aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2008-07-31 12:43:41 -0400
committerIngo Molnar <mingo@elte.hu>2008-07-31 12:43:41 -0400
commit85e9ca333d03fbd56b9e123c8456f0d98e20faad (patch)
tree7bb15ada5f536950efa23ad60ea9eea60380ca1c /arch/x86/kernel
parenta300bec952127d9a15e666b391bb35c9aecb3002 (diff)
parent6e86841d05f371b5b9b86ce76c02aaee83352298 (diff)
Merge branch 'linus' into timers/hpet
Diffstat (limited to 'arch/x86/kernel')
-rw-r--r--arch/x86/kernel/.gitignore1
-rw-r--r--arch/x86/kernel/Makefile31
-rw-r--r--arch/x86/kernel/acpi/boot.c442
-rw-r--r--arch/x86/kernel/acpi/processor.c6
-rw-r--r--arch/x86/kernel/acpi/realmode/wakeup.S38
-rw-r--r--arch/x86/kernel/acpi/realmode/wakeup.h5
-rw-r--r--arch/x86/kernel/acpi/sleep.c36
-rw-r--r--arch/x86/kernel/alternative.c22
-rw-r--r--arch/x86/kernel/amd_iommu.c1164
-rw-r--r--arch/x86/kernel/amd_iommu_init.c1060
-rw-r--r--arch/x86/kernel/aperture_64.c314
-rw-r--r--arch/x86/kernel/apic_32.c298
-rw-r--r--arch/x86/kernel/apic_64.c94
-rw-r--r--arch/x86/kernel/apm_32.c38
-rw-r--r--arch/x86/kernel/asm-offsets_32.c2
-rw-r--r--arch/x86/kernel/asm-offsets_64.c18
-rw-r--r--arch/x86/kernel/bios_uv.c48
-rw-r--r--arch/x86/kernel/cpu/Makefile4
-rw-r--r--arch/x86/kernel/cpu/addon_cpuid_features.c17
-rw-r--r--arch/x86/kernel/cpu/amd.c44
-rw-r--r--arch/x86/kernel/cpu/amd_64.c224
-rw-r--r--arch/x86/kernel/cpu/bugs.c50
-rw-r--r--arch/x86/kernel/cpu/bugs_64.c (renamed from arch/x86/kernel/bugs_64.c)0
-rw-r--r--arch/x86/kernel/cpu/centaur_64.c35
-rw-r--r--arch/x86/kernel/cpu/common.c2
-rw-r--r--arch/x86/kernel/cpu/common_64.c670
-rw-r--r--arch/x86/kernel/cpu/cpu.h5
-rw-r--r--arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c6
-rw-r--r--arch/x86/kernel/cpu/cpufreq/cpufreq-nforce2.c44
-rw-r--r--arch/x86/kernel/cpu/cpufreq/p4-clockmod.c6
-rw-r--r--arch/x86/kernel/cpu/cpufreq/powernow-k7.h1
-rw-r--r--arch/x86/kernel/cpu/cpufreq/powernow-k8.c8
-rw-r--r--arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c149
-rw-r--r--arch/x86/kernel/cpu/cpufreq/speedstep-ich.c4
-rw-r--r--arch/x86/kernel/cpu/intel.c14
-rw-r--r--arch/x86/kernel/cpu/intel_64.c95
-rw-r--r--arch/x86/kernel/cpu/intel_cacheinfo.c10
-rw-r--r--arch/x86/kernel/cpu/mcheck/k7.c36
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_64.c54
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_amd_64.c4
-rw-r--r--arch/x86/kernel/cpu/mcheck/non-fatal.c2
-rw-r--r--arch/x86/kernel/cpu/mcheck/p4.c92
-rw-r--r--arch/x86/kernel/cpu/mcheck/therm_throt.c1
-rw-r--r--arch/x86/kernel/cpu/mtrr/generic.c38
-rw-r--r--arch/x86/kernel/cpu/mtrr/main.c905
-rw-r--r--arch/x86/kernel/cpu/mtrr/mtrr.h3
-rw-r--r--arch/x86/kernel/cpu/perfctr-watchdog.c214
-rw-r--r--arch/x86/kernel/cpu/proc.c2
-rw-r--r--arch/x86/kernel/cpuid.c31
-rw-r--r--arch/x86/kernel/e820.c1365
-rw-r--r--arch/x86/kernel/e820_32.c775
-rw-r--r--arch/x86/kernel/e820_64.c952
-rw-r--r--arch/x86/kernel/early-quirks.c46
-rw-r--r--arch/x86/kernel/early_printk.c2
-rw-r--r--arch/x86/kernel/efi.c67
-rw-r--r--arch/x86/kernel/efi_32.c8
-rw-r--r--arch/x86/kernel/efi_64.c8
-rw-r--r--arch/x86/kernel/entry_32.S162
-rw-r--r--arch/x86/kernel/entry_64.S339
-rw-r--r--arch/x86/kernel/ftrace.c141
-rw-r--r--arch/x86/kernel/genapic_64.c2
-rw-r--r--arch/x86/kernel/genapic_flat_64.c2
-rw-r--r--arch/x86/kernel/genx2apic_uv_x.c241
-rw-r--r--arch/x86/kernel/geode_32.c5
-rw-r--r--arch/x86/kernel/head.c55
-rw-r--r--arch/x86/kernel/head32.c27
-rw-r--r--arch/x86/kernel/head64.c101
-rw-r--r--arch/x86/kernel/head_32.S15
-rw-r--r--arch/x86/kernel/head_64.S102
-rw-r--r--arch/x86/kernel/hpet.c73
-rw-r--r--arch/x86/kernel/i386_ksyms_32.c9
-rw-r--r--arch/x86/kernel/i8259.c (renamed from arch/x86/kernel/i8259_32.c)136
-rw-r--r--arch/x86/kernel/i8259_64.c512
-rw-r--r--arch/x86/kernel/io_apic_32.c723
-rw-r--r--arch/x86/kernel/io_apic_64.c327
-rw-r--r--arch/x86/kernel/io_delay.c3
-rw-r--r--arch/x86/kernel/ipi.c7
-rw-r--r--arch/x86/kernel/irq_32.c253
-rw-r--r--arch/x86/kernel/irq_64.c28
-rw-r--r--arch/x86/kernel/irqinit_32.c114
-rw-r--r--arch/x86/kernel/irqinit_64.c222
-rw-r--r--arch/x86/kernel/kdebugfs.c8
-rw-r--r--arch/x86/kernel/kprobes.c7
-rw-r--r--arch/x86/kernel/kvmclock.c89
-rw-r--r--arch/x86/kernel/ldt.c12
-rw-r--r--arch/x86/kernel/machine_kexec_32.c47
-rw-r--r--arch/x86/kernel/machine_kexec_64.c8
-rw-r--r--arch/x86/kernel/microcode.c49
-rw-r--r--arch/x86/kernel/mmconf-fam10h_64.c1
-rw-r--r--arch/x86/kernel/module_64.c11
-rw-r--r--arch/x86/kernel/mpparse.c883
-rw-r--r--arch/x86/kernel/msr.c20
-rw-r--r--arch/x86/kernel/nmi.c (renamed from arch/x86/kernel/nmi_64.c)225
-rw-r--r--arch/x86/kernel/nmi_32.c467
-rw-r--r--arch/x86/kernel/numaq_32.c215
-rw-r--r--arch/x86/kernel/paravirt.c65
-rw-r--r--arch/x86/kernel/paravirt_patch_32.c4
-rw-r--r--arch/x86/kernel/paravirt_patch_64.c9
-rw-r--r--arch/x86/kernel/pci-calgary_64.c160
-rw-r--r--arch/x86/kernel/pci-dma.c194
-rw-r--r--arch/x86/kernel/pci-gart_64.c114
-rw-r--r--arch/x86/kernel/pci-nommu.c16
-rw-r--r--arch/x86/kernel/pci-swiotlb_64.c6
-rw-r--r--arch/x86/kernel/probe_roms_32.c166
-rw-r--r--arch/x86/kernel/process.c225
-rw-r--r--arch/x86/kernel/process_32.c71
-rw-r--r--arch/x86/kernel/process_64.c143
-rw-r--r--arch/x86/kernel/ptrace.c155
-rw-r--r--arch/x86/kernel/pvclock.c141
-rw-r--r--arch/x86/kernel/quirks.c56
-rw-r--r--arch/x86/kernel/reboot.c29
-rw-r--r--arch/x86/kernel/reboot_fixups_32.c4
-rw-r--r--arch/x86/kernel/relocate_kernel_32.S174
-rw-r--r--arch/x86/kernel/setup.c924
-rw-r--r--arch/x86/kernel/setup64.c287
-rw-r--r--arch/x86/kernel/setup_32.c958
-rw-r--r--arch/x86/kernel/setup_64.c1194
-rw-r--r--arch/x86/kernel/setup_percpu.c378
-rw-r--r--arch/x86/kernel/signal_32.c11
-rw-r--r--arch/x86/kernel/signal_64.c62
-rw-r--r--arch/x86/kernel/smp.c158
-rw-r--r--arch/x86/kernel/smpboot.c331
-rw-r--r--arch/x86/kernel/smpcommon.c56
-rw-r--r--arch/x86/kernel/smpcommon_32.c1
-rw-r--r--arch/x86/kernel/srat_32.c358
-rw-r--r--arch/x86/kernel/stacktrace.c2
-rw-r--r--arch/x86/kernel/step.c35
-rw-r--r--arch/x86/kernel/summit_32.c2
-rw-r--r--arch/x86/kernel/sys_i386_32.c64
-rw-r--r--arch/x86/kernel/syscall_table_32.S6
-rw-r--r--arch/x86/kernel/time_32.c7
-rw-r--r--arch/x86/kernel/time_64.c16
-rw-r--r--arch/x86/kernel/tlb_32.c2
-rw-r--r--arch/x86/kernel/tlb_64.c7
-rw-r--r--arch/x86/kernel/tlb_uv.c792
-rw-r--r--arch/x86/kernel/trampoline.c2
-rw-r--r--arch/x86/kernel/traps_32.c306
-rw-r--r--arch/x86/kernel/traps_64.c571
-rw-r--r--arch/x86/kernel/tsc.c535
-rw-r--r--arch/x86/kernel/tsc_32.c453
-rw-r--r--arch/x86/kernel/tsc_64.c357
-rw-r--r--arch/x86/kernel/visws_quirks.c707
-rw-r--r--arch/x86/kernel/vmi_32.c7
-rw-r--r--arch/x86/kernel/vmiclock_32.c7
-rw-r--r--arch/x86/kernel/vmlinux_32.lds.S15
-rw-r--r--arch/x86/kernel/vmlinux_64.lds.S18
-rw-r--r--arch/x86/kernel/vsmp_64.c3
-rw-r--r--arch/x86/kernel/vsyscall_64.c19
-rw-r--r--arch/x86/kernel/x8664_ksyms_64.c16
149 files changed, 15259 insertions, 10391 deletions
diff --git a/arch/x86/kernel/.gitignore b/arch/x86/kernel/.gitignore
index 4ea38a39aed4..08f4fd731469 100644
--- a/arch/x86/kernel/.gitignore
+++ b/arch/x86/kernel/.gitignore
@@ -1,2 +1,3 @@
1vsyscall.lds 1vsyscall.lds
2vsyscall_32.lds 2vsyscall_32.lds
3vmlinux.lds
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 5e618c3b4720..3db651fc8ec5 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -2,10 +2,17 @@
2# Makefile for the linux kernel. 2# Makefile for the linux kernel.
3# 3#
4 4
5extra-y := head_$(BITS).o head$(BITS).o init_task.o vmlinux.lds 5extra-y := head_$(BITS).o head$(BITS).o head.o init_task.o vmlinux.lds
6 6
7CPPFLAGS_vmlinux.lds += -U$(UTS_MACHINE) 7CPPFLAGS_vmlinux.lds += -U$(UTS_MACHINE)
8 8
9ifdef CONFIG_FTRACE
10# Do not profile debug and lowlevel utilities
11CFLAGS_REMOVE_tsc.o = -pg
12CFLAGS_REMOVE_rtc.o = -pg
13CFLAGS_REMOVE_paravirt.o = -pg
14endif
15
9# 16#
10# vsyscalls (which work on the user stack) should have 17# vsyscalls (which work on the user stack) should have
11# no stack-protector checks: 18# no stack-protector checks:
@@ -13,20 +20,21 @@ CPPFLAGS_vmlinux.lds += -U$(UTS_MACHINE)
13nostackp := $(call cc-option, -fno-stack-protector) 20nostackp := $(call cc-option, -fno-stack-protector)
14CFLAGS_vsyscall_64.o := $(PROFILING) -g0 $(nostackp) 21CFLAGS_vsyscall_64.o := $(PROFILING) -g0 $(nostackp)
15CFLAGS_hpet.o := $(nostackp) 22CFLAGS_hpet.o := $(nostackp)
16CFLAGS_tsc_64.o := $(nostackp) 23CFLAGS_tsc.o := $(nostackp)
17 24
18obj-y := process_$(BITS).o signal_$(BITS).o entry_$(BITS).o 25obj-y := process_$(BITS).o signal_$(BITS).o entry_$(BITS).o
19obj-y += traps_$(BITS).o irq_$(BITS).o 26obj-y += traps_$(BITS).o irq_$(BITS).o
20obj-y += time_$(BITS).o ioport.o ldt.o 27obj-y += time_$(BITS).o ioport.o ldt.o
21obj-y += setup_$(BITS).o i8259_$(BITS).o setup.o 28obj-y += setup.o i8259.o irqinit_$(BITS).o setup_percpu.o
29obj-$(CONFIG_X86_VISWS) += visws_quirks.o
30obj-$(CONFIG_X86_32) += probe_roms_32.o
22obj-$(CONFIG_X86_32) += sys_i386_32.o i386_ksyms_32.o 31obj-$(CONFIG_X86_32) += sys_i386_32.o i386_ksyms_32.o
23obj-$(CONFIG_X86_64) += sys_x86_64.o x8664_ksyms_64.o 32obj-$(CONFIG_X86_64) += sys_x86_64.o x8664_ksyms_64.o
24obj-$(CONFIG_X86_64) += syscall_64.o vsyscall_64.o setup64.o 33obj-$(CONFIG_X86_64) += syscall_64.o vsyscall_64.o
25obj-y += bootflag.o e820_$(BITS).o 34obj-y += bootflag.o e820.o
26obj-y += pci-dma.o quirks.o i8237.o topology.o kdebugfs.o 35obj-y += pci-dma.o quirks.o i8237.o topology.o kdebugfs.o
27obj-y += alternative.o i8253.o pci-nommu.o 36obj-y += alternative.o i8253.o pci-nommu.o
28obj-$(CONFIG_X86_64) += bugs_64.o 37obj-y += tsc.o io_delay.o rtc.o
29obj-y += tsc_$(BITS).o io_delay.o rtc.o
30 38
31obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o 39obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o
32obj-y += process.o 40obj-y += process.o
@@ -53,9 +61,10 @@ obj-$(CONFIG_X86_32_SMP) += smpcommon.o
53obj-$(CONFIG_X86_64_SMP) += tsc_sync.o smpcommon.o 61obj-$(CONFIG_X86_64_SMP) += tsc_sync.o smpcommon.o
54obj-$(CONFIG_X86_TRAMPOLINE) += trampoline_$(BITS).o 62obj-$(CONFIG_X86_TRAMPOLINE) += trampoline_$(BITS).o
55obj-$(CONFIG_X86_MPPARSE) += mpparse.o 63obj-$(CONFIG_X86_MPPARSE) += mpparse.o
56obj-$(CONFIG_X86_LOCAL_APIC) += apic_$(BITS).o nmi_$(BITS).o 64obj-$(CONFIG_X86_LOCAL_APIC) += apic_$(BITS).o nmi.o
57obj-$(CONFIG_X86_IO_APIC) += io_apic_$(BITS).o 65obj-$(CONFIG_X86_IO_APIC) += io_apic_$(BITS).o
58obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o 66obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o
67obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o
59obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o 68obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o
60obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o 69obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o
61obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o 70obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o
@@ -64,7 +73,6 @@ obj-$(CONFIG_X86_SUMMIT_NUMA) += summit_32.o
64obj-y += vsmp_64.o 73obj-y += vsmp_64.o
65obj-$(CONFIG_KPROBES) += kprobes.o 74obj-$(CONFIG_KPROBES) += kprobes.o
66obj-$(CONFIG_MODULES) += module_$(BITS).o 75obj-$(CONFIG_MODULES) += module_$(BITS).o
67obj-$(CONFIG_ACPI_SRAT) += srat_32.o
68obj-$(CONFIG_EFI) += efi.o efi_$(BITS).o efi_stub_$(BITS).o 76obj-$(CONFIG_EFI) += efi.o efi_$(BITS).o efi_stub_$(BITS).o
69obj-$(CONFIG_DOUBLEFAULT) += doublefault_32.o 77obj-$(CONFIG_DOUBLEFAULT) += doublefault_32.o
70obj-$(CONFIG_KGDB) += kgdb.o 78obj-$(CONFIG_KGDB) += kgdb.o
@@ -82,6 +90,7 @@ obj-$(CONFIG_VMI) += vmi_32.o vmiclock_32.o
82obj-$(CONFIG_KVM_GUEST) += kvm.o 90obj-$(CONFIG_KVM_GUEST) += kvm.o
83obj-$(CONFIG_KVM_CLOCK) += kvmclock.o 91obj-$(CONFIG_KVM_CLOCK) += kvmclock.o
84obj-$(CONFIG_PARAVIRT) += paravirt.o paravirt_patch_$(BITS).o 92obj-$(CONFIG_PARAVIRT) += paravirt.o paravirt_patch_$(BITS).o
93obj-$(CONFIG_PARAVIRT_CLOCK) += pvclock.o
85 94
86obj-$(CONFIG_PCSPKR_PLATFORM) += pcspeaker.o 95obj-$(CONFIG_PCSPKR_PLATFORM) += pcspeaker.o
87 96
@@ -93,12 +102,14 @@ obj-$(CONFIG_OLPC) += olpc.o
93### 102###
94# 64 bit specific files 103# 64 bit specific files
95ifeq ($(CONFIG_X86_64),y) 104ifeq ($(CONFIG_X86_64),y)
96 obj-y += genapic_64.o genapic_flat_64.o genx2apic_uv_x.o 105 obj-y += genapic_64.o genapic_flat_64.o genx2apic_uv_x.o tlb_uv.o
106 obj-y += bios_uv.o
97 obj-$(CONFIG_X86_PM_TIMER) += pmtimer_64.o 107 obj-$(CONFIG_X86_PM_TIMER) += pmtimer_64.o
98 obj-$(CONFIG_AUDIT) += audit_64.o 108 obj-$(CONFIG_AUDIT) += audit_64.o
99 109
100 obj-$(CONFIG_GART_IOMMU) += pci-gart_64.o aperture_64.o 110 obj-$(CONFIG_GART_IOMMU) += pci-gart_64.o aperture_64.o
101 obj-$(CONFIG_CALGARY_IOMMU) += pci-calgary_64.o tce_64.o 111 obj-$(CONFIG_CALGARY_IOMMU) += pci-calgary_64.o tce_64.o
112 obj-$(CONFIG_AMD_IOMMU) += amd_iommu_init.o amd_iommu.o
102 obj-$(CONFIG_SWIOTLB) += pci-swiotlb_64.o 113 obj-$(CONFIG_SWIOTLB) += pci-swiotlb_64.o
103 114
104 obj-$(CONFIG_PCI_MMCONFIG) += mmconf-fam10h_64.o 115 obj-$(CONFIG_PCI_MMCONFIG) += mmconf-fam10h_64.o
diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 33c5216fd3e1..fa88a1d71290 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -37,6 +37,7 @@
37#include <asm/pgtable.h> 37#include <asm/pgtable.h>
38#include <asm/io_apic.h> 38#include <asm/io_apic.h>
39#include <asm/apic.h> 39#include <asm/apic.h>
40#include <asm/genapic.h>
40#include <asm/io.h> 41#include <asm/io.h>
41#include <asm/mpspec.h> 42#include <asm/mpspec.h>
42#include <asm/smp.h> 43#include <asm/smp.h>
@@ -106,21 +107,6 @@ static u64 acpi_lapic_addr __initdata = APIC_DEFAULT_PHYS_BASE;
106 */ 107 */
107enum acpi_irq_model_id acpi_irq_model = ACPI_IRQ_MODEL_PIC; 108enum acpi_irq_model_id acpi_irq_model = ACPI_IRQ_MODEL_PIC;
108 109
109#ifdef CONFIG_X86_64
110
111/* rely on all ACPI tables being in the direct mapping */
112char *__init __acpi_map_table(unsigned long phys_addr, unsigned long size)
113{
114 if (!phys_addr || !size)
115 return NULL;
116
117 if (phys_addr+size <= (max_pfn_mapped << PAGE_SHIFT) + PAGE_SIZE)
118 return __va(phys_addr);
119
120 return NULL;
121}
122
123#else
124 110
125/* 111/*
126 * Temporarily use the virtual area starting from FIX_IO_APIC_BASE_END, 112 * Temporarily use the virtual area starting from FIX_IO_APIC_BASE_END,
@@ -139,11 +125,15 @@ char *__init __acpi_map_table(unsigned long phys, unsigned long size)
139 unsigned long base, offset, mapped_size; 125 unsigned long base, offset, mapped_size;
140 int idx; 126 int idx;
141 127
142 if (phys + size < 8 * 1024 * 1024) 128 if (!phys || !size)
129 return NULL;
130
131 if (phys+size <= (max_low_pfn_mapped << PAGE_SHIFT))
143 return __va(phys); 132 return __va(phys);
144 133
145 offset = phys & (PAGE_SIZE - 1); 134 offset = phys & (PAGE_SIZE - 1);
146 mapped_size = PAGE_SIZE - offset; 135 mapped_size = PAGE_SIZE - offset;
136 clear_fixmap(FIX_ACPI_END);
147 set_fixmap(FIX_ACPI_END, phys); 137 set_fixmap(FIX_ACPI_END, phys);
148 base = fix_to_virt(FIX_ACPI_END); 138 base = fix_to_virt(FIX_ACPI_END);
149 139
@@ -155,13 +145,13 @@ char *__init __acpi_map_table(unsigned long phys, unsigned long size)
155 if (--idx < FIX_ACPI_BEGIN) 145 if (--idx < FIX_ACPI_BEGIN)
156 return NULL; /* cannot handle this */ 146 return NULL; /* cannot handle this */
157 phys += PAGE_SIZE; 147 phys += PAGE_SIZE;
148 clear_fixmap(idx);
158 set_fixmap(idx, phys); 149 set_fixmap(idx, phys);
159 mapped_size += PAGE_SIZE; 150 mapped_size += PAGE_SIZE;
160 } 151 }
161 152
162 return ((unsigned char *)base + offset); 153 return ((unsigned char *)base + offset);
163} 154}
164#endif
165 155
166#ifdef CONFIG_PCI_MMCONFIG 156#ifdef CONFIG_PCI_MMCONFIG
167/* The physical address of the MMCONFIG aperture. Set from ACPI tables. */ 157/* The physical address of the MMCONFIG aperture. Set from ACPI tables. */
@@ -338,8 +328,6 @@ acpi_parse_lapic_nmi(struct acpi_subtable_header * header, const unsigned long e
338 328
339#ifdef CONFIG_X86_IO_APIC 329#ifdef CONFIG_X86_IO_APIC
340 330
341struct mp_ioapic_routing mp_ioapic_routing[MAX_IO_APICS];
342
343static int __init 331static int __init
344acpi_parse_ioapic(struct acpi_subtable_header * header, const unsigned long end) 332acpi_parse_ioapic(struct acpi_subtable_header * header, const unsigned long end)
345{ 333{
@@ -514,8 +502,6 @@ int acpi_register_gsi(u32 gsi, int triggering, int polarity)
514 * Make sure all (legacy) PCI IRQs are set as level-triggered. 502 * Make sure all (legacy) PCI IRQs are set as level-triggered.
515 */ 503 */
516 if (acpi_irq_model == ACPI_IRQ_MODEL_PIC) { 504 if (acpi_irq_model == ACPI_IRQ_MODEL_PIC) {
517 extern void eisa_set_level_irq(unsigned int irq);
518
519 if (triggering == ACPI_LEVEL_SENSITIVE) 505 if (triggering == ACPI_LEVEL_SENSITIVE)
520 eisa_set_level_irq(gsi); 506 eisa_set_level_irq(gsi);
521 } 507 }
@@ -860,6 +846,364 @@ static int __init acpi_parse_madt_lapic_entries(void)
860#endif /* CONFIG_X86_LOCAL_APIC */ 846#endif /* CONFIG_X86_LOCAL_APIC */
861 847
862#ifdef CONFIG_X86_IO_APIC 848#ifdef CONFIG_X86_IO_APIC
849#define MP_ISA_BUS 0
850
851#ifdef CONFIG_X86_ES7000
852extern int es7000_plat;
853#endif
854
855static struct {
856 int apic_id;
857 int gsi_base;
858 int gsi_end;
859 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
860} mp_ioapic_routing[MAX_IO_APICS];
861
862static int mp_find_ioapic(int gsi)
863{
864 int i = 0;
865
866 /* Find the IOAPIC that manages this GSI. */
867 for (i = 0; i < nr_ioapics; i++) {
868 if ((gsi >= mp_ioapic_routing[i].gsi_base)
869 && (gsi <= mp_ioapic_routing[i].gsi_end))
870 return i;
871 }
872
873 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
874 return -1;
875}
876
877static u8 __init uniq_ioapic_id(u8 id)
878{
879#ifdef CONFIG_X86_32
880 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
881 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
882 return io_apic_get_unique_id(nr_ioapics, id);
883 else
884 return id;
885#else
886 int i;
887 DECLARE_BITMAP(used, 256);
888 bitmap_zero(used, 256);
889 for (i = 0; i < nr_ioapics; i++) {
890 struct mp_config_ioapic *ia = &mp_ioapics[i];
891 __set_bit(ia->mp_apicid, used);
892 }
893 if (!test_bit(id, used))
894 return id;
895 return find_first_zero_bit(used, 256);
896#endif
897}
898
899static int bad_ioapic(unsigned long address)
900{
901 if (nr_ioapics >= MAX_IO_APICS) {
902 printk(KERN_ERR "ERROR: Max # of I/O APICs (%d) exceeded "
903 "(found %d)\n", MAX_IO_APICS, nr_ioapics);
904 panic("Recompile kernel with bigger MAX_IO_APICS!\n");
905 }
906 if (!address) {
907 printk(KERN_ERR "WARNING: Bogus (zero) I/O APIC address"
908 " found in table, skipping!\n");
909 return 1;
910 }
911 return 0;
912}
913
914void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
915{
916 int idx = 0;
917
918 if (bad_ioapic(address))
919 return;
920
921 idx = nr_ioapics;
922
923 mp_ioapics[idx].mp_type = MP_IOAPIC;
924 mp_ioapics[idx].mp_flags = MPC_APIC_USABLE;
925 mp_ioapics[idx].mp_apicaddr = address;
926
927 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
928 mp_ioapics[idx].mp_apicid = uniq_ioapic_id(id);
929#ifdef CONFIG_X86_32
930 mp_ioapics[idx].mp_apicver = io_apic_get_version(idx);
931#else
932 mp_ioapics[idx].mp_apicver = 0;
933#endif
934 /*
935 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
936 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
937 */
938 mp_ioapic_routing[idx].apic_id = mp_ioapics[idx].mp_apicid;
939 mp_ioapic_routing[idx].gsi_base = gsi_base;
940 mp_ioapic_routing[idx].gsi_end = gsi_base +
941 io_apic_get_redir_entries(idx);
942
943 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%lx, "
944 "GSI %d-%d\n", idx, mp_ioapics[idx].mp_apicid,
945 mp_ioapics[idx].mp_apicver, mp_ioapics[idx].mp_apicaddr,
946 mp_ioapic_routing[idx].gsi_base, mp_ioapic_routing[idx].gsi_end);
947
948 nr_ioapics++;
949}
950
951static void assign_to_mp_irq(struct mp_config_intsrc *m,
952 struct mp_config_intsrc *mp_irq)
953{
954 memcpy(mp_irq, m, sizeof(struct mp_config_intsrc));
955}
956
957static int mp_irq_cmp(struct mp_config_intsrc *mp_irq,
958 struct mp_config_intsrc *m)
959{
960 return memcmp(mp_irq, m, sizeof(struct mp_config_intsrc));
961}
962
963static void save_mp_irq(struct mp_config_intsrc *m)
964{
965 int i;
966
967 for (i = 0; i < mp_irq_entries; i++) {
968 if (!mp_irq_cmp(&mp_irqs[i], m))
969 return;
970 }
971
972 assign_to_mp_irq(m, &mp_irqs[mp_irq_entries]);
973 if (++mp_irq_entries == MAX_IRQ_SOURCES)
974 panic("Max # of irq sources exceeded!!\n");
975}
976
977void __init mp_override_legacy_irq(u8 bus_irq, u8 polarity, u8 trigger, u32 gsi)
978{
979 int ioapic;
980 int pin;
981 struct mp_config_intsrc mp_irq;
982
983 /*
984 * Convert 'gsi' to 'ioapic.pin'.
985 */
986 ioapic = mp_find_ioapic(gsi);
987 if (ioapic < 0)
988 return;
989 pin = gsi - mp_ioapic_routing[ioapic].gsi_base;
990
991 /*
992 * TBD: This check is for faulty timer entries, where the override
993 * erroneously sets the trigger to level, resulting in a HUGE
994 * increase of timer interrupts!
995 */
996 if ((bus_irq == 0) && (trigger == 3))
997 trigger = 1;
998
999 mp_irq.mp_type = MP_INTSRC;
1000 mp_irq.mp_irqtype = mp_INT;
1001 mp_irq.mp_irqflag = (trigger << 2) | polarity;
1002 mp_irq.mp_srcbus = MP_ISA_BUS;
1003 mp_irq.mp_srcbusirq = bus_irq; /* IRQ */
1004 mp_irq.mp_dstapic = mp_ioapics[ioapic].mp_apicid; /* APIC ID */
1005 mp_irq.mp_dstirq = pin; /* INTIN# */
1006
1007 save_mp_irq(&mp_irq);
1008}
1009
1010void __init mp_config_acpi_legacy_irqs(void)
1011{
1012 int i;
1013 int ioapic;
1014 unsigned int dstapic;
1015 struct mp_config_intsrc mp_irq;
1016
1017#if defined (CONFIG_MCA) || defined (CONFIG_EISA)
1018 /*
1019 * Fabricate the legacy ISA bus (bus #31).
1020 */
1021 mp_bus_id_to_type[MP_ISA_BUS] = MP_BUS_ISA;
1022#endif
1023 set_bit(MP_ISA_BUS, mp_bus_not_pci);
1024 pr_debug("Bus #%d is ISA\n", MP_ISA_BUS);
1025
1026#ifdef CONFIG_X86_ES7000
1027 /*
1028 * Older generations of ES7000 have no legacy identity mappings
1029 */
1030 if (es7000_plat == 1)
1031 return;
1032#endif
1033
1034 /*
1035 * Locate the IOAPIC that manages the ISA IRQs (0-15).
1036 */
1037 ioapic = mp_find_ioapic(0);
1038 if (ioapic < 0)
1039 return;
1040 dstapic = mp_ioapics[ioapic].mp_apicid;
1041
1042 /*
1043 * Use the default configuration for the IRQs 0-15. Unless
1044 * overridden by (MADT) interrupt source override entries.
1045 */
1046 for (i = 0; i < 16; i++) {
1047 int idx;
1048
1049 for (idx = 0; idx < mp_irq_entries; idx++) {
1050 struct mp_config_intsrc *irq = mp_irqs + idx;
1051
1052 /* Do we already have a mapping for this ISA IRQ? */
1053 if (irq->mp_srcbus == MP_ISA_BUS
1054 && irq->mp_srcbusirq == i)
1055 break;
1056
1057 /* Do we already have a mapping for this IOAPIC pin */
1058 if (irq->mp_dstapic == dstapic &&
1059 irq->mp_dstirq == i)
1060 break;
1061 }
1062
1063 if (idx != mp_irq_entries) {
1064 printk(KERN_DEBUG "ACPI: IRQ%d used by override.\n", i);
1065 continue; /* IRQ already used */
1066 }
1067
1068 mp_irq.mp_type = MP_INTSRC;
1069 mp_irq.mp_irqflag = 0; /* Conforming */
1070 mp_irq.mp_srcbus = MP_ISA_BUS;
1071 mp_irq.mp_dstapic = dstapic;
1072 mp_irq.mp_irqtype = mp_INT;
1073 mp_irq.mp_srcbusirq = i; /* Identity mapped */
1074 mp_irq.mp_dstirq = i;
1075
1076 save_mp_irq(&mp_irq);
1077 }
1078}
1079
1080int mp_register_gsi(u32 gsi, int triggering, int polarity)
1081{
1082 int ioapic;
1083 int ioapic_pin;
1084#ifdef CONFIG_X86_32
1085#define MAX_GSI_NUM 4096
1086#define IRQ_COMPRESSION_START 64
1087
1088 static int pci_irq = IRQ_COMPRESSION_START;
1089 /*
1090 * Mapping between Global System Interrupts, which
1091 * represent all possible interrupts, and IRQs
1092 * assigned to actual devices.
1093 */
1094 static int gsi_to_irq[MAX_GSI_NUM];
1095#else
1096
1097 if (acpi_irq_model != ACPI_IRQ_MODEL_IOAPIC)
1098 return gsi;
1099#endif
1100
1101 /* Don't set up the ACPI SCI because it's already set up */
1102 if (acpi_gbl_FADT.sci_interrupt == gsi)
1103 return gsi;
1104
1105 ioapic = mp_find_ioapic(gsi);
1106 if (ioapic < 0) {
1107 printk(KERN_WARNING "No IOAPIC for GSI %u\n", gsi);
1108 return gsi;
1109 }
1110
1111 ioapic_pin = gsi - mp_ioapic_routing[ioapic].gsi_base;
1112
1113#ifdef CONFIG_X86_32
1114 if (ioapic_renumber_irq)
1115 gsi = ioapic_renumber_irq(ioapic, gsi);
1116#endif
1117
1118 /*
1119 * Avoid pin reprogramming. PRTs typically include entries
1120 * with redundant pin->gsi mappings (but unique PCI devices);
1121 * we only program the IOAPIC on the first.
1122 */
1123 if (ioapic_pin > MP_MAX_IOAPIC_PIN) {
1124 printk(KERN_ERR "Invalid reference to IOAPIC pin "
1125 "%d-%d\n", mp_ioapic_routing[ioapic].apic_id,
1126 ioapic_pin);
1127 return gsi;
1128 }
1129 if (test_bit(ioapic_pin, mp_ioapic_routing[ioapic].pin_programmed)) {
1130 pr_debug(KERN_DEBUG "Pin %d-%d already programmed\n",
1131 mp_ioapic_routing[ioapic].apic_id, ioapic_pin);
1132#ifdef CONFIG_X86_32
1133 return (gsi < IRQ_COMPRESSION_START ? gsi : gsi_to_irq[gsi]);
1134#else
1135 return gsi;
1136#endif
1137 }
1138
1139 set_bit(ioapic_pin, mp_ioapic_routing[ioapic].pin_programmed);
1140#ifdef CONFIG_X86_32
1141 /*
1142 * For GSI >= 64, use IRQ compression
1143 */
1144 if ((gsi >= IRQ_COMPRESSION_START)
1145 && (triggering == ACPI_LEVEL_SENSITIVE)) {
1146 /*
1147 * For PCI devices assign IRQs in order, avoiding gaps
1148 * due to unused I/O APIC pins.
1149 */
1150 int irq = gsi;
1151 if (gsi < MAX_GSI_NUM) {
1152 /*
1153 * Retain the VIA chipset work-around (gsi > 15), but
1154 * avoid a problem where the 8254 timer (IRQ0) is setup
1155 * via an override (so it's not on pin 0 of the ioapic),
1156 * and at the same time, the pin 0 interrupt is a PCI
1157 * type. The gsi > 15 test could cause these two pins
1158 * to be shared as IRQ0, and they are not shareable.
1159 * So test for this condition, and if necessary, avoid
1160 * the pin collision.
1161 */
1162 gsi = pci_irq++;
1163 /*
1164 * Don't assign IRQ used by ACPI SCI
1165 */
1166 if (gsi == acpi_gbl_FADT.sci_interrupt)
1167 gsi = pci_irq++;
1168 gsi_to_irq[irq] = gsi;
1169 } else {
1170 printk(KERN_ERR "GSI %u is too high\n", gsi);
1171 return gsi;
1172 }
1173 }
1174#endif
1175 io_apic_set_pci_routing(ioapic, ioapic_pin, gsi,
1176 triggering == ACPI_EDGE_SENSITIVE ? 0 : 1,
1177 polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
1178 return gsi;
1179}
1180
1181int mp_config_acpi_gsi(unsigned char number, unsigned int devfn, u8 pin,
1182 u32 gsi, int triggering, int polarity)
1183{
1184#ifdef CONFIG_X86_MPPARSE
1185 struct mp_config_intsrc mp_irq;
1186 int ioapic;
1187
1188 if (!acpi_ioapic)
1189 return 0;
1190
1191 /* print the entry should happen on mptable identically */
1192 mp_irq.mp_type = MP_INTSRC;
1193 mp_irq.mp_irqtype = mp_INT;
1194 mp_irq.mp_irqflag = (triggering == ACPI_EDGE_SENSITIVE ? 4 : 0x0c) |
1195 (polarity == ACPI_ACTIVE_HIGH ? 1 : 3);
1196 mp_irq.mp_srcbus = number;
1197 mp_irq.mp_srcbusirq = (((devfn >> 3) & 0x1f) << 2) | ((pin - 1) & 3);
1198 ioapic = mp_find_ioapic(gsi);
1199 mp_irq.mp_dstapic = mp_ioapic_routing[ioapic].apic_id;
1200 mp_irq.mp_dstirq = gsi - mp_ioapic_routing[ioapic].gsi_base;
1201
1202 save_mp_irq(&mp_irq);
1203#endif
1204 return 0;
1205}
1206
863/* 1207/*
864 * Parse IOAPIC related entries in MADT 1208 * Parse IOAPIC related entries in MADT
865 * returns 0 on success, < 0 on error 1209 * returns 0 on success, < 0 on error
@@ -1009,8 +1353,6 @@ static void __init acpi_process_madt(void)
1009 return; 1353 return;
1010} 1354}
1011 1355
1012#ifdef __i386__
1013
1014static int __init disable_acpi_irq(const struct dmi_system_id *d) 1356static int __init disable_acpi_irq(const struct dmi_system_id *d)
1015{ 1357{
1016 if (!acpi_force) { 1358 if (!acpi_force) {
@@ -1061,6 +1403,16 @@ static int __init force_acpi_ht(const struct dmi_system_id *d)
1061} 1403}
1062 1404
1063/* 1405/*
1406 * Force ignoring BIOS IRQ0 pin2 override
1407 */
1408static int __init dmi_ignore_irq0_timer_override(const struct dmi_system_id *d)
1409{
1410 pr_notice("%s detected: Ignoring BIOS IRQ0 pin2 override\n", d->ident);
1411 acpi_skip_timer_override = 1;
1412 return 0;
1413}
1414
1415/*
1064 * If your system is blacklisted here, but you find that acpi=force 1416 * If your system is blacklisted here, but you find that acpi=force
1065 * works for you, please contact acpi-devel@sourceforge.net 1417 * works for you, please contact acpi-devel@sourceforge.net
1066 */ 1418 */
@@ -1227,11 +1579,35 @@ static struct dmi_system_id __initdata acpi_dmi_table[] = {
1227 DMI_MATCH(DMI_PRODUCT_NAME, "TravelMate 360"), 1579 DMI_MATCH(DMI_PRODUCT_NAME, "TravelMate 360"),
1228 }, 1580 },
1229 }, 1581 },
1582 /*
1583 * HP laptops which use a DSDT reporting as HP/SB400/10000,
1584 * which includes some code which overrides all temperature
1585 * trip points to 16C if the INTIN2 input of the I/O APIC
1586 * is enabled. This input is incorrectly designated the
1587 * ISA IRQ 0 via an interrupt source override even though
1588 * it is wired to the output of the master 8259A and INTIN0
1589 * is not connected at all. Force ignoring BIOS IRQ0 pin2
1590 * override in that cases.
1591 */
1592 {
1593 .callback = dmi_ignore_irq0_timer_override,
1594 .ident = "HP NX6125 laptop",
1595 .matches = {
1596 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1597 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq nx6125"),
1598 },
1599 },
1600 {
1601 .callback = dmi_ignore_irq0_timer_override,
1602 .ident = "HP NX6325 laptop",
1603 .matches = {
1604 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
1605 DMI_MATCH(DMI_PRODUCT_NAME, "HP Compaq nx6325"),
1606 },
1607 },
1230 {} 1608 {}
1231}; 1609};
1232 1610
1233#endif /* __i386__ */
1234
1235/* 1611/*
1236 * acpi_boot_table_init() and acpi_boot_init() 1612 * acpi_boot_table_init() and acpi_boot_init()
1237 * called from setup_arch(), always. 1613 * called from setup_arch(), always.
@@ -1259,9 +1635,7 @@ int __init acpi_boot_table_init(void)
1259{ 1635{
1260 int error; 1636 int error;
1261 1637
1262#ifdef __i386__
1263 dmi_check_system(acpi_dmi_table); 1638 dmi_check_system(acpi_dmi_table);
1264#endif
1265 1639
1266 /* 1640 /*
1267 * If acpi_disabled, bail out 1641 * If acpi_disabled, bail out
@@ -1386,6 +1760,20 @@ static int __init parse_pci(char *arg)
1386} 1760}
1387early_param("pci", parse_pci); 1761early_param("pci", parse_pci);
1388 1762
1763int __init acpi_mps_check(void)
1764{
1765#if defined(CONFIG_X86_LOCAL_APIC) && !defined(CONFIG_X86_MPPARSE)
1766/* mptable code is not built-in*/
1767 if (acpi_disabled || acpi_noirq) {
1768 printk(KERN_WARNING "MPS support code is not built-in.\n"
1769 "Using acpi=off or acpi=noirq or pci=noacpi "
1770 "may have problem\n");
1771 return 1;
1772 }
1773#endif
1774 return 0;
1775}
1776
1389#ifdef CONFIG_X86_IO_APIC 1777#ifdef CONFIG_X86_IO_APIC
1390static int __init parse_acpi_skip_timer_override(char *arg) 1778static int __init parse_acpi_skip_timer_override(char *arg)
1391{ 1779{
diff --git a/arch/x86/kernel/acpi/processor.c b/arch/x86/kernel/acpi/processor.c
index de2d2e4ebad9..7c074eec39fb 100644
--- a/arch/x86/kernel/acpi/processor.c
+++ b/arch/x86/kernel/acpi/processor.c
@@ -56,6 +56,12 @@ static void init_intel_pdc(struct acpi_processor *pr, struct cpuinfo_x86 *c)
56 if (cpu_has(c, X86_FEATURE_ACPI)) 56 if (cpu_has(c, X86_FEATURE_ACPI))
57 buf[2] |= ACPI_PDC_T_FFH; 57 buf[2] |= ACPI_PDC_T_FFH;
58 58
59 /*
60 * If mwait/monitor is unsupported, C2/C3_FFH will be disabled
61 */
62 if (!cpu_has(c, X86_FEATURE_MWAIT))
63 buf[2] &= ~(ACPI_PDC_C_C2C3_FFH);
64
59 obj->type = ACPI_TYPE_BUFFER; 65 obj->type = ACPI_TYPE_BUFFER;
60 obj->buffer.length = 12; 66 obj->buffer.length = 12;
61 obj->buffer.pointer = (u8 *) buf; 67 obj->buffer.pointer = (u8 *) buf;
diff --git a/arch/x86/kernel/acpi/realmode/wakeup.S b/arch/x86/kernel/acpi/realmode/wakeup.S
index f9b77fb37e5b..3355973b12ac 100644
--- a/arch/x86/kernel/acpi/realmode/wakeup.S
+++ b/arch/x86/kernel/acpi/realmode/wakeup.S
@@ -5,6 +5,7 @@
5#include <asm/msr-index.h> 5#include <asm/msr-index.h>
6#include <asm/page.h> 6#include <asm/page.h>
7#include <asm/pgtable.h> 7#include <asm/pgtable.h>
8#include <asm/processor-flags.h>
8 9
9 .code16 10 .code16
10 .section ".header", "a" 11 .section ".header", "a"
@@ -24,6 +25,11 @@ pmode_gdt: .quad 0
24realmode_flags: .long 0 25realmode_flags: .long 0
25real_magic: .long 0 26real_magic: .long 0
26trampoline_segment: .word 0 27trampoline_segment: .word 0
28_pad1: .byte 0
29wakeup_jmp: .byte 0xea /* ljmpw */
30wakeup_jmp_off: .word 3f
31wakeup_jmp_seg: .word 0
32wakeup_gdt: .quad 0, 0, 0
27signature: .long 0x51ee1111 33signature: .long 0x51ee1111
28 34
29 .text 35 .text
@@ -34,11 +40,34 @@ _start:
34 cli 40 cli
35 cld 41 cld
36 42
43 /* Apparently some dimwit BIOS programmers don't know how to
44 program a PM to RM transition, and we might end up here with
45 junk in the data segment descriptor registers. The only way
46 to repair that is to go into PM and fix it ourselves... */
47 movw $16, %cx
48 lgdtl %cs:wakeup_gdt
49 movl %cr0, %eax
50 orb $X86_CR0_PE, %al
51 movl %eax, %cr0
52 jmp 1f
531: ljmpw $8, $2f
542:
55 movw %cx, %ds
56 movw %cx, %es
57 movw %cx, %ss
58 movw %cx, %fs
59 movw %cx, %gs
60
61 andb $~X86_CR0_PE, %al
62 movl %eax, %cr0
63 jmp wakeup_jmp
643:
37 /* Set up segments */ 65 /* Set up segments */
38 movw %cs, %ax 66 movw %cs, %ax
39 movw %ax, %ds 67 movw %ax, %ds
40 movw %ax, %es 68 movw %ax, %es
41 movw %ax, %ss 69 movw %ax, %ss
70 lidtl wakeup_idt
42 71
43 movl $wakeup_stack_end, %esp 72 movl $wakeup_stack_end, %esp
44 73
@@ -98,7 +127,14 @@ bogus_real_magic:
98 jmp 1b 127 jmp 1b
99 128
100 .data 129 .data
101 .balign 4 130 .balign 8
131
132 /* This is the standard real-mode IDT */
133wakeup_idt:
134 .word 0xffff /* limit */
135 .long 0 /* address */
136 .word 0
137
102 .globl HEAP, heap_end 138 .globl HEAP, heap_end
103HEAP: 139HEAP:
104 .long wakeup_heap 140 .long wakeup_heap
diff --git a/arch/x86/kernel/acpi/realmode/wakeup.h b/arch/x86/kernel/acpi/realmode/wakeup.h
index ef8166fe8020..69d38d0b2b64 100644
--- a/arch/x86/kernel/acpi/realmode/wakeup.h
+++ b/arch/x86/kernel/acpi/realmode/wakeup.h
@@ -24,6 +24,11 @@ struct wakeup_header {
24 u32 realmode_flags; 24 u32 realmode_flags;
25 u32 real_magic; 25 u32 real_magic;
26 u16 trampoline_segment; /* segment with trampoline code, 64-bit only */ 26 u16 trampoline_segment; /* segment with trampoline code, 64-bit only */
27 u8 _pad1;
28 u8 wakeup_jmp;
29 u16 wakeup_jmp_off;
30 u16 wakeup_jmp_seg;
31 u64 wakeup_gdt[3];
27 u32 signature; /* To check we have correct structure */ 32 u32 signature; /* To check we have correct structure */
28} __attribute__((__packed__)); 33} __attribute__((__packed__));
29 34
diff --git a/arch/x86/kernel/acpi/sleep.c b/arch/x86/kernel/acpi/sleep.c
index afc25ee9964b..fa2161d5003b 100644
--- a/arch/x86/kernel/acpi/sleep.c
+++ b/arch/x86/kernel/acpi/sleep.c
@@ -9,6 +9,7 @@
9#include <linux/bootmem.h> 9#include <linux/bootmem.h>
10#include <linux/dmi.h> 10#include <linux/dmi.h>
11#include <linux/cpumask.h> 11#include <linux/cpumask.h>
12#include <asm/segment.h>
12 13
13#include "realmode/wakeup.h" 14#include "realmode/wakeup.h"
14#include "sleep.h" 15#include "sleep.h"
@@ -50,6 +51,29 @@ int acpi_save_state_mem(void)
50 51
51 header->video_mode = saved_video_mode; 52 header->video_mode = saved_video_mode;
52 53
54 header->wakeup_jmp_seg = acpi_wakeup_address >> 4;
55
56 /*
57 * Set up the wakeup GDT. We set these up as Big Real Mode,
58 * that is, with limits set to 4 GB. At least the Lenovo
59 * Thinkpad X61 is known to need this for the video BIOS
60 * initialization quirk to work; this is likely to also
61 * be the case for other laptops or integrated video devices.
62 */
63
64 /* GDT[0]: GDT self-pointer */
65 header->wakeup_gdt[0] =
66 (u64)(sizeof(header->wakeup_gdt) - 1) +
67 ((u64)(acpi_wakeup_address +
68 ((char *)&header->wakeup_gdt - (char *)acpi_realmode))
69 << 16);
70 /* GDT[1]: big real mode-like code segment */
71 header->wakeup_gdt[1] =
72 GDT_ENTRY(0x809b, acpi_wakeup_address, 0xfffff);
73 /* GDT[2]: big real mode-like data segment */
74 header->wakeup_gdt[2] =
75 GDT_ENTRY(0x8093, acpi_wakeup_address, 0xfffff);
76
53#ifndef CONFIG_64BIT 77#ifndef CONFIG_64BIT
54 store_gdt((struct desc_ptr *)&header->pmode_gdt); 78 store_gdt((struct desc_ptr *)&header->pmode_gdt);
55 79
@@ -72,7 +96,9 @@ int acpi_save_state_mem(void)
72 saved_magic = 0x12345678; 96 saved_magic = 0x12345678;
73#else /* CONFIG_64BIT */ 97#else /* CONFIG_64BIT */
74 header->trampoline_segment = setup_trampoline() >> 4; 98 header->trampoline_segment = setup_trampoline() >> 4;
75 init_rsp = (unsigned long)temp_stack + 4096; 99#ifdef CONFIG_SMP
100 stack_start.sp = temp_stack + 4096;
101#endif
76 initial_code = (unsigned long)wakeup_long64; 102 initial_code = (unsigned long)wakeup_long64;
77 saved_magic = 0x123456789abcdef0; 103 saved_magic = 0x123456789abcdef0;
78#endif /* CONFIG_64BIT */ 104#endif /* CONFIG_64BIT */
@@ -111,7 +137,7 @@ void __init acpi_reserve_bootmem(void)
111 return; 137 return;
112 } 138 }
113 139
114 acpi_wakeup_address = acpi_realmode; 140 acpi_wakeup_address = virt_to_phys((void *)acpi_realmode);
115} 141}
116 142
117 143
@@ -124,6 +150,12 @@ static int __init acpi_sleep_setup(char *str)
124 acpi_realmode_flags |= 2; 150 acpi_realmode_flags |= 2;
125 if (strncmp(str, "s3_beep", 7) == 0) 151 if (strncmp(str, "s3_beep", 7) == 0)
126 acpi_realmode_flags |= 4; 152 acpi_realmode_flags |= 4;
153#ifdef CONFIG_HIBERNATION
154 if (strncmp(str, "s4_nohwsig", 10) == 0)
155 acpi_no_s4_hw_signature();
156#endif
157 if (strncmp(str, "old_ordering", 12) == 0)
158 acpi_old_suspend_ordering();
127 str = strchr(str, ','); 159 str = strchr(str, ',');
128 if (str != NULL) 160 if (str != NULL)
129 str += strspn(str, ", \t"); 161 str += strspn(str, ", \t");
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index 65c7857a90dd..2763cb37b553 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -1,6 +1,6 @@
1#include <linux/module.h> 1#include <linux/module.h>
2#include <linux/sched.h> 2#include <linux/sched.h>
3#include <linux/spinlock.h> 3#include <linux/mutex.h>
4#include <linux/list.h> 4#include <linux/list.h>
5#include <linux/kprobes.h> 5#include <linux/kprobes.h>
6#include <linux/mm.h> 6#include <linux/mm.h>
@@ -143,7 +143,7 @@ static const unsigned char *const p6_nops[ASM_NOP_MAX+1] = {
143#ifdef CONFIG_X86_64 143#ifdef CONFIG_X86_64
144 144
145extern char __vsyscall_0; 145extern char __vsyscall_0;
146static inline const 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 return boot_cpu_data.x86_vendor != X86_VENDOR_INTEL ||
149 boot_cpu_data.x86 < 6 ? k8_nops : p6_nops; 149 boot_cpu_data.x86 < 6 ? k8_nops : p6_nops;
@@ -162,7 +162,7 @@ static const struct nop {
162 { -1, NULL } 162 { -1, NULL }
163}; 163};
164 164
165static const unsigned char*const * find_nop_table(void) 165const unsigned char *const *find_nop_table(void)
166{ 166{
167 const unsigned char *const *noptable = intel_nops; 167 const unsigned char *const *noptable = intel_nops;
168 int i; 168 int i;
@@ -279,7 +279,7 @@ struct smp_alt_module {
279 struct list_head next; 279 struct list_head next;
280}; 280};
281static LIST_HEAD(smp_alt_modules); 281static LIST_HEAD(smp_alt_modules);
282static DEFINE_SPINLOCK(smp_alt); 282static DEFINE_MUTEX(smp_alt);
283static int smp_mode = 1; /* protected by smp_alt */ 283static int smp_mode = 1; /* protected by smp_alt */
284 284
285void alternatives_smp_module_add(struct module *mod, char *name, 285void alternatives_smp_module_add(struct module *mod, char *name,
@@ -312,12 +312,12 @@ void alternatives_smp_module_add(struct module *mod, char *name,
312 __func__, smp->locks, smp->locks_end, 312 __func__, smp->locks, smp->locks_end,
313 smp->text, smp->text_end, smp->name); 313 smp->text, smp->text_end, smp->name);
314 314
315 spin_lock(&smp_alt); 315 mutex_lock(&smp_alt);
316 list_add_tail(&smp->next, &smp_alt_modules); 316 list_add_tail(&smp->next, &smp_alt_modules);
317 if (boot_cpu_has(X86_FEATURE_UP)) 317 if (boot_cpu_has(X86_FEATURE_UP))
318 alternatives_smp_unlock(smp->locks, smp->locks_end, 318 alternatives_smp_unlock(smp->locks, smp->locks_end,
319 smp->text, smp->text_end); 319 smp->text, smp->text_end);
320 spin_unlock(&smp_alt); 320 mutex_unlock(&smp_alt);
321} 321}
322 322
323void alternatives_smp_module_del(struct module *mod) 323void alternatives_smp_module_del(struct module *mod)
@@ -327,17 +327,17 @@ void alternatives_smp_module_del(struct module *mod)
327 if (smp_alt_once || noreplace_smp) 327 if (smp_alt_once || noreplace_smp)
328 return; 328 return;
329 329
330 spin_lock(&smp_alt); 330 mutex_lock(&smp_alt);
331 list_for_each_entry(item, &smp_alt_modules, next) { 331 list_for_each_entry(item, &smp_alt_modules, next) {
332 if (mod != item->mod) 332 if (mod != item->mod)
333 continue; 333 continue;
334 list_del(&item->next); 334 list_del(&item->next);
335 spin_unlock(&smp_alt); 335 mutex_unlock(&smp_alt);
336 DPRINTK("%s: %s\n", __func__, item->name); 336 DPRINTK("%s: %s\n", __func__, item->name);
337 kfree(item); 337 kfree(item);
338 return; 338 return;
339 } 339 }
340 spin_unlock(&smp_alt); 340 mutex_unlock(&smp_alt);
341} 341}
342 342
343void alternatives_smp_switch(int smp) 343void alternatives_smp_switch(int smp)
@@ -359,7 +359,7 @@ void alternatives_smp_switch(int smp)
359 return; 359 return;
360 BUG_ON(!smp && (num_online_cpus() > 1)); 360 BUG_ON(!smp && (num_online_cpus() > 1));
361 361
362 spin_lock(&smp_alt); 362 mutex_lock(&smp_alt);
363 363
364 /* 364 /*
365 * Avoid unnecessary switches because it forces JIT based VMs to 365 * Avoid unnecessary switches because it forces JIT based VMs to
@@ -383,7 +383,7 @@ void alternatives_smp_switch(int smp)
383 mod->text, mod->text_end); 383 mod->text, mod->text_end);
384 } 384 }
385 smp_mode = smp; 385 smp_mode = smp;
386 spin_unlock(&smp_alt); 386 mutex_unlock(&smp_alt);
387} 387}
388 388
389#endif 389#endif
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
new file mode 100644
index 000000000000..22d7d050905d
--- /dev/null
+++ b/arch/x86/kernel/amd_iommu.c
@@ -0,0 +1,1164 @@
1/*
2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
4 * Leo Duran <leo.duran@amd.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/pci.h>
21#include <linux/gfp.h>
22#include <linux/bitops.h>
23#include <linux/scatterlist.h>
24#include <linux/iommu-helper.h>
25#include <asm/proto.h>
26#include <asm/iommu.h>
27#include <asm/amd_iommu_types.h>
28#include <asm/amd_iommu.h>
29
30#define CMD_SET_TYPE(cmd, t) ((cmd)->data[1] |= ((t) << 28))
31
32#define EXIT_LOOP_COUNT 10000000
33
34static DEFINE_RWLOCK(amd_iommu_devtable_lock);
35
36/*
37 * general struct to manage commands send to an IOMMU
38 */
39struct iommu_cmd {
40 u32 data[4];
41};
42
43static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
44 struct unity_map_entry *e);
45
46/* returns !0 if the IOMMU is caching non-present entries in its TLB */
47static int iommu_has_npcache(struct amd_iommu *iommu)
48{
49 return iommu->cap & IOMMU_CAP_NPCACHE;
50}
51
52/****************************************************************************
53 *
54 * IOMMU command queuing functions
55 *
56 ****************************************************************************/
57
58/*
59 * Writes the command to the IOMMUs command buffer and informs the
60 * hardware about the new command. Must be called with iommu->lock held.
61 */
62static int __iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd)
63{
64 u32 tail, head;
65 u8 *target;
66
67 tail = readl(iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
68 target = (iommu->cmd_buf + tail);
69 memcpy_toio(target, cmd, sizeof(*cmd));
70 tail = (tail + sizeof(*cmd)) % iommu->cmd_buf_size;
71 head = readl(iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
72 if (tail == head)
73 return -ENOMEM;
74 writel(tail, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
75
76 return 0;
77}
78
79/*
80 * General queuing function for commands. Takes iommu->lock and calls
81 * __iommu_queue_command().
82 */
83static int iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd)
84{
85 unsigned long flags;
86 int ret;
87
88 spin_lock_irqsave(&iommu->lock, flags);
89 ret = __iommu_queue_command(iommu, cmd);
90 spin_unlock_irqrestore(&iommu->lock, flags);
91
92 return ret;
93}
94
95/*
96 * This function is called whenever we need to ensure that the IOMMU has
97 * completed execution of all commands we sent. It sends a
98 * COMPLETION_WAIT command and waits for it to finish. The IOMMU informs
99 * us about that by writing a value to a physical address we pass with
100 * the command.
101 */
102static int iommu_completion_wait(struct amd_iommu *iommu)
103{
104 int ret;
105 struct iommu_cmd cmd;
106 volatile u64 ready = 0;
107 unsigned long ready_phys = virt_to_phys(&ready);
108 unsigned long i = 0;
109
110 memset(&cmd, 0, sizeof(cmd));
111 cmd.data[0] = LOW_U32(ready_phys) | CMD_COMPL_WAIT_STORE_MASK;
112 cmd.data[1] = upper_32_bits(ready_phys);
113 cmd.data[2] = 1; /* value written to 'ready' */
114 CMD_SET_TYPE(&cmd, CMD_COMPL_WAIT);
115
116 iommu->need_sync = 0;
117
118 ret = iommu_queue_command(iommu, &cmd);
119
120 if (ret)
121 return ret;
122
123 while (!ready && (i < EXIT_LOOP_COUNT)) {
124 ++i;
125 cpu_relax();
126 }
127
128 if (unlikely((i == EXIT_LOOP_COUNT) && printk_ratelimit()))
129 printk(KERN_WARNING "AMD IOMMU: Completion wait loop failed\n");
130
131 return 0;
132}
133
134/*
135 * Command send function for invalidating a device table entry
136 */
137static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
138{
139 struct iommu_cmd cmd;
140
141 BUG_ON(iommu == NULL);
142
143 memset(&cmd, 0, sizeof(cmd));
144 CMD_SET_TYPE(&cmd, CMD_INV_DEV_ENTRY);
145 cmd.data[0] = devid;
146
147 iommu->need_sync = 1;
148
149 return iommu_queue_command(iommu, &cmd);
150}
151
152/*
153 * Generic command send function for invalidaing TLB entries
154 */
155static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
156 u64 address, u16 domid, int pde, int s)
157{
158 struct iommu_cmd cmd;
159
160 memset(&cmd, 0, sizeof(cmd));
161 address &= PAGE_MASK;
162 CMD_SET_TYPE(&cmd, CMD_INV_IOMMU_PAGES);
163 cmd.data[1] |= domid;
164 cmd.data[2] = LOW_U32(address);
165 cmd.data[3] = upper_32_bits(address);
166 if (s) /* size bit - we flush more than one 4kb page */
167 cmd.data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK;
168 if (pde) /* PDE bit - we wan't flush everything not only the PTEs */
169 cmd.data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK;
170
171 iommu->need_sync = 1;
172
173 return iommu_queue_command(iommu, &cmd);
174}
175
176/*
177 * TLB invalidation function which is called from the mapping functions.
178 * It invalidates a single PTE if the range to flush is within a single
179 * page. Otherwise it flushes the whole TLB of the IOMMU.
180 */
181static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
182 u64 address, size_t size)
183{
184 int s = 0;
185 unsigned pages = iommu_num_pages(address, size);
186
187 address &= PAGE_MASK;
188
189 if (pages > 1) {
190 /*
191 * If we have to flush more than one page, flush all
192 * TLB entries for this domain
193 */
194 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
195 s = 1;
196 }
197
198 iommu_queue_inv_iommu_pages(iommu, address, domid, 0, s);
199
200 return 0;
201}
202
203/****************************************************************************
204 *
205 * The functions below are used the create the page table mappings for
206 * unity mapped regions.
207 *
208 ****************************************************************************/
209
210/*
211 * Generic mapping functions. It maps a physical address into a DMA
212 * address space. It allocates the page table pages if necessary.
213 * In the future it can be extended to a generic mapping function
214 * supporting all features of AMD IOMMU page tables like level skipping
215 * and full 64 bit address spaces.
216 */
217static int iommu_map(struct protection_domain *dom,
218 unsigned long bus_addr,
219 unsigned long phys_addr,
220 int prot)
221{
222 u64 __pte, *pte, *page;
223
224 bus_addr = PAGE_ALIGN(bus_addr);
225 phys_addr = PAGE_ALIGN(bus_addr);
226
227 /* only support 512GB address spaces for now */
228 if (bus_addr > IOMMU_MAP_SIZE_L3 || !(prot & IOMMU_PROT_MASK))
229 return -EINVAL;
230
231 pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(bus_addr)];
232
233 if (!IOMMU_PTE_PRESENT(*pte)) {
234 page = (u64 *)get_zeroed_page(GFP_KERNEL);
235 if (!page)
236 return -ENOMEM;
237 *pte = IOMMU_L2_PDE(virt_to_phys(page));
238 }
239
240 pte = IOMMU_PTE_PAGE(*pte);
241 pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)];
242
243 if (!IOMMU_PTE_PRESENT(*pte)) {
244 page = (u64 *)get_zeroed_page(GFP_KERNEL);
245 if (!page)
246 return -ENOMEM;
247 *pte = IOMMU_L1_PDE(virt_to_phys(page));
248 }
249
250 pte = IOMMU_PTE_PAGE(*pte);
251 pte = &pte[IOMMU_PTE_L0_INDEX(bus_addr)];
252
253 if (IOMMU_PTE_PRESENT(*pte))
254 return -EBUSY;
255
256 __pte = phys_addr | IOMMU_PTE_P;
257 if (prot & IOMMU_PROT_IR)
258 __pte |= IOMMU_PTE_IR;
259 if (prot & IOMMU_PROT_IW)
260 __pte |= IOMMU_PTE_IW;
261
262 *pte = __pte;
263
264 return 0;
265}
266
267/*
268 * This function checks if a specific unity mapping entry is needed for
269 * this specific IOMMU.
270 */
271static int iommu_for_unity_map(struct amd_iommu *iommu,
272 struct unity_map_entry *entry)
273{
274 u16 bdf, i;
275
276 for (i = entry->devid_start; i <= entry->devid_end; ++i) {
277 bdf = amd_iommu_alias_table[i];
278 if (amd_iommu_rlookup_table[bdf] == iommu)
279 return 1;
280 }
281
282 return 0;
283}
284
285/*
286 * Init the unity mappings for a specific IOMMU in the system
287 *
288 * Basically iterates over all unity mapping entries and applies them to
289 * the default domain DMA of that IOMMU if necessary.
290 */
291static int iommu_init_unity_mappings(struct amd_iommu *iommu)
292{
293 struct unity_map_entry *entry;
294 int ret;
295
296 list_for_each_entry(entry, &amd_iommu_unity_map, list) {
297 if (!iommu_for_unity_map(iommu, entry))
298 continue;
299 ret = dma_ops_unity_map(iommu->default_dom, entry);
300 if (ret)
301 return ret;
302 }
303
304 return 0;
305}
306
307/*
308 * This function actually applies the mapping to the page table of the
309 * dma_ops domain.
310 */
311static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
312 struct unity_map_entry *e)
313{
314 u64 addr;
315 int ret;
316
317 for (addr = e->address_start; addr < e->address_end;
318 addr += PAGE_SIZE) {
319 ret = iommu_map(&dma_dom->domain, addr, addr, e->prot);
320 if (ret)
321 return ret;
322 /*
323 * if unity mapping is in aperture range mark the page
324 * as allocated in the aperture
325 */
326 if (addr < dma_dom->aperture_size)
327 __set_bit(addr >> PAGE_SHIFT, dma_dom->bitmap);
328 }
329
330 return 0;
331}
332
333/*
334 * Inits the unity mappings required for a specific device
335 */
336static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
337 u16 devid)
338{
339 struct unity_map_entry *e;
340 int ret;
341
342 list_for_each_entry(e, &amd_iommu_unity_map, list) {
343 if (!(devid >= e->devid_start && devid <= e->devid_end))
344 continue;
345 ret = dma_ops_unity_map(dma_dom, e);
346 if (ret)
347 return ret;
348 }
349
350 return 0;
351}
352
353/****************************************************************************
354 *
355 * The next functions belong to the address allocator for the dma_ops
356 * interface functions. They work like the allocators in the other IOMMU
357 * drivers. Its basically a bitmap which marks the allocated pages in
358 * the aperture. Maybe it could be enhanced in the future to a more
359 * efficient allocator.
360 *
361 ****************************************************************************/
362static unsigned long dma_mask_to_pages(unsigned long mask)
363{
364 return (mask >> PAGE_SHIFT) +
365 (PAGE_ALIGN(mask & ~PAGE_MASK) >> PAGE_SHIFT);
366}
367
368/*
369 * The address allocator core function.
370 *
371 * called with domain->lock held
372 */
373static unsigned long dma_ops_alloc_addresses(struct device *dev,
374 struct dma_ops_domain *dom,
375 unsigned int pages)
376{
377 unsigned long limit = dma_mask_to_pages(*dev->dma_mask);
378 unsigned long address;
379 unsigned long size = dom->aperture_size >> PAGE_SHIFT;
380 unsigned long boundary_size;
381
382 boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
383 PAGE_SIZE) >> PAGE_SHIFT;
384 limit = limit < size ? limit : size;
385
386 if (dom->next_bit >= limit)
387 dom->next_bit = 0;
388
389 address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
390 0 , boundary_size, 0);
391 if (address == -1)
392 address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
393 0, boundary_size, 0);
394
395 if (likely(address != -1)) {
396 dom->next_bit = address + pages;
397 address <<= PAGE_SHIFT;
398 } else
399 address = bad_dma_address;
400
401 WARN_ON((address + (PAGE_SIZE*pages)) > dom->aperture_size);
402
403 return address;
404}
405
406/*
407 * The address free function.
408 *
409 * called with domain->lock held
410 */
411static void dma_ops_free_addresses(struct dma_ops_domain *dom,
412 unsigned long address,
413 unsigned int pages)
414{
415 address >>= PAGE_SHIFT;
416 iommu_area_free(dom->bitmap, address, pages);
417}
418
419/****************************************************************************
420 *
421 * The next functions belong to the domain allocation. A domain is
422 * allocated for every IOMMU as the default domain. If device isolation
423 * is enabled, every device get its own domain. The most important thing
424 * about domains is the page table mapping the DMA address space they
425 * contain.
426 *
427 ****************************************************************************/
428
429static u16 domain_id_alloc(void)
430{
431 unsigned long flags;
432 int id;
433
434 write_lock_irqsave(&amd_iommu_devtable_lock, flags);
435 id = find_first_zero_bit(amd_iommu_pd_alloc_bitmap, MAX_DOMAIN_ID);
436 BUG_ON(id == 0);
437 if (id > 0 && id < MAX_DOMAIN_ID)
438 __set_bit(id, amd_iommu_pd_alloc_bitmap);
439 else
440 id = 0;
441 write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
442
443 return id;
444}
445
446/*
447 * Used to reserve address ranges in the aperture (e.g. for exclusion
448 * ranges.
449 */
450static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
451 unsigned long start_page,
452 unsigned int pages)
453{
454 unsigned int last_page = dom->aperture_size >> PAGE_SHIFT;
455
456 if (start_page + pages > last_page)
457 pages = last_page - start_page;
458
459 set_bit_string(dom->bitmap, start_page, pages);
460}
461
462static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom)
463{
464 int i, j;
465 u64 *p1, *p2, *p3;
466
467 p1 = dma_dom->domain.pt_root;
468
469 if (!p1)
470 return;
471
472 for (i = 0; i < 512; ++i) {
473 if (!IOMMU_PTE_PRESENT(p1[i]))
474 continue;
475
476 p2 = IOMMU_PTE_PAGE(p1[i]);
477 for (j = 0; j < 512; ++i) {
478 if (!IOMMU_PTE_PRESENT(p2[j]))
479 continue;
480 p3 = IOMMU_PTE_PAGE(p2[j]);
481 free_page((unsigned long)p3);
482 }
483
484 free_page((unsigned long)p2);
485 }
486
487 free_page((unsigned long)p1);
488}
489
490/*
491 * Free a domain, only used if something went wrong in the
492 * allocation path and we need to free an already allocated page table
493 */
494static void dma_ops_domain_free(struct dma_ops_domain *dom)
495{
496 if (!dom)
497 return;
498
499 dma_ops_free_pagetable(dom);
500
501 kfree(dom->pte_pages);
502
503 kfree(dom->bitmap);
504
505 kfree(dom);
506}
507
508/*
509 * Allocates a new protection domain usable for the dma_ops functions.
510 * It also intializes the page table and the address allocator data
511 * structures required for the dma_ops interface
512 */
513static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
514 unsigned order)
515{
516 struct dma_ops_domain *dma_dom;
517 unsigned i, num_pte_pages;
518 u64 *l2_pde;
519 u64 address;
520
521 /*
522 * Currently the DMA aperture must be between 32 MB and 1GB in size
523 */
524 if ((order < 25) || (order > 30))
525 return NULL;
526
527 dma_dom = kzalloc(sizeof(struct dma_ops_domain), GFP_KERNEL);
528 if (!dma_dom)
529 return NULL;
530
531 spin_lock_init(&dma_dom->domain.lock);
532
533 dma_dom->domain.id = domain_id_alloc();
534 if (dma_dom->domain.id == 0)
535 goto free_dma_dom;
536 dma_dom->domain.mode = PAGE_MODE_3_LEVEL;
537 dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL);
538 dma_dom->domain.priv = dma_dom;
539 if (!dma_dom->domain.pt_root)
540 goto free_dma_dom;
541 dma_dom->aperture_size = (1ULL << order);
542 dma_dom->bitmap = kzalloc(dma_dom->aperture_size / (PAGE_SIZE * 8),
543 GFP_KERNEL);
544 if (!dma_dom->bitmap)
545 goto free_dma_dom;
546 /*
547 * mark the first page as allocated so we never return 0 as
548 * a valid dma-address. So we can use 0 as error value
549 */
550 dma_dom->bitmap[0] = 1;
551 dma_dom->next_bit = 0;
552
553 /* Intialize the exclusion range if necessary */
554 if (iommu->exclusion_start &&
555 iommu->exclusion_start < dma_dom->aperture_size) {
556 unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
557 int pages = iommu_num_pages(iommu->exclusion_start,
558 iommu->exclusion_length);
559 dma_ops_reserve_addresses(dma_dom, startpage, pages);
560 }
561
562 /*
563 * At the last step, build the page tables so we don't need to
564 * allocate page table pages in the dma_ops mapping/unmapping
565 * path.
566 */
567 num_pte_pages = dma_dom->aperture_size / (PAGE_SIZE * 512);
568 dma_dom->pte_pages = kzalloc(num_pte_pages * sizeof(void *),
569 GFP_KERNEL);
570 if (!dma_dom->pte_pages)
571 goto free_dma_dom;
572
573 l2_pde = (u64 *)get_zeroed_page(GFP_KERNEL);
574 if (l2_pde == NULL)
575 goto free_dma_dom;
576
577 dma_dom->domain.pt_root[0] = IOMMU_L2_PDE(virt_to_phys(l2_pde));
578
579 for (i = 0; i < num_pte_pages; ++i) {
580 dma_dom->pte_pages[i] = (u64 *)get_zeroed_page(GFP_KERNEL);
581 if (!dma_dom->pte_pages[i])
582 goto free_dma_dom;
583 address = virt_to_phys(dma_dom->pte_pages[i]);
584 l2_pde[i] = IOMMU_L1_PDE(address);
585 }
586
587 return dma_dom;
588
589free_dma_dom:
590 dma_ops_domain_free(dma_dom);
591
592 return NULL;
593}
594
595/*
596 * Find out the protection domain structure for a given PCI device. This
597 * will give us the pointer to the page table root for example.
598 */
599static struct protection_domain *domain_for_device(u16 devid)
600{
601 struct protection_domain *dom;
602 unsigned long flags;
603
604 read_lock_irqsave(&amd_iommu_devtable_lock, flags);
605 dom = amd_iommu_pd_table[devid];
606 read_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
607
608 return dom;
609}
610
611/*
612 * If a device is not yet associated with a domain, this function does
613 * assigns it visible for the hardware
614 */
615static void set_device_domain(struct amd_iommu *iommu,
616 struct protection_domain *domain,
617 u16 devid)
618{
619 unsigned long flags;
620
621 u64 pte_root = virt_to_phys(domain->pt_root);
622
623 pte_root |= (domain->mode & 0x07) << 9;
624 pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | 2;
625
626 write_lock_irqsave(&amd_iommu_devtable_lock, flags);
627 amd_iommu_dev_table[devid].data[0] = pte_root;
628 amd_iommu_dev_table[devid].data[1] = pte_root >> 32;
629 amd_iommu_dev_table[devid].data[2] = domain->id;
630
631 amd_iommu_pd_table[devid] = domain;
632 write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
633
634 iommu_queue_inv_dev_entry(iommu, devid);
635
636 iommu->need_sync = 1;
637}
638
639/*****************************************************************************
640 *
641 * The next functions belong to the dma_ops mapping/unmapping code.
642 *
643 *****************************************************************************/
644
645/*
646 * In the dma_ops path we only have the struct device. This function
647 * finds the corresponding IOMMU, the protection domain and the
648 * requestor id for a given device.
649 * If the device is not yet associated with a domain this is also done
650 * in this function.
651 */
652static int get_device_resources(struct device *dev,
653 struct amd_iommu **iommu,
654 struct protection_domain **domain,
655 u16 *bdf)
656{
657 struct dma_ops_domain *dma_dom;
658 struct pci_dev *pcidev;
659 u16 _bdf;
660
661 BUG_ON(!dev || dev->bus != &pci_bus_type || !dev->dma_mask);
662
663 pcidev = to_pci_dev(dev);
664 _bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
665
666 /* device not translated by any IOMMU in the system? */
667 if (_bdf > amd_iommu_last_bdf) {
668 *iommu = NULL;
669 *domain = NULL;
670 *bdf = 0xffff;
671 return 0;
672 }
673
674 *bdf = amd_iommu_alias_table[_bdf];
675
676 *iommu = amd_iommu_rlookup_table[*bdf];
677 if (*iommu == NULL)
678 return 0;
679 dma_dom = (*iommu)->default_dom;
680 *domain = domain_for_device(*bdf);
681 if (*domain == NULL) {
682 *domain = &dma_dom->domain;
683 set_device_domain(*iommu, *domain, *bdf);
684 printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
685 "device ", (*domain)->id);
686 print_devid(_bdf, 1);
687 }
688
689 return 1;
690}
691
692/*
693 * This is the generic map function. It maps one 4kb page at paddr to
694 * the given address in the DMA address space for the domain.
695 */
696static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
697 struct dma_ops_domain *dom,
698 unsigned long address,
699 phys_addr_t paddr,
700 int direction)
701{
702 u64 *pte, __pte;
703
704 WARN_ON(address > dom->aperture_size);
705
706 paddr &= PAGE_MASK;
707
708 pte = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
709 pte += IOMMU_PTE_L0_INDEX(address);
710
711 __pte = paddr | IOMMU_PTE_P | IOMMU_PTE_FC;
712
713 if (direction == DMA_TO_DEVICE)
714 __pte |= IOMMU_PTE_IR;
715 else if (direction == DMA_FROM_DEVICE)
716 __pte |= IOMMU_PTE_IW;
717 else if (direction == DMA_BIDIRECTIONAL)
718 __pte |= IOMMU_PTE_IR | IOMMU_PTE_IW;
719
720 WARN_ON(*pte);
721
722 *pte = __pte;
723
724 return (dma_addr_t)address;
725}
726
727/*
728 * The generic unmapping function for on page in the DMA address space.
729 */
730static void dma_ops_domain_unmap(struct amd_iommu *iommu,
731 struct dma_ops_domain *dom,
732 unsigned long address)
733{
734 u64 *pte;
735
736 if (address >= dom->aperture_size)
737 return;
738
739 WARN_ON(address & 0xfffULL || address > dom->aperture_size);
740
741 pte = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
742 pte += IOMMU_PTE_L0_INDEX(address);
743
744 WARN_ON(!*pte);
745
746 *pte = 0ULL;
747}
748
749/*
750 * This function contains common code for mapping of a physically
751 * contiguous memory region into DMA address space. It is uses by all
752 * mapping functions provided by this IOMMU driver.
753 * Must be called with the domain lock held.
754 */
755static dma_addr_t __map_single(struct device *dev,
756 struct amd_iommu *iommu,
757 struct dma_ops_domain *dma_dom,
758 phys_addr_t paddr,
759 size_t size,
760 int dir)
761{
762 dma_addr_t offset = paddr & ~PAGE_MASK;
763 dma_addr_t address, start;
764 unsigned int pages;
765 int i;
766
767 pages = iommu_num_pages(paddr, size);
768 paddr &= PAGE_MASK;
769
770 address = dma_ops_alloc_addresses(dev, dma_dom, pages);
771 if (unlikely(address == bad_dma_address))
772 goto out;
773
774 start = address;
775 for (i = 0; i < pages; ++i) {
776 dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
777 paddr += PAGE_SIZE;
778 start += PAGE_SIZE;
779 }
780 address += offset;
781
782out:
783 return address;
784}
785
786/*
787 * Does the reverse of the __map_single function. Must be called with
788 * the domain lock held too
789 */
790static void __unmap_single(struct amd_iommu *iommu,
791 struct dma_ops_domain *dma_dom,
792 dma_addr_t dma_addr,
793 size_t size,
794 int dir)
795{
796 dma_addr_t i, start;
797 unsigned int pages;
798
799 if ((dma_addr == 0) || (dma_addr + size > dma_dom->aperture_size))
800 return;
801
802 pages = iommu_num_pages(dma_addr, size);
803 dma_addr &= PAGE_MASK;
804 start = dma_addr;
805
806 for (i = 0; i < pages; ++i) {
807 dma_ops_domain_unmap(iommu, dma_dom, start);
808 start += PAGE_SIZE;
809 }
810
811 dma_ops_free_addresses(dma_dom, dma_addr, pages);
812}
813
814/*
815 * The exported map_single function for dma_ops.
816 */
817static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
818 size_t size, int dir)
819{
820 unsigned long flags;
821 struct amd_iommu *iommu;
822 struct protection_domain *domain;
823 u16 devid;
824 dma_addr_t addr;
825
826 get_device_resources(dev, &iommu, &domain, &devid);
827
828 if (iommu == NULL || domain == NULL)
829 /* device not handled by any AMD IOMMU */
830 return (dma_addr_t)paddr;
831
832 spin_lock_irqsave(&domain->lock, flags);
833 addr = __map_single(dev, iommu, domain->priv, paddr, size, dir);
834 if (addr == bad_dma_address)
835 goto out;
836
837 if (iommu_has_npcache(iommu))
838 iommu_flush_pages(iommu, domain->id, addr, size);
839
840 if (iommu->need_sync)
841 iommu_completion_wait(iommu);
842
843out:
844 spin_unlock_irqrestore(&domain->lock, flags);
845
846 return addr;
847}
848
849/*
850 * The exported unmap_single function for dma_ops.
851 */
852static void unmap_single(struct device *dev, dma_addr_t dma_addr,
853 size_t size, int dir)
854{
855 unsigned long flags;
856 struct amd_iommu *iommu;
857 struct protection_domain *domain;
858 u16 devid;
859
860 if (!get_device_resources(dev, &iommu, &domain, &devid))
861 /* device not handled by any AMD IOMMU */
862 return;
863
864 spin_lock_irqsave(&domain->lock, flags);
865
866 __unmap_single(iommu, domain->priv, dma_addr, size, dir);
867
868 iommu_flush_pages(iommu, domain->id, dma_addr, size);
869
870 if (iommu->need_sync)
871 iommu_completion_wait(iommu);
872
873 spin_unlock_irqrestore(&domain->lock, flags);
874}
875
876/*
877 * This is a special map_sg function which is used if we should map a
878 * device which is not handled by an AMD IOMMU in the system.
879 */
880static int map_sg_no_iommu(struct device *dev, struct scatterlist *sglist,
881 int nelems, int dir)
882{
883 struct scatterlist *s;
884 int i;
885
886 for_each_sg(sglist, s, nelems, i) {
887 s->dma_address = (dma_addr_t)sg_phys(s);
888 s->dma_length = s->length;
889 }
890
891 return nelems;
892}
893
894/*
895 * The exported map_sg function for dma_ops (handles scatter-gather
896 * lists).
897 */
898static int map_sg(struct device *dev, struct scatterlist *sglist,
899 int nelems, int dir)
900{
901 unsigned long flags;
902 struct amd_iommu *iommu;
903 struct protection_domain *domain;
904 u16 devid;
905 int i;
906 struct scatterlist *s;
907 phys_addr_t paddr;
908 int mapped_elems = 0;
909
910 get_device_resources(dev, &iommu, &domain, &devid);
911
912 if (!iommu || !domain)
913 return map_sg_no_iommu(dev, sglist, nelems, dir);
914
915 spin_lock_irqsave(&domain->lock, flags);
916
917 for_each_sg(sglist, s, nelems, i) {
918 paddr = sg_phys(s);
919
920 s->dma_address = __map_single(dev, iommu, domain->priv,
921 paddr, s->length, dir);
922
923 if (s->dma_address) {
924 s->dma_length = s->length;
925 mapped_elems++;
926 } else
927 goto unmap;
928 if (iommu_has_npcache(iommu))
929 iommu_flush_pages(iommu, domain->id, s->dma_address,
930 s->dma_length);
931 }
932
933 if (iommu->need_sync)
934 iommu_completion_wait(iommu);
935
936out:
937 spin_unlock_irqrestore(&domain->lock, flags);
938
939 return mapped_elems;
940unmap:
941 for_each_sg(sglist, s, mapped_elems, i) {
942 if (s->dma_address)
943 __unmap_single(iommu, domain->priv, s->dma_address,
944 s->dma_length, dir);
945 s->dma_address = s->dma_length = 0;
946 }
947
948 mapped_elems = 0;
949
950 goto out;
951}
952
953/*
954 * The exported map_sg function for dma_ops (handles scatter-gather
955 * lists).
956 */
957static void unmap_sg(struct device *dev, struct scatterlist *sglist,
958 int nelems, int dir)
959{
960 unsigned long flags;
961 struct amd_iommu *iommu;
962 struct protection_domain *domain;
963 struct scatterlist *s;
964 u16 devid;
965 int i;
966
967 if (!get_device_resources(dev, &iommu, &domain, &devid))
968 return;
969
970 spin_lock_irqsave(&domain->lock, flags);
971
972 for_each_sg(sglist, s, nelems, i) {
973 __unmap_single(iommu, domain->priv, s->dma_address,
974 s->dma_length, dir);
975 iommu_flush_pages(iommu, domain->id, s->dma_address,
976 s->dma_length);
977 s->dma_address = s->dma_length = 0;
978 }
979
980 if (iommu->need_sync)
981 iommu_completion_wait(iommu);
982
983 spin_unlock_irqrestore(&domain->lock, flags);
984}
985
986/*
987 * The exported alloc_coherent function for dma_ops.
988 */
989static void *alloc_coherent(struct device *dev, size_t size,
990 dma_addr_t *dma_addr, gfp_t flag)
991{
992 unsigned long flags;
993 void *virt_addr;
994 struct amd_iommu *iommu;
995 struct protection_domain *domain;
996 u16 devid;
997 phys_addr_t paddr;
998
999 virt_addr = (void *)__get_free_pages(flag, get_order(size));
1000 if (!virt_addr)
1001 return 0;
1002
1003 memset(virt_addr, 0, size);
1004 paddr = virt_to_phys(virt_addr);
1005
1006 get_device_resources(dev, &iommu, &domain, &devid);
1007
1008 if (!iommu || !domain) {
1009 *dma_addr = (dma_addr_t)paddr;
1010 return virt_addr;
1011 }
1012
1013 spin_lock_irqsave(&domain->lock, flags);
1014
1015 *dma_addr = __map_single(dev, iommu, domain->priv, paddr,
1016 size, DMA_BIDIRECTIONAL);
1017
1018 if (*dma_addr == bad_dma_address) {
1019 free_pages((unsigned long)virt_addr, get_order(size));
1020 virt_addr = NULL;
1021 goto out;
1022 }
1023
1024 if (iommu_has_npcache(iommu))
1025 iommu_flush_pages(iommu, domain->id, *dma_addr, size);
1026
1027 if (iommu->need_sync)
1028 iommu_completion_wait(iommu);
1029
1030out:
1031 spin_unlock_irqrestore(&domain->lock, flags);
1032
1033 return virt_addr;
1034}
1035
1036/*
1037 * The exported free_coherent function for dma_ops.
1038 * FIXME: fix the generic x86 DMA layer so that it actually calls that
1039 * function.
1040 */
1041static void free_coherent(struct device *dev, size_t size,
1042 void *virt_addr, dma_addr_t dma_addr)
1043{
1044 unsigned long flags;
1045 struct amd_iommu *iommu;
1046 struct protection_domain *domain;
1047 u16 devid;
1048
1049 get_device_resources(dev, &iommu, &domain, &devid);
1050
1051 if (!iommu || !domain)
1052 goto free_mem;
1053
1054 spin_lock_irqsave(&domain->lock, flags);
1055
1056 __unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
1057 iommu_flush_pages(iommu, domain->id, dma_addr, size);
1058
1059 if (iommu->need_sync)
1060 iommu_completion_wait(iommu);
1061
1062 spin_unlock_irqrestore(&domain->lock, flags);
1063
1064free_mem:
1065 free_pages((unsigned long)virt_addr, get_order(size));
1066}
1067
1068/*
1069 * The function for pre-allocating protection domains.
1070 *
1071 * If the driver core informs the DMA layer if a driver grabs a device
1072 * we don't need to preallocate the protection domains anymore.
1073 * For now we have to.
1074 */
1075void prealloc_protection_domains(void)
1076{
1077 struct pci_dev *dev = NULL;
1078 struct dma_ops_domain *dma_dom;
1079 struct amd_iommu *iommu;
1080 int order = amd_iommu_aperture_order;
1081 u16 devid;
1082
1083 while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
1084 devid = (dev->bus->number << 8) | dev->devfn;
1085 if (devid > amd_iommu_last_bdf)
1086 continue;
1087 devid = amd_iommu_alias_table[devid];
1088 if (domain_for_device(devid))
1089 continue;
1090 iommu = amd_iommu_rlookup_table[devid];
1091 if (!iommu)
1092 continue;
1093 dma_dom = dma_ops_domain_alloc(iommu, order);
1094 if (!dma_dom)
1095 continue;
1096 init_unity_mappings_for_device(dma_dom, devid);
1097 set_device_domain(iommu, &dma_dom->domain, devid);
1098 printk(KERN_INFO "AMD IOMMU: Allocated domain %d for device ",
1099 dma_dom->domain.id);
1100 print_devid(devid, 1);
1101 }
1102}
1103
1104static struct dma_mapping_ops amd_iommu_dma_ops = {
1105 .alloc_coherent = alloc_coherent,
1106 .free_coherent = free_coherent,
1107 .map_single = map_single,
1108 .unmap_single = unmap_single,
1109 .map_sg = map_sg,
1110 .unmap_sg = unmap_sg,
1111};
1112
1113/*
1114 * The function which clues the AMD IOMMU driver into dma_ops.
1115 */
1116int __init amd_iommu_init_dma_ops(void)
1117{
1118 struct amd_iommu *iommu;
1119 int order = amd_iommu_aperture_order;
1120 int ret;
1121
1122 /*
1123 * first allocate a default protection domain for every IOMMU we
1124 * found in the system. Devices not assigned to any other
1125 * protection domain will be assigned to the default one.
1126 */
1127 list_for_each_entry(iommu, &amd_iommu_list, list) {
1128 iommu->default_dom = dma_ops_domain_alloc(iommu, order);
1129 if (iommu->default_dom == NULL)
1130 return -ENOMEM;
1131 ret = iommu_init_unity_mappings(iommu);
1132 if (ret)
1133 goto free_domains;
1134 }
1135
1136 /*
1137 * If device isolation is enabled, pre-allocate the protection
1138 * domains for each device.
1139 */
1140 if (amd_iommu_isolate)
1141 prealloc_protection_domains();
1142
1143 iommu_detected = 1;
1144 force_iommu = 1;
1145 bad_dma_address = 0;
1146#ifdef CONFIG_GART_IOMMU
1147 gart_iommu_aperture_disabled = 1;
1148 gart_iommu_aperture = 0;
1149#endif
1150
1151 /* Make the driver finally visible to the drivers */
1152 dma_ops = &amd_iommu_dma_ops;
1153
1154 return 0;
1155
1156free_domains:
1157
1158 list_for_each_entry(iommu, &amd_iommu_list, list) {
1159 if (iommu->default_dom)
1160 dma_ops_domain_free(iommu->default_dom);
1161 }
1162
1163 return ret;
1164}
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
new file mode 100644
index 000000000000..d9a9da597e79
--- /dev/null
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -0,0 +1,1060 @@
1/*
2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
4 * Leo Duran <leo.duran@amd.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/pci.h>
21#include <linux/acpi.h>
22#include <linux/gfp.h>
23#include <linux/list.h>
24#include <linux/sysdev.h>
25#include <asm/pci-direct.h>
26#include <asm/amd_iommu_types.h>
27#include <asm/amd_iommu.h>
28#include <asm/iommu.h>
29
30/*
31 * definitions for the ACPI scanning code
32 */
33#define PCI_BUS(x) (((x) >> 8) & 0xff)
34#define IVRS_HEADER_LENGTH 48
35
36#define ACPI_IVHD_TYPE 0x10
37#define ACPI_IVMD_TYPE_ALL 0x20
38#define ACPI_IVMD_TYPE 0x21
39#define ACPI_IVMD_TYPE_RANGE 0x22
40
41#define IVHD_DEV_ALL 0x01
42#define IVHD_DEV_SELECT 0x02
43#define IVHD_DEV_SELECT_RANGE_START 0x03
44#define IVHD_DEV_RANGE_END 0x04
45#define IVHD_DEV_ALIAS 0x42
46#define IVHD_DEV_ALIAS_RANGE 0x43
47#define IVHD_DEV_EXT_SELECT 0x46
48#define IVHD_DEV_EXT_SELECT_RANGE 0x47
49
50#define IVHD_FLAG_HT_TUN_EN 0x00
51#define IVHD_FLAG_PASSPW_EN 0x01
52#define IVHD_FLAG_RESPASSPW_EN 0x02
53#define IVHD_FLAG_ISOC_EN 0x03
54
55#define IVMD_FLAG_EXCL_RANGE 0x08
56#define IVMD_FLAG_UNITY_MAP 0x01
57
58#define ACPI_DEVFLAG_INITPASS 0x01
59#define ACPI_DEVFLAG_EXTINT 0x02
60#define ACPI_DEVFLAG_NMI 0x04
61#define ACPI_DEVFLAG_SYSMGT1 0x10
62#define ACPI_DEVFLAG_SYSMGT2 0x20
63#define ACPI_DEVFLAG_LINT0 0x40
64#define ACPI_DEVFLAG_LINT1 0x80
65#define ACPI_DEVFLAG_ATSDIS 0x10000000
66
67/*
68 * ACPI table definitions
69 *
70 * These data structures are laid over the table to parse the important values
71 * out of it.
72 */
73
74/*
75 * structure describing one IOMMU in the ACPI table. Typically followed by one
76 * or more ivhd_entrys.
77 */
78struct ivhd_header {
79 u8 type;
80 u8 flags;
81 u16 length;
82 u16 devid;
83 u16 cap_ptr;
84 u64 mmio_phys;
85 u16 pci_seg;
86 u16 info;
87 u32 reserved;
88} __attribute__((packed));
89
90/*
91 * A device entry describing which devices a specific IOMMU translates and
92 * which requestor ids they use.
93 */
94struct ivhd_entry {
95 u8 type;
96 u16 devid;
97 u8 flags;
98 u32 ext;
99} __attribute__((packed));
100
101/*
102 * An AMD IOMMU memory definition structure. It defines things like exclusion
103 * ranges for devices and regions that should be unity mapped.
104 */
105struct ivmd_header {
106 u8 type;
107 u8 flags;
108 u16 length;
109 u16 devid;
110 u16 aux;
111 u64 resv;
112 u64 range_start;
113 u64 range_length;
114} __attribute__((packed));
115
116static int __initdata amd_iommu_detected;
117
118u16 amd_iommu_last_bdf; /* largest PCI device id we have
119 to handle */
120LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
121 we find in ACPI */
122unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
123int amd_iommu_isolate; /* if 1, device isolation is enabled */
124
125LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
126 system */
127
128/*
129 * Pointer to the device table which is shared by all AMD IOMMUs
130 * it is indexed by the PCI device id or the HT unit id and contains
131 * information about the domain the device belongs to as well as the
132 * page table root pointer.
133 */
134struct dev_table_entry *amd_iommu_dev_table;
135
136/*
137 * The alias table is a driver specific data structure which contains the
138 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
139 * More than one device can share the same requestor id.
140 */
141u16 *amd_iommu_alias_table;
142
143/*
144 * The rlookup table is used to find the IOMMU which is responsible
145 * for a specific device. It is also indexed by the PCI device id.
146 */
147struct amd_iommu **amd_iommu_rlookup_table;
148
149/*
150 * The pd table (protection domain table) is used to find the protection domain
151 * data structure a device belongs to. Indexed with the PCI device id too.
152 */
153struct protection_domain **amd_iommu_pd_table;
154
155/*
156 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
157 * to know which ones are already in use.
158 */
159unsigned long *amd_iommu_pd_alloc_bitmap;
160
161static u32 dev_table_size; /* size of the device table */
162static u32 alias_table_size; /* size of the alias table */
163static u32 rlookup_table_size; /* size if the rlookup table */
164
165static inline void update_last_devid(u16 devid)
166{
167 if (devid > amd_iommu_last_bdf)
168 amd_iommu_last_bdf = devid;
169}
170
171static inline unsigned long tbl_size(int entry_size)
172{
173 unsigned shift = PAGE_SHIFT +
174 get_order(amd_iommu_last_bdf * entry_size);
175
176 return 1UL << shift;
177}
178
179/****************************************************************************
180 *
181 * AMD IOMMU MMIO register space handling functions
182 *
183 * These functions are used to program the IOMMU device registers in
184 * MMIO space required for that driver.
185 *
186 ****************************************************************************/
187
188/*
189 * This function set the exclusion range in the IOMMU. DMA accesses to the
190 * exclusion range are passed through untranslated
191 */
192static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
193{
194 u64 start = iommu->exclusion_start & PAGE_MASK;
195 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
196 u64 entry;
197
198 if (!iommu->exclusion_start)
199 return;
200
201 entry = start | MMIO_EXCL_ENABLE_MASK;
202 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
203 &entry, sizeof(entry));
204
205 entry = limit;
206 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
207 &entry, sizeof(entry));
208}
209
210/* Programs the physical address of the device table into the IOMMU hardware */
211static void __init iommu_set_device_table(struct amd_iommu *iommu)
212{
213 u32 entry;
214
215 BUG_ON(iommu->mmio_base == NULL);
216
217 entry = virt_to_phys(amd_iommu_dev_table);
218 entry |= (dev_table_size >> 12) - 1;
219 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
220 &entry, sizeof(entry));
221}
222
223/* Generic functions to enable/disable certain features of the IOMMU. */
224static void __init iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
225{
226 u32 ctrl;
227
228 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
229 ctrl |= (1 << bit);
230 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
231}
232
233static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
234{
235 u32 ctrl;
236
237 ctrl = (u64)readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
238 ctrl &= ~(1 << bit);
239 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
240}
241
242/* Function to enable the hardware */
243void __init iommu_enable(struct amd_iommu *iommu)
244{
245 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at ");
246 print_devid(iommu->devid, 0);
247 printk(" cap 0x%hx\n", iommu->cap_ptr);
248
249 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
250}
251
252/*
253 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
254 * the system has one.
255 */
256static u8 * __init iommu_map_mmio_space(u64 address)
257{
258 u8 *ret;
259
260 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu"))
261 return NULL;
262
263 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
264 if (ret != NULL)
265 return ret;
266
267 release_mem_region(address, MMIO_REGION_LENGTH);
268
269 return NULL;
270}
271
272static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
273{
274 if (iommu->mmio_base)
275 iounmap(iommu->mmio_base);
276 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
277}
278
279/****************************************************************************
280 *
281 * The functions below belong to the first pass of AMD IOMMU ACPI table
282 * parsing. In this pass we try to find out the highest device id this
283 * code has to handle. Upon this information the size of the shared data
284 * structures is determined later.
285 *
286 ****************************************************************************/
287
288/*
289 * This function reads the last device id the IOMMU has to handle from the PCI
290 * capability header for this IOMMU
291 */
292static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
293{
294 u32 cap;
295
296 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
297 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
298
299 return 0;
300}
301
302/*
303 * After reading the highest device id from the IOMMU PCI capability header
304 * this function looks if there is a higher device id defined in the ACPI table
305 */
306static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
307{
308 u8 *p = (void *)h, *end = (void *)h;
309 struct ivhd_entry *dev;
310
311 p += sizeof(*h);
312 end += h->length;
313
314 find_last_devid_on_pci(PCI_BUS(h->devid),
315 PCI_SLOT(h->devid),
316 PCI_FUNC(h->devid),
317 h->cap_ptr);
318
319 while (p < end) {
320 dev = (struct ivhd_entry *)p;
321 switch (dev->type) {
322 case IVHD_DEV_SELECT:
323 case IVHD_DEV_RANGE_END:
324 case IVHD_DEV_ALIAS:
325 case IVHD_DEV_EXT_SELECT:
326 /* all the above subfield types refer to device ids */
327 update_last_devid(dev->devid);
328 break;
329 default:
330 break;
331 }
332 p += 0x04 << (*p >> 6);
333 }
334
335 WARN_ON(p != end);
336
337 return 0;
338}
339
340/*
341 * Iterate over all IVHD entries in the ACPI table and find the highest device
342 * id which we need to handle. This is the first of three functions which parse
343 * the ACPI table. So we check the checksum here.
344 */
345static int __init find_last_devid_acpi(struct acpi_table_header *table)
346{
347 int i;
348 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
349 struct ivhd_header *h;
350
351 /*
352 * Validate checksum here so we don't need to do it when
353 * we actually parse the table
354 */
355 for (i = 0; i < table->length; ++i)
356 checksum += p[i];
357 if (checksum != 0)
358 /* ACPI table corrupt */
359 return -ENODEV;
360
361 p += IVRS_HEADER_LENGTH;
362
363 end += table->length;
364 while (p < end) {
365 h = (struct ivhd_header *)p;
366 switch (h->type) {
367 case ACPI_IVHD_TYPE:
368 find_last_devid_from_ivhd(h);
369 break;
370 default:
371 break;
372 }
373 p += h->length;
374 }
375 WARN_ON(p != end);
376
377 return 0;
378}
379
380/****************************************************************************
381 *
382 * The following functions belong the the code path which parses the ACPI table
383 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
384 * data structures, initialize the device/alias/rlookup table and also
385 * basically initialize the hardware.
386 *
387 ****************************************************************************/
388
389/*
390 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
391 * write commands to that buffer later and the IOMMU will execute them
392 * asynchronously
393 */
394static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
395{
396 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
397 get_order(CMD_BUFFER_SIZE));
398 u64 entry;
399
400 if (cmd_buf == NULL)
401 return NULL;
402
403 iommu->cmd_buf_size = CMD_BUFFER_SIZE;
404
405 entry = (u64)virt_to_phys(cmd_buf);
406 entry |= MMIO_CMD_SIZE_512;
407 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
408 &entry, sizeof(entry));
409
410 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
411
412 return cmd_buf;
413}
414
415static void __init free_command_buffer(struct amd_iommu *iommu)
416{
417 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
418}
419
420/* sets a specific bit in the device table entry. */
421static void set_dev_entry_bit(u16 devid, u8 bit)
422{
423 int i = (bit >> 5) & 0x07;
424 int _bit = bit & 0x1f;
425
426 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
427}
428
429/* Writes the specific IOMMU for a device into the rlookup table */
430static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
431{
432 amd_iommu_rlookup_table[devid] = iommu;
433}
434
435/*
436 * This function takes the device specific flags read from the ACPI
437 * table and sets up the device table entry with that information
438 */
439static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
440 u16 devid, u32 flags, u32 ext_flags)
441{
442 if (flags & ACPI_DEVFLAG_INITPASS)
443 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
444 if (flags & ACPI_DEVFLAG_EXTINT)
445 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
446 if (flags & ACPI_DEVFLAG_NMI)
447 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
448 if (flags & ACPI_DEVFLAG_SYSMGT1)
449 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
450 if (flags & ACPI_DEVFLAG_SYSMGT2)
451 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
452 if (flags & ACPI_DEVFLAG_LINT0)
453 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
454 if (flags & ACPI_DEVFLAG_LINT1)
455 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
456
457 set_iommu_for_device(iommu, devid);
458}
459
460/*
461 * Reads the device exclusion range from ACPI and initialize IOMMU with
462 * it
463 */
464static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
465{
466 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
467
468 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
469 return;
470
471 if (iommu) {
472 /*
473 * We only can configure exclusion ranges per IOMMU, not
474 * per device. But we can enable the exclusion range per
475 * device. This is done here
476 */
477 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
478 iommu->exclusion_start = m->range_start;
479 iommu->exclusion_length = m->range_length;
480 }
481}
482
483/*
484 * This function reads some important data from the IOMMU PCI space and
485 * initializes the driver data structure with it. It reads the hardware
486 * capabilities and the first/last device entries
487 */
488static void __init init_iommu_from_pci(struct amd_iommu *iommu)
489{
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;
494 u32 range;
495
496 iommu->cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_CAP_HDR_OFFSET);
497
498 range = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
499 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
500 MMIO_GET_FD(range));
501 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
502 MMIO_GET_LD(range));
503}
504
505/*
506 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
507 * initializes the hardware and our data structures with it.
508 */
509static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
510 struct ivhd_header *h)
511{
512 u8 *p = (u8 *)h;
513 u8 *end = p, flags = 0;
514 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
515 u32 ext_flags = 0;
516 bool alias = false;
517 struct ivhd_entry *e;
518
519 /*
520 * First set the recommended feature enable bits from ACPI
521 * into the IOMMU control registers
522 */
523 h->flags & IVHD_FLAG_HT_TUN_EN ?
524 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
525 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
526
527 h->flags & IVHD_FLAG_PASSPW_EN ?
528 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
529 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
530
531 h->flags & IVHD_FLAG_RESPASSPW_EN ?
532 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
533 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
534
535 h->flags & IVHD_FLAG_ISOC_EN ?
536 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
537 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
538
539 /*
540 * make IOMMU memory accesses cache coherent
541 */
542 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
543
544 /*
545 * Done. Now parse the device entries
546 */
547 p += sizeof(struct ivhd_header);
548 end += h->length;
549
550 while (p < end) {
551 e = (struct ivhd_entry *)p;
552 switch (e->type) {
553 case IVHD_DEV_ALL:
554 for (dev_i = iommu->first_device;
555 dev_i <= iommu->last_device; ++dev_i)
556 set_dev_entry_from_acpi(iommu, dev_i,
557 e->flags, 0);
558 break;
559 case IVHD_DEV_SELECT:
560 devid = e->devid;
561 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
562 break;
563 case IVHD_DEV_SELECT_RANGE_START:
564 devid_start = e->devid;
565 flags = e->flags;
566 ext_flags = 0;
567 alias = false;
568 break;
569 case IVHD_DEV_ALIAS:
570 devid = e->devid;
571 devid_to = e->ext >> 8;
572 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
573 amd_iommu_alias_table[devid] = devid_to;
574 break;
575 case IVHD_DEV_ALIAS_RANGE:
576 devid_start = e->devid;
577 flags = e->flags;
578 devid_to = e->ext >> 8;
579 ext_flags = 0;
580 alias = true;
581 break;
582 case IVHD_DEV_EXT_SELECT:
583 devid = e->devid;
584 set_dev_entry_from_acpi(iommu, devid, e->flags,
585 e->ext);
586 break;
587 case IVHD_DEV_EXT_SELECT_RANGE:
588 devid_start = e->devid;
589 flags = e->flags;
590 ext_flags = e->ext;
591 alias = false;
592 break;
593 case IVHD_DEV_RANGE_END:
594 devid = e->devid;
595 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
596 if (alias)
597 amd_iommu_alias_table[dev_i] = devid_to;
598 set_dev_entry_from_acpi(iommu,
599 amd_iommu_alias_table[dev_i],
600 flags, ext_flags);
601 }
602 break;
603 default:
604 break;
605 }
606
607 p += 0x04 << (e->type >> 6);
608 }
609}
610
611/* Initializes the device->iommu mapping for the driver */
612static int __init init_iommu_devices(struct amd_iommu *iommu)
613{
614 u16 i;
615
616 for (i = iommu->first_device; i <= iommu->last_device; ++i)
617 set_iommu_for_device(iommu, i);
618
619 return 0;
620}
621
622static void __init free_iommu_one(struct amd_iommu *iommu)
623{
624 free_command_buffer(iommu);
625 iommu_unmap_mmio_space(iommu);
626}
627
628static void __init free_iommu_all(void)
629{
630 struct amd_iommu *iommu, *next;
631
632 list_for_each_entry_safe(iommu, next, &amd_iommu_list, list) {
633 list_del(&iommu->list);
634 free_iommu_one(iommu);
635 kfree(iommu);
636 }
637}
638
639/*
640 * This function clues the initialization function for one IOMMU
641 * together and also allocates the command buffer and programs the
642 * hardware. It does NOT enable the IOMMU. This is done afterwards.
643 */
644static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
645{
646 spin_lock_init(&iommu->lock);
647 list_add_tail(&iommu->list, &amd_iommu_list);
648
649 /*
650 * Copy data from ACPI table entry to the iommu struct
651 */
652 iommu->devid = h->devid;
653 iommu->cap_ptr = h->cap_ptr;
654 iommu->mmio_phys = h->mmio_phys;
655 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
656 if (!iommu->mmio_base)
657 return -ENOMEM;
658
659 iommu_set_device_table(iommu);
660 iommu->cmd_buf = alloc_command_buffer(iommu);
661 if (!iommu->cmd_buf)
662 return -ENOMEM;
663
664 init_iommu_from_pci(iommu);
665 init_iommu_from_acpi(iommu, h);
666 init_iommu_devices(iommu);
667
668 return 0;
669}
670
671/*
672 * Iterates over all IOMMU entries in the ACPI table, allocates the
673 * IOMMU structure and initializes it with init_iommu_one()
674 */
675static int __init init_iommu_all(struct acpi_table_header *table)
676{
677 u8 *p = (u8 *)table, *end = (u8 *)table;
678 struct ivhd_header *h;
679 struct amd_iommu *iommu;
680 int ret;
681
682 end += table->length;
683 p += IVRS_HEADER_LENGTH;
684
685 while (p < end) {
686 h = (struct ivhd_header *)p;
687 switch (*p) {
688 case ACPI_IVHD_TYPE:
689 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
690 if (iommu == NULL)
691 return -ENOMEM;
692 ret = init_iommu_one(iommu, h);
693 if (ret)
694 return ret;
695 break;
696 default:
697 break;
698 }
699 p += h->length;
700
701 }
702 WARN_ON(p != end);
703
704 return 0;
705}
706
707/****************************************************************************
708 *
709 * The next functions belong to the third pass of parsing the ACPI
710 * table. In this last pass the memory mapping requirements are
711 * gathered (like exclusion and unity mapping reanges).
712 *
713 ****************************************************************************/
714
715static void __init free_unity_maps(void)
716{
717 struct unity_map_entry *entry, *next;
718
719 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
720 list_del(&entry->list);
721 kfree(entry);
722 }
723}
724
725/* called when we find an exclusion range definition in ACPI */
726static int __init init_exclusion_range(struct ivmd_header *m)
727{
728 int i;
729
730 switch (m->type) {
731 case ACPI_IVMD_TYPE:
732 set_device_exclusion_range(m->devid, m);
733 break;
734 case ACPI_IVMD_TYPE_ALL:
735 for (i = 0; i <= amd_iommu_last_bdf; ++i)
736 set_device_exclusion_range(i, m);
737 break;
738 case ACPI_IVMD_TYPE_RANGE:
739 for (i = m->devid; i <= m->aux; ++i)
740 set_device_exclusion_range(i, m);
741 break;
742 default:
743 break;
744 }
745
746 return 0;
747}
748
749/* called for unity map ACPI definition */
750static int __init init_unity_map_range(struct ivmd_header *m)
751{
752 struct unity_map_entry *e = 0;
753
754 e = kzalloc(sizeof(*e), GFP_KERNEL);
755 if (e == NULL)
756 return -ENOMEM;
757
758 switch (m->type) {
759 default:
760 case ACPI_IVMD_TYPE:
761 e->devid_start = e->devid_end = m->devid;
762 break;
763 case ACPI_IVMD_TYPE_ALL:
764 e->devid_start = 0;
765 e->devid_end = amd_iommu_last_bdf;
766 break;
767 case ACPI_IVMD_TYPE_RANGE:
768 e->devid_start = m->devid;
769 e->devid_end = m->aux;
770 break;
771 }
772 e->address_start = PAGE_ALIGN(m->range_start);
773 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
774 e->prot = m->flags >> 1;
775
776 list_add_tail(&e->list, &amd_iommu_unity_map);
777
778 return 0;
779}
780
781/* iterates over all memory definitions we find in the ACPI table */
782static int __init init_memory_definitions(struct acpi_table_header *table)
783{
784 u8 *p = (u8 *)table, *end = (u8 *)table;
785 struct ivmd_header *m;
786
787 end += table->length;
788 p += IVRS_HEADER_LENGTH;
789
790 while (p < end) {
791 m = (struct ivmd_header *)p;
792 if (m->flags & IVMD_FLAG_EXCL_RANGE)
793 init_exclusion_range(m);
794 else if (m->flags & IVMD_FLAG_UNITY_MAP)
795 init_unity_map_range(m);
796
797 p += m->length;
798 }
799
800 return 0;
801}
802
803/*
804 * This function finally enables all IOMMUs found in the system after
805 * they have been initialized
806 */
807static void __init enable_iommus(void)
808{
809 struct amd_iommu *iommu;
810
811 list_for_each_entry(iommu, &amd_iommu_list, list) {
812 iommu_set_exclusion_range(iommu);
813 iommu_enable(iommu);
814 }
815}
816
817/*
818 * Suspend/Resume support
819 * disable suspend until real resume implemented
820 */
821
822static int amd_iommu_resume(struct sys_device *dev)
823{
824 return 0;
825}
826
827static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
828{
829 return -EINVAL;
830}
831
832static struct sysdev_class amd_iommu_sysdev_class = {
833 .name = "amd_iommu",
834 .suspend = amd_iommu_suspend,
835 .resume = amd_iommu_resume,
836};
837
838static struct sys_device device_amd_iommu = {
839 .id = 0,
840 .cls = &amd_iommu_sysdev_class,
841};
842
843/*
844 * This is the core init function for AMD IOMMU hardware in the system.
845 * This function is called from the generic x86 DMA layer initialization
846 * code.
847 *
848 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
849 * three times:
850 *
851 * 1 pass) Find the highest PCI device id the driver has to handle.
852 * Upon this information the size of the data structures is
853 * determined that needs to be allocated.
854 *
855 * 2 pass) Initialize the data structures just allocated with the
856 * information in the ACPI table about available AMD IOMMUs
857 * in the system. It also maps the PCI devices in the
858 * system to specific IOMMUs
859 *
860 * 3 pass) After the basic data structures are allocated and
861 * initialized we update them with information about memory
862 * remapping requirements parsed out of the ACPI table in
863 * this last pass.
864 *
865 * After that the hardware is initialized and ready to go. In the last
866 * step we do some Linux specific things like registering the driver in
867 * the dma_ops interface and initializing the suspend/resume support
868 * functions. Finally it prints some information about AMD IOMMUs and
869 * the driver state and enables the hardware.
870 */
871int __init amd_iommu_init(void)
872{
873 int i, ret = 0;
874
875
876 if (no_iommu) {
877 printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n");
878 return 0;
879 }
880
881 if (!amd_iommu_detected)
882 return -ENODEV;
883
884 /*
885 * First parse ACPI tables to find the largest Bus/Dev/Func
886 * we need to handle. Upon this information the shared data
887 * structures for the IOMMUs in the system will be allocated
888 */
889 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
890 return -ENODEV;
891
892 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
893 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
894 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
895
896 ret = -ENOMEM;
897
898 /* Device table - directly used by all IOMMUs */
899 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
900 get_order(dev_table_size));
901 if (amd_iommu_dev_table == NULL)
902 goto out;
903
904 /*
905 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
906 * IOMMU see for that device
907 */
908 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
909 get_order(alias_table_size));
910 if (amd_iommu_alias_table == NULL)
911 goto free;
912
913 /* IOMMU rlookup table - find the IOMMU for a specific device */
914 amd_iommu_rlookup_table = (void *)__get_free_pages(GFP_KERNEL,
915 get_order(rlookup_table_size));
916 if (amd_iommu_rlookup_table == NULL)
917 goto free;
918
919 /*
920 * Protection Domain table - maps devices to protection domains
921 * This table has the same size as the rlookup_table
922 */
923 amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
924 get_order(rlookup_table_size));
925 if (amd_iommu_pd_table == NULL)
926 goto free;
927
928 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
929 GFP_KERNEL | __GFP_ZERO,
930 get_order(MAX_DOMAIN_ID/8));
931 if (amd_iommu_pd_alloc_bitmap == NULL)
932 goto free;
933
934 /*
935 * let all alias entries point to itself
936 */
937 for (i = 0; i <= amd_iommu_last_bdf; ++i)
938 amd_iommu_alias_table[i] = i;
939
940 /*
941 * never allocate domain 0 because its used as the non-allocated and
942 * error value placeholder
943 */
944 amd_iommu_pd_alloc_bitmap[0] = 1;
945
946 /*
947 * now the data structures are allocated and basically initialized
948 * start the real acpi table scan
949 */
950 ret = -ENODEV;
951 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
952 goto free;
953
954 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
955 goto free;
956
957 ret = amd_iommu_init_dma_ops();
958 if (ret)
959 goto free;
960
961 ret = sysdev_class_register(&amd_iommu_sysdev_class);
962 if (ret)
963 goto free;
964
965 ret = sysdev_register(&device_amd_iommu);
966 if (ret)
967 goto free;
968
969 enable_iommus();
970
971 printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n",
972 (1 << (amd_iommu_aperture_order-20)));
973
974 printk(KERN_INFO "AMD IOMMU: device isolation ");
975 if (amd_iommu_isolate)
976 printk("enabled\n");
977 else
978 printk("disabled\n");
979
980out:
981 return ret;
982
983free:
984 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 1);
985
986 free_pages((unsigned long)amd_iommu_pd_table,
987 get_order(rlookup_table_size));
988
989 free_pages((unsigned long)amd_iommu_rlookup_table,
990 get_order(rlookup_table_size));
991
992 free_pages((unsigned long)amd_iommu_alias_table,
993 get_order(alias_table_size));
994
995 free_pages((unsigned long)amd_iommu_dev_table,
996 get_order(dev_table_size));
997
998 free_iommu_all();
999
1000 free_unity_maps();
1001
1002 goto out;
1003}
1004
1005/****************************************************************************
1006 *
1007 * Early detect code. This code runs at IOMMU detection time in the DMA
1008 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1009 * IOMMUs
1010 *
1011 ****************************************************************************/
1012static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1013{
1014 return 0;
1015}
1016
1017void __init amd_iommu_detect(void)
1018{
1019 if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
1020 return;
1021
1022 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1023 iommu_detected = 1;
1024 amd_iommu_detected = 1;
1025#ifdef CONFIG_GART_IOMMU
1026 gart_iommu_aperture_disabled = 1;
1027 gart_iommu_aperture = 0;
1028#endif
1029 }
1030}
1031
1032/****************************************************************************
1033 *
1034 * Parsing functions for the AMD IOMMU specific kernel command line
1035 * options.
1036 *
1037 ****************************************************************************/
1038
1039static int __init parse_amd_iommu_options(char *str)
1040{
1041 for (; *str; ++str) {
1042 if (strcmp(str, "isolate") == 0)
1043 amd_iommu_isolate = 1;
1044 }
1045
1046 return 1;
1047}
1048
1049static int __init parse_amd_iommu_size_options(char *str)
1050{
1051 unsigned order = PAGE_SHIFT + get_order(memparse(str, &str));
1052
1053 if ((order > 24) && (order < 31))
1054 amd_iommu_aperture_order = order;
1055
1056 return 1;
1057}
1058
1059__setup("amd_iommu=", parse_amd_iommu_options);
1060__setup("amd_iommu_size=", parse_amd_iommu_size_options);
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 479926d9e004..44e21826db11 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -21,6 +21,7 @@
21#include <linux/suspend.h> 21#include <linux/suspend.h>
22#include <asm/e820.h> 22#include <asm/e820.h>
23#include <asm/io.h> 23#include <asm/io.h>
24#include <asm/iommu.h>
24#include <asm/gart.h> 25#include <asm/gart.h>
25#include <asm/pci-direct.h> 26#include <asm/pci-direct.h>
26#include <asm/dma.h> 27#include <asm/dma.h>
@@ -35,6 +36,18 @@ int fallback_aper_force __initdata;
35 36
36int fix_aperture __initdata = 1; 37int fix_aperture __initdata = 1;
37 38
39struct bus_dev_range {
40 int bus;
41 int dev_base;
42 int dev_limit;
43};
44
45static struct bus_dev_range bus_dev_ranges[] __initdata = {
46 { 0x00, 0x18, 0x20},
47 { 0xff, 0x00, 0x20},
48 { 0xfe, 0x00, 0x20}
49};
50
38static struct resource gart_resource = { 51static struct resource gart_resource = {
39 .name = "GART", 52 .name = "GART",
40 .flags = IORESOURCE_MEM, 53 .flags = IORESOURCE_MEM,
@@ -55,8 +68,9 @@ static u32 __init allocate_aperture(void)
55 u32 aper_size; 68 u32 aper_size;
56 void *p; 69 void *p;
57 70
58 if (fallback_aper_order > 7) 71 /* aper_size should <= 1G */
59 fallback_aper_order = 7; 72 if (fallback_aper_order > 5)
73 fallback_aper_order = 5;
60 aper_size = (32 * 1024 * 1024) << fallback_aper_order; 74 aper_size = (32 * 1024 * 1024) << fallback_aper_order;
61 75
62 /* 76 /*
@@ -65,7 +79,20 @@ static u32 __init allocate_aperture(void)
65 * memory. Unfortunately we cannot move it up because that would 79 * memory. Unfortunately we cannot move it up because that would
66 * make the IOMMU useless. 80 * make the IOMMU useless.
67 */ 81 */
68 p = __alloc_bootmem_nopanic(aper_size, aper_size, 0); 82 /*
83 * using 512M as goal, in case kexec will load kernel_big
84 * that will do the on position decompress, and could overlap with
85 * that positon with gart that is used.
86 * sequende:
87 * kernel_small
88 * ==> kexec (with kdump trigger path or previous doesn't shutdown gart)
89 * ==> kernel_small(gart area become e820_reserved)
90 * ==> kexec (with kdump trigger path or previous doesn't shutdown gart)
91 * ==> kerne_big (uncompressed size will be big than 64M or 128M)
92 * so don't use 512M below as gart iommu, leave the space for kernel
93 * code for safe
94 */
95 p = __alloc_bootmem_nopanic(aper_size, aper_size, 512ULL<<20);
69 if (!p || __pa(p)+aper_size > 0xffffffff) { 96 if (!p || __pa(p)+aper_size > 0xffffffff) {
70 printk(KERN_ERR 97 printk(KERN_ERR
71 "Cannot allocate aperture memory hole (%p,%uK)\n", 98 "Cannot allocate aperture memory hole (%p,%uK)\n",
@@ -83,69 +110,53 @@ static u32 __init allocate_aperture(void)
83 return (u32)__pa(p); 110 return (u32)__pa(p);
84} 111}
85 112
86static int __init aperture_valid(u64 aper_base, u32 aper_size)
87{
88 if (!aper_base)
89 return 0;
90
91 if (aper_base + aper_size > 0x100000000UL) {
92 printk(KERN_ERR "Aperture beyond 4GB. Ignoring.\n");
93 return 0;
94 }
95 if (e820_any_mapped(aper_base, aper_base + aper_size, E820_RAM)) {
96 printk(KERN_ERR "Aperture pointing to e820 RAM. Ignoring.\n");
97 return 0;
98 }
99 if (aper_size < 64*1024*1024) {
100 printk(KERN_ERR "Aperture too small (%d MB)\n", aper_size>>20);
101 return 0;
102 }
103
104 return 1;
105}
106 113
107/* Find a PCI capability */ 114/* Find a PCI capability */
108static __u32 __init find_cap(int num, int slot, int func, int cap) 115static u32 __init find_cap(int bus, int slot, int func, int cap)
109{ 116{
110 int bytes; 117 int bytes;
111 u8 pos; 118 u8 pos;
112 119
113 if (!(read_pci_config_16(num, slot, func, PCI_STATUS) & 120 if (!(read_pci_config_16(bus, slot, func, PCI_STATUS) &
114 PCI_STATUS_CAP_LIST)) 121 PCI_STATUS_CAP_LIST))
115 return 0; 122 return 0;
116 123
117 pos = read_pci_config_byte(num, slot, func, PCI_CAPABILITY_LIST); 124 pos = read_pci_config_byte(bus, slot, func, PCI_CAPABILITY_LIST);
118 for (bytes = 0; bytes < 48 && pos >= 0x40; bytes++) { 125 for (bytes = 0; bytes < 48 && pos >= 0x40; bytes++) {
119 u8 id; 126 u8 id;
120 127
121 pos &= ~3; 128 pos &= ~3;
122 id = read_pci_config_byte(num, slot, func, pos+PCI_CAP_LIST_ID); 129 id = read_pci_config_byte(bus, slot, func, pos+PCI_CAP_LIST_ID);
123 if (id == 0xff) 130 if (id == 0xff)
124 break; 131 break;
125 if (id == cap) 132 if (id == cap)
126 return pos; 133 return pos;
127 pos = read_pci_config_byte(num, slot, func, 134 pos = read_pci_config_byte(bus, slot, func,
128 pos+PCI_CAP_LIST_NEXT); 135 pos+PCI_CAP_LIST_NEXT);
129 } 136 }
130 return 0; 137 return 0;
131} 138}
132 139
133/* Read a standard AGPv3 bridge header */ 140/* Read a standard AGPv3 bridge header */
134static __u32 __init read_agp(int num, int slot, int func, int cap, u32 *order) 141static u32 __init read_agp(int bus, int slot, int func, int cap, u32 *order)
135{ 142{
136 u32 apsize; 143 u32 apsize;
137 u32 apsizereg; 144 u32 apsizereg;
138 int nbits; 145 int nbits;
139 u32 aper_low, aper_hi; 146 u32 aper_low, aper_hi;
140 u64 aper; 147 u64 aper;
148 u32 old_order;
141 149
142 printk(KERN_INFO "AGP bridge at %02x:%02x:%02x\n", num, slot, func); 150 printk(KERN_INFO "AGP bridge at %02x:%02x:%02x\n", bus, slot, func);
143 apsizereg = read_pci_config_16(num, slot, func, cap + 0x14); 151 apsizereg = read_pci_config_16(bus, slot, func, cap + 0x14);
144 if (apsizereg == 0xffffffff) { 152 if (apsizereg == 0xffffffff) {
145 printk(KERN_ERR "APSIZE in AGP bridge unreadable\n"); 153 printk(KERN_ERR "APSIZE in AGP bridge unreadable\n");
146 return 0; 154 return 0;
147 } 155 }
148 156
157 /* old_order could be the value from NB gart setting */
158 old_order = *order;
159
149 apsize = apsizereg & 0xfff; 160 apsize = apsizereg & 0xfff;
150 /* Some BIOS use weird encodings not in the AGPv3 table. */ 161 /* Some BIOS use weird encodings not in the AGPv3 table. */
151 if (apsize & 0xff) 162 if (apsize & 0xff)
@@ -155,14 +166,26 @@ static __u32 __init read_agp(int num, int slot, int func, int cap, u32 *order)
155 if ((int)*order < 0) /* < 32MB */ 166 if ((int)*order < 0) /* < 32MB */
156 *order = 0; 167 *order = 0;
157 168
158 aper_low = read_pci_config(num, slot, func, 0x10); 169 aper_low = read_pci_config(bus, slot, func, 0x10);
159 aper_hi = read_pci_config(num, slot, func, 0x14); 170 aper_hi = read_pci_config(bus, slot, func, 0x14);
160 aper = (aper_low & ~((1<<22)-1)) | ((u64)aper_hi << 32); 171 aper = (aper_low & ~((1<<22)-1)) | ((u64)aper_hi << 32);
161 172
173 /*
174 * On some sick chips, APSIZE is 0. It means it wants 4G
175 * so let double check that order, and lets trust AMD NB settings:
176 */
177 printk(KERN_INFO "Aperture from AGP @ %Lx old size %u MB\n",
178 aper, 32 << old_order);
179 if (aper + (32ULL<<(20 + *order)) > 0x100000000ULL) {
180 printk(KERN_INFO "Aperture size %u MB (APSIZE %x) is not right, using settings from NB\n",
181 32 << *order, apsizereg);
182 *order = old_order;
183 }
184
162 printk(KERN_INFO "Aperture from AGP @ %Lx size %u MB (APSIZE %x)\n", 185 printk(KERN_INFO "Aperture from AGP @ %Lx size %u MB (APSIZE %x)\n",
163 aper, 32 << *order, apsizereg); 186 aper, 32 << *order, apsizereg);
164 187
165 if (!aperture_valid(aper, (32*1024*1024) << *order)) 188 if (!aperture_valid(aper, (32*1024*1024) << *order, 32<<20))
166 return 0; 189 return 0;
167 return (u32)aper; 190 return (u32)aper;
168} 191}
@@ -180,17 +203,17 @@ static __u32 __init read_agp(int num, int slot, int func, int cap, u32 *order)
180 * the AGP bridges should be always an own bus on the HT hierarchy, 203 * the AGP bridges should be always an own bus on the HT hierarchy,
181 * but do it here for future safety. 204 * but do it here for future safety.
182 */ 205 */
183static __u32 __init search_agp_bridge(u32 *order, int *valid_agp) 206static u32 __init search_agp_bridge(u32 *order, int *valid_agp)
184{ 207{
185 int num, slot, func; 208 int bus, slot, func;
186 209
187 /* Poor man's PCI discovery */ 210 /* Poor man's PCI discovery */
188 for (num = 0; num < 256; num++) { 211 for (bus = 0; bus < 256; bus++) {
189 for (slot = 0; slot < 32; slot++) { 212 for (slot = 0; slot < 32; slot++) {
190 for (func = 0; func < 8; func++) { 213 for (func = 0; func < 8; func++) {
191 u32 class, cap; 214 u32 class, cap;
192 u8 type; 215 u8 type;
193 class = read_pci_config(num, slot, func, 216 class = read_pci_config(bus, slot, func,
194 PCI_CLASS_REVISION); 217 PCI_CLASS_REVISION);
195 if (class == 0xffffffff) 218 if (class == 0xffffffff)
196 break; 219 break;
@@ -199,17 +222,17 @@ static __u32 __init search_agp_bridge(u32 *order, int *valid_agp)
199 case PCI_CLASS_BRIDGE_HOST: 222 case PCI_CLASS_BRIDGE_HOST:
200 case PCI_CLASS_BRIDGE_OTHER: /* needed? */ 223 case PCI_CLASS_BRIDGE_OTHER: /* needed? */
201 /* AGP bridge? */ 224 /* AGP bridge? */
202 cap = find_cap(num, slot, func, 225 cap = find_cap(bus, slot, func,
203 PCI_CAP_ID_AGP); 226 PCI_CAP_ID_AGP);
204 if (!cap) 227 if (!cap)
205 break; 228 break;
206 *valid_agp = 1; 229 *valid_agp = 1;
207 return read_agp(num, slot, func, cap, 230 return read_agp(bus, slot, func, cap,
208 order); 231 order);
209 } 232 }
210 233
211 /* No multi-function device? */ 234 /* No multi-function device? */
212 type = read_pci_config_byte(num, slot, func, 235 type = read_pci_config_byte(bus, slot, func,
213 PCI_HEADER_TYPE); 236 PCI_HEADER_TYPE);
214 if (!(type & 0x80)) 237 if (!(type & 0x80))
215 break; 238 break;
@@ -249,36 +272,50 @@ void __init early_gart_iommu_check(void)
249 * or BIOS forget to put that in reserved. 272 * or BIOS forget to put that in reserved.
250 * try to update e820 to make that region as reserved. 273 * try to update e820 to make that region as reserved.
251 */ 274 */
252 int fix, num; 275 int i, fix, slot;
253 u32 ctl; 276 u32 ctl;
254 u32 aper_size = 0, aper_order = 0, last_aper_order = 0; 277 u32 aper_size = 0, aper_order = 0, last_aper_order = 0;
255 u64 aper_base = 0, last_aper_base = 0; 278 u64 aper_base = 0, last_aper_base = 0;
256 int aper_enabled = 0, last_aper_enabled = 0; 279 int aper_enabled = 0, last_aper_enabled = 0, last_valid = 0;
257 280
258 if (!early_pci_allowed()) 281 if (!early_pci_allowed())
259 return; 282 return;
260 283
284 /* This is mostly duplicate of iommu_hole_init */
261 fix = 0; 285 fix = 0;
262 for (num = 24; num < 32; num++) { 286 for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
263 if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00))) 287 int bus;
264 continue; 288 int dev_base, dev_limit;
265 289
266 ctl = read_pci_config(0, num, 3, 0x90); 290 bus = bus_dev_ranges[i].bus;
267 aper_enabled = ctl & 1; 291 dev_base = bus_dev_ranges[i].dev_base;
268 aper_order = (ctl >> 1) & 7; 292 dev_limit = bus_dev_ranges[i].dev_limit;
269 aper_size = (32 * 1024 * 1024) << aper_order; 293
270 aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff; 294 for (slot = dev_base; slot < dev_limit; slot++) {
271 aper_base <<= 25; 295 if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
272 296 continue;
273 if ((last_aper_order && aper_order != last_aper_order) || 297
274 (last_aper_base && aper_base != last_aper_base) || 298 ctl = read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL);
275 (last_aper_enabled && aper_enabled != last_aper_enabled)) { 299 aper_enabled = ctl & AMD64_GARTEN;
276 fix = 1; 300 aper_order = (ctl >> 1) & 7;
277 break; 301 aper_size = (32 * 1024 * 1024) << aper_order;
302 aper_base = read_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE) & 0x7fff;
303 aper_base <<= 25;
304
305 if (last_valid) {
306 if ((aper_order != last_aper_order) ||
307 (aper_base != last_aper_base) ||
308 (aper_enabled != last_aper_enabled)) {
309 fix = 1;
310 break;
311 }
312 }
313
314 last_aper_order = aper_order;
315 last_aper_base = aper_base;
316 last_aper_enabled = aper_enabled;
317 last_valid = 1;
278 } 318 }
279 last_aper_order = aper_order;
280 last_aper_base = aper_base;
281 last_aper_enabled = aper_enabled;
282 } 319 }
283 320
284 if (!fix && !aper_enabled) 321 if (!fix && !aper_enabled)
@@ -290,32 +327,46 @@ void __init early_gart_iommu_check(void)
290 if (gart_fix_e820 && !fix && aper_enabled) { 327 if (gart_fix_e820 && !fix && aper_enabled) {
291 if (e820_any_mapped(aper_base, aper_base + aper_size, 328 if (e820_any_mapped(aper_base, aper_base + aper_size,
292 E820_RAM)) { 329 E820_RAM)) {
293 /* reserved it, so we can resuse it in second kernel */ 330 /* reserve it, so we can reuse it in second kernel */
294 printk(KERN_INFO "update e820 for GART\n"); 331 printk(KERN_INFO "update e820 for GART\n");
295 add_memory_region(aper_base, aper_size, E820_RESERVED); 332 e820_add_region(aper_base, aper_size, E820_RESERVED);
296 update_e820(); 333 update_e820();
297 } 334 }
298 return;
299 } 335 }
300 336
337 if (!fix)
338 return;
339
301 /* different nodes have different setting, disable them all at first*/ 340 /* different nodes have different setting, disable them all at first*/
302 for (num = 24; num < 32; num++) { 341 for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
303 if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00))) 342 int bus;
304 continue; 343 int dev_base, dev_limit;
344
345 bus = bus_dev_ranges[i].bus;
346 dev_base = bus_dev_ranges[i].dev_base;
347 dev_limit = bus_dev_ranges[i].dev_limit;
348
349 for (slot = dev_base; slot < dev_limit; slot++) {
350 if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
351 continue;
305 352
306 ctl = read_pci_config(0, num, 3, 0x90); 353 ctl = read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL);
307 ctl &= ~1; 354 ctl &= ~AMD64_GARTEN;
308 write_pci_config(0, num, 3, 0x90, ctl); 355 write_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL, ctl);
356 }
309 } 357 }
310 358
311} 359}
312 360
361static int __initdata printed_gart_size_msg;
362
313void __init gart_iommu_hole_init(void) 363void __init gart_iommu_hole_init(void)
314{ 364{
365 u32 agp_aper_base = 0, agp_aper_order = 0;
315 u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0; 366 u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
316 u64 aper_base, last_aper_base = 0; 367 u64 aper_base, last_aper_base = 0;
317 int fix, num, valid_agp = 0; 368 int fix, slot, valid_agp = 0;
318 int node; 369 int i, node;
319 370
320 if (gart_iommu_aperture_disabled || !fix_aperture || 371 if (gart_iommu_aperture_disabled || !fix_aperture ||
321 !early_pci_allowed()) 372 !early_pci_allowed())
@@ -323,38 +374,65 @@ void __init gart_iommu_hole_init(void)
323 374
324 printk(KERN_INFO "Checking aperture...\n"); 375 printk(KERN_INFO "Checking aperture...\n");
325 376
377 if (!fallback_aper_force)
378 agp_aper_base = search_agp_bridge(&agp_aper_order, &valid_agp);
379
326 fix = 0; 380 fix = 0;
327 node = 0; 381 node = 0;
328 for (num = 24; num < 32; num++) { 382 for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
329 if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00))) 383 int bus;
330 continue; 384 int dev_base, dev_limit;
331 385
332 iommu_detected = 1; 386 bus = bus_dev_ranges[i].bus;
333 gart_iommu_aperture = 1; 387 dev_base = bus_dev_ranges[i].dev_base;
334 388 dev_limit = bus_dev_ranges[i].dev_limit;
335 aper_order = (read_pci_config(0, num, 3, 0x90) >> 1) & 7; 389
336 aper_size = (32 * 1024 * 1024) << aper_order; 390 for (slot = dev_base; slot < dev_limit; slot++) {
337 aper_base = read_pci_config(0, num, 3, 0x94) & 0x7fff; 391 if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
338 aper_base <<= 25; 392 continue;
339 393
340 printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n", 394 iommu_detected = 1;
341 node, aper_base, aper_size >> 20); 395 gart_iommu_aperture = 1;
342 node++; 396
343 397 aper_order = (read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL) >> 1) & 7;
344 if (!aperture_valid(aper_base, aper_size)) { 398 aper_size = (32 * 1024 * 1024) << aper_order;
345 fix = 1; 399 aper_base = read_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE) & 0x7fff;
346 break; 400 aper_base <<= 25;
347 } 401
402 printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n",
403 node, aper_base, aper_size >> 20);
404 node++;
405
406 if (!aperture_valid(aper_base, aper_size, 64<<20)) {
407 if (valid_agp && agp_aper_base &&
408 agp_aper_base == aper_base &&
409 agp_aper_order == aper_order) {
410 /* the same between two setting from NB and agp */
411 if (!no_iommu &&
412 max_pfn > MAX_DMA32_PFN &&
413 !printed_gart_size_msg) {
414 printk(KERN_ERR "you are using iommu with agp, but GART size is less than 64M\n");
415 printk(KERN_ERR "please increase GART size in your BIOS setup\n");
416 printk(KERN_ERR "if BIOS doesn't have that option, contact your HW vendor!\n");
417 printed_gart_size_msg = 1;
418 }
419 } else {
420 fix = 1;
421 goto out;
422 }
423 }
348 424
349 if ((last_aper_order && aper_order != last_aper_order) || 425 if ((last_aper_order && aper_order != last_aper_order) ||
350 (last_aper_base && aper_base != last_aper_base)) { 426 (last_aper_base && aper_base != last_aper_base)) {
351 fix = 1; 427 fix = 1;
352 break; 428 goto out;
429 }
430 last_aper_order = aper_order;
431 last_aper_base = aper_base;
353 } 432 }
354 last_aper_order = aper_order;
355 last_aper_base = aper_base;
356 } 433 }
357 434
435out:
358 if (!fix && !fallback_aper_force) { 436 if (!fix && !fallback_aper_force) {
359 if (last_aper_base) { 437 if (last_aper_base) {
360 unsigned long n = (32 * 1024 * 1024) << last_aper_order; 438 unsigned long n = (32 * 1024 * 1024) << last_aper_order;
@@ -364,14 +442,16 @@ void __init gart_iommu_hole_init(void)
364 return; 442 return;
365 } 443 }
366 444
367 if (!fallback_aper_force) 445 if (!fallback_aper_force) {
368 aper_alloc = search_agp_bridge(&aper_order, &valid_agp); 446 aper_alloc = agp_aper_base;
447 aper_order = agp_aper_order;
448 }
369 449
370 if (aper_alloc) { 450 if (aper_alloc) {
371 /* Got the aperture from the AGP bridge */ 451 /* Got the aperture from the AGP bridge */
372 } else if (swiotlb && !valid_agp) { 452 } else if (swiotlb && !valid_agp) {
373 /* Do nothing */ 453 /* Do nothing */
374 } else if ((!no_iommu && end_pfn > MAX_DMA32_PFN) || 454 } else if ((!no_iommu && max_pfn > MAX_DMA32_PFN) ||
375 force_iommu || 455 force_iommu ||
376 valid_agp || 456 valid_agp ||
377 fallback_aper_force) { 457 fallback_aper_force) {
@@ -401,16 +481,24 @@ void __init gart_iommu_hole_init(void)
401 } 481 }
402 482
403 /* Fix up the north bridges */ 483 /* Fix up the north bridges */
404 for (num = 24; num < 32; num++) { 484 for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
405 if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00))) 485 int bus;
406 continue; 486 int dev_base, dev_limit;
407 487
408 /* 488 bus = bus_dev_ranges[i].bus;
409 * Don't enable translation yet. That is done later. 489 dev_base = bus_dev_ranges[i].dev_base;
410 * Assume this BIOS didn't initialise the GART so 490 dev_limit = bus_dev_ranges[i].dev_limit;
411 * just overwrite all previous bits 491 for (slot = dev_base; slot < dev_limit; slot++) {
412 */ 492 if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
413 write_pci_config(0, num, 3, 0x90, aper_order<<1); 493 continue;
414 write_pci_config(0, num, 3, 0x94, aper_alloc>>25); 494
495 /* Don't enable translation yet. That is done later.
496 Assume this BIOS didn't initialise the GART so
497 just overwrite all previous bits */
498 write_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL, aper_order << 1);
499 write_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE, aper_alloc >> 25);
500 }
415 } 501 }
502
503 set_up_gart_resume(aper_order, aper_alloc);
416} 504}
diff --git a/arch/x86/kernel/apic_32.c b/arch/x86/kernel/apic_32.c
index 4b99b1bdeb6c..d6c898358371 100644
--- a/arch/x86/kernel/apic_32.c
+++ b/arch/x86/kernel/apic_32.c
@@ -52,29 +52,40 @@
52 52
53unsigned long mp_lapic_addr; 53unsigned long mp_lapic_addr;
54 54
55DEFINE_PER_CPU(u16, x86_bios_cpu_apicid) = BAD_APICID;
56EXPORT_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
57
58/* 55/*
59 * Knob to control our willingness to enable the local APIC. 56 * Knob to control our willingness to enable the local APIC.
60 * 57 *
61 * -1=force-disable, +1=force-enable 58 * +1=force-enable
62 */ 59 */
63static int enable_local_apic __initdata; 60static int force_enable_local_apic;
61int disable_apic;
64 62
65/* Local APIC timer verification ok */ 63/* Local APIC timer verification ok */
66static int local_apic_timer_verify_ok; 64static int local_apic_timer_verify_ok;
67/* Disable local APIC timer from the kernel commandline or via dmi quirk 65/* Disable local APIC timer from the kernel commandline or via dmi quirk */
68 or using CPU MSR check */ 66static int local_apic_timer_disabled;
69int local_apic_timer_disabled;
70/* Local APIC timer works in C2 */ 67/* Local APIC timer works in C2 */
71int local_apic_timer_c2_ok; 68int local_apic_timer_c2_ok;
72EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok); 69EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
73 70
71int first_system_vector = 0xfe;
72
73char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};
74
74/* 75/*
75 * Debug level, exported for io_apic.c 76 * Debug level, exported for io_apic.c
76 */ 77 */
77int apic_verbosity; 78unsigned int apic_verbosity;
79
80int pic_mode;
81
82/* Have we found an MP table */
83int smp_found_config;
84
85static struct resource lapic_resource = {
86 .name = "Local APIC",
87 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
88};
78 89
79static unsigned int calibration_result; 90static unsigned int calibration_result;
80 91
@@ -166,7 +177,7 @@ void __cpuinit enable_NMI_through_LVT0(void)
166 /* Level triggered for 82489DX */ 177 /* Level triggered for 82489DX */
167 if (!lapic_is_integrated()) 178 if (!lapic_is_integrated())
168 v |= APIC_LVT_LEVEL_TRIGGER; 179 v |= APIC_LVT_LEVEL_TRIGGER;
169 apic_write_around(APIC_LVT0, v); 180 apic_write(APIC_LVT0, v);
170} 181}
171 182
172/** 183/**
@@ -201,9 +212,6 @@ int lapic_get_maxlvt(void)
201 * this function twice on the boot CPU, once with a bogus timeout 212 * this function twice on the boot CPU, once with a bogus timeout
202 * value, second time for real. The other (noncalibrating) CPUs 213 * value, second time for real. The other (noncalibrating) CPUs
203 * call this function only once, with the real, calibrated value. 214 * call this function only once, with the real, calibrated value.
204 *
205 * We do reads before writes even if unnecessary, to get around the
206 * P5 APIC double write bug.
207 */ 215 */
208static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen) 216static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
209{ 217{
@@ -218,18 +226,18 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
218 if (!irqen) 226 if (!irqen)
219 lvtt_value |= APIC_LVT_MASKED; 227 lvtt_value |= APIC_LVT_MASKED;
220 228
221 apic_write_around(APIC_LVTT, lvtt_value); 229 apic_write(APIC_LVTT, lvtt_value);
222 230
223 /* 231 /*
224 * Divide PICLK by 16 232 * Divide PICLK by 16
225 */ 233 */
226 tmp_value = apic_read(APIC_TDCR); 234 tmp_value = apic_read(APIC_TDCR);
227 apic_write_around(APIC_TDCR, (tmp_value 235 apic_write(APIC_TDCR,
228 & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) 236 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
229 | APIC_TDR_DIV_16); 237 APIC_TDR_DIV_16);
230 238
231 if (!oneshot) 239 if (!oneshot)
232 apic_write_around(APIC_TMICT, clocks/APIC_DIVISOR); 240 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
233} 241}
234 242
235/* 243/*
@@ -238,7 +246,7 @@ static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
238static int lapic_next_event(unsigned long delta, 246static int lapic_next_event(unsigned long delta,
239 struct clock_event_device *evt) 247 struct clock_event_device *evt)
240{ 248{
241 apic_write_around(APIC_TMICT, delta); 249 apic_write(APIC_TMICT, delta);
242 return 0; 250 return 0;
243} 251}
244 252
@@ -267,7 +275,7 @@ static void lapic_timer_setup(enum clock_event_mode mode,
267 case CLOCK_EVT_MODE_SHUTDOWN: 275 case CLOCK_EVT_MODE_SHUTDOWN:
268 v = apic_read(APIC_LVTT); 276 v = apic_read(APIC_LVTT);
269 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 277 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
270 apic_write_around(APIC_LVTT, v); 278 apic_write(APIC_LVTT, v);
271 break; 279 break;
272 case CLOCK_EVT_MODE_RESUME: 280 case CLOCK_EVT_MODE_RESUME:
273 /* Nothing to do here */ 281 /* Nothing to do here */
@@ -361,12 +369,7 @@ static void __init lapic_cal_handler(struct clock_event_device *dev)
361 } 369 }
362} 370}
363 371
364/* 372static int __init calibrate_APIC_clock(void)
365 * Setup the boot APIC
366 *
367 * Calibrate and verify the result.
368 */
369void __init setup_boot_APIC_clock(void)
370{ 373{
371 struct clock_event_device *levt = &__get_cpu_var(lapic_events); 374 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
372 const long pm_100ms = PMTMR_TICKS_PER_SEC/10; 375 const long pm_100ms = PMTMR_TICKS_PER_SEC/10;
@@ -376,24 +379,6 @@ void __init setup_boot_APIC_clock(void)
376 long delta, deltapm; 379 long delta, deltapm;
377 int pm_referenced = 0; 380 int pm_referenced = 0;
378 381
379 /*
380 * The local apic timer can be disabled via the kernel
381 * commandline or from the CPU detection code. Register the lapic
382 * timer as a dummy clock event source on SMP systems, so the
383 * broadcast mechanism is used. On UP systems simply ignore it.
384 */
385 if (local_apic_timer_disabled) {
386 /* No broadcast on UP ! */
387 if (num_possible_cpus() > 1) {
388 lapic_clockevent.mult = 1;
389 setup_APIC_timer();
390 }
391 return;
392 }
393
394 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
395 "calibrating APIC timer ...\n");
396
397 local_irq_disable(); 382 local_irq_disable();
398 383
399 /* Replace the global interrupt handler */ 384 /* Replace the global interrupt handler */
@@ -478,8 +463,6 @@ void __init setup_boot_APIC_clock(void)
478 calibration_result / (1000000 / HZ), 463 calibration_result / (1000000 / HZ),
479 calibration_result % (1000000 / HZ)); 464 calibration_result % (1000000 / HZ));
480 465
481 local_apic_timer_verify_ok = 1;
482
483 /* 466 /*
484 * Do a sanity check on the APIC calibration result 467 * Do a sanity check on the APIC calibration result
485 */ 468 */
@@ -487,12 +470,11 @@ void __init setup_boot_APIC_clock(void)
487 local_irq_enable(); 470 local_irq_enable();
488 printk(KERN_WARNING 471 printk(KERN_WARNING
489 "APIC frequency too slow, disabling apic timer\n"); 472 "APIC frequency too slow, disabling apic timer\n");
490 /* No broadcast on UP ! */ 473 return -1;
491 if (num_possible_cpus() > 1)
492 setup_APIC_timer();
493 return;
494 } 474 }
495 475
476 local_apic_timer_verify_ok = 1;
477
496 /* We trust the pm timer based calibration */ 478 /* We trust the pm timer based calibration */
497 if (!pm_referenced) { 479 if (!pm_referenced) {
498 apic_printk(APIC_VERBOSE, "... verify APIC timer\n"); 480 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
@@ -532,22 +514,55 @@ void __init setup_boot_APIC_clock(void)
532 if (!local_apic_timer_verify_ok) { 514 if (!local_apic_timer_verify_ok) {
533 printk(KERN_WARNING 515 printk(KERN_WARNING
534 "APIC timer disabled due to verification failure.\n"); 516 "APIC timer disabled due to verification failure.\n");
517 return -1;
518 }
519
520 return 0;
521}
522
523/*
524 * Setup the boot APIC
525 *
526 * Calibrate and verify the result.
527 */
528void __init setup_boot_APIC_clock(void)
529{
530 /*
531 * The local apic timer can be disabled via the kernel
532 * commandline or from the CPU detection code. Register the lapic
533 * timer as a dummy clock event source on SMP systems, so the
534 * broadcast mechanism is used. On UP systems simply ignore it.
535 */
536 if (local_apic_timer_disabled) {
535 /* No broadcast on UP ! */ 537 /* No broadcast on UP ! */
536 if (num_possible_cpus() == 1) 538 if (num_possible_cpus() > 1) {
537 return; 539 lapic_clockevent.mult = 1;
538 } else { 540 setup_APIC_timer();
539 /* 541 }
540 * If nmi_watchdog is set to IO_APIC, we need the 542 return;
541 * PIT/HPET going. Otherwise register lapic as a dummy 543 }
542 * device. 544
543 */ 545 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
544 if (nmi_watchdog != NMI_IO_APIC) 546 "calibrating APIC timer ...\n");
545 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; 547
546 else 548 if (calibrate_APIC_clock()) {
547 printk(KERN_WARNING "APIC timer registered as dummy," 549 /* No broadcast on UP ! */
548 " due to nmi_watchdog=1!\n"); 550 if (num_possible_cpus() > 1)
551 setup_APIC_timer();
552 return;
549 } 553 }
550 554
555 /*
556 * If nmi_watchdog is set to IO_APIC, we need the
557 * PIT/HPET going. Otherwise register lapic as a dummy
558 * device.
559 */
560 if (nmi_watchdog != NMI_IO_APIC)
561 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
562 else
563 printk(KERN_WARNING "APIC timer registered as dummy,"
564 " due to nmi_watchdog=%d!\n", nmi_watchdog);
565
551 /* Setup the lapic or request the broadcast */ 566 /* Setup the lapic or request the broadcast */
552 setup_APIC_timer(); 567 setup_APIC_timer();
553} 568}
@@ -682,44 +697,44 @@ void clear_local_APIC(void)
682 */ 697 */
683 if (maxlvt >= 3) { 698 if (maxlvt >= 3) {
684 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */ 699 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
685 apic_write_around(APIC_LVTERR, v | APIC_LVT_MASKED); 700 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
686 } 701 }
687 /* 702 /*
688 * Careful: we have to set masks only first to deassert 703 * Careful: we have to set masks only first to deassert
689 * any level-triggered sources. 704 * any level-triggered sources.
690 */ 705 */
691 v = apic_read(APIC_LVTT); 706 v = apic_read(APIC_LVTT);
692 apic_write_around(APIC_LVTT, v | APIC_LVT_MASKED); 707 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
693 v = apic_read(APIC_LVT0); 708 v = apic_read(APIC_LVT0);
694 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED); 709 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
695 v = apic_read(APIC_LVT1); 710 v = apic_read(APIC_LVT1);
696 apic_write_around(APIC_LVT1, v | APIC_LVT_MASKED); 711 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
697 if (maxlvt >= 4) { 712 if (maxlvt >= 4) {
698 v = apic_read(APIC_LVTPC); 713 v = apic_read(APIC_LVTPC);
699 apic_write_around(APIC_LVTPC, v | APIC_LVT_MASKED); 714 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
700 } 715 }
701 716
702 /* lets not touch this if we didn't frob it */ 717 /* lets not touch this if we didn't frob it */
703#ifdef CONFIG_X86_MCE_P4THERMAL 718#ifdef CONFIG_X86_MCE_P4THERMAL
704 if (maxlvt >= 5) { 719 if (maxlvt >= 5) {
705 v = apic_read(APIC_LVTTHMR); 720 v = apic_read(APIC_LVTTHMR);
706 apic_write_around(APIC_LVTTHMR, v | APIC_LVT_MASKED); 721 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
707 } 722 }
708#endif 723#endif
709 /* 724 /*
710 * Clean APIC state for other OSs: 725 * Clean APIC state for other OSs:
711 */ 726 */
712 apic_write_around(APIC_LVTT, APIC_LVT_MASKED); 727 apic_write(APIC_LVTT, APIC_LVT_MASKED);
713 apic_write_around(APIC_LVT0, APIC_LVT_MASKED); 728 apic_write(APIC_LVT0, APIC_LVT_MASKED);
714 apic_write_around(APIC_LVT1, APIC_LVT_MASKED); 729 apic_write(APIC_LVT1, APIC_LVT_MASKED);
715 if (maxlvt >= 3) 730 if (maxlvt >= 3)
716 apic_write_around(APIC_LVTERR, APIC_LVT_MASKED); 731 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
717 if (maxlvt >= 4) 732 if (maxlvt >= 4)
718 apic_write_around(APIC_LVTPC, APIC_LVT_MASKED); 733 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
719 734
720#ifdef CONFIG_X86_MCE_P4THERMAL 735#ifdef CONFIG_X86_MCE_P4THERMAL
721 if (maxlvt >= 5) 736 if (maxlvt >= 5)
722 apic_write_around(APIC_LVTTHMR, APIC_LVT_MASKED); 737 apic_write(APIC_LVTTHMR, APIC_LVT_MASKED);
723#endif 738#endif
724 /* Integrated APIC (!82489DX) ? */ 739 /* Integrated APIC (!82489DX) ? */
725 if (lapic_is_integrated()) { 740 if (lapic_is_integrated()) {
@@ -745,7 +760,7 @@ void disable_local_APIC(void)
745 */ 760 */
746 value = apic_read(APIC_SPIV); 761 value = apic_read(APIC_SPIV);
747 value &= ~APIC_SPIV_APIC_ENABLED; 762 value &= ~APIC_SPIV_APIC_ENABLED;
748 apic_write_around(APIC_SPIV, value); 763 apic_write(APIC_SPIV, value);
749 764
750 /* 765 /*
751 * When LAPIC was disabled by the BIOS and enabled by the kernel, 766 * When LAPIC was disabled by the BIOS and enabled by the kernel,
@@ -854,8 +869,8 @@ void __init sync_Arb_IDs(void)
854 apic_wait_icr_idle(); 869 apic_wait_icr_idle();
855 870
856 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n"); 871 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
857 apic_write_around(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG 872 apic_write(APIC_ICR,
858 | APIC_DM_INIT); 873 APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT);
859} 874}
860 875
861/* 876/*
@@ -891,16 +906,16 @@ void __init init_bsp_APIC(void)
891 else 906 else
892 value |= APIC_SPIV_FOCUS_DISABLED; 907 value |= APIC_SPIV_FOCUS_DISABLED;
893 value |= SPURIOUS_APIC_VECTOR; 908 value |= SPURIOUS_APIC_VECTOR;
894 apic_write_around(APIC_SPIV, value); 909 apic_write(APIC_SPIV, value);
895 910
896 /* 911 /*
897 * Set up the virtual wire mode. 912 * Set up the virtual wire mode.
898 */ 913 */
899 apic_write_around(APIC_LVT0, APIC_DM_EXTINT); 914 apic_write(APIC_LVT0, APIC_DM_EXTINT);
900 value = APIC_DM_NMI; 915 value = APIC_DM_NMI;
901 if (!lapic_is_integrated()) /* 82489DX */ 916 if (!lapic_is_integrated()) /* 82489DX */
902 value |= APIC_LVT_LEVEL_TRIGGER; 917 value |= APIC_LVT_LEVEL_TRIGGER;
903 apic_write_around(APIC_LVT1, value); 918 apic_write(APIC_LVT1, value);
904} 919}
905 920
906static void __cpuinit lapic_setup_esr(void) 921static void __cpuinit lapic_setup_esr(void)
@@ -915,7 +930,7 @@ static void __cpuinit lapic_setup_esr(void)
915 930
916 /* enables sending errors */ 931 /* enables sending errors */
917 value = ERROR_APIC_VECTOR; 932 value = ERROR_APIC_VECTOR;
918 apic_write_around(APIC_LVTERR, value); 933 apic_write(APIC_LVTERR, value);
919 /* 934 /*
920 * spec says clear errors after enabling vector. 935 * spec says clear errors after enabling vector.
921 */ 936 */
@@ -963,7 +978,7 @@ void __cpuinit setup_local_APIC(void)
963 * Double-check whether this APIC is really registered. 978 * Double-check whether this APIC is really registered.
964 */ 979 */
965 if (!apic_id_registered()) 980 if (!apic_id_registered())
966 BUG(); 981 WARN_ON_ONCE(1);
967 982
968 /* 983 /*
969 * Intel recommends to set DFR, LDR and TPR before enabling 984 * Intel recommends to set DFR, LDR and TPR before enabling
@@ -978,7 +993,7 @@ void __cpuinit setup_local_APIC(void)
978 */ 993 */
979 value = apic_read(APIC_TASKPRI); 994 value = apic_read(APIC_TASKPRI);
980 value &= ~APIC_TPRI_MASK; 995 value &= ~APIC_TPRI_MASK;
981 apic_write_around(APIC_TASKPRI, value); 996 apic_write(APIC_TASKPRI, value);
982 997
983 /* 998 /*
984 * After a crash, we no longer service the interrupts and a pending 999 * After a crash, we no longer service the interrupts and a pending
@@ -1036,7 +1051,7 @@ void __cpuinit setup_local_APIC(void)
1036 * Set spurious IRQ vector 1051 * Set spurious IRQ vector
1037 */ 1052 */
1038 value |= SPURIOUS_APIC_VECTOR; 1053 value |= SPURIOUS_APIC_VECTOR;
1039 apic_write_around(APIC_SPIV, value); 1054 apic_write(APIC_SPIV, value);
1040 1055
1041 /* 1056 /*
1042 * Set up LVT0, LVT1: 1057 * Set up LVT0, LVT1:
@@ -1058,7 +1073,7 @@ void __cpuinit setup_local_APIC(void)
1058 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", 1073 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n",
1059 smp_processor_id()); 1074 smp_processor_id());
1060 } 1075 }
1061 apic_write_around(APIC_LVT0, value); 1076 apic_write(APIC_LVT0, value);
1062 1077
1063 /* 1078 /*
1064 * only the BP should see the LINT1 NMI signal, obviously. 1079 * only the BP should see the LINT1 NMI signal, obviously.
@@ -1069,7 +1084,7 @@ void __cpuinit setup_local_APIC(void)
1069 value = APIC_DM_NMI | APIC_LVT_MASKED; 1084 value = APIC_DM_NMI | APIC_LVT_MASKED;
1070 if (!integrated) /* 82489DX */ 1085 if (!integrated) /* 82489DX */
1071 value |= APIC_LVT_LEVEL_TRIGGER; 1086 value |= APIC_LVT_LEVEL_TRIGGER;
1072 apic_write_around(APIC_LVT1, value); 1087 apic_write(APIC_LVT1, value);
1073} 1088}
1074 1089
1075void __cpuinit end_local_APIC_setup(void) 1090void __cpuinit end_local_APIC_setup(void)
@@ -1080,7 +1095,7 @@ void __cpuinit end_local_APIC_setup(void)
1080 /* Disable the local apic timer */ 1095 /* Disable the local apic timer */
1081 value = apic_read(APIC_LVTT); 1096 value = apic_read(APIC_LVTT);
1082 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 1097 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1083 apic_write_around(APIC_LVTT, value); 1098 apic_write(APIC_LVTT, value);
1084 1099
1085 setup_apic_nmi_watchdog(NULL); 1100 setup_apic_nmi_watchdog(NULL);
1086 apic_pm_activate(); 1101 apic_pm_activate();
@@ -1094,7 +1109,7 @@ static int __init detect_init_APIC(void)
1094 u32 h, l, features; 1109 u32 h, l, features;
1095 1110
1096 /* Disabled by kernel option? */ 1111 /* Disabled by kernel option? */
1097 if (enable_local_apic < 0) 1112 if (disable_apic)
1098 return -1; 1113 return -1;
1099 1114
1100 switch (boot_cpu_data.x86_vendor) { 1115 switch (boot_cpu_data.x86_vendor) {
@@ -1117,7 +1132,7 @@ static int __init detect_init_APIC(void)
1117 * Over-ride BIOS and try to enable the local APIC only if 1132 * Over-ride BIOS and try to enable the local APIC only if
1118 * "lapic" specified. 1133 * "lapic" specified.
1119 */ 1134 */
1120 if (enable_local_apic <= 0) { 1135 if (!force_enable_local_apic) {
1121 printk(KERN_INFO "Local APIC disabled by BIOS -- " 1136 printk(KERN_INFO "Local APIC disabled by BIOS -- "
1122 "you can enable it with \"lapic\"\n"); 1137 "you can enable it with \"lapic\"\n");
1123 return -1; 1138 return -1;
@@ -1154,9 +1169,6 @@ static int __init detect_init_APIC(void)
1154 if (l & MSR_IA32_APICBASE_ENABLE) 1169 if (l & MSR_IA32_APICBASE_ENABLE)
1155 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE; 1170 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1156 1171
1157 if (nmi_watchdog != NMI_NONE && nmi_watchdog != NMI_DISABLED)
1158 nmi_watchdog = NMI_LOCAL_APIC;
1159
1160 printk(KERN_INFO "Found and enabled local APIC!\n"); 1172 printk(KERN_INFO "Found and enabled local APIC!\n");
1161 1173
1162 apic_pm_activate(); 1174 apic_pm_activate();
@@ -1195,36 +1207,6 @@ void __init init_apic_mappings(void)
1195 if (boot_cpu_physical_apicid == -1U) 1207 if (boot_cpu_physical_apicid == -1U)
1196 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1208 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id());
1197 1209
1198#ifdef CONFIG_X86_IO_APIC
1199 {
1200 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
1201 int i;
1202
1203 for (i = 0; i < nr_ioapics; i++) {
1204 if (smp_found_config) {
1205 ioapic_phys = mp_ioapics[i].mpc_apicaddr;
1206 if (!ioapic_phys) {
1207 printk(KERN_ERR
1208 "WARNING: bogus zero IO-APIC "
1209 "address found in MPTABLE, "
1210 "disabling IO/APIC support!\n");
1211 smp_found_config = 0;
1212 skip_ioapic_setup = 1;
1213 goto fake_ioapic_page;
1214 }
1215 } else {
1216fake_ioapic_page:
1217 ioapic_phys = (unsigned long)
1218 alloc_bootmem_pages(PAGE_SIZE);
1219 ioapic_phys = __pa(ioapic_phys);
1220 }
1221 set_fixmap_nocache(idx, ioapic_phys);
1222 printk(KERN_DEBUG "mapped IOAPIC to %08lx (%08lx)\n",
1223 __fix_to_virt(idx), ioapic_phys);
1224 idx++;
1225 }
1226 }
1227#endif
1228} 1210}
1229 1211
1230/* 1212/*
@@ -1236,9 +1218,6 @@ int apic_version[MAX_APICS];
1236 1218
1237int __init APIC_init_uniprocessor(void) 1219int __init APIC_init_uniprocessor(void)
1238{ 1220{
1239 if (enable_local_apic < 0)
1240 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1241
1242 if (!smp_found_config && !cpu_has_apic) 1221 if (!smp_found_config && !cpu_has_apic)
1243 return -1; 1222 return -1;
1244 1223
@@ -1265,10 +1244,14 @@ int __init APIC_init_uniprocessor(void)
1265#ifdef CONFIG_CRASH_DUMP 1244#ifdef CONFIG_CRASH_DUMP
1266 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id()); 1245 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id());
1267#endif 1246#endif
1268 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid); 1247 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1269 1248
1270 setup_local_APIC(); 1249 setup_local_APIC();
1271 1250
1251#ifdef CONFIG_X86_IO_APIC
1252 if (!smp_found_config || skip_ioapic_setup || !nr_ioapics)
1253#endif
1254 localise_nmi_watchdog();
1272 end_local_APIC_setup(); 1255 end_local_APIC_setup();
1273#ifdef CONFIG_X86_IO_APIC 1256#ifdef CONFIG_X86_IO_APIC
1274 if (smp_found_config) 1257 if (smp_found_config)
@@ -1351,13 +1334,17 @@ void __init smp_intr_init(void)
1351 * The reschedule interrupt is a CPU-to-CPU reschedule-helper 1334 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
1352 * IPI, driven by wakeup. 1335 * IPI, driven by wakeup.
1353 */ 1336 */
1354 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt); 1337 alloc_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
1355 1338
1356 /* IPI for invalidation */ 1339 /* IPI for invalidation */
1357 set_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt); 1340 alloc_intr_gate(INVALIDATE_TLB_VECTOR, invalidate_interrupt);
1358 1341
1359 /* IPI for generic function call */ 1342 /* IPI for generic function call */
1360 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt); 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);
1361} 1348}
1362#endif 1349#endif
1363 1350
@@ -1370,15 +1357,15 @@ void __init apic_intr_init(void)
1370 smp_intr_init(); 1357 smp_intr_init();
1371#endif 1358#endif
1372 /* self generated IPI for local APIC timer */ 1359 /* self generated IPI for local APIC timer */
1373 set_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt); 1360 alloc_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
1374 1361
1375 /* IPI vectors for APIC spurious and error interrupts */ 1362 /* IPI vectors for APIC spurious and error interrupts */
1376 set_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt); 1363 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
1377 set_intr_gate(ERROR_APIC_VECTOR, error_interrupt); 1364 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
1378 1365
1379 /* thermal monitor LVT interrupt */ 1366 /* thermal monitor LVT interrupt */
1380#ifdef CONFIG_X86_MCE_P4THERMAL 1367#ifdef CONFIG_X86_MCE_P4THERMAL
1381 set_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt); 1368 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
1382#endif 1369#endif
1383} 1370}
1384 1371
@@ -1433,7 +1420,7 @@ void disconnect_bsp_APIC(int virt_wire_setup)
1433 value &= ~APIC_VECTOR_MASK; 1420 value &= ~APIC_VECTOR_MASK;
1434 value |= APIC_SPIV_APIC_ENABLED; 1421 value |= APIC_SPIV_APIC_ENABLED;
1435 value |= 0xf; 1422 value |= 0xf;
1436 apic_write_around(APIC_SPIV, value); 1423 apic_write(APIC_SPIV, value);
1437 1424
1438 if (!virt_wire_setup) { 1425 if (!virt_wire_setup) {
1439 /* 1426 /*
@@ -1446,10 +1433,10 @@ void disconnect_bsp_APIC(int virt_wire_setup)
1446 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); 1433 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1447 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; 1434 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1448 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT); 1435 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1449 apic_write_around(APIC_LVT0, value); 1436 apic_write(APIC_LVT0, value);
1450 } else { 1437 } else {
1451 /* Disable LVT0 */ 1438 /* Disable LVT0 */
1452 apic_write_around(APIC_LVT0, APIC_LVT_MASKED); 1439 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1453 } 1440 }
1454 1441
1455 /* 1442 /*
@@ -1463,7 +1450,7 @@ void disconnect_bsp_APIC(int virt_wire_setup)
1463 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); 1450 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1464 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; 1451 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1465 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI); 1452 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1466 apic_write_around(APIC_LVT1, value); 1453 apic_write(APIC_LVT1, value);
1467 } 1454 }
1468} 1455}
1469 1456
@@ -1513,6 +1500,9 @@ void __cpuinit generic_processor_info(int apicid, int version)
1513 */ 1500 */
1514 cpu = 0; 1501 cpu = 0;
1515 1502
1503 if (apicid > max_physical_apicid)
1504 max_physical_apicid = apicid;
1505
1516 /* 1506 /*
1517 * Would be preferable to switch to bigsmp when CONFIG_HOTPLUG_CPU=y 1507 * Would be preferable to switch to bigsmp when CONFIG_HOTPLUG_CPU=y
1518 * but we need to work other dependencies like SMP_SUSPEND etc 1508 * but we need to work other dependencies like SMP_SUSPEND etc
@@ -1520,7 +1510,7 @@ void __cpuinit generic_processor_info(int apicid, int version)
1520 * if (CPU_HOTPLUG_ENABLED || num_processors > 8) 1510 * if (CPU_HOTPLUG_ENABLED || num_processors > 8)
1521 * - Ashok Raj <ashok.raj@intel.com> 1511 * - Ashok Raj <ashok.raj@intel.com>
1522 */ 1512 */
1523 if (num_processors > 8) { 1513 if (max_physical_apicid >= 8) {
1524 switch (boot_cpu_data.x86_vendor) { 1514 switch (boot_cpu_data.x86_vendor) {
1525 case X86_VENDOR_INTEL: 1515 case X86_VENDOR_INTEL:
1526 if (!APIC_XAPIC(version)) { 1516 if (!APIC_XAPIC(version)) {
@@ -1534,9 +1524,9 @@ void __cpuinit generic_processor_info(int apicid, int version)
1534 } 1524 }
1535#ifdef CONFIG_SMP 1525#ifdef CONFIG_SMP
1536 /* are we being called early in kernel startup? */ 1526 /* are we being called early in kernel startup? */
1537 if (x86_cpu_to_apicid_early_ptr) { 1527 if (early_per_cpu_ptr(x86_cpu_to_apicid)) {
1538 u16 *cpu_to_apicid = x86_cpu_to_apicid_early_ptr; 1528 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
1539 u16 *bios_cpu_apicid = x86_bios_cpu_apicid_early_ptr; 1529 u16 *bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
1540 1530
1541 cpu_to_apicid[cpu] = apicid; 1531 cpu_to_apicid[cpu] = apicid;
1542 bios_cpu_apicid[cpu] = apicid; 1532 bios_cpu_apicid[cpu] = apicid;
@@ -1703,15 +1693,15 @@ static void apic_pm_activate(void) { }
1703 */ 1693 */
1704static int __init parse_lapic(char *arg) 1694static int __init parse_lapic(char *arg)
1705{ 1695{
1706 enable_local_apic = 1; 1696 force_enable_local_apic = 1;
1707 return 0; 1697 return 0;
1708} 1698}
1709early_param("lapic", parse_lapic); 1699early_param("lapic", parse_lapic);
1710 1700
1711static int __init parse_nolapic(char *arg) 1701static int __init parse_nolapic(char *arg)
1712{ 1702{
1713 enable_local_apic = -1; 1703 disable_apic = 1;
1714 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); 1704 setup_clear_cpu_cap(X86_FEATURE_APIC);
1715 return 0; 1705 return 0;
1716} 1706}
1717early_param("nolapic", parse_nolapic); 1707early_param("nolapic", parse_nolapic);
@@ -1740,3 +1730,21 @@ static int __init apic_set_verbosity(char *str)
1740} 1730}
1741__setup("apic=", apic_set_verbosity); 1731__setup("apic=", apic_set_verbosity);
1742 1732
1733static int __init lapic_insert_resource(void)
1734{
1735 if (!apic_phys)
1736 return -1;
1737
1738 /* Put local APIC into the resource map. */
1739 lapic_resource.start = apic_phys;
1740 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
1741 insert_resource(&iomem_resource, &lapic_resource);
1742
1743 return 0;
1744}
1745
1746/*
1747 * need call insert after e820_reserve_resources()
1748 * that is using request_resource
1749 */
1750late_initcall(lapic_insert_resource);
diff --git a/arch/x86/kernel/apic_64.c b/arch/x86/kernel/apic_64.c
index 0633cfd0dc29..7f1f030da7ee 100644
--- a/arch/x86/kernel/apic_64.c
+++ b/arch/x86/kernel/apic_64.c
@@ -43,7 +43,7 @@
43#include <mach_ipi.h> 43#include <mach_ipi.h>
44#include <mach_apic.h> 44#include <mach_apic.h>
45 45
46int disable_apic_timer __cpuinitdata; 46static int disable_apic_timer __cpuinitdata;
47static int apic_calibrate_pmtmr __initdata; 47static int apic_calibrate_pmtmr __initdata;
48int disable_apic; 48int disable_apic;
49 49
@@ -54,7 +54,10 @@ EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
54/* 54/*
55 * Debug level, exported for io_apic.c 55 * Debug level, exported for io_apic.c
56 */ 56 */
57int apic_verbosity; 57unsigned int apic_verbosity;
58
59/* Have we found an MP table */
60int smp_found_config;
58 61
59static struct resource lapic_resource = { 62static struct resource lapic_resource = {
60 .name = "Local APIC", 63 .name = "Local APIC",
@@ -87,9 +90,6 @@ static unsigned long apic_phys;
87 90
88unsigned long mp_lapic_addr; 91unsigned long mp_lapic_addr;
89 92
90DEFINE_PER_CPU(u16, x86_bios_cpu_apicid) = BAD_APICID;
91EXPORT_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
92
93unsigned int __cpuinitdata maxcpus = NR_CPUS; 93unsigned int __cpuinitdata maxcpus = NR_CPUS;
94/* 94/*
95 * Get the LAPIC version 95 * Get the LAPIC version
@@ -314,7 +314,7 @@ static void setup_APIC_timer(void)
314 314
315#define TICK_COUNT 100000000 315#define TICK_COUNT 100000000
316 316
317static void __init calibrate_APIC_clock(void) 317static int __init calibrate_APIC_clock(void)
318{ 318{
319 unsigned apic, apic_start; 319 unsigned apic, apic_start;
320 unsigned long tsc, tsc_start; 320 unsigned long tsc, tsc_start;
@@ -368,6 +368,17 @@ static void __init calibrate_APIC_clock(void)
368 clockevent_delta2ns(0xF, &lapic_clockevent); 368 clockevent_delta2ns(0xF, &lapic_clockevent);
369 369
370 calibration_result = result / HZ; 370 calibration_result = result / HZ;
371
372 /*
373 * Do a sanity check on the APIC calibration result
374 */
375 if (calibration_result < (1000000 / HZ)) {
376 printk(KERN_WARNING
377 "APIC frequency too slow, disabling apic timer\n");
378 return -1;
379 }
380
381 return 0;
371} 382}
372 383
373/* 384/*
@@ -394,14 +405,7 @@ void __init setup_boot_APIC_clock(void)
394 } 405 }
395 406
396 printk(KERN_INFO "Using local APIC timer interrupts.\n"); 407 printk(KERN_INFO "Using local APIC timer interrupts.\n");
397 calibrate_APIC_clock(); 408 if (calibrate_APIC_clock()) {
398
399 /*
400 * Do a sanity check on the APIC calibration result
401 */
402 if (calibration_result < (1000000 / HZ)) {
403 printk(KERN_WARNING
404 "APIC frequency too slow, disabling apic timer\n");
405 /* No broadcast on UP ! */ 409 /* No broadcast on UP ! */
406 if (num_possible_cpus() > 1) 410 if (num_possible_cpus() > 1)
407 setup_APIC_timer(); 411 setup_APIC_timer();
@@ -417,37 +421,13 @@ void __init setup_boot_APIC_clock(void)
417 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; 421 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
418 else 422 else
419 printk(KERN_WARNING "APIC timer registered as dummy," 423 printk(KERN_WARNING "APIC timer registered as dummy,"
420 " due to nmi_watchdog=1!\n"); 424 " due to nmi_watchdog=%d!\n", nmi_watchdog);
421 425
422 setup_APIC_timer(); 426 setup_APIC_timer();
423} 427}
424 428
425/*
426 * AMD C1E enabled CPUs have a real nasty problem: Some BIOSes set the
427 * C1E flag only in the secondary CPU, so when we detect the wreckage
428 * we already have enabled the boot CPU local apic timer. Check, if
429 * disable_apic_timer is set and the DUMMY flag is cleared. If yes,
430 * set the DUMMY flag again and force the broadcast mode in the
431 * clockevents layer.
432 */
433static void __cpuinit check_boot_apic_timer_broadcast(void)
434{
435 if (!disable_apic_timer ||
436 (lapic_clockevent.features & CLOCK_EVT_FEAT_DUMMY))
437 return;
438
439 printk(KERN_INFO "AMD C1E detected late. Force timer broadcast.\n");
440 lapic_clockevent.features |= CLOCK_EVT_FEAT_DUMMY;
441
442 local_irq_enable();
443 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
444 &boot_cpu_physical_apicid);
445 local_irq_disable();
446}
447
448void __cpuinit setup_secondary_APIC_clock(void) 429void __cpuinit setup_secondary_APIC_clock(void)
449{ 430{
450 check_boot_apic_timer_broadcast();
451 setup_APIC_timer(); 431 setup_APIC_timer();
452} 432}
453 433
@@ -850,7 +830,6 @@ static void __cpuinit lapic_setup_esr(void)
850void __cpuinit end_local_APIC_setup(void) 830void __cpuinit end_local_APIC_setup(void)
851{ 831{
852 lapic_setup_esr(); 832 lapic_setup_esr();
853 nmi_watchdog_default();
854 setup_apic_nmi_watchdog(NULL); 833 setup_apic_nmi_watchdog(NULL);
855 apic_pm_activate(); 834 apic_pm_activate();
856} 835}
@@ -875,7 +854,7 @@ static int __init detect_init_APIC(void)
875 854
876void __init early_init_lapic_mapping(void) 855void __init early_init_lapic_mapping(void)
877{ 856{
878 unsigned long apic_phys; 857 unsigned long phys_addr;
879 858
880 /* 859 /*
881 * If no local APIC can be found then go out 860 * If no local APIC can be found then go out
@@ -884,11 +863,11 @@ void __init early_init_lapic_mapping(void)
884 if (!smp_found_config) 863 if (!smp_found_config)
885 return; 864 return;
886 865
887 apic_phys = mp_lapic_addr; 866 phys_addr = mp_lapic_addr;
888 867
889 set_fixmap_nocache(FIX_APIC_BASE, apic_phys); 868 set_fixmap_nocache(FIX_APIC_BASE, phys_addr);
890 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n", 869 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
891 APIC_BASE, apic_phys); 870 APIC_BASE, phys_addr);
892 871
893 /* 872 /*
894 * Fetch the APIC ID of the BSP in case we have a 873 * Fetch the APIC ID of the BSP in case we have a
@@ -942,7 +921,9 @@ int __init APIC_init_uniprocessor(void)
942 921
943 verify_local_APIC(); 922 verify_local_APIC();
944 923
945 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid); 924 connect_bsp_APIC();
925
926 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
946 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid)); 927 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
947 928
948 setup_local_APIC(); 929 setup_local_APIC();
@@ -954,6 +935,8 @@ int __init APIC_init_uniprocessor(void)
954 if (!skip_ioapic_setup && nr_ioapics) 935 if (!skip_ioapic_setup && nr_ioapics)
955 enable_IO_APIC(); 936 enable_IO_APIC();
956 937
938 if (!smp_found_config || skip_ioapic_setup || !nr_ioapics)
939 localise_nmi_watchdog();
957 end_local_APIC_setup(); 940 end_local_APIC_setup();
958 941
959 if (smp_found_config && !skip_ioapic_setup && nr_ioapics) 942 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
@@ -1021,6 +1004,14 @@ asmlinkage void smp_error_interrupt(void)
1021 irq_exit(); 1004 irq_exit();
1022} 1005}
1023 1006
1007/**
1008 * * connect_bsp_APIC - attach the APIC to the interrupt system
1009 * */
1010void __init connect_bsp_APIC(void)
1011{
1012 enable_apic_mode();
1013}
1014
1024void disconnect_bsp_APIC(int virt_wire_setup) 1015void disconnect_bsp_APIC(int virt_wire_setup)
1025{ 1016{
1026 /* Go back to Virtual Wire compatibility mode */ 1017 /* Go back to Virtual Wire compatibility mode */
@@ -1090,10 +1081,13 @@ void __cpuinit generic_processor_info(int apicid, int version)
1090 */ 1081 */
1091 cpu = 0; 1082 cpu = 0;
1092 } 1083 }
1084 if (apicid > max_physical_apicid)
1085 max_physical_apicid = apicid;
1086
1093 /* are we being called early in kernel startup? */ 1087 /* are we being called early in kernel startup? */
1094 if (x86_cpu_to_apicid_early_ptr) { 1088 if (early_per_cpu_ptr(x86_cpu_to_apicid)) {
1095 u16 *cpu_to_apicid = x86_cpu_to_apicid_early_ptr; 1089 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
1096 u16 *bios_cpu_apicid = x86_bios_cpu_apicid_early_ptr; 1090 u16 *bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
1097 1091
1098 cpu_to_apicid[cpu] = apicid; 1092 cpu_to_apicid[cpu] = apicid;
1099 bios_cpu_apicid[cpu] = apicid; 1093 bios_cpu_apicid[cpu] = apicid;
@@ -1269,7 +1263,7 @@ __cpuinit int apic_is_clustered_box(void)
1269 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && !is_vsmp_box()) 1263 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && !is_vsmp_box())
1270 return 0; 1264 return 0;
1271 1265
1272 bios_cpu_apicid = x86_bios_cpu_apicid_early_ptr; 1266 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
1273 bitmap_zero(clustermap, NUM_APIC_CLUSTERS); 1267 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
1274 1268
1275 for (i = 0; i < NR_CPUS; i++) { 1269 for (i = 0; i < NR_CPUS; i++) {
@@ -1347,7 +1341,7 @@ early_param("apic", apic_set_verbosity);
1347static __init int setup_disableapic(char *str) 1341static __init int setup_disableapic(char *str)
1348{ 1342{
1349 disable_apic = 1; 1343 disable_apic = 1;
1350 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); 1344 setup_clear_cpu_cap(X86_FEATURE_APIC);
1351 return 0; 1345 return 0;
1352} 1346}
1353early_param("disableapic", setup_disableapic); 1347early_param("disableapic", setup_disableapic);
diff --git a/arch/x86/kernel/apm_32.c b/arch/x86/kernel/apm_32.c
index bf9290e29013..9ee24e6bc4b0 100644
--- a/arch/x86/kernel/apm_32.c
+++ b/arch/x86/kernel/apm_32.c
@@ -204,6 +204,7 @@
204#include <linux/module.h> 204#include <linux/module.h>
205 205
206#include <linux/poll.h> 206#include <linux/poll.h>
207#include <linux/smp_lock.h>
207#include <linux/types.h> 208#include <linux/types.h>
208#include <linux/stddef.h> 209#include <linux/stddef.h>
209#include <linux/timer.h> 210#include <linux/timer.h>
@@ -218,7 +219,6 @@
218#include <linux/time.h> 219#include <linux/time.h>
219#include <linux/sched.h> 220#include <linux/sched.h>
220#include <linux/pm.h> 221#include <linux/pm.h>
221#include <linux/pm_legacy.h>
222#include <linux/capability.h> 222#include <linux/capability.h>
223#include <linux/device.h> 223#include <linux/device.h>
224#include <linux/kernel.h> 224#include <linux/kernel.h>
@@ -228,6 +228,7 @@
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>
231 232
232#include <asm/system.h> 233#include <asm/system.h>
233#include <asm/uaccess.h> 234#include <asm/uaccess.h>
@@ -1149,7 +1150,7 @@ static void queue_event(apm_event_t event, struct apm_user *sender)
1149 as->event_tail = 0; 1150 as->event_tail = 0;
1150 } 1151 }
1151 as->events[as->event_head] = event; 1152 as->events[as->event_head] = event;
1152 if ((!as->suser) || (!as->writer)) 1153 if (!as->suser || !as->writer)
1153 continue; 1154 continue;
1154 switch (event) { 1155 switch (event) {
1155 case APM_SYS_SUSPEND: 1156 case APM_SYS_SUSPEND:
@@ -1211,9 +1212,9 @@ static int suspend(int vetoable)
1211 if (err != APM_SUCCESS) 1212 if (err != APM_SUCCESS)
1212 apm_error("suspend", err); 1213 apm_error("suspend", err);
1213 err = (err == APM_SUCCESS) ? 0 : -EIO; 1214 err = (err == APM_SUCCESS) ? 0 : -EIO;
1214 device_power_up(); 1215 device_power_up(PMSG_RESUME);
1215 local_irq_enable(); 1216 local_irq_enable();
1216 device_resume(); 1217 device_resume(PMSG_RESUME);
1217 queue_event(APM_NORMAL_RESUME, NULL); 1218 queue_event(APM_NORMAL_RESUME, NULL);
1218 spin_lock(&user_list_lock); 1219 spin_lock(&user_list_lock);
1219 for (as = user_list; as != NULL; as = as->next) { 1220 for (as = user_list; as != NULL; as = as->next) {
@@ -1238,7 +1239,7 @@ static void standby(void)
1238 apm_error("standby", err); 1239 apm_error("standby", err);
1239 1240
1240 local_irq_disable(); 1241 local_irq_disable();
1241 device_power_up(); 1242 device_power_up(PMSG_RESUME);
1242 local_irq_enable(); 1243 local_irq_enable();
1243} 1244}
1244 1245
@@ -1324,7 +1325,7 @@ static void check_events(void)
1324 ignore_bounce = 1; 1325 ignore_bounce = 1;
1325 if ((event != APM_NORMAL_RESUME) 1326 if ((event != APM_NORMAL_RESUME)
1326 || (ignore_normal_resume == 0)) { 1327 || (ignore_normal_resume == 0)) {
1327 device_resume(); 1328 device_resume(PMSG_RESUME);
1328 queue_event(event, NULL); 1329 queue_event(event, NULL);
1329 } 1330 }
1330 ignore_normal_resume = 0; 1331 ignore_normal_resume = 0;
@@ -1396,7 +1397,7 @@ static void apm_mainloop(void)
1396 1397
1397static int check_apm_user(struct apm_user *as, const char *func) 1398static int check_apm_user(struct apm_user *as, const char *func)
1398{ 1399{
1399 if ((as == NULL) || (as->magic != APM_BIOS_MAGIC)) { 1400 if (as == NULL || as->magic != APM_BIOS_MAGIC) {
1400 printk(KERN_ERR "apm: %s passed bad filp\n", func); 1401 printk(KERN_ERR "apm: %s passed bad filp\n", func);
1401 return 1; 1402 return 1;
1402 } 1403 }
@@ -1459,18 +1460,19 @@ static unsigned int do_poll(struct file *fp, poll_table *wait)
1459 return 0; 1460 return 0;
1460} 1461}
1461 1462
1462static int do_ioctl(struct inode *inode, struct file *filp, 1463static long do_ioctl(struct file *filp, u_int cmd, u_long arg)
1463 u_int cmd, u_long arg)
1464{ 1464{
1465 struct apm_user *as; 1465 struct apm_user *as;
1466 int ret;
1466 1467
1467 as = filp->private_data; 1468 as = filp->private_data;
1468 if (check_apm_user(as, "ioctl")) 1469 if (check_apm_user(as, "ioctl"))
1469 return -EIO; 1470 return -EIO;
1470 if ((!as->suser) || (!as->writer)) 1471 if (!as->suser || !as->writer)
1471 return -EPERM; 1472 return -EPERM;
1472 switch (cmd) { 1473 switch (cmd) {
1473 case APM_IOC_STANDBY: 1474 case APM_IOC_STANDBY:
1475 lock_kernel();
1474 if (as->standbys_read > 0) { 1476 if (as->standbys_read > 0) {
1475 as->standbys_read--; 1477 as->standbys_read--;
1476 as->standbys_pending--; 1478 as->standbys_pending--;
@@ -1479,8 +1481,10 @@ static int do_ioctl(struct inode *inode, struct file *filp,
1479 queue_event(APM_USER_STANDBY, as); 1481 queue_event(APM_USER_STANDBY, as);
1480 if (standbys_pending <= 0) 1482 if (standbys_pending <= 0)
1481 standby(); 1483 standby();
1484 unlock_kernel();
1482 break; 1485 break;
1483 case APM_IOC_SUSPEND: 1486 case APM_IOC_SUSPEND:
1487 lock_kernel();
1484 if (as->suspends_read > 0) { 1488 if (as->suspends_read > 0) {
1485 as->suspends_read--; 1489 as->suspends_read--;
1486 as->suspends_pending--; 1490 as->suspends_pending--;
@@ -1488,16 +1492,17 @@ static int do_ioctl(struct inode *inode, struct file *filp,
1488 } else 1492 } else
1489 queue_event(APM_USER_SUSPEND, as); 1493 queue_event(APM_USER_SUSPEND, as);
1490 if (suspends_pending <= 0) { 1494 if (suspends_pending <= 0) {
1491 return suspend(1); 1495 ret = suspend(1);
1492 } else { 1496 } else {
1493 as->suspend_wait = 1; 1497 as->suspend_wait = 1;
1494 wait_event_interruptible(apm_suspend_waitqueue, 1498 wait_event_interruptible(apm_suspend_waitqueue,
1495 as->suspend_wait == 0); 1499 as->suspend_wait == 0);
1496 return as->suspend_result; 1500 ret = as->suspend_result;
1497 } 1501 }
1498 break; 1502 unlock_kernel();
1503 return ret;
1499 default: 1504 default:
1500 return -EINVAL; 1505 return -ENOTTY;
1501 } 1506 }
1502 return 0; 1507 return 0;
1503} 1508}
@@ -1544,10 +1549,12 @@ static int do_open(struct inode *inode, struct file *filp)
1544{ 1549{
1545 struct apm_user *as; 1550 struct apm_user *as;
1546 1551
1552 lock_kernel();
1547 as = kmalloc(sizeof(*as), GFP_KERNEL); 1553 as = kmalloc(sizeof(*as), GFP_KERNEL);
1548 if (as == NULL) { 1554 if (as == NULL) {
1549 printk(KERN_ERR "apm: cannot allocate struct of size %d bytes\n", 1555 printk(KERN_ERR "apm: cannot allocate struct of size %d bytes\n",
1550 sizeof(*as)); 1556 sizeof(*as));
1557 unlock_kernel();
1551 return -ENOMEM; 1558 return -ENOMEM;
1552 } 1559 }
1553 as->magic = APM_BIOS_MAGIC; 1560 as->magic = APM_BIOS_MAGIC;
@@ -1569,6 +1576,7 @@ static int do_open(struct inode *inode, struct file *filp)
1569 user_list = as; 1576 user_list = as;
1570 spin_unlock(&user_list_lock); 1577 spin_unlock(&user_list_lock);
1571 filp->private_data = as; 1578 filp->private_data = as;
1579 unlock_kernel();
1572 return 0; 1580 return 0;
1573} 1581}
1574 1582
@@ -1860,7 +1868,7 @@ static const struct file_operations apm_bios_fops = {
1860 .owner = THIS_MODULE, 1868 .owner = THIS_MODULE,
1861 .read = do_read, 1869 .read = do_read,
1862 .poll = do_poll, 1870 .poll = do_poll,
1863 .ioctl = do_ioctl, 1871 .unlocked_ioctl = do_ioctl,
1864 .open = do_open, 1872 .open = do_open,
1865 .release = do_release, 1873 .release = do_release,
1866}; 1874};
diff --git a/arch/x86/kernel/asm-offsets_32.c b/arch/x86/kernel/asm-offsets_32.c
index 92588083950f..6649d09ad88f 100644
--- a/arch/x86/kernel/asm-offsets_32.c
+++ b/arch/x86/kernel/asm-offsets_32.c
@@ -111,7 +111,7 @@ void foo(void)
111 OFFSET(PV_IRQ_irq_disable, pv_irq_ops, irq_disable); 111 OFFSET(PV_IRQ_irq_disable, pv_irq_ops, irq_disable);
112 OFFSET(PV_IRQ_irq_enable, pv_irq_ops, irq_enable); 112 OFFSET(PV_IRQ_irq_enable, pv_irq_ops, irq_enable);
113 OFFSET(PV_CPU_iret, pv_cpu_ops, iret); 113 OFFSET(PV_CPU_iret, pv_cpu_ops, iret);
114 OFFSET(PV_CPU_irq_enable_syscall_ret, pv_cpu_ops, irq_enable_syscall_ret); 114 OFFSET(PV_CPU_irq_enable_sysexit, pv_cpu_ops, irq_enable_sysexit);
115 OFFSET(PV_CPU_read_cr0, pv_cpu_ops, read_cr0); 115 OFFSET(PV_CPU_read_cr0, pv_cpu_ops, read_cr0);
116#endif 116#endif
117 117
diff --git a/arch/x86/kernel/asm-offsets_64.c b/arch/x86/kernel/asm-offsets_64.c
index f126c05d6170..aa89387006fe 100644
--- a/arch/x86/kernel/asm-offsets_64.c
+++ b/arch/x86/kernel/asm-offsets_64.c
@@ -18,6 +18,8 @@
18#include <asm/ia32.h> 18#include <asm/ia32.h>
19#include <asm/bootparam.h> 19#include <asm/bootparam.h>
20 20
21#include <xen/interface/xen.h>
22
21#define __NO_STUBS 1 23#define __NO_STUBS 1
22#undef __SYSCALL 24#undef __SYSCALL
23#undef _ASM_X86_64_UNISTD_H_ 25#undef _ASM_X86_64_UNISTD_H_
@@ -34,7 +36,7 @@ int main(void)
34 ENTRY(pid); 36 ENTRY(pid);
35 BLANK(); 37 BLANK();
36#undef ENTRY 38#undef ENTRY
37#define ENTRY(entry) DEFINE(threadinfo_ ## entry, offsetof(struct thread_info, entry)) 39#define ENTRY(entry) DEFINE(TI_ ## entry, offsetof(struct thread_info, entry))
38 ENTRY(flags); 40 ENTRY(flags);
39 ENTRY(addr_limit); 41 ENTRY(addr_limit);
40 ENTRY(preempt_count); 42 ENTRY(preempt_count);
@@ -61,8 +63,11 @@ int main(void)
61 OFFSET(PARAVIRT_PATCH_pv_irq_ops, paravirt_patch_template, pv_irq_ops); 63 OFFSET(PARAVIRT_PATCH_pv_irq_ops, paravirt_patch_template, pv_irq_ops);
62 OFFSET(PV_IRQ_irq_disable, pv_irq_ops, irq_disable); 64 OFFSET(PV_IRQ_irq_disable, pv_irq_ops, irq_disable);
63 OFFSET(PV_IRQ_irq_enable, pv_irq_ops, irq_enable); 65 OFFSET(PV_IRQ_irq_enable, pv_irq_ops, irq_enable);
66 OFFSET(PV_IRQ_adjust_exception_frame, pv_irq_ops, adjust_exception_frame);
64 OFFSET(PV_CPU_iret, pv_cpu_ops, iret); 67 OFFSET(PV_CPU_iret, pv_cpu_ops, iret);
65 OFFSET(PV_CPU_irq_enable_syscall_ret, pv_cpu_ops, irq_enable_syscall_ret); 68 OFFSET(PV_CPU_usergs_sysret32, pv_cpu_ops, usergs_sysret32);
69 OFFSET(PV_CPU_usergs_sysret64, pv_cpu_ops, usergs_sysret64);
70 OFFSET(PV_CPU_irq_enable_sysexit, pv_cpu_ops, irq_enable_sysexit);
66 OFFSET(PV_CPU_swapgs, pv_cpu_ops, swapgs); 71 OFFSET(PV_CPU_swapgs, pv_cpu_ops, swapgs);
67 OFFSET(PV_MMU_read_cr2, pv_mmu_ops, read_cr2); 72 OFFSET(PV_MMU_read_cr2, pv_mmu_ops, read_cr2);
68#endif 73#endif
@@ -128,5 +133,14 @@ int main(void)
128 OFFSET(BP_loadflags, boot_params, hdr.loadflags); 133 OFFSET(BP_loadflags, boot_params, hdr.loadflags);
129 OFFSET(BP_hardware_subarch, boot_params, hdr.hardware_subarch); 134 OFFSET(BP_hardware_subarch, boot_params, hdr.hardware_subarch);
130 OFFSET(BP_version, boot_params, hdr.version); 135 OFFSET(BP_version, boot_params, hdr.version);
136
137 BLANK();
138 DEFINE(PAGE_SIZE_asm, PAGE_SIZE);
139#ifdef CONFIG_XEN
140 BLANK();
141 OFFSET(XEN_vcpu_info_mask, vcpu_info, evtchn_upcall_mask);
142 OFFSET(XEN_vcpu_info_pending, vcpu_info, evtchn_upcall_pending);
143#undef ENTRY
144#endif
131 return 0; 145 return 0;
132} 146}
diff --git a/arch/x86/kernel/bios_uv.c b/arch/x86/kernel/bios_uv.c
new file mode 100644
index 000000000000..c639bd55391c
--- /dev/null
+++ b/arch/x86/kernel/bios_uv.c
@@ -0,0 +1,48 @@
1/*
2 * BIOS run time interface routines.
3 *
4 * Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
20
21#include <asm/uv/bios.h>
22
23const char *
24x86_bios_strerror(long status)
25{
26 const char *str;
27 switch (status) {
28 case 0: str = "Call completed without error"; break;
29 case -1: str = "Not implemented"; break;
30 case -2: str = "Invalid argument"; break;
31 case -3: str = "Call completed with error"; break;
32 default: str = "Unknown BIOS status code"; break;
33 }
34 return str;
35}
36
37long
38x86_bios_freq_base(unsigned long which, unsigned long *ticks_per_second,
39 unsigned long *drift_info)
40{
41 struct uv_bios_retval isrv;
42
43 BIOS_CALL(isrv, BIOS_FREQ_BASE, which, 0, 0, 0, 0, 0, 0);
44 *ticks_per_second = isrv.v0;
45 *drift_info = isrv.v1;
46 return isrv.status;
47}
48EXPORT_SYMBOL_GPL(x86_bios_freq_base);
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index a0c6f8190887..ee76eaad3001 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -6,11 +6,15 @@ obj-y := intel_cacheinfo.o addon_cpuid_features.o
6obj-y += proc.o feature_names.o 6obj-y += proc.o feature_names.o
7 7
8obj-$(CONFIG_X86_32) += common.o bugs.o 8obj-$(CONFIG_X86_32) += common.o bugs.o
9obj-$(CONFIG_X86_64) += common_64.o bugs_64.o
9obj-$(CONFIG_X86_32) += amd.o 10obj-$(CONFIG_X86_32) += amd.o
11obj-$(CONFIG_X86_64) += amd_64.o
10obj-$(CONFIG_X86_32) += cyrix.o 12obj-$(CONFIG_X86_32) += cyrix.o
11obj-$(CONFIG_X86_32) += centaur.o 13obj-$(CONFIG_X86_32) += centaur.o
14obj-$(CONFIG_X86_64) += centaur_64.o
12obj-$(CONFIG_X86_32) += transmeta.o 15obj-$(CONFIG_X86_32) += transmeta.o
13obj-$(CONFIG_X86_32) += intel.o 16obj-$(CONFIG_X86_32) += intel.o
17obj-$(CONFIG_X86_64) += intel_64.o
14obj-$(CONFIG_X86_32) += umc.o 18obj-$(CONFIG_X86_32) += umc.o
15 19
16obj-$(CONFIG_X86_MCE) += mcheck/ 20obj-$(CONFIG_X86_MCE) += mcheck/
diff --git a/arch/x86/kernel/cpu/addon_cpuid_features.c b/arch/x86/kernel/cpu/addon_cpuid_features.c
index c2e1ce33c7cb..84a8220a6072 100644
--- a/arch/x86/kernel/cpu/addon_cpuid_features.c
+++ b/arch/x86/kernel/cpu/addon_cpuid_features.c
@@ -1,9 +1,7 @@
1
2/* 1/*
3 * Routines to indentify additional cpu features that are scattered in 2 * Routines to indentify additional cpu features that are scattered in
4 * cpuid space. 3 * cpuid space.
5 */ 4 */
6
7#include <linux/cpu.h> 5#include <linux/cpu.h>
8 6
9#include <asm/pat.h> 7#include <asm/pat.h>
@@ -53,19 +51,20 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c)
53#ifdef CONFIG_X86_PAT 51#ifdef CONFIG_X86_PAT
54void __cpuinit validate_pat_support(struct cpuinfo_x86 *c) 52void __cpuinit validate_pat_support(struct cpuinfo_x86 *c)
55{ 53{
54 if (!cpu_has_pat)
55 pat_disable("PAT not supported by CPU.");
56
56 switch (c->x86_vendor) { 57 switch (c->x86_vendor) {
57 case X86_VENDOR_AMD:
58 if (c->x86 >= 0xf && c->x86 <= 0x11)
59 return;
60 break;
61 case X86_VENDOR_INTEL: 58 case X86_VENDOR_INTEL:
62 if (c->x86 == 0xF || (c->x86 == 6 && c->x86_model >= 15)) 59 if (c->x86 == 0xF || (c->x86 == 6 && c->x86_model >= 15))
63 return; 60 return;
64 break; 61 break;
62 case X86_VENDOR_AMD:
63 case X86_VENDOR_CENTAUR:
64 case X86_VENDOR_TRANSMETA:
65 return;
65 } 66 }
66 67
67 pat_disable(cpu_has_pat ? 68 pat_disable("PAT disabled. Not yet verified on this CPU type.");
68 "PAT disabled. Not yet verified on this CPU type." :
69 "PAT not supported by CPU.");
70} 69}
71#endif 70#endif
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 245866828294..cae9cabc3031 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -24,45 +24,6 @@
24extern void vide(void); 24extern void vide(void);
25__asm__(".align 4\nvide: ret"); 25__asm__(".align 4\nvide: ret");
26 26
27#ifdef CONFIG_X86_LOCAL_APIC
28#define ENABLE_C1E_MASK 0x18000000
29#define CPUID_PROCESSOR_SIGNATURE 1
30#define CPUID_XFAM 0x0ff00000
31#define CPUID_XFAM_K8 0x00000000
32#define CPUID_XFAM_10H 0x00100000
33#define CPUID_XFAM_11H 0x00200000
34#define CPUID_XMOD 0x000f0000
35#define CPUID_XMOD_REV_F 0x00040000
36
37/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
38static __cpuinit int amd_apic_timer_broken(void)
39{
40 u32 lo, hi;
41 u32 eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
42 switch (eax & CPUID_XFAM) {
43 case CPUID_XFAM_K8:
44 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
45 break;
46 case CPUID_XFAM_10H:
47 case CPUID_XFAM_11H:
48 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
49 if (lo & ENABLE_C1E_MASK) {
50 if (smp_processor_id() != boot_cpu_physical_apicid)
51 printk(KERN_INFO "AMD C1E detected late. "
52 " Force timer broadcast.\n");
53 return 1;
54 }
55 break;
56 default:
57 /* err on the side of caution */
58 return 1;
59 }
60 return 0;
61}
62#endif
63
64int force_mwait __cpuinitdata;
65
66static void __cpuinit early_init_amd(struct cpuinfo_x86 *c) 27static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
67{ 28{
68 if (cpuid_eax(0x80000000) >= 0x80000007) { 29 if (cpuid_eax(0x80000000) >= 0x80000007) {
@@ -297,11 +258,6 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
297 num_cache_leaves = 3; 258 num_cache_leaves = 3;
298 } 259 }
299 260
300#ifdef CONFIG_X86_LOCAL_APIC
301 if (amd_apic_timer_broken())
302 local_apic_timer_disabled = 1;
303#endif
304
305 /* K6s reports MCEs but don't actually have all the MSRs */ 261 /* K6s reports MCEs but don't actually have all the MSRs */
306 if (c->x86 < 6) 262 if (c->x86 < 6)
307 clear_cpu_cap(c, X86_FEATURE_MCE); 263 clear_cpu_cap(c, X86_FEATURE_MCE);
diff --git a/arch/x86/kernel/cpu/amd_64.c b/arch/x86/kernel/cpu/amd_64.c
new file mode 100644
index 000000000000..d1692b2a41ff
--- /dev/null
+++ b/arch/x86/kernel/cpu/amd_64.c
@@ -0,0 +1,224 @@
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/bugs.c b/arch/x86/kernel/cpu/bugs.c
index 170d2f5523b2..c9b58a806e85 100644
--- a/arch/x86/kernel/cpu/bugs.c
+++ b/arch/x86/kernel/cpu/bugs.c
@@ -59,8 +59,12 @@ static void __init check_fpu(void)
59 return; 59 return;
60 } 60 }
61 61
62/* trap_init() enabled FXSR and company _before_ testing for FP problems here. */ 62 /*
63 /* Test for the divl bug.. */ 63 * trap_init() enabled FXSR and company _before_ testing for FP
64 * problems here.
65 *
66 * Test for the divl bug..
67 */
64 __asm__("fninit\n\t" 68 __asm__("fninit\n\t"
65 "fldl %1\n\t" 69 "fldl %1\n\t"
66 "fdivl %2\n\t" 70 "fdivl %2\n\t"
@@ -108,10 +112,15 @@ static void __init check_popad(void)
108 "movl $12345678,%%eax; movl $0,%%edi; pusha; popa; movl (%%edx,%%edi),%%ecx " 112 "movl $12345678,%%eax; movl $0,%%edi; pusha; popa; movl (%%edx,%%edi),%%ecx "
109 : "=&a" (res) 113 : "=&a" (res)
110 : "d" (inp) 114 : "d" (inp)
111 : "ecx", "edi" ); 115 : "ecx", "edi");
112 /* If this fails, it means that any user program may lock the CPU hard. Too bad. */ 116 /*
113 if (res != 12345678) printk( "Buggy.\n" ); 117 * If this fails, it means that any user program may lock the
114 else printk( "OK.\n" ); 118 * CPU hard. Too bad.
119 */
120 if (res != 12345678)
121 printk("Buggy.\n");
122 else
123 printk("OK.\n");
115#endif 124#endif
116} 125}
117 126
@@ -122,13 +131,7 @@ static void __init check_popad(void)
122 * (for due to lack of "invlpg" and working WP on a i386) 131 * (for due to lack of "invlpg" and working WP on a i386)
123 * - In order to run on anything without a TSC, we need to be 132 * - In order to run on anything without a TSC, we need to be
124 * compiled for a i486. 133 * compiled for a i486.
125 * - In order to support the local APIC on a buggy Pentium machine, 134 */
126 * we need to be compiled with CONFIG_X86_GOOD_APIC disabled,
127 * which happens implicitly if compiled for a Pentium or lower
128 * (unless an advanced selection of CPU features is used) as an
129 * otherwise config implies a properly working local APIC without
130 * the need to do extra reads from the APIC.
131*/
132 135
133static void __init check_config(void) 136static void __init check_config(void)
134{ 137{
@@ -137,25 +140,11 @@ static void __init check_config(void)
137 * i486+ only features! (WP works in supervisor mode and the 140 * i486+ only features! (WP works in supervisor mode and the
138 * new "invlpg" and "bswap" instructions) 141 * new "invlpg" and "bswap" instructions)
139 */ 142 */
140#if defined(CONFIG_X86_WP_WORKS_OK) || defined(CONFIG_X86_INVLPG) || defined(CONFIG_X86_BSWAP) 143#if defined(CONFIG_X86_WP_WORKS_OK) || defined(CONFIG_X86_INVLPG) || \
144 defined(CONFIG_X86_BSWAP)
141 if (boot_cpu_data.x86 == 3) 145 if (boot_cpu_data.x86 == 3)
142 panic("Kernel requires i486+ for 'invlpg' and other features"); 146 panic("Kernel requires i486+ for 'invlpg' and other features");
143#endif 147#endif
144
145/*
146 * If we were told we had a good local APIC, check for buggy Pentia,
147 * i.e. all B steppings and the C2 stepping of P54C when using their
148 * integrated APIC (see 11AP erratum in "Pentium Processor
149 * Specification Update").
150 */
151#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86_GOOD_APIC)
152 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL
153 && cpu_has_apic
154 && boot_cpu_data.x86 == 5
155 && boot_cpu_data.x86_model == 2
156 && (boot_cpu_data.x86_mask < 6 || boot_cpu_data.x86_mask == 11))
157 panic("Kernel compiled for PMMX+, assumes a local APIC without the read-before-write bug!");
158#endif
159} 148}
160 149
161 150
@@ -170,6 +159,7 @@ void __init check_bugs(void)
170 check_fpu(); 159 check_fpu();
171 check_hlt(); 160 check_hlt();
172 check_popad(); 161 check_popad();
173 init_utsname()->machine[1] = '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86); 162 init_utsname()->machine[1] =
163 '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
174 alternative_instructions(); 164 alternative_instructions();
175} 165}
diff --git a/arch/x86/kernel/bugs_64.c b/arch/x86/kernel/cpu/bugs_64.c
index 9a3ed0649d4e..9a3ed0649d4e 100644
--- a/arch/x86/kernel/bugs_64.c
+++ b/arch/x86/kernel/cpu/bugs_64.c
diff --git a/arch/x86/kernel/cpu/centaur_64.c b/arch/x86/kernel/cpu/centaur_64.c
new file mode 100644
index 000000000000..1d181c40e2e1
--- /dev/null
+++ b/arch/x86/kernel/cpu/centaur_64.c
@@ -0,0 +1,35 @@
1#include <linux/init.h>
2#include <linux/smp.h>
3
4#include <asm/cpufeature.h>
5#include <asm/processor.h>
6
7#include "cpu.h"
8
9static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
10{
11 if (c->x86 == 0x6 && c->x86_model >= 0xf)
12 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
13
14 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
15}
16
17static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
18{
19 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
20 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 }
24 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
25}
26
27static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
28 .c_vendor = "Centaur",
29 .c_ident = { "CentaurHauls" },
30 .c_early_init = early_init_centaur,
31 .c_init = init_centaur,
32};
33
34cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev);
35
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index d0463a946247..80ab20d4fa39 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -427,7 +427,7 @@ __setup("serialnumber", x86_serial_nr_setup);
427/* 427/*
428 * This does the hard work of actually picking apart the CPU stuff... 428 * This does the hard work of actually picking apart the CPU stuff...
429 */ 429 */
430void __cpuinit identify_cpu(struct cpuinfo_x86 *c) 430static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
431{ 431{
432 int i; 432 int i;
433 433
diff --git a/arch/x86/kernel/cpu/common_64.c b/arch/x86/kernel/cpu/common_64.c
new file mode 100644
index 000000000000..dd6e3f15017e
--- /dev/null
+++ b/arch/x86/kernel/cpu/common_64.c
@@ -0,0 +1,670 @@
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 783691b2a738..4d894e8565fe 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -1,3 +1,6 @@
1#ifndef ARCH_X86_CPU_H
2
3#define ARCH_X86_CPU_H
1 4
2struct cpu_model_info { 5struct cpu_model_info {
3 int vendor; 6 int vendor;
@@ -36,3 +39,5 @@ extern struct cpu_vendor_dev __x86cpuvendor_start[], __x86cpuvendor_end[];
36 39
37extern int get_model_name(struct cpuinfo_x86 *c); 40extern int get_model_name(struct cpuinfo_x86 *c);
38extern void display_cacheinfo(struct cpuinfo_x86 *c); 41extern void display_cacheinfo(struct cpuinfo_x86 *c);
42
43#endif
diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index b0c8208df9fa..dd097b835839 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -202,7 +202,7 @@ static void drv_write(struct drv_cmd *cmd)
202 cpumask_t saved_mask = current->cpus_allowed; 202 cpumask_t saved_mask = current->cpus_allowed;
203 unsigned int i; 203 unsigned int i;
204 204
205 for_each_cpu_mask(i, cmd->mask) { 205 for_each_cpu_mask_nr(i, cmd->mask) {
206 set_cpus_allowed_ptr(current, &cpumask_of_cpu(i)); 206 set_cpus_allowed_ptr(current, &cpumask_of_cpu(i));
207 do_drv_write(cmd); 207 do_drv_write(cmd);
208 } 208 }
@@ -451,7 +451,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
451 451
452 freqs.old = perf->states[perf->state].core_frequency * 1000; 452 freqs.old = perf->states[perf->state].core_frequency * 1000;
453 freqs.new = data->freq_table[next_state].frequency; 453 freqs.new = data->freq_table[next_state].frequency;
454 for_each_cpu_mask(i, cmd.mask) { 454 for_each_cpu_mask_nr(i, cmd.mask) {
455 freqs.cpu = i; 455 freqs.cpu = i;
456 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 456 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
457 } 457 }
@@ -466,7 +466,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
466 } 466 }
467 } 467 }
468 468
469 for_each_cpu_mask(i, cmd.mask) { 469 for_each_cpu_mask_nr(i, cmd.mask) {
470 freqs.cpu = i; 470 freqs.cpu = i;
471 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 471 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
472 } 472 }
diff --git a/arch/x86/kernel/cpu/cpufreq/cpufreq-nforce2.c b/arch/x86/kernel/cpu/cpufreq/cpufreq-nforce2.c
index f03e9153618e..965ea52767ac 100644
--- a/arch/x86/kernel/cpu/cpufreq/cpufreq-nforce2.c
+++ b/arch/x86/kernel/cpu/cpufreq/cpufreq-nforce2.c
@@ -26,9 +26,10 @@
26#define NFORCE2_SAFE_DISTANCE 50 26#define NFORCE2_SAFE_DISTANCE 50
27 27
28/* Delay in ms between FSB changes */ 28/* Delay in ms between FSB changes */
29//#define NFORCE2_DELAY 10 29/* #define NFORCE2_DELAY 10 */
30 30
31/* nforce2_chipset: 31/*
32 * nforce2_chipset:
32 * FSB is changed using the chipset 33 * FSB is changed using the chipset
33 */ 34 */
34static struct pci_dev *nforce2_chipset_dev; 35static struct pci_dev *nforce2_chipset_dev;
@@ -36,13 +37,13 @@ static struct pci_dev *nforce2_chipset_dev;
36/* fid: 37/* fid:
37 * multiplier * 10 38 * multiplier * 10
38 */ 39 */
39static int fid = 0; 40static int fid;
40 41
41/* min_fsb, max_fsb: 42/* min_fsb, max_fsb:
42 * minimum and maximum FSB (= FSB at boot time) 43 * minimum and maximum FSB (= FSB at boot time)
43 */ 44 */
44static int min_fsb = 0; 45static int min_fsb;
45static int max_fsb = 0; 46static int max_fsb;
46 47
47MODULE_AUTHOR("Sebastian Witt <se.witt@gmx.net>"); 48MODULE_AUTHOR("Sebastian Witt <se.witt@gmx.net>");
48MODULE_DESCRIPTION("nForce2 FSB changing cpufreq driver"); 49MODULE_DESCRIPTION("nForce2 FSB changing cpufreq driver");
@@ -53,7 +54,7 @@ module_param(min_fsb, int, 0444);
53 54
54MODULE_PARM_DESC(fid, "CPU multiplier to use (11.5 = 115)"); 55MODULE_PARM_DESC(fid, "CPU multiplier to use (11.5 = 115)");
55MODULE_PARM_DESC(min_fsb, 56MODULE_PARM_DESC(min_fsb,
56 "Minimum FSB to use, if not defined: current FSB - 50"); 57 "Minimum FSB to use, if not defined: current FSB - 50");
57 58
58#define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "cpufreq-nforce2", msg) 59#define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "cpufreq-nforce2", msg)
59 60
@@ -139,7 +140,7 @@ static unsigned int nforce2_fsb_read(int bootfsb)
139 140
140 /* Get chipset boot FSB from subdevice 5 (FSB at boot-time) */ 141 /* Get chipset boot FSB from subdevice 5 (FSB at boot-time) */
141 nforce2_sub5 = pci_get_subsys(PCI_VENDOR_ID_NVIDIA, 142 nforce2_sub5 = pci_get_subsys(PCI_VENDOR_ID_NVIDIA,
142 0x01EF,PCI_ANY_ID,PCI_ANY_ID,NULL); 143 0x01EF, PCI_ANY_ID, PCI_ANY_ID, NULL);
143 if (!nforce2_sub5) 144 if (!nforce2_sub5)
144 return 0; 145 return 0;
145 146
@@ -147,13 +148,13 @@ static unsigned int nforce2_fsb_read(int bootfsb)
147 fsb /= 1000000; 148 fsb /= 1000000;
148 149
149 /* Check if PLL register is already set */ 150 /* Check if PLL register is already set */
150 pci_read_config_byte(nforce2_chipset_dev,NFORCE2_PLLENABLE, (u8 *)&temp); 151 pci_read_config_byte(nforce2_chipset_dev, NFORCE2_PLLENABLE, (u8 *)&temp);
151 152
152 if(bootfsb || !temp) 153 if (bootfsb || !temp)
153 return fsb; 154 return fsb;
154 155
155 /* Use PLL register FSB value */ 156 /* Use PLL register FSB value */
156 pci_read_config_dword(nforce2_chipset_dev,NFORCE2_PLLREG, &temp); 157 pci_read_config_dword(nforce2_chipset_dev, NFORCE2_PLLREG, &temp);
157 fsb = nforce2_calc_fsb(temp); 158 fsb = nforce2_calc_fsb(temp);
158 159
159 return fsb; 160 return fsb;
@@ -184,7 +185,7 @@ static int nforce2_set_fsb(unsigned int fsb)
184 } 185 }
185 186
186 /* First write? Then set actual value */ 187 /* First write? Then set actual value */
187 pci_read_config_byte(nforce2_chipset_dev,NFORCE2_PLLENABLE, (u8 *)&temp); 188 pci_read_config_byte(nforce2_chipset_dev, NFORCE2_PLLENABLE, (u8 *)&temp);
188 if (!temp) { 189 if (!temp) {
189 pll = nforce2_calc_pll(tfsb); 190 pll = nforce2_calc_pll(tfsb);
190 191
@@ -210,7 +211,8 @@ static int nforce2_set_fsb(unsigned int fsb)
210 tfsb--; 211 tfsb--;
211 212
212 /* Calculate the PLL reg. value */ 213 /* Calculate the PLL reg. value */
213 if ((pll = nforce2_calc_pll(tfsb)) == -1) 214 pll = nforce2_calc_pll(tfsb);
215 if (pll == -1)
214 return -EINVAL; 216 return -EINVAL;
215 217
216 nforce2_write_pll(pll); 218 nforce2_write_pll(pll);
@@ -249,7 +251,7 @@ static unsigned int nforce2_get(unsigned int cpu)
249static int nforce2_target(struct cpufreq_policy *policy, 251static int nforce2_target(struct cpufreq_policy *policy,
250 unsigned int target_freq, unsigned int relation) 252 unsigned int target_freq, unsigned int relation)
251{ 253{
252// unsigned long flags; 254/* unsigned long flags; */
253 struct cpufreq_freqs freqs; 255 struct cpufreq_freqs freqs;
254 unsigned int target_fsb; 256 unsigned int target_fsb;
255 257
@@ -271,17 +273,17 @@ static int nforce2_target(struct cpufreq_policy *policy,
271 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 273 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
272 274
273 /* Disable IRQs */ 275 /* Disable IRQs */
274 //local_irq_save(flags); 276 /* local_irq_save(flags); */
275 277
276 if (nforce2_set_fsb(target_fsb) < 0) 278 if (nforce2_set_fsb(target_fsb) < 0)
277 printk(KERN_ERR "cpufreq: Changing FSB to %d failed\n", 279 printk(KERN_ERR "cpufreq: Changing FSB to %d failed\n",
278 target_fsb); 280 target_fsb);
279 else 281 else
280 dprintk("Changed FSB successfully to %d\n", 282 dprintk("Changed FSB successfully to %d\n",
281 target_fsb); 283 target_fsb);
282 284
283 /* Enable IRQs */ 285 /* Enable IRQs */
284 //local_irq_restore(flags); 286 /* local_irq_restore(flags); */
285 287
286 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 288 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
287 289
@@ -302,8 +304,8 @@ static int nforce2_verify(struct cpufreq_policy *policy)
302 policy->max = (fsb_pol_max + 1) * fid * 100; 304 policy->max = (fsb_pol_max + 1) * fid * 100;
303 305
304 cpufreq_verify_within_limits(policy, 306 cpufreq_verify_within_limits(policy,
305 policy->cpuinfo.min_freq, 307 policy->cpuinfo.min_freq,
306 policy->cpuinfo.max_freq); 308 policy->cpuinfo.max_freq);
307 return 0; 309 return 0;
308} 310}
309 311
@@ -347,7 +349,7 @@ static int nforce2_cpu_init(struct cpufreq_policy *policy)
347 /* Set maximum FSB to FSB at boot time */ 349 /* Set maximum FSB to FSB at boot time */
348 max_fsb = nforce2_fsb_read(1); 350 max_fsb = nforce2_fsb_read(1);
349 351
350 if(!max_fsb) 352 if (!max_fsb)
351 return -EIO; 353 return -EIO;
352 354
353 if (!min_fsb) 355 if (!min_fsb)
diff --git a/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c b/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
index 199e4e05e5dc..f1685fb91fbd 100644
--- a/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
+++ b/arch/x86/kernel/cpu/cpufreq/p4-clockmod.c
@@ -122,7 +122,7 @@ static int cpufreq_p4_target(struct cpufreq_policy *policy,
122 return 0; 122 return 0;
123 123
124 /* notifiers */ 124 /* notifiers */
125 for_each_cpu_mask(i, policy->cpus) { 125 for_each_cpu_mask_nr(i, policy->cpus) {
126 freqs.cpu = i; 126 freqs.cpu = i;
127 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 127 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
128 } 128 }
@@ -130,11 +130,11 @@ static int cpufreq_p4_target(struct cpufreq_policy *policy,
130 /* run on each logical CPU, see section 13.15.3 of IA32 Intel Architecture Software 130 /* run on each logical CPU, see section 13.15.3 of IA32 Intel Architecture Software
131 * Developer's Manual, Volume 3 131 * Developer's Manual, Volume 3
132 */ 132 */
133 for_each_cpu_mask(i, policy->cpus) 133 for_each_cpu_mask_nr(i, policy->cpus)
134 cpufreq_p4_setdc(i, p4clockmod_table[newstate].index); 134 cpufreq_p4_setdc(i, p4clockmod_table[newstate].index);
135 135
136 /* notifiers */ 136 /* notifiers */
137 for_each_cpu_mask(i, policy->cpus) { 137 for_each_cpu_mask_nr(i, policy->cpus) {
138 freqs.cpu = i; 138 freqs.cpu = i;
139 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 139 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
140 } 140 }
diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k7.h b/arch/x86/kernel/cpu/cpufreq/powernow-k7.h
index f8a63b3664e3..35fb4eaf6e1c 100644
--- a/arch/x86/kernel/cpu/cpufreq/powernow-k7.h
+++ b/arch/x86/kernel/cpu/cpufreq/powernow-k7.h
@@ -1,5 +1,4 @@
1/* 1/*
2 * $Id: powernow-k7.h,v 1.2 2003/02/10 18:26:01 davej Exp $
3 * (C) 2003 Dave Jones. 2 * (C) 2003 Dave Jones.
4 * 3 *
5 * Licensed under the terms of the GNU GPL License version 2. 4 * Licensed under the terms of the GNU GPL License version 2.
diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
index 206791eb46e3..c45ca6d4dce1 100644
--- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
+++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c
@@ -966,7 +966,7 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data, unsigned i
966 freqs.old = find_khz_freq_from_fid(data->currfid); 966 freqs.old = find_khz_freq_from_fid(data->currfid);
967 freqs.new = find_khz_freq_from_fid(fid); 967 freqs.new = find_khz_freq_from_fid(fid);
968 968
969 for_each_cpu_mask(i, *(data->available_cores)) { 969 for_each_cpu_mask_nr(i, *(data->available_cores)) {
970 freqs.cpu = i; 970 freqs.cpu = i;
971 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 971 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
972 } 972 }
@@ -974,7 +974,7 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data, unsigned i
974 res = transition_fid_vid(data, fid, vid); 974 res = transition_fid_vid(data, fid, vid);
975 freqs.new = find_khz_freq_from_fid(data->currfid); 975 freqs.new = find_khz_freq_from_fid(data->currfid);
976 976
977 for_each_cpu_mask(i, *(data->available_cores)) { 977 for_each_cpu_mask_nr(i, *(data->available_cores)) {
978 freqs.cpu = i; 978 freqs.cpu = i;
979 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 979 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
980 } 980 }
@@ -997,7 +997,7 @@ static int transition_frequency_pstate(struct powernow_k8_data *data, unsigned i
997 freqs.old = find_khz_freq_from_pstate(data->powernow_table, data->currpstate); 997 freqs.old = find_khz_freq_from_pstate(data->powernow_table, data->currpstate);
998 freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate); 998 freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
999 999
1000 for_each_cpu_mask(i, *(data->available_cores)) { 1000 for_each_cpu_mask_nr(i, *(data->available_cores)) {
1001 freqs.cpu = i; 1001 freqs.cpu = i;
1002 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 1002 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1003 } 1003 }
@@ -1005,7 +1005,7 @@ static int transition_frequency_pstate(struct powernow_k8_data *data, unsigned i
1005 res = transition_pstate(data, pstate); 1005 res = transition_pstate(data, pstate);
1006 freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate); 1006 freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
1007 1007
1008 for_each_cpu_mask(i, *(data->available_cores)) { 1008 for_each_cpu_mask_nr(i, *(data->available_cores)) {
1009 freqs.cpu = i; 1009 freqs.cpu = i;
1010 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 1010 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1011 } 1011 }
diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
index 908dd347c67e..15e13c01cc36 100644
--- a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
+++ b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c
@@ -28,7 +28,8 @@
28#define PFX "speedstep-centrino: " 28#define PFX "speedstep-centrino: "
29#define MAINTAINER "cpufreq@lists.linux.org.uk" 29#define MAINTAINER "cpufreq@lists.linux.org.uk"
30 30
31#define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "speedstep-centrino", msg) 31#define dprintk(msg...) \
32 cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "speedstep-centrino", msg)
32 33
33#define INTEL_MSR_RANGE (0xffff) 34#define INTEL_MSR_RANGE (0xffff)
34 35
@@ -66,11 +67,12 @@ struct cpu_model
66 67
67 struct cpufreq_frequency_table *op_points; /* clock/voltage pairs */ 68 struct cpufreq_frequency_table *op_points; /* clock/voltage pairs */
68}; 69};
69static int centrino_verify_cpu_id(const struct cpuinfo_x86 *c, const struct cpu_id *x); 70static int centrino_verify_cpu_id(const struct cpuinfo_x86 *c,
71 const struct cpu_id *x);
70 72
71/* Operating points for current CPU */ 73/* Operating points for current CPU */
72static struct cpu_model *centrino_model[NR_CPUS]; 74static DEFINE_PER_CPU(struct cpu_model *, centrino_model);
73static const struct cpu_id *centrino_cpu[NR_CPUS]; 75static DEFINE_PER_CPU(const struct cpu_id *, centrino_cpu);
74 76
75static struct cpufreq_driver centrino_driver; 77static struct cpufreq_driver centrino_driver;
76 78
@@ -255,7 +257,7 @@ static int centrino_cpu_init_table(struct cpufreq_policy *policy)
255 return -ENOENT; 257 return -ENOENT;
256 } 258 }
257 259
258 centrino_model[policy->cpu] = model; 260 per_cpu(centrino_model, policy->cpu) = model;
259 261
260 dprintk("found \"%s\": max frequency: %dkHz\n", 262 dprintk("found \"%s\": max frequency: %dkHz\n",
261 model->model_name, model->max_freq); 263 model->model_name, model->max_freq);
@@ -264,10 +266,14 @@ static int centrino_cpu_init_table(struct cpufreq_policy *policy)
264} 266}
265 267
266#else 268#else
267static inline int centrino_cpu_init_table(struct cpufreq_policy *policy) { return -ENODEV; } 269static inline int centrino_cpu_init_table(struct cpufreq_policy *policy)
270{
271 return -ENODEV;
272}
268#endif /* CONFIG_X86_SPEEDSTEP_CENTRINO_TABLE */ 273#endif /* CONFIG_X86_SPEEDSTEP_CENTRINO_TABLE */
269 274
270static int centrino_verify_cpu_id(const struct cpuinfo_x86 *c, const struct cpu_id *x) 275static int centrino_verify_cpu_id(const struct cpuinfo_x86 *c,
276 const struct cpu_id *x)
271{ 277{
272 if ((c->x86 == x->x86) && 278 if ((c->x86 == x->x86) &&
273 (c->x86_model == x->x86_model) && 279 (c->x86_model == x->x86_model) &&
@@ -286,23 +292,28 @@ static unsigned extract_clock(unsigned msr, unsigned int cpu, int failsafe)
286 * for centrino, as some DSDTs are buggy. 292 * for centrino, as some DSDTs are buggy.
287 * Ideally, this can be done using the acpi_data structure. 293 * Ideally, this can be done using the acpi_data structure.
288 */ 294 */
289 if ((centrino_cpu[cpu] == &cpu_ids[CPU_BANIAS]) || 295 if ((per_cpu(centrino_cpu, cpu) == &cpu_ids[CPU_BANIAS]) ||
290 (centrino_cpu[cpu] == &cpu_ids[CPU_DOTHAN_A1]) || 296 (per_cpu(centrino_cpu, cpu) == &cpu_ids[CPU_DOTHAN_A1]) ||
291 (centrino_cpu[cpu] == &cpu_ids[CPU_DOTHAN_B0])) { 297 (per_cpu(centrino_cpu, cpu) == &cpu_ids[CPU_DOTHAN_B0])) {
292 msr = (msr >> 8) & 0xff; 298 msr = (msr >> 8) & 0xff;
293 return msr * 100000; 299 return msr * 100000;
294 } 300 }
295 301
296 if ((!centrino_model[cpu]) || (!centrino_model[cpu]->op_points)) 302 if ((!per_cpu(centrino_model, cpu)) ||
303 (!per_cpu(centrino_model, cpu)->op_points))
297 return 0; 304 return 0;
298 305
299 msr &= 0xffff; 306 msr &= 0xffff;
300 for (i=0;centrino_model[cpu]->op_points[i].frequency != CPUFREQ_TABLE_END; i++) { 307 for (i = 0;
301 if (msr == centrino_model[cpu]->op_points[i].index) 308 per_cpu(centrino_model, cpu)->op_points[i].frequency
302 return centrino_model[cpu]->op_points[i].frequency; 309 != CPUFREQ_TABLE_END;
310 i++) {
311 if (msr == per_cpu(centrino_model, cpu)->op_points[i].index)
312 return per_cpu(centrino_model, cpu)->
313 op_points[i].frequency;
303 } 314 }
304 if (failsafe) 315 if (failsafe)
305 return centrino_model[cpu]->op_points[i-1].frequency; 316 return per_cpu(centrino_model, cpu)->op_points[i-1].frequency;
306 else 317 else
307 return 0; 318 return 0;
308} 319}
@@ -347,7 +358,8 @@ static int centrino_cpu_init(struct cpufreq_policy *policy)
347 int i; 358 int i;
348 359
349 /* Only Intel makes Enhanced Speedstep-capable CPUs */ 360 /* Only Intel makes Enhanced Speedstep-capable CPUs */
350 if (cpu->x86_vendor != X86_VENDOR_INTEL || !cpu_has(cpu, X86_FEATURE_EST)) 361 if (cpu->x86_vendor != X86_VENDOR_INTEL ||
362 !cpu_has(cpu, X86_FEATURE_EST))
351 return -ENODEV; 363 return -ENODEV;
352 364
353 if (cpu_has(cpu, X86_FEATURE_CONSTANT_TSC)) 365 if (cpu_has(cpu, X86_FEATURE_CONSTANT_TSC))
@@ -361,9 +373,9 @@ static int centrino_cpu_init(struct cpufreq_policy *policy)
361 break; 373 break;
362 374
363 if (i != N_IDS) 375 if (i != N_IDS)
364 centrino_cpu[policy->cpu] = &cpu_ids[i]; 376 per_cpu(centrino_cpu, policy->cpu) = &cpu_ids[i];
365 377
366 if (!centrino_cpu[policy->cpu]) { 378 if (!per_cpu(centrino_cpu, policy->cpu)) {
367 dprintk("found unsupported CPU with " 379 dprintk("found unsupported CPU with "
368 "Enhanced SpeedStep: send /proc/cpuinfo to " 380 "Enhanced SpeedStep: send /proc/cpuinfo to "
369 MAINTAINER "\n"); 381 MAINTAINER "\n");
@@ -386,23 +398,26 @@ static int centrino_cpu_init(struct cpufreq_policy *policy)
386 /* check to see if it stuck */ 398 /* check to see if it stuck */
387 rdmsr(MSR_IA32_MISC_ENABLE, l, h); 399 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
388 if (!(l & (1<<16))) { 400 if (!(l & (1<<16))) {
389 printk(KERN_INFO PFX "couldn't enable Enhanced SpeedStep\n"); 401 printk(KERN_INFO PFX
402 "couldn't enable Enhanced SpeedStep\n");
390 return -ENODEV; 403 return -ENODEV;
391 } 404 }
392 } 405 }
393 406
394 freq = get_cur_freq(policy->cpu); 407 freq = get_cur_freq(policy->cpu);
395 408 policy->cpuinfo.transition_latency = 10000;
396 policy->cpuinfo.transition_latency = 10000; /* 10uS transition latency */ 409 /* 10uS transition latency */
397 policy->cur = freq; 410 policy->cur = freq;
398 411
399 dprintk("centrino_cpu_init: cur=%dkHz\n", policy->cur); 412 dprintk("centrino_cpu_init: cur=%dkHz\n", policy->cur);
400 413
401 ret = cpufreq_frequency_table_cpuinfo(policy, centrino_model[policy->cpu]->op_points); 414 ret = cpufreq_frequency_table_cpuinfo(policy,
415 per_cpu(centrino_model, policy->cpu)->op_points);
402 if (ret) 416 if (ret)
403 return (ret); 417 return (ret);
404 418
405 cpufreq_frequency_table_get_attr(centrino_model[policy->cpu]->op_points, policy->cpu); 419 cpufreq_frequency_table_get_attr(
420 per_cpu(centrino_model, policy->cpu)->op_points, policy->cpu);
406 421
407 return 0; 422 return 0;
408} 423}
@@ -411,12 +426,12 @@ static int centrino_cpu_exit(struct cpufreq_policy *policy)
411{ 426{
412 unsigned int cpu = policy->cpu; 427 unsigned int cpu = policy->cpu;
413 428
414 if (!centrino_model[cpu]) 429 if (!per_cpu(centrino_model, cpu))
415 return -ENODEV; 430 return -ENODEV;
416 431
417 cpufreq_frequency_table_put_attr(cpu); 432 cpufreq_frequency_table_put_attr(cpu);
418 433
419 centrino_model[cpu] = NULL; 434 per_cpu(centrino_model, cpu) = NULL;
420 435
421 return 0; 436 return 0;
422} 437}
@@ -430,17 +445,26 @@ static int centrino_cpu_exit(struct cpufreq_policy *policy)
430 */ 445 */
431static int centrino_verify (struct cpufreq_policy *policy) 446static int centrino_verify (struct cpufreq_policy *policy)
432{ 447{
433 return cpufreq_frequency_table_verify(policy, centrino_model[policy->cpu]->op_points); 448 return cpufreq_frequency_table_verify(policy,
449 per_cpu(centrino_model, policy->cpu)->op_points);
434} 450}
435 451
436/** 452/**
437 * centrino_setpolicy - set a new CPUFreq policy 453 * centrino_setpolicy - set a new CPUFreq policy
438 * @policy: new policy 454 * @policy: new policy
439 * @target_freq: the target frequency 455 * @target_freq: the target frequency
440 * @relation: how that frequency relates to achieved frequency (CPUFREQ_RELATION_L or CPUFREQ_RELATION_H) 456 * @relation: how that frequency relates to achieved frequency
457 * (CPUFREQ_RELATION_L or CPUFREQ_RELATION_H)
441 * 458 *
442 * Sets a new CPUFreq policy. 459 * Sets a new CPUFreq policy.
443 */ 460 */
461struct allmasks {
462 cpumask_t online_policy_cpus;
463 cpumask_t saved_mask;
464 cpumask_t set_mask;
465 cpumask_t covered_cpus;
466};
467
444static int centrino_target (struct cpufreq_policy *policy, 468static int centrino_target (struct cpufreq_policy *policy,
445 unsigned int target_freq, 469 unsigned int target_freq,
446 unsigned int relation) 470 unsigned int relation)
@@ -448,48 +472,55 @@ static int centrino_target (struct cpufreq_policy *policy,
448 unsigned int newstate = 0; 472 unsigned int newstate = 0;
449 unsigned int msr, oldmsr = 0, h = 0, cpu = policy->cpu; 473 unsigned int msr, oldmsr = 0, h = 0, cpu = policy->cpu;
450 struct cpufreq_freqs freqs; 474 struct cpufreq_freqs freqs;
451 cpumask_t online_policy_cpus;
452 cpumask_t saved_mask;
453 cpumask_t set_mask;
454 cpumask_t covered_cpus;
455 int retval = 0; 475 int retval = 0;
456 unsigned int j, k, first_cpu, tmp; 476 unsigned int j, k, first_cpu, tmp;
457 477 CPUMASK_ALLOC(allmasks);
458 if (unlikely(centrino_model[cpu] == NULL)) 478 CPUMASK_PTR(online_policy_cpus, allmasks);
459 return -ENODEV; 479 CPUMASK_PTR(saved_mask, allmasks);
480 CPUMASK_PTR(set_mask, allmasks);
481 CPUMASK_PTR(covered_cpus, allmasks);
482
483 if (unlikely(allmasks == NULL))
484 return -ENOMEM;
485
486 if (unlikely(per_cpu(centrino_model, cpu) == NULL)) {
487 retval = -ENODEV;
488 goto out;
489 }
460 490
461 if (unlikely(cpufreq_frequency_table_target(policy, 491 if (unlikely(cpufreq_frequency_table_target(policy,
462 centrino_model[cpu]->op_points, 492 per_cpu(centrino_model, cpu)->op_points,
463 target_freq, 493 target_freq,
464 relation, 494 relation,
465 &newstate))) { 495 &newstate))) {
466 return -EINVAL; 496 retval = -EINVAL;
497 goto out;
467 } 498 }
468 499
469#ifdef CONFIG_HOTPLUG_CPU 500#ifdef CONFIG_HOTPLUG_CPU
470 /* cpufreq holds the hotplug lock, so we are safe from here on */ 501 /* cpufreq holds the hotplug lock, so we are safe from here on */
471 cpus_and(online_policy_cpus, cpu_online_map, policy->cpus); 502 cpus_and(*online_policy_cpus, cpu_online_map, policy->cpus);
472#else 503#else
473 online_policy_cpus = policy->cpus; 504 *online_policy_cpus = policy->cpus;
474#endif 505#endif
475 506
476 saved_mask = current->cpus_allowed; 507 *saved_mask = current->cpus_allowed;
477 first_cpu = 1; 508 first_cpu = 1;
478 cpus_clear(covered_cpus); 509 cpus_clear(*covered_cpus);
479 for_each_cpu_mask(j, online_policy_cpus) { 510 for_each_cpu_mask_nr(j, *online_policy_cpus) {
480 /* 511 /*
481 * Support for SMP systems. 512 * Support for SMP systems.
482 * Make sure we are running on CPU that wants to change freq 513 * Make sure we are running on CPU that wants to change freq
483 */ 514 */
484 cpus_clear(set_mask); 515 cpus_clear(*set_mask);
485 if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) 516 if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
486 cpus_or(set_mask, set_mask, online_policy_cpus); 517 cpus_or(*set_mask, *set_mask, *online_policy_cpus);
487 else 518 else
488 cpu_set(j, set_mask); 519 cpu_set(j, *set_mask);
489 520
490 set_cpus_allowed_ptr(current, &set_mask); 521 set_cpus_allowed_ptr(current, set_mask);
491 preempt_disable(); 522 preempt_disable();
492 if (unlikely(!cpu_isset(smp_processor_id(), set_mask))) { 523 if (unlikely(!cpu_isset(smp_processor_id(), *set_mask))) {
493 dprintk("couldn't limit to CPUs in this domain\n"); 524 dprintk("couldn't limit to CPUs in this domain\n");
494 retval = -EAGAIN; 525 retval = -EAGAIN;
495 if (first_cpu) { 526 if (first_cpu) {
@@ -500,7 +531,7 @@ static int centrino_target (struct cpufreq_policy *policy,
500 break; 531 break;
501 } 532 }
502 533
503 msr = centrino_model[cpu]->op_points[newstate].index; 534 msr = per_cpu(centrino_model, cpu)->op_points[newstate].index;
504 535
505 if (first_cpu) { 536 if (first_cpu) {
506 rdmsr(MSR_IA32_PERF_CTL, oldmsr, h); 537 rdmsr(MSR_IA32_PERF_CTL, oldmsr, h);
@@ -517,7 +548,7 @@ static int centrino_target (struct cpufreq_policy *policy,
517 dprintk("target=%dkHz old=%d new=%d msr=%04x\n", 548 dprintk("target=%dkHz old=%d new=%d msr=%04x\n",
518 target_freq, freqs.old, freqs.new, msr); 549 target_freq, freqs.old, freqs.new, msr);
519 550
520 for_each_cpu_mask(k, online_policy_cpus) { 551 for_each_cpu_mask_nr(k, *online_policy_cpus) {
521 freqs.cpu = k; 552 freqs.cpu = k;
522 cpufreq_notify_transition(&freqs, 553 cpufreq_notify_transition(&freqs,
523 CPUFREQ_PRECHANGE); 554 CPUFREQ_PRECHANGE);
@@ -536,11 +567,11 @@ static int centrino_target (struct cpufreq_policy *policy,
536 break; 567 break;
537 } 568 }
538 569
539 cpu_set(j, covered_cpus); 570 cpu_set(j, *covered_cpus);
540 preempt_enable(); 571 preempt_enable();
541 } 572 }
542 573
543 for_each_cpu_mask(k, online_policy_cpus) { 574 for_each_cpu_mask_nr(k, *online_policy_cpus) {
544 freqs.cpu = k; 575 freqs.cpu = k;
545 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 576 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
546 } 577 }
@@ -553,30 +584,32 @@ static int centrino_target (struct cpufreq_policy *policy,
553 * Best effort undo.. 584 * Best effort undo..
554 */ 585 */
555 586
556 if (!cpus_empty(covered_cpus)) { 587 if (!cpus_empty(*covered_cpus))
557 for_each_cpu_mask(j, covered_cpus) { 588 for_each_cpu_mask_nr(j, *covered_cpus) {
558 set_cpus_allowed_ptr(current, 589 set_cpus_allowed_ptr(current,
559 &cpumask_of_cpu(j)); 590 &cpumask_of_cpu(j));
560 wrmsr(MSR_IA32_PERF_CTL, oldmsr, h); 591 wrmsr(MSR_IA32_PERF_CTL, oldmsr, h);
561 } 592 }
562 }
563 593
564 tmp = freqs.new; 594 tmp = freqs.new;
565 freqs.new = freqs.old; 595 freqs.new = freqs.old;
566 freqs.old = tmp; 596 freqs.old = tmp;
567 for_each_cpu_mask(j, online_policy_cpus) { 597 for_each_cpu_mask_nr(j, *online_policy_cpus) {
568 freqs.cpu = j; 598 freqs.cpu = j;
569 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 599 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
570 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 600 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
571 } 601 }
572 } 602 }
573 set_cpus_allowed_ptr(current, &saved_mask); 603 set_cpus_allowed_ptr(current, saved_mask);
574 return 0; 604 retval = 0;
605 goto out;
575 606
576migrate_end: 607migrate_end:
577 preempt_enable(); 608 preempt_enable();
578 set_cpus_allowed_ptr(current, &saved_mask); 609 set_cpus_allowed_ptr(current, saved_mask);
579 return 0; 610out:
611 CPUMASK_FREE(allmasks);
612 return retval;
580} 613}
581 614
582static struct freq_attr* centrino_attr[] = { 615static struct freq_attr* centrino_attr[] = {
diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c b/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c
index 1b50244b1fdf..191f7263c61d 100644
--- a/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c
+++ b/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c
@@ -279,7 +279,7 @@ static int speedstep_target (struct cpufreq_policy *policy,
279 279
280 cpus_allowed = current->cpus_allowed; 280 cpus_allowed = current->cpus_allowed;
281 281
282 for_each_cpu_mask(i, policy->cpus) { 282 for_each_cpu_mask_nr(i, policy->cpus) {
283 freqs.cpu = i; 283 freqs.cpu = i;
284 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); 284 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
285 } 285 }
@@ -292,7 +292,7 @@ static int speedstep_target (struct cpufreq_policy *policy,
292 /* allow to be run on all CPUs */ 292 /* allow to be run on all CPUs */
293 set_cpus_allowed_ptr(current, &cpus_allowed); 293 set_cpus_allowed_ptr(current, &cpus_allowed);
294 294
295 for_each_cpu_mask(i, policy->cpus) { 295 for_each_cpu_mask_nr(i, policy->cpus) {
296 freqs.cpu = i; 296 freqs.cpu = i;
297 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); 297 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
298 } 298 }
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index fe9224c51d37..b75f2569b8f8 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -226,6 +226,20 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
226 226
227 if (cpu_has_bts) 227 if (cpu_has_bts)
228 ds_init_intel(c); 228 ds_init_intel(c);
229
230 /*
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
240#ifdef CONFIG_X86_NUMAQ
241 numaq_tsc_disable();
242#endif
229} 243}
230 244
231static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size) 245static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
diff --git a/arch/x86/kernel/cpu/intel_64.c b/arch/x86/kernel/cpu/intel_64.c
new file mode 100644
index 000000000000..1019c58d39f0
--- /dev/null
+++ b/arch/x86/kernel/cpu/intel_64.c
@@ -0,0 +1,95 @@
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 26d615dcb149..6b0a10b002f1 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -62,6 +62,7 @@ static struct _cache_table cache_table[] __cpuinitdata =
62 { 0x4b, LVL_3, 8192 }, /* 16-way set assoc, 64 byte line size */ 62 { 0x4b, LVL_3, 8192 }, /* 16-way set assoc, 64 byte line size */
63 { 0x4c, LVL_3, 12288 }, /* 12-way set assoc, 64 byte line size */ 63 { 0x4c, LVL_3, 12288 }, /* 12-way set assoc, 64 byte line size */
64 { 0x4d, LVL_3, 16384 }, /* 16-way set assoc, 64 byte line size */ 64 { 0x4d, LVL_3, 16384 }, /* 16-way set assoc, 64 byte line size */
65 { 0x4e, LVL_2, 6144 }, /* 24-way set assoc, 64 byte line size */
65 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */ 66 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
66 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */ 67 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
67 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */ 68 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
@@ -488,7 +489,7 @@ static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index)
488 int sibling; 489 int sibling;
489 490
490 this_leaf = CPUID4_INFO_IDX(cpu, index); 491 this_leaf = CPUID4_INFO_IDX(cpu, index);
491 for_each_cpu_mask(sibling, this_leaf->shared_cpu_map) { 492 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) {
492 sibling_leaf = CPUID4_INFO_IDX(sibling, index); 493 sibling_leaf = CPUID4_INFO_IDX(sibling, index);
493 cpu_clear(cpu, sibling_leaf->shared_cpu_map); 494 cpu_clear(cpu, sibling_leaf->shared_cpu_map);
494 } 495 }
@@ -779,15 +780,14 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev)
779 } 780 }
780 kobject_put(per_cpu(cache_kobject, cpu)); 781 kobject_put(per_cpu(cache_kobject, cpu));
781 cpuid4_cache_sysfs_exit(cpu); 782 cpuid4_cache_sysfs_exit(cpu);
782 break; 783 return retval;
783 } 784 }
784 kobject_uevent(&(this_object->kobj), KOBJ_ADD); 785 kobject_uevent(&(this_object->kobj), KOBJ_ADD);
785 } 786 }
786 if (!retval) 787 cpu_set(cpu, cache_dev_map);
787 cpu_set(cpu, cache_dev_map);
788 788
789 kobject_uevent(per_cpu(cache_kobject, cpu), KOBJ_ADD); 789 kobject_uevent(per_cpu(cache_kobject, cpu), KOBJ_ADD);
790 return retval; 790 return 0;
791} 791}
792 792
793static void __cpuinit cache_remove_dev(struct sys_device * sys_dev) 793static void __cpuinit cache_remove_dev(struct sys_device * sys_dev)
diff --git a/arch/x86/kernel/cpu/mcheck/k7.c b/arch/x86/kernel/cpu/mcheck/k7.c
index e633c9c2b764..f390c9f66351 100644
--- a/arch/x86/kernel/cpu/mcheck/k7.c
+++ b/arch/x86/kernel/cpu/mcheck/k7.c
@@ -9,23 +9,23 @@
9#include <linux/interrupt.h> 9#include <linux/interrupt.h>
10#include <linux/smp.h> 10#include <linux/smp.h>
11 11
12#include <asm/processor.h> 12#include <asm/processor.h>
13#include <asm/system.h> 13#include <asm/system.h>
14#include <asm/msr.h> 14#include <asm/msr.h>
15 15
16#include "mce.h" 16#include "mce.h"
17 17
18/* Machine Check Handler For AMD Athlon/Duron */ 18/* Machine Check Handler For AMD Athlon/Duron */
19static void k7_machine_check(struct pt_regs * regs, long error_code) 19static void k7_machine_check(struct pt_regs *regs, long error_code)
20{ 20{
21 int recover=1; 21 int recover = 1;
22 u32 alow, ahigh, high, low; 22 u32 alow, ahigh, high, low;
23 u32 mcgstl, mcgsth; 23 u32 mcgstl, mcgsth;
24 int i; 24 int i;
25 25
26 rdmsr (MSR_IA32_MCG_STATUS, mcgstl, mcgsth); 26 rdmsr(MSR_IA32_MCG_STATUS, mcgstl, mcgsth);
27 if (mcgstl & (1<<0)) /* Recoverable ? */ 27 if (mcgstl & (1<<0)) /* Recoverable ? */
28 recover=0; 28 recover = 0;
29 29
30 printk(KERN_EMERG "CPU %d: Machine Check Exception: %08x%08x\n", 30 printk(KERN_EMERG "CPU %d: Machine Check Exception: %08x%08x\n",
31 smp_processor_id(), mcgsth, mcgstl); 31 smp_processor_id(), mcgsth, mcgstl);
@@ -60,12 +60,12 @@ static void k7_machine_check(struct pt_regs * regs, long error_code)
60 } 60 }
61 61
62 if (recover&2) 62 if (recover&2)
63 panic ("CPU context corrupt"); 63 panic("CPU context corrupt");
64 if (recover&1) 64 if (recover&1)
65 panic ("Unable to continue"); 65 panic("Unable to continue");
66 printk (KERN_EMERG "Attempting to continue.\n"); 66 printk(KERN_EMERG "Attempting to continue.\n");
67 mcgstl &= ~(1<<2); 67 mcgstl &= ~(1<<2);
68 wrmsr (MSR_IA32_MCG_STATUS,mcgstl, mcgsth); 68 wrmsr(MSR_IA32_MCG_STATUS, mcgstl, mcgsth);
69} 69}
70 70
71 71
@@ -81,25 +81,25 @@ void amd_mcheck_init(struct cpuinfo_x86 *c)
81 machine_check_vector = k7_machine_check; 81 machine_check_vector = k7_machine_check;
82 wmb(); 82 wmb();
83 83
84 printk (KERN_INFO "Intel machine check architecture supported.\n"); 84 printk(KERN_INFO "Intel machine check architecture supported.\n");
85 rdmsr (MSR_IA32_MCG_CAP, l, h); 85 rdmsr(MSR_IA32_MCG_CAP, l, h);
86 if (l & (1<<8)) /* Control register present ? */ 86 if (l & (1<<8)) /* Control register present ? */
87 wrmsr (MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); 87 wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
88 nr_mce_banks = l & 0xff; 88 nr_mce_banks = l & 0xff;
89 89
90 /* Clear status for MC index 0 separately, we don't touch CTL, 90 /* Clear status for MC index 0 separately, we don't touch CTL,
91 * as some K7 Athlons cause spurious MCEs when its enabled. */ 91 * as some K7 Athlons cause spurious MCEs when its enabled. */
92 if (boot_cpu_data.x86 == 6) { 92 if (boot_cpu_data.x86 == 6) {
93 wrmsr (MSR_IA32_MC0_STATUS, 0x0, 0x0); 93 wrmsr(MSR_IA32_MC0_STATUS, 0x0, 0x0);
94 i = 1; 94 i = 1;
95 } else 95 } else
96 i = 0; 96 i = 0;
97 for (; i<nr_mce_banks; i++) { 97 for (; i < nr_mce_banks; i++) {
98 wrmsr (MSR_IA32_MC0_CTL+4*i, 0xffffffff, 0xffffffff); 98 wrmsr(MSR_IA32_MC0_CTL+4*i, 0xffffffff, 0xffffffff);
99 wrmsr (MSR_IA32_MC0_STATUS+4*i, 0x0, 0x0); 99 wrmsr(MSR_IA32_MC0_STATUS+4*i, 0x0, 0x0);
100 } 100 }
101 101
102 set_in_cr4 (X86_CR4_MCE); 102 set_in_cr4(X86_CR4_MCE);
103 printk (KERN_INFO "Intel machine check reporting enabled on CPU#%d.\n", 103 printk(KERN_INFO "Intel machine check reporting enabled on CPU#%d.\n",
104 smp_processor_id()); 104 smp_processor_id());
105} 105}
diff --git a/arch/x86/kernel/cpu/mcheck/mce_64.c b/arch/x86/kernel/cpu/mcheck/mce_64.c
index e07e8c068ae0..65a339678ece 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_64.c
@@ -9,6 +9,7 @@
9#include <linux/types.h> 9#include <linux/types.h>
10#include <linux/kernel.h> 10#include <linux/kernel.h>
11#include <linux/sched.h> 11#include <linux/sched.h>
12#include <linux/smp_lock.h>
12#include <linux/string.h> 13#include <linux/string.h>
13#include <linux/rcupdate.h> 14#include <linux/rcupdate.h>
14#include <linux/kallsyms.h> 15#include <linux/kallsyms.h>
@@ -31,7 +32,7 @@
31#include <asm/idle.h> 32#include <asm/idle.h>
32 33
33#define MISC_MCELOG_MINOR 227 34#define MISC_MCELOG_MINOR 227
34#define NR_BANKS 6 35#define NR_SYSFS_BANKS 6
35 36
36atomic_t mce_entry; 37atomic_t mce_entry;
37 38
@@ -46,7 +47,7 @@ static int mce_dont_init;
46 */ 47 */
47static int tolerant = 1; 48static int tolerant = 1;
48static int banks; 49static int banks;
49static unsigned long bank[NR_BANKS] = { [0 ... NR_BANKS-1] = ~0UL }; 50static unsigned long bank[NR_SYSFS_BANKS] = { [0 ... NR_SYSFS_BANKS-1] = ~0UL };
50static unsigned long notify_user; 51static unsigned long notify_user;
51static int rip_msr; 52static int rip_msr;
52static int mce_bootlog = -1; 53static int mce_bootlog = -1;
@@ -209,7 +210,7 @@ void do_machine_check(struct pt_regs * regs, long error_code)
209 barrier(); 210 barrier();
210 211
211 for (i = 0; i < banks; i++) { 212 for (i = 0; i < banks; i++) {
212 if (!bank[i]) 213 if (i < NR_SYSFS_BANKS && !bank[i])
213 continue; 214 continue;
214 215
215 m.misc = 0; 216 m.misc = 0;
@@ -363,7 +364,7 @@ static void mcheck_check_cpu(void *info)
363 364
364static void mcheck_timer(struct work_struct *work) 365static void mcheck_timer(struct work_struct *work)
365{ 366{
366 on_each_cpu(mcheck_check_cpu, NULL, 1, 1); 367 on_each_cpu(mcheck_check_cpu, NULL, 1);
367 368
368 /* 369 /*
369 * Alert userspace if needed. If we logged an MCE, reduce the 370 * Alert userspace if needed. If we logged an MCE, reduce the
@@ -444,9 +445,10 @@ static void mce_init(void *dummy)
444 445
445 rdmsrl(MSR_IA32_MCG_CAP, cap); 446 rdmsrl(MSR_IA32_MCG_CAP, cap);
446 banks = cap & 0xff; 447 banks = cap & 0xff;
447 if (banks > NR_BANKS) { 448 if (banks > MCE_EXTENDED_BANK) {
448 printk(KERN_INFO "MCE: warning: using only %d banks\n", banks); 449 banks = MCE_EXTENDED_BANK;
449 banks = NR_BANKS; 450 printk(KERN_INFO "MCE: warning: using only %d banks\n",
451 MCE_EXTENDED_BANK);
450 } 452 }
451 /* Use accurate RIP reporting if available. */ 453 /* Use accurate RIP reporting if available. */
452 if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9) 454 if ((cap & (1<<9)) && ((cap >> 16) & 0xff) >= 9)
@@ -462,7 +464,11 @@ static void mce_init(void *dummy)
462 wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); 464 wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
463 465
464 for (i = 0; i < banks; i++) { 466 for (i = 0; i < banks; i++) {
465 wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]); 467 if (i < NR_SYSFS_BANKS)
468 wrmsrl(MSR_IA32_MC0_CTL+4*i, bank[i]);
469 else
470 wrmsrl(MSR_IA32_MC0_CTL+4*i, ~0UL);
471
466 wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0); 472 wrmsrl(MSR_IA32_MC0_STATUS+4*i, 0);
467 } 473 }
468} 474}
@@ -527,10 +533,12 @@ static int open_exclu; /* already open exclusive? */
527 533
528static int mce_open(struct inode *inode, struct file *file) 534static int mce_open(struct inode *inode, struct file *file)
529{ 535{
536 lock_kernel();
530 spin_lock(&mce_state_lock); 537 spin_lock(&mce_state_lock);
531 538
532 if (open_exclu || (open_count && (file->f_flags & O_EXCL))) { 539 if (open_exclu || (open_count && (file->f_flags & O_EXCL))) {
533 spin_unlock(&mce_state_lock); 540 spin_unlock(&mce_state_lock);
541 unlock_kernel();
534 return -EBUSY; 542 return -EBUSY;
535 } 543 }
536 544
@@ -539,6 +547,7 @@ static int mce_open(struct inode *inode, struct file *file)
539 open_count++; 547 open_count++;
540 548
541 spin_unlock(&mce_state_lock); 549 spin_unlock(&mce_state_lock);
550 unlock_kernel();
542 551
543 return nonseekable_open(inode, file); 552 return nonseekable_open(inode, file);
544} 553}
@@ -571,7 +580,7 @@ static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize,
571 char __user *buf = ubuf; 580 char __user *buf = ubuf;
572 int i, err; 581 int i, err;
573 582
574 cpu_tsc = kmalloc(NR_CPUS * sizeof(long), GFP_KERNEL); 583 cpu_tsc = kmalloc(nr_cpu_ids * sizeof(long), GFP_KERNEL);
575 if (!cpu_tsc) 584 if (!cpu_tsc)
576 return -ENOMEM; 585 return -ENOMEM;
577 586
@@ -612,7 +621,7 @@ static ssize_t mce_read(struct file *filp, char __user *ubuf, size_t usize,
612 * Collect entries that were still getting written before the 621 * Collect entries that were still getting written before the
613 * synchronize. 622 * synchronize.
614 */ 623 */
615 on_each_cpu(collect_tscs, cpu_tsc, 1, 1); 624 on_each_cpu(collect_tscs, cpu_tsc, 1);
616 for (i = next; i < MCE_LOG_LEN; i++) { 625 for (i = next; i < MCE_LOG_LEN; i++) {
617 if (mcelog.entry[i].finished && 626 if (mcelog.entry[i].finished &&
618 mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) { 627 mcelog.entry[i].tsc < cpu_tsc[mcelog.entry[i].cpu]) {
@@ -737,7 +746,7 @@ static void mce_restart(void)
737 if (next_interval) 746 if (next_interval)
738 cancel_delayed_work(&mcheck_work); 747 cancel_delayed_work(&mcheck_work);
739 /* Timer race is harmless here */ 748 /* Timer race is harmless here */
740 on_each_cpu(mce_init, NULL, 1, 1); 749 on_each_cpu(mce_init, NULL, 1);
741 next_interval = check_interval * HZ; 750 next_interval = check_interval * HZ;
742 if (next_interval) 751 if (next_interval)
743 schedule_delayed_work(&mcheck_work, 752 schedule_delayed_work(&mcheck_work,
@@ -753,10 +762,14 @@ DEFINE_PER_CPU(struct sys_device, device_mce);
753 762
754/* Why are there no generic functions for this? */ 763/* Why are there no generic functions for this? */
755#define ACCESSOR(name, var, start) \ 764#define ACCESSOR(name, var, start) \
756 static ssize_t show_ ## name(struct sys_device *s, char *buf) { \ 765 static ssize_t show_ ## name(struct sys_device *s, \
766 struct sysdev_attribute *attr, \
767 char *buf) { \
757 return sprintf(buf, "%lx\n", (unsigned long)var); \ 768 return sprintf(buf, "%lx\n", (unsigned long)var); \
758 } \ 769 } \
759 static ssize_t set_ ## name(struct sys_device *s,const char *buf,size_t siz) { \ 770 static ssize_t set_ ## name(struct sys_device *s, \
771 struct sysdev_attribute *attr, \
772 const char *buf, size_t siz) { \
760 char *end; \ 773 char *end; \
761 unsigned long new = simple_strtoul(buf, &end, 0); \ 774 unsigned long new = simple_strtoul(buf, &end, 0); \
762 if (end == buf) return -EINVAL; \ 775 if (end == buf) return -EINVAL; \
@@ -766,7 +779,10 @@ DEFINE_PER_CPU(struct sys_device, device_mce);
766 } \ 779 } \
767 static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name); 780 static SYSDEV_ATTR(name, 0644, show_ ## name, set_ ## name);
768 781
769/* TBD should generate these dynamically based on number of available banks */ 782/*
783 * TBD should generate these dynamically based on number of available banks.
784 * Have only 6 contol banks in /sysfs until then.
785 */
770ACCESSOR(bank0ctl,bank[0],mce_restart()) 786ACCESSOR(bank0ctl,bank[0],mce_restart())
771ACCESSOR(bank1ctl,bank[1],mce_restart()) 787ACCESSOR(bank1ctl,bank[1],mce_restart())
772ACCESSOR(bank2ctl,bank[2],mce_restart()) 788ACCESSOR(bank2ctl,bank[2],mce_restart())
@@ -774,14 +790,16 @@ ACCESSOR(bank3ctl,bank[3],mce_restart())
774ACCESSOR(bank4ctl,bank[4],mce_restart()) 790ACCESSOR(bank4ctl,bank[4],mce_restart())
775ACCESSOR(bank5ctl,bank[5],mce_restart()) 791ACCESSOR(bank5ctl,bank[5],mce_restart())
776 792
777static ssize_t show_trigger(struct sys_device *s, char *buf) 793static ssize_t show_trigger(struct sys_device *s, struct sysdev_attribute *attr,
794 char *buf)
778{ 795{
779 strcpy(buf, trigger); 796 strcpy(buf, trigger);
780 strcat(buf, "\n"); 797 strcat(buf, "\n");
781 return strlen(trigger) + 1; 798 return strlen(trigger) + 1;
782} 799}
783 800
784static ssize_t set_trigger(struct sys_device *s,const char *buf,size_t siz) 801static ssize_t set_trigger(struct sys_device *s, struct sysdev_attribute *attr,
802 const char *buf,size_t siz)
785{ 803{
786 char *p; 804 char *p;
787 int len; 805 int len;
@@ -794,12 +812,12 @@ static ssize_t set_trigger(struct sys_device *s,const char *buf,size_t siz)
794} 812}
795 813
796static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger); 814static SYSDEV_ATTR(trigger, 0644, show_trigger, set_trigger);
797ACCESSOR(tolerant,tolerant,) 815static SYSDEV_INT_ATTR(tolerant, 0644, tolerant);
798ACCESSOR(check_interval,check_interval,mce_restart()) 816ACCESSOR(check_interval,check_interval,mce_restart())
799static struct sysdev_attribute *mce_attributes[] = { 817static struct sysdev_attribute *mce_attributes[] = {
800 &attr_bank0ctl, &attr_bank1ctl, &attr_bank2ctl, 818 &attr_bank0ctl, &attr_bank1ctl, &attr_bank2ctl,
801 &attr_bank3ctl, &attr_bank4ctl, &attr_bank5ctl, 819 &attr_bank3ctl, &attr_bank4ctl, &attr_bank5ctl,
802 &attr_tolerant, &attr_check_interval, &attr_trigger, 820 &attr_tolerant.attr, &attr_check_interval, &attr_trigger,
803 NULL 821 NULL
804}; 822};
805 823
diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
index 7c9a813e1193..88736cadbaa6 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
@@ -527,7 +527,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
527 if (err) 527 if (err)
528 goto out_free; 528 goto out_free;
529 529
530 for_each_cpu_mask(i, b->cpus) { 530 for_each_cpu_mask_nr(i, b->cpus) {
531 if (i == cpu) 531 if (i == cpu)
532 continue; 532 continue;
533 533
@@ -617,7 +617,7 @@ static void threshold_remove_bank(unsigned int cpu, int bank)
617#endif 617#endif
618 618
619 /* remove all sibling symlinks before unregistering */ 619 /* remove all sibling symlinks before unregistering */
620 for_each_cpu_mask(i, b->cpus) { 620 for_each_cpu_mask_nr(i, b->cpus) {
621 if (i == cpu) 621 if (i == cpu)
622 continue; 622 continue;
623 623
diff --git a/arch/x86/kernel/cpu/mcheck/non-fatal.c b/arch/x86/kernel/cpu/mcheck/non-fatal.c
index 00ccb6c14ec2..cc1fccdd31e0 100644
--- a/arch/x86/kernel/cpu/mcheck/non-fatal.c
+++ b/arch/x86/kernel/cpu/mcheck/non-fatal.c
@@ -59,7 +59,7 @@ static DECLARE_DELAYED_WORK(mce_work, mce_work_fn);
59 59
60static void mce_work_fn(struct work_struct *work) 60static void mce_work_fn(struct work_struct *work)
61{ 61{
62 on_each_cpu(mce_checkregs, NULL, 1, 1); 62 on_each_cpu(mce_checkregs, NULL, 1);
63 schedule_delayed_work(&mce_work, round_jiffies_relative(MCE_RATE)); 63 schedule_delayed_work(&mce_work, round_jiffies_relative(MCE_RATE));
64} 64}
65 65
diff --git a/arch/x86/kernel/cpu/mcheck/p4.c b/arch/x86/kernel/cpu/mcheck/p4.c
index cb03345554a5..9b60fce09f75 100644
--- a/arch/x86/kernel/cpu/mcheck/p4.c
+++ b/arch/x86/kernel/cpu/mcheck/p4.c
@@ -8,7 +8,7 @@
8#include <linux/interrupt.h> 8#include <linux/interrupt.h>
9#include <linux/smp.h> 9#include <linux/smp.h>
10 10
11#include <asm/processor.h> 11#include <asm/processor.h>
12#include <asm/system.h> 12#include <asm/system.h>
13#include <asm/msr.h> 13#include <asm/msr.h>
14#include <asm/apic.h> 14#include <asm/apic.h>
@@ -32,12 +32,12 @@ struct intel_mce_extended_msrs {
32 /* u32 *reserved[]; */ 32 /* u32 *reserved[]; */
33}; 33};
34 34
35static int mce_num_extended_msrs = 0; 35static int mce_num_extended_msrs;
36 36
37 37
38#ifdef CONFIG_X86_MCE_P4THERMAL 38#ifdef CONFIG_X86_MCE_P4THERMAL
39static void unexpected_thermal_interrupt(struct pt_regs *regs) 39static void unexpected_thermal_interrupt(struct pt_regs *regs)
40{ 40{
41 printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n", 41 printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
42 smp_processor_id()); 42 smp_processor_id());
43 add_taint(TAINT_MACHINE_CHECK); 43 add_taint(TAINT_MACHINE_CHECK);
@@ -83,7 +83,7 @@ static void intel_init_thermal(struct cpuinfo_x86 *c)
83 * be some SMM goo which handles it, so we can't even put a handler 83 * be some SMM goo which handles it, so we can't even put a handler
84 * since it might be delivered via SMI already -zwanem. 84 * since it might be delivered via SMI already -zwanem.
85 */ 85 */
86 rdmsr (MSR_IA32_MISC_ENABLE, l, h); 86 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
87 h = apic_read(APIC_LVTTHMR); 87 h = apic_read(APIC_LVTTHMR);
88 if ((l & (1<<3)) && (h & APIC_DM_SMI)) { 88 if ((l & (1<<3)) && (h & APIC_DM_SMI)) {
89 printk(KERN_DEBUG "CPU%d: Thermal monitoring handled by SMI\n", 89 printk(KERN_DEBUG "CPU%d: Thermal monitoring handled by SMI\n",
@@ -91,7 +91,7 @@ static void intel_init_thermal(struct cpuinfo_x86 *c)
91 return; /* -EBUSY */ 91 return; /* -EBUSY */
92 } 92 }
93 93
94 /* check whether a vector already exists, temporarily masked? */ 94 /* check whether a vector already exists, temporarily masked? */
95 if (h & APIC_VECTOR_MASK) { 95 if (h & APIC_VECTOR_MASK) {
96 printk(KERN_DEBUG "CPU%d: Thermal LVT vector (%#x) already " 96 printk(KERN_DEBUG "CPU%d: Thermal LVT vector (%#x) already "
97 "installed\n", 97 "installed\n",
@@ -102,20 +102,20 @@ static void intel_init_thermal(struct cpuinfo_x86 *c)
102 /* The temperature transition interrupt handler setup */ 102 /* The temperature transition interrupt handler setup */
103 h = THERMAL_APIC_VECTOR; /* our delivery vector */ 103 h = THERMAL_APIC_VECTOR; /* our delivery vector */
104 h |= (APIC_DM_FIXED | APIC_LVT_MASKED); /* we'll mask till we're ready */ 104 h |= (APIC_DM_FIXED | APIC_LVT_MASKED); /* we'll mask till we're ready */
105 apic_write_around(APIC_LVTTHMR, h); 105 apic_write(APIC_LVTTHMR, h);
106 106
107 rdmsr (MSR_IA32_THERM_INTERRUPT, l, h); 107 rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
108 wrmsr (MSR_IA32_THERM_INTERRUPT, l | 0x03 , h); 108 wrmsr(MSR_IA32_THERM_INTERRUPT, l | 0x03 , h);
109 109
110 /* ok we're good to go... */ 110 /* ok we're good to go... */
111 vendor_thermal_interrupt = intel_thermal_interrupt; 111 vendor_thermal_interrupt = intel_thermal_interrupt;
112
113 rdmsr (MSR_IA32_MISC_ENABLE, l, h);
114 wrmsr (MSR_IA32_MISC_ENABLE, l | (1<<3), h);
115 112
116 l = apic_read (APIC_LVTTHMR); 113 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
117 apic_write_around (APIC_LVTTHMR, l & ~APIC_LVT_MASKED); 114 wrmsr(MSR_IA32_MISC_ENABLE, l | (1<<3), h);
118 printk (KERN_INFO "CPU%d: Thermal monitoring enabled\n", cpu); 115
116 l = apic_read(APIC_LVTTHMR);
117 apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
118 printk(KERN_INFO "CPU%d: Thermal monitoring enabled\n", cpu);
119 119
120 /* enable thermal throttle processing */ 120 /* enable thermal throttle processing */
121 atomic_set(&therm_throt_en, 1); 121 atomic_set(&therm_throt_en, 1);
@@ -129,28 +129,28 @@ static inline void intel_get_extended_msrs(struct intel_mce_extended_msrs *r)
129{ 129{
130 u32 h; 130 u32 h;
131 131
132 rdmsr (MSR_IA32_MCG_EAX, r->eax, h); 132 rdmsr(MSR_IA32_MCG_EAX, r->eax, h);
133 rdmsr (MSR_IA32_MCG_EBX, r->ebx, h); 133 rdmsr(MSR_IA32_MCG_EBX, r->ebx, h);
134 rdmsr (MSR_IA32_MCG_ECX, r->ecx, h); 134 rdmsr(MSR_IA32_MCG_ECX, r->ecx, h);
135 rdmsr (MSR_IA32_MCG_EDX, r->edx, h); 135 rdmsr(MSR_IA32_MCG_EDX, r->edx, h);
136 rdmsr (MSR_IA32_MCG_ESI, r->esi, h); 136 rdmsr(MSR_IA32_MCG_ESI, r->esi, h);
137 rdmsr (MSR_IA32_MCG_EDI, r->edi, h); 137 rdmsr(MSR_IA32_MCG_EDI, r->edi, h);
138 rdmsr (MSR_IA32_MCG_EBP, r->ebp, h); 138 rdmsr(MSR_IA32_MCG_EBP, r->ebp, h);
139 rdmsr (MSR_IA32_MCG_ESP, r->esp, h); 139 rdmsr(MSR_IA32_MCG_ESP, r->esp, h);
140 rdmsr (MSR_IA32_MCG_EFLAGS, r->eflags, h); 140 rdmsr(MSR_IA32_MCG_EFLAGS, r->eflags, h);
141 rdmsr (MSR_IA32_MCG_EIP, r->eip, h); 141 rdmsr(MSR_IA32_MCG_EIP, r->eip, h);
142} 142}
143 143
144static void intel_machine_check(struct pt_regs * regs, long error_code) 144static void intel_machine_check(struct pt_regs *regs, long error_code)
145{ 145{
146 int recover=1; 146 int recover = 1;
147 u32 alow, ahigh, high, low; 147 u32 alow, ahigh, high, low;
148 u32 mcgstl, mcgsth; 148 u32 mcgstl, mcgsth;
149 int i; 149 int i;
150 150
151 rdmsr (MSR_IA32_MCG_STATUS, mcgstl, mcgsth); 151 rdmsr(MSR_IA32_MCG_STATUS, mcgstl, mcgsth);
152 if (mcgstl & (1<<0)) /* Recoverable ? */ 152 if (mcgstl & (1<<0)) /* Recoverable ? */
153 recover=0; 153 recover = 0;
154 154
155 printk(KERN_EMERG "CPU %d: Machine Check Exception: %08x%08x\n", 155 printk(KERN_EMERG "CPU %d: Machine Check Exception: %08x%08x\n",
156 smp_processor_id(), mcgsth, mcgstl); 156 smp_processor_id(), mcgsth, mcgstl);
@@ -191,20 +191,20 @@ static void intel_machine_check(struct pt_regs * regs, long error_code)
191 } 191 }
192 192
193 if (recover & 2) 193 if (recover & 2)
194 panic ("CPU context corrupt"); 194 panic("CPU context corrupt");
195 if (recover & 1) 195 if (recover & 1)
196 panic ("Unable to continue"); 196 panic("Unable to continue");
197 197
198 printk(KERN_EMERG "Attempting to continue.\n"); 198 printk(KERN_EMERG "Attempting to continue.\n");
199 /* 199 /*
200 * Do not clear the MSR_IA32_MCi_STATUS if the error is not 200 * Do not clear the MSR_IA32_MCi_STATUS if the error is not
201 * recoverable/continuable.This will allow BIOS to look at the MSRs 201 * recoverable/continuable.This will allow BIOS to look at the MSRs
202 * for errors if the OS could not log the error. 202 * for errors if the OS could not log the error.
203 */ 203 */
204 for (i=0; i<nr_mce_banks; i++) { 204 for (i = 0; i < nr_mce_banks; i++) {
205 u32 msr; 205 u32 msr;
206 msr = MSR_IA32_MC0_STATUS+i*4; 206 msr = MSR_IA32_MC0_STATUS+i*4;
207 rdmsr (msr, low, high); 207 rdmsr(msr, low, high);
208 if (high&(1<<31)) { 208 if (high&(1<<31)) {
209 /* Clear it */ 209 /* Clear it */
210 wrmsr(msr, 0UL, 0UL); 210 wrmsr(msr, 0UL, 0UL);
@@ -214,7 +214,7 @@ static void intel_machine_check(struct pt_regs * regs, long error_code)
214 } 214 }
215 } 215 }
216 mcgstl &= ~(1<<2); 216 mcgstl &= ~(1<<2);
217 wrmsr (MSR_IA32_MCG_STATUS,mcgstl, mcgsth); 217 wrmsr(MSR_IA32_MCG_STATUS, mcgstl, mcgsth);
218} 218}
219 219
220 220
@@ -222,30 +222,30 @@ void intel_p4_mcheck_init(struct cpuinfo_x86 *c)
222{ 222{
223 u32 l, h; 223 u32 l, h;
224 int i; 224 int i;
225 225
226 machine_check_vector = intel_machine_check; 226 machine_check_vector = intel_machine_check;
227 wmb(); 227 wmb();
228 228
229 printk (KERN_INFO "Intel machine check architecture supported.\n"); 229 printk(KERN_INFO "Intel machine check architecture supported.\n");
230 rdmsr (MSR_IA32_MCG_CAP, l, h); 230 rdmsr(MSR_IA32_MCG_CAP, l, h);
231 if (l & (1<<8)) /* Control register present ? */ 231 if (l & (1<<8)) /* Control register present ? */
232 wrmsr (MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff); 232 wrmsr(MSR_IA32_MCG_CTL, 0xffffffff, 0xffffffff);
233 nr_mce_banks = l & 0xff; 233 nr_mce_banks = l & 0xff;
234 234
235 for (i=0; i<nr_mce_banks; i++) { 235 for (i = 0; i < nr_mce_banks; i++) {
236 wrmsr (MSR_IA32_MC0_CTL+4*i, 0xffffffff, 0xffffffff); 236 wrmsr(MSR_IA32_MC0_CTL+4*i, 0xffffffff, 0xffffffff);
237 wrmsr (MSR_IA32_MC0_STATUS+4*i, 0x0, 0x0); 237 wrmsr(MSR_IA32_MC0_STATUS+4*i, 0x0, 0x0);
238 } 238 }
239 239
240 set_in_cr4 (X86_CR4_MCE); 240 set_in_cr4(X86_CR4_MCE);
241 printk (KERN_INFO "Intel machine check reporting enabled on CPU#%d.\n", 241 printk(KERN_INFO "Intel machine check reporting enabled on CPU#%d.\n",
242 smp_processor_id()); 242 smp_processor_id());
243 243
244 /* Check for P4/Xeon extended MCE MSRs */ 244 /* Check for P4/Xeon extended MCE MSRs */
245 rdmsr (MSR_IA32_MCG_CAP, l, h); 245 rdmsr(MSR_IA32_MCG_CAP, l, h);
246 if (l & (1<<9)) {/* MCG_EXT_P */ 246 if (l & (1<<9)) {/* MCG_EXT_P */
247 mce_num_extended_msrs = (l >> 16) & 0xff; 247 mce_num_extended_msrs = (l >> 16) & 0xff;
248 printk (KERN_INFO "CPU%d: Intel P4/Xeon Extended MCE MSRs (%d)" 248 printk(KERN_INFO "CPU%d: Intel P4/Xeon Extended MCE MSRs (%d)"
249 " available\n", 249 " available\n",
250 smp_processor_id(), mce_num_extended_msrs); 250 smp_processor_id(), mce_num_extended_msrs);
251 251
diff --git a/arch/x86/kernel/cpu/mcheck/therm_throt.c b/arch/x86/kernel/cpu/mcheck/therm_throt.c
index 1f4cc48c14c6..d5ae2243f0b9 100644
--- a/arch/x86/kernel/cpu/mcheck/therm_throt.c
+++ b/arch/x86/kernel/cpu/mcheck/therm_throt.c
@@ -35,6 +35,7 @@ atomic_t therm_throt_en = ATOMIC_INIT(0);
35 35
36#define define_therm_throt_sysdev_show_func(name) \ 36#define define_therm_throt_sysdev_show_func(name) \
37static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev, \ 37static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev, \
38 struct sysdev_attribute *attr, \
38 char *buf) \ 39 char *buf) \
39{ \ 40{ \
40 unsigned int cpu = dev->id; \ 41 unsigned int cpu = dev->id; \
diff --git a/arch/x86/kernel/cpu/mtrr/generic.c b/arch/x86/kernel/cpu/mtrr/generic.c
index 5d241ce94a44..509bd3d9eacd 100644
--- a/arch/x86/kernel/cpu/mtrr/generic.c
+++ b/arch/x86/kernel/cpu/mtrr/generic.c
@@ -37,7 +37,7 @@ static struct fixed_range_block fixed_range_blocks[] = {
37static unsigned long smp_changes_mask; 37static unsigned long smp_changes_mask;
38static struct mtrr_state mtrr_state = {}; 38static struct mtrr_state mtrr_state = {};
39static int mtrr_state_set; 39static int mtrr_state_set;
40static u64 tom2; 40u64 mtrr_tom2;
41 41
42#undef MODULE_PARAM_PREFIX 42#undef MODULE_PARAM_PREFIX
43#define MODULE_PARAM_PREFIX "mtrr." 43#define MODULE_PARAM_PREFIX "mtrr."
@@ -139,8 +139,8 @@ u8 mtrr_type_lookup(u64 start, u64 end)
139 } 139 }
140 } 140 }
141 141
142 if (tom2) { 142 if (mtrr_tom2) {
143 if (start >= (1ULL<<32) && (end < tom2)) 143 if (start >= (1ULL<<32) && (end < mtrr_tom2))
144 return MTRR_TYPE_WRBACK; 144 return MTRR_TYPE_WRBACK;
145 } 145 }
146 146
@@ -158,6 +158,20 @@ get_mtrr_var_range(unsigned int index, struct mtrr_var_range *vr)
158 rdmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi); 158 rdmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi);
159} 159}
160 160
161/* fill the MSR pair relating to a var range */
162void fill_mtrr_var_range(unsigned int index,
163 u32 base_lo, u32 base_hi, u32 mask_lo, u32 mask_hi)
164{
165 struct mtrr_var_range *vr;
166
167 vr = mtrr_state.var_ranges;
168
169 vr[index].base_lo = base_lo;
170 vr[index].base_hi = base_hi;
171 vr[index].mask_lo = mask_lo;
172 vr[index].mask_hi = mask_hi;
173}
174
161static void 175static void
162get_fixed_ranges(mtrr_type * frs) 176get_fixed_ranges(mtrr_type * frs)
163{ 177{
@@ -213,13 +227,13 @@ void __init get_mtrr_state(void)
213 mtrr_state.enabled = (lo & 0xc00) >> 10; 227 mtrr_state.enabled = (lo & 0xc00) >> 10;
214 228
215 if (amd_special_default_mtrr()) { 229 if (amd_special_default_mtrr()) {
216 unsigned lo, hi; 230 unsigned low, high;
217 /* TOP_MEM2 */ 231 /* TOP_MEM2 */
218 rdmsr(MSR_K8_TOP_MEM2, lo, hi); 232 rdmsr(MSR_K8_TOP_MEM2, low, high);
219 tom2 = hi; 233 mtrr_tom2 = high;
220 tom2 <<= 32; 234 mtrr_tom2 <<= 32;
221 tom2 |= lo; 235 mtrr_tom2 |= low;
222 tom2 &= 0xffffff8000000ULL; 236 mtrr_tom2 &= 0xffffff800000ULL;
223 } 237 }
224 if (mtrr_show) { 238 if (mtrr_show) {
225 int high_width; 239 int high_width;
@@ -251,9 +265,9 @@ void __init get_mtrr_state(void)
251 else 265 else
252 printk(KERN_INFO "MTRR %u disabled\n", i); 266 printk(KERN_INFO "MTRR %u disabled\n", i);
253 } 267 }
254 if (tom2) { 268 if (mtrr_tom2) {
255 printk(KERN_INFO "TOM2: %016llx aka %lldM\n", 269 printk(KERN_INFO "TOM2: %016llx aka %lldM\n",
256 tom2, tom2>>20); 270 mtrr_tom2, mtrr_tom2>>20);
257 } 271 }
258 } 272 }
259 mtrr_state_set = 1; 273 mtrr_state_set = 1;
@@ -328,7 +342,7 @@ static void set_fixed_range(int msr, bool *changed, unsigned int *msrwords)
328 342
329 if (lo != msrwords[0] || hi != msrwords[1]) { 343 if (lo != msrwords[0] || hi != msrwords[1]) {
330 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD && 344 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
331 boot_cpu_data.x86 == 15 && 345 (boot_cpu_data.x86 >= 0x0f && boot_cpu_data.x86 <= 0x11) &&
332 ((msrwords[0] | msrwords[1]) & K8_MTRR_RDMEM_WRMEM_MASK)) 346 ((msrwords[0] | msrwords[1]) & K8_MTRR_RDMEM_WRMEM_MASK))
333 k8_enable_fixed_iorrs(); 347 k8_enable_fixed_iorrs();
334 mtrr_wrmsr(msr, msrwords[0], msrwords[1]); 348 mtrr_wrmsr(msr, msrwords[0], msrwords[1]);
diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c
index 6a1e278d9323..6f23969c8faf 100644
--- a/arch/x86/kernel/cpu/mtrr/main.c
+++ b/arch/x86/kernel/cpu/mtrr/main.c
@@ -37,6 +37,7 @@
37#include <linux/smp.h> 37#include <linux/smp.h>
38#include <linux/cpu.h> 38#include <linux/cpu.h>
39#include <linux/mutex.h> 39#include <linux/mutex.h>
40#include <linux/sort.h>
40 41
41#include <asm/e820.h> 42#include <asm/e820.h>
42#include <asm/mtrr.h> 43#include <asm/mtrr.h>
@@ -222,7 +223,7 @@ static void set_mtrr(unsigned int reg, unsigned long base,
222 atomic_set(&data.gate,0); 223 atomic_set(&data.gate,0);
223 224
224 /* Start the ball rolling on other CPUs */ 225 /* Start the ball rolling on other CPUs */
225 if (smp_call_function(ipi_handler, &data, 1, 0) != 0) 226 if (smp_call_function(ipi_handler, &data, 0) != 0)
226 panic("mtrr: timed out waiting for other CPUs\n"); 227 panic("mtrr: timed out waiting for other CPUs\n");
227 228
228 local_irq_save(flags); 229 local_irq_save(flags);
@@ -609,6 +610,787 @@ static struct sysdev_driver mtrr_sysdev_driver = {
609 .resume = mtrr_restore, 610 .resume = mtrr_restore,
610}; 611};
611 612
613/* should be related to MTRR_VAR_RANGES nums */
614#define RANGE_NUM 256
615
616struct res_range {
617 unsigned long start;
618 unsigned long end;
619};
620
621static int __init
622add_range(struct res_range *range, int nr_range, unsigned long start,
623 unsigned long end)
624{
625 /* out of slots */
626 if (nr_range >= RANGE_NUM)
627 return nr_range;
628
629 range[nr_range].start = start;
630 range[nr_range].end = end;
631
632 nr_range++;
633
634 return nr_range;
635}
636
637static int __init
638add_range_with_merge(struct res_range *range, int nr_range, unsigned long start,
639 unsigned long end)
640{
641 int i;
642
643 /* try to merge it with old one */
644 for (i = 0; i < nr_range; i++) {
645 unsigned long final_start, final_end;
646 unsigned long common_start, common_end;
647
648 if (!range[i].end)
649 continue;
650
651 common_start = max(range[i].start, start);
652 common_end = min(range[i].end, end);
653 if (common_start > common_end + 1)
654 continue;
655
656 final_start = min(range[i].start, start);
657 final_end = max(range[i].end, end);
658
659 range[i].start = final_start;
660 range[i].end = final_end;
661 return nr_range;
662 }
663
664 /* need to add that */
665 return add_range(range, nr_range, start, end);
666}
667
668static void __init
669subtract_range(struct res_range *range, unsigned long start, unsigned long end)
670{
671 int i, j;
672
673 for (j = 0; j < RANGE_NUM; j++) {
674 if (!range[j].end)
675 continue;
676
677 if (start <= range[j].start && end >= range[j].end) {
678 range[j].start = 0;
679 range[j].end = 0;
680 continue;
681 }
682
683 if (start <= range[j].start && end < range[j].end &&
684 range[j].start < end + 1) {
685 range[j].start = end + 1;
686 continue;
687 }
688
689
690 if (start > range[j].start && end >= range[j].end &&
691 range[j].end > start - 1) {
692 range[j].end = start - 1;
693 continue;
694 }
695
696 if (start > range[j].start && end < range[j].end) {
697 /* find the new spare */
698 for (i = 0; i < RANGE_NUM; i++) {
699 if (range[i].end == 0)
700 break;
701 }
702 if (i < RANGE_NUM) {
703 range[i].end = range[j].end;
704 range[i].start = end + 1;
705 } else {
706 printk(KERN_ERR "run of slot in ranges\n");
707 }
708 range[j].end = start - 1;
709 continue;
710 }
711 }
712}
713
714static int __init cmp_range(const void *x1, const void *x2)
715{
716 const struct res_range *r1 = x1;
717 const struct res_range *r2 = x2;
718 long start1, start2;
719
720 start1 = r1->start;
721 start2 = r2->start;
722
723 return start1 - start2;
724}
725
726struct var_mtrr_range_state {
727 unsigned long base_pfn;
728 unsigned long size_pfn;
729 mtrr_type type;
730};
731
732struct var_mtrr_range_state __initdata range_state[RANGE_NUM];
733static int __initdata debug_print;
734
735static int __init
736x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
737 unsigned long extra_remove_base,
738 unsigned long extra_remove_size)
739{
740 unsigned long i, base, size;
741 mtrr_type type;
742
743 for (i = 0; i < num_var_ranges; i++) {
744 type = range_state[i].type;
745 if (type != MTRR_TYPE_WRBACK)
746 continue;
747 base = range_state[i].base_pfn;
748 size = range_state[i].size_pfn;
749 nr_range = add_range_with_merge(range, nr_range, base,
750 base + size - 1);
751 }
752 if (debug_print) {
753 printk(KERN_DEBUG "After WB checking\n");
754 for (i = 0; i < nr_range; i++)
755 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
756 range[i].start, range[i].end + 1);
757 }
758
759 /* take out UC ranges */
760 for (i = 0; i < num_var_ranges; i++) {
761 type = range_state[i].type;
762 if (type != MTRR_TYPE_UNCACHABLE)
763 continue;
764 size = range_state[i].size_pfn;
765 if (!size)
766 continue;
767 base = range_state[i].base_pfn;
768 subtract_range(range, base, base + size - 1);
769 }
770 if (extra_remove_size)
771 subtract_range(range, extra_remove_base,
772 extra_remove_base + extra_remove_size - 1);
773
774 /* get new range num */
775 nr_range = 0;
776 for (i = 0; i < RANGE_NUM; i++) {
777 if (!range[i].end)
778 continue;
779 nr_range++;
780 }
781 if (debug_print) {
782 printk(KERN_DEBUG "After UC checking\n");
783 for (i = 0; i < nr_range; i++)
784 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
785 range[i].start, range[i].end + 1);
786 }
787
788 /* sort the ranges */
789 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
790 if (debug_print) {
791 printk(KERN_DEBUG "After sorting\n");
792 for (i = 0; i < nr_range; i++)
793 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
794 range[i].start, range[i].end + 1);
795 }
796
797 /* clear those is not used */
798 for (i = nr_range; i < RANGE_NUM; i++)
799 memset(&range[i], 0, sizeof(range[i]));
800
801 return nr_range;
802}
803
804static struct res_range __initdata range[RANGE_NUM];
805
806#ifdef CONFIG_MTRR_SANITIZER
807
808static unsigned long __init sum_ranges(struct res_range *range, int nr_range)
809{
810 unsigned long sum;
811 int i;
812
813 sum = 0;
814 for (i = 0; i < nr_range; i++)
815 sum += range[i].end + 1 - range[i].start;
816
817 return sum;
818}
819
820static int enable_mtrr_cleanup __initdata =
821 CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT;
822
823static int __init disable_mtrr_cleanup_setup(char *str)
824{
825 if (enable_mtrr_cleanup != -1)
826 enable_mtrr_cleanup = 0;
827 return 0;
828}
829early_param("disable_mtrr_cleanup", disable_mtrr_cleanup_setup);
830
831static int __init enable_mtrr_cleanup_setup(char *str)
832{
833 if (enable_mtrr_cleanup != -1)
834 enable_mtrr_cleanup = 1;
835 return 0;
836}
837early_param("enble_mtrr_cleanup", enable_mtrr_cleanup_setup);
838
839struct var_mtrr_state {
840 unsigned long range_startk;
841 unsigned long range_sizek;
842 unsigned long chunk_sizek;
843 unsigned long gran_sizek;
844 unsigned int reg;
845};
846
847static void __init
848set_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
849 unsigned char type, unsigned int address_bits)
850{
851 u32 base_lo, base_hi, mask_lo, mask_hi;
852 u64 base, mask;
853
854 if (!sizek) {
855 fill_mtrr_var_range(reg, 0, 0, 0, 0);
856 return;
857 }
858
859 mask = (1ULL << address_bits) - 1;
860 mask &= ~((((u64)sizek) << 10) - 1);
861
862 base = ((u64)basek) << 10;
863
864 base |= type;
865 mask |= 0x800;
866
867 base_lo = base & ((1ULL<<32) - 1);
868 base_hi = base >> 32;
869
870 mask_lo = mask & ((1ULL<<32) - 1);
871 mask_hi = mask >> 32;
872
873 fill_mtrr_var_range(reg, base_lo, base_hi, mask_lo, mask_hi);
874}
875
876static void __init
877save_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
878 unsigned char type)
879{
880 range_state[reg].base_pfn = basek >> (PAGE_SHIFT - 10);
881 range_state[reg].size_pfn = sizek >> (PAGE_SHIFT - 10);
882 range_state[reg].type = type;
883}
884
885static void __init
886set_var_mtrr_all(unsigned int address_bits)
887{
888 unsigned long basek, sizek;
889 unsigned char type;
890 unsigned int reg;
891
892 for (reg = 0; reg < num_var_ranges; reg++) {
893 basek = range_state[reg].base_pfn << (PAGE_SHIFT - 10);
894 sizek = range_state[reg].size_pfn << (PAGE_SHIFT - 10);
895 type = range_state[reg].type;
896
897 set_var_mtrr(reg, basek, sizek, type, address_bits);
898 }
899}
900
901static unsigned int __init
902range_to_mtrr(unsigned int reg, unsigned long range_startk,
903 unsigned long range_sizek, unsigned char type)
904{
905 if (!range_sizek || (reg >= num_var_ranges))
906 return reg;
907
908 while (range_sizek) {
909 unsigned long max_align, align;
910 unsigned long sizek;
911
912 /* Compute the maximum size I can make a range */
913 if (range_startk)
914 max_align = ffs(range_startk) - 1;
915 else
916 max_align = 32;
917 align = fls(range_sizek) - 1;
918 if (align > max_align)
919 align = max_align;
920
921 sizek = 1 << align;
922 if (debug_print)
923 printk(KERN_DEBUG "Setting variable MTRR %d, "
924 "base: %ldMB, range: %ldMB, type %s\n",
925 reg, range_startk >> 10, sizek >> 10,
926 (type == MTRR_TYPE_UNCACHABLE)?"UC":
927 ((type == MTRR_TYPE_WRBACK)?"WB":"Other")
928 );
929 save_var_mtrr(reg++, range_startk, sizek, type);
930 range_startk += sizek;
931 range_sizek -= sizek;
932 if (reg >= num_var_ranges)
933 break;
934 }
935 return reg;
936}
937
938static unsigned __init
939range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
940 unsigned long sizek)
941{
942 unsigned long hole_basek, hole_sizek;
943 unsigned long second_basek, second_sizek;
944 unsigned long range0_basek, range0_sizek;
945 unsigned long range_basek, range_sizek;
946 unsigned long chunk_sizek;
947 unsigned long gran_sizek;
948
949 hole_basek = 0;
950 hole_sizek = 0;
951 second_basek = 0;
952 second_sizek = 0;
953 chunk_sizek = state->chunk_sizek;
954 gran_sizek = state->gran_sizek;
955
956 /* align with gran size, prevent small block used up MTRRs */
957 range_basek = ALIGN(state->range_startk, gran_sizek);
958 if ((range_basek > basek) && basek)
959 return second_sizek;
960 state->range_sizek -= (range_basek - state->range_startk);
961 range_sizek = ALIGN(state->range_sizek, gran_sizek);
962
963 while (range_sizek > state->range_sizek) {
964 range_sizek -= gran_sizek;
965 if (!range_sizek)
966 return 0;
967 }
968 state->range_sizek = range_sizek;
969
970 /* try to append some small hole */
971 range0_basek = state->range_startk;
972 range0_sizek = ALIGN(state->range_sizek, chunk_sizek);
973 if (range0_sizek == state->range_sizek) {
974 if (debug_print)
975 printk(KERN_DEBUG "rangeX: %016lx - %016lx\n",
976 range0_basek<<10,
977 (range0_basek + state->range_sizek)<<10);
978 state->reg = range_to_mtrr(state->reg, range0_basek,
979 state->range_sizek, MTRR_TYPE_WRBACK);
980 return 0;
981 }
982
983 range0_sizek -= chunk_sizek;
984 if (range0_sizek && sizek) {
985 while (range0_basek + range0_sizek > (basek + sizek)) {
986 range0_sizek -= chunk_sizek;
987 if (!range0_sizek)
988 break;
989 }
990 }
991
992 if (range0_sizek) {
993 if (debug_print)
994 printk(KERN_DEBUG "range0: %016lx - %016lx\n",
995 range0_basek<<10,
996 (range0_basek + range0_sizek)<<10);
997 state->reg = range_to_mtrr(state->reg, range0_basek,
998 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 }
1011
1012 /* if last piece, only could one hole near end */
1013 if ((second_basek || !basek) &&
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;
1027 second_basek = 0;
1028 second_sizek = 0;
1029 }
1030
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) {
1037 if (debug_print)
1038 printk(KERN_DEBUG "hole: %016lx - %016lx\n",
1039 hole_basek<<10, (hole_basek + hole_sizek)<<10);
1040 state->reg = range_to_mtrr(state->reg, hole_basek, hole_sizek,
1041 MTRR_TYPE_UNCACHABLE);
1042
1043 }
1044
1045 return second_sizek;
1046}
1047
1048static void __init
1049set_var_mtrr_range(struct var_mtrr_state *state, unsigned long base_pfn,
1050 unsigned long size_pfn)
1051{
1052 unsigned long basek, sizek;
1053 unsigned long second_sizek = 0;
1054
1055 if (state->reg >= num_var_ranges)
1056 return;
1057
1058 basek = base_pfn << (PAGE_SHIFT - 10);
1059 sizek = size_pfn << (PAGE_SHIFT - 10);
1060
1061 /* See if I can merge with the last range */
1062 if ((basek <= 1024) ||
1063 (state->range_startk + state->range_sizek == basek)) {
1064 unsigned long endk = basek + sizek;
1065 state->range_sizek = endk - state->range_startk;
1066 return;
1067 }
1068 /* Write the range mtrrs */
1069 if (state->range_sizek != 0)
1070 second_sizek = range_to_mtrr_with_hole(state, basek, sizek);
1071
1072 /* Allocate an msr */
1073 state->range_startk = basek + second_sizek;
1074 state->range_sizek = sizek - second_sizek;
1075}
1076
1077/* mininum size of mtrr block that can take hole */
1078static u64 mtrr_chunk_size __initdata = (256ULL<<20);
1079
1080static int __init parse_mtrr_chunk_size_opt(char *p)
1081{
1082 if (!p)
1083 return -EINVAL;
1084 mtrr_chunk_size = memparse(p, &p);
1085 return 0;
1086}
1087early_param("mtrr_chunk_size", parse_mtrr_chunk_size_opt);
1088
1089/* granity of mtrr of block */
1090static u64 mtrr_gran_size __initdata;
1091
1092static int __init parse_mtrr_gran_size_opt(char *p)
1093{
1094 if (!p)
1095 return -EINVAL;
1096 mtrr_gran_size = memparse(p, &p);
1097 return 0;
1098}
1099early_param("mtrr_gran_size", parse_mtrr_gran_size_opt);
1100
1101static int nr_mtrr_spare_reg __initdata =
1102 CONFIG_MTRR_SANITIZER_SPARE_REG_NR_DEFAULT;
1103
1104static int __init parse_mtrr_spare_reg(char *arg)
1105{
1106 if (arg)
1107 nr_mtrr_spare_reg = simple_strtoul(arg, NULL, 0);
1108 return 0;
1109}
1110
1111early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg);
1112
1113static int __init
1114x86_setup_var_mtrrs(struct res_range *range, int nr_range,
1115 u64 chunk_size, u64 gran_size)
1116{
1117 struct var_mtrr_state var_state;
1118 int i;
1119 int num_reg;
1120
1121 var_state.range_startk = 0;
1122 var_state.range_sizek = 0;
1123 var_state.reg = 0;
1124 var_state.chunk_sizek = chunk_size >> 10;
1125 var_state.gran_sizek = gran_size >> 10;
1126
1127 memset(range_state, 0, sizeof(range_state));
1128
1129 /* Write the range etc */
1130 for (i = 0; i < nr_range; i++)
1131 set_var_mtrr_range(&var_state, range[i].start,
1132 range[i].end - range[i].start + 1);
1133
1134 /* Write the last range */
1135 if (var_state.range_sizek != 0)
1136 range_to_mtrr_with_hole(&var_state, 0, 0);
1137
1138 num_reg = var_state.reg;
1139 /* Clear out the extra MTRR's */
1140 while (var_state.reg < num_var_ranges) {
1141 save_var_mtrr(var_state.reg, 0, 0, 0);
1142 var_state.reg++;
1143 }
1144
1145 return num_reg;
1146}
1147
1148struct mtrr_cleanup_result {
1149 unsigned long gran_sizek;
1150 unsigned long chunk_sizek;
1151 unsigned long lose_cover_sizek;
1152 unsigned int num_reg;
1153 int bad;
1154};
1155
1156/*
1157 * gran_size: 1M, 2M, ..., 2G
1158 * chunk size: gran_size, ..., 4G
1159 * so we need (2+13)*6
1160 */
1161#define NUM_RESULT 90
1162#define PSHIFT (PAGE_SHIFT - 10)
1163
1164static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
1165static struct res_range __initdata range_new[RANGE_NUM];
1166static unsigned long __initdata min_loss_pfn[RANGE_NUM];
1167
1168static int __init mtrr_cleanup(unsigned address_bits)
1169{
1170 unsigned long extra_remove_base, extra_remove_size;
1171 unsigned long i, base, size, def, dummy;
1172 mtrr_type type;
1173 int nr_range, nr_range_new;
1174 u64 chunk_size, gran_size;
1175 unsigned long range_sums, range_sums_new;
1176 int index_good;
1177 int num_reg_good;
1178
1179 /* extra one for all 0 */
1180 int num[MTRR_NUM_TYPES + 1];
1181
1182 if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1)
1183 return 0;
1184 rdmsr(MTRRdefType_MSR, def, dummy);
1185 def &= 0xff;
1186 if (def != MTRR_TYPE_UNCACHABLE)
1187 return 0;
1188
1189 /* get it and store it aside */
1190 memset(range_state, 0, sizeof(range_state));
1191 for (i = 0; i < num_var_ranges; i++) {
1192 mtrr_if->get(i, &base, &size, &type);
1193 range_state[i].base_pfn = base;
1194 range_state[i].size_pfn = size;
1195 range_state[i].type = type;
1196 }
1197
1198 /* check entries number */
1199 memset(num, 0, sizeof(num));
1200 for (i = 0; i < num_var_ranges; i++) {
1201 type = range_state[i].type;
1202 size = range_state[i].size_pfn;
1203 if (type >= MTRR_NUM_TYPES)
1204 continue;
1205 if (!size)
1206 type = MTRR_NUM_TYPES;
1207 num[type]++;
1208 }
1209
1210 /* check if we got UC entries */
1211 if (!num[MTRR_TYPE_UNCACHABLE])
1212 return 0;
1213
1214 /* check if we only had WB and UC */
1215 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
1216 num_var_ranges - num[MTRR_NUM_TYPES])
1217 return 0;
1218
1219 memset(range, 0, sizeof(range));
1220 extra_remove_size = 0;
1221 if (mtrr_tom2) {
1222 extra_remove_base = 1 << (32 - PAGE_SHIFT);
1223 extra_remove_size =
1224 (mtrr_tom2 >> PAGE_SHIFT) - extra_remove_base;
1225 }
1226 nr_range = x86_get_mtrr_mem_range(range, 0, extra_remove_base,
1227 extra_remove_size);
1228 range_sums = sum_ranges(range, nr_range);
1229 printk(KERN_INFO "total RAM coverred: %ldM\n",
1230 range_sums >> (20 - PAGE_SHIFT));
1231
1232 if (mtrr_chunk_size && mtrr_gran_size) {
1233 int num_reg;
1234
1235 debug_print = 1;
1236 /* convert ranges to var ranges state */
1237 num_reg = x86_setup_var_mtrrs(range, nr_range, mtrr_chunk_size,
1238 mtrr_gran_size);
1239
1240 /* we got new setting in range_state, check it */
1241 memset(range_new, 0, sizeof(range_new));
1242 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
1243 extra_remove_base,
1244 extra_remove_size);
1245 range_sums_new = sum_ranges(range_new, nr_range_new);
1246
1247 i = 0;
1248 result[i].chunk_sizek = mtrr_chunk_size >> 10;
1249 result[i].gran_sizek = mtrr_gran_size >> 10;
1250 result[i].num_reg = num_reg;
1251 if (range_sums < range_sums_new) {
1252 result[i].lose_cover_sizek =
1253 (range_sums_new - range_sums) << PSHIFT;
1254 result[i].bad = 1;
1255 } else
1256 result[i].lose_cover_sizek =
1257 (range_sums - range_sums_new) << PSHIFT;
1258
1259 printk(KERN_INFO "%sgran_size: %ldM \tchunk_size: %ldM \t",
1260 result[i].bad?"*BAD*":" ", result[i].gran_sizek >> 10,
1261 result[i].chunk_sizek >> 10);
1262 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ldM \n",
1263 result[i].num_reg, result[i].bad?"-":"",
1264 result[i].lose_cover_sizek >> 10);
1265 if (!result[i].bad) {
1266 set_var_mtrr_all(address_bits);
1267 return 1;
1268 }
1269 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, "
1270 "will find optimal one\n");
1271 debug_print = 0;
1272 memset(result, 0, sizeof(result[0]));
1273 }
1274
1275 i = 0;
1276 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
1277 memset(result, 0, sizeof(result));
1278 for (gran_size = (1ULL<<20); gran_size < (1ULL<<32); gran_size <<= 1) {
1279 for (chunk_size = gran_size; chunk_size < (1ULL<<33);
1280 chunk_size <<= 1) {
1281 int num_reg;
1282
1283 if (debug_print)
1284 printk(KERN_INFO
1285 "\ngran_size: %lldM chunk_size_size: %lldM\n",
1286 gran_size >> 20, chunk_size >> 20);
1287 if (i >= NUM_RESULT)
1288 continue;
1289
1290 /* convert ranges to var ranges state */
1291 num_reg = x86_setup_var_mtrrs(range, nr_range,
1292 chunk_size, gran_size);
1293
1294 /* we got new setting in range_state, check it */
1295 memset(range_new, 0, sizeof(range_new));
1296 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
1297 extra_remove_base, extra_remove_size);
1298 range_sums_new = sum_ranges(range_new, nr_range_new);
1299
1300 result[i].chunk_sizek = chunk_size >> 10;
1301 result[i].gran_sizek = gran_size >> 10;
1302 result[i].num_reg = num_reg;
1303 if (range_sums < range_sums_new) {
1304 result[i].lose_cover_sizek =
1305 (range_sums_new - range_sums) << PSHIFT;
1306 result[i].bad = 1;
1307 } else
1308 result[i].lose_cover_sizek =
1309 (range_sums - range_sums_new) << PSHIFT;
1310
1311 /* double check it */
1312 if (!result[i].bad && !result[i].lose_cover_sizek) {
1313 if (nr_range_new != nr_range ||
1314 memcmp(range, range_new, sizeof(range)))
1315 result[i].bad = 1;
1316 }
1317
1318 if (!result[i].bad && (range_sums - range_sums_new <
1319 min_loss_pfn[num_reg])) {
1320 min_loss_pfn[num_reg] =
1321 range_sums - range_sums_new;
1322 }
1323 i++;
1324 }
1325 }
1326
1327 /* print out all */
1328 for (i = 0; i < NUM_RESULT; i++) {
1329 printk(KERN_INFO "%sgran_size: %ldM \tchunk_size: %ldM \t",
1330 result[i].bad?"*BAD* ":" ", result[i].gran_sizek >> 10,
1331 result[i].chunk_sizek >> 10);
1332 printk(KERN_CONT "num_reg: %d \tlose RAM: %s%ldM\n",
1333 result[i].num_reg, result[i].bad?"-":"",
1334 result[i].lose_cover_sizek >> 10);
1335 }
1336
1337 /* try to find the optimal index */
1338 if (nr_mtrr_spare_reg >= num_var_ranges)
1339 nr_mtrr_spare_reg = num_var_ranges - 1;
1340 num_reg_good = -1;
1341 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
1342 if (!min_loss_pfn[i]) {
1343 num_reg_good = i;
1344 break;
1345 }
1346 }
1347
1348 index_good = -1;
1349 if (num_reg_good != -1) {
1350 for (i = 0; i < NUM_RESULT; i++) {
1351 if (!result[i].bad &&
1352 result[i].num_reg == num_reg_good &&
1353 !result[i].lose_cover_sizek) {
1354 index_good = i;
1355 break;
1356 }
1357 }
1358 }
1359
1360 if (index_good != -1) {
1361 printk(KERN_INFO "Found optimal setting for mtrr clean up\n");
1362 i = index_good;
1363 printk(KERN_INFO "gran_size: %ldM \tchunk_size: %ldM \t",
1364 result[i].gran_sizek >> 10,
1365 result[i].chunk_sizek >> 10);
1366 printk(KERN_CONT "num_reg: %d \tlose RAM: %ldM\n",
1367 result[i].num_reg,
1368 result[i].lose_cover_sizek >> 10);
1369 /* convert ranges to var ranges state */
1370 chunk_size = result[i].chunk_sizek;
1371 chunk_size <<= 10;
1372 gran_size = result[i].gran_sizek;
1373 gran_size <<= 10;
1374 debug_print = 1;
1375 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
1376 set_var_mtrr_all(address_bits);
1377 return 1;
1378 }
1379
1380 printk(KERN_INFO "mtrr_cleanup: can not find optimal value\n");
1381 printk(KERN_INFO "please specify mtrr_gran_size/mtrr_chunk_size\n");
1382
1383 return 0;
1384}
1385#else
1386static int __init mtrr_cleanup(unsigned address_bits)
1387{
1388 return 0;
1389}
1390#endif
1391
1392static int __initdata changed_by_mtrr_cleanup;
1393
612static int disable_mtrr_trim; 1394static int disable_mtrr_trim;
613 1395
614static int __init disable_mtrr_trim_setup(char *str) 1396static int __init disable_mtrr_trim_setup(char *str)
@@ -648,6 +1430,19 @@ int __init amd_special_default_mtrr(void)
648 return 0; 1430 return 0;
649} 1431}
650 1432
1433static u64 __init real_trim_memory(unsigned long start_pfn,
1434 unsigned long limit_pfn)
1435{
1436 u64 trim_start, trim_size;
1437 trim_start = start_pfn;
1438 trim_start <<= PAGE_SHIFT;
1439 trim_size = limit_pfn;
1440 trim_size <<= PAGE_SHIFT;
1441 trim_size -= trim_start;
1442
1443 return e820_update_range(trim_start, trim_size, E820_RAM,
1444 E820_RESERVED);
1445}
651/** 1446/**
652 * mtrr_trim_uncached_memory - trim RAM not covered by MTRRs 1447 * mtrr_trim_uncached_memory - trim RAM not covered by MTRRs
653 * @end_pfn: ending page frame number 1448 * @end_pfn: ending page frame number
@@ -663,8 +1458,11 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
663{ 1458{
664 unsigned long i, base, size, highest_pfn = 0, def, dummy; 1459 unsigned long i, base, size, highest_pfn = 0, def, dummy;
665 mtrr_type type; 1460 mtrr_type type;
666 u64 trim_start, trim_size; 1461 int nr_range;
1462 u64 total_trim_size;
667 1463
1464 /* extra one for all 0 */
1465 int num[MTRR_NUM_TYPES + 1];
668 /* 1466 /*
669 * Make sure we only trim uncachable memory on machines that 1467 * Make sure we only trim uncachable memory on machines that
670 * support the Intel MTRR architecture: 1468 * support the Intel MTRR architecture:
@@ -676,14 +1474,22 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
676 if (def != MTRR_TYPE_UNCACHABLE) 1474 if (def != MTRR_TYPE_UNCACHABLE)
677 return 0; 1475 return 0;
678 1476
679 if (amd_special_default_mtrr()) 1477 /* get it and store it aside */
680 return 0; 1478 memset(range_state, 0, sizeof(range_state));
1479 for (i = 0; i < num_var_ranges; i++) {
1480 mtrr_if->get(i, &base, &size, &type);
1481 range_state[i].base_pfn = base;
1482 range_state[i].size_pfn = size;
1483 range_state[i].type = type;
1484 }
681 1485
682 /* Find highest cached pfn */ 1486 /* Find highest cached pfn */
683 for (i = 0; i < num_var_ranges; i++) { 1487 for (i = 0; i < num_var_ranges; i++) {
684 mtrr_if->get(i, &base, &size, &type); 1488 type = range_state[i].type;
685 if (type != MTRR_TYPE_WRBACK) 1489 if (type != MTRR_TYPE_WRBACK)
686 continue; 1490 continue;
1491 base = range_state[i].base_pfn;
1492 size = range_state[i].size_pfn;
687 if (highest_pfn < base + size) 1493 if (highest_pfn < base + size)
688 highest_pfn = base + size; 1494 highest_pfn = base + size;
689 } 1495 }
@@ -698,22 +1504,65 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
698 return 0; 1504 return 0;
699 } 1505 }
700 1506
701 if (highest_pfn < end_pfn) { 1507 /* check entries number */
1508 memset(num, 0, sizeof(num));
1509 for (i = 0; i < num_var_ranges; i++) {
1510 type = range_state[i].type;
1511 if (type >= MTRR_NUM_TYPES)
1512 continue;
1513 size = range_state[i].size_pfn;
1514 if (!size)
1515 type = MTRR_NUM_TYPES;
1516 num[type]++;
1517 }
1518
1519 /* no entry for WB? */
1520 if (!num[MTRR_TYPE_WRBACK])
1521 return 0;
1522
1523 /* check if we only had WB and UC */
1524 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
1525 num_var_ranges - num[MTRR_NUM_TYPES])
1526 return 0;
1527
1528 memset(range, 0, sizeof(range));
1529 nr_range = 0;
1530 if (mtrr_tom2) {
1531 range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT));
1532 range[nr_range].end = (mtrr_tom2 >> PAGE_SHIFT) - 1;
1533 if (highest_pfn < range[nr_range].end + 1)
1534 highest_pfn = range[nr_range].end + 1;
1535 nr_range++;
1536 }
1537 nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0);
1538
1539 total_trim_size = 0;
1540 /* check the head */
1541 if (range[0].start)
1542 total_trim_size += real_trim_memory(0, range[0].start);
1543 /* check the holes */
1544 for (i = 0; i < nr_range - 1; i++) {
1545 if (range[i].end + 1 < range[i+1].start)
1546 total_trim_size += real_trim_memory(range[i].end + 1,
1547 range[i+1].start);
1548 }
1549 /* check the top */
1550 i = nr_range - 1;
1551 if (range[i].end + 1 < end_pfn)
1552 total_trim_size += real_trim_memory(range[i].end + 1,
1553 end_pfn);
1554
1555 if (total_trim_size) {
702 printk(KERN_WARNING "WARNING: BIOS bug: CPU MTRRs don't cover" 1556 printk(KERN_WARNING "WARNING: BIOS bug: CPU MTRRs don't cover"
703 " all of memory, losing %luMB of RAM.\n", 1557 " all of memory, losing %lluMB of RAM.\n",
704 (end_pfn - highest_pfn) >> (20 - PAGE_SHIFT)); 1558 total_trim_size >> 20);
705 1559
706 WARN_ON(1); 1560 if (!changed_by_mtrr_cleanup)
1561 WARN_ON(1);
707 1562
708 printk(KERN_INFO "update e820 for mtrr\n"); 1563 printk(KERN_INFO "update e820 for mtrr\n");
709 trim_start = highest_pfn;
710 trim_start <<= PAGE_SHIFT;
711 trim_size = end_pfn;
712 trim_size <<= PAGE_SHIFT;
713 trim_size -= trim_start;
714 update_memory_range(trim_start, trim_size, E820_RAM,
715 E820_RESERVED);
716 update_e820(); 1564 update_e820();
1565
717 return 1; 1566 return 1;
718 } 1567 }
719 1568
@@ -729,18 +1578,21 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
729 */ 1578 */
730void __init mtrr_bp_init(void) 1579void __init mtrr_bp_init(void)
731{ 1580{
1581 u32 phys_addr;
732 init_ifs(); 1582 init_ifs();
733 1583
1584 phys_addr = 32;
1585
734 if (cpu_has_mtrr) { 1586 if (cpu_has_mtrr) {
735 mtrr_if = &generic_mtrr_ops; 1587 mtrr_if = &generic_mtrr_ops;
736 size_or_mask = 0xff000000; /* 36 bits */ 1588 size_or_mask = 0xff000000; /* 36 bits */
737 size_and_mask = 0x00f00000; 1589 size_and_mask = 0x00f00000;
1590 phys_addr = 36;
738 1591
739 /* This is an AMD specific MSR, but we assume(hope?) that 1592 /* This is an AMD specific MSR, but we assume(hope?) that
740 Intel will implement it to when they extend the address 1593 Intel will implement it to when they extend the address
741 bus of the Xeon. */ 1594 bus of the Xeon. */
742 if (cpuid_eax(0x80000000) >= 0x80000008) { 1595 if (cpuid_eax(0x80000000) >= 0x80000008) {
743 u32 phys_addr;
744 phys_addr = cpuid_eax(0x80000008) & 0xff; 1596 phys_addr = cpuid_eax(0x80000008) & 0xff;
745 /* CPUID workaround for Intel 0F33/0F34 CPU */ 1597 /* CPUID workaround for Intel 0F33/0F34 CPU */
746 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && 1598 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
@@ -758,6 +1610,7 @@ void __init mtrr_bp_init(void)
758 don't support PAE */ 1610 don't support PAE */
759 size_or_mask = 0xfff00000; /* 32 bits */ 1611 size_or_mask = 0xfff00000; /* 32 bits */
760 size_and_mask = 0; 1612 size_and_mask = 0;
1613 phys_addr = 32;
761 } 1614 }
762 } else { 1615 } else {
763 switch (boot_cpu_data.x86_vendor) { 1616 switch (boot_cpu_data.x86_vendor) {
@@ -791,8 +1644,15 @@ void __init mtrr_bp_init(void)
791 if (mtrr_if) { 1644 if (mtrr_if) {
792 set_num_var_ranges(); 1645 set_num_var_ranges();
793 init_table(); 1646 init_table();
794 if (use_intel()) 1647 if (use_intel()) {
795 get_mtrr_state(); 1648 get_mtrr_state();
1649
1650 if (mtrr_cleanup(phys_addr)) {
1651 changed_by_mtrr_cleanup = 1;
1652 mtrr_if->set_all();
1653 }
1654
1655 }
796 } 1656 }
797} 1657}
798 1658
@@ -822,16 +1682,17 @@ void mtrr_ap_init(void)
822 */ 1682 */
823void mtrr_save_state(void) 1683void mtrr_save_state(void)
824{ 1684{
825 smp_call_function_single(0, mtrr_save_fixed_ranges, NULL, 1, 1); 1685 smp_call_function_single(0, mtrr_save_fixed_ranges, NULL, 1);
826} 1686}
827 1687
828static int __init mtrr_init_finialize(void) 1688static int __init mtrr_init_finialize(void)
829{ 1689{
830 if (!mtrr_if) 1690 if (!mtrr_if)
831 return 0; 1691 return 0;
832 if (use_intel()) 1692 if (use_intel()) {
833 mtrr_state_warn(); 1693 if (!changed_by_mtrr_cleanup)
834 else { 1694 mtrr_state_warn();
1695 } else {
835 /* The CPUs haven't MTRR and seem to not support SMP. They have 1696 /* The CPUs haven't MTRR and seem to not support SMP. They have
836 * specific drivers, we use a tricky method to support 1697 * specific drivers, we use a tricky method to support
837 * suspend/resume for them. 1698 * suspend/resume for them.
diff --git a/arch/x86/kernel/cpu/mtrr/mtrr.h b/arch/x86/kernel/cpu/mtrr/mtrr.h
index 2cc77eb6fea3..2dc4ec656b23 100644
--- a/arch/x86/kernel/cpu/mtrr/mtrr.h
+++ b/arch/x86/kernel/cpu/mtrr/mtrr.h
@@ -81,6 +81,8 @@ void set_mtrr_done(struct set_mtrr_context *ctxt);
81void set_mtrr_cache_disable(struct set_mtrr_context *ctxt); 81void set_mtrr_cache_disable(struct set_mtrr_context *ctxt);
82void set_mtrr_prepare_save(struct set_mtrr_context *ctxt); 82void set_mtrr_prepare_save(struct set_mtrr_context *ctxt);
83 83
84void fill_mtrr_var_range(unsigned int index,
85 u32 base_lo, u32 base_hi, u32 mask_lo, u32 mask_hi);
84void get_mtrr_state(void); 86void get_mtrr_state(void);
85 87
86extern void set_mtrr_ops(struct mtrr_ops * ops); 88extern void set_mtrr_ops(struct mtrr_ops * ops);
@@ -92,6 +94,7 @@ extern struct mtrr_ops * mtrr_if;
92#define use_intel() (mtrr_if && mtrr_if->use_intel_if == 1) 94#define use_intel() (mtrr_if && mtrr_if->use_intel_if == 1)
93 95
94extern unsigned int num_var_ranges; 96extern unsigned int num_var_ranges;
97extern u64 mtrr_tom2;
95 98
96void mtrr_state_warn(void); 99void mtrr_state_warn(void);
97const char *mtrr_attrib_to_str(int x); 100const char *mtrr_attrib_to_str(int x);
diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c
index f9ae93adffe5..de7439f82b92 100644
--- a/arch/x86/kernel/cpu/perfctr-watchdog.c
+++ b/arch/x86/kernel/cpu/perfctr-watchdog.c
@@ -1,11 +1,15 @@
1/* local apic based NMI watchdog for various CPUs. 1/*
2 This file also handles reservation of performance counters for coordination 2 * local apic based NMI watchdog for various CPUs.
3 with other users (like oprofile). 3 *
4 4 * This file also handles reservation of performance counters for coordination
5 Note that these events normally don't tick when the CPU idles. This means 5 * with other users (like oprofile).
6 the frequency varies with CPU load. 6 *
7 7 * Note that these events normally don't tick when the CPU idles. This means
8 Original code for K7/P6 written by Keith Owens */ 8 * the frequency varies with CPU load.
9 *
10 * Original code for K7/P6 written by Keith Owens
11 *
12 */
9 13
10#include <linux/percpu.h> 14#include <linux/percpu.h>
11#include <linux/module.h> 15#include <linux/module.h>
@@ -36,12 +40,16 @@ struct wd_ops {
36 40
37static const struct wd_ops *wd_ops; 41static const struct wd_ops *wd_ops;
38 42
39/* this number is calculated from Intel's MSR_P4_CRU_ESCR5 register and it's 43/*
40 * offset from MSR_P4_BSU_ESCR0. It will be the max for all platforms (for now) 44 * this number is calculated from Intel's MSR_P4_CRU_ESCR5 register and it's
45 * offset from MSR_P4_BSU_ESCR0.
46 *
47 * It will be the max for all platforms (for now)
41 */ 48 */
42#define NMI_MAX_COUNTER_BITS 66 49#define NMI_MAX_COUNTER_BITS 66
43 50
44/* perfctr_nmi_owner tracks the ownership of the perfctr registers: 51/*
52 * perfctr_nmi_owner tracks the ownership of the perfctr registers:
45 * evtsel_nmi_owner tracks the ownership of the event selection 53 * evtsel_nmi_owner tracks the ownership of the event selection
46 * - different performance counters/ event selection may be reserved for 54 * - different performance counters/ event selection may be reserved for
47 * different subsystems this reservation system just tries to coordinate 55 * different subsystems this reservation system just tries to coordinate
@@ -73,8 +81,10 @@ static inline unsigned int nmi_perfctr_msr_to_bit(unsigned int msr)
73 return 0; 81 return 0;
74} 82}
75 83
76/* converts an msr to an appropriate reservation bit */ 84/*
77/* returns the bit offset of the event selection register */ 85 * converts an msr to an appropriate reservation bit
86 * returns the bit offset of the event selection register
87 */
78static inline unsigned int nmi_evntsel_msr_to_bit(unsigned int msr) 88static inline unsigned int nmi_evntsel_msr_to_bit(unsigned int msr)
79{ 89{
80 /* returns the bit offset of the event selection register */ 90 /* returns the bit offset of the event selection register */
@@ -114,6 +124,7 @@ int avail_to_resrv_perfctr_nmi(unsigned int msr)
114 124
115 return (!test_bit(counter, perfctr_nmi_owner)); 125 return (!test_bit(counter, perfctr_nmi_owner));
116} 126}
127EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi_bit);
117 128
118int reserve_perfctr_nmi(unsigned int msr) 129int reserve_perfctr_nmi(unsigned int msr)
119{ 130{
@@ -128,6 +139,7 @@ int reserve_perfctr_nmi(unsigned int msr)
128 return 1; 139 return 1;
129 return 0; 140 return 0;
130} 141}
142EXPORT_SYMBOL(reserve_perfctr_nmi);
131 143
132void release_perfctr_nmi(unsigned int msr) 144void release_perfctr_nmi(unsigned int msr)
133{ 145{
@@ -140,6 +152,7 @@ void release_perfctr_nmi(unsigned int msr)
140 152
141 clear_bit(counter, perfctr_nmi_owner); 153 clear_bit(counter, perfctr_nmi_owner);
142} 154}
155EXPORT_SYMBOL(release_perfctr_nmi);
143 156
144int reserve_evntsel_nmi(unsigned int msr) 157int reserve_evntsel_nmi(unsigned int msr)
145{ 158{
@@ -154,6 +167,7 @@ int reserve_evntsel_nmi(unsigned int msr)
154 return 1; 167 return 1;
155 return 0; 168 return 0;
156} 169}
170EXPORT_SYMBOL(reserve_evntsel_nmi);
157 171
158void release_evntsel_nmi(unsigned int msr) 172void release_evntsel_nmi(unsigned int msr)
159{ 173{
@@ -166,11 +180,6 @@ void release_evntsel_nmi(unsigned int msr)
166 180
167 clear_bit(counter, evntsel_nmi_owner); 181 clear_bit(counter, evntsel_nmi_owner);
168} 182}
169
170EXPORT_SYMBOL(avail_to_resrv_perfctr_nmi_bit);
171EXPORT_SYMBOL(reserve_perfctr_nmi);
172EXPORT_SYMBOL(release_perfctr_nmi);
173EXPORT_SYMBOL(reserve_evntsel_nmi);
174EXPORT_SYMBOL(release_evntsel_nmi); 183EXPORT_SYMBOL(release_evntsel_nmi);
175 184
176void disable_lapic_nmi_watchdog(void) 185void disable_lapic_nmi_watchdog(void)
@@ -180,8 +189,10 @@ void disable_lapic_nmi_watchdog(void)
180 if (atomic_read(&nmi_active) <= 0) 189 if (atomic_read(&nmi_active) <= 0)
181 return; 190 return;
182 191
183 on_each_cpu(stop_apic_nmi_watchdog, NULL, 0, 1); 192 on_each_cpu(stop_apic_nmi_watchdog, NULL, 1);
184 wd_ops->unreserve(); 193
194 if (wd_ops)
195 wd_ops->unreserve();
185 196
186 BUG_ON(atomic_read(&nmi_active) != 0); 197 BUG_ON(atomic_read(&nmi_active) != 0);
187} 198}
@@ -202,7 +213,7 @@ void enable_lapic_nmi_watchdog(void)
202 return; 213 return;
203 } 214 }
204 215
205 on_each_cpu(setup_apic_nmi_watchdog, NULL, 0, 1); 216 on_each_cpu(setup_apic_nmi_watchdog, NULL, 1);
206 touch_nmi_watchdog(); 217 touch_nmi_watchdog();
207} 218}
208 219
@@ -232,31 +243,32 @@ static unsigned int adjust_for_32bit_ctr(unsigned int hz)
232 return retval; 243 return retval;
233} 244}
234 245
235static void 246static void write_watchdog_counter(unsigned int perfctr_msr,
236write_watchdog_counter(unsigned int perfctr_msr, const char *descr, unsigned nmi_hz) 247 const char *descr, unsigned nmi_hz)
237{ 248{
238 u64 count = (u64)cpu_khz * 1000; 249 u64 count = (u64)cpu_khz * 1000;
239 250
240 do_div(count, nmi_hz); 251 do_div(count, nmi_hz);
241 if(descr) 252 if(descr)
242 Dprintk("setting %s to -0x%08Lx\n", descr, count); 253 pr_debug("setting %s to -0x%08Lx\n", descr, count);
243 wrmsrl(perfctr_msr, 0 - count); 254 wrmsrl(perfctr_msr, 0 - count);
244} 255}
245 256
246static void write_watchdog_counter32(unsigned int perfctr_msr, 257static void write_watchdog_counter32(unsigned int perfctr_msr,
247 const char *descr, unsigned nmi_hz) 258 const char *descr, unsigned nmi_hz)
248{ 259{
249 u64 count = (u64)cpu_khz * 1000; 260 u64 count = (u64)cpu_khz * 1000;
250 261
251 do_div(count, nmi_hz); 262 do_div(count, nmi_hz);
252 if(descr) 263 if(descr)
253 Dprintk("setting %s to -0x%08Lx\n", descr, count); 264 pr_debug("setting %s to -0x%08Lx\n", descr, count);
254 wrmsr(perfctr_msr, (u32)(-count), 0); 265 wrmsr(perfctr_msr, (u32)(-count), 0);
255} 266}
256 267
257/* AMD K7/K8/Family10h/Family11h support. AMD keeps this interface 268/*
258 nicely stable so there is not much variety */ 269 * AMD K7/K8/Family10h/Family11h support.
259 270 * AMD keeps this interface nicely stable so there is not much variety
271 */
260#define K7_EVNTSEL_ENABLE (1 << 22) 272#define K7_EVNTSEL_ENABLE (1 << 22)
261#define K7_EVNTSEL_INT (1 << 20) 273#define K7_EVNTSEL_INT (1 << 20)
262#define K7_EVNTSEL_OS (1 << 17) 274#define K7_EVNTSEL_OS (1 << 17)
@@ -289,7 +301,7 @@ static int setup_k7_watchdog(unsigned nmi_hz)
289 301
290 wd->perfctr_msr = perfctr_msr; 302 wd->perfctr_msr = perfctr_msr;
291 wd->evntsel_msr = evntsel_msr; 303 wd->evntsel_msr = evntsel_msr;
292 wd->cccr_msr = 0; //unused 304 wd->cccr_msr = 0; /* unused */
293 return 1; 305 return 1;
294} 306}
295 307
@@ -325,18 +337,19 @@ static void single_msr_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz)
325} 337}
326 338
327static const struct wd_ops k7_wd_ops = { 339static const struct wd_ops k7_wd_ops = {
328 .reserve = single_msr_reserve, 340 .reserve = single_msr_reserve,
329 .unreserve = single_msr_unreserve, 341 .unreserve = single_msr_unreserve,
330 .setup = setup_k7_watchdog, 342 .setup = setup_k7_watchdog,
331 .rearm = single_msr_rearm, 343 .rearm = single_msr_rearm,
332 .stop = single_msr_stop_watchdog, 344 .stop = single_msr_stop_watchdog,
333 .perfctr = MSR_K7_PERFCTR0, 345 .perfctr = MSR_K7_PERFCTR0,
334 .evntsel = MSR_K7_EVNTSEL0, 346 .evntsel = MSR_K7_EVNTSEL0,
335 .checkbit = 1ULL<<47, 347 .checkbit = 1ULL << 47,
336}; 348};
337 349
338/* Intel Model 6 (PPro+,P2,P3,P-M,Core1) */ 350/*
339 351 * Intel Model 6 (PPro+,P2,P3,P-M,Core1)
352 */
340#define P6_EVNTSEL0_ENABLE (1 << 22) 353#define P6_EVNTSEL0_ENABLE (1 << 22)
341#define P6_EVNTSEL_INT (1 << 20) 354#define P6_EVNTSEL_INT (1 << 20)
342#define P6_EVNTSEL_OS (1 << 17) 355#define P6_EVNTSEL_OS (1 << 17)
@@ -372,52 +385,58 @@ static int setup_p6_watchdog(unsigned nmi_hz)
372 385
373 wd->perfctr_msr = perfctr_msr; 386 wd->perfctr_msr = perfctr_msr;
374 wd->evntsel_msr = evntsel_msr; 387 wd->evntsel_msr = evntsel_msr;
375 wd->cccr_msr = 0; //unused 388 wd->cccr_msr = 0; /* unused */
376 return 1; 389 return 1;
377} 390}
378 391
379static void p6_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz) 392static void p6_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz)
380{ 393{
381 /* P6 based Pentium M need to re-unmask 394 /*
395 * P6 based Pentium M need to re-unmask
382 * the apic vector but it doesn't hurt 396 * the apic vector but it doesn't hurt
383 * other P6 variant. 397 * other P6 variant.
384 * ArchPerfom/Core Duo also needs this */ 398 * ArchPerfom/Core Duo also needs this
399 */
385 apic_write(APIC_LVTPC, APIC_DM_NMI); 400 apic_write(APIC_LVTPC, APIC_DM_NMI);
401
386 /* P6/ARCH_PERFMON has 32 bit counter write */ 402 /* P6/ARCH_PERFMON has 32 bit counter write */
387 write_watchdog_counter32(wd->perfctr_msr, NULL,nmi_hz); 403 write_watchdog_counter32(wd->perfctr_msr, NULL,nmi_hz);
388} 404}
389 405
390static const struct wd_ops p6_wd_ops = { 406static const struct wd_ops p6_wd_ops = {
391 .reserve = single_msr_reserve, 407 .reserve = single_msr_reserve,
392 .unreserve = single_msr_unreserve, 408 .unreserve = single_msr_unreserve,
393 .setup = setup_p6_watchdog, 409 .setup = setup_p6_watchdog,
394 .rearm = p6_rearm, 410 .rearm = p6_rearm,
395 .stop = single_msr_stop_watchdog, 411 .stop = single_msr_stop_watchdog,
396 .perfctr = MSR_P6_PERFCTR0, 412 .perfctr = MSR_P6_PERFCTR0,
397 .evntsel = MSR_P6_EVNTSEL0, 413 .evntsel = MSR_P6_EVNTSEL0,
398 .checkbit = 1ULL<<39, 414 .checkbit = 1ULL << 39,
399}; 415};
400 416
401/* Intel P4 performance counters. By far the most complicated of all. */ 417/*
402 418 * Intel P4 performance counters.
403#define MSR_P4_MISC_ENABLE_PERF_AVAIL (1<<7) 419 * By far the most complicated of all.
404#define P4_ESCR_EVENT_SELECT(N) ((N)<<25) 420 */
405#define P4_ESCR_OS (1<<3) 421#define MSR_P4_MISC_ENABLE_PERF_AVAIL (1 << 7)
406#define P4_ESCR_USR (1<<2) 422#define P4_ESCR_EVENT_SELECT(N) ((N) << 25)
407#define P4_CCCR_OVF_PMI0 (1<<26) 423#define P4_ESCR_OS (1 << 3)
408#define P4_CCCR_OVF_PMI1 (1<<27) 424#define P4_ESCR_USR (1 << 2)
409#define P4_CCCR_THRESHOLD(N) ((N)<<20) 425#define P4_CCCR_OVF_PMI0 (1 << 26)
410#define P4_CCCR_COMPLEMENT (1<<19) 426#define P4_CCCR_OVF_PMI1 (1 << 27)
411#define P4_CCCR_COMPARE (1<<18) 427#define P4_CCCR_THRESHOLD(N) ((N) << 20)
412#define P4_CCCR_REQUIRED (3<<16) 428#define P4_CCCR_COMPLEMENT (1 << 19)
413#define P4_CCCR_ESCR_SELECT(N) ((N)<<13) 429#define P4_CCCR_COMPARE (1 << 18)
414#define P4_CCCR_ENABLE (1<<12) 430#define P4_CCCR_REQUIRED (3 << 16)
415#define P4_CCCR_OVF (1<<31) 431#define P4_CCCR_ESCR_SELECT(N) ((N) << 13)
416 432#define P4_CCCR_ENABLE (1 << 12)
417/* Set up IQ_COUNTER0 to behave like a clock, by having IQ_CCCR0 filter 433#define P4_CCCR_OVF (1 << 31)
418 CRU_ESCR0 (with any non-null event selector) through a complemented
419 max threshold. [IA32-Vol3, Section 14.9.9] */
420 434
435/*
436 * 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
438 * max threshold. [IA32-Vol3, Section 14.9.9]
439 */
421static int setup_p4_watchdog(unsigned nmi_hz) 440static int setup_p4_watchdog(unsigned nmi_hz)
422{ 441{
423 unsigned int perfctr_msr, evntsel_msr, cccr_msr; 442 unsigned int perfctr_msr, evntsel_msr, cccr_msr;
@@ -442,7 +461,8 @@ static int setup_p4_watchdog(unsigned nmi_hz)
442#endif 461#endif
443 ht_num = 0; 462 ht_num = 0;
444 463
445 /* performance counters are shared resources 464 /*
465 * performance counters are shared resources
446 * assign each hyperthread its own set 466 * assign each hyperthread its own set
447 * (re-use the ESCR0 register, seems safe 467 * (re-use the ESCR0 register, seems safe
448 * and keeps the cccr_val the same) 468 * and keeps the cccr_val the same)
@@ -540,20 +560,21 @@ static void p4_rearm(struct nmi_watchdog_ctlblk *wd, unsigned nmi_hz)
540} 560}
541 561
542static const struct wd_ops p4_wd_ops = { 562static const struct wd_ops p4_wd_ops = {
543 .reserve = p4_reserve, 563 .reserve = p4_reserve,
544 .unreserve = p4_unreserve, 564 .unreserve = p4_unreserve,
545 .setup = setup_p4_watchdog, 565 .setup = setup_p4_watchdog,
546 .rearm = p4_rearm, 566 .rearm = p4_rearm,
547 .stop = stop_p4_watchdog, 567 .stop = stop_p4_watchdog,
548 /* RED-PEN this is wrong for the other sibling */ 568 /* RED-PEN this is wrong for the other sibling */
549 .perfctr = MSR_P4_BPU_PERFCTR0, 569 .perfctr = MSR_P4_BPU_PERFCTR0,
550 .evntsel = MSR_P4_BSU_ESCR0, 570 .evntsel = MSR_P4_BSU_ESCR0,
551 .checkbit = 1ULL<<39, 571 .checkbit = 1ULL << 39,
552}; 572};
553 573
554/* Watchdog using the Intel architected PerfMon. Used for Core2 and hopefully 574/*
555 all future Intel CPUs. */ 575 * Watchdog using the Intel architected PerfMon.
556 576 * Used for Core2 and hopefully all future Intel CPUs.
577 */
557#define ARCH_PERFMON_NMI_EVENT_SEL ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 578#define ARCH_PERFMON_NMI_EVENT_SEL ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL
558#define ARCH_PERFMON_NMI_EVENT_UMASK ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK 579#define ARCH_PERFMON_NMI_EVENT_UMASK ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK
559 580
@@ -599,19 +620,19 @@ static int setup_intel_arch_watchdog(unsigned nmi_hz)
599 620
600 wd->perfctr_msr = perfctr_msr; 621 wd->perfctr_msr = perfctr_msr;
601 wd->evntsel_msr = evntsel_msr; 622 wd->evntsel_msr = evntsel_msr;
602 wd->cccr_msr = 0; //unused 623 wd->cccr_msr = 0; /* unused */
603 intel_arch_wd_ops.checkbit = 1ULL << (eax.split.bit_width - 1); 624 intel_arch_wd_ops.checkbit = 1ULL << (eax.split.bit_width - 1);
604 return 1; 625 return 1;
605} 626}
606 627
607static struct wd_ops intel_arch_wd_ops __read_mostly = { 628static struct wd_ops intel_arch_wd_ops __read_mostly = {
608 .reserve = single_msr_reserve, 629 .reserve = single_msr_reserve,
609 .unreserve = single_msr_unreserve, 630 .unreserve = single_msr_unreserve,
610 .setup = setup_intel_arch_watchdog, 631 .setup = setup_intel_arch_watchdog,
611 .rearm = p6_rearm, 632 .rearm = p6_rearm,
612 .stop = single_msr_stop_watchdog, 633 .stop = single_msr_stop_watchdog,
613 .perfctr = MSR_ARCH_PERFMON_PERFCTR1, 634 .perfctr = MSR_ARCH_PERFMON_PERFCTR1,
614 .evntsel = MSR_ARCH_PERFMON_EVENTSEL1, 635 .evntsel = MSR_ARCH_PERFMON_EVENTSEL1,
615}; 636};
616 637
617static void probe_nmi_watchdog(void) 638static void probe_nmi_watchdog(void)
@@ -624,8 +645,10 @@ static void probe_nmi_watchdog(void)
624 wd_ops = &k7_wd_ops; 645 wd_ops = &k7_wd_ops;
625 break; 646 break;
626 case X86_VENDOR_INTEL: 647 case X86_VENDOR_INTEL:
627 /* Work around Core Duo (Yonah) errata AE49 where perfctr1 648 /*
628 doesn't have a working enable bit. */ 649 * Work around Core Duo (Yonah) errata AE49 where perfctr1
650 * doesn't have a working enable bit.
651 */
629 if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 14) { 652 if (boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model == 14) {
630 intel_arch_wd_ops.perfctr = MSR_ARCH_PERFMON_PERFCTR0; 653 intel_arch_wd_ops.perfctr = MSR_ARCH_PERFMON_PERFCTR0;
631 intel_arch_wd_ops.evntsel = MSR_ARCH_PERFMON_EVENTSEL0; 654 intel_arch_wd_ops.evntsel = MSR_ARCH_PERFMON_EVENTSEL0;
@@ -636,7 +659,7 @@ static void probe_nmi_watchdog(void)
636 } 659 }
637 switch (boot_cpu_data.x86) { 660 switch (boot_cpu_data.x86) {
638 case 6: 661 case 6:
639 if (boot_cpu_data.x86_model > 0xd) 662 if (boot_cpu_data.x86_model > 13)
640 return; 663 return;
641 664
642 wd_ops = &p6_wd_ops; 665 wd_ops = &p6_wd_ops;
@@ -697,10 +720,11 @@ int lapic_wd_event(unsigned nmi_hz)
697{ 720{
698 struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk); 721 struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
699 u64 ctr; 722 u64 ctr;
723
700 rdmsrl(wd->perfctr_msr, ctr); 724 rdmsrl(wd->perfctr_msr, ctr);
701 if (ctr & wd_ops->checkbit) { /* perfctr still running? */ 725 if (ctr & wd_ops->checkbit) /* perfctr still running? */
702 return 0; 726 return 0;
703 } 727
704 wd_ops->rearm(wd, nmi_hz); 728 wd_ops->rearm(wd, nmi_hz);
705 return 1; 729 return 1;
706} 730}
diff --git a/arch/x86/kernel/cpu/proc.c b/arch/x86/kernel/cpu/proc.c
index 0d0d9057e7c0..a26c480b9491 100644
--- a/arch/x86/kernel/cpu/proc.c
+++ b/arch/x86/kernel/cpu/proc.c
@@ -160,7 +160,7 @@ static void *c_start(struct seq_file *m, loff_t *pos)
160{ 160{
161 if (*pos == 0) /* just in case, cpu 0 is not the first */ 161 if (*pos == 0) /* just in case, cpu 0 is not the first */
162 *pos = first_cpu(cpu_online_map); 162 *pos = first_cpu(cpu_online_map);
163 if ((*pos) < NR_CPUS && cpu_online(*pos)) 163 if ((*pos) < nr_cpu_ids && cpu_online(*pos))
164 return &cpu_data(*pos); 164 return &cpu_data(*pos);
165 return NULL; 165 return NULL;
166} 166}
diff --git a/arch/x86/kernel/cpuid.c b/arch/x86/kernel/cpuid.c
index daff52a62248..14b11b3be31c 100644
--- a/arch/x86/kernel/cpuid.c
+++ b/arch/x86/kernel/cpuid.c
@@ -33,6 +33,7 @@
33#include <linux/init.h> 33#include <linux/init.h>
34#include <linux/poll.h> 34#include <linux/poll.h>
35#include <linux/smp.h> 35#include <linux/smp.h>
36#include <linux/smp_lock.h>
36#include <linux/major.h> 37#include <linux/major.h>
37#include <linux/fs.h> 38#include <linux/fs.h>
38#include <linux/smp_lock.h> 39#include <linux/smp_lock.h>
@@ -95,7 +96,7 @@ static ssize_t cpuid_read(struct file *file, char __user *buf,
95 for (; count; count -= 16) { 96 for (; count; count -= 16) {
96 cmd.eax = pos; 97 cmd.eax = pos;
97 cmd.ecx = pos >> 32; 98 cmd.ecx = pos >> 32;
98 smp_call_function_single(cpu, cpuid_smp_cpuid, &cmd, 1, 1); 99 smp_call_function_single(cpu, cpuid_smp_cpuid, &cmd, 1);
99 if (copy_to_user(tmp, &cmd, 16)) 100 if (copy_to_user(tmp, &cmd, 16))
100 return -EFAULT; 101 return -EFAULT;
101 tmp += 16; 102 tmp += 16;
@@ -107,15 +108,23 @@ static ssize_t cpuid_read(struct file *file, char __user *buf,
107 108
108static int cpuid_open(struct inode *inode, struct file *file) 109static int cpuid_open(struct inode *inode, struct file *file)
109{ 110{
110 unsigned int cpu = iminor(file->f_path.dentry->d_inode); 111 unsigned int cpu;
111 struct cpuinfo_x86 *c = &cpu_data(cpu); 112 struct cpuinfo_x86 *c;
112 113 int ret = 0;
113 if (cpu >= NR_CPUS || !cpu_online(cpu)) 114
114 return -ENXIO; /* No such CPU */ 115 lock_kernel();
116
117 cpu = iminor(file->f_path.dentry->d_inode);
118 if (cpu >= NR_CPUS || !cpu_online(cpu)) {
119 ret = -ENXIO; /* No such CPU */
120 goto out;
121 }
122 c = &cpu_data(cpu);
115 if (c->cpuid_level < 0) 123 if (c->cpuid_level < 0)
116 return -EIO; /* CPUID not supported */ 124 ret = -EIO; /* CPUID not supported */
117 125out:
118 return 0; 126 unlock_kernel();
127 return ret;
119} 128}
120 129
121/* 130/*
@@ -132,8 +141,8 @@ static __cpuinit int cpuid_device_create(int cpu)
132{ 141{
133 struct device *dev; 142 struct device *dev;
134 143
135 dev = device_create(cpuid_class, NULL, MKDEV(CPUID_MAJOR, cpu), 144 dev = device_create_drvdata(cpuid_class, NULL, MKDEV(CPUID_MAJOR, cpu),
136 "cpu%d", cpu); 145 NULL, "cpu%d", cpu);
137 return IS_ERR(dev) ? PTR_ERR(dev) : 0; 146 return IS_ERR(dev) ? PTR_ERR(dev) : 0;
138} 147}
139 148
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
new file mode 100644
index 000000000000..9af89078f7bb
--- /dev/null
+++ b/arch/x86/kernel/e820.c
@@ -0,0 +1,1365 @@
1/*
2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
4 *
5 * Getting sanitize_e820_map() in sync with i386 version by applying change:
6 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
7 * Alex Achenbach <xela@slit.de>, December 2002.
8 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
9 *
10 */
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/init.h>
14#include <linux/bootmem.h>
15#include <linux/ioport.h>
16#include <linux/string.h>
17#include <linux/kexec.h>
18#include <linux/module.h>
19#include <linux/mm.h>
20#include <linux/pfn.h>
21#include <linux/suspend.h>
22#include <linux/firmware-map.h>
23
24#include <asm/pgtable.h>
25#include <asm/page.h>
26#include <asm/e820.h>
27#include <asm/proto.h>
28#include <asm/setup.h>
29#include <asm/trampoline.h>
30
31/*
32 * The e820 map is the map that gets modified e.g. with command line parameters
33 * and that is also registered with modifications in the kernel resource tree
34 * with the iomem_resource as parent.
35 *
36 * The e820_saved is directly saved after the BIOS-provided memory map is
37 * copied. It doesn't get modified afterwards. It's registered for the
38 * /sys/firmware/memmap interface.
39 *
40 * That memory map is not modified and is used as base for kexec. The kexec'd
41 * kernel should get the same memory map as the firmware provides. Then the
42 * user can e.g. boot the original kernel with mem=1G while still booting the
43 * next kernel with full memory.
44 */
45struct e820map e820;
46struct e820map e820_saved;
47
48/* For PCI or other memory-mapped resources */
49unsigned long pci_mem_start = 0xaeedbabe;
50#ifdef CONFIG_PCI
51EXPORT_SYMBOL(pci_mem_start);
52#endif
53
54/*
55 * This function checks if any part of the range <start,end> is mapped
56 * with type.
57 */
58int
59e820_any_mapped(u64 start, u64 end, unsigned type)
60{
61 int i;
62
63 for (i = 0; i < e820.nr_map; i++) {
64 struct e820entry *ei = &e820.map[i];
65
66 if (type && ei->type != type)
67 continue;
68 if (ei->addr >= end || ei->addr + ei->size <= start)
69 continue;
70 return 1;
71 }
72 return 0;
73}
74EXPORT_SYMBOL_GPL(e820_any_mapped);
75
76/*
77 * This function checks if the entire range <start,end> is mapped with type.
78 *
79 * Note: this function only works correct if the e820 table is sorted and
80 * not-overlapping, which is the case
81 */
82int __init e820_all_mapped(u64 start, u64 end, unsigned type)
83{
84 int i;
85
86 for (i = 0; i < e820.nr_map; i++) {
87 struct e820entry *ei = &e820.map[i];
88
89 if (type && ei->type != type)
90 continue;
91 /* is the region (part) in overlap with the current region ?*/
92 if (ei->addr >= end || ei->addr + ei->size <= start)
93 continue;
94
95 /* if the region is at the beginning of <start,end> we move
96 * start to the end of the region since it's ok until there
97 */
98 if (ei->addr <= start)
99 start = ei->addr + ei->size;
100 /*
101 * if start is now at or beyond end, we're done, full
102 * coverage
103 */
104 if (start >= end)
105 return 1;
106 }
107 return 0;
108}
109
110/*
111 * Add a memory region to the kernel e820 map.
112 */
113void __init e820_add_region(u64 start, u64 size, int type)
114{
115 int x = e820.nr_map;
116
117 if (x == ARRAY_SIZE(e820.map)) {
118 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
119 return;
120 }
121
122 e820.map[x].addr = start;
123 e820.map[x].size = size;
124 e820.map[x].type = type;
125 e820.nr_map++;
126}
127
128void __init e820_print_map(char *who)
129{
130 int i;
131
132 for (i = 0; i < e820.nr_map; i++) {
133 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
134 (unsigned long long) e820.map[i].addr,
135 (unsigned long long)
136 (e820.map[i].addr + e820.map[i].size));
137 switch (e820.map[i].type) {
138 case E820_RAM:
139 case E820_RESERVED_KERN:
140 printk(KERN_CONT "(usable)\n");
141 break;
142 case E820_RESERVED:
143 printk(KERN_CONT "(reserved)\n");
144 break;
145 case E820_ACPI:
146 printk(KERN_CONT "(ACPI data)\n");
147 break;
148 case E820_NVS:
149 printk(KERN_CONT "(ACPI NVS)\n");
150 break;
151 default:
152 printk(KERN_CONT "type %u\n", e820.map[i].type);
153 break;
154 }
155 }
156}
157
158/*
159 * Sanitize the BIOS e820 map.
160 *
161 * Some e820 responses include overlapping entries. The following
162 * replaces the original e820 map with a new one, removing overlaps,
163 * and resolving conflicting memory types in favor of highest
164 * numbered type.
165 *
166 * The input parameter biosmap points to an array of 'struct
167 * e820entry' which on entry has elements in the range [0, *pnr_map)
168 * valid, and which has space for up to max_nr_map entries.
169 * On return, the resulting sanitized e820 map entries will be in
170 * overwritten in the same location, starting at biosmap.
171 *
172 * The integer pointed to by pnr_map must be valid on entry (the
173 * current number of valid entries located at biosmap) and will
174 * be updated on return, with the new number of valid entries
175 * (something no more than max_nr_map.)
176 *
177 * The return value from sanitize_e820_map() is zero if it
178 * successfully 'sanitized' the map entries passed in, and is -1
179 * if it did nothing, which can happen if either of (1) it was
180 * only passed one map entry, or (2) any of the input map entries
181 * were invalid (start + size < start, meaning that the size was
182 * so big the described memory range wrapped around through zero.)
183 *
184 * Visually we're performing the following
185 * (1,2,3,4 = memory types)...
186 *
187 * Sample memory map (w/overlaps):
188 * ____22__________________
189 * ______________________4_
190 * ____1111________________
191 * _44_____________________
192 * 11111111________________
193 * ____________________33__
194 * ___________44___________
195 * __________33333_________
196 * ______________22________
197 * ___________________2222_
198 * _________111111111______
199 * _____________________11_
200 * _________________4______
201 *
202 * Sanitized equivalent (no overlap):
203 * 1_______________________
204 * _44_____________________
205 * ___1____________________
206 * ____22__________________
207 * ______11________________
208 * _________1______________
209 * __________3_____________
210 * ___________44___________
211 * _____________33_________
212 * _______________2________
213 * ________________1_______
214 * _________________4______
215 * ___________________2____
216 * ____________________33__
217 * ______________________4_
218 */
219
220int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
221 int *pnr_map)
222{
223 struct change_member {
224 struct e820entry *pbios; /* pointer to original bios entry */
225 unsigned long long addr; /* address for this change point */
226 };
227 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
228 static struct change_member *change_point[2*E820_X_MAX] __initdata;
229 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
230 static struct e820entry new_bios[E820_X_MAX] __initdata;
231 struct change_member *change_tmp;
232 unsigned long current_type, last_type;
233 unsigned long long last_addr;
234 int chgidx, still_changing;
235 int overlap_entries;
236 int new_bios_entry;
237 int old_nr, new_nr, chg_nr;
238 int i;
239
240 /* if there's only one memory region, don't bother */
241 if (*pnr_map < 2)
242 return -1;
243
244 old_nr = *pnr_map;
245 BUG_ON(old_nr > max_nr_map);
246
247 /* bail out if we find any unreasonable addresses in bios map */
248 for (i = 0; i < old_nr; i++)
249 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
250 return -1;
251
252 /* create pointers for initial change-point information (for sorting) */
253 for (i = 0; i < 2 * old_nr; i++)
254 change_point[i] = &change_point_list[i];
255
256 /* record all known change-points (starting and ending addresses),
257 omitting those that are for empty memory regions */
258 chgidx = 0;
259 for (i = 0; i < old_nr; i++) {
260 if (biosmap[i].size != 0) {
261 change_point[chgidx]->addr = biosmap[i].addr;
262 change_point[chgidx++]->pbios = &biosmap[i];
263 change_point[chgidx]->addr = biosmap[i].addr +
264 biosmap[i].size;
265 change_point[chgidx++]->pbios = &biosmap[i];
266 }
267 }
268 chg_nr = chgidx;
269
270 /* sort change-point list by memory addresses (low -> high) */
271 still_changing = 1;
272 while (still_changing) {
273 still_changing = 0;
274 for (i = 1; i < chg_nr; i++) {
275 unsigned long long curaddr, lastaddr;
276 unsigned long long curpbaddr, lastpbaddr;
277
278 curaddr = change_point[i]->addr;
279 lastaddr = change_point[i - 1]->addr;
280 curpbaddr = change_point[i]->pbios->addr;
281 lastpbaddr = change_point[i - 1]->pbios->addr;
282
283 /*
284 * swap entries, when:
285 *
286 * curaddr > lastaddr or
287 * curaddr == lastaddr and curaddr == curpbaddr and
288 * lastaddr != lastpbaddr
289 */
290 if (curaddr < lastaddr ||
291 (curaddr == lastaddr && curaddr == curpbaddr &&
292 lastaddr != lastpbaddr)) {
293 change_tmp = change_point[i];
294 change_point[i] = change_point[i-1];
295 change_point[i-1] = change_tmp;
296 still_changing = 1;
297 }
298 }
299 }
300
301 /* create a new bios memory map, removing overlaps */
302 overlap_entries = 0; /* number of entries in the overlap table */
303 new_bios_entry = 0; /* index for creating new bios map entries */
304 last_type = 0; /* start with undefined memory type */
305 last_addr = 0; /* start with 0 as last starting address */
306
307 /* loop through change-points, determining affect on the new bios map */
308 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
309 /* keep track of all overlapping bios entries */
310 if (change_point[chgidx]->addr ==
311 change_point[chgidx]->pbios->addr) {
312 /*
313 * add map entry to overlap list (> 1 entry
314 * implies an overlap)
315 */
316 overlap_list[overlap_entries++] =
317 change_point[chgidx]->pbios;
318 } else {
319 /*
320 * remove entry from list (order independent,
321 * so swap with last)
322 */
323 for (i = 0; i < overlap_entries; i++) {
324 if (overlap_list[i] ==
325 change_point[chgidx]->pbios)
326 overlap_list[i] =
327 overlap_list[overlap_entries-1];
328 }
329 overlap_entries--;
330 }
331 /*
332 * if there are overlapping entries, decide which
333 * "type" to use (larger value takes precedence --
334 * 1=usable, 2,3,4,4+=unusable)
335 */
336 current_type = 0;
337 for (i = 0; i < overlap_entries; i++)
338 if (overlap_list[i]->type > current_type)
339 current_type = overlap_list[i]->type;
340 /*
341 * continue building up new bios map based on this
342 * information
343 */
344 if (current_type != last_type) {
345 if (last_type != 0) {
346 new_bios[new_bios_entry].size =
347 change_point[chgidx]->addr - last_addr;
348 /*
349 * move forward only if the new size
350 * was non-zero
351 */
352 if (new_bios[new_bios_entry].size != 0)
353 /*
354 * no more space left for new
355 * bios entries ?
356 */
357 if (++new_bios_entry >= max_nr_map)
358 break;
359 }
360 if (current_type != 0) {
361 new_bios[new_bios_entry].addr =
362 change_point[chgidx]->addr;
363 new_bios[new_bios_entry].type = current_type;
364 last_addr = change_point[chgidx]->addr;
365 }
366 last_type = current_type;
367 }
368 }
369 /* retain count for new bios entries */
370 new_nr = new_bios_entry;
371
372 /* copy new bios mapping into original location */
373 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
374 *pnr_map = new_nr;
375
376 return 0;
377}
378
379static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
380{
381 while (nr_map) {
382 u64 start = biosmap->addr;
383 u64 size = biosmap->size;
384 u64 end = start + size;
385 u32 type = biosmap->type;
386
387 /* Overflow in 64 bits? Ignore the memory map. */
388 if (start > end)
389 return -1;
390
391 e820_add_region(start, size, type);
392
393 biosmap++;
394 nr_map--;
395 }
396 return 0;
397}
398
399/*
400 * Copy the BIOS e820 map into a safe place.
401 *
402 * Sanity-check it while we're at it..
403 *
404 * If we're lucky and live on a modern system, the setup code
405 * will have given us a memory map that we can use to properly
406 * set up memory. If we aren't, we'll fake a memory map.
407 */
408static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
409{
410 /* Only one memory region (or negative)? Ignore it */
411 if (nr_map < 2)
412 return -1;
413
414 return __append_e820_map(biosmap, nr_map);
415}
416
417static u64 __init e820_update_range_map(struct e820map *e820x, u64 start,
418 u64 size, unsigned old_type,
419 unsigned new_type)
420{
421 int i;
422 u64 real_updated_size = 0;
423
424 BUG_ON(old_type == new_type);
425
426 if (size > (ULLONG_MAX - start))
427 size = ULLONG_MAX - start;
428
429 for (i = 0; i < e820.nr_map; i++) {
430 struct e820entry *ei = &e820x->map[i];
431 u64 final_start, final_end;
432 if (ei->type != old_type)
433 continue;
434 /* totally covered? */
435 if (ei->addr >= start &&
436 (ei->addr + ei->size) <= (start + size)) {
437 ei->type = new_type;
438 real_updated_size += ei->size;
439 continue;
440 }
441 /* partially covered */
442 final_start = max(start, ei->addr);
443 final_end = min(start + size, ei->addr + ei->size);
444 if (final_start >= final_end)
445 continue;
446 e820_add_region(final_start, final_end - final_start,
447 new_type);
448 real_updated_size += final_end - final_start;
449
450 ei->size -= final_end - final_start;
451 if (ei->addr < final_start)
452 continue;
453 ei->addr = final_end;
454 }
455 return real_updated_size;
456}
457
458u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
459 unsigned new_type)
460{
461 return e820_update_range_map(&e820, start, size, old_type, new_type);
462}
463
464static u64 __init e820_update_range_saved(u64 start, u64 size,
465 unsigned old_type, unsigned new_type)
466{
467 return e820_update_range_map(&e820_saved, start, size, old_type,
468 new_type);
469}
470
471/* make e820 not cover the range */
472u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
473 int checktype)
474{
475 int i;
476 u64 real_removed_size = 0;
477
478 if (size > (ULLONG_MAX - start))
479 size = ULLONG_MAX - start;
480
481 for (i = 0; i < e820.nr_map; i++) {
482 struct e820entry *ei = &e820.map[i];
483 u64 final_start, final_end;
484
485 if (checktype && ei->type != old_type)
486 continue;
487 /* totally covered? */
488 if (ei->addr >= start &&
489 (ei->addr + ei->size) <= (start + size)) {
490 real_removed_size += ei->size;
491 memset(ei, 0, sizeof(struct e820entry));
492 continue;
493 }
494 /* partially covered */
495 final_start = max(start, ei->addr);
496 final_end = min(start + size, ei->addr + ei->size);
497 if (final_start >= final_end)
498 continue;
499 real_removed_size += final_end - final_start;
500
501 ei->size -= final_end - final_start;
502 if (ei->addr < final_start)
503 continue;
504 ei->addr = final_end;
505 }
506 return real_removed_size;
507}
508
509void __init update_e820(void)
510{
511 int nr_map;
512
513 nr_map = e820.nr_map;
514 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
515 return;
516 e820.nr_map = nr_map;
517 printk(KERN_INFO "modified physical RAM map:\n");
518 e820_print_map("modified");
519}
520static void __init update_e820_saved(void)
521{
522 int nr_map;
523
524 nr_map = e820_saved.nr_map;
525 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
526 return;
527 e820_saved.nr_map = nr_map;
528}
529#define MAX_GAP_END 0x100000000ull
530/*
531 * Search for a gap in the e820 memory space from start_addr to end_addr.
532 */
533__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
534 unsigned long start_addr, unsigned long long end_addr)
535{
536 unsigned long long last;
537 int i = e820.nr_map;
538 int found = 0;
539
540 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
541
542 while (--i >= 0) {
543 unsigned long long start = e820.map[i].addr;
544 unsigned long long end = start + e820.map[i].size;
545
546 if (end < start_addr)
547 continue;
548
549 /*
550 * Since "last" is at most 4GB, we know we'll
551 * fit in 32 bits if this condition is true
552 */
553 if (last > end) {
554 unsigned long gap = last - end;
555
556 if (gap >= *gapsize) {
557 *gapsize = gap;
558 *gapstart = end;
559 found = 1;
560 }
561 }
562 if (start < last)
563 last = start;
564 }
565 return found;
566}
567
568/*
569 * Search for the biggest gap in the low 32 bits of the e820
570 * memory space. We pass this space to PCI to assign MMIO resources
571 * for hotplug or unconfigured devices in.
572 * Hopefully the BIOS let enough space left.
573 */
574__init void e820_setup_gap(void)
575{
576 unsigned long gapstart, gapsize, round;
577 int found;
578
579 gapstart = 0x10000000;
580 gapsize = 0x400000;
581 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
582
583#ifdef CONFIG_X86_64
584 if (!found) {
585 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
586 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
587 "address range\n"
588 KERN_ERR "PCI: Unassigned devices with 32bit resource "
589 "registers may break!\n");
590 }
591#endif
592
593 /*
594 * See how much we want to round up: start off with
595 * rounding to the next 1MB area.
596 */
597 round = 0x100000;
598 while ((gapsize >> 4) > round)
599 round += round;
600 /* Fun with two's complement */
601 pci_mem_start = (gapstart + round) & -round;
602
603 printk(KERN_INFO
604 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
605 pci_mem_start, gapstart, gapsize);
606}
607
608/**
609 * Because of the size limitation of struct boot_params, only first
610 * 128 E820 memory entries are passed to kernel via
611 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
612 * linked list of struct setup_data, which is parsed here.
613 */
614void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
615{
616 u32 map_len;
617 int entries;
618 struct e820entry *extmap;
619
620 entries = sdata->len / sizeof(struct e820entry);
621 map_len = sdata->len + sizeof(struct setup_data);
622 if (map_len > PAGE_SIZE)
623 sdata = early_ioremap(pa_data, map_len);
624 extmap = (struct e820entry *)(sdata->data);
625 __append_e820_map(extmap, entries);
626 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
627 if (map_len > PAGE_SIZE)
628 early_iounmap(sdata, map_len);
629 printk(KERN_INFO "extended physical RAM map:\n");
630 e820_print_map("extended");
631}
632
633#if defined(CONFIG_X86_64) || \
634 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
635/**
636 * Find the ranges of physical addresses that do not correspond to
637 * e820 RAM areas and mark the corresponding pages as nosave for
638 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
639 *
640 * This function requires the e820 map to be sorted and without any
641 * overlapping entries and assumes the first e820 area to be RAM.
642 */
643void __init e820_mark_nosave_regions(unsigned long limit_pfn)
644{
645 int i;
646 unsigned long pfn;
647
648 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
649 for (i = 1; i < e820.nr_map; i++) {
650 struct e820entry *ei = &e820.map[i];
651
652 if (pfn < PFN_UP(ei->addr))
653 register_nosave_region(pfn, PFN_UP(ei->addr));
654
655 pfn = PFN_DOWN(ei->addr + ei->size);
656 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
657 register_nosave_region(PFN_UP(ei->addr), pfn);
658
659 if (pfn >= limit_pfn)
660 break;
661 }
662}
663#endif
664
665/*
666 * Early reserved memory areas.
667 */
668#define MAX_EARLY_RES 20
669
670struct early_res {
671 u64 start, end;
672 char name[16];
673 char overlap_ok;
674};
675static struct early_res early_res[MAX_EARLY_RES] __initdata = {
676 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
677#if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
678 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
679#endif
680#if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
681 /*
682 * But first pinch a few for the stack/trampoline stuff
683 * FIXME: Don't need the extra page at 4K, but need to fix
684 * trampoline before removing it. (see the GDT stuff)
685 */
686 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
687 /*
688 * Has to be in very low memory so we can execute
689 * real-mode AP code.
690 */
691 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
692#endif
693 {}
694};
695
696static int __init find_overlapped_early(u64 start, u64 end)
697{
698 int i;
699 struct early_res *r;
700
701 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
702 r = &early_res[i];
703 if (end > r->start && start < r->end)
704 break;
705 }
706
707 return i;
708}
709
710/*
711 * Drop the i-th range from the early reservation map,
712 * by copying any higher ranges down one over it, and
713 * clearing what had been the last slot.
714 */
715static void __init drop_range(int i)
716{
717 int j;
718
719 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
720 ;
721
722 memmove(&early_res[i], &early_res[i + 1],
723 (j - 1 - i) * sizeof(struct early_res));
724
725 early_res[j - 1].end = 0;
726}
727
728/*
729 * Split any existing ranges that:
730 * 1) are marked 'overlap_ok', and
731 * 2) overlap with the stated range [start, end)
732 * into whatever portion (if any) of the existing range is entirely
733 * below or entirely above the stated range. Drop the portion
734 * of the existing range that overlaps with the stated range,
735 * which will allow the caller of this routine to then add that
736 * stated range without conflicting with any existing range.
737 */
738static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
739{
740 int i;
741 struct early_res *r;
742 u64 lower_start, lower_end;
743 u64 upper_start, upper_end;
744 char name[16];
745
746 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
747 r = &early_res[i];
748
749 /* Continue past non-overlapping ranges */
750 if (end <= r->start || start >= r->end)
751 continue;
752
753 /*
754 * Leave non-ok overlaps as is; let caller
755 * panic "Overlapping early reservations"
756 * when it hits this overlap.
757 */
758 if (!r->overlap_ok)
759 return;
760
761 /*
762 * We have an ok overlap. We will drop it from the early
763 * reservation map, and add back in any non-overlapping
764 * portions (lower or upper) as separate, overlap_ok,
765 * non-overlapping ranges.
766 */
767
768 /* 1. Note any non-overlapping (lower or upper) ranges. */
769 strncpy(name, r->name, sizeof(name) - 1);
770
771 lower_start = lower_end = 0;
772 upper_start = upper_end = 0;
773 if (r->start < start) {
774 lower_start = r->start;
775 lower_end = start;
776 }
777 if (r->end > end) {
778 upper_start = end;
779 upper_end = r->end;
780 }
781
782 /* 2. Drop the original ok overlapping range */
783 drop_range(i);
784
785 i--; /* resume for-loop on copied down entry */
786
787 /* 3. Add back in any non-overlapping ranges. */
788 if (lower_end)
789 reserve_early_overlap_ok(lower_start, lower_end, name);
790 if (upper_end)
791 reserve_early_overlap_ok(upper_start, upper_end, name);
792 }
793}
794
795static void __init __reserve_early(u64 start, u64 end, char *name,
796 int overlap_ok)
797{
798 int i;
799 struct early_res *r;
800
801 i = find_overlapped_early(start, end);
802 if (i >= MAX_EARLY_RES)
803 panic("Too many early reservations");
804 r = &early_res[i];
805 if (r->end)
806 panic("Overlapping early reservations "
807 "%llx-%llx %s to %llx-%llx %s\n",
808 start, end - 1, name?name:"", r->start,
809 r->end - 1, r->name);
810 r->start = start;
811 r->end = end;
812 r->overlap_ok = overlap_ok;
813 if (name)
814 strncpy(r->name, name, sizeof(r->name) - 1);
815}
816
817/*
818 * A few early reservtations come here.
819 *
820 * The 'overlap_ok' in the name of this routine does -not- mean it
821 * is ok for these reservations to overlap an earlier reservation.
822 * Rather it means that it is ok for subsequent reservations to
823 * overlap this one.
824 *
825 * Use this entry point to reserve early ranges when you are doing
826 * so out of "Paranoia", reserving perhaps more memory than you need,
827 * just in case, and don't mind a subsequent overlapping reservation
828 * that is known to be needed.
829 *
830 * The drop_overlaps_that_are_ok() call here isn't really needed.
831 * It would be needed if we had two colliding 'overlap_ok'
832 * reservations, so that the second such would not panic on the
833 * overlap with the first. We don't have any such as of this
834 * writing, but might as well tolerate such if it happens in
835 * the future.
836 */
837void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
838{
839 drop_overlaps_that_are_ok(start, end);
840 __reserve_early(start, end, name, 1);
841}
842
843/*
844 * Most early reservations come here.
845 *
846 * We first have drop_overlaps_that_are_ok() drop any pre-existing
847 * 'overlap_ok' ranges, so that we can then reserve this memory
848 * range without risk of panic'ing on an overlapping overlap_ok
849 * early reservation.
850 */
851void __init reserve_early(u64 start, u64 end, char *name)
852{
853 drop_overlaps_that_are_ok(start, end);
854 __reserve_early(start, end, name, 0);
855}
856
857void __init free_early(u64 start, u64 end)
858{
859 struct early_res *r;
860 int i;
861
862 i = find_overlapped_early(start, end);
863 r = &early_res[i];
864 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
865 panic("free_early on not reserved area: %llx-%llx!",
866 start, end - 1);
867
868 drop_range(i);
869}
870
871void __init early_res_to_bootmem(u64 start, u64 end)
872{
873 int i, count;
874 u64 final_start, final_end;
875
876 count = 0;
877 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
878 count++;
879
880 printk(KERN_INFO "(%d early reservations) ==> bootmem [%010llx - %010llx]\n",
881 count, start, end);
882 for (i = 0; i < count; i++) {
883 struct early_res *r = &early_res[i];
884 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
885 r->start, r->end, r->name);
886 final_start = max(start, r->start);
887 final_end = min(end, r->end);
888 if (final_start >= final_end) {
889 printk(KERN_CONT "\n");
890 continue;
891 }
892 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
893 final_start, final_end);
894 reserve_bootmem_generic(final_start, final_end - final_start,
895 BOOTMEM_DEFAULT);
896 }
897}
898
899/* Check for already reserved areas */
900static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
901{
902 int i;
903 u64 addr = *addrp;
904 int changed = 0;
905 struct early_res *r;
906again:
907 i = find_overlapped_early(addr, addr + size);
908 r = &early_res[i];
909 if (i < MAX_EARLY_RES && r->end) {
910 *addrp = addr = round_up(r->end, align);
911 changed = 1;
912 goto again;
913 }
914 return changed;
915}
916
917/* Check for already reserved areas */
918static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
919{
920 int i;
921 u64 addr = *addrp, last;
922 u64 size = *sizep;
923 int changed = 0;
924again:
925 last = addr + size;
926 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
927 struct early_res *r = &early_res[i];
928 if (last > r->start && addr < r->start) {
929 size = r->start - addr;
930 changed = 1;
931 goto again;
932 }
933 if (last > r->end && addr < r->end) {
934 addr = round_up(r->end, align);
935 size = last - addr;
936 changed = 1;
937 goto again;
938 }
939 if (last <= r->end && addr >= r->start) {
940 (*sizep)++;
941 return 0;
942 }
943 }
944 if (changed) {
945 *addrp = addr;
946 *sizep = size;
947 }
948 return changed;
949}
950
951/*
952 * Find a free area with specified alignment in a specific range.
953 */
954u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
955{
956 int i;
957
958 for (i = 0; i < e820.nr_map; i++) {
959 struct e820entry *ei = &e820.map[i];
960 u64 addr, last;
961 u64 ei_last;
962
963 if (ei->type != E820_RAM)
964 continue;
965 addr = round_up(ei->addr, align);
966 ei_last = ei->addr + ei->size;
967 if (addr < start)
968 addr = round_up(start, align);
969 if (addr >= ei_last)
970 continue;
971 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
972 ;
973 last = addr + size;
974 if (last > ei_last)
975 continue;
976 if (last > end)
977 continue;
978 return addr;
979 }
980 return -1ULL;
981}
982
983/*
984 * Find next free range after *start
985 */
986u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
987{
988 int i;
989
990 for (i = 0; i < e820.nr_map; i++) {
991 struct e820entry *ei = &e820.map[i];
992 u64 addr, last;
993 u64 ei_last;
994
995 if (ei->type != E820_RAM)
996 continue;
997 addr = round_up(ei->addr, align);
998 ei_last = ei->addr + ei->size;
999 if (addr < start)
1000 addr = round_up(start, align);
1001 if (addr >= ei_last)
1002 continue;
1003 *sizep = ei_last - addr;
1004 while (bad_addr_size(&addr, sizep, align) &&
1005 addr + *sizep <= ei_last)
1006 ;
1007 last = addr + *sizep;
1008 if (last > ei_last)
1009 continue;
1010 return addr;
1011 }
1012 return -1UL;
1013
1014}
1015
1016/*
1017 * pre allocated 4k and reserved it in e820
1018 */
1019u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
1020{
1021 u64 size = 0;
1022 u64 addr;
1023 u64 start;
1024
1025 start = startt;
1026 while (size < sizet)
1027 start = find_e820_area_size(start, &size, align);
1028
1029 if (size < sizet)
1030 return 0;
1031
1032 addr = round_down(start + size - sizet, align);
1033 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
1034 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
1035 printk(KERN_INFO "update e820 for early_reserve_e820\n");
1036 update_e820();
1037 update_e820_saved();
1038
1039 return addr;
1040}
1041
1042#ifdef CONFIG_X86_32
1043# ifdef CONFIG_X86_PAE
1044# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1045# else
1046# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1047# endif
1048#else /* CONFIG_X86_32 */
1049# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1050#endif
1051
1052/*
1053 * Find the highest page frame number we have available
1054 */
1055static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
1056{
1057 int i;
1058 unsigned long last_pfn = 0;
1059 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1060
1061 for (i = 0; i < e820.nr_map; i++) {
1062 struct e820entry *ei = &e820.map[i];
1063 unsigned long start_pfn;
1064 unsigned long end_pfn;
1065
1066 if (ei->type != type)
1067 continue;
1068
1069 start_pfn = ei->addr >> PAGE_SHIFT;
1070 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
1071
1072 if (start_pfn >= limit_pfn)
1073 continue;
1074 if (end_pfn > limit_pfn) {
1075 last_pfn = limit_pfn;
1076 break;
1077 }
1078 if (end_pfn > last_pfn)
1079 last_pfn = end_pfn;
1080 }
1081
1082 if (last_pfn > max_arch_pfn)
1083 last_pfn = max_arch_pfn;
1084
1085 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1086 last_pfn, max_arch_pfn);
1087 return last_pfn;
1088}
1089unsigned long __init e820_end_of_ram_pfn(void)
1090{
1091 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
1092}
1093
1094unsigned long __init e820_end_of_low_ram_pfn(void)
1095{
1096 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
1097}
1098/*
1099 * Finds an active region in the address range from start_pfn to last_pfn and
1100 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1101 */
1102int __init e820_find_active_region(const struct e820entry *ei,
1103 unsigned long start_pfn,
1104 unsigned long last_pfn,
1105 unsigned long *ei_startpfn,
1106 unsigned long *ei_endpfn)
1107{
1108 u64 align = PAGE_SIZE;
1109
1110 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1111 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1112
1113 /* Skip map entries smaller than a page */
1114 if (*ei_startpfn >= *ei_endpfn)
1115 return 0;
1116
1117 /* Skip if map is outside the node */
1118 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1119 *ei_startpfn >= last_pfn)
1120 return 0;
1121
1122 /* Check for overlaps */
1123 if (*ei_startpfn < start_pfn)
1124 *ei_startpfn = start_pfn;
1125 if (*ei_endpfn > last_pfn)
1126 *ei_endpfn = last_pfn;
1127
1128 return 1;
1129}
1130
1131/* Walk the e820 map and register active regions within a node */
1132void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1133 unsigned long last_pfn)
1134{
1135 unsigned long ei_startpfn;
1136 unsigned long ei_endpfn;
1137 int i;
1138
1139 for (i = 0; i < e820.nr_map; i++)
1140 if (e820_find_active_region(&e820.map[i],
1141 start_pfn, last_pfn,
1142 &ei_startpfn, &ei_endpfn))
1143 add_active_range(nid, ei_startpfn, ei_endpfn);
1144}
1145
1146/*
1147 * Find the hole size (in bytes) in the memory range.
1148 * @start: starting address of the memory range to scan
1149 * @end: ending address of the memory range to scan
1150 */
1151u64 __init e820_hole_size(u64 start, u64 end)
1152{
1153 unsigned long start_pfn = start >> PAGE_SHIFT;
1154 unsigned long last_pfn = end >> PAGE_SHIFT;
1155 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1156 int i;
1157
1158 for (i = 0; i < e820.nr_map; i++) {
1159 if (e820_find_active_region(&e820.map[i],
1160 start_pfn, last_pfn,
1161 &ei_startpfn, &ei_endpfn))
1162 ram += ei_endpfn - ei_startpfn;
1163 }
1164 return end - start - ((u64)ram << PAGE_SHIFT);
1165}
1166
1167static void early_panic(char *msg)
1168{
1169 early_printk(msg);
1170 panic(msg);
1171}
1172
1173static int userdef __initdata;
1174
1175/* "mem=nopentium" disables the 4MB page tables. */
1176static int __init parse_memopt(char *p)
1177{
1178 u64 mem_size;
1179
1180 if (!p)
1181 return -EINVAL;
1182
1183#ifdef CONFIG_X86_32
1184 if (!strcmp(p, "nopentium")) {
1185 setup_clear_cpu_cap(X86_FEATURE_PSE);
1186 return 0;
1187 }
1188#endif
1189
1190 userdef = 1;
1191 mem_size = memparse(p, &p);
1192 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1193
1194 return 0;
1195}
1196early_param("mem", parse_memopt);
1197
1198static int __init parse_memmap_opt(char *p)
1199{
1200 char *oldp;
1201 u64 start_at, mem_size;
1202
1203 if (!p)
1204 return -EINVAL;
1205
1206 if (!strcmp(p, "exactmap")) {
1207#ifdef CONFIG_CRASH_DUMP
1208 /*
1209 * If we are doing a crash dump, we still need to know
1210 * the real mem size before original memory map is
1211 * reset.
1212 */
1213 saved_max_pfn = e820_end_of_ram_pfn();
1214#endif
1215 e820.nr_map = 0;
1216 userdef = 1;
1217 return 0;
1218 }
1219
1220 oldp = p;
1221 mem_size = memparse(p, &p);
1222 if (p == oldp)
1223 return -EINVAL;
1224
1225 userdef = 1;
1226 if (*p == '@') {
1227 start_at = memparse(p+1, &p);
1228 e820_add_region(start_at, mem_size, E820_RAM);
1229 } else if (*p == '#') {
1230 start_at = memparse(p+1, &p);
1231 e820_add_region(start_at, mem_size, E820_ACPI);
1232 } else if (*p == '$') {
1233 start_at = memparse(p+1, &p);
1234 e820_add_region(start_at, mem_size, E820_RESERVED);
1235 } else
1236 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1237
1238 return *p == '\0' ? 0 : -EINVAL;
1239}
1240early_param("memmap", parse_memmap_opt);
1241
1242void __init finish_e820_parsing(void)
1243{
1244 if (userdef) {
1245 int nr = e820.nr_map;
1246
1247 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1248 early_panic("Invalid user supplied memory map");
1249 e820.nr_map = nr;
1250
1251 printk(KERN_INFO "user-defined physical RAM map:\n");
1252 e820_print_map("user");
1253 }
1254}
1255
1256static inline const char *e820_type_to_string(int e820_type)
1257{
1258 switch (e820_type) {
1259 case E820_RESERVED_KERN:
1260 case E820_RAM: return "System RAM";
1261 case E820_ACPI: return "ACPI Tables";
1262 case E820_NVS: return "ACPI Non-volatile Storage";
1263 default: return "reserved";
1264 }
1265}
1266
1267/*
1268 * Mark e820 reserved areas as busy for the resource manager.
1269 */
1270void __init e820_reserve_resources(void)
1271{
1272 int i;
1273 struct resource *res;
1274 u64 end;
1275
1276 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1277 for (i = 0; i < e820.nr_map; i++) {
1278 end = e820.map[i].addr + e820.map[i].size - 1;
1279#ifndef CONFIG_RESOURCES_64BIT
1280 if (end > 0x100000000ULL) {
1281 res++;
1282 continue;
1283 }
1284#endif
1285 res->name = e820_type_to_string(e820.map[i].type);
1286 res->start = e820.map[i].addr;
1287 res->end = end;
1288
1289 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1290 insert_resource(&iomem_resource, res);
1291 res++;
1292 }
1293
1294 for (i = 0; i < e820_saved.nr_map; i++) {
1295 struct e820entry *entry = &e820_saved.map[i];
1296 firmware_map_add_early(entry->addr,
1297 entry->addr + entry->size - 1,
1298 e820_type_to_string(entry->type));
1299 }
1300}
1301
1302char *__init default_machine_specific_memory_setup(void)
1303{
1304 char *who = "BIOS-e820";
1305 int new_nr;
1306 /*
1307 * Try to copy the BIOS-supplied E820-map.
1308 *
1309 * Otherwise fake a memory map; one section from 0k->640k,
1310 * the next section from 1mb->appropriate_mem_k
1311 */
1312 new_nr = boot_params.e820_entries;
1313 sanitize_e820_map(boot_params.e820_map,
1314 ARRAY_SIZE(boot_params.e820_map),
1315 &new_nr);
1316 boot_params.e820_entries = new_nr;
1317 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1318 < 0) {
1319 u64 mem_size;
1320
1321 /* compare results from other methods and take the greater */
1322 if (boot_params.alt_mem_k
1323 < boot_params.screen_info.ext_mem_k) {
1324 mem_size = boot_params.screen_info.ext_mem_k;
1325 who = "BIOS-88";
1326 } else {
1327 mem_size = boot_params.alt_mem_k;
1328 who = "BIOS-e801";
1329 }
1330
1331 e820.nr_map = 0;
1332 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1333 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1334 }
1335
1336 /* In case someone cares... */
1337 return who;
1338}
1339
1340char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1341{
1342 if (x86_quirks->arch_memory_setup) {
1343 char *who = x86_quirks->arch_memory_setup();
1344
1345 if (who)
1346 return who;
1347 }
1348 return default_machine_specific_memory_setup();
1349}
1350
1351/* Overridden in paravirt.c if CONFIG_PARAVIRT */
1352char * __init __attribute__((weak)) memory_setup(void)
1353{
1354 return machine_specific_memory_setup();
1355}
1356
1357void __init setup_memory_map(void)
1358{
1359 char *who;
1360
1361 who = memory_setup();
1362 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1363 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1364 e820_print_map(who);
1365}
diff --git a/arch/x86/kernel/e820_32.c b/arch/x86/kernel/e820_32.c
deleted file mode 100644
index ed733e7cf4e6..000000000000
--- a/arch/x86/kernel/e820_32.c
+++ /dev/null
@@ -1,775 +0,0 @@
1#include <linux/kernel.h>
2#include <linux/types.h>
3#include <linux/init.h>
4#include <linux/bootmem.h>
5#include <linux/ioport.h>
6#include <linux/string.h>
7#include <linux/kexec.h>
8#include <linux/module.h>
9#include <linux/mm.h>
10#include <linux/pfn.h>
11#include <linux/uaccess.h>
12#include <linux/suspend.h>
13
14#include <asm/pgtable.h>
15#include <asm/page.h>
16#include <asm/e820.h>
17#include <asm/setup.h>
18
19struct e820map e820;
20struct change_member {
21 struct e820entry *pbios; /* pointer to original bios entry */
22 unsigned long long addr; /* address for this change point */
23};
24static struct change_member change_point_list[2*E820MAX] __initdata;
25static struct change_member *change_point[2*E820MAX] __initdata;
26static struct e820entry *overlap_list[E820MAX] __initdata;
27static struct e820entry new_bios[E820MAX] __initdata;
28/* For PCI or other memory-mapped resources */
29unsigned long pci_mem_start = 0x10000000;
30#ifdef CONFIG_PCI
31EXPORT_SYMBOL(pci_mem_start);
32#endif
33extern int user_defined_memmap;
34
35static struct resource system_rom_resource = {
36 .name = "System ROM",
37 .start = 0xf0000,
38 .end = 0xfffff,
39 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
40};
41
42static struct resource extension_rom_resource = {
43 .name = "Extension ROM",
44 .start = 0xe0000,
45 .end = 0xeffff,
46 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
47};
48
49static struct resource adapter_rom_resources[] = { {
50 .name = "Adapter ROM",
51 .start = 0xc8000,
52 .end = 0,
53 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
54}, {
55 .name = "Adapter ROM",
56 .start = 0,
57 .end = 0,
58 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
59}, {
60 .name = "Adapter ROM",
61 .start = 0,
62 .end = 0,
63 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
64}, {
65 .name = "Adapter ROM",
66 .start = 0,
67 .end = 0,
68 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
69}, {
70 .name = "Adapter ROM",
71 .start = 0,
72 .end = 0,
73 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
74}, {
75 .name = "Adapter ROM",
76 .start = 0,
77 .end = 0,
78 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
79} };
80
81static struct resource video_rom_resource = {
82 .name = "Video ROM",
83 .start = 0xc0000,
84 .end = 0xc7fff,
85 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
86};
87
88#define ROMSIGNATURE 0xaa55
89
90static int __init romsignature(const unsigned char *rom)
91{
92 const unsigned short * const ptr = (const unsigned short *)rom;
93 unsigned short sig;
94
95 return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
96}
97
98static int __init romchecksum(const unsigned char *rom, unsigned long length)
99{
100 unsigned char sum, c;
101
102 for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
103 sum += c;
104 return !length && !sum;
105}
106
107static void __init probe_roms(void)
108{
109 const unsigned char *rom;
110 unsigned long start, length, upper;
111 unsigned char c;
112 int i;
113
114 /* video rom */
115 upper = adapter_rom_resources[0].start;
116 for (start = video_rom_resource.start; start < upper; start += 2048) {
117 rom = isa_bus_to_virt(start);
118 if (!romsignature(rom))
119 continue;
120
121 video_rom_resource.start = start;
122
123 if (probe_kernel_address(rom + 2, c) != 0)
124 continue;
125
126 /* 0 < length <= 0x7f * 512, historically */
127 length = c * 512;
128
129 /* if checksum okay, trust length byte */
130 if (length && romchecksum(rom, length))
131 video_rom_resource.end = start + length - 1;
132
133 request_resource(&iomem_resource, &video_rom_resource);
134 break;
135 }
136
137 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
138 if (start < upper)
139 start = upper;
140
141 /* system rom */
142 request_resource(&iomem_resource, &system_rom_resource);
143 upper = system_rom_resource.start;
144
145 /* check for extension rom (ignore length byte!) */
146 rom = isa_bus_to_virt(extension_rom_resource.start);
147 if (romsignature(rom)) {
148 length = extension_rom_resource.end - extension_rom_resource.start + 1;
149 if (romchecksum(rom, length)) {
150 request_resource(&iomem_resource, &extension_rom_resource);
151 upper = extension_rom_resource.start;
152 }
153 }
154
155 /* check for adapter roms on 2k boundaries */
156 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
157 rom = isa_bus_to_virt(start);
158 if (!romsignature(rom))
159 continue;
160
161 if (probe_kernel_address(rom + 2, c) != 0)
162 continue;
163
164 /* 0 < length <= 0x7f * 512, historically */
165 length = c * 512;
166
167 /* but accept any length that fits if checksum okay */
168 if (!length || start + length > upper || !romchecksum(rom, length))
169 continue;
170
171 adapter_rom_resources[i].start = start;
172 adapter_rom_resources[i].end = start + length - 1;
173 request_resource(&iomem_resource, &adapter_rom_resources[i]);
174
175 start = adapter_rom_resources[i++].end & ~2047UL;
176 }
177}
178
179/*
180 * Request address space for all standard RAM and ROM resources
181 * and also for regions reported as reserved by the e820.
182 */
183void __init init_iomem_resources(struct resource *code_resource,
184 struct resource *data_resource,
185 struct resource *bss_resource)
186{
187 int i;
188
189 probe_roms();
190 for (i = 0; i < e820.nr_map; i++) {
191 struct resource *res;
192#ifndef CONFIG_RESOURCES_64BIT
193 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
194 continue;
195#endif
196 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
197 switch (e820.map[i].type) {
198 case E820_RAM: res->name = "System RAM"; break;
199 case E820_ACPI: res->name = "ACPI Tables"; break;
200 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
201 default: res->name = "reserved";
202 }
203 res->start = e820.map[i].addr;
204 res->end = res->start + e820.map[i].size - 1;
205 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
206 if (request_resource(&iomem_resource, res)) {
207 kfree(res);
208 continue;
209 }
210 if (e820.map[i].type == E820_RAM) {
211 /*
212 * We don't know which RAM region contains kernel data,
213 * so we try it repeatedly and let the resource manager
214 * test it.
215 */
216 request_resource(res, code_resource);
217 request_resource(res, data_resource);
218 request_resource(res, bss_resource);
219#ifdef CONFIG_KEXEC
220 if (crashk_res.start != crashk_res.end)
221 request_resource(res, &crashk_res);
222#endif
223 }
224 }
225}
226
227#if defined(CONFIG_PM) && defined(CONFIG_HIBERNATION)
228/**
229 * e820_mark_nosave_regions - Find the ranges of physical addresses that do not
230 * correspond to e820 RAM areas and mark the corresponding pages as nosave for
231 * hibernation.
232 *
233 * This function requires the e820 map to be sorted and without any
234 * overlapping entries and assumes the first e820 area to be RAM.
235 */
236void __init e820_mark_nosave_regions(void)
237{
238 int i;
239 unsigned long pfn;
240
241 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
242 for (i = 1; i < e820.nr_map; i++) {
243 struct e820entry *ei = &e820.map[i];
244
245 if (pfn < PFN_UP(ei->addr))
246 register_nosave_region(pfn, PFN_UP(ei->addr));
247
248 pfn = PFN_DOWN(ei->addr + ei->size);
249 if (ei->type != E820_RAM)
250 register_nosave_region(PFN_UP(ei->addr), pfn);
251
252 if (pfn >= max_low_pfn)
253 break;
254 }
255}
256#endif
257
258void __init add_memory_region(unsigned long long start,
259 unsigned long long size, int type)
260{
261 int x;
262
263 x = e820.nr_map;
264
265 if (x == E820MAX) {
266 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
267 return;
268 }
269
270 e820.map[x].addr = start;
271 e820.map[x].size = size;
272 e820.map[x].type = type;
273 e820.nr_map++;
274} /* add_memory_region */
275
276/*
277 * Sanitize the BIOS e820 map.
278 *
279 * Some e820 responses include overlapping entries. The following
280 * replaces the original e820 map with a new one, removing overlaps.
281 *
282 */
283int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
284{
285 struct change_member *change_tmp;
286 unsigned long current_type, last_type;
287 unsigned long long last_addr;
288 int chgidx, still_changing;
289 int overlap_entries;
290 int new_bios_entry;
291 int old_nr, new_nr, chg_nr;
292 int i;
293
294 /*
295 Visually we're performing the following (1,2,3,4 = memory types)...
296
297 Sample memory map (w/overlaps):
298 ____22__________________
299 ______________________4_
300 ____1111________________
301 _44_____________________
302 11111111________________
303 ____________________33__
304 ___________44___________
305 __________33333_________
306 ______________22________
307 ___________________2222_
308 _________111111111______
309 _____________________11_
310 _________________4______
311
312 Sanitized equivalent (no overlap):
313 1_______________________
314 _44_____________________
315 ___1____________________
316 ____22__________________
317 ______11________________
318 _________1______________
319 __________3_____________
320 ___________44___________
321 _____________33_________
322 _______________2________
323 ________________1_______
324 _________________4______
325 ___________________2____
326 ____________________33__
327 ______________________4_
328 */
329 /* if there's only one memory region, don't bother */
330 if (*pnr_map < 2) {
331 return -1;
332 }
333
334 old_nr = *pnr_map;
335
336 /* bail out if we find any unreasonable addresses in bios map */
337 for (i=0; i<old_nr; i++)
338 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
339 return -1;
340 }
341
342 /* create pointers for initial change-point information (for sorting) */
343 for (i=0; i < 2*old_nr; i++)
344 change_point[i] = &change_point_list[i];
345
346 /* record all known change-points (starting and ending addresses),
347 omitting those that are for empty memory regions */
348 chgidx = 0;
349 for (i=0; i < old_nr; i++) {
350 if (biosmap[i].size != 0) {
351 change_point[chgidx]->addr = biosmap[i].addr;
352 change_point[chgidx++]->pbios = &biosmap[i];
353 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
354 change_point[chgidx++]->pbios = &biosmap[i];
355 }
356 }
357 chg_nr = chgidx; /* true number of change-points */
358
359 /* sort change-point list by memory addresses (low -> high) */
360 still_changing = 1;
361 while (still_changing) {
362 still_changing = 0;
363 for (i=1; i < chg_nr; i++) {
364 /* if <current_addr> > <last_addr>, swap */
365 /* or, if current=<start_addr> & last=<end_addr>, swap */
366 if ((change_point[i]->addr < change_point[i-1]->addr) ||
367 ((change_point[i]->addr == change_point[i-1]->addr) &&
368 (change_point[i]->addr == change_point[i]->pbios->addr) &&
369 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
370 )
371 {
372 change_tmp = change_point[i];
373 change_point[i] = change_point[i-1];
374 change_point[i-1] = change_tmp;
375 still_changing=1;
376 }
377 }
378 }
379
380 /* create a new bios memory map, removing overlaps */
381 overlap_entries=0; /* number of entries in the overlap table */
382 new_bios_entry=0; /* index for creating new bios map entries */
383 last_type = 0; /* start with undefined memory type */
384 last_addr = 0; /* start with 0 as last starting address */
385 /* loop through change-points, determining affect on the new bios map */
386 for (chgidx=0; chgidx < chg_nr; chgidx++)
387 {
388 /* keep track of all overlapping bios entries */
389 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
390 {
391 /* add map entry to overlap list (> 1 entry implies an overlap) */
392 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
393 }
394 else
395 {
396 /* remove entry from list (order independent, so swap with last) */
397 for (i=0; i<overlap_entries; i++)
398 {
399 if (overlap_list[i] == change_point[chgidx]->pbios)
400 overlap_list[i] = overlap_list[overlap_entries-1];
401 }
402 overlap_entries--;
403 }
404 /* if there are overlapping entries, decide which "type" to use */
405 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
406 current_type = 0;
407 for (i=0; i<overlap_entries; i++)
408 if (overlap_list[i]->type > current_type)
409 current_type = overlap_list[i]->type;
410 /* continue building up new bios map based on this information */
411 if (current_type != last_type) {
412 if (last_type != 0) {
413 new_bios[new_bios_entry].size =
414 change_point[chgidx]->addr - last_addr;
415 /* move forward only if the new size was non-zero */
416 if (new_bios[new_bios_entry].size != 0)
417 if (++new_bios_entry >= E820MAX)
418 break; /* no more space left for new bios entries */
419 }
420 if (current_type != 0) {
421 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
422 new_bios[new_bios_entry].type = current_type;
423 last_addr=change_point[chgidx]->addr;
424 }
425 last_type = current_type;
426 }
427 }
428 new_nr = new_bios_entry; /* retain count for new bios entries */
429
430 /* copy new bios mapping into original location */
431 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
432 *pnr_map = new_nr;
433
434 return 0;
435}
436
437/*
438 * Copy the BIOS e820 map into a safe place.
439 *
440 * Sanity-check it while we're at it..
441 *
442 * If we're lucky and live on a modern system, the setup code
443 * will have given us a memory map that we can use to properly
444 * set up memory. If we aren't, we'll fake a memory map.
445 *
446 * We check to see that the memory map contains at least 2 elements
447 * before we'll use it, because the detection code in setup.S may
448 * not be perfect and most every PC known to man has two memory
449 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
450 * thinkpad 560x, for example, does not cooperate with the memory
451 * detection code.)
452 */
453int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
454{
455 /* Only one memory region (or negative)? Ignore it */
456 if (nr_map < 2)
457 return -1;
458
459 do {
460 u64 start = biosmap->addr;
461 u64 size = biosmap->size;
462 u64 end = start + size;
463 u32 type = biosmap->type;
464
465 /* Overflow in 64 bits? Ignore the memory map. */
466 if (start > end)
467 return -1;
468
469 add_memory_region(start, size, type);
470 } while (biosmap++, --nr_map);
471
472 return 0;
473}
474
475/*
476 * Find the highest page frame number we have available
477 */
478void __init propagate_e820_map(void)
479{
480 int i;
481
482 max_pfn = 0;
483
484 for (i = 0; i < e820.nr_map; i++) {
485 unsigned long start, end;
486 /* RAM? */
487 if (e820.map[i].type != E820_RAM)
488 continue;
489 start = PFN_UP(e820.map[i].addr);
490 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
491 if (start >= end)
492 continue;
493 if (end > max_pfn)
494 max_pfn = end;
495 memory_present(0, start, end);
496 }
497}
498
499/*
500 * Register fully available low RAM pages with the bootmem allocator.
501 */
502void __init register_bootmem_low_pages(unsigned long max_low_pfn)
503{
504 int i;
505
506 for (i = 0; i < e820.nr_map; i++) {
507 unsigned long curr_pfn, last_pfn, size;
508 /*
509 * Reserve usable low memory
510 */
511 if (e820.map[i].type != E820_RAM)
512 continue;
513 /*
514 * We are rounding up the start address of usable memory:
515 */
516 curr_pfn = PFN_UP(e820.map[i].addr);
517 if (curr_pfn >= max_low_pfn)
518 continue;
519 /*
520 * ... and at the end of the usable range downwards:
521 */
522 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
523
524 if (last_pfn > max_low_pfn)
525 last_pfn = max_low_pfn;
526
527 /*
528 * .. finally, did all the rounding and playing
529 * around just make the area go away?
530 */
531 if (last_pfn <= curr_pfn)
532 continue;
533
534 size = last_pfn - curr_pfn;
535 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
536 }
537}
538
539void __init e820_register_memory(void)
540{
541 unsigned long gapstart, gapsize, round;
542 unsigned long long last;
543 int i;
544
545 /*
546 * Search for the biggest gap in the low 32 bits of the e820
547 * memory space.
548 */
549 last = 0x100000000ull;
550 gapstart = 0x10000000;
551 gapsize = 0x400000;
552 i = e820.nr_map;
553 while (--i >= 0) {
554 unsigned long long start = e820.map[i].addr;
555 unsigned long long end = start + e820.map[i].size;
556
557 /*
558 * Since "last" is at most 4GB, we know we'll
559 * fit in 32 bits if this condition is true
560 */
561 if (last > end) {
562 unsigned long gap = last - end;
563
564 if (gap > gapsize) {
565 gapsize = gap;
566 gapstart = end;
567 }
568 }
569 if (start < last)
570 last = start;
571 }
572
573 /*
574 * See how much we want to round up: start off with
575 * rounding to the next 1MB area.
576 */
577 round = 0x100000;
578 while ((gapsize >> 4) > round)
579 round += round;
580 /* Fun with two's complement */
581 pci_mem_start = (gapstart + round) & -round;
582
583 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
584 pci_mem_start, gapstart, gapsize);
585}
586
587void __init print_memory_map(char *who)
588{
589 int i;
590
591 for (i = 0; i < e820.nr_map; i++) {
592 printk(" %s: %016Lx - %016Lx ", who,
593 e820.map[i].addr,
594 e820.map[i].addr + e820.map[i].size);
595 switch (e820.map[i].type) {
596 case E820_RAM: printk("(usable)\n");
597 break;
598 case E820_RESERVED:
599 printk("(reserved)\n");
600 break;
601 case E820_ACPI:
602 printk("(ACPI data)\n");
603 break;
604 case E820_NVS:
605 printk("(ACPI NVS)\n");
606 break;
607 default: printk("type %u\n", e820.map[i].type);
608 break;
609 }
610 }
611}
612
613void __init limit_regions(unsigned long long size)
614{
615 unsigned long long current_addr;
616 int i;
617
618 print_memory_map("limit_regions start");
619 for (i = 0; i < e820.nr_map; i++) {
620 current_addr = e820.map[i].addr + e820.map[i].size;
621 if (current_addr < size)
622 continue;
623
624 if (e820.map[i].type != E820_RAM)
625 continue;
626
627 if (e820.map[i].addr >= size) {
628 /*
629 * This region starts past the end of the
630 * requested size, skip it completely.
631 */
632 e820.nr_map = i;
633 } else {
634 e820.nr_map = i + 1;
635 e820.map[i].size -= current_addr - size;
636 }
637 print_memory_map("limit_regions endfor");
638 return;
639 }
640 print_memory_map("limit_regions endfunc");
641}
642
643/*
644 * This function checks if any part of the range <start,end> is mapped
645 * with type.
646 */
647int
648e820_any_mapped(u64 start, u64 end, unsigned type)
649{
650 int i;
651 for (i = 0; i < e820.nr_map; i++) {
652 const struct e820entry *ei = &e820.map[i];
653 if (type && ei->type != type)
654 continue;
655 if (ei->addr >= end || ei->addr + ei->size <= start)
656 continue;
657 return 1;
658 }
659 return 0;
660}
661EXPORT_SYMBOL_GPL(e820_any_mapped);
662
663 /*
664 * This function checks if the entire range <start,end> is mapped with type.
665 *
666 * Note: this function only works correct if the e820 table is sorted and
667 * not-overlapping, which is the case
668 */
669int __init
670e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
671{
672 u64 start = s;
673 u64 end = e;
674 int i;
675 for (i = 0; i < e820.nr_map; i++) {
676 struct e820entry *ei = &e820.map[i];
677 if (type && ei->type != type)
678 continue;
679 /* is the region (part) in overlap with the current region ?*/
680 if (ei->addr >= end || ei->addr + ei->size <= start)
681 continue;
682 /* if the region is at the beginning of <start,end> we move
683 * start to the end of the region since it's ok until there
684 */
685 if (ei->addr <= start)
686 start = ei->addr + ei->size;
687 /* if start is now at or beyond end, we're done, full
688 * coverage */
689 if (start >= end)
690 return 1; /* we're done */
691 }
692 return 0;
693}
694
695static int __init parse_memmap(char *arg)
696{
697 if (!arg)
698 return -EINVAL;
699
700 if (strcmp(arg, "exactmap") == 0) {
701#ifdef CONFIG_CRASH_DUMP
702 /* If we are doing a crash dump, we
703 * still need to know the real mem
704 * size before original memory map is
705 * reset.
706 */
707 propagate_e820_map();
708 saved_max_pfn = max_pfn;
709#endif
710 e820.nr_map = 0;
711 user_defined_memmap = 1;
712 } else {
713 /* If the user specifies memory size, we
714 * limit the BIOS-provided memory map to
715 * that size. exactmap can be used to specify
716 * the exact map. mem=number can be used to
717 * trim the existing memory map.
718 */
719 unsigned long long start_at, mem_size;
720
721 mem_size = memparse(arg, &arg);
722 if (*arg == '@') {
723 start_at = memparse(arg+1, &arg);
724 add_memory_region(start_at, mem_size, E820_RAM);
725 } else if (*arg == '#') {
726 start_at = memparse(arg+1, &arg);
727 add_memory_region(start_at, mem_size, E820_ACPI);
728 } else if (*arg == '$') {
729 start_at = memparse(arg+1, &arg);
730 add_memory_region(start_at, mem_size, E820_RESERVED);
731 } else {
732 limit_regions(mem_size);
733 user_defined_memmap = 1;
734 }
735 }
736 return 0;
737}
738early_param("memmap", parse_memmap);
739void __init update_memory_range(u64 start, u64 size, unsigned old_type,
740 unsigned new_type)
741{
742 int i;
743
744 BUG_ON(old_type == new_type);
745
746 for (i = 0; i < e820.nr_map; i++) {
747 struct e820entry *ei = &e820.map[i];
748 u64 final_start, final_end;
749 if (ei->type != old_type)
750 continue;
751 /* totally covered? */
752 if (ei->addr >= start && ei->size <= size) {
753 ei->type = new_type;
754 continue;
755 }
756 /* partially covered */
757 final_start = max(start, ei->addr);
758 final_end = min(start + size, ei->addr + ei->size);
759 if (final_start >= final_end)
760 continue;
761 add_memory_region(final_start, final_end - final_start,
762 new_type);
763 }
764}
765void __init update_e820(void)
766{
767 u8 nr_map;
768
769 nr_map = e820.nr_map;
770 if (sanitize_e820_map(e820.map, &nr_map))
771 return;
772 e820.nr_map = nr_map;
773 printk(KERN_INFO "modified physical RAM map:\n");
774 print_memory_map("modified");
775}
diff --git a/arch/x86/kernel/e820_64.c b/arch/x86/kernel/e820_64.c
deleted file mode 100644
index 124480c0008d..000000000000
--- a/arch/x86/kernel/e820_64.c
+++ /dev/null
@@ -1,952 +0,0 @@
1/*
2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
4 *
5 * Getting sanitize_e820_map() in sync with i386 version by applying change:
6 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
7 * Alex Achenbach <xela@slit.de>, December 2002.
8 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
9 *
10 */
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/init.h>
14#include <linux/bootmem.h>
15#include <linux/ioport.h>
16#include <linux/string.h>
17#include <linux/kexec.h>
18#include <linux/module.h>
19#include <linux/mm.h>
20#include <linux/suspend.h>
21#include <linux/pfn.h>
22
23#include <asm/pgtable.h>
24#include <asm/page.h>
25#include <asm/e820.h>
26#include <asm/proto.h>
27#include <asm/setup.h>
28#include <asm/sections.h>
29#include <asm/kdebug.h>
30#include <asm/trampoline.h>
31
32struct e820map e820;
33
34/*
35 * PFN of last memory page.
36 */
37unsigned long end_pfn;
38
39/*
40 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
41 * The direct mapping extends to max_pfn_mapped, so that we can directly access
42 * apertures, ACPI and other tables without having to play with fixmaps.
43 */
44unsigned long max_pfn_mapped;
45
46/*
47 * Last pfn which the user wants to use.
48 */
49static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
50
51/*
52 * Early reserved memory areas.
53 */
54#define MAX_EARLY_RES 20
55
56struct early_res {
57 unsigned long start, end;
58 char name[16];
59};
60static struct early_res early_res[MAX_EARLY_RES] __initdata = {
61 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
62#ifdef CONFIG_X86_TRAMPOLINE
63 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
64#endif
65 {}
66};
67
68void __init reserve_early(unsigned long start, unsigned long end, char *name)
69{
70 int i;
71 struct early_res *r;
72 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
73 r = &early_res[i];
74 if (end > r->start && start < r->end)
75 panic("Overlapping early reservations %lx-%lx %s to %lx-%lx %s\n",
76 start, end - 1, name?name:"", r->start, r->end - 1, r->name);
77 }
78 if (i >= MAX_EARLY_RES)
79 panic("Too many early reservations");
80 r = &early_res[i];
81 r->start = start;
82 r->end = end;
83 if (name)
84 strncpy(r->name, name, sizeof(r->name) - 1);
85}
86
87void __init free_early(unsigned long start, unsigned long end)
88{
89 struct early_res *r;
90 int i, j;
91
92 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
93 r = &early_res[i];
94 if (start == r->start && end == r->end)
95 break;
96 }
97 if (i >= MAX_EARLY_RES || !early_res[i].end)
98 panic("free_early on not reserved area: %lx-%lx!", start, end);
99
100 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
101 ;
102
103 memmove(&early_res[i], &early_res[i + 1],
104 (j - 1 - i) * sizeof(struct early_res));
105
106 early_res[j - 1].end = 0;
107}
108
109void __init early_res_to_bootmem(unsigned long start, unsigned long end)
110{
111 int i;
112 unsigned long final_start, final_end;
113 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
114 struct early_res *r = &early_res[i];
115 final_start = max(start, r->start);
116 final_end = min(end, r->end);
117 if (final_start >= final_end)
118 continue;
119 printk(KERN_INFO " early res: %d [%lx-%lx] %s\n", i,
120 final_start, final_end - 1, r->name);
121 reserve_bootmem_generic(final_start, final_end - final_start);
122 }
123}
124
125/* Check for already reserved areas */
126static inline int __init
127bad_addr(unsigned long *addrp, unsigned long size, unsigned long align)
128{
129 int i;
130 unsigned long addr = *addrp, last;
131 int changed = 0;
132again:
133 last = addr + size;
134 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
135 struct early_res *r = &early_res[i];
136 if (last >= r->start && addr < r->end) {
137 *addrp = addr = round_up(r->end, align);
138 changed = 1;
139 goto again;
140 }
141 }
142 return changed;
143}
144
145/* Check for already reserved areas */
146static inline int __init
147bad_addr_size(unsigned long *addrp, unsigned long *sizep, unsigned long align)
148{
149 int i;
150 unsigned long addr = *addrp, last;
151 unsigned long size = *sizep;
152 int changed = 0;
153again:
154 last = addr + size;
155 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
156 struct early_res *r = &early_res[i];
157 if (last > r->start && addr < r->start) {
158 size = r->start - addr;
159 changed = 1;
160 goto again;
161 }
162 if (last > r->end && addr < r->end) {
163 addr = round_up(r->end, align);
164 size = last - addr;
165 changed = 1;
166 goto again;
167 }
168 if (last <= r->end && addr >= r->start) {
169 (*sizep)++;
170 return 0;
171 }
172 }
173 if (changed) {
174 *addrp = addr;
175 *sizep = size;
176 }
177 return changed;
178}
179/*
180 * This function checks if any part of the range <start,end> is mapped
181 * with type.
182 */
183int
184e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
185{
186 int i;
187
188 for (i = 0; i < e820.nr_map; i++) {
189 struct e820entry *ei = &e820.map[i];
190
191 if (type && ei->type != type)
192 continue;
193 if (ei->addr >= end || ei->addr + ei->size <= start)
194 continue;
195 return 1;
196 }
197 return 0;
198}
199EXPORT_SYMBOL_GPL(e820_any_mapped);
200
201/*
202 * This function checks if the entire range <start,end> is mapped with type.
203 *
204 * Note: this function only works correct if the e820 table is sorted and
205 * not-overlapping, which is the case
206 */
207int __init e820_all_mapped(unsigned long start, unsigned long end,
208 unsigned type)
209{
210 int i;
211
212 for (i = 0; i < e820.nr_map; i++) {
213 struct e820entry *ei = &e820.map[i];
214
215 if (type && ei->type != type)
216 continue;
217 /* is the region (part) in overlap with the current region ?*/
218 if (ei->addr >= end || ei->addr + ei->size <= start)
219 continue;
220
221 /* if the region is at the beginning of <start,end> we move
222 * start to the end of the region since it's ok until there
223 */
224 if (ei->addr <= start)
225 start = ei->addr + ei->size;
226 /*
227 * if start is now at or beyond end, we're done, full
228 * coverage
229 */
230 if (start >= end)
231 return 1;
232 }
233 return 0;
234}
235
236/*
237 * Find a free area with specified alignment in a specific range.
238 */
239unsigned long __init find_e820_area(unsigned long start, unsigned long end,
240 unsigned long size, unsigned long align)
241{
242 int i;
243
244 for (i = 0; i < e820.nr_map; i++) {
245 struct e820entry *ei = &e820.map[i];
246 unsigned long addr, last;
247 unsigned long ei_last;
248
249 if (ei->type != E820_RAM)
250 continue;
251 addr = round_up(ei->addr, align);
252 ei_last = ei->addr + ei->size;
253 if (addr < start)
254 addr = round_up(start, align);
255 if (addr >= ei_last)
256 continue;
257 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
258 ;
259 last = addr + size;
260 if (last > ei_last)
261 continue;
262 if (last > end)
263 continue;
264 return addr;
265 }
266 return -1UL;
267}
268
269/*
270 * Find next free range after *start
271 */
272unsigned long __init find_e820_area_size(unsigned long start,
273 unsigned long *sizep,
274 unsigned long align)
275{
276 int i;
277
278 for (i = 0; i < e820.nr_map; i++) {
279 struct e820entry *ei = &e820.map[i];
280 unsigned long addr, last;
281 unsigned long ei_last;
282
283 if (ei->type != E820_RAM)
284 continue;
285 addr = round_up(ei->addr, align);
286 ei_last = ei->addr + ei->size;
287 if (addr < start)
288 addr = round_up(start, align);
289 if (addr >= ei_last)
290 continue;
291 *sizep = ei_last - addr;
292 while (bad_addr_size(&addr, sizep, align) &&
293 addr + *sizep <= ei_last)
294 ;
295 last = addr + *sizep;
296 if (last > ei_last)
297 continue;
298 return addr;
299 }
300 return -1UL;
301
302}
303/*
304 * Find the highest page frame number we have available
305 */
306unsigned long __init e820_end_of_ram(void)
307{
308 unsigned long end_pfn;
309
310 end_pfn = find_max_pfn_with_active_regions();
311
312 if (end_pfn > max_pfn_mapped)
313 max_pfn_mapped = end_pfn;
314 if (max_pfn_mapped > MAXMEM>>PAGE_SHIFT)
315 max_pfn_mapped = MAXMEM>>PAGE_SHIFT;
316 if (end_pfn > end_user_pfn)
317 end_pfn = end_user_pfn;
318 if (end_pfn > max_pfn_mapped)
319 end_pfn = max_pfn_mapped;
320
321 printk(KERN_INFO "max_pfn_mapped = %lu\n", max_pfn_mapped);
322 return end_pfn;
323}
324
325/*
326 * Mark e820 reserved areas as busy for the resource manager.
327 */
328void __init e820_reserve_resources(void)
329{
330 int i;
331 struct resource *res;
332
333 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
334 for (i = 0; i < e820.nr_map; i++) {
335 switch (e820.map[i].type) {
336 case E820_RAM: res->name = "System RAM"; break;
337 case E820_ACPI: res->name = "ACPI Tables"; break;
338 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
339 default: res->name = "reserved";
340 }
341 res->start = e820.map[i].addr;
342 res->end = res->start + e820.map[i].size - 1;
343 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
344 insert_resource(&iomem_resource, res);
345 res++;
346 }
347}
348
349/*
350 * Find the ranges of physical addresses that do not correspond to
351 * e820 RAM areas and mark the corresponding pages as nosave for software
352 * suspend and suspend to RAM.
353 *
354 * This function requires the e820 map to be sorted and without any
355 * overlapping entries and assumes the first e820 area to be RAM.
356 */
357void __init e820_mark_nosave_regions(void)
358{
359 int i;
360 unsigned long paddr;
361
362 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
363 for (i = 1; i < e820.nr_map; i++) {
364 struct e820entry *ei = &e820.map[i];
365
366 if (paddr < ei->addr)
367 register_nosave_region(PFN_DOWN(paddr),
368 PFN_UP(ei->addr));
369
370 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
371 if (ei->type != E820_RAM)
372 register_nosave_region(PFN_UP(ei->addr),
373 PFN_DOWN(paddr));
374
375 if (paddr >= (end_pfn << PAGE_SHIFT))
376 break;
377 }
378}
379
380/*
381 * Finds an active region in the address range from start_pfn to end_pfn and
382 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
383 */
384static int __init e820_find_active_region(const struct e820entry *ei,
385 unsigned long start_pfn,
386 unsigned long end_pfn,
387 unsigned long *ei_startpfn,
388 unsigned long *ei_endpfn)
389{
390 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
391 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
392
393 /* Skip map entries smaller than a page */
394 if (*ei_startpfn >= *ei_endpfn)
395 return 0;
396
397 /* Check if max_pfn_mapped should be updated */
398 if (ei->type != E820_RAM && *ei_endpfn > max_pfn_mapped)
399 max_pfn_mapped = *ei_endpfn;
400
401 /* Skip if map is outside the node */
402 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
403 *ei_startpfn >= end_pfn)
404 return 0;
405
406 /* Check for overlaps */
407 if (*ei_startpfn < start_pfn)
408 *ei_startpfn = start_pfn;
409 if (*ei_endpfn > end_pfn)
410 *ei_endpfn = end_pfn;
411
412 /* Obey end_user_pfn to save on memmap */
413 if (*ei_startpfn >= end_user_pfn)
414 return 0;
415 if (*ei_endpfn > end_user_pfn)
416 *ei_endpfn = end_user_pfn;
417
418 return 1;
419}
420
421/* Walk the e820 map and register active regions within a node */
422void __init
423e820_register_active_regions(int nid, unsigned long start_pfn,
424 unsigned long end_pfn)
425{
426 unsigned long ei_startpfn;
427 unsigned long ei_endpfn;
428 int i;
429
430 for (i = 0; i < e820.nr_map; i++)
431 if (e820_find_active_region(&e820.map[i],
432 start_pfn, end_pfn,
433 &ei_startpfn, &ei_endpfn))
434 add_active_range(nid, ei_startpfn, ei_endpfn);
435}
436
437/*
438 * Add a memory region to the kernel e820 map.
439 */
440void __init add_memory_region(unsigned long start, unsigned long size, int type)
441{
442 int x = e820.nr_map;
443
444 if (x == E820MAX) {
445 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
446 return;
447 }
448
449 e820.map[x].addr = start;
450 e820.map[x].size = size;
451 e820.map[x].type = type;
452 e820.nr_map++;
453}
454
455/*
456 * Find the hole size (in bytes) in the memory range.
457 * @start: starting address of the memory range to scan
458 * @end: ending address of the memory range to scan
459 */
460unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
461{
462 unsigned long start_pfn = start >> PAGE_SHIFT;
463 unsigned long end_pfn = end >> PAGE_SHIFT;
464 unsigned long ei_startpfn, ei_endpfn, ram = 0;
465 int i;
466
467 for (i = 0; i < e820.nr_map; i++) {
468 if (e820_find_active_region(&e820.map[i],
469 start_pfn, end_pfn,
470 &ei_startpfn, &ei_endpfn))
471 ram += ei_endpfn - ei_startpfn;
472 }
473 return end - start - (ram << PAGE_SHIFT);
474}
475
476static void __init e820_print_map(char *who)
477{
478 int i;
479
480 for (i = 0; i < e820.nr_map; i++) {
481 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
482 (unsigned long long) e820.map[i].addr,
483 (unsigned long long)
484 (e820.map[i].addr + e820.map[i].size));
485 switch (e820.map[i].type) {
486 case E820_RAM:
487 printk(KERN_CONT "(usable)\n");
488 break;
489 case E820_RESERVED:
490 printk(KERN_CONT "(reserved)\n");
491 break;
492 case E820_ACPI:
493 printk(KERN_CONT "(ACPI data)\n");
494 break;
495 case E820_NVS:
496 printk(KERN_CONT "(ACPI NVS)\n");
497 break;
498 default:
499 printk(KERN_CONT "type %u\n", e820.map[i].type);
500 break;
501 }
502 }
503}
504
505/*
506 * Sanitize the BIOS e820 map.
507 *
508 * Some e820 responses include overlapping entries. The following
509 * replaces the original e820 map with a new one, removing overlaps.
510 *
511 */
512static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
513{
514 struct change_member {
515 struct e820entry *pbios; /* pointer to original bios entry */
516 unsigned long long addr; /* address for this change point */
517 };
518 static struct change_member change_point_list[2*E820MAX] __initdata;
519 static struct change_member *change_point[2*E820MAX] __initdata;
520 static struct e820entry *overlap_list[E820MAX] __initdata;
521 static struct e820entry new_bios[E820MAX] __initdata;
522 struct change_member *change_tmp;
523 unsigned long current_type, last_type;
524 unsigned long long last_addr;
525 int chgidx, still_changing;
526 int overlap_entries;
527 int new_bios_entry;
528 int old_nr, new_nr, chg_nr;
529 int i;
530
531 /*
532 Visually we're performing the following
533 (1,2,3,4 = memory types)...
534
535 Sample memory map (w/overlaps):
536 ____22__________________
537 ______________________4_
538 ____1111________________
539 _44_____________________
540 11111111________________
541 ____________________33__
542 ___________44___________
543 __________33333_________
544 ______________22________
545 ___________________2222_
546 _________111111111______
547 _____________________11_
548 _________________4______
549
550 Sanitized equivalent (no overlap):
551 1_______________________
552 _44_____________________
553 ___1____________________
554 ____22__________________
555 ______11________________
556 _________1______________
557 __________3_____________
558 ___________44___________
559 _____________33_________
560 _______________2________
561 ________________1_______
562 _________________4______
563 ___________________2____
564 ____________________33__
565 ______________________4_
566 */
567
568 /* if there's only one memory region, don't bother */
569 if (*pnr_map < 2)
570 return -1;
571
572 old_nr = *pnr_map;
573
574 /* bail out if we find any unreasonable addresses in bios map */
575 for (i = 0; i < old_nr; i++)
576 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
577 return -1;
578
579 /* create pointers for initial change-point information (for sorting) */
580 for (i = 0; i < 2 * old_nr; i++)
581 change_point[i] = &change_point_list[i];
582
583 /* record all known change-points (starting and ending addresses),
584 omitting those that are for empty memory regions */
585 chgidx = 0;
586 for (i = 0; i < old_nr; i++) {
587 if (biosmap[i].size != 0) {
588 change_point[chgidx]->addr = biosmap[i].addr;
589 change_point[chgidx++]->pbios = &biosmap[i];
590 change_point[chgidx]->addr = biosmap[i].addr +
591 biosmap[i].size;
592 change_point[chgidx++]->pbios = &biosmap[i];
593 }
594 }
595 chg_nr = chgidx;
596
597 /* sort change-point list by memory addresses (low -> high) */
598 still_changing = 1;
599 while (still_changing) {
600 still_changing = 0;
601 for (i = 1; i < chg_nr; i++) {
602 unsigned long long curaddr, lastaddr;
603 unsigned long long curpbaddr, lastpbaddr;
604
605 curaddr = change_point[i]->addr;
606 lastaddr = change_point[i - 1]->addr;
607 curpbaddr = change_point[i]->pbios->addr;
608 lastpbaddr = change_point[i - 1]->pbios->addr;
609
610 /*
611 * swap entries, when:
612 *
613 * curaddr > lastaddr or
614 * curaddr == lastaddr and curaddr == curpbaddr and
615 * lastaddr != lastpbaddr
616 */
617 if (curaddr < lastaddr ||
618 (curaddr == lastaddr && curaddr == curpbaddr &&
619 lastaddr != lastpbaddr)) {
620 change_tmp = change_point[i];
621 change_point[i] = change_point[i-1];
622 change_point[i-1] = change_tmp;
623 still_changing = 1;
624 }
625 }
626 }
627
628 /* create a new bios memory map, removing overlaps */
629 overlap_entries = 0; /* number of entries in the overlap table */
630 new_bios_entry = 0; /* index for creating new bios map entries */
631 last_type = 0; /* start with undefined memory type */
632 last_addr = 0; /* start with 0 as last starting address */
633
634 /* loop through change-points, determining affect on the new bios map */
635 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
636 /* keep track of all overlapping bios entries */
637 if (change_point[chgidx]->addr ==
638 change_point[chgidx]->pbios->addr) {
639 /*
640 * add map entry to overlap list (> 1 entry
641 * implies an overlap)
642 */
643 overlap_list[overlap_entries++] =
644 change_point[chgidx]->pbios;
645 } else {
646 /*
647 * remove entry from list (order independent,
648 * so swap with last)
649 */
650 for (i = 0; i < overlap_entries; i++) {
651 if (overlap_list[i] ==
652 change_point[chgidx]->pbios)
653 overlap_list[i] =
654 overlap_list[overlap_entries-1];
655 }
656 overlap_entries--;
657 }
658 /*
659 * if there are overlapping entries, decide which
660 * "type" to use (larger value takes precedence --
661 * 1=usable, 2,3,4,4+=unusable)
662 */
663 current_type = 0;
664 for (i = 0; i < overlap_entries; i++)
665 if (overlap_list[i]->type > current_type)
666 current_type = overlap_list[i]->type;
667 /*
668 * continue building up new bios map based on this
669 * information
670 */
671 if (current_type != last_type) {
672 if (last_type != 0) {
673 new_bios[new_bios_entry].size =
674 change_point[chgidx]->addr - last_addr;
675 /*
676 * move forward only if the new size
677 * was non-zero
678 */
679 if (new_bios[new_bios_entry].size != 0)
680 /*
681 * no more space left for new
682 * bios entries ?
683 */
684 if (++new_bios_entry >= E820MAX)
685 break;
686 }
687 if (current_type != 0) {
688 new_bios[new_bios_entry].addr =
689 change_point[chgidx]->addr;
690 new_bios[new_bios_entry].type = current_type;
691 last_addr = change_point[chgidx]->addr;
692 }
693 last_type = current_type;
694 }
695 }
696 /* retain count for new bios entries */
697 new_nr = new_bios_entry;
698
699 /* copy new bios mapping into original location */
700 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
701 *pnr_map = new_nr;
702
703 return 0;
704}
705
706/*
707 * Copy the BIOS e820 map into a safe place.
708 *
709 * Sanity-check it while we're at it..
710 *
711 * If we're lucky and live on a modern system, the setup code
712 * will have given us a memory map that we can use to properly
713 * set up memory. If we aren't, we'll fake a memory map.
714 */
715static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
716{
717 /* Only one memory region (or negative)? Ignore it */
718 if (nr_map < 2)
719 return -1;
720
721 do {
722 u64 start = biosmap->addr;
723 u64 size = biosmap->size;
724 u64 end = start + size;
725 u32 type = biosmap->type;
726
727 /* Overflow in 64 bits? Ignore the memory map. */
728 if (start > end)
729 return -1;
730
731 add_memory_region(start, size, type);
732 } while (biosmap++, --nr_map);
733 return 0;
734}
735
736static void early_panic(char *msg)
737{
738 early_printk(msg);
739 panic(msg);
740}
741
742/* We're not void only for x86 32-bit compat */
743char * __init machine_specific_memory_setup(void)
744{
745 char *who = "BIOS-e820";
746 /*
747 * Try to copy the BIOS-supplied E820-map.
748 *
749 * Otherwise fake a memory map; one section from 0k->640k,
750 * the next section from 1mb->appropriate_mem_k
751 */
752 sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
753 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
754 early_panic("Cannot find a valid memory map");
755 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
756 e820_print_map(who);
757
758 /* In case someone cares... */
759 return who;
760}
761
762static int __init parse_memopt(char *p)
763{
764 if (!p)
765 return -EINVAL;
766 end_user_pfn = memparse(p, &p);
767 end_user_pfn >>= PAGE_SHIFT;
768 return 0;
769}
770early_param("mem", parse_memopt);
771
772static int userdef __initdata;
773
774static int __init parse_memmap_opt(char *p)
775{
776 char *oldp;
777 unsigned long long start_at, mem_size;
778
779 if (!strcmp(p, "exactmap")) {
780#ifdef CONFIG_CRASH_DUMP
781 /*
782 * If we are doing a crash dump, we still need to know
783 * the real mem size before original memory map is
784 * reset.
785 */
786 e820_register_active_regions(0, 0, -1UL);
787 saved_max_pfn = e820_end_of_ram();
788 remove_all_active_ranges();
789#endif
790 max_pfn_mapped = 0;
791 e820.nr_map = 0;
792 userdef = 1;
793 return 0;
794 }
795
796 oldp = p;
797 mem_size = memparse(p, &p);
798 if (p == oldp)
799 return -EINVAL;
800
801 userdef = 1;
802 if (*p == '@') {
803 start_at = memparse(p+1, &p);
804 add_memory_region(start_at, mem_size, E820_RAM);
805 } else if (*p == '#') {
806 start_at = memparse(p+1, &p);
807 add_memory_region(start_at, mem_size, E820_ACPI);
808 } else if (*p == '$') {
809 start_at = memparse(p+1, &p);
810 add_memory_region(start_at, mem_size, E820_RESERVED);
811 } else {
812 end_user_pfn = (mem_size >> PAGE_SHIFT);
813 }
814 return *p == '\0' ? 0 : -EINVAL;
815}
816early_param("memmap", parse_memmap_opt);
817
818void __init finish_e820_parsing(void)
819{
820 if (userdef) {
821 char nr = e820.nr_map;
822
823 if (sanitize_e820_map(e820.map, &nr) < 0)
824 early_panic("Invalid user supplied memory map");
825 e820.nr_map = nr;
826
827 printk(KERN_INFO "user-defined physical RAM map:\n");
828 e820_print_map("user");
829 }
830}
831
832void __init update_memory_range(u64 start, u64 size, unsigned old_type,
833 unsigned new_type)
834{
835 int i;
836
837 BUG_ON(old_type == new_type);
838
839 for (i = 0; i < e820.nr_map; i++) {
840 struct e820entry *ei = &e820.map[i];
841 u64 final_start, final_end;
842 if (ei->type != old_type)
843 continue;
844 /* totally covered? */
845 if (ei->addr >= start && ei->size <= size) {
846 ei->type = new_type;
847 continue;
848 }
849 /* partially covered */
850 final_start = max(start, ei->addr);
851 final_end = min(start + size, ei->addr + ei->size);
852 if (final_start >= final_end)
853 continue;
854 add_memory_region(final_start, final_end - final_start,
855 new_type);
856 }
857}
858
859void __init update_e820(void)
860{
861 u8 nr_map;
862
863 nr_map = e820.nr_map;
864 if (sanitize_e820_map(e820.map, &nr_map))
865 return;
866 e820.nr_map = nr_map;
867 printk(KERN_INFO "modified physical RAM map:\n");
868 e820_print_map("modified");
869}
870
871unsigned long pci_mem_start = 0xaeedbabe;
872EXPORT_SYMBOL(pci_mem_start);
873
874/*
875 * Search for the biggest gap in the low 32 bits of the e820
876 * memory space. We pass this space to PCI to assign MMIO resources
877 * for hotplug or unconfigured devices in.
878 * Hopefully the BIOS let enough space left.
879 */
880__init void e820_setup_gap(void)
881{
882 unsigned long gapstart, gapsize, round;
883 unsigned long last;
884 int i;
885 int found = 0;
886
887 last = 0x100000000ull;
888 gapstart = 0x10000000;
889 gapsize = 0x400000;
890 i = e820.nr_map;
891 while (--i >= 0) {
892 unsigned long long start = e820.map[i].addr;
893 unsigned long long end = start + e820.map[i].size;
894
895 /*
896 * Since "last" is at most 4GB, we know we'll
897 * fit in 32 bits if this condition is true
898 */
899 if (last > end) {
900 unsigned long gap = last - end;
901
902 if (gap > gapsize) {
903 gapsize = gap;
904 gapstart = end;
905 found = 1;
906 }
907 }
908 if (start < last)
909 last = start;
910 }
911
912 if (!found) {
913 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
914 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
915 "address range\n"
916 KERN_ERR "PCI: Unassigned devices with 32bit resource "
917 "registers may break!\n");
918 }
919
920 /*
921 * See how much we want to round up: start off with
922 * rounding to the next 1MB area.
923 */
924 round = 0x100000;
925 while ((gapsize >> 4) > round)
926 round += round;
927 /* Fun with two's complement */
928 pci_mem_start = (gapstart + round) & -round;
929
930 printk(KERN_INFO
931 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
932 pci_mem_start, gapstart, gapsize);
933}
934
935int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
936{
937 int i;
938
939 if (slot < 0 || slot >= e820.nr_map)
940 return -1;
941 for (i = slot; i < e820.nr_map; i++) {
942 if (e820.map[i].type != E820_RAM)
943 continue;
944 break;
945 }
946 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
947 return -1;
948 *addr = e820.map[i].addr;
949 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
950 max_pfn << PAGE_SHIFT) - *addr;
951 return i + 1;
952}
diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
index 9f51e1ea9e82..4353cf5e6fac 100644
--- a/arch/x86/kernel/early-quirks.c
+++ b/arch/x86/kernel/early-quirks.c
@@ -16,10 +16,7 @@
16#include <asm/dma.h> 16#include <asm/dma.h>
17#include <asm/io_apic.h> 17#include <asm/io_apic.h>
18#include <asm/apic.h> 18#include <asm/apic.h>
19 19#include <asm/iommu.h>
20#ifdef CONFIG_GART_IOMMU
21#include <asm/gart.h>
22#endif
23 20
24static void __init fix_hypertransport_config(int num, int slot, int func) 21static void __init fix_hypertransport_config(int num, int slot, int func)
25{ 22{
@@ -50,7 +47,7 @@ static void __init fix_hypertransport_config(int num, int slot, int func)
50static void __init via_bugs(int num, int slot, int func) 47static void __init via_bugs(int num, int slot, int func)
51{ 48{
52#ifdef CONFIG_GART_IOMMU 49#ifdef CONFIG_GART_IOMMU
53 if ((end_pfn > MAX_DMA32_PFN || force_iommu) && 50 if ((max_pfn > MAX_DMA32_PFN || force_iommu) &&
54 !gart_iommu_aperture_allowed) { 51 !gart_iommu_aperture_allowed) {
55 printk(KERN_INFO 52 printk(KERN_INFO
56 "Looks like a VIA chipset. Disabling IOMMU." 53 "Looks like a VIA chipset. Disabling IOMMU."
@@ -98,17 +95,6 @@ static void __init nvidia_bugs(int num, int slot, int func)
98 95
99} 96}
100 97
101static void __init ati_bugs(int num, int slot, int func)
102{
103#ifdef CONFIG_X86_IO_APIC
104 if (timer_over_8254 == 1) {
105 timer_over_8254 = 0;
106 printk(KERN_INFO
107 "ATI board detected. Disabling timer routing over 8254.\n");
108 }
109#endif
110}
111
112#define QFLAG_APPLY_ONCE 0x1 98#define QFLAG_APPLY_ONCE 0x1
113#define QFLAG_APPLIED 0x2 99#define QFLAG_APPLIED 0x2
114#define QFLAG_DONE (QFLAG_APPLY_ONCE|QFLAG_APPLIED) 100#define QFLAG_DONE (QFLAG_APPLY_ONCE|QFLAG_APPLIED)
@@ -126,14 +112,23 @@ static struct chipset early_qrk[] __initdata = {
126 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, nvidia_bugs }, 112 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, nvidia_bugs },
127 { PCI_VENDOR_ID_VIA, PCI_ANY_ID, 113 { PCI_VENDOR_ID_VIA, PCI_ANY_ID,
128 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, via_bugs }, 114 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, via_bugs },
129 { PCI_VENDOR_ID_ATI, PCI_ANY_ID,
130 PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, ati_bugs },
131 { PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB, 115 { PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB,
132 PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, fix_hypertransport_config }, 116 PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, fix_hypertransport_config },
133 {} 117 {}
134}; 118};
135 119
136static void __init check_dev_quirk(int num, int slot, int func) 120/**
121 * check_dev_quirk - apply early quirks to a given PCI device
122 * @num: bus number
123 * @slot: slot number
124 * @func: PCI function
125 *
126 * Check the vendor & device ID against the early quirks table.
127 *
128 * If the device is single function, let early_quirks() know so we don't
129 * poke at this device again.
130 */
131static int __init check_dev_quirk(int num, int slot, int func)
137{ 132{
138 u16 class; 133 u16 class;
139 u16 vendor; 134 u16 vendor;
@@ -144,7 +139,7 @@ static void __init check_dev_quirk(int num, int slot, int func)
144 class = read_pci_config_16(num, slot, func, PCI_CLASS_DEVICE); 139 class = read_pci_config_16(num, slot, func, PCI_CLASS_DEVICE);
145 140
146 if (class == 0xffff) 141 if (class == 0xffff)
147 return; 142 return -1; /* no class, treat as single function */
148 143
149 vendor = read_pci_config_16(num, slot, func, PCI_VENDOR_ID); 144 vendor = read_pci_config_16(num, slot, func, PCI_VENDOR_ID);
150 145
@@ -167,7 +162,9 @@ static void __init check_dev_quirk(int num, int slot, int func)
167 type = read_pci_config_byte(num, slot, func, 162 type = read_pci_config_byte(num, slot, func,
168 PCI_HEADER_TYPE); 163 PCI_HEADER_TYPE);
169 if (!(type & 0x80)) 164 if (!(type & 0x80))
170 return; 165 return -1;
166
167 return 0;
171} 168}
172 169
173void __init early_quirks(void) 170void __init early_quirks(void)
@@ -180,6 +177,9 @@ void __init early_quirks(void)
180 /* Poor man's PCI discovery */ 177 /* Poor man's PCI discovery */
181 for (num = 0; num < 32; num++) 178 for (num = 0; num < 32; num++)
182 for (slot = 0; slot < 32; slot++) 179 for (slot = 0; slot < 32; slot++)
183 for (func = 0; func < 8; func++) 180 for (func = 0; func < 8; func++) {
184 check_dev_quirk(num, slot, func); 181 /* Only probe function 0 on single fn devices */
182 if (check_dev_quirk(num, slot, func))
183 break;
184 }
185} 185}
diff --git a/arch/x86/kernel/early_printk.c b/arch/x86/kernel/early_printk.c
index 643fd861b724..ff9e7350da54 100644
--- a/arch/x86/kernel/early_printk.c
+++ b/arch/x86/kernel/early_printk.c
@@ -196,7 +196,7 @@ static struct console simnow_console = {
196static struct console *early_console = &early_vga_console; 196static struct console *early_console = &early_vga_console;
197static int early_console_initialized; 197static int early_console_initialized;
198 198
199void early_printk(const char *fmt, ...) 199asmlinkage void early_printk(const char *fmt, ...)
200{ 200{
201 char buf[512]; 201 char buf[512];
202 int n; 202 int n;
diff --git a/arch/x86/kernel/efi.c b/arch/x86/kernel/efi.c
index 77d424cf68b3..06cc8d4254b1 100644
--- a/arch/x86/kernel/efi.c
+++ b/arch/x86/kernel/efi.c
@@ -64,6 +64,17 @@ static int __init setup_noefi(char *arg)
64} 64}
65early_param("noefi", setup_noefi); 65early_param("noefi", setup_noefi);
66 66
67int add_efi_memmap;
68EXPORT_SYMBOL(add_efi_memmap);
69
70static int __init setup_add_efi_memmap(char *arg)
71{
72 add_efi_memmap = 1;
73 return 0;
74}
75early_param("add_efi_memmap", setup_add_efi_memmap);
76
77
67static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc) 78static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
68{ 79{
69 return efi_call_virt2(get_time, tm, tc); 80 return efi_call_virt2(get_time, tm, tc);
@@ -213,6 +224,50 @@ unsigned long efi_get_time(void)
213 eft.minute, eft.second); 224 eft.minute, eft.second);
214} 225}
215 226
227/*
228 * Tell the kernel about the EFI memory map. This might include
229 * more than the max 128 entries that can fit in the e820 legacy
230 * (zeropage) memory map.
231 */
232
233static void __init do_add_efi_memmap(void)
234{
235 void *p;
236
237 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
238 efi_memory_desc_t *md = p;
239 unsigned long long start = md->phys_addr;
240 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
241 int e820_type;
242
243 if (md->attribute & EFI_MEMORY_WB)
244 e820_type = E820_RAM;
245 else
246 e820_type = E820_RESERVED;
247 e820_add_region(start, size, e820_type);
248 }
249 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
250}
251
252void __init efi_reserve_early(void)
253{
254 unsigned long pmap;
255
256#ifdef CONFIG_X86_32
257 pmap = boot_params.efi_info.efi_memmap;
258#else
259 pmap = (boot_params.efi_info.efi_memmap |
260 ((__u64)boot_params.efi_info.efi_memmap_hi<<32));
261#endif
262 memmap.phys_map = (void *)pmap;
263 memmap.nr_map = boot_params.efi_info.efi_memmap_size /
264 boot_params.efi_info.efi_memdesc_size;
265 memmap.desc_version = boot_params.efi_info.efi_memdesc_version;
266 memmap.desc_size = boot_params.efi_info.efi_memdesc_size;
267 reserve_early(pmap, pmap + memmap.nr_map * memmap.desc_size,
268 "EFI memmap");
269}
270
216#if EFI_DEBUG 271#if EFI_DEBUG
217static void __init print_efi_memmap(void) 272static void __init print_efi_memmap(void)
218{ 273{
@@ -244,19 +299,11 @@ void __init efi_init(void)
244 299
245#ifdef CONFIG_X86_32 300#ifdef CONFIG_X86_32
246 efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab; 301 efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
247 memmap.phys_map = (void *)boot_params.efi_info.efi_memmap;
248#else 302#else
249 efi_phys.systab = (efi_system_table_t *) 303 efi_phys.systab = (efi_system_table_t *)
250 (boot_params.efi_info.efi_systab | 304 (boot_params.efi_info.efi_systab |
251 ((__u64)boot_params.efi_info.efi_systab_hi<<32)); 305 ((__u64)boot_params.efi_info.efi_systab_hi<<32));
252 memmap.phys_map = (void *)
253 (boot_params.efi_info.efi_memmap |
254 ((__u64)boot_params.efi_info.efi_memmap_hi<<32));
255#endif 306#endif
256 memmap.nr_map = boot_params.efi_info.efi_memmap_size /
257 boot_params.efi_info.efi_memdesc_size;
258 memmap.desc_version = boot_params.efi_info.efi_memdesc_version;
259 memmap.desc_size = boot_params.efi_info.efi_memdesc_size;
260 307
261 efi.systab = early_ioremap((unsigned long)efi_phys.systab, 308 efi.systab = early_ioremap((unsigned long)efi_phys.systab,
262 sizeof(efi_system_table_t)); 309 sizeof(efi_system_table_t));
@@ -370,6 +417,8 @@ void __init efi_init(void)
370 if (memmap.desc_size != sizeof(efi_memory_desc_t)) 417 if (memmap.desc_size != sizeof(efi_memory_desc_t))
371 printk(KERN_WARNING "Kernel-defined memdesc" 418 printk(KERN_WARNING "Kernel-defined memdesc"
372 "doesn't match the one from EFI!\n"); 419 "doesn't match the one from EFI!\n");
420 if (add_efi_memmap)
421 do_add_efi_memmap();
373 422
374 /* Setup for EFI runtime service */ 423 /* Setup for EFI runtime service */
375 reboot_type = BOOT_EFI; 424 reboot_type = BOOT_EFI;
@@ -424,7 +473,7 @@ void __init efi_enter_virtual_mode(void)
424 size = md->num_pages << EFI_PAGE_SHIFT; 473 size = md->num_pages << EFI_PAGE_SHIFT;
425 end = md->phys_addr + size; 474 end = md->phys_addr + size;
426 475
427 if (PFN_UP(end) <= max_pfn_mapped) 476 if (PFN_UP(end) <= max_low_pfn_mapped)
428 va = __va(md->phys_addr); 477 va = __va(md->phys_addr);
429 else 478 else
430 va = efi_ioremap(md->phys_addr, size); 479 va = efi_ioremap(md->phys_addr, size);
diff --git a/arch/x86/kernel/efi_32.c b/arch/x86/kernel/efi_32.c
index 5d23d85624d4..4b63c8e1f13b 100644
--- a/arch/x86/kernel/efi_32.c
+++ b/arch/x86/kernel/efi_32.c
@@ -49,13 +49,13 @@ void efi_call_phys_prelog(void)
49 local_irq_save(efi_rt_eflags); 49 local_irq_save(efi_rt_eflags);
50 50
51 /* 51 /*
52 * If I don't have PSE, I should just duplicate two entries in page 52 * If I don't have PAE, I should just duplicate two entries in page
53 * directory. If I have PSE, I just need to duplicate one entry in 53 * directory. If I have PAE, I just need to duplicate one entry in
54 * page directory. 54 * page directory.
55 */ 55 */
56 cr4 = read_cr4(); 56 cr4 = read_cr4();
57 57
58 if (cr4 & X86_CR4_PSE) { 58 if (cr4 & X86_CR4_PAE) {
59 efi_bak_pg_dir_pointer[0].pgd = 59 efi_bak_pg_dir_pointer[0].pgd =
60 swapper_pg_dir[pgd_index(0)].pgd; 60 swapper_pg_dir[pgd_index(0)].pgd;
61 swapper_pg_dir[0].pgd = 61 swapper_pg_dir[0].pgd =
@@ -93,7 +93,7 @@ void efi_call_phys_epilog(void)
93 93
94 cr4 = read_cr4(); 94 cr4 = read_cr4();
95 95
96 if (cr4 & X86_CR4_PSE) { 96 if (cr4 & X86_CR4_PAE) {
97 swapper_pg_dir[pgd_index(0)].pgd = 97 swapper_pg_dir[pgd_index(0)].pgd =
98 efi_bak_pg_dir_pointer[0].pgd; 98 efi_bak_pg_dir_pointer[0].pgd;
99 } else { 99 } else {
diff --git a/arch/x86/kernel/efi_64.c b/arch/x86/kernel/efi_64.c
index d0060fdcccac..652c5287215f 100644
--- a/arch/x86/kernel/efi_64.c
+++ b/arch/x86/kernel/efi_64.c
@@ -97,13 +97,7 @@ void __init efi_call_phys_epilog(void)
97 early_runtime_code_mapping_set_exec(0); 97 early_runtime_code_mapping_set_exec(0);
98} 98}
99 99
100void __init efi_reserve_bootmem(void) 100void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size)
101{
102 reserve_bootmem_generic((unsigned long)memmap.phys_map,
103 memmap.nr_map * memmap.desc_size);
104}
105
106void __iomem * __init efi_ioremap(unsigned long phys_addr, unsigned long size)
107{ 101{
108 static unsigned pages_mapped __initdata; 102 static unsigned pages_mapped __initdata;
109 unsigned i, pages; 103 unsigned i, pages;
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index c778e4fa55a2..109792bc7cfa 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -51,14 +51,25 @@
51#include <asm/percpu.h> 51#include <asm/percpu.h>
52#include <asm/dwarf2.h> 52#include <asm/dwarf2.h>
53#include <asm/processor-flags.h> 53#include <asm/processor-flags.h>
54#include "irq_vectors.h" 54#include <asm/ftrace.h>
55#include <asm/irq_vectors.h>
56
57/* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58#include <linux/elf-em.h>
59#define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
60#define __AUDIT_ARCH_LE 0x40000000
61
62#ifndef CONFIG_AUDITSYSCALL
63#define sysenter_audit syscall_trace_entry
64#define sysexit_audit syscall_exit_work
65#endif
55 66
56/* 67/*
57 * We use macros for low-level operations which need to be overridden 68 * We use macros for low-level operations which need to be overridden
58 * for paravirtualization. The following will never clobber any registers: 69 * for paravirtualization. The following will never clobber any registers:
59 * INTERRUPT_RETURN (aka. "iret") 70 * INTERRUPT_RETURN (aka. "iret")
60 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax") 71 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
61 * ENABLE_INTERRUPTS_SYSCALL_RET (aka "sti; sysexit"). 72 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
62 * 73 *
63 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must 74 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
64 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY). 75 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
@@ -331,8 +342,9 @@ sysenter_past_esp:
331 GET_THREAD_INFO(%ebp) 342 GET_THREAD_INFO(%ebp)
332 343
333 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ 344 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
334 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) 345 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
335 jnz syscall_trace_entry 346 jnz sysenter_audit
347sysenter_do_call:
336 cmpl $(nr_syscalls), %eax 348 cmpl $(nr_syscalls), %eax
337 jae syscall_badsys 349 jae syscall_badsys
338 call *sys_call_table(,%eax,4) 350 call *sys_call_table(,%eax,4)
@@ -342,14 +354,54 @@ sysenter_past_esp:
342 TRACE_IRQS_OFF 354 TRACE_IRQS_OFF
343 movl TI_flags(%ebp), %ecx 355 movl TI_flags(%ebp), %ecx
344 testw $_TIF_ALLWORK_MASK, %cx 356 testw $_TIF_ALLWORK_MASK, %cx
345 jne syscall_exit_work 357 jne sysexit_audit
358sysenter_exit:
346/* if something modifies registers it must also disable sysexit */ 359/* if something modifies registers it must also disable sysexit */
347 movl PT_EIP(%esp), %edx 360 movl PT_EIP(%esp), %edx
348 movl PT_OLDESP(%esp), %ecx 361 movl PT_OLDESP(%esp), %ecx
349 xorl %ebp,%ebp 362 xorl %ebp,%ebp
350 TRACE_IRQS_ON 363 TRACE_IRQS_ON
3511: mov PT_FS(%esp), %fs 3641: mov PT_FS(%esp), %fs
352 ENABLE_INTERRUPTS_SYSCALL_RET 365 ENABLE_INTERRUPTS_SYSEXIT
366
367#ifdef CONFIG_AUDITSYSCALL
368sysenter_audit:
369 testw $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
370 jnz syscall_trace_entry
371 addl $4,%esp
372 CFI_ADJUST_CFA_OFFSET -4
373 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
374 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
375 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
376 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
377 movl %eax,%edx /* 2nd arg: syscall number */
378 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
379 call audit_syscall_entry
380 pushl %ebx
381 CFI_ADJUST_CFA_OFFSET 4
382 movl PT_EAX(%esp),%eax /* reload syscall number */
383 jmp sysenter_do_call
384
385sysexit_audit:
386 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
387 jne syscall_exit_work
388 TRACE_IRQS_ON
389 ENABLE_INTERRUPTS(CLBR_ANY)
390 movl %eax,%edx /* second arg, syscall return value */
391 cmpl $0,%eax /* is it < 0? */
392 setl %al /* 1 if so, 0 if not */
393 movzbl %al,%eax /* zero-extend that */
394 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
395 call audit_syscall_exit
396 DISABLE_INTERRUPTS(CLBR_ANY)
397 TRACE_IRQS_OFF
398 movl TI_flags(%ebp), %ecx
399 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
400 jne syscall_exit_work
401 movl PT_EAX(%esp),%eax /* reload syscall return value */
402 jmp sysenter_exit
403#endif
404
353 CFI_ENDPROC 405 CFI_ENDPROC
354.pushsection .fixup,"ax" 406.pushsection .fixup,"ax"
3552: movl $0,PT_FS(%esp) 4072: movl $0,PT_FS(%esp)
@@ -369,7 +421,7 @@ ENTRY(system_call)
369 GET_THREAD_INFO(%ebp) 421 GET_THREAD_INFO(%ebp)
370 # system call tracing in operation / emulation 422 # system call tracing in operation / emulation
371 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ 423 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
372 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) 424 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
373 jnz syscall_trace_entry 425 jnz syscall_trace_entry
374 cmpl $(nr_syscalls), %eax 426 cmpl $(nr_syscalls), %eax
375 jae syscall_badsys 427 jae syscall_badsys
@@ -382,10 +434,6 @@ syscall_exit:
382 # setting need_resched or sigpending 434 # setting need_resched or sigpending
383 # between sampling and the iret 435 # between sampling and the iret
384 TRACE_IRQS_OFF 436 TRACE_IRQS_OFF
385 testl $X86_EFLAGS_TF,PT_EFLAGS(%esp) # If tracing set singlestep flag on exit
386 jz no_singlestep
387 orl $_TIF_SINGLESTEP,TI_flags(%ebp)
388no_singlestep:
389 movl TI_flags(%ebp), %ecx 437 movl TI_flags(%ebp), %ecx
390 testw $_TIF_ALLWORK_MASK, %cx # current->work 438 testw $_TIF_ALLWORK_MASK, %cx # current->work
391 jne syscall_exit_work 439 jne syscall_exit_work
@@ -513,12 +561,8 @@ END(work_pending)
513syscall_trace_entry: 561syscall_trace_entry:
514 movl $-ENOSYS,PT_EAX(%esp) 562 movl $-ENOSYS,PT_EAX(%esp)
515 movl %esp, %eax 563 movl %esp, %eax
516 xorl %edx,%edx 564 call syscall_trace_enter
517 call do_syscall_trace 565 /* What it returned is what we'll actually use. */
518 cmpl $0, %eax
519 jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU,
520 # so must skip actual syscall
521 movl PT_ORIG_EAX(%esp), %eax
522 cmpl $(nr_syscalls), %eax 566 cmpl $(nr_syscalls), %eax
523 jnae syscall_call 567 jnae syscall_call
524 jmp syscall_exit 568 jmp syscall_exit
@@ -527,14 +571,13 @@ END(syscall_trace_entry)
527 # perform syscall exit tracing 571 # perform syscall exit tracing
528 ALIGN 572 ALIGN
529syscall_exit_work: 573syscall_exit_work:
530 testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl 574 testb $_TIF_WORK_SYSCALL_EXIT, %cl
531 jz work_pending 575 jz work_pending
532 TRACE_IRQS_ON 576 TRACE_IRQS_ON
533 ENABLE_INTERRUPTS(CLBR_ANY) # could let do_syscall_trace() call 577 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
534 # schedule() instead 578 # schedule() instead
535 movl %esp, %eax 579 movl %esp, %eax
536 movl $1, %edx 580 call syscall_trace_leave
537 call do_syscall_trace
538 jmp resume_userspace 581 jmp resume_userspace
539END(syscall_exit_work) 582END(syscall_exit_work)
540 CFI_ENDPROC 583 CFI_ENDPROC
@@ -874,10 +917,10 @@ ENTRY(native_iret)
874.previous 917.previous
875END(native_iret) 918END(native_iret)
876 919
877ENTRY(native_irq_enable_syscall_ret) 920ENTRY(native_irq_enable_sysexit)
878 sti 921 sti
879 sysexit 922 sysexit
880END(native_irq_enable_syscall_ret) 923END(native_irq_enable_sysexit)
881#endif 924#endif
882 925
883KPROBE_ENTRY(int3) 926KPROBE_ENTRY(int3)
@@ -1023,7 +1066,9 @@ ENDPROC(kernel_thread_helper)
1023ENTRY(xen_sysenter_target) 1066ENTRY(xen_sysenter_target)
1024 RING0_INT_FRAME 1067 RING0_INT_FRAME
1025 addl $5*4, %esp /* remove xen-provided frame */ 1068 addl $5*4, %esp /* remove xen-provided frame */
1069 CFI_ADJUST_CFA_OFFSET -5*4
1026 jmp sysenter_past_esp 1070 jmp sysenter_past_esp
1071 CFI_ENDPROC
1027 1072
1028ENTRY(xen_hypervisor_callback) 1073ENTRY(xen_hypervisor_callback)
1029 CFI_STARTPROC 1074 CFI_STARTPROC
@@ -1110,6 +1155,77 @@ ENDPROC(xen_failsafe_callback)
1110 1155
1111#endif /* CONFIG_XEN */ 1156#endif /* CONFIG_XEN */
1112 1157
1158#ifdef CONFIG_FTRACE
1159#ifdef CONFIG_DYNAMIC_FTRACE
1160
1161ENTRY(mcount)
1162 pushl %eax
1163 pushl %ecx
1164 pushl %edx
1165 movl 0xc(%esp), %eax
1166 subl $MCOUNT_INSN_SIZE, %eax
1167
1168.globl mcount_call
1169mcount_call:
1170 call ftrace_stub
1171
1172 popl %edx
1173 popl %ecx
1174 popl %eax
1175
1176 ret
1177END(mcount)
1178
1179ENTRY(ftrace_caller)
1180 pushl %eax
1181 pushl %ecx
1182 pushl %edx
1183 movl 0xc(%esp), %eax
1184 movl 0x4(%ebp), %edx
1185 subl $MCOUNT_INSN_SIZE, %eax
1186
1187.globl ftrace_call
1188ftrace_call:
1189 call ftrace_stub
1190
1191 popl %edx
1192 popl %ecx
1193 popl %eax
1194
1195.globl ftrace_stub
1196ftrace_stub:
1197 ret
1198END(ftrace_caller)
1199
1200#else /* ! CONFIG_DYNAMIC_FTRACE */
1201
1202ENTRY(mcount)
1203 cmpl $ftrace_stub, ftrace_trace_function
1204 jnz trace
1205.globl ftrace_stub
1206ftrace_stub:
1207 ret
1208
1209 /* taken from glibc */
1210trace:
1211 pushl %eax
1212 pushl %ecx
1213 pushl %edx
1214 movl 0xc(%esp), %eax
1215 movl 0x4(%ebp), %edx
1216 subl $MCOUNT_INSN_SIZE, %eax
1217
1218 call *ftrace_trace_function
1219
1220 popl %edx
1221 popl %ecx
1222 popl %eax
1223
1224 jmp ftrace_stub
1225END(mcount)
1226#endif /* CONFIG_DYNAMIC_FTRACE */
1227#endif /* CONFIG_FTRACE */
1228
1113.section .rodata,"a" 1229.section .rodata,"a"
1114#include "syscall_table_32.S" 1230#include "syscall_table_32.S"
1115 1231
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 556a8df522a7..89434d439605 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -51,16 +51,127 @@
51#include <asm/page.h> 51#include <asm/page.h>
52#include <asm/irqflags.h> 52#include <asm/irqflags.h>
53#include <asm/paravirt.h> 53#include <asm/paravirt.h>
54#include <asm/ftrace.h>
55
56/* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
57#include <linux/elf-em.h>
58#define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
59#define __AUDIT_ARCH_64BIT 0x80000000
60#define __AUDIT_ARCH_LE 0x40000000
54 61
55 .code64 62 .code64
56 63
64#ifdef CONFIG_FTRACE
65#ifdef CONFIG_DYNAMIC_FTRACE
66ENTRY(mcount)
67
68 subq $0x38, %rsp
69 movq %rax, (%rsp)
70 movq %rcx, 8(%rsp)
71 movq %rdx, 16(%rsp)
72 movq %rsi, 24(%rsp)
73 movq %rdi, 32(%rsp)
74 movq %r8, 40(%rsp)
75 movq %r9, 48(%rsp)
76
77 movq 0x38(%rsp), %rdi
78 subq $MCOUNT_INSN_SIZE, %rdi
79
80.globl mcount_call
81mcount_call:
82 call ftrace_stub
83
84 movq 48(%rsp), %r9
85 movq 40(%rsp), %r8
86 movq 32(%rsp), %rdi
87 movq 24(%rsp), %rsi
88 movq 16(%rsp), %rdx
89 movq 8(%rsp), %rcx
90 movq (%rsp), %rax
91 addq $0x38, %rsp
92
93 retq
94END(mcount)
95
96ENTRY(ftrace_caller)
97
98 /* taken from glibc */
99 subq $0x38, %rsp
100 movq %rax, (%rsp)
101 movq %rcx, 8(%rsp)
102 movq %rdx, 16(%rsp)
103 movq %rsi, 24(%rsp)
104 movq %rdi, 32(%rsp)
105 movq %r8, 40(%rsp)
106 movq %r9, 48(%rsp)
107
108 movq 0x38(%rsp), %rdi
109 movq 8(%rbp), %rsi
110 subq $MCOUNT_INSN_SIZE, %rdi
111
112.globl ftrace_call
113ftrace_call:
114 call ftrace_stub
115
116 movq 48(%rsp), %r9
117 movq 40(%rsp), %r8
118 movq 32(%rsp), %rdi
119 movq 24(%rsp), %rsi
120 movq 16(%rsp), %rdx
121 movq 8(%rsp), %rcx
122 movq (%rsp), %rax
123 addq $0x38, %rsp
124
125.globl ftrace_stub
126ftrace_stub:
127 retq
128END(ftrace_caller)
129
130#else /* ! CONFIG_DYNAMIC_FTRACE */
131ENTRY(mcount)
132 cmpq $ftrace_stub, ftrace_trace_function
133 jnz trace
134.globl ftrace_stub
135ftrace_stub:
136 retq
137
138trace:
139 /* taken from glibc */
140 subq $0x38, %rsp
141 movq %rax, (%rsp)
142 movq %rcx, 8(%rsp)
143 movq %rdx, 16(%rsp)
144 movq %rsi, 24(%rsp)
145 movq %rdi, 32(%rsp)
146 movq %r8, 40(%rsp)
147 movq %r9, 48(%rsp)
148
149 movq 0x38(%rsp), %rdi
150 movq 8(%rbp), %rsi
151 subq $MCOUNT_INSN_SIZE, %rdi
152
153 call *ftrace_trace_function
154
155 movq 48(%rsp), %r9
156 movq 40(%rsp), %r8
157 movq 32(%rsp), %rdi
158 movq 24(%rsp), %rsi
159 movq 16(%rsp), %rdx
160 movq 8(%rsp), %rcx
161 movq (%rsp), %rax
162 addq $0x38, %rsp
163
164 jmp ftrace_stub
165END(mcount)
166#endif /* CONFIG_DYNAMIC_FTRACE */
167#endif /* CONFIG_FTRACE */
168
57#ifndef CONFIG_PREEMPT 169#ifndef CONFIG_PREEMPT
58#define retint_kernel retint_restore_args 170#define retint_kernel retint_restore_args
59#endif 171#endif
60 172
61#ifdef CONFIG_PARAVIRT 173#ifdef CONFIG_PARAVIRT
62ENTRY(native_irq_enable_syscall_ret) 174ENTRY(native_usergs_sysret64)
63 movq %gs:pda_oldrsp,%rsp
64 swapgs 175 swapgs
65 sysretq 176 sysretq
66#endif /* CONFIG_PARAVIRT */ 177#endif /* CONFIG_PARAVIRT */
@@ -104,7 +215,7 @@ ENTRY(native_irq_enable_syscall_ret)
104 .macro FAKE_STACK_FRAME child_rip 215 .macro FAKE_STACK_FRAME child_rip
105 /* push in order ss, rsp, eflags, cs, rip */ 216 /* push in order ss, rsp, eflags, cs, rip */
106 xorl %eax, %eax 217 xorl %eax, %eax
107 pushq %rax /* ss */ 218 pushq $__KERNEL_DS /* ss */
108 CFI_ADJUST_CFA_OFFSET 8 219 CFI_ADJUST_CFA_OFFSET 8
109 /*CFI_REL_OFFSET ss,0*/ 220 /*CFI_REL_OFFSET ss,0*/
110 pushq %rax /* rsp */ 221 pushq %rax /* rsp */
@@ -169,13 +280,13 @@ ENTRY(ret_from_fork)
169 CFI_ADJUST_CFA_OFFSET -4 280 CFI_ADJUST_CFA_OFFSET -4
170 call schedule_tail 281 call schedule_tail
171 GET_THREAD_INFO(%rcx) 282 GET_THREAD_INFO(%rcx)
172 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),threadinfo_flags(%rcx) 283 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
173 jnz rff_trace 284 jnz rff_trace
174rff_action: 285rff_action:
175 RESTORE_REST 286 RESTORE_REST
176 testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread? 287 testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread?
177 je int_ret_from_sys_call 288 je int_ret_from_sys_call
178 testl $_TIF_IA32,threadinfo_flags(%rcx) 289 testl $_TIF_IA32,TI_flags(%rcx)
179 jnz int_ret_from_sys_call 290 jnz int_ret_from_sys_call
180 RESTORE_TOP_OF_STACK %rdi,ARGOFFSET 291 RESTORE_TOP_OF_STACK %rdi,ARGOFFSET
181 jmp ret_from_sys_call 292 jmp ret_from_sys_call
@@ -244,8 +355,9 @@ ENTRY(system_call_after_swapgs)
244 movq %rcx,RIP-ARGOFFSET(%rsp) 355 movq %rcx,RIP-ARGOFFSET(%rsp)
245 CFI_REL_OFFSET rip,RIP-ARGOFFSET 356 CFI_REL_OFFSET rip,RIP-ARGOFFSET
246 GET_THREAD_INFO(%rcx) 357 GET_THREAD_INFO(%rcx)
247 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP),threadinfo_flags(%rcx) 358 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
248 jnz tracesys 359 jnz tracesys
360system_call_fastpath:
249 cmpq $__NR_syscall_max,%rax 361 cmpq $__NR_syscall_max,%rax
250 ja badsys 362 ja badsys
251 movq %r10,%rcx 363 movq %r10,%rcx
@@ -263,7 +375,7 @@ sysret_check:
263 GET_THREAD_INFO(%rcx) 375 GET_THREAD_INFO(%rcx)
264 DISABLE_INTERRUPTS(CLBR_NONE) 376 DISABLE_INTERRUPTS(CLBR_NONE)
265 TRACE_IRQS_OFF 377 TRACE_IRQS_OFF
266 movl threadinfo_flags(%rcx),%edx 378 movl TI_flags(%rcx),%edx
267 andl %edi,%edx 379 andl %edi,%edx
268 jnz sysret_careful 380 jnz sysret_careful
269 CFI_REMEMBER_STATE 381 CFI_REMEMBER_STATE
@@ -275,7 +387,8 @@ sysret_check:
275 CFI_REGISTER rip,rcx 387 CFI_REGISTER rip,rcx
276 RESTORE_ARGS 0,-ARG_SKIP,1 388 RESTORE_ARGS 0,-ARG_SKIP,1
277 /*CFI_REGISTER rflags,r11*/ 389 /*CFI_REGISTER rflags,r11*/
278 ENABLE_INTERRUPTS_SYSCALL_RET 390 movq %gs:pda_oldrsp, %rsp
391 USERGS_SYSRET64
279 392
280 CFI_RESTORE_STATE 393 CFI_RESTORE_STATE
281 /* Handle reschedules */ 394 /* Handle reschedules */
@@ -296,16 +409,16 @@ sysret_careful:
296sysret_signal: 409sysret_signal:
297 TRACE_IRQS_ON 410 TRACE_IRQS_ON
298 ENABLE_INTERRUPTS(CLBR_NONE) 411 ENABLE_INTERRUPTS(CLBR_NONE)
299 testl $_TIF_DO_NOTIFY_MASK,%edx 412#ifdef CONFIG_AUDITSYSCALL
300 jz 1f 413 bt $TIF_SYSCALL_AUDIT,%edx
301 414 jc sysret_audit
302 /* Really a signal */ 415#endif
303 /* edx: work flags (arg3) */ 416 /* edx: work flags (arg3) */
304 leaq do_notify_resume(%rip),%rax 417 leaq do_notify_resume(%rip),%rax
305 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1 418 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
306 xorl %esi,%esi # oldset -> arg2 419 xorl %esi,%esi # oldset -> arg2
307 call ptregscall_common 420 call ptregscall_common
3081: movl $_TIF_NEED_RESCHED,%edi 421 movl $_TIF_WORK_MASK,%edi
309 /* Use IRET because user could have changed frame. This 422 /* Use IRET because user could have changed frame. This
310 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */ 423 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
311 DISABLE_INTERRUPTS(CLBR_NONE) 424 DISABLE_INTERRUPTS(CLBR_NONE)
@@ -316,14 +429,56 @@ badsys:
316 movq $-ENOSYS,RAX-ARGOFFSET(%rsp) 429 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
317 jmp ret_from_sys_call 430 jmp ret_from_sys_call
318 431
432#ifdef CONFIG_AUDITSYSCALL
433 /*
434 * Fast path for syscall audit without full syscall trace.
435 * We just call audit_syscall_entry() directly, and then
436 * jump back to the normal fast path.
437 */
438auditsys:
439 movq %r10,%r9 /* 6th arg: 4th syscall arg */
440 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
441 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
442 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
443 movq %rax,%rsi /* 2nd arg: syscall number */
444 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
445 call audit_syscall_entry
446 LOAD_ARGS 0 /* reload call-clobbered registers */
447 jmp system_call_fastpath
448
449 /*
450 * Return fast path for syscall audit. Call audit_syscall_exit()
451 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
452 * masked off.
453 */
454sysret_audit:
455 movq %rax,%rsi /* second arg, syscall return value */
456 cmpq $0,%rax /* is it < 0? */
457 setl %al /* 1 if so, 0 if not */
458 movzbl %al,%edi /* zero-extend that into %edi */
459 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
460 call audit_syscall_exit
461 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
462 jmp sysret_check
463#endif /* CONFIG_AUDITSYSCALL */
464
319 /* Do syscall tracing */ 465 /* Do syscall tracing */
320tracesys: 466tracesys:
467#ifdef CONFIG_AUDITSYSCALL
468 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
469 jz auditsys
470#endif
321 SAVE_REST 471 SAVE_REST
322 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */ 472 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
323 FIXUP_TOP_OF_STACK %rdi 473 FIXUP_TOP_OF_STACK %rdi
324 movq %rsp,%rdi 474 movq %rsp,%rdi
325 call syscall_trace_enter 475 call syscall_trace_enter
326 LOAD_ARGS ARGOFFSET /* reload args from stack in case ptrace changed it */ 476 /*
477 * Reload arg registers from stack in case ptrace changed them.
478 * We don't reload %rax because syscall_trace_enter() returned
479 * the value it wants us to use in the table lookup.
480 */
481 LOAD_ARGS ARGOFFSET, 1
327 RESTORE_REST 482 RESTORE_REST
328 cmpq $__NR_syscall_max,%rax 483 cmpq $__NR_syscall_max,%rax
329 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */ 484 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
@@ -337,6 +492,7 @@ tracesys:
337 * Has correct top of stack, but partial stack frame. 492 * Has correct top of stack, but partial stack frame.
338 */ 493 */
339 .globl int_ret_from_sys_call 494 .globl int_ret_from_sys_call
495 .globl int_with_check
340int_ret_from_sys_call: 496int_ret_from_sys_call:
341 DISABLE_INTERRUPTS(CLBR_NONE) 497 DISABLE_INTERRUPTS(CLBR_NONE)
342 TRACE_IRQS_OFF 498 TRACE_IRQS_OFF
@@ -347,10 +503,10 @@ int_ret_from_sys_call:
347int_with_check: 503int_with_check:
348 LOCKDEP_SYS_EXIT_IRQ 504 LOCKDEP_SYS_EXIT_IRQ
349 GET_THREAD_INFO(%rcx) 505 GET_THREAD_INFO(%rcx)
350 movl threadinfo_flags(%rcx),%edx 506 movl TI_flags(%rcx),%edx
351 andl %edi,%edx 507 andl %edi,%edx
352 jnz int_careful 508 jnz int_careful
353 andl $~TS_COMPAT,threadinfo_status(%rcx) 509 andl $~TS_COMPAT,TI_status(%rcx)
354 jmp retint_swapgs 510 jmp retint_swapgs
355 511
356 /* Either reschedule or signal or syscall exit tracking needed. */ 512 /* Either reschedule or signal or syscall exit tracking needed. */
@@ -376,7 +532,7 @@ int_very_careful:
376 ENABLE_INTERRUPTS(CLBR_NONE) 532 ENABLE_INTERRUPTS(CLBR_NONE)
377 SAVE_REST 533 SAVE_REST
378 /* Check for syscall exit trace */ 534 /* Check for syscall exit trace */
379 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx 535 testl $_TIF_WORK_SYSCALL_EXIT,%edx
380 jz int_signal 536 jz int_signal
381 pushq %rdi 537 pushq %rdi
382 CFI_ADJUST_CFA_OFFSET 8 538 CFI_ADJUST_CFA_OFFSET 8
@@ -384,7 +540,7 @@ int_very_careful:
384 call syscall_trace_leave 540 call syscall_trace_leave
385 popq %rdi 541 popq %rdi
386 CFI_ADJUST_CFA_OFFSET -8 542 CFI_ADJUST_CFA_OFFSET -8
387 andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi 543 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
388 jmp int_restore_rest 544 jmp int_restore_rest
389 545
390int_signal: 546int_signal:
@@ -393,7 +549,7 @@ int_signal:
393 movq %rsp,%rdi # &ptregs -> arg1 549 movq %rsp,%rdi # &ptregs -> arg1
394 xorl %esi,%esi # oldset -> arg2 550 xorl %esi,%esi # oldset -> arg2
395 call do_notify_resume 551 call do_notify_resume
3961: movl $_TIF_NEED_RESCHED,%edi 5521: movl $_TIF_WORK_MASK,%edi
397int_restore_rest: 553int_restore_rest:
398 RESTORE_REST 554 RESTORE_REST
399 DISABLE_INTERRUPTS(CLBR_NONE) 555 DISABLE_INTERRUPTS(CLBR_NONE)
@@ -420,7 +576,6 @@ END(\label)
420 PTREGSCALL stub_clone, sys_clone, %r8 576 PTREGSCALL stub_clone, sys_clone, %r8
421 PTREGSCALL stub_fork, sys_fork, %rdi 577 PTREGSCALL stub_fork, sys_fork, %rdi
422 PTREGSCALL stub_vfork, sys_vfork, %rdi 578 PTREGSCALL stub_vfork, sys_vfork, %rdi
423 PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx
424 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx 579 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
425 PTREGSCALL stub_iopl, sys_iopl, %rsi 580 PTREGSCALL stub_iopl, sys_iopl, %rsi
426 581
@@ -559,7 +714,7 @@ retint_with_reschedule:
559 movl $_TIF_WORK_MASK,%edi 714 movl $_TIF_WORK_MASK,%edi
560retint_check: 715retint_check:
561 LOCKDEP_SYS_EXIT_IRQ 716 LOCKDEP_SYS_EXIT_IRQ
562 movl threadinfo_flags(%rcx),%edx 717 movl TI_flags(%rcx),%edx
563 andl %edi,%edx 718 andl %edi,%edx
564 CFI_REMEMBER_STATE 719 CFI_REMEMBER_STATE
565 jnz retint_careful 720 jnz retint_careful
@@ -647,17 +802,16 @@ retint_signal:
647 RESTORE_REST 802 RESTORE_REST
648 DISABLE_INTERRUPTS(CLBR_NONE) 803 DISABLE_INTERRUPTS(CLBR_NONE)
649 TRACE_IRQS_OFF 804 TRACE_IRQS_OFF
650 movl $_TIF_NEED_RESCHED,%edi
651 GET_THREAD_INFO(%rcx) 805 GET_THREAD_INFO(%rcx)
652 jmp retint_check 806 jmp retint_with_reschedule
653 807
654#ifdef CONFIG_PREEMPT 808#ifdef CONFIG_PREEMPT
655 /* Returning to kernel space. Check if we need preemption */ 809 /* Returning to kernel space. Check if we need preemption */
656 /* rcx: threadinfo. interrupts off. */ 810 /* rcx: threadinfo. interrupts off. */
657ENTRY(retint_kernel) 811ENTRY(retint_kernel)
658 cmpl $0,threadinfo_preempt_count(%rcx) 812 cmpl $0,TI_preempt_count(%rcx)
659 jnz retint_restore_args 813 jnz retint_restore_args
660 bt $TIF_NEED_RESCHED,threadinfo_flags(%rcx) 814 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
661 jnc retint_restore_args 815 jnc retint_restore_args
662 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */ 816 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
663 jnc retint_restore_args 817 jnc retint_restore_args
@@ -711,6 +865,9 @@ END(invalidate_interrupt\num)
711ENTRY(call_function_interrupt) 865ENTRY(call_function_interrupt)
712 apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt 866 apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
713END(call_function_interrupt) 867END(call_function_interrupt)
868ENTRY(call_function_single_interrupt)
869 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR,smp_call_function_single_interrupt
870END(call_function_single_interrupt)
714ENTRY(irq_move_cleanup_interrupt) 871ENTRY(irq_move_cleanup_interrupt)
715 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt 872 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
716END(irq_move_cleanup_interrupt) 873END(irq_move_cleanup_interrupt)
@@ -720,6 +877,10 @@ ENTRY(apic_timer_interrupt)
720 apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt 877 apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
721END(apic_timer_interrupt) 878END(apic_timer_interrupt)
722 879
880ENTRY(uv_bau_message_intr1)
881 apicinterrupt 220,uv_bau_message_interrupt
882END(uv_bau_message_intr1)
883
723ENTRY(error_interrupt) 884ENTRY(error_interrupt)
724 apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt 885 apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
725END(error_interrupt) 886END(error_interrupt)
@@ -733,6 +894,7 @@ END(spurious_interrupt)
733 */ 894 */
734 .macro zeroentry sym 895 .macro zeroentry sym
735 INTR_FRAME 896 INTR_FRAME
897 PARAVIRT_ADJUST_EXCEPTION_FRAME
736 pushq $0 /* push error code/oldrax */ 898 pushq $0 /* push error code/oldrax */
737 CFI_ADJUST_CFA_OFFSET 8 899 CFI_ADJUST_CFA_OFFSET 8
738 pushq %rax /* push real oldrax to the rdi slot */ 900 pushq %rax /* push real oldrax to the rdi slot */
@@ -745,6 +907,7 @@ END(spurious_interrupt)
745 907
746 .macro errorentry sym 908 .macro errorentry sym
747 XCPT_FRAME 909 XCPT_FRAME
910 PARAVIRT_ADJUST_EXCEPTION_FRAME
748 pushq %rax 911 pushq %rax
749 CFI_ADJUST_CFA_OFFSET 8 912 CFI_ADJUST_CFA_OFFSET 8
750 CFI_REL_OFFSET rax,0 913 CFI_REL_OFFSET rax,0
@@ -814,7 +977,7 @@ paranoid_restore\trace:
814 jmp irq_return 977 jmp irq_return
815paranoid_userspace\trace: 978paranoid_userspace\trace:
816 GET_THREAD_INFO(%rcx) 979 GET_THREAD_INFO(%rcx)
817 movl threadinfo_flags(%rcx),%ebx 980 movl TI_flags(%rcx),%ebx
818 andl $_TIF_WORK_MASK,%ebx 981 andl $_TIF_WORK_MASK,%ebx
819 jz paranoid_swapgs\trace 982 jz paranoid_swapgs\trace
820 movq %rsp,%rdi /* &pt_regs */ 983 movq %rsp,%rdi /* &pt_regs */
@@ -912,7 +1075,7 @@ error_exit:
912 testl %eax,%eax 1075 testl %eax,%eax
913 jne retint_kernel 1076 jne retint_kernel
914 LOCKDEP_SYS_EXIT_IRQ 1077 LOCKDEP_SYS_EXIT_IRQ
915 movl threadinfo_flags(%rcx),%edx 1078 movl TI_flags(%rcx),%edx
916 movl $_TIF_WORK_MASK,%edi 1079 movl $_TIF_WORK_MASK,%edi
917 andl %edi,%edx 1080 andl %edi,%edx
918 jnz retint_careful 1081 jnz retint_careful
@@ -926,11 +1089,11 @@ error_kernelspace:
926 iret run with kernel gs again, so don't set the user space flag. 1089 iret run with kernel gs again, so don't set the user space flag.
927 B stepping K8s sometimes report an truncated RIP for IRET 1090 B stepping K8s sometimes report an truncated RIP for IRET
928 exceptions returning to compat mode. Check for these here too. */ 1091 exceptions returning to compat mode. Check for these here too. */
929 leaq irq_return(%rip),%rbp 1092 leaq irq_return(%rip),%rcx
930 cmpq %rbp,RIP(%rsp) 1093 cmpq %rcx,RIP(%rsp)
931 je error_swapgs 1094 je error_swapgs
932 movl %ebp,%ebp /* zero extend */ 1095 movl %ecx,%ecx /* zero extend */
933 cmpq %rbp,RIP(%rsp) 1096 cmpq %rcx,RIP(%rsp)
934 je error_swapgs 1097 je error_swapgs
935 cmpq $gs_change,RIP(%rsp) 1098 cmpq $gs_change,RIP(%rsp)
936 je error_swapgs 1099 je error_swapgs
@@ -939,7 +1102,7 @@ KPROBE_END(error_entry)
939 1102
940 /* Reload gs selector with exception handling */ 1103 /* Reload gs selector with exception handling */
941 /* edi: new selector */ 1104 /* edi: new selector */
942ENTRY(load_gs_index) 1105ENTRY(native_load_gs_index)
943 CFI_STARTPROC 1106 CFI_STARTPROC
944 pushf 1107 pushf
945 CFI_ADJUST_CFA_OFFSET 8 1108 CFI_ADJUST_CFA_OFFSET 8
@@ -953,7 +1116,7 @@ gs_change:
953 CFI_ADJUST_CFA_OFFSET -8 1116 CFI_ADJUST_CFA_OFFSET -8
954 ret 1117 ret
955 CFI_ENDPROC 1118 CFI_ENDPROC
956ENDPROC(load_gs_index) 1119ENDPROC(native_load_gs_index)
957 1120
958 .section __ex_table,"a" 1121 .section __ex_table,"a"
959 .align 8 1122 .align 8
@@ -1075,6 +1238,7 @@ END(device_not_available)
1075 /* runs on exception stack */ 1238 /* runs on exception stack */
1076KPROBE_ENTRY(debug) 1239KPROBE_ENTRY(debug)
1077 INTR_FRAME 1240 INTR_FRAME
1241 PARAVIRT_ADJUST_EXCEPTION_FRAME
1078 pushq $0 1242 pushq $0
1079 CFI_ADJUST_CFA_OFFSET 8 1243 CFI_ADJUST_CFA_OFFSET 8
1080 paranoidentry do_debug, DEBUG_STACK 1244 paranoidentry do_debug, DEBUG_STACK
@@ -1084,6 +1248,7 @@ KPROBE_END(debug)
1084 /* runs on exception stack */ 1248 /* runs on exception stack */
1085KPROBE_ENTRY(nmi) 1249KPROBE_ENTRY(nmi)
1086 INTR_FRAME 1250 INTR_FRAME
1251 PARAVIRT_ADJUST_EXCEPTION_FRAME
1087 pushq $-1 1252 pushq $-1
1088 CFI_ADJUST_CFA_OFFSET 8 1253 CFI_ADJUST_CFA_OFFSET 8
1089 paranoidentry do_nmi, 0, 0 1254 paranoidentry do_nmi, 0, 0
@@ -1097,6 +1262,7 @@ KPROBE_END(nmi)
1097 1262
1098KPROBE_ENTRY(int3) 1263KPROBE_ENTRY(int3)
1099 INTR_FRAME 1264 INTR_FRAME
1265 PARAVIRT_ADJUST_EXCEPTION_FRAME
1100 pushq $0 1266 pushq $0
1101 CFI_ADJUST_CFA_OFFSET 8 1267 CFI_ADJUST_CFA_OFFSET 8
1102 paranoidentry do_int3, DEBUG_STACK 1268 paranoidentry do_int3, DEBUG_STACK
@@ -1120,13 +1286,10 @@ ENTRY(coprocessor_segment_overrun)
1120 zeroentry do_coprocessor_segment_overrun 1286 zeroentry do_coprocessor_segment_overrun
1121END(coprocessor_segment_overrun) 1287END(coprocessor_segment_overrun)
1122 1288
1123ENTRY(reserved)
1124 zeroentry do_reserved
1125END(reserved)
1126
1127 /* runs on exception stack */ 1289 /* runs on exception stack */
1128ENTRY(double_fault) 1290ENTRY(double_fault)
1129 XCPT_FRAME 1291 XCPT_FRAME
1292 PARAVIRT_ADJUST_EXCEPTION_FRAME
1130 paranoidentry do_double_fault 1293 paranoidentry do_double_fault
1131 jmp paranoid_exit1 1294 jmp paranoid_exit1
1132 CFI_ENDPROC 1295 CFI_ENDPROC
@@ -1143,6 +1306,7 @@ END(segment_not_present)
1143 /* runs on exception stack */ 1306 /* runs on exception stack */
1144ENTRY(stack_segment) 1307ENTRY(stack_segment)
1145 XCPT_FRAME 1308 XCPT_FRAME
1309 PARAVIRT_ADJUST_EXCEPTION_FRAME
1146 paranoidentry do_stack_segment 1310 paranoidentry do_stack_segment
1147 jmp paranoid_exit1 1311 jmp paranoid_exit1
1148 CFI_ENDPROC 1312 CFI_ENDPROC
@@ -1168,6 +1332,7 @@ END(spurious_interrupt_bug)
1168 /* runs on exception stack */ 1332 /* runs on exception stack */
1169ENTRY(machine_check) 1333ENTRY(machine_check)
1170 INTR_FRAME 1334 INTR_FRAME
1335 PARAVIRT_ADJUST_EXCEPTION_FRAME
1171 pushq $0 1336 pushq $0
1172 CFI_ADJUST_CFA_OFFSET 8 1337 CFI_ADJUST_CFA_OFFSET 8
1173 paranoidentry do_machine_check 1338 paranoidentry do_machine_check
@@ -1202,3 +1367,103 @@ KPROBE_ENTRY(ignore_sysret)
1202 sysret 1367 sysret
1203 CFI_ENDPROC 1368 CFI_ENDPROC
1204ENDPROC(ignore_sysret) 1369ENDPROC(ignore_sysret)
1370
1371#ifdef CONFIG_XEN
1372ENTRY(xen_hypervisor_callback)
1373 zeroentry xen_do_hypervisor_callback
1374END(xen_hypervisor_callback)
1375
1376/*
1377# A note on the "critical region" in our callback handler.
1378# We want to avoid stacking callback handlers due to events occurring
1379# during handling of the last event. To do this, we keep events disabled
1380# until we've done all processing. HOWEVER, we must enable events before
1381# popping the stack frame (can't be done atomically) and so it would still
1382# be possible to get enough handler activations to overflow the stack.
1383# Although unlikely, bugs of that kind are hard to track down, so we'd
1384# like to avoid the possibility.
1385# So, on entry to the handler we detect whether we interrupted an
1386# existing activation in its critical region -- if so, we pop the current
1387# activation and restart the handler using the previous one.
1388*/
1389ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1390 CFI_STARTPROC
1391/* Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1392 see the correct pointer to the pt_regs */
1393 movq %rdi, %rsp # we don't return, adjust the stack frame
1394 CFI_ENDPROC
1395 CFI_DEFAULT_STACK
139611: incl %gs:pda_irqcount
1397 movq %rsp,%rbp
1398 CFI_DEF_CFA_REGISTER rbp
1399 cmovzq %gs:pda_irqstackptr,%rsp
1400 pushq %rbp # backlink for old unwinder
1401 call xen_evtchn_do_upcall
1402 popq %rsp
1403 CFI_DEF_CFA_REGISTER rsp
1404 decl %gs:pda_irqcount
1405 jmp error_exit
1406 CFI_ENDPROC
1407END(do_hypervisor_callback)
1408
1409/*
1410# Hypervisor uses this for application faults while it executes.
1411# We get here for two reasons:
1412# 1. Fault while reloading DS, ES, FS or GS
1413# 2. Fault while executing IRET
1414# Category 1 we do not need to fix up as Xen has already reloaded all segment
1415# registers that could be reloaded and zeroed the others.
1416# Category 2 we fix up by killing the current process. We cannot use the
1417# normal Linux return path in this case because if we use the IRET hypercall
1418# to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1419# We distinguish between categories by comparing each saved segment register
1420# with its current contents: any discrepancy means we in category 1.
1421*/
1422ENTRY(xen_failsafe_callback)
1423 framesz = (RIP-0x30) /* workaround buggy gas */
1424 _frame framesz
1425 CFI_REL_OFFSET rcx, 0
1426 CFI_REL_OFFSET r11, 8
1427 movw %ds,%cx
1428 cmpw %cx,0x10(%rsp)
1429 CFI_REMEMBER_STATE
1430 jne 1f
1431 movw %es,%cx
1432 cmpw %cx,0x18(%rsp)
1433 jne 1f
1434 movw %fs,%cx
1435 cmpw %cx,0x20(%rsp)
1436 jne 1f
1437 movw %gs,%cx
1438 cmpw %cx,0x28(%rsp)
1439 jne 1f
1440 /* All segments match their saved values => Category 2 (Bad IRET). */
1441 movq (%rsp),%rcx
1442 CFI_RESTORE rcx
1443 movq 8(%rsp),%r11
1444 CFI_RESTORE r11
1445 addq $0x30,%rsp
1446 CFI_ADJUST_CFA_OFFSET -0x30
1447 pushq $0
1448 CFI_ADJUST_CFA_OFFSET 8
1449 pushq %r11
1450 CFI_ADJUST_CFA_OFFSET 8
1451 pushq %rcx
1452 CFI_ADJUST_CFA_OFFSET 8
1453 jmp general_protection
1454 CFI_RESTORE_STATE
14551: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1456 movq (%rsp),%rcx
1457 CFI_RESTORE rcx
1458 movq 8(%rsp),%r11
1459 CFI_RESTORE r11
1460 addq $0x30,%rsp
1461 CFI_ADJUST_CFA_OFFSET -0x30
1462 pushq $0
1463 CFI_ADJUST_CFA_OFFSET 8
1464 SAVE_ALL
1465 jmp error_exit
1466 CFI_ENDPROC
1467END(xen_failsafe_callback)
1468
1469#endif /* CONFIG_XEN */
diff --git a/arch/x86/kernel/ftrace.c b/arch/x86/kernel/ftrace.c
new file mode 100644
index 000000000000..ab115cd15fdf
--- /dev/null
+++ b/arch/x86/kernel/ftrace.c
@@ -0,0 +1,141 @@
1/*
2 * Code for replacing ftrace calls with jumps.
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 *
6 * Thanks goes to Ingo Molnar, for suggesting the idea.
7 * Mathieu Desnoyers, for suggesting postponing the modifications.
8 * Arjan van de Ven, for keeping me straight, and explaining to me
9 * the dangers of modifying code on the run.
10 */
11
12#include <linux/spinlock.h>
13#include <linux/hardirq.h>
14#include <linux/ftrace.h>
15#include <linux/percpu.h>
16#include <linux/init.h>
17#include <linux/list.h>
18
19#include <asm/alternative.h>
20#include <asm/ftrace.h>
21
22
23/* Long is fine, even if it is only 4 bytes ;-) */
24static long *ftrace_nop;
25
26union ftrace_code_union {
27 char code[MCOUNT_INSN_SIZE];
28 struct {
29 char e8;
30 int offset;
31 } __attribute__((packed));
32};
33
34
35static int notrace ftrace_calc_offset(long ip, long addr)
36{
37 return (int)(addr - ip);
38}
39
40notrace unsigned char *ftrace_nop_replace(void)
41{
42 return (char *)ftrace_nop;
43}
44
45notrace unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr)
46{
47 static union ftrace_code_union calc;
48
49 calc.e8 = 0xe8;
50 calc.offset = ftrace_calc_offset(ip + MCOUNT_INSN_SIZE, addr);
51
52 /*
53 * No locking needed, this must be called via kstop_machine
54 * which in essence is like running on a uniprocessor machine.
55 */
56 return calc.code;
57}
58
59notrace int
60ftrace_modify_code(unsigned long ip, unsigned char *old_code,
61 unsigned char *new_code)
62{
63 unsigned replaced;
64 unsigned old = *(unsigned *)old_code; /* 4 bytes */
65 unsigned new = *(unsigned *)new_code; /* 4 bytes */
66 unsigned char newch = new_code[4];
67 int faulted = 0;
68
69 /*
70 * Note: Due to modules and __init, code can
71 * disappear and change, we need to protect against faulting
72 * as well as code changing.
73 *
74 * No real locking needed, this code is run through
75 * kstop_machine.
76 */
77 asm volatile (
78 "1: lock\n"
79 " cmpxchg %3, (%2)\n"
80 " jnz 2f\n"
81 " movb %b4, 4(%2)\n"
82 "2:\n"
83 ".section .fixup, \"ax\"\n"
84 "3: movl $1, %0\n"
85 " jmp 2b\n"
86 ".previous\n"
87 _ASM_EXTABLE(1b, 3b)
88 : "=r"(faulted), "=a"(replaced)
89 : "r"(ip), "r"(new), "c"(newch),
90 "0"(faulted), "a"(old)
91 : "memory");
92 sync_core();
93
94 if (replaced != old && replaced != new)
95 faulted = 2;
96
97 return faulted;
98}
99
100notrace int ftrace_update_ftrace_func(ftrace_func_t func)
101{
102 unsigned long ip = (unsigned long)(&ftrace_call);
103 unsigned char old[MCOUNT_INSN_SIZE], *new;
104 int ret;
105
106 memcpy(old, &ftrace_call, MCOUNT_INSN_SIZE);
107 new = ftrace_call_replace(ip, (unsigned long)func);
108 ret = ftrace_modify_code(ip, old, new);
109
110 return ret;
111}
112
113notrace int ftrace_mcount_set(unsigned long *data)
114{
115 unsigned long ip = (long)(&mcount_call);
116 unsigned long *addr = data;
117 unsigned char old[MCOUNT_INSN_SIZE], *new;
118
119 /*
120 * Replace the mcount stub with a pointer to the
121 * ip recorder function.
122 */
123 memcpy(old, &mcount_call, MCOUNT_INSN_SIZE);
124 new = ftrace_call_replace(ip, *addr);
125 *addr = ftrace_modify_code(ip, old, new);
126
127 return 0;
128}
129
130int __init ftrace_dyn_arch_init(void *data)
131{
132 const unsigned char *const *noptable = find_nop_table();
133
134 /* This is running in kstop_machine */
135
136 ftrace_mcount_set(data);
137
138 ftrace_nop = (unsigned long *)noptable[MCOUNT_INSN_SIZE];
139
140 return 0;
141}
diff --git a/arch/x86/kernel/genapic_64.c b/arch/x86/kernel/genapic_64.c
index cbaaf69bedb2..1fa8be5bd217 100644
--- a/arch/x86/kernel/genapic_64.c
+++ b/arch/x86/kernel/genapic_64.c
@@ -51,7 +51,7 @@ void __init setup_apic_routing(void)
51 else 51 else
52#endif 52#endif
53 53
54 if (num_possible_cpus() <= 8) 54 if (max_physical_apicid < 8)
55 genapic = &apic_flat; 55 genapic = &apic_flat;
56 else 56 else
57 genapic = &apic_physflat; 57 genapic = &apic_physflat;
diff --git a/arch/x86/kernel/genapic_flat_64.c b/arch/x86/kernel/genapic_flat_64.c
index 1a9c68845ee8..786548a62d38 100644
--- a/arch/x86/kernel/genapic_flat_64.c
+++ b/arch/x86/kernel/genapic_flat_64.c
@@ -168,7 +168,7 @@ static unsigned int physflat_cpu_mask_to_apicid(cpumask_t cpumask)
168 * May as well be the first. 168 * May as well be the first.
169 */ 169 */
170 cpu = first_cpu(cpumask); 170 cpu = first_cpu(cpumask);
171 if ((unsigned)cpu < NR_CPUS) 171 if ((unsigned)cpu < nr_cpu_ids)
172 return per_cpu(x86_cpu_to_apicid, cpu); 172 return per_cpu(x86_cpu_to_apicid, cpu);
173 else 173 else
174 return BAD_APICID; 174 return BAD_APICID;
diff --git a/arch/x86/kernel/genx2apic_uv_x.c b/arch/x86/kernel/genx2apic_uv_x.c
index ebf13908a743..2cfcbded888a 100644
--- a/arch/x86/kernel/genx2apic_uv_x.c
+++ b/arch/x86/kernel/genx2apic_uv_x.c
@@ -5,9 +5,10 @@
5 * 5 *
6 * SGI UV APIC functions (note: not an Intel compatible APIC) 6 * SGI UV APIC functions (note: not an Intel compatible APIC)
7 * 7 *
8 * Copyright (C) 2007 Silicon Graphics, Inc. All rights reserved. 8 * Copyright (C) 2007-2008 Silicon Graphics, Inc. All rights reserved.
9 */ 9 */
10 10
11#include <linux/kernel.h>
11#include <linux/threads.h> 12#include <linux/threads.h>
12#include <linux/cpumask.h> 13#include <linux/cpumask.h>
13#include <linux/string.h> 14#include <linux/string.h>
@@ -20,8 +21,10 @@
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>
24#include <asm/pgtable.h>
23#include <asm/uv/uv_mmrs.h> 25#include <asm/uv/uv_mmrs.h>
24#include <asm/uv/uv_hub.h> 26#include <asm/uv/uv_hub.h>
27#include <asm/uv/bios.h>
25 28
26DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info); 29DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
27EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info); 30EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
@@ -38,6 +41,9 @@ EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
38short uv_possible_blades; 41short uv_possible_blades;
39EXPORT_SYMBOL_GPL(uv_possible_blades); 42EXPORT_SYMBOL_GPL(uv_possible_blades);
40 43
44unsigned long sn_rtc_cycles_per_second;
45EXPORT_SYMBOL(sn_rtc_cycles_per_second);
46
41/* Start with all IRQs pointing to boot CPU. IRQ balancing will shift them. */ 47/* Start with all IRQs pointing to boot CPU. IRQ balancing will shift them. */
42 48
43static cpumask_t uv_target_cpus(void) 49static cpumask_t uv_target_cpus(void)
@@ -55,44 +61,44 @@ static cpumask_t uv_vector_allocation_domain(int cpu)
55int uv_wakeup_secondary(int phys_apicid, unsigned int start_rip) 61int uv_wakeup_secondary(int phys_apicid, unsigned int start_rip)
56{ 62{
57 unsigned long val; 63 unsigned long val;
58 int nasid; 64 int pnode;
59 65
60 nasid = uv_apicid_to_nasid(phys_apicid); 66 pnode = uv_apicid_to_pnode(phys_apicid);
61 val = (1UL << UVH_IPI_INT_SEND_SHFT) | 67 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
62 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) | 68 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
63 (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) | 69 (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
64 APIC_DM_INIT; 70 APIC_DM_INIT;
65 uv_write_global_mmr64(nasid, UVH_IPI_INT, val); 71 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
66 mdelay(10); 72 mdelay(10);
67 73
68 val = (1UL << UVH_IPI_INT_SEND_SHFT) | 74 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
69 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) | 75 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
70 (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) | 76 (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
71 APIC_DM_STARTUP; 77 APIC_DM_STARTUP;
72 uv_write_global_mmr64(nasid, UVH_IPI_INT, val); 78 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
73 return 0; 79 return 0;
74} 80}
75 81
76static void uv_send_IPI_one(int cpu, int vector) 82static void uv_send_IPI_one(int cpu, int vector)
77{ 83{
78 unsigned long val, apicid, lapicid; 84 unsigned long val, apicid, lapicid;
79 int nasid; 85 int pnode;
80 86
81 apicid = per_cpu(x86_cpu_to_apicid, cpu); /* ZZZ - cache node-local ? */ 87 apicid = per_cpu(x86_cpu_to_apicid, cpu); /* ZZZ - cache node-local ? */
82 lapicid = apicid & 0x3f; /* ZZZ macro needed */ 88 lapicid = apicid & 0x3f; /* ZZZ macro needed */
83 nasid = uv_apicid_to_nasid(apicid); 89 pnode = uv_apicid_to_pnode(apicid);
84 val = 90 val =
85 (1UL << UVH_IPI_INT_SEND_SHFT) | (lapicid << 91 (1UL << UVH_IPI_INT_SEND_SHFT) | (lapicid <<
86 UVH_IPI_INT_APIC_ID_SHFT) | 92 UVH_IPI_INT_APIC_ID_SHFT) |
87 (vector << UVH_IPI_INT_VECTOR_SHFT); 93 (vector << UVH_IPI_INT_VECTOR_SHFT);
88 uv_write_global_mmr64(nasid, UVH_IPI_INT, val); 94 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
89} 95}
90 96
91static void uv_send_IPI_mask(cpumask_t mask, int vector) 97static void uv_send_IPI_mask(cpumask_t mask, int vector)
92{ 98{
93 unsigned int cpu; 99 unsigned int cpu;
94 100
95 for (cpu = 0; cpu < NR_CPUS; ++cpu) 101 for_each_possible_cpu(cpu)
96 if (cpu_isset(cpu, mask)) 102 if (cpu_isset(cpu, mask))
97 uv_send_IPI_one(cpu, vector); 103 uv_send_IPI_one(cpu, vector);
98} 104}
@@ -126,7 +132,7 @@ static unsigned int uv_cpu_mask_to_apicid(cpumask_t cpumask)
126 * May as well be the first. 132 * May as well be the first.
127 */ 133 */
128 cpu = first_cpu(cpumask); 134 cpu = first_cpu(cpumask);
129 if ((unsigned)cpu < NR_CPUS) 135 if ((unsigned)cpu < nr_cpu_ids)
130 return per_cpu(x86_cpu_to_apicid, cpu); 136 return per_cpu(x86_cpu_to_apicid, cpu);
131 else 137 else
132 return BAD_APICID; 138 return BAD_APICID;
@@ -159,39 +165,163 @@ struct genapic apic_x2apic_uv_x = {
159 .phys_pkg_id = phys_pkg_id, /* Fixme ZZZ */ 165 .phys_pkg_id = phys_pkg_id, /* Fixme ZZZ */
160}; 166};
161 167
162static __cpuinit void set_x2apic_extra_bits(int nasid) 168static __cpuinit void set_x2apic_extra_bits(int pnode)
163{ 169{
164 __get_cpu_var(x2apic_extra_bits) = ((nasid >> 1) << 6); 170 __get_cpu_var(x2apic_extra_bits) = (pnode << 6);
165} 171}
166 172
167/* 173/*
168 * Called on boot cpu. 174 * Called on boot cpu.
169 */ 175 */
176static __init int boot_pnode_to_blade(int pnode)
177{
178 int blade;
179
180 for (blade = 0; blade < uv_num_possible_blades(); blade++)
181 if (pnode == uv_blade_info[blade].pnode)
182 return blade;
183 BUG();
184}
185
186struct redir_addr {
187 unsigned long redirect;
188 unsigned long alias;
189};
190
191#define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
192
193static __initdata struct redir_addr redir_addrs[] = {
194 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_SI_ALIAS0_OVERLAY_CONFIG},
195 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_SI_ALIAS1_OVERLAY_CONFIG},
196 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_SI_ALIAS2_OVERLAY_CONFIG},
197};
198
199static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
200{
201 union uvh_si_alias0_overlay_config_u alias;
202 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
203 int i;
204
205 for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
206 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
207 if (alias.s.base == 0) {
208 *size = (1UL << alias.s.m_alias);
209 redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
210 *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
211 return;
212 }
213 }
214 BUG();
215}
216
217static __init void map_low_mmrs(void)
218{
219 init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
220 init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
221}
222
223enum map_type {map_wb, map_uc};
224
225static void map_high(char *id, unsigned long base, int shift, enum map_type map_type)
226{
227 unsigned long bytes, paddr;
228
229 paddr = base << shift;
230 bytes = (1UL << shift);
231 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
232 paddr + bytes);
233 if (map_type == map_uc)
234 init_extra_mapping_uc(paddr, bytes);
235 else
236 init_extra_mapping_wb(paddr, bytes);
237
238}
239static __init void map_gru_high(int max_pnode)
240{
241 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
242 int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
243
244 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
245 if (gru.s.enable)
246 map_high("GRU", gru.s.base, shift, map_wb);
247}
248
249static __init void map_config_high(int max_pnode)
250{
251 union uvh_rh_gam_cfg_overlay_config_mmr_u cfg;
252 int shift = UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR_BASE_SHFT;
253
254 cfg.v = uv_read_local_mmr(UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR);
255 if (cfg.s.enable)
256 map_high("CONFIG", cfg.s.base, shift, map_uc);
257}
258
259static __init void map_mmr_high(int max_pnode)
260{
261 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
262 int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
263
264 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
265 if (mmr.s.enable)
266 map_high("MMR", mmr.s.base, shift, map_uc);
267}
268
269static __init void map_mmioh_high(int max_pnode)
270{
271 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
272 int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
273
274 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
275 if (mmioh.s.enable)
276 map_high("MMIOH", mmioh.s.base, shift, map_uc);
277}
278
279static __init void uv_rtc_init(void)
280{
281 long status, ticks_per_sec, drift;
282
283 status =
284 x86_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK, &ticks_per_sec,
285 &drift);
286 if (status != 0 || ticks_per_sec < 100000) {
287 printk(KERN_WARNING
288 "unable to determine platform RTC clock frequency, "
289 "guessing.\n");
290 /* BIOS gives wrong value for clock freq. so guess */
291 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
292 } else
293 sn_rtc_cycles_per_second = ticks_per_sec;
294}
295
170static __init void uv_system_init(void) 296static __init void uv_system_init(void)
171{ 297{
172 union uvh_si_addr_map_config_u m_n_config; 298 union uvh_si_addr_map_config_u m_n_config;
173 int bytes, nid, cpu, lcpu, nasid, last_nasid, blade; 299 union uvh_node_id_u node_id;
174 unsigned long mmr_base; 300 unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
301 int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
302 int max_pnode = 0;
303 unsigned long mmr_base, present;
304
305 map_low_mmrs();
175 306
176 m_n_config.v = uv_read_local_mmr(UVH_SI_ADDR_MAP_CONFIG); 307 m_n_config.v = uv_read_local_mmr(UVH_SI_ADDR_MAP_CONFIG);
308 m_val = m_n_config.s.m_skt;
309 n_val = m_n_config.s.n_skt;
177 mmr_base = 310 mmr_base =
178 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) & 311 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
179 ~UV_MMR_ENABLE; 312 ~UV_MMR_ENABLE;
180 printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base); 313 printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
181 314
182 last_nasid = -1; 315 for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
183 for_each_possible_cpu(cpu) { 316 uv_possible_blades +=
184 nid = cpu_to_node(cpu); 317 hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
185 nasid = uv_apicid_to_nasid(per_cpu(x86_cpu_to_apicid, cpu));
186 if (nasid != last_nasid)
187 uv_possible_blades++;
188 last_nasid = nasid;
189 }
190 printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades()); 318 printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
191 319
192 bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades(); 320 bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
193 uv_blade_info = alloc_bootmem_pages(bytes); 321 uv_blade_info = alloc_bootmem_pages(bytes);
194 322
323 get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
324
195 bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes(); 325 bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
196 uv_node_to_blade = alloc_bootmem_pages(bytes); 326 uv_node_to_blade = alloc_bootmem_pages(bytes);
197 memset(uv_node_to_blade, 255, bytes); 327 memset(uv_node_to_blade, 255, bytes);
@@ -200,43 +330,64 @@ static __init void uv_system_init(void)
200 uv_cpu_to_blade = alloc_bootmem_pages(bytes); 330 uv_cpu_to_blade = alloc_bootmem_pages(bytes);
201 memset(uv_cpu_to_blade, 255, bytes); 331 memset(uv_cpu_to_blade, 255, bytes);
202 332
203 last_nasid = -1; 333 blade = 0;
204 blade = -1; 334 for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
205 lcpu = -1; 335 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
206 for_each_possible_cpu(cpu) { 336 for (j = 0; j < 64; j++) {
207 nid = cpu_to_node(cpu); 337 if (!test_bit(j, &present))
208 nasid = uv_apicid_to_nasid(per_cpu(x86_cpu_to_apicid, cpu)); 338 continue;
209 if (nasid != last_nasid) { 339 uv_blade_info[blade].pnode = (i * 64 + j);
210 blade++; 340 uv_blade_info[blade].nr_possible_cpus = 0;
211 lcpu = -1;
212 uv_blade_info[blade].nr_posible_cpus = 0;
213 uv_blade_info[blade].nr_online_cpus = 0; 341 uv_blade_info[blade].nr_online_cpus = 0;
342 blade++;
214 } 343 }
215 last_nasid = nasid; 344 }
216 lcpu++; 345
346 node_id.v = uv_read_local_mmr(UVH_NODE_ID);
347 gnode_upper = (((unsigned long)node_id.s.node_id) &
348 ~((1 << n_val) - 1)) << m_val;
217 349
218 uv_cpu_hub_info(cpu)->m_val = m_n_config.s.m_skt; 350 uv_rtc_init();
219 uv_cpu_hub_info(cpu)->n_val = m_n_config.s.n_skt; 351
352 for_each_present_cpu(cpu) {
353 nid = cpu_to_node(cpu);
354 pnode = uv_apicid_to_pnode(per_cpu(x86_cpu_to_apicid, cpu));
355 blade = boot_pnode_to_blade(pnode);
356 lcpu = uv_blade_info[blade].nr_possible_cpus;
357 uv_blade_info[blade].nr_possible_cpus++;
358
359 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
360 uv_cpu_hub_info(cpu)->lowmem_remap_top =
361 lowmem_redir_base + lowmem_redir_size;
362 uv_cpu_hub_info(cpu)->m_val = m_val;
363 uv_cpu_hub_info(cpu)->n_val = m_val;
220 uv_cpu_hub_info(cpu)->numa_blade_id = blade; 364 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
221 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu; 365 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
222 uv_cpu_hub_info(cpu)->local_nasid = nasid; 366 uv_cpu_hub_info(cpu)->pnode = pnode;
223 uv_cpu_hub_info(cpu)->gnode_upper = 367 uv_cpu_hub_info(cpu)->pnode_mask = (1 << n_val) - 1;
224 nasid & ~((1 << uv_hub_info->n_val) - 1); 368 uv_cpu_hub_info(cpu)->gpa_mask = (1 << (m_val + n_val)) - 1;
369 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
225 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base; 370 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
226 uv_cpu_hub_info(cpu)->coherency_domain_number = 0;/* ZZZ */ 371 uv_cpu_hub_info(cpu)->coherency_domain_number = 0;/* ZZZ */
227 uv_blade_info[blade].nasid = nasid;
228 uv_blade_info[blade].nr_posible_cpus++;
229 uv_node_to_blade[nid] = blade; 372 uv_node_to_blade[nid] = blade;
230 uv_cpu_to_blade[cpu] = blade; 373 uv_cpu_to_blade[cpu] = blade;
374 max_pnode = max(pnode, max_pnode);
231 375
232 printk(KERN_DEBUG "UV cpu %d, apicid 0x%x, nasid %d, nid %d\n", 376 printk(KERN_DEBUG "UV: cpu %d, apicid 0x%x, pnode %d, nid %d, "
233 cpu, per_cpu(x86_cpu_to_apicid, cpu), nasid, nid); 377 "lcpu %d, blade %d\n",
234 printk(KERN_DEBUG "UV lcpu %d, blade %d\n", lcpu, blade); 378 cpu, per_cpu(x86_cpu_to_apicid, cpu), pnode, nid,
379 lcpu, blade);
235 } 380 }
381
382 map_gru_high(max_pnode);
383 map_mmr_high(max_pnode);
384 map_config_high(max_pnode);
385 map_mmioh_high(max_pnode);
236} 386}
237 387
238/* 388/*
239 * Called on each cpu to initialize the per_cpu UV data area. 389 * Called on each cpu to initialize the per_cpu UV data area.
390 * ZZZ hotplug not supported yet
240 */ 391 */
241void __cpuinit uv_cpu_init(void) 392void __cpuinit uv_cpu_init(void)
242{ 393{
@@ -246,5 +397,5 @@ void __cpuinit uv_cpu_init(void)
246 uv_blade_info[uv_numa_blade_id()].nr_online_cpus++; 397 uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
247 398
248 if (get_uv_system_type() == UV_NON_UNIQUE_APIC) 399 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
249 set_x2apic_extra_bits(uv_hub_info->local_nasid); 400 set_x2apic_extra_bits(uv_hub_info->pnode);
250} 401}
diff --git a/arch/x86/kernel/geode_32.c b/arch/x86/kernel/geode_32.c
index e8edd63ab000..9b08e852fd1a 100644
--- a/arch/x86/kernel/geode_32.c
+++ b/arch/x86/kernel/geode_32.c
@@ -166,6 +166,8 @@ int geode_has_vsa2(void)
166 static int has_vsa2 = -1; 166 static int has_vsa2 = -1;
167 167
168 if (has_vsa2 == -1) { 168 if (has_vsa2 == -1) {
169 u16 val;
170
169 /* 171 /*
170 * The VSA has virtual registers that we can query for a 172 * The VSA has virtual registers that we can query for a
171 * signature. 173 * signature.
@@ -173,7 +175,8 @@ int geode_has_vsa2(void)
173 outw(VSA_VR_UNLOCK, VSA_VRC_INDEX); 175 outw(VSA_VR_UNLOCK, VSA_VRC_INDEX);
174 outw(VSA_VR_SIGNATURE, VSA_VRC_INDEX); 176 outw(VSA_VR_SIGNATURE, VSA_VRC_INDEX);
175 177
176 has_vsa2 = (inw(VSA_VRC_DATA) == VSA_SIG); 178 val = inw(VSA_VRC_DATA);
179 has_vsa2 = (val == AMD_VSA_SIG || val == GSW_VSA_SIG);
177 } 180 }
178 181
179 return has_vsa2; 182 return has_vsa2;
diff --git a/arch/x86/kernel/head.c b/arch/x86/kernel/head.c
new file mode 100644
index 000000000000..3e66bd364a9d
--- /dev/null
+++ b/arch/x86/kernel/head.c
@@ -0,0 +1,55 @@
1#include <linux/kernel.h>
2#include <linux/init.h>
3
4#include <asm/setup.h>
5#include <asm/bios_ebda.h>
6
7#define BIOS_LOWMEM_KILOBYTES 0x413
8
9/*
10 * The BIOS places the EBDA/XBDA at the top of conventional
11 * memory, and usually decreases the reported amount of
12 * conventional memory (int 0x12) too. This also contains a
13 * workaround for Dell systems that neglect to reserve EBDA.
14 * The same workaround also avoids a problem with the AMD768MPX
15 * chipset: reserve a page before VGA to prevent PCI prefetch
16 * into it (errata #56). Usually the page is reserved anyways,
17 * unless you have no PS/2 mouse plugged in.
18 */
19void __init reserve_ebda_region(void)
20{
21 unsigned int lowmem, ebda_addr;
22
23 /* To determine the position of the EBDA and the */
24 /* end of conventional memory, we need to look at */
25 /* the BIOS data area. In a paravirtual environment */
26 /* that area is absent. We'll just have to assume */
27 /* that the paravirt case can handle memory setup */
28 /* correctly, without our help. */
29 if (paravirt_enabled())
30 return;
31
32 /* end of low (conventional) memory */
33 lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
34 lowmem <<= 10;
35
36 /* start of EBDA area */
37 ebda_addr = get_bios_ebda();
38
39 /* Fixup: bios puts an EBDA in the top 64K segment */
40 /* of conventional memory, but does not adjust lowmem. */
41 if ((lowmem - ebda_addr) <= 0x10000)
42 lowmem = ebda_addr;
43
44 /* Fixup: bios does not report an EBDA at all. */
45 /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
46 if ((ebda_addr == 0) && (lowmem >= 0x9f000))
47 lowmem = 0x9f000;
48
49 /* Paranoia: should never happen, but... */
50 if ((lowmem == 0) || (lowmem >= 0x100000))
51 lowmem = 0x9f000;
52
53 /* reserve all memory between lowmem and the 1MB mark */
54 reserve_early_overlap_ok(lowmem, 0x100000, "BIOS reserved");
55}
diff --git a/arch/x86/kernel/head32.c b/arch/x86/kernel/head32.c
index 3db059058927..fa1d25dd83e3 100644
--- a/arch/x86/kernel/head32.c
+++ b/arch/x86/kernel/head32.c
@@ -8,7 +8,34 @@
8#include <linux/init.h> 8#include <linux/init.h>
9#include <linux/start_kernel.h> 9#include <linux/start_kernel.h>
10 10
11#include <asm/setup.h>
12#include <asm/sections.h>
13#include <asm/e820.h>
14#include <asm/bios_ebda.h>
15
11void __init i386_start_kernel(void) 16void __init i386_start_kernel(void)
12{ 17{
18 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS");
19
20#ifdef CONFIG_BLK_DEV_INITRD
21 /* Reserve INITRD */
22 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
23 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
24 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
25 u64 ramdisk_end = ramdisk_image + ramdisk_size;
26 reserve_early(ramdisk_image, ramdisk_end, "RAMDISK");
27 }
28#endif
29 reserve_early(init_pg_tables_start, init_pg_tables_end,
30 "INIT_PG_TABLE");
31
32 reserve_ebda_region();
33
34 /*
35 * At this point everything still needed from the boot loader
36 * or BIOS or kernel text should be early reserved or marked not
37 * RAM in e820. All other memory is free game.
38 */
39
13 start_kernel(); 40 start_kernel();
14} 41}
diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index e25c57b8aa84..1b318e903bf6 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -25,6 +25,27 @@
25#include <asm/e820.h> 25#include <asm/e820.h>
26#include <asm/bios_ebda.h> 26#include <asm/bios_ebda.h>
27 27
28/* boot cpu pda */
29static struct x8664_pda _boot_cpu_pda __read_mostly;
30
31#ifdef CONFIG_SMP
32/*
33 * We install an empty cpu_pda pointer table to indicate to early users
34 * (numa_set_node) that the cpu_pda pointer table for cpus other than
35 * the boot cpu is not yet setup.
36 */
37static struct x8664_pda *__cpu_pda[NR_CPUS] __initdata;
38#else
39static struct x8664_pda *__cpu_pda[NR_CPUS] __read_mostly;
40#endif
41
42void __init x86_64_init_pda(void)
43{
44 _cpu_pda = __cpu_pda;
45 cpu_pda(0) = &_boot_cpu_pda;
46 pda_init(0);
47}
48
28static void __init zap_identity_mappings(void) 49static void __init zap_identity_mappings(void)
29{ 50{
30 pgd_t *pgd = pgd_offset_k(0UL); 51 pgd_t *pgd = pgd_offset_k(0UL);
@@ -51,74 +72,6 @@ static void __init copy_bootdata(char *real_mode_data)
51 } 72 }
52} 73}
53 74
54#define BIOS_LOWMEM_KILOBYTES 0x413
55
56/*
57 * The BIOS places the EBDA/XBDA at the top of conventional
58 * memory, and usually decreases the reported amount of
59 * conventional memory (int 0x12) too. This also contains a
60 * workaround for Dell systems that neglect to reserve EBDA.
61 * The same workaround also avoids a problem with the AMD768MPX
62 * chipset: reserve a page before VGA to prevent PCI prefetch
63 * into it (errata #56). Usually the page is reserved anyways,
64 * unless you have no PS/2 mouse plugged in.
65 */
66static void __init reserve_ebda_region(void)
67{
68 unsigned int lowmem, ebda_addr;
69
70 /* To determine the position of the EBDA and the */
71 /* end of conventional memory, we need to look at */
72 /* the BIOS data area. In a paravirtual environment */
73 /* that area is absent. We'll just have to assume */
74 /* that the paravirt case can handle memory setup */
75 /* correctly, without our help. */
76 if (paravirt_enabled())
77 return;
78
79 /* end of low (conventional) memory */
80 lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
81 lowmem <<= 10;
82
83 /* start of EBDA area */
84 ebda_addr = get_bios_ebda();
85
86 /* Fixup: bios puts an EBDA in the top 64K segment */
87 /* of conventional memory, but does not adjust lowmem. */
88 if ((lowmem - ebda_addr) <= 0x10000)
89 lowmem = ebda_addr;
90
91 /* Fixup: bios does not report an EBDA at all. */
92 /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
93 if ((ebda_addr == 0) && (lowmem >= 0x9f000))
94 lowmem = 0x9f000;
95
96 /* Paranoia: should never happen, but... */
97 if ((lowmem == 0) || (lowmem >= 0x100000))
98 lowmem = 0x9f000;
99
100 /* reserve all memory between lowmem and the 1MB mark */
101 reserve_early(lowmem, 0x100000, "BIOS reserved");
102}
103
104static void __init reserve_setup_data(void)
105{
106 struct setup_data *data;
107 unsigned long pa_data;
108 char buf[32];
109
110 if (boot_params.hdr.version < 0x0209)
111 return;
112 pa_data = boot_params.hdr.setup_data;
113 while (pa_data) {
114 data = early_ioremap(pa_data, sizeof(*data));
115 sprintf(buf, "setup data %x", data->type);
116 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
117 pa_data = data->next;
118 early_iounmap(data, sizeof(*data));
119 }
120}
121
122void __init x86_64_start_kernel(char * real_mode_data) 75void __init x86_64_start_kernel(char * real_mode_data)
123{ 76{
124 int i; 77 int i;
@@ -156,10 +109,15 @@ void __init x86_64_start_kernel(char * real_mode_data)
156 109
157 early_printk("Kernel alive\n"); 110 early_printk("Kernel alive\n");
158 111
159 for (i = 0; i < NR_CPUS; i++) 112 x86_64_init_pda();
160 cpu_pda(i) = &boot_cpu_pda[i];
161 113
162 pda_init(0); 114 early_printk("Kernel really alive\n");
115
116 x86_64_start_reservations(real_mode_data);
117}
118
119void __init x86_64_start_reservations(char *real_mode_data)
120{
163 copy_bootdata(__va(real_mode_data)); 121 copy_bootdata(__va(real_mode_data));
164 122
165 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS"); 123 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS");
@@ -175,7 +133,6 @@ void __init x86_64_start_kernel(char * real_mode_data)
175#endif 133#endif
176 134
177 reserve_ebda_region(); 135 reserve_ebda_region();
178 reserve_setup_data();
179 136
180 /* 137 /*
181 * At this point everything still needed from the boot loader 138 * At this point everything still needed from the boot loader
diff --git a/arch/x86/kernel/head_32.S b/arch/x86/kernel/head_32.S
index f7357cc0162c..a7010c3a377a 100644
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -194,6 +194,7 @@ default_entry:
194 xorl %ebx,%ebx /* %ebx is kept at zero */ 194 xorl %ebx,%ebx /* %ebx is kept at zero */
195 195
196 movl $pa(pg0), %edi 196 movl $pa(pg0), %edi
197 movl %edi, pa(init_pg_tables_start)
197 movl $pa(swapper_pg_pmd), %edx 198 movl $pa(swapper_pg_pmd), %edx
198 movl $PTE_ATTR, %eax 199 movl $PTE_ATTR, %eax
19910: 20010:
@@ -219,6 +220,8 @@ default_entry:
219 jb 10b 220 jb 10b
2201: 2211:
221 movl %edi,pa(init_pg_tables_end) 222 movl %edi,pa(init_pg_tables_end)
223 shrl $12, %eax
224 movl %eax, pa(max_pfn_mapped)
222 225
223 /* Do early initialization of the fixmap area */ 226 /* Do early initialization of the fixmap area */
224 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax 227 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
@@ -228,6 +231,7 @@ default_entry:
228page_pde_offset = (__PAGE_OFFSET >> 20); 231page_pde_offset = (__PAGE_OFFSET >> 20);
229 232
230 movl $pa(pg0), %edi 233 movl $pa(pg0), %edi
234 movl %edi, pa(init_pg_tables_start)
231 movl $pa(swapper_pg_dir), %edx 235 movl $pa(swapper_pg_dir), %edx
232 movl $PTE_ATTR, %eax 236 movl $PTE_ATTR, %eax
23310: 23710:
@@ -249,6 +253,8 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
249 cmpl %ebp,%eax 253 cmpl %ebp,%eax
250 jb 10b 254 jb 10b
251 movl %edi,pa(init_pg_tables_end) 255 movl %edi,pa(init_pg_tables_end)
256 shrl $12, %eax
257 movl %eax, pa(max_pfn_mapped)
252 258
253 /* Do early initialization of the fixmap area */ 259 /* Do early initialization of the fixmap area */
254 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax 260 movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
@@ -446,10 +452,10 @@ is386: movl $2,%ecx # set MP
446 je 1f 452 je 1f
447 movl $(__KERNEL_PERCPU), %eax 453 movl $(__KERNEL_PERCPU), %eax
448 movl %eax,%fs # set this cpu's percpu 454 movl %eax,%fs # set this cpu's percpu
449 jmp initialize_secondary # all other CPUs call initialize_secondary 455 movl (stack_start), %esp
4501: 4561:
451#endif /* CONFIG_SMP */ 457#endif /* CONFIG_SMP */
452 jmp i386_start_kernel 458 jmp *(initial_code)
453 459
454/* 460/*
455 * We depend on ET to be correct. This checks for 287/387. 461 * We depend on ET to be correct. This checks for 287/387.
@@ -592,6 +598,11 @@ ignore_int:
592#endif 598#endif
593 iret 599 iret
594 600
601.section .cpuinit.data,"wa"
602.align 4
603ENTRY(initial_code)
604 .long i386_start_kernel
605
595.section .text 606.section .text
596/* 607/*
597 * Real beginning of normal "text" segment 608 * Real beginning of normal "text" segment
diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index 10a1955bb1d1..db3280afe886 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -18,6 +18,7 @@
18#include <asm/page.h> 18#include <asm/page.h>
19#include <asm/msr.h> 19#include <asm/msr.h>
20#include <asm/cache.h> 20#include <asm/cache.h>
21#include <asm/processor-flags.h>
21 22
22#ifdef CONFIG_PARAVIRT 23#ifdef CONFIG_PARAVIRT
23#include <asm/asm-offsets.h> 24#include <asm/asm-offsets.h>
@@ -31,6 +32,13 @@
31 * 32 *
32 */ 33 */
33 34
35#define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
36
37L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
38L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
39L4_START_KERNEL = pgd_index(__START_KERNEL_map)
40L3_START_KERNEL = pud_index(__START_KERNEL_map)
41
34 .text 42 .text
35 .section .text.head 43 .section .text.head
36 .code64 44 .code64
@@ -76,8 +84,8 @@ startup_64:
76 /* Fixup the physical addresses in the page table 84 /* Fixup the physical addresses in the page table
77 */ 85 */
78 addq %rbp, init_level4_pgt + 0(%rip) 86 addq %rbp, init_level4_pgt + 0(%rip)
79 addq %rbp, init_level4_pgt + (258*8)(%rip) 87 addq %rbp, init_level4_pgt + (L4_PAGE_OFFSET*8)(%rip)
80 addq %rbp, init_level4_pgt + (511*8)(%rip) 88 addq %rbp, init_level4_pgt + (L4_START_KERNEL*8)(%rip)
81 89
82 addq %rbp, level3_ident_pgt + 0(%rip) 90 addq %rbp, level3_ident_pgt + 0(%rip)
83 91
@@ -128,7 +136,7 @@ ident_complete:
128 /* Fixup phys_base */ 136 /* Fixup phys_base */
129 addq %rbp, phys_base(%rip) 137 addq %rbp, phys_base(%rip)
130 138
131#ifdef CONFIG_SMP 139#ifdef CONFIG_X86_TRAMPOLINE
132 addq %rbp, trampoline_level4_pgt + 0(%rip) 140 addq %rbp, trampoline_level4_pgt + 0(%rip)
133 addq %rbp, trampoline_level4_pgt + (511*8)(%rip) 141 addq %rbp, trampoline_level4_pgt + (511*8)(%rip)
134#endif 142#endif
@@ -154,9 +162,7 @@ ENTRY(secondary_startup_64)
154 */ 162 */
155 163
156 /* Enable PAE mode and PGE */ 164 /* Enable PAE mode and PGE */
157 xorq %rax, %rax 165 movl $(X86_CR4_PAE | X86_CR4_PGE), %eax
158 btsq $5, %rax
159 btsq $7, %rax
160 movq %rax, %cr4 166 movq %rax, %cr4
161 167
162 /* Setup early boot stage 4 level pagetables. */ 168 /* Setup early boot stage 4 level pagetables. */
@@ -184,19 +190,15 @@ ENTRY(secondary_startup_64)
1841: wrmsr /* Make changes effective */ 1901: wrmsr /* Make changes effective */
185 191
186 /* Setup cr0 */ 192 /* Setup cr0 */
187#define CR0_PM 1 /* protected mode */ 193#define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
188#define CR0_MP (1<<1) 194 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
189#define CR0_ET (1<<4) 195 X86_CR0_PG)
190#define CR0_NE (1<<5) 196 movl $CR0_STATE, %eax
191#define CR0_WP (1<<16)
192#define CR0_AM (1<<18)
193#define CR0_PAGING (1<<31)
194 movl $CR0_PM|CR0_MP|CR0_ET|CR0_NE|CR0_WP|CR0_AM|CR0_PAGING,%eax
195 /* Make changes effective */ 197 /* Make changes effective */
196 movq %rax, %cr0 198 movq %rax, %cr0
197 199
198 /* Setup a boot time stack */ 200 /* Setup a boot time stack */
199 movq init_rsp(%rip),%rsp 201 movq stack_start(%rip),%rsp
200 202
201 /* zero EFLAGS after setting rsp */ 203 /* zero EFLAGS after setting rsp */
202 pushq $0 204 pushq $0
@@ -208,7 +210,7 @@ ENTRY(secondary_startup_64)
208 * addresses where we're currently running on. We have to do that here 210 * addresses where we're currently running on. We have to do that here
209 * because in 32bit we couldn't load a 64bit linear address. 211 * because in 32bit we couldn't load a 64bit linear address.
210 */ 212 */
211 lgdt cpu_gdt_descr(%rip) 213 lgdt early_gdt_descr(%rip)
212 214
213 /* set up data segments. actually 0 would do too */ 215 /* set up data segments. actually 0 would do too */
214 movl $__KERNEL_DS,%eax 216 movl $__KERNEL_DS,%eax
@@ -257,8 +259,9 @@ ENTRY(secondary_startup_64)
257 .quad x86_64_start_kernel 259 .quad x86_64_start_kernel
258 __FINITDATA 260 __FINITDATA
259 261
260 ENTRY(init_rsp) 262 ENTRY(stack_start)
261 .quad init_thread_union+THREAD_SIZE-8 263 .quad init_thread_union+THREAD_SIZE-8
264 .word 0
262 265
263bad_address: 266bad_address:
264 jmp bad_address 267 jmp bad_address
@@ -327,11 +330,11 @@ early_idt_ripmsg:
327ENTRY(name) 330ENTRY(name)
328 331
329/* Automate the creation of 1 to 1 mapping pmd entries */ 332/* Automate the creation of 1 to 1 mapping pmd entries */
330#define PMDS(START, PERM, COUNT) \ 333#define PMDS(START, PERM, COUNT) \
331 i = 0 ; \ 334 i = 0 ; \
332 .rept (COUNT) ; \ 335 .rept (COUNT) ; \
333 .quad (START) + (i << 21) + (PERM) ; \ 336 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
334 i = i + 1 ; \ 337 i = i + 1 ; \
335 .endr 338 .endr
336 339
337 /* 340 /*
@@ -342,9 +345,9 @@ ENTRY(name)
342 */ 345 */
343NEXT_PAGE(init_level4_pgt) 346NEXT_PAGE(init_level4_pgt)
344 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE 347 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
345 .fill 257,8,0 348 .org init_level4_pgt + L4_PAGE_OFFSET*8, 0
346 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE 349 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
347 .fill 252,8,0 350 .org init_level4_pgt + L4_START_KERNEL*8, 0
348 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ 351 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
349 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE 352 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
350 353
@@ -353,7 +356,7 @@ NEXT_PAGE(level3_ident_pgt)
353 .fill 511,8,0 356 .fill 511,8,0
354 357
355NEXT_PAGE(level3_kernel_pgt) 358NEXT_PAGE(level3_kernel_pgt)
356 .fill 510,8,0 359 .fill L3_START_KERNEL,8,0
357 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */ 360 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
358 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE 361 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
359 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE 362 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
@@ -384,7 +387,7 @@ NEXT_PAGE(level2_kernel_pgt)
384 * If you want to increase this then increase MODULES_VADDR 387 * If you want to increase this then increase MODULES_VADDR
385 * too.) 388 * too.)
386 */ 389 */
387 PMDS(0, __PAGE_KERNEL_LARGE_EXEC|_PAGE_GLOBAL, 390 PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
388 KERNEL_IMAGE_SIZE/PMD_SIZE) 391 KERNEL_IMAGE_SIZE/PMD_SIZE)
389 392
390NEXT_PAGE(level2_spare_pgt) 393NEXT_PAGE(level2_spare_pgt)
@@ -395,54 +398,17 @@ NEXT_PAGE(level2_spare_pgt)
395 398
396 .data 399 .data
397 .align 16 400 .align 16
398 .globl cpu_gdt_descr 401 .globl early_gdt_descr
399cpu_gdt_descr: 402early_gdt_descr:
400 .word gdt_end-cpu_gdt_table-1 403 .word GDT_ENTRIES*8-1
401gdt: 404 .quad per_cpu__gdt_page
402 .quad cpu_gdt_table
403#ifdef CONFIG_SMP
404 .rept NR_CPUS-1
405 .word 0
406 .quad 0
407 .endr
408#endif
409 405
410ENTRY(phys_base) 406ENTRY(phys_base)
411 /* This must match the first entry in level2_kernel_pgt */ 407 /* This must match the first entry in level2_kernel_pgt */
412 .quad 0x0000000000000000 408 .quad 0x0000000000000000
413 409
414/* We need valid kernel segments for data and code in long mode too 410#include "../../x86/xen/xen-head.S"
415 * IRET will check the segment types kkeil 2000/10/28
416 * Also sysret mandates a special GDT layout
417 */
418
419 .section .data.page_aligned, "aw"
420 .align PAGE_SIZE
421
422/* The TLS descriptors are currently at a different place compared to i386.
423 Hopefully nobody expects them at a fixed place (Wine?) */
424 411
425ENTRY(cpu_gdt_table)
426 .quad 0x0000000000000000 /* NULL descriptor */
427 .quad 0x00cf9b000000ffff /* __KERNEL32_CS */
428 .quad 0x00af9b000000ffff /* __KERNEL_CS */
429 .quad 0x00cf93000000ffff /* __KERNEL_DS */
430 .quad 0x00cffb000000ffff /* __USER32_CS */
431 .quad 0x00cff3000000ffff /* __USER_DS, __USER32_DS */
432 .quad 0x00affb000000ffff /* __USER_CS */
433 .quad 0x0 /* unused */
434 .quad 0,0 /* TSS */
435 .quad 0,0 /* LDT */
436 .quad 0,0,0 /* three TLS descriptors */
437 .quad 0x0000f40000000000 /* node/CPU stored in limit */
438gdt_end:
439 /* asm/segment.h:GDT_ENTRIES must match this */
440 /* This should be a multiple of the cache line size */
441 /* GDTs of other CPUs are now dynamically allocated */
442
443 /* zero the remaining page */
444 .fill PAGE_SIZE / 8 - GDT_ENTRIES,8,0
445
446 .section .bss, "aw", @nobits 412 .section .bss, "aw", @nobits
447 .align L1_CACHE_BYTES 413 .align L1_CACHE_BYTES
448ENTRY(idt_table) 414ENTRY(idt_table)
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index 9b5cfcdfc426..ad2b15a1334d 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -17,7 +17,7 @@
17 17
18/* FSEC = 10^-15 18/* FSEC = 10^-15
19 NSEC = 10^-9 */ 19 NSEC = 10^-9 */
20#define FSEC_PER_NSEC 1000000 20#define FSEC_PER_NSEC 1000000L
21 21
22/* 22/*
23 * HPET address is set in acpi/boot.c, when an ACPI entry exists 23 * HPET address is set in acpi/boot.c, when an ACPI entry exists
@@ -36,26 +36,15 @@ static inline void hpet_writel(unsigned long d, unsigned long a)
36} 36}
37 37
38#ifdef CONFIG_X86_64 38#ifdef CONFIG_X86_64
39
40#include <asm/pgtable.h> 39#include <asm/pgtable.h>
41 40#endif
42static inline void hpet_set_mapping(void)
43{
44 set_fixmap_nocache(FIX_HPET_BASE, hpet_address);
45 __set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
46 hpet_virt_address = (void __iomem *)fix_to_virt(FIX_HPET_BASE);
47}
48
49static inline void hpet_clear_mapping(void)
50{
51 hpet_virt_address = NULL;
52}
53
54#else
55 41
56static inline void hpet_set_mapping(void) 42static inline void hpet_set_mapping(void)
57{ 43{
58 hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE); 44 hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
45#ifdef CONFIG_X86_64
46 __set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
47#endif
59} 48}
60 49
61static inline void hpet_clear_mapping(void) 50static inline void hpet_clear_mapping(void)
@@ -63,7 +52,6 @@ static inline void hpet_clear_mapping(void)
63 iounmap(hpet_virt_address); 52 iounmap(hpet_virt_address);
64 hpet_virt_address = NULL; 53 hpet_virt_address = NULL;
65} 54}
66#endif
67 55
68/* 56/*
69 * HPET command line enable / disable 57 * HPET command line enable / disable
@@ -206,20 +194,19 @@ static void hpet_enable_legacy_int(void)
206 194
207static void hpet_legacy_clockevent_register(void) 195static void hpet_legacy_clockevent_register(void)
208{ 196{
209 uint64_t hpet_freq;
210
211 /* Start HPET legacy interrupts */ 197 /* Start HPET legacy interrupts */
212 hpet_enable_legacy_int(); 198 hpet_enable_legacy_int();
213 199
214 /* 200 /*
215 * The period is a femto seconds value. We need to calculate the 201 * The mult factor is defined as (include/linux/clockchips.h)
216 * scaled math multiplication factor for nanosecond to hpet tick 202 * mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
217 * conversion. 203 * hpet_period is in units of femtoseconds (per cycle), so
204 * mult/2^shift = cyc/ns = 10^6/hpet_period
205 * mult = (10^6 * 2^shift)/hpet_period
206 * mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
218 */ 207 */
219 hpet_freq = 1000000000000000ULL; 208 hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
220 do_div(hpet_freq, hpet_period); 209 hpet_period, hpet_clockevent.shift);
221 hpet_clockevent.mult = div_sc((unsigned long) hpet_freq,
222 NSEC_PER_SEC, hpet_clockevent.shift);
223 /* Calculate the min / max delta */ 210 /* Calculate the min / max delta */
224 hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, 211 hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
225 &hpet_clockevent); 212 &hpet_clockevent);
@@ -324,7 +311,7 @@ static struct clocksource clocksource_hpet = {
324 311
325static int hpet_clocksource_register(void) 312static int hpet_clocksource_register(void)
326{ 313{
327 u64 tmp, start, now; 314 u64 start, now;
328 cycle_t t1; 315 cycle_t t1;
329 316
330 /* Start the counter */ 317 /* Start the counter */
@@ -351,21 +338,15 @@ static int hpet_clocksource_register(void)
351 return -ENODEV; 338 return -ENODEV;
352 } 339 }
353 340
354 /* Initialize and register HPET clocksource 341 /*
355 * 342 * The definition of mult is (include/linux/clocksource.h)
356 * hpet period is in femto seconds per cycle 343 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
357 * so we need to convert this to ns/cyc units 344 * so we first need to convert hpet_period to ns/cyc units:
358 * approximated by mult/2^shift 345 * mult/2^shift = ns/cyc = hpet_period/10^6
359 * 346 * mult = (hpet_period * 2^shift)/10^6
360 * fsec/cyc * 1nsec/1000000fsec = nsec/cyc = mult/2^shift 347 * mult = (hpet_period << shift)/FSEC_PER_NSEC
361 * fsec/cyc * 1ns/1000000fsec * 2^shift = mult
362 * fsec/cyc * 2^shift * 1nsec/1000000fsec = mult
363 * (fsec/cyc << shift)/1000000 = mult
364 * (hpet_period << shift)/FSEC_PER_NSEC = mult
365 */ 348 */
366 tmp = (u64)hpet_period << HPET_SHIFT; 349 clocksource_hpet.mult = div_sc(hpet_period, FSEC_PER_NSEC, HPET_SHIFT);
367 do_div(tmp, FSEC_PER_NSEC);
368 clocksource_hpet.mult = (u32)tmp;
369 350
370 clocksource_register(&clocksource_hpet); 351 clocksource_register(&clocksource_hpet);
371 352
@@ -487,7 +468,7 @@ void hpet_disable(void)
487#define RTC_NUM_INTS 1 468#define RTC_NUM_INTS 1
488 469
489static unsigned long hpet_rtc_flags; 470static unsigned long hpet_rtc_flags;
490static unsigned long hpet_prev_update_sec; 471static int hpet_prev_update_sec;
491static struct rtc_time hpet_alarm_time; 472static struct rtc_time hpet_alarm_time;
492static unsigned long hpet_pie_count; 473static unsigned long hpet_pie_count;
493static unsigned long hpet_t1_cmp; 474static unsigned long hpet_t1_cmp;
@@ -594,6 +575,9 @@ int hpet_set_rtc_irq_bit(unsigned long bit_mask)
594 575
595 hpet_rtc_flags |= bit_mask; 576 hpet_rtc_flags |= bit_mask;
596 577
578 if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
579 hpet_prev_update_sec = -1;
580
597 if (!oldbits) 581 if (!oldbits)
598 hpet_rtc_timer_init(); 582 hpet_rtc_timer_init();
599 583
@@ -671,7 +655,7 @@ static void hpet_rtc_timer_reinit(void)
671 if (hpet_rtc_flags & RTC_PIE) 655 if (hpet_rtc_flags & RTC_PIE)
672 hpet_pie_count += lost_ints; 656 hpet_pie_count += lost_ints;
673 if (printk_ratelimit()) 657 if (printk_ratelimit())
674 printk(KERN_WARNING "rtc: lost %d interrupts\n", 658 printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
675 lost_ints); 659 lost_ints);
676 } 660 }
677} 661}
@@ -689,7 +673,8 @@ irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
689 673
690 if (hpet_rtc_flags & RTC_UIE && 674 if (hpet_rtc_flags & RTC_UIE &&
691 curr_time.tm_sec != hpet_prev_update_sec) { 675 curr_time.tm_sec != hpet_prev_update_sec) {
692 rtc_int_flag = RTC_UF; 676 if (hpet_prev_update_sec >= 0)
677 rtc_int_flag = RTC_UF;
693 hpet_prev_update_sec = curr_time.tm_sec; 678 hpet_prev_update_sec = curr_time.tm_sec;
694 } 679 }
695 680
diff --git a/arch/x86/kernel/i386_ksyms_32.c b/arch/x86/kernel/i386_ksyms_32.c
index deb43785e923..dd7ebee446af 100644
--- a/arch/x86/kernel/i386_ksyms_32.c
+++ b/arch/x86/kernel/i386_ksyms_32.c
@@ -1,7 +1,14 @@
1#include <linux/module.h> 1#include <linux/module.h>
2
2#include <asm/checksum.h> 3#include <asm/checksum.h>
3#include <asm/desc.h>
4#include <asm/pgtable.h> 4#include <asm/pgtable.h>
5#include <asm/desc.h>
6#include <asm/ftrace.h>
7
8#ifdef CONFIG_FTRACE
9/* mcount is defined in assembly */
10EXPORT_SYMBOL(mcount);
11#endif
5 12
6/* Networking helper routines. */ 13/* Networking helper routines. */
7EXPORT_SYMBOL(csum_partial_copy_generic); 14EXPORT_SYMBOL(csum_partial_copy_generic);
diff --git a/arch/x86/kernel/i8259_32.c b/arch/x86/kernel/i8259.c
index fe631967d625..dc92b49d9204 100644
--- a/arch/x86/kernel/i8259_32.c
+++ b/arch/x86/kernel/i8259.c
@@ -1,8 +1,10 @@
1#include <linux/linkage.h>
1#include <linux/errno.h> 2#include <linux/errno.h>
2#include <linux/signal.h> 3#include <linux/signal.h>
3#include <linux/sched.h> 4#include <linux/sched.h>
4#include <linux/ioport.h> 5#include <linux/ioport.h>
5#include <linux/interrupt.h> 6#include <linux/interrupt.h>
7#include <linux/timex.h>
6#include <linux/slab.h> 8#include <linux/slab.h>
7#include <linux/random.h> 9#include <linux/random.h>
8#include <linux/init.h> 10#include <linux/init.h>
@@ -10,10 +12,12 @@
10#include <linux/sysdev.h> 12#include <linux/sysdev.h>
11#include <linux/bitops.h> 13#include <linux/bitops.h>
12 14
15#include <asm/acpi.h>
13#include <asm/atomic.h> 16#include <asm/atomic.h>
14#include <asm/system.h> 17#include <asm/system.h>
15#include <asm/io.h> 18#include <asm/io.h>
16#include <asm/timer.h> 19#include <asm/timer.h>
20#include <asm/hw_irq.h>
17#include <asm/pgtable.h> 21#include <asm/pgtable.h>
18#include <asm/delay.h> 22#include <asm/delay.h>
19#include <asm/desc.h> 23#include <asm/desc.h>
@@ -32,7 +36,7 @@ static int i8259A_auto_eoi;
32DEFINE_SPINLOCK(i8259A_lock); 36DEFINE_SPINLOCK(i8259A_lock);
33static void mask_and_ack_8259A(unsigned int); 37static void mask_and_ack_8259A(unsigned int);
34 38
35static struct irq_chip i8259A_chip = { 39struct irq_chip i8259A_chip = {
36 .name = "XT-PIC", 40 .name = "XT-PIC",
37 .mask = disable_8259A_irq, 41 .mask = disable_8259A_irq,
38 .disable = disable_8259A_irq, 42 .disable = disable_8259A_irq,
@@ -125,14 +129,14 @@ static inline int i8259A_irq_real(unsigned int irq)
125 int irqmask = 1<<irq; 129 int irqmask = 1<<irq;
126 130
127 if (irq < 8) { 131 if (irq < 8) {
128 outb(0x0B,PIC_MASTER_CMD); /* ISR register */ 132 outb(0x0B, PIC_MASTER_CMD); /* ISR register */
129 value = inb(PIC_MASTER_CMD) & irqmask; 133 value = inb(PIC_MASTER_CMD) & irqmask;
130 outb(0x0A,PIC_MASTER_CMD); /* back to the IRR register */ 134 outb(0x0A, PIC_MASTER_CMD); /* back to the IRR register */
131 return value; 135 return value;
132 } 136 }
133 outb(0x0B,PIC_SLAVE_CMD); /* ISR register */ 137 outb(0x0B, PIC_SLAVE_CMD); /* ISR register */
134 value = inb(PIC_SLAVE_CMD) & (irqmask >> 8); 138 value = inb(PIC_SLAVE_CMD) & (irqmask >> 8);
135 outb(0x0A,PIC_SLAVE_CMD); /* back to the IRR register */ 139 outb(0x0A, PIC_SLAVE_CMD); /* back to the IRR register */
136 return value; 140 return value;
137} 141}
138 142
@@ -171,12 +175,14 @@ handle_real_irq:
171 if (irq & 8) { 175 if (irq & 8) {
172 inb(PIC_SLAVE_IMR); /* DUMMY - (do we need this?) */ 176 inb(PIC_SLAVE_IMR); /* DUMMY - (do we need this?) */
173 outb(cached_slave_mask, PIC_SLAVE_IMR); 177 outb(cached_slave_mask, PIC_SLAVE_IMR);
174 outb(0x60+(irq&7),PIC_SLAVE_CMD);/* 'Specific EOI' to slave */ 178 /* 'Specific EOI' to slave */
175 outb(0x60+PIC_CASCADE_IR,PIC_MASTER_CMD); /* 'Specific EOI' to master-IRQ2 */ 179 outb(0x60+(irq&7), PIC_SLAVE_CMD);
180 /* 'Specific EOI' to master-IRQ2 */
181 outb(0x60+PIC_CASCADE_IR, PIC_MASTER_CMD);
176 } else { 182 } else {
177 inb(PIC_MASTER_IMR); /* DUMMY - (do we need this?) */ 183 inb(PIC_MASTER_IMR); /* DUMMY - (do we need this?) */
178 outb(cached_master_mask, PIC_MASTER_IMR); 184 outb(cached_master_mask, PIC_MASTER_IMR);
179 outb(0x60+irq,PIC_MASTER_CMD); /* 'Specific EOI to master */ 185 outb(0x60+irq, PIC_MASTER_CMD); /* 'Specific EOI to master */
180 } 186 }
181 spin_unlock_irqrestore(&i8259A_lock, flags); 187 spin_unlock_irqrestore(&i8259A_lock, flags);
182 return; 188 return;
@@ -199,7 +205,8 @@ spurious_8259A_irq:
199 * lets ACK and report it. [once per IRQ] 205 * lets ACK and report it. [once per IRQ]
200 */ 206 */
201 if (!(spurious_irq_mask & irqmask)) { 207 if (!(spurious_irq_mask & irqmask)) {
202 printk(KERN_DEBUG "spurious 8259A interrupt: IRQ%d.\n", irq); 208 printk(KERN_DEBUG
209 "spurious 8259A interrupt: IRQ%d.\n", irq);
203 spurious_irq_mask |= irqmask; 210 spurious_irq_mask |= irqmask;
204 } 211 }
205 atomic_inc(&irq_err_count); 212 atomic_inc(&irq_err_count);
@@ -290,17 +297,28 @@ void init_8259A(int auto_eoi)
290 * outb_pic - this has to work on a wide range of PC hardware. 297 * outb_pic - this has to work on a wide range of PC hardware.
291 */ 298 */
292 outb_pic(0x11, PIC_MASTER_CMD); /* ICW1: select 8259A-1 init */ 299 outb_pic(0x11, PIC_MASTER_CMD); /* ICW1: select 8259A-1 init */
293 outb_pic(0x20 + 0, PIC_MASTER_IMR); /* ICW2: 8259A-1 IR0-7 mapped to 0x20-0x27 */ 300
294 outb_pic(1U << PIC_CASCADE_IR, PIC_MASTER_IMR); /* 8259A-1 (the master) has a slave on IR2 */ 301 /* ICW2: 8259A-1 IR0-7 mapped to 0x30-0x37 on x86-64,
302 to 0x20-0x27 on i386 */
303 outb_pic(IRQ0_VECTOR, PIC_MASTER_IMR);
304
305 /* 8259A-1 (the master) has a slave on IR2 */
306 outb_pic(1U << PIC_CASCADE_IR, PIC_MASTER_IMR);
307
295 if (auto_eoi) /* master does Auto EOI */ 308 if (auto_eoi) /* master does Auto EOI */
296 outb_pic(MASTER_ICW4_DEFAULT | PIC_ICW4_AEOI, PIC_MASTER_IMR); 309 outb_pic(MASTER_ICW4_DEFAULT | PIC_ICW4_AEOI, PIC_MASTER_IMR);
297 else /* master expects normal EOI */ 310 else /* master expects normal EOI */
298 outb_pic(MASTER_ICW4_DEFAULT, PIC_MASTER_IMR); 311 outb_pic(MASTER_ICW4_DEFAULT, PIC_MASTER_IMR);
299 312
300 outb_pic(0x11, PIC_SLAVE_CMD); /* ICW1: select 8259A-2 init */ 313 outb_pic(0x11, PIC_SLAVE_CMD); /* ICW1: select 8259A-2 init */
301 outb_pic(0x20 + 8, PIC_SLAVE_IMR); /* ICW2: 8259A-2 IR0-7 mapped to 0x28-0x2f */ 314
302 outb_pic(PIC_CASCADE_IR, PIC_SLAVE_IMR); /* 8259A-2 is a slave on master's IR2 */ 315 /* ICW2: 8259A-2 IR0-7 mapped to IRQ8_VECTOR */
303 outb_pic(SLAVE_ICW4_DEFAULT, PIC_SLAVE_IMR); /* (slave's support for AEOI in flat mode is to be investigated) */ 316 outb_pic(IRQ8_VECTOR, PIC_SLAVE_IMR);
317 /* 8259A-2 is a slave on master's IR2 */
318 outb_pic(PIC_CASCADE_IR, PIC_SLAVE_IMR);
319 /* (slave's support for AEOI in flat mode is to be investigated) */
320 outb_pic(SLAVE_ICW4_DEFAULT, PIC_SLAVE_IMR);
321
304 if (auto_eoi) 322 if (auto_eoi)
305 /* 323 /*
306 * In AEOI mode we just have to mask the interrupt 324 * In AEOI mode we just have to mask the interrupt
@@ -317,93 +335,3 @@ void init_8259A(int auto_eoi)
317 335
318 spin_unlock_irqrestore(&i8259A_lock, flags); 336 spin_unlock_irqrestore(&i8259A_lock, flags);
319} 337}
320
321/*
322 * Note that on a 486, we don't want to do a SIGFPE on an irq13
323 * as the irq is unreliable, and exception 16 works correctly
324 * (ie as explained in the intel literature). On a 386, you
325 * can't use exception 16 due to bad IBM design, so we have to
326 * rely on the less exact irq13.
327 *
328 * Careful.. Not only is IRQ13 unreliable, but it is also
329 * leads to races. IBM designers who came up with it should
330 * be shot.
331 */
332
333
334static irqreturn_t math_error_irq(int cpl, void *dev_id)
335{
336 extern void math_error(void __user *);
337 outb(0,0xF0);
338 if (ignore_fpu_irq || !boot_cpu_data.hard_math)
339 return IRQ_NONE;
340 math_error((void __user *)get_irq_regs()->ip);
341 return IRQ_HANDLED;
342}
343
344/*
345 * New motherboards sometimes make IRQ 13 be a PCI interrupt,
346 * so allow interrupt sharing.
347 */
348static struct irqaction fpu_irq = {
349 .handler = math_error_irq,
350 .mask = CPU_MASK_NONE,
351 .name = "fpu",
352};
353
354void __init init_ISA_irqs (void)
355{
356 int i;
357
358#ifdef CONFIG_X86_LOCAL_APIC
359 init_bsp_APIC();
360#endif
361 init_8259A(0);
362
363 /*
364 * 16 old-style INTA-cycle interrupts:
365 */
366 for (i = 0; i < 16; i++) {
367 set_irq_chip_and_handler_name(i, &i8259A_chip,
368 handle_level_irq, "XT");
369 }
370}
371
372/* Overridden in paravirt.c */
373void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
374
375void __init native_init_IRQ(void)
376{
377 int i;
378
379 /* all the set up before the call gates are initialised */
380 pre_intr_init_hook();
381
382 /*
383 * Cover the whole vector space, no vector can escape
384 * us. (some of these will be overridden and become
385 * 'special' SMP interrupts)
386 */
387 for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
388 int vector = FIRST_EXTERNAL_VECTOR + i;
389 if (i >= NR_IRQS)
390 break;
391 /* SYSCALL_VECTOR was reserved in trap_init. */
392 if (!test_bit(vector, used_vectors))
393 set_intr_gate(vector, interrupt[i]);
394 }
395
396 /* setup after call gates are initialised (usually add in
397 * the architecture specific gates)
398 */
399 intr_init_hook();
400
401 /*
402 * External FPU? Set up irq13 if so, for
403 * original braindamaged IBM FERR coupling.
404 */
405 if (boot_cpu_data.hard_math && !cpu_has_fpu)
406 setup_irq(FPU_IRQ, &fpu_irq);
407
408 irq_ctx_init(smp_processor_id());
409}
diff --git a/arch/x86/kernel/i8259_64.c b/arch/x86/kernel/i8259_64.c
deleted file mode 100644
index fa57a1568508..000000000000
--- a/arch/x86/kernel/i8259_64.c
+++ /dev/null
@@ -1,512 +0,0 @@
1#include <linux/linkage.h>
2#include <linux/errno.h>
3#include <linux/signal.h>
4#include <linux/sched.h>
5#include <linux/ioport.h>
6#include <linux/interrupt.h>
7#include <linux/timex.h>
8#include <linux/slab.h>
9#include <linux/random.h>
10#include <linux/init.h>
11#include <linux/kernel_stat.h>
12#include <linux/sysdev.h>
13#include <linux/bitops.h>
14
15#include <asm/acpi.h>
16#include <asm/atomic.h>
17#include <asm/system.h>
18#include <asm/io.h>
19#include <asm/hw_irq.h>
20#include <asm/pgtable.h>
21#include <asm/delay.h>
22#include <asm/desc.h>
23#include <asm/apic.h>
24#include <asm/i8259.h>
25
26/*
27 * Common place to define all x86 IRQ vectors
28 *
29 * This builds up the IRQ handler stubs using some ugly macros in irq.h
30 *
31 * These macros create the low-level assembly IRQ routines that save
32 * register context and call do_IRQ(). do_IRQ() then does all the
33 * operations that are needed to keep the AT (or SMP IOAPIC)
34 * interrupt-controller happy.
35 */
36
37#define BI(x,y) \
38 BUILD_IRQ(x##y)
39
40#define BUILD_16_IRQS(x) \
41 BI(x,0) BI(x,1) BI(x,2) BI(x,3) \
42 BI(x,4) BI(x,5) BI(x,6) BI(x,7) \
43 BI(x,8) BI(x,9) BI(x,a) BI(x,b) \
44 BI(x,c) BI(x,d) BI(x,e) BI(x,f)
45
46/*
47 * ISA PIC or low IO-APIC triggered (INTA-cycle or APIC) interrupts:
48 * (these are usually mapped to vectors 0x30-0x3f)
49 */
50
51/*
52 * The IO-APIC gives us many more interrupt sources. Most of these
53 * are unused but an SMP system is supposed to have enough memory ...
54 * sometimes (mostly wrt. hw bugs) we get corrupted vectors all
55 * across the spectrum, so we really want to be prepared to get all
56 * of these. Plus, more powerful systems might have more than 64
57 * IO-APIC registers.
58 *
59 * (these are usually mapped into the 0x30-0xff vector range)
60 */
61 BUILD_16_IRQS(0x2) BUILD_16_IRQS(0x3)
62BUILD_16_IRQS(0x4) BUILD_16_IRQS(0x5) BUILD_16_IRQS(0x6) BUILD_16_IRQS(0x7)
63BUILD_16_IRQS(0x8) BUILD_16_IRQS(0x9) BUILD_16_IRQS(0xa) BUILD_16_IRQS(0xb)
64BUILD_16_IRQS(0xc) BUILD_16_IRQS(0xd) BUILD_16_IRQS(0xe) BUILD_16_IRQS(0xf)
65
66#undef BUILD_16_IRQS
67#undef BI
68
69
70#define IRQ(x,y) \
71 IRQ##x##y##_interrupt
72
73#define IRQLIST_16(x) \
74 IRQ(x,0), IRQ(x,1), IRQ(x,2), IRQ(x,3), \
75 IRQ(x,4), IRQ(x,5), IRQ(x,6), IRQ(x,7), \
76 IRQ(x,8), IRQ(x,9), IRQ(x,a), IRQ(x,b), \
77 IRQ(x,c), IRQ(x,d), IRQ(x,e), IRQ(x,f)
78
79/* for the irq vectors */
80static void (*__initdata interrupt[NR_VECTORS - FIRST_EXTERNAL_VECTOR])(void) = {
81 IRQLIST_16(0x2), IRQLIST_16(0x3),
82 IRQLIST_16(0x4), IRQLIST_16(0x5), IRQLIST_16(0x6), IRQLIST_16(0x7),
83 IRQLIST_16(0x8), IRQLIST_16(0x9), IRQLIST_16(0xa), IRQLIST_16(0xb),
84 IRQLIST_16(0xc), IRQLIST_16(0xd), IRQLIST_16(0xe), IRQLIST_16(0xf)
85};
86
87#undef IRQ
88#undef IRQLIST_16
89
90/*
91 * This is the 'legacy' 8259A Programmable Interrupt Controller,
92 * present in the majority of PC/AT boxes.
93 * plus some generic x86 specific things if generic specifics makes
94 * any sense at all.
95 * this file should become arch/i386/kernel/irq.c when the old irq.c
96 * moves to arch independent land
97 */
98
99static int i8259A_auto_eoi;
100DEFINE_SPINLOCK(i8259A_lock);
101static void mask_and_ack_8259A(unsigned int);
102
103static struct irq_chip i8259A_chip = {
104 .name = "XT-PIC",
105 .mask = disable_8259A_irq,
106 .disable = disable_8259A_irq,
107 .unmask = enable_8259A_irq,
108 .mask_ack = mask_and_ack_8259A,
109};
110
111/*
112 * 8259A PIC functions to handle ISA devices:
113 */
114
115/*
116 * This contains the irq mask for both 8259A irq controllers,
117 */
118unsigned int cached_irq_mask = 0xffff;
119
120/*
121 * Not all IRQs can be routed through the IO-APIC, eg. on certain (older)
122 * boards the timer interrupt is not really connected to any IO-APIC pin,
123 * it's fed to the master 8259A's IR0 line only.
124 *
125 * Any '1' bit in this mask means the IRQ is routed through the IO-APIC.
126 * this 'mixed mode' IRQ handling costs nothing because it's only used
127 * at IRQ setup time.
128 */
129unsigned long io_apic_irqs;
130
131void disable_8259A_irq(unsigned int irq)
132{
133 unsigned int mask = 1 << irq;
134 unsigned long flags;
135
136 spin_lock_irqsave(&i8259A_lock, flags);
137 cached_irq_mask |= mask;
138 if (irq & 8)
139 outb(cached_slave_mask, PIC_SLAVE_IMR);
140 else
141 outb(cached_master_mask, PIC_MASTER_IMR);
142 spin_unlock_irqrestore(&i8259A_lock, flags);
143}
144
145void enable_8259A_irq(unsigned int irq)
146{
147 unsigned int mask = ~(1 << irq);
148 unsigned long flags;
149
150 spin_lock_irqsave(&i8259A_lock, flags);
151 cached_irq_mask &= mask;
152 if (irq & 8)
153 outb(cached_slave_mask, PIC_SLAVE_IMR);
154 else
155 outb(cached_master_mask, PIC_MASTER_IMR);
156 spin_unlock_irqrestore(&i8259A_lock, flags);
157}
158
159int i8259A_irq_pending(unsigned int irq)
160{
161 unsigned int mask = 1<<irq;
162 unsigned long flags;
163 int ret;
164
165 spin_lock_irqsave(&i8259A_lock, flags);
166 if (irq < 8)
167 ret = inb(PIC_MASTER_CMD) & mask;
168 else
169 ret = inb(PIC_SLAVE_CMD) & (mask >> 8);
170 spin_unlock_irqrestore(&i8259A_lock, flags);
171
172 return ret;
173}
174
175void make_8259A_irq(unsigned int irq)
176{
177 disable_irq_nosync(irq);
178 io_apic_irqs &= ~(1<<irq);
179 set_irq_chip_and_handler_name(irq, &i8259A_chip, handle_level_irq,
180 "XT");
181 enable_irq(irq);
182}
183
184/*
185 * This function assumes to be called rarely. Switching between
186 * 8259A registers is slow.
187 * This has to be protected by the irq controller spinlock
188 * before being called.
189 */
190static inline int i8259A_irq_real(unsigned int irq)
191{
192 int value;
193 int irqmask = 1<<irq;
194
195 if (irq < 8) {
196 outb(0x0B,PIC_MASTER_CMD); /* ISR register */
197 value = inb(PIC_MASTER_CMD) & irqmask;
198 outb(0x0A,PIC_MASTER_CMD); /* back to the IRR register */
199 return value;
200 }
201 outb(0x0B,PIC_SLAVE_CMD); /* ISR register */
202 value = inb(PIC_SLAVE_CMD) & (irqmask >> 8);
203 outb(0x0A,PIC_SLAVE_CMD); /* back to the IRR register */
204 return value;
205}
206
207/*
208 * Careful! The 8259A is a fragile beast, it pretty
209 * much _has_ to be done exactly like this (mask it
210 * first, _then_ send the EOI, and the order of EOI
211 * to the two 8259s is important!
212 */
213static void mask_and_ack_8259A(unsigned int irq)
214{
215 unsigned int irqmask = 1 << irq;
216 unsigned long flags;
217
218 spin_lock_irqsave(&i8259A_lock, flags);
219 /*
220 * Lightweight spurious IRQ detection. We do not want
221 * to overdo spurious IRQ handling - it's usually a sign
222 * of hardware problems, so we only do the checks we can
223 * do without slowing down good hardware unnecessarily.
224 *
225 * Note that IRQ7 and IRQ15 (the two spurious IRQs
226 * usually resulting from the 8259A-1|2 PICs) occur
227 * even if the IRQ is masked in the 8259A. Thus we
228 * can check spurious 8259A IRQs without doing the
229 * quite slow i8259A_irq_real() call for every IRQ.
230 * This does not cover 100% of spurious interrupts,
231 * but should be enough to warn the user that there
232 * is something bad going on ...
233 */
234 if (cached_irq_mask & irqmask)
235 goto spurious_8259A_irq;
236 cached_irq_mask |= irqmask;
237
238handle_real_irq:
239 if (irq & 8) {
240 inb(PIC_SLAVE_IMR); /* DUMMY - (do we need this?) */
241 outb(cached_slave_mask, PIC_SLAVE_IMR);
242 /* 'Specific EOI' to slave */
243 outb(0x60+(irq&7),PIC_SLAVE_CMD);
244 /* 'Specific EOI' to master-IRQ2 */
245 outb(0x60+PIC_CASCADE_IR,PIC_MASTER_CMD);
246 } else {
247 inb(PIC_MASTER_IMR); /* DUMMY - (do we need this?) */
248 outb(cached_master_mask, PIC_MASTER_IMR);
249 /* 'Specific EOI' to master */
250 outb(0x60+irq,PIC_MASTER_CMD);
251 }
252 spin_unlock_irqrestore(&i8259A_lock, flags);
253 return;
254
255spurious_8259A_irq:
256 /*
257 * this is the slow path - should happen rarely.
258 */
259 if (i8259A_irq_real(irq))
260 /*
261 * oops, the IRQ _is_ in service according to the
262 * 8259A - not spurious, go handle it.
263 */
264 goto handle_real_irq;
265
266 {
267 static int spurious_irq_mask;
268 /*
269 * At this point we can be sure the IRQ is spurious,
270 * lets ACK and report it. [once per IRQ]
271 */
272 if (!(spurious_irq_mask & irqmask)) {
273 printk(KERN_DEBUG
274 "spurious 8259A interrupt: IRQ%d.\n", irq);
275 spurious_irq_mask |= irqmask;
276 }
277 atomic_inc(&irq_err_count);
278 /*
279 * Theoretically we do not have to handle this IRQ,
280 * but in Linux this does not cause problems and is
281 * simpler for us.
282 */
283 goto handle_real_irq;
284 }
285}
286
287static char irq_trigger[2];
288/**
289 * ELCR registers (0x4d0, 0x4d1) control edge/level of IRQ
290 */
291static void restore_ELCR(char *trigger)
292{
293 outb(trigger[0], 0x4d0);
294 outb(trigger[1], 0x4d1);
295}
296
297static void save_ELCR(char *trigger)
298{
299 /* IRQ 0,1,2,8,13 are marked as reserved */
300 trigger[0] = inb(0x4d0) & 0xF8;
301 trigger[1] = inb(0x4d1) & 0xDE;
302}
303
304static int i8259A_resume(struct sys_device *dev)
305{
306 init_8259A(i8259A_auto_eoi);
307 restore_ELCR(irq_trigger);
308 return 0;
309}
310
311static int i8259A_suspend(struct sys_device *dev, pm_message_t state)
312{
313 save_ELCR(irq_trigger);
314 return 0;
315}
316
317static int i8259A_shutdown(struct sys_device *dev)
318{
319 /* Put the i8259A into a quiescent state that
320 * the kernel initialization code can get it
321 * out of.
322 */
323 outb(0xff, PIC_MASTER_IMR); /* mask all of 8259A-1 */
324 outb(0xff, PIC_SLAVE_IMR); /* mask all of 8259A-1 */
325 return 0;
326}
327
328static struct sysdev_class i8259_sysdev_class = {
329 .name = "i8259",
330 .suspend = i8259A_suspend,
331 .resume = i8259A_resume,
332 .shutdown = i8259A_shutdown,
333};
334
335static struct sys_device device_i8259A = {
336 .id = 0,
337 .cls = &i8259_sysdev_class,
338};
339
340static int __init i8259A_init_sysfs(void)
341{
342 int error = sysdev_class_register(&i8259_sysdev_class);
343 if (!error)
344 error = sysdev_register(&device_i8259A);
345 return error;
346}
347
348device_initcall(i8259A_init_sysfs);
349
350void init_8259A(int auto_eoi)
351{
352 unsigned long flags;
353
354 i8259A_auto_eoi = auto_eoi;
355
356 spin_lock_irqsave(&i8259A_lock, flags);
357
358 outb(0xff, PIC_MASTER_IMR); /* mask all of 8259A-1 */
359 outb(0xff, PIC_SLAVE_IMR); /* mask all of 8259A-2 */
360
361 /*
362 * outb_pic - this has to work on a wide range of PC hardware.
363 */
364 outb_pic(0x11, PIC_MASTER_CMD); /* ICW1: select 8259A-1 init */
365 /* ICW2: 8259A-1 IR0-7 mapped to 0x30-0x37 */
366 outb_pic(IRQ0_VECTOR, PIC_MASTER_IMR);
367 /* 8259A-1 (the master) has a slave on IR2 */
368 outb_pic(0x04, PIC_MASTER_IMR);
369 if (auto_eoi) /* master does Auto EOI */
370 outb_pic(MASTER_ICW4_DEFAULT | PIC_ICW4_AEOI, PIC_MASTER_IMR);
371 else /* master expects normal EOI */
372 outb_pic(MASTER_ICW4_DEFAULT, PIC_MASTER_IMR);
373
374 outb_pic(0x11, PIC_SLAVE_CMD); /* ICW1: select 8259A-2 init */
375 /* ICW2: 8259A-2 IR0-7 mapped to 0x38-0x3f */
376 outb_pic(IRQ8_VECTOR, PIC_SLAVE_IMR);
377 /* 8259A-2 is a slave on master's IR2 */
378 outb_pic(PIC_CASCADE_IR, PIC_SLAVE_IMR);
379 /* (slave's support for AEOI in flat mode is to be investigated) */
380 outb_pic(SLAVE_ICW4_DEFAULT, PIC_SLAVE_IMR);
381
382 if (auto_eoi)
383 /*
384 * In AEOI mode we just have to mask the interrupt
385 * when acking.
386 */
387 i8259A_chip.mask_ack = disable_8259A_irq;
388 else
389 i8259A_chip.mask_ack = mask_and_ack_8259A;
390
391 udelay(100); /* wait for 8259A to initialize */
392
393 outb(cached_master_mask, PIC_MASTER_IMR); /* restore master IRQ mask */
394 outb(cached_slave_mask, PIC_SLAVE_IMR); /* restore slave IRQ mask */
395
396 spin_unlock_irqrestore(&i8259A_lock, flags);
397}
398
399
400
401
402/*
403 * IRQ2 is cascade interrupt to second interrupt controller
404 */
405
406static struct irqaction irq2 = {
407 .handler = no_action,
408 .mask = CPU_MASK_NONE,
409 .name = "cascade",
410};
411DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
412 [0 ... IRQ0_VECTOR - 1] = -1,
413 [IRQ0_VECTOR] = 0,
414 [IRQ1_VECTOR] = 1,
415 [IRQ2_VECTOR] = 2,
416 [IRQ3_VECTOR] = 3,
417 [IRQ4_VECTOR] = 4,
418 [IRQ5_VECTOR] = 5,
419 [IRQ6_VECTOR] = 6,
420 [IRQ7_VECTOR] = 7,
421 [IRQ8_VECTOR] = 8,
422 [IRQ9_VECTOR] = 9,
423 [IRQ10_VECTOR] = 10,
424 [IRQ11_VECTOR] = 11,
425 [IRQ12_VECTOR] = 12,
426 [IRQ13_VECTOR] = 13,
427 [IRQ14_VECTOR] = 14,
428 [IRQ15_VECTOR] = 15,
429 [IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
430};
431
432void __init init_ISA_irqs (void)
433{
434 int i;
435
436 init_bsp_APIC();
437 init_8259A(0);
438
439 for (i = 0; i < NR_IRQS; i++) {
440 irq_desc[i].status = IRQ_DISABLED;
441 irq_desc[i].action = NULL;
442 irq_desc[i].depth = 1;
443
444 if (i < 16) {
445 /*
446 * 16 old-style INTA-cycle interrupts:
447 */
448 set_irq_chip_and_handler_name(i, &i8259A_chip,
449 handle_level_irq, "XT");
450 } else {
451 /*
452 * 'high' PCI IRQs filled in on demand
453 */
454 irq_desc[i].chip = &no_irq_chip;
455 }
456 }
457}
458
459void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
460
461void __init native_init_IRQ(void)
462{
463 int i;
464
465 init_ISA_irqs();
466 /*
467 * Cover the whole vector space, no vector can escape
468 * us. (some of these will be overridden and become
469 * 'special' SMP interrupts)
470 */
471 for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
472 int vector = FIRST_EXTERNAL_VECTOR + i;
473 if (vector != IA32_SYSCALL_VECTOR)
474 set_intr_gate(vector, interrupt[i]);
475 }
476
477#ifdef CONFIG_SMP
478 /*
479 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
480 * IPI, driven by wakeup.
481 */
482 set_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
483
484 /* IPIs for invalidation */
485 set_intr_gate(INVALIDATE_TLB_VECTOR_START+0, invalidate_interrupt0);
486 set_intr_gate(INVALIDATE_TLB_VECTOR_START+1, invalidate_interrupt1);
487 set_intr_gate(INVALIDATE_TLB_VECTOR_START+2, invalidate_interrupt2);
488 set_intr_gate(INVALIDATE_TLB_VECTOR_START+3, invalidate_interrupt3);
489 set_intr_gate(INVALIDATE_TLB_VECTOR_START+4, invalidate_interrupt4);
490 set_intr_gate(INVALIDATE_TLB_VECTOR_START+5, invalidate_interrupt5);
491 set_intr_gate(INVALIDATE_TLB_VECTOR_START+6, invalidate_interrupt6);
492 set_intr_gate(INVALIDATE_TLB_VECTOR_START+7, invalidate_interrupt7);
493
494 /* IPI for generic function call */
495 set_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
496
497 /* Low priority IPI to cleanup after moving an irq */
498 set_intr_gate(IRQ_MOVE_CLEANUP_VECTOR, irq_move_cleanup_interrupt);
499#endif
500 set_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
501 set_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt);
502
503 /* self generated IPI for local APIC timer */
504 set_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
505
506 /* IPI vectors for APIC spurious and error interrupts */
507 set_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
508 set_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
509
510 if (!acpi_ioapic)
511 setup_irq(2, &irq2);
512}
diff --git a/arch/x86/kernel/io_apic_32.c b/arch/x86/kernel/io_apic_32.c
index 4dc8600d9d20..de9aa0e3a9c5 100644
--- a/arch/x86/kernel/io_apic_32.c
+++ b/arch/x86/kernel/io_apic_32.c
@@ -25,6 +25,7 @@
25#include <linux/init.h> 25#include <linux/init.h>
26#include <linux/delay.h> 26#include <linux/delay.h>
27#include <linux/sched.h> 27#include <linux/sched.h>
28#include <linux/bootmem.h>
28#include <linux/mc146818rtc.h> 29#include <linux/mc146818rtc.h>
29#include <linux/compiler.h> 30#include <linux/compiler.h>
30#include <linux/acpi.h> 31#include <linux/acpi.h>
@@ -58,7 +59,7 @@ static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
58static DEFINE_SPINLOCK(ioapic_lock); 59static DEFINE_SPINLOCK(ioapic_lock);
59static DEFINE_SPINLOCK(vector_lock); 60static DEFINE_SPINLOCK(vector_lock);
60 61
61int timer_over_8254 __initdata = 1; 62int timer_through_8259 __initdata;
62 63
63/* 64/*
64 * Is the SiS APIC rmw bug present ? 65 * Is the SiS APIC rmw bug present ?
@@ -72,15 +73,21 @@ int sis_apic_bug = -1;
72int nr_ioapic_registers[MAX_IO_APICS]; 73int nr_ioapic_registers[MAX_IO_APICS];
73 74
74/* I/O APIC entries */ 75/* I/O APIC entries */
75struct mpc_config_ioapic mp_ioapics[MAX_IO_APICS]; 76struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
76int nr_ioapics; 77int nr_ioapics;
77 78
78/* MP IRQ source entries */ 79/* MP IRQ source entries */
79struct mpc_config_intsrc mp_irqs[MAX_IRQ_SOURCES]; 80struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
80 81
81/* # of MP IRQ source entries */ 82/* # of MP IRQ source entries */
82int mp_irq_entries; 83int mp_irq_entries;
83 84
85#if defined (CONFIG_MCA) || defined (CONFIG_EISA)
86int mp_bus_id_to_type[MAX_MP_BUSSES];
87#endif
88
89DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
90
84static int disable_timer_pin_1 __initdata; 91static int disable_timer_pin_1 __initdata;
85 92
86/* 93/*
@@ -110,7 +117,7 @@ struct io_apic {
110static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx) 117static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
111{ 118{
112 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx) 119 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
113 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK); 120 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
114} 121}
115 122
116static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg) 123static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
@@ -239,7 +246,7 @@ static void __init replace_pin_at_irq(unsigned int irq,
239 } 246 }
240} 247}
241 248
242static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable) 249static void __modify_IO_APIC_irq(unsigned int irq, unsigned long enable, unsigned long disable)
243{ 250{
244 struct irq_pin_list *entry = irq_2_pin + irq; 251 struct irq_pin_list *entry = irq_2_pin + irq;
245 unsigned int pin, reg; 252 unsigned int pin, reg;
@@ -259,30 +266,32 @@ static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsign
259} 266}
260 267
261/* mask = 1 */ 268/* mask = 1 */
262static void __mask_IO_APIC_irq (unsigned int irq) 269static void __mask_IO_APIC_irq(unsigned int irq)
263{ 270{
264 __modify_IO_APIC_irq(irq, 0x00010000, 0); 271 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_MASKED, 0);
265} 272}
266 273
267/* mask = 0 */ 274/* mask = 0 */
268static void __unmask_IO_APIC_irq (unsigned int irq) 275static void __unmask_IO_APIC_irq(unsigned int irq)
269{ 276{
270 __modify_IO_APIC_irq(irq, 0, 0x00010000); 277 __modify_IO_APIC_irq(irq, 0, IO_APIC_REDIR_MASKED);
271} 278}
272 279
273/* mask = 1, trigger = 0 */ 280/* mask = 1, trigger = 0 */
274static void __mask_and_edge_IO_APIC_irq (unsigned int irq) 281static void __mask_and_edge_IO_APIC_irq(unsigned int irq)
275{ 282{
276 __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000); 283 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_MASKED,
284 IO_APIC_REDIR_LEVEL_TRIGGER);
277} 285}
278 286
279/* mask = 0, trigger = 1 */ 287/* mask = 0, trigger = 1 */
280static void __unmask_and_level_IO_APIC_irq (unsigned int irq) 288static void __unmask_and_level_IO_APIC_irq(unsigned int irq)
281{ 289{
282 __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000); 290 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_LEVEL_TRIGGER,
291 IO_APIC_REDIR_MASKED);
283} 292}
284 293
285static void mask_IO_APIC_irq (unsigned int irq) 294static void mask_IO_APIC_irq(unsigned int irq)
286{ 295{
287 unsigned long flags; 296 unsigned long flags;
288 297
@@ -291,7 +300,7 @@ static void mask_IO_APIC_irq (unsigned int irq)
291 spin_unlock_irqrestore(&ioapic_lock, flags); 300 spin_unlock_irqrestore(&ioapic_lock, flags);
292} 301}
293 302
294static void unmask_IO_APIC_irq (unsigned int irq) 303static void unmask_IO_APIC_irq(unsigned int irq)
295{ 304{
296 unsigned long flags; 305 unsigned long flags;
297 306
@@ -303,7 +312,7 @@ static void unmask_IO_APIC_irq (unsigned int irq)
303static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin) 312static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
304{ 313{
305 struct IO_APIC_route_entry entry; 314 struct IO_APIC_route_entry entry;
306 315
307 /* Check delivery_mode to be sure we're not clearing an SMI pin */ 316 /* Check delivery_mode to be sure we're not clearing an SMI pin */
308 entry = ioapic_read_entry(apic, pin); 317 entry = ioapic_read_entry(apic, pin);
309 if (entry.delivery_mode == dest_SMI) 318 if (entry.delivery_mode == dest_SMI)
@@ -315,7 +324,7 @@ static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
315 ioapic_mask_entry(apic, pin); 324 ioapic_mask_entry(apic, pin);
316} 325}
317 326
318static void clear_IO_APIC (void) 327static void clear_IO_APIC(void)
319{ 328{
320 int apic, pin; 329 int apic, pin;
321 330
@@ -332,7 +341,7 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
332 struct irq_pin_list *entry = irq_2_pin + irq; 341 struct irq_pin_list *entry = irq_2_pin + irq;
333 unsigned int apicid_value; 342 unsigned int apicid_value;
334 cpumask_t tmp; 343 cpumask_t tmp;
335 344
336 cpus_and(tmp, cpumask, cpu_online_map); 345 cpus_and(tmp, cpumask, cpu_online_map);
337 if (cpus_empty(tmp)) 346 if (cpus_empty(tmp))
338 tmp = TARGET_CPUS; 347 tmp = TARGET_CPUS;
@@ -361,7 +370,7 @@ static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
361# include <linux/kernel_stat.h> /* kstat */ 370# include <linux/kernel_stat.h> /* kstat */
362# include <linux/slab.h> /* kmalloc() */ 371# include <linux/slab.h> /* kmalloc() */
363# include <linux/timer.h> 372# include <linux/timer.h>
364 373
365#define IRQBALANCE_CHECK_ARCH -999 374#define IRQBALANCE_CHECK_ARCH -999
366#define MAX_BALANCED_IRQ_INTERVAL (5*HZ) 375#define MAX_BALANCED_IRQ_INTERVAL (5*HZ)
367#define MIN_BALANCED_IRQ_INTERVAL (HZ/2) 376#define MIN_BALANCED_IRQ_INTERVAL (HZ/2)
@@ -373,14 +382,14 @@ static int physical_balance __read_mostly;
373static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL; 382static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL;
374 383
375static struct irq_cpu_info { 384static struct irq_cpu_info {
376 unsigned long * last_irq; 385 unsigned long *last_irq;
377 unsigned long * irq_delta; 386 unsigned long *irq_delta;
378 unsigned long irq; 387 unsigned long irq;
379} irq_cpu_data[NR_CPUS]; 388} irq_cpu_data[NR_CPUS];
380 389
381#define CPU_IRQ(cpu) (irq_cpu_data[cpu].irq) 390#define CPU_IRQ(cpu) (irq_cpu_data[cpu].irq)
382#define LAST_CPU_IRQ(cpu,irq) (irq_cpu_data[cpu].last_irq[irq]) 391#define LAST_CPU_IRQ(cpu, irq) (irq_cpu_data[cpu].last_irq[irq])
383#define IRQ_DELTA(cpu,irq) (irq_cpu_data[cpu].irq_delta[irq]) 392#define IRQ_DELTA(cpu, irq) (irq_cpu_data[cpu].irq_delta[irq])
384 393
385#define IDLE_ENOUGH(cpu,now) \ 394#define IDLE_ENOUGH(cpu,now) \
386 (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1)) 395 (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
@@ -419,8 +428,8 @@ inside:
419 if (cpu == -1) 428 if (cpu == -1)
420 cpu = NR_CPUS-1; 429 cpu = NR_CPUS-1;
421 } 430 }
422 } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) || 431 } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu, allowed_mask) ||
423 (search_idle && !IDLE_ENOUGH(cpu,now))); 432 (search_idle && !IDLE_ENOUGH(cpu, now)));
424 433
425 return cpu; 434 return cpu;
426} 435}
@@ -430,15 +439,14 @@ static inline void balance_irq(int cpu, int irq)
430 unsigned long now = jiffies; 439 unsigned long now = jiffies;
431 cpumask_t allowed_mask; 440 cpumask_t allowed_mask;
432 unsigned int new_cpu; 441 unsigned int new_cpu;
433 442
434 if (irqbalance_disabled) 443 if (irqbalance_disabled)
435 return; 444 return;
436 445
437 cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]); 446 cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]);
438 new_cpu = move(cpu, allowed_mask, now, 1); 447 new_cpu = move(cpu, allowed_mask, now, 1);
439 if (cpu != new_cpu) { 448 if (cpu != new_cpu)
440 set_pending_irq(irq, cpumask_of_cpu(new_cpu)); 449 set_pending_irq(irq, cpumask_of_cpu(new_cpu));
441 }
442} 450}
443 451
444static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold) 452static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
@@ -450,14 +458,14 @@ static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
450 if (!irq_desc[j].action) 458 if (!irq_desc[j].action)
451 continue; 459 continue;
452 /* Is it a significant load ? */ 460 /* Is it a significant load ? */
453 if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) < 461 if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i), j) <
454 useful_load_threshold) 462 useful_load_threshold)
455 continue; 463 continue;
456 balance_irq(i, j); 464 balance_irq(i, j);
457 } 465 }
458 } 466 }
459 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL, 467 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
460 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA); 468 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
461 return; 469 return;
462} 470}
463 471
@@ -486,22 +494,22 @@ static void do_irq_balance(void)
486 /* Is this an active IRQ or balancing disabled ? */ 494 /* Is this an active IRQ or balancing disabled ? */
487 if (!irq_desc[j].action || irq_balancing_disabled(j)) 495 if (!irq_desc[j].action || irq_balancing_disabled(j))
488 continue; 496 continue;
489 if ( package_index == i ) 497 if (package_index == i)
490 IRQ_DELTA(package_index,j) = 0; 498 IRQ_DELTA(package_index, j) = 0;
491 /* Determine the total count per processor per IRQ */ 499 /* Determine the total count per processor per IRQ */
492 value_now = (unsigned long) kstat_cpu(i).irqs[j]; 500 value_now = (unsigned long) kstat_cpu(i).irqs[j];
493 501
494 /* Determine the activity per processor per IRQ */ 502 /* Determine the activity per processor per IRQ */
495 delta = value_now - LAST_CPU_IRQ(i,j); 503 delta = value_now - LAST_CPU_IRQ(i, j);
496 504
497 /* Update last_cpu_irq[][] for the next time */ 505 /* Update last_cpu_irq[][] for the next time */
498 LAST_CPU_IRQ(i,j) = value_now; 506 LAST_CPU_IRQ(i, j) = value_now;
499 507
500 /* Ignore IRQs whose rate is less than the clock */ 508 /* Ignore IRQs whose rate is less than the clock */
501 if (delta < useful_load_threshold) 509 if (delta < useful_load_threshold)
502 continue; 510 continue;
503 /* update the load for the processor or package total */ 511 /* update the load for the processor or package total */
504 IRQ_DELTA(package_index,j) += delta; 512 IRQ_DELTA(package_index, j) += delta;
505 513
506 /* Keep track of the higher numbered sibling as well */ 514 /* Keep track of the higher numbered sibling as well */
507 if (i != package_index) 515 if (i != package_index)
@@ -527,7 +535,8 @@ static void do_irq_balance(void)
527 max_cpu_irq = ULONG_MAX; 535 max_cpu_irq = ULONG_MAX;
528 536
529tryanothercpu: 537tryanothercpu:
530 /* Look for heaviest loaded processor. 538 /*
539 * Look for heaviest loaded processor.
531 * We may come back to get the next heaviest loaded processor. 540 * We may come back to get the next heaviest loaded processor.
532 * Skip processors with trivial loads. 541 * Skip processors with trivial loads.
533 */ 542 */
@@ -536,7 +545,7 @@ tryanothercpu:
536 for_each_online_cpu(i) { 545 for_each_online_cpu(i) {
537 if (i != CPU_TO_PACKAGEINDEX(i)) 546 if (i != CPU_TO_PACKAGEINDEX(i))
538 continue; 547 continue;
539 if (max_cpu_irq <= CPU_IRQ(i)) 548 if (max_cpu_irq <= CