
49 53
54/* flush reasons 0- slots, 1- args, 2- callbacks */
55enum flush_reasons
56{
57 FL_SLOTS,
58 FL_ARGS,
59 FL_CALLBACKS,
60
61 FL_N_REASONS
62};
63
64#ifdef CONFIG_XEN_DEBUG_FS
65#define NHYPERCALLS 40 /* not really */
66
67static struct {
68 unsigned histo[MC_BATCH+1];
69
70 unsigned issued;
71 unsigned arg_total;
72 unsigned hypercalls;
73 unsigned histo_hypercalls[NHYPERCALLS];
74
75 unsigned flush[FL_N_REASONS];
76} mc_stats;
77
78static u8 zero_stats;
79
80static inline void check_zero(void)
81{
82 if (unlikely(zero_stats)) {
83 memset(&mc_stats, 0, sizeof(mc_stats));
84 zero_stats = 0;
85 }
86}
87
88static void mc_add_stats(const struct mc_buffer *mc)
89{
90 int i;
91
92 check_zero();
93
94 mc_stats.issued++;
95 mc_stats.hypercalls += mc->mcidx;
96 mc_stats.arg_total += mc->argidx;
97
98 mc_stats.histo[mc->mcidx]++;
99 for(i = 0; i < mc->mcidx; i++) {
100 unsigned op = mc->entries[i].op;
101 if (op < NHYPERCALLS)
102 mc_stats.histo_hypercalls[op]++;
103 }
104}
105
106static void mc_stats_flush(enum flush_reasons idx)
107{
108 check_zero();
109
110 mc_stats.flush[idx]++;
111}
112
113#else /* !CONFIG_XEN_DEBUG_FS */
114
115static inline void mc_add_stats(const struct mc_buffer *mc)
116{
117}
118
119static inline void mc_stats_flush(enum flush_reasons idx)
120{
121}
122#endif /* CONFIG_XEN_DEBUG_FS */
123
50void xen_mc_flush(void) 124void xen_mc_flush(void)
51{ 125{
52 struct mc_buffer *b = &__get_cpu_var(mc_buffer); 126 struct mc_buffer *b = &__get_cpu_var(mc_buffer);
@@ -60,6 +134,8 @@ void xen_mc_flush(void)
60 something in the middle */ 134 something in the middle */
61 local_irq_save(flags); 135 local_irq_save(flags);
62 136
137 mc_add_stats(b);
138
63 if (b->mcidx) { 139 if (b->mcidx) {
64#if MC_DEBUG 140#if MC_DEBUG
65 memcpy(b->debug, b->entries, 141 memcpy(b->debug, b->entries,
@@ -115,6 +191,7 @@ struct multicall_space __xen_mc_entry(size_t args)
115 191
116 if (b->mcidx == MC_BATCH || 192 if (b->mcidx == MC_BATCH ||
117 (argidx + args) > MC_ARGS) { 193 (argidx + args) > MC_ARGS) {
194 mc_stats_flush(b->mcidx == MC_BATCH ? FL_SLOTS : FL_ARGS);
118 xen_mc_flush(); 195 xen_mc_flush();
119 argidx = roundup(b->argidx, sizeof(u64)); 196 argidx = roundup(b->argidx, sizeof(u64));
120 } 197 }
@@ -158,10 +235,44 @@ void xen_mc_callback(void (*fn)(void *), void *data)
158 struct mc_buffer *b = &__get_cpu_var(mc_buffer); 235 struct mc_buffer *b = &__get_cpu_var(mc_buffer);
159 struct callback *cb; 236 struct callback *cb;
160 237
161 if (b->cbidx == MC_BATCH) 238 if (b->cbidx == MC_BATCH) {
239 mc_stats_flush(FL_CALLBACKS);
162 xen_mc_flush(); 240 xen_mc_flush();
241 }
163 242
164 cb = &b->callbacks[b->cbidx++]; 243 cb = &b->callbacks[b->cbidx++];
165 cb->fn = fn; 244 cb->fn = fn;
166 cb->data = data; 245 cb->data = data;
167} 246}
247
248#ifdef CONFIG_XEN_DEBUG_FS
249
250static struct dentry *d_mc_debug;
251
252static int __init xen_mc_debugfs(void)
253{
254 struct dentry *d_xen = xen_init_debugfs();
255
256 if (d_xen == NULL)
257 return -ENOMEM;
258
259 d_mc_debug = debugfs_create_dir("multicalls", d_xen);
260
261 debugfs_create_u8("zero_stats", 0644, d_mc_debug, &zero_stats);
262
263 debugfs_create_u32("batches", 0444, d_mc_debug, &mc_stats.issued);
264 debugfs_create_u32("hypercalls", 0444, d_mc_debug, &mc_stats.hypercalls);
265 debugfs_create_u32("arg_total", 0444, d_mc_debug, &mc_stats.arg_total);
266
267 xen_debugfs_create_u32_array("batch_histo", 0444, d_mc_debug,
268 mc_stats.histo, MC_BATCH);
269 xen_debugfs_create_u32_array("hypercall_histo", 0444, d_mc_debug,
270 mc_stats.histo_hypercalls, NHYPERCALLS);
271 xen_debugfs_create_u32_array("flush_reasons", 0444, d_mc_debug,
272 mc_stats.flush, FL_N_REASONS);
273
274 return 0;
275}
276fs_initcall(xen_mc_debugfs);
277
278#endif /* CONFIG_XEN_DEBUG_FS */
diff --git a/arch/x86/xen/setup.c b/arch/x86/xen/setup.c
index b6acc3a0af46..d67901083888 100644
--- a/arch/x86/xen/setup.c
+++ b/arch/x86/xen/setup.c
@@ -42,7 +42,7 @@ char * __init xen_memory_setup(void)
42 42
43 e820.nr_map = 0; 43 e820.nr_map = 0;
44 44
45 e820_add_region(0, PFN_PHYS(max_pfn), E820_RAM); 45 e820_add_region(0, PFN_PHYS((u64)max_pfn), E820_RAM);
46 46
47 /* 47 /*
48 * Even though this is normal, usable memory under Xen, reserve 48 * Even though this is normal, usable memory under Xen, reserve
diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
index d8faf79a0a1d..d77da613b1d2 100644
--- a/arch/x86/xen/smp.c
+++ b/arch/x86/xen/smp.c
@@ -11,11 +11,8 @@
11 * useful topology information for the kernel to make use of. As a 11 * useful topology information for the kernel to make use of. As a
12 * result, all CPUs are treated as if they're single-core and 12 * result, all CPUs are treated as if they're single-core and
13 * single-threaded. 13 * single-threaded.
14 *
15 * This does not handle HOTPLUG_CPU yet.
16 */ 14 */
17#include <linux/sched.h> 15#include <linux/sched.h>
18#include <linux/kernel_stat.h>
19#include <linux/err.h> 16#include <linux/err.h>
20#include <linux/smp.h> 17#include <linux/smp.h>
21 18
@@ -36,8 +33,6 @@
36#include "xen-ops.h" 33#include "xen-ops.h"
37#include "mmu.h" 34#include "mmu.h"
38 35
39static void __cpuinit xen_init_lock_cpu(int cpu);
40
41cpumask_t xen_cpu_initialized_map; 36cpumask_t xen_cpu_initialized_map;
42 37
43static DEFINE_PER_CPU(int, resched_irq); 38static DEFINE_PER_CPU(int, resched_irq);
@@ -64,11 +59,12 @@ static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
64 return IRQ_HANDLED; 59 return IRQ_HANDLED;
65} 60}
66 61
67static __cpuinit void cpu_bringup_and_idle(void) 62static __cpuinit void cpu_bringup(void)
68{ 63{
69 int cpu = smp_processor_id(); 64 int cpu = smp_processor_id();
70 65
71 cpu_init(); 66 cpu_init();
67 touch_softlockup_watchdog();
72 preempt_disable(); 68 preempt_disable();
73 69
74 xen_enable_sysenter(); 70 xen_enable_sysenter();
@@ -89,6 +85,11 @@ static __cpuinit void cpu_bringup_and_idle(void)
89 local_irq_enable(); 85 local_irq_enable();
90 86
91 wmb(); /* make sure everything is out */ 87 wmb(); /* make sure everything is out */
88}
89
90static __cpuinit void cpu_bringup_and_idle(void)
91{
92 cpu_bringup();
92 cpu_idle(); 93 cpu_idle();
93} 94}
94 95
@@ -212,8 +213,6 @@ static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
212 213
213 cpu_set(cpu, cpu_present_map); 214 cpu_set(cpu, cpu_present_map);
214 } 215 }
215
216 //init_xenbus_allowed_cpumask();
217} 216}
218 217
219static __cpuinit int 218static __cpuinit int
@@ -281,12 +280,6 @@ static int __cpuinit xen_cpu_up(unsigned int cpu)
281 struct task_struct *idle = idle_task(cpu); 280 struct task_struct *idle = idle_task(cpu);
282 int rc; 281 int rc;
283 282
284#if 0
285 rc = cpu_up_check(cpu);
286 if (rc)
287 return rc;
288#endif
289
290#ifdef CONFIG_X86_64 283#ifdef CONFIG_X86_64
291 /* Allocate node local memory for AP pdas */ 284 /* Allocate node local memory for AP pdas */
292 WARN_ON(cpu == 0); 285 WARN_ON(cpu == 0);
@@ -339,6 +332,60 @@ static void xen_smp_cpus_done(unsigned int max_cpus)
339{ 332{
340} 333}
341 334
335#ifdef CONFIG_HOTPLUG_CPU
336static int xen_cpu_disable(void)
337{
338 unsigned int cpu = smp_processor_id();
339 if (cpu == 0)
340 return -EBUSY;
341
342 cpu_disable_common();
343
344 load_cr3(swapper_pg_dir);
345 return 0;
346}
347
348static void xen_cpu_die(unsigned int cpu)
349{
350 while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
351 current->state = TASK_UNINTERRUPTIBLE;
352 schedule_timeout(HZ/10);
353 }
354 unbind_from_irqhandler(per_cpu(resched_irq, cpu), NULL);
355 unbind_from_irqhandler(per_cpu(callfunc_irq, cpu), NULL);
356 unbind_from_irqhandler(per_cpu(debug_irq, cpu), NULL);
357 unbind_from_irqhandler(per_cpu(callfuncsingle_irq, cpu), NULL);
358 xen_uninit_lock_cpu(cpu);
359 xen_teardown_timer(cpu);
360
361 if (num_online_cpus() == 1)
362 alternatives_smp_switch(0);
363}
364
365static void xen_play_dead(void)
366{
367 play_dead_common();
368 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
369 cpu_bringup();
370}
371
372#else /* !CONFIG_HOTPLUG_CPU */
373static int xen_cpu_disable(void)
374{
375 return -ENOSYS;
376}
377
378static void xen_cpu_die(unsigned int cpu)
379{
380 BUG();
381}
382
383static void xen_play_dead(void)
384{
385 BUG();
386}
387
388#endif
342static void stop_self(void *v) 389static void stop_self(void *v)
343{ 390{
344 int cpu = smp_processor_id(); 391 int cpu = smp_processor_id();
@@ -419,176 +466,16 @@ static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
419 return IRQ_HANDLED; 466 return IRQ_HANDLED;
420} 467}
421 468
422struct xen_spinlock {
423 unsigned char lock; /* 0 -> free; 1 -> locked */
424 unsigned short spinners; /* count of waiting cpus */
425};
426
427static int xen_spin_is_locked(struct raw_spinlock *lock)
428{
429 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
430
431 return xl->lock != 0;
432}
433
434static int xen_spin_is_contended(struct raw_spinlock *lock)
435{
436 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
437
438 /* Not strictly true; this is only the count of contended
439 lock-takers entering the slow path. */
440 return xl->spinners != 0;
441}
442
443static int xen_spin_trylock(struct raw_spinlock *lock)
444{
445 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
446 u8 old = 1;
447
448 asm("xchgb %b0,%1"
449 : "+q" (old), "+m" (xl->lock) : : "memory");
450
451 return old == 0;
452}
453
454static DEFINE_PER_CPU(int, lock_kicker_irq) = -1;
455static DEFINE_PER_CPU(struct xen_spinlock *, lock_spinners);
456
457static inline void spinning_lock(struct xen_spinlock *xl)
458{
459 __get_cpu_var(lock_spinners) = xl;
460 wmb(); /* set lock of interest before count */
461 asm(LOCK_PREFIX " incw %0"
462 : "+m" (xl->spinners) : : "memory");
463}
464
465static inline void unspinning_lock(struct xen_spinlock *xl)
466{
467 asm(LOCK_PREFIX " decw %0"
468 : "+m" (xl->spinners) : : "memory");
469 wmb(); /* decrement count before clearing lock */
470 __get_cpu_var(lock_spinners) = NULL;
471}
472
473static noinline int xen_spin_lock_slow(struct raw_spinlock *lock)
474{
475 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
476 int irq = __get_cpu_var(lock_kicker_irq);
477 int ret;
478
479 /* If kicker interrupts not initialized yet, just spin */
480 if (irq == -1)
481 return 0;
482
483 /* announce we're spinning */
484 spinning_lock(xl);
485
486 /* clear pending */
487 xen_clear_irq_pending(irq);
488
489 /* check again make sure it didn't become free while
490 we weren't looking */
491 ret = xen_spin_trylock(lock);
492 if (ret)
493 goto out;
494
495 /* block until irq becomes pending */
496 xen_poll_irq(irq);
497 kstat_this_cpu.irqs[irq]++;
498
499out:
500 unspinning_lock(xl);
501 return ret;
502}
503
504static void xen_spin_lock(struct raw_spinlock *lock)
505{
506 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
507 int timeout;
508 u8 oldval;
509
510 do {
511 timeout = 1 << 10;
512
513 asm("1: xchgb %1,%0\n"
514 " testb %1,%1\n"
515 " jz 3f\n"
516 "2: rep;nop\n"
517 " cmpb $0,%0\n"
518 " je 1b\n"
519 " dec %2\n"
520 " jnz 2b\n"
521 "3:\n"
522 : "+m" (xl->lock), "=q" (oldval), "+r" (timeout)
523 : "1" (1)
524 : "memory");
525
526 } while (unlikely(oldval != 0 && !xen_spin_lock_slow(lock)));
527}
528
529static noinline void xen_spin_unlock_slow(struct xen_spinlock *xl)
530{
531 int cpu;
532
533 for_each_online_cpu(cpu) {
534 /* XXX should mix up next cpu selection */
535 if (per_cpu(lock_spinners, cpu) == xl) {
536 xen_send_IPI_one(cpu, XEN_SPIN_UNLOCK_VECTOR);
537 break;
538 }
539 }
540}
541
542static void xen_spin_unlock(struct raw_spinlock *lock)
543{
544 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
545
546 smp_wmb(); /* make sure no writes get moved after unlock */
547 xl->lock = 0; /* release lock */
548
549 /* make sure unlock happens before kick */
550 barrier();
551
552 if (unlikely(xl->spinners))
553 xen_spin_unlock_slow(xl);
554}
555
556static __cpuinit void xen_init_lock_cpu(int cpu)
557{
558 int irq;
559 const char *name;
560
561 name = kasprintf(GFP_KERNEL, "spinlock%d", cpu);
562 irq = bind_ipi_to_irqhandler(XEN_SPIN_UNLOCK_VECTOR,
563 cpu,
564 xen_reschedule_interrupt,
565 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
566 name,
567 NULL);
568
569 if (irq >= 0) {
570 disable_irq(irq); /* make sure it's never delivered */
571 per_cpu(lock_kicker_irq, cpu) = irq;
572 }
573
574 printk("cpu %d spinlock event irq %d\n", cpu, irq);
575}
576
577static void __init xen_init_spinlocks(void)
578{
579 pv_lock_ops.spin_is_locked = xen_spin_is_locked;
580 pv_lock_ops.spin_is_contended = xen_spin_is_contended;
581 pv_lock_ops.spin_lock = xen_spin_lock;
582 pv_lock_ops.spin_trylock = xen_spin_trylock;
583 pv_lock_ops.spin_unlock = xen_spin_unlock;
584}
585
586static const struct smp_ops xen_smp_ops __initdata = { 469static const struct smp_ops xen_smp_ops __initdata = {
587 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, 470 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
588 .smp_prepare_cpus = xen_smp_prepare_cpus, 471 .smp_prepare_cpus = xen_smp_prepare_cpus,
589 .cpu_up = xen_cpu_up,
590 .smp_cpus_done = xen_smp_cpus_done, 472 .smp_cpus_done = xen_smp_cpus_done,
591 473
474 .cpu_up = xen_cpu_up,
475 .cpu_die = xen_cpu_die,
476 .cpu_disable = xen_cpu_disable,
477 .play_dead = xen_play_dead,
478
592 .smp_send_stop = xen_smp_send_stop, 479 .smp_send_stop = xen_smp_send_stop,
593 .smp_send_reschedule = xen_smp_send_reschedule, 480 .smp_send_reschedule = xen_smp_send_reschedule,
594 481
diff --git a/arch/x86/xen/spinlock.c b/arch/x86/xen/spinlock.c
new file mode 100644
index 000000000000..dd71e3a021cd
--- /dev/null
+++ b/arch/x86/xen/spinlock.c
@@ -0,0 +1,428 @@
1/*
2 * Split spinlock implementation out into its own file, so it can be
3 * compiled in a FTRACE-compatible way.
4 */
5#include <linux/kernel_stat.h>
6#include <linux/spinlock.h>
7#include <linux/debugfs.h>
8#include <linux/log2.h>
9
10#include <asm/paravirt.h>
11
12#include <xen/interface/xen.h>
13#include <xen/events.h>
14
15#include "xen-ops.h"
16#include "debugfs.h"
17
18#ifdef CONFIG_XEN_DEBUG_FS
19static struct xen_spinlock_stats
20{
21 u64 taken;
22 u32 taken_slow;
23 u32 taken_slow_nested;
24 u32 taken_slow_pickup;
25 u32 taken_slow_spurious;
26 u32 taken_slow_irqenable;
27
28 u64 released;
29 u32 released_slow;
30 u32 released_slow_kicked;
31
32#define HISTO_BUCKETS 30
33 u32 histo_spin_total[HISTO_BUCKETS+1];
34 u32 histo_spin_spinning[HISTO_BUCKETS+1];
35 u32 histo_spin_blocked[HISTO_BUCKETS+1];
36
37 u64 time_total;
38 u64 time_spinning;
39 u64 time_blocked;
40} spinlock_stats;
41
42static u8 zero_stats;
43
44static unsigned lock_timeout = 1 << 10;
45#define TIMEOUT lock_timeout
46
47static inline void check_zero(void)
48{
49 if (unlikely(zero_stats)) {
50 memset(&spinlock_stats, 0, sizeof(spinlock_stats));
51 zero_stats = 0;
52 }
53}
54
55#define ADD_STATS(elem, val) \
56 do { check_zero(); spinlock_stats.elem += (val); } while(0)
57
58static inline u64 spin_time_start(void)
59{
60 return xen_clocksource_read();
61}
62
63static void __spin_time_accum(u64 delta, u32 *array)
64{
65 unsigned index = ilog2(delta);
66
67 check_zero();
68
69 if (index < HISTO_BUCKETS)
70 array[index]++;
71 else
72 array[HISTO_BUCKETS]++;
73}
74
75static inline void spin_time_accum_spinning(u64 start)
76{
77 u32 delta = xen_clocksource_read() - start;
78
79 __spin_time_accum(delta, spinlock_stats.histo_spin_spinning);
80 spinlock_stats.time_spinning += delta;
81}
82
83static inline void spin_time_accum_total(u64 start)
84{
85 u32 delta = xen_clocksource_read() - start;
86
87 __spin_time_accum(delta, spinlock_stats.histo_spin_total);
88 spinlock_stats.time_total += delta;
89}
90
91static inline void spin_time_accum_blocked(u64 start)
92{
93 u32 delta = xen_clocksource_read() - start;
94
95 __spin_time_accum(delta, spinlock_stats.histo_spin_blocked);
96 spinlock_stats.time_blocked += delta;
97}
98#else /* !CONFIG_XEN_DEBUG_FS */
99#define TIMEOUT (1 << 10)
100#define ADD_STATS(elem, val) do { (void)(val); } while(0)
101
102static inline u64 spin_time_start(void)
103{
104 return 0;
105}
106
107static inline void spin_time_accum_total(u64 start)
108{
109}
110static inline void spin_time_accum_spinning(u64 start)
111{
112}
113static inline void spin_time_accum_blocked(u64 start)
114{
115}
116#endif /* CONFIG_XEN_DEBUG_FS */
117
118struct xen_spinlock {
119 unsigned char lock; /* 0 -> free; 1 -> locked */
120 unsigned short spinners; /* count of waiting cpus */
121};
122
123static int xen_spin_is_locked(struct raw_spinlock *lock)
124{
125 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
126
127 return xl->lock != 0;
128}
129
130static int xen_spin_is_contended(struct raw_spinlock *lock)
131{
132 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
133
134 /* Not strictly true; this is only the count of contended
135 lock-takers entering the slow path. */
136 return xl->spinners != 0;
137}
138
139static int xen_spin_trylock(struct raw_spinlock *lock)
140{
141 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
142 u8 old = 1;
143
144 asm("xchgb %b0,%1"
145 : "+q" (old), "+m" (xl->lock) : : "memory");
146
147 return old == 0;
148}
149
150static DEFINE_PER_CPU(int, lock_kicker_irq) = -1;
151static DEFINE_PER_CPU(struct xen_spinlock *, lock_spinners);
152
153/*
154 * Mark a cpu as interested in a lock. Returns the CPU's previous
155 * lock of interest, in case we got preempted by an interrupt.
156 */
157static inline struct xen_spinlock *spinning_lock(struct xen_spinlock *xl)
158{
159 struct xen_spinlock *prev;
160
161 prev = __get_cpu_var(lock_spinners);
162 __get_cpu_var(lock_spinners) = xl;
163
164 wmb(); /* set lock of interest before count */
165
166 asm(LOCK_PREFIX " incw %0"
167 : "+m" (xl->spinners) : : "memory");
168
169 return prev;
170}
171
172/*
173 * Mark a cpu as no longer interested in a lock. Restores previous
174 * lock of interest (NULL for none).
175 */
176static inline void unspinning_lock(struct xen_spinlock *xl, struct xen_spinlock *prev)
177{
178 asm(LOCK_PREFIX " decw %0"
179 : "+m" (xl->spinners) : : "memory");
180 wmb(); /* decrement count before restoring lock */
181 __get_cpu_var(lock_spinners) = prev;
182}
183
184static noinline int xen_spin_lock_slow(struct raw_spinlock *lock, bool irq_enable)
185{
186 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
187 struct xen_spinlock *prev;
188 int irq = __get_cpu_var(lock_kicker_irq);
189 int ret;
190 unsigned long flags;
191 u64 start;
192
193 /* If kicker interrupts not initialized yet, just spin */
194 if (irq == -1)
195 return 0;
196
197 start = spin_time_start();
198
199 /* announce we're spinning */
200 prev = spinning_lock(xl);
201
202 flags = __raw_local_save_flags();
203 if (irq_enable) {
204 ADD_STATS(taken_slow_irqenable, 1);
205 raw_local_irq_enable();
206 }
207
208 ADD_STATS(taken_slow, 1);
209 ADD_STATS(taken_slow_nested, prev != NULL);
210
211 do {
212 /* clear pending */
213 xen_clear_irq_pending(irq);
214
215 /* check again make sure it didn't become free while
216 we weren't looking */
217 ret = xen_spin_trylock(lock);
218 if (ret) {
219 ADD_STATS(taken_slow_pickup, 1);
220
221 /*
222 * If we interrupted another spinlock while it
223 * was blocking, make sure it doesn't block
224 * without rechecking the lock.
225 */
226 if (prev != NULL)
227 xen_set_irq_pending(irq);
228 goto out;
229 }
230
231 /*
232 * Block until irq becomes pending. If we're
233 * interrupted at this point (after the trylock but
234 * before entering the block), then the nested lock
235 * handler guarantees that the irq will be left
236 * pending if there's any chance the lock became free;
237 * xen_poll_irq() returns immediately if the irq is
238 * pending.
239 */
240 xen_poll_irq(irq);
241 ADD_STATS(taken_slow_spurious, !xen_test_irq_pending(irq));
242 } while (!xen_test_irq_pending(irq)); /* check for spurious wakeups */
243
244 kstat_this_cpu.irqs[irq]++;
245
246out:
247 raw_local_irq_restore(flags);
248 unspinning_lock(xl, prev);
249 spin_time_accum_blocked(start);
250
251 return ret;
252}
253
254static inline void __xen_spin_lock(struct raw_spinlock *lock, bool irq_enable)
255{
256 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
257 unsigned timeout;
258 u8 oldval;
259 u64 start_spin;
260
261 ADD_STATS(taken, 1);
262
263 start_spin = spin_time_start();
264
265 do {
266 u64 start_spin_fast = spin_time_start();
267
268 timeout = TIMEOUT;
269
270 asm("1: xchgb %1,%0\n"
271 " testb %1,%1\n"
272 " jz 3f\n"
273 "2: rep;nop\n"
274 " cmpb $0,%0\n"
275 " je 1b\n"
276 " dec %2\n"
277 " jnz 2b\n"
278 "3:\n"
279 : "+m" (xl->lock), "=q" (oldval), "+r" (timeout)
280 : "1" (1)
281 : "memory");
282
283 spin_time_accum_spinning(start_spin_fast);
284
285 } while (unlikely(oldval != 0 &&
286 (TIMEOUT == ~0 || !xen_spin_lock_slow(lock, irq_enable))));
287
288 spin_time_accum_total(start_spin);
289}
290
291static void xen_spin_lock(struct raw_spinlock *lock)
292{
293 __xen_spin_lock(lock, false);
294}
295
296static void xen_spin_lock_flags(struct raw_spinlock *lock, unsigned long flags)
297{
298 __xen_spin_lock(lock, !raw_irqs_disabled_flags(flags));
299}
300
301static noinline void xen_spin_unlock_slow(struct xen_spinlock *xl)
302{
303 int cpu;
304
305 ADD_STATS(released_slow, 1);
306
307 for_each_online_cpu(cpu) {
308 /* XXX should mix up next cpu selection */
309 if (per_cpu(lock_spinners, cpu) == xl) {
310 ADD_STATS(released_slow_kicked, 1);
311 xen_send_IPI_one(cpu, XEN_SPIN_UNLOCK_VECTOR);
312 break;
313 }
314 }
315}
316
317static void xen_spin_unlock(struct raw_spinlock *lock)
318{
319 struct xen_spinlock *xl = (struct xen_spinlock *)lock;
320
321 ADD_STATS(released, 1);
322
323 smp_wmb(); /* make sure no writes get moved after unlock */
324 xl->lock = 0; /* release lock */
325
326 /* make sure unlock happens before kick */
327 barrier();
328
329 if (unlikely(xl->spinners))
330 xen_spin_unlock_slow(xl);
331}
332
333static irqreturn_t dummy_handler(int irq, void *dev_id)
334{
335 BUG();
336 return IRQ_HANDLED;
337}
338
339void __cpuinit xen_init_lock_cpu(int cpu)
340{
341 int irq;
342 const char *name;
343
344 name = kasprintf(GFP_KERNEL, "spinlock%d", cpu);
345 irq = bind_ipi_to_irqhandler(XEN_SPIN_UNLOCK_VECTOR,
346 cpu,
347 dummy_handler,
348 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
349 name,
350 NULL);
351
352 if (irq >= 0) {
353 disable_irq(irq); /* make sure it's never delivered */
354 per_cpu(lock_kicker_irq, cpu) = irq;
355 }
356
357 printk("cpu %d spinlock event irq %d\n", cpu, irq);
358}
359
360void xen_uninit_lock_cpu(int cpu)
361{
362 unbind_from_irqhandler(per_cpu(lock_kicker_irq, cpu), NULL);
363}
364
365void __init xen_init_spinlocks(void)
366{
367 pv_lock_ops.spin_is_locked = xen_spin_is_locked;
368 pv_lock_ops.spin_is_contended = xen_spin_is_contended;
369 pv_lock_ops.spin_lock = xen_spin_lock;
370 pv_lock_ops.spin_lock_flags = xen_spin_lock_flags;
371 pv_lock_ops.spin_trylock = xen_spin_trylock;
372 pv_lock_ops.spin_unlock = xen_spin_unlock;
373}
374
375#ifdef CONFIG_XEN_DEBUG_FS
376
377static struct dentry *d_spin_debug;
378
379static int __init xen_spinlock_debugfs(void)
380{
381 struct dentry *d_xen = xen_init_debugfs();
382
383 if (d_xen == NULL)
384 return -ENOMEM;
385
386 d_spin_debug = debugfs_create_dir("spinlocks", d_xen);
387
388 debugfs_create_u8("zero_stats", 0644, d_spin_debug, &zero_stats);
389
390 debugfs_create_u32("timeout", 0644, d_spin_debug, &lock_timeout);
391
392 debugfs_create_u64("taken", 0444, d_spin_debug, &spinlock_stats.taken);
393 debugfs_create_u32("taken_slow", 0444, d_spin_debug,
394 &spinlock_stats.taken_slow);
395 debugfs_create_u32("taken_slow_nested", 0444, d_spin_debug,
396 &spinlock_stats.taken_slow_nested);
397 debugfs_create_u32("taken_slow_pickup", 0444, d_spin_debug,
398 &spinlock_stats.taken_slow_pickup);
399 debugfs_create_u32("taken_slow_spurious", 0444, d_spin_debug,
400 &spinlock_stats.taken_slow_spurious);
401 debugfs_create_u32("taken_slow_irqenable", 0444, d_spin_debug,
402 &spinlock_stats.taken_slow_irqenable);
403
404 debugfs_create_u64("released", 0444, d_spin_debug, &spinlock_stats.released);
405 debugfs_create_u32("released_slow", 0444, d_spin_debug,
406 &spinlock_stats.released_slow);
407 debugfs_create_u32("released_slow_kicked", 0444, d_spin_debug,
408 &spinlock_stats.released_slow_kicked);
409
410 debugfs_create_u64("time_spinning", 0444, d_spin_debug,
411 &spinlock_stats.time_spinning);
412 debugfs_create_u64("time_blocked", 0444, d_spin_debug,
413 &spinlock_stats.time_blocked);
414 debugfs_create_u64("time_total", 0444, d_spin_debug,
415 &spinlock_stats.time_total);
416
417 xen_debugfs_create_u32_array("histo_total", 0444, d_spin_debug,
418 spinlock_stats.histo_spin_total, HISTO_BUCKETS + 1);
419 xen_debugfs_create_u32_array("histo_spinning", 0444, d_spin_debug,
420 spinlock_stats.histo_spin_spinning, HISTO_BUCKETS + 1);
421 xen_debugfs_create_u32_array("histo_blocked", 0444, d_spin_debug,
422 spinlock_stats.histo_spin_blocked, HISTO_BUCKETS + 1);
423
424 return 0;
425}
426fs_initcall(xen_spinlock_debugfs);
427
428#endif /* CONFIG_XEN_DEBUG_FS */
diff --git a/arch/x86/xen/time.c b/arch/x86/xen/time.c
index 685b77470fc3..c9f7cda48ed7 100644
--- a/arch/x86/xen/time.c
+++ b/arch/x86/xen/time.c
@@ -30,8 +30,6 @@
30#define TIMER_SLOP 100000 30#define TIMER_SLOP 100000
31#define NS_PER_TICK (1000000000LL / HZ) 31#define NS_PER_TICK (1000000000LL / HZ)
32 32
33static cycle_t xen_clocksource_read(void);
34
35/* runstate info updated by Xen */ 33/* runstate info updated by Xen */
36static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate); 34static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate);
37 35
@@ -200,20 +198,13 @@ unsigned long long xen_sched_clock(void)
200/* Get the TSC speed from Xen */ 198/* Get the TSC speed from Xen */
201unsigned long xen_tsc_khz(void) 199unsigned long xen_tsc_khz(void)
202{ 200{
203 u64 xen_khz = 1000000ULL << 32; 201 struct pvclock_vcpu_time_info *info =
204 const struct pvclock_vcpu_time_info *info =
205 &HYPERVISOR_shared_info->vcpu_info[0].time; 202 &HYPERVISOR_shared_info->vcpu_info[0].time;
206 203
207 do_div(xen_khz, info->tsc_to_system_mul); 204 return pvclock_tsc_khz(info);
208 if (info->tsc_shift < 0)
209 xen_khz <<= -info->tsc_shift;
210 else
211 xen_khz >>= info->tsc_shift;
212
213 return xen_khz;
214} 205}
215 206
216static cycle_t xen_clocksource_read(void) 207cycle_t xen_clocksource_read(void)
217{ 208{
218 struct pvclock_vcpu_time_info *src; 209 struct pvclock_vcpu_time_info *src;
219 cycle_t ret; 210 cycle_t ret;
@@ -452,6 +443,14 @@ void xen_setup_timer(int cpu)
452 setup_runstate_info(cpu); 443 setup_runstate_info(cpu);
453} 444}
454 445
446void xen_teardown_timer(int cpu)
447{
448 struct clock_event_device *evt;
449 BUG_ON(cpu == 0);
450 evt = &per_cpu(xen_clock_events, cpu);
451 unbind_from_irqhandler(evt->irq, NULL);
452}
453
455void xen_setup_cpu_clockevents(void) 454void xen_setup_cpu_clockevents(void)
456{ 455{
457 BUG_ON(preemptible()); 456 BUG_ON(preemptible());
diff --git a/arch/x86/xen/xen-asm_32.S b/arch/x86/xen/xen-asm_32.S
index 2497a30f41de..42786f59d9c0 100644
--- a/arch/x86/xen/xen-asm_32.S
+++ b/arch/x86/xen/xen-asm_32.S
@@ -298,7 +298,7 @@ check_events:
298 push %eax 298 push %eax
299 push %ecx 299 push %ecx
300 push %edx 300 push %edx
301 call force_evtchn_callback 301 call xen_force_evtchn_callback
302 pop %edx 302 pop %edx
303 pop %ecx 303 pop %ecx
304 pop %eax 304 pop %eax
diff --git a/arch/x86/xen/xen-asm_64.S b/arch/x86/xen/xen-asm_64.S
index 7f58304fafb3..05794c566e87 100644
--- a/arch/x86/xen/xen-asm_64.S
+++ b/arch/x86/xen/xen-asm_64.S
@@ -26,8 +26,15 @@
26/* Pseudo-flag used for virtual NMI, which we don't implement yet */ 26/* Pseudo-flag used for virtual NMI, which we don't implement yet */
27#define XEN_EFLAGS_NMI 0x80000000 27#define XEN_EFLAGS_NMI 0x80000000
28 28
29#if 0 29#if 1
30#include <asm/percpu.h> 30/*
31 x86-64 does not yet support direct access to percpu variables
32 via a segment override, so we just need to make sure this code
33 never gets used
34 */
35#define BUG ud2a
36#define PER_CPU_VAR(var, off) 0xdeadbeef
37#endif
31 38
32/* 39/*
33 Enable events. This clears the event mask and tests the pending 40 Enable events. This clears the event mask and tests the pending
@@ -35,6 +42,8 @@
35 events, then enter the hypervisor to get them handled. 42 events, then enter the hypervisor to get them handled.
36 */ 43 */
37ENTRY(xen_irq_enable_direct) 44ENTRY(xen_irq_enable_direct)
45 BUG
46
38 /* Unmask events */ 47 /* Unmask events */
39 movb $0, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask) 48 movb $0, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask)
40 49
@@ -58,6 +67,8 @@ ENDPATCH(xen_irq_enable_direct)
58 non-zero. 67 non-zero.
59 */ 68 */
60ENTRY(xen_irq_disable_direct) 69ENTRY(xen_irq_disable_direct)
70 BUG
71
61 movb $1, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask) 72 movb $1, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask)
62ENDPATCH(xen_irq_disable_direct) 73ENDPATCH(xen_irq_disable_direct)
63 ret 74 ret
@@ -74,6 +85,8 @@ ENDPATCH(xen_irq_disable_direct)
74 Xen and x86 use opposite senses (mask vs enable). 85 Xen and x86 use opposite senses (mask vs enable).
75 */ 86 */
76ENTRY(xen_save_fl_direct) 87ENTRY(xen_save_fl_direct)
88 BUG
89
77 testb $0xff, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask) 90 testb $0xff, PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask)
78 setz %ah 91 setz %ah
79 addb %ah,%ah 92 addb %ah,%ah
@@ -91,6 +104,8 @@ ENDPATCH(xen_save_fl_direct)
91 if so. 104 if so.
92 */ 105 */
93ENTRY(xen_restore_fl_direct) 106ENTRY(xen_restore_fl_direct)
107 BUG
108
94 testb $X86_EFLAGS_IF>>8, %ah 109 testb $X86_EFLAGS_IF>>8, %ah
95 setz PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask) 110 setz PER_CPU_VAR(xen_vcpu_info, XEN_vcpu_info_mask)
96 /* Preempt here doesn't matter because that will deal with 111 /* Preempt here doesn't matter because that will deal with
@@ -122,7 +137,7 @@ check_events:
122 push %r9 137 push %r9
123 push %r10 138 push %r10
124 push %r11 139 push %r11
125 call force_evtchn_callback 140 call xen_force_evtchn_callback
126 pop %r11 141 pop %r11
127 pop %r10 142 pop %r10
128 pop %r9 143 pop %r9
@@ -133,7 +148,6 @@ check_events:
133 pop %rcx 148 pop %rcx
134 pop %rax 149 pop %rax
135 ret 150 ret
136#endif
137 151
138ENTRY(xen_adjust_exception_frame) 152ENTRY(xen_adjust_exception_frame)
139 mov 8+0(%rsp),%rcx 153 mov 8+0(%rsp),%rcx
diff --git a/arch/x86/xen/xen-ops.h b/arch/x86/xen/xen-ops.h
index dd3c23152a2e..d7422dc2a55c 100644
--- a/arch/x86/xen/xen-ops.h
+++ b/arch/x86/xen/xen-ops.h
@@ -2,6 +2,7 @@
2#define XEN_OPS_H 2#define XEN_OPS_H
3 3
4#include <linux/init.h> 4#include <linux/init.h>
5#include <linux/clocksource.h>
5#include <linux/irqreturn.h> 6#include <linux/irqreturn.h>
6#include <xen/xen-ops.h> 7#include <xen/xen-ops.h>
7 8
@@ -31,7 +32,10 @@ void xen_vcpu_restore(void);
31 32
32void __init xen_build_dynamic_phys_to_machine(void); 33void __init xen_build_dynamic_phys_to_machine(void);
33 34
35void xen_init_irq_ops(void);
34void xen_setup_timer(int cpu); 36void xen_setup_timer(int cpu);
37void xen_teardown_timer(int cpu);
38cycle_t xen_clocksource_read(void);
35void xen_setup_cpu_clockevents(void); 39void xen_setup_cpu_clockevents(void);
36unsigned long xen_tsc_khz(void); 40unsigned long xen_tsc_khz(void);
37void __init xen_time_init(void); 41void __init xen_time_init(void);
@@ -50,6 +54,10 @@ void __init xen_setup_vcpu_info_placement(void);
50#ifdef CONFIG_SMP 54#ifdef CONFIG_SMP
51void xen_smp_init(void); 55void xen_smp_init(void);
52 56
57void __init xen_init_spinlocks(void);
58__cpuinit void xen_init_lock_cpu(int cpu);
59void xen_uninit_lock_cpu(int cpu);
60
53extern cpumask_t xen_cpu_initialized_map; 61extern cpumask_t xen_cpu_initialized_map;
54#else 62#else
55static inline void xen_smp_init(void) {} 63static inline void xen_smp_init(void) {}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-08 03:07:04 -0500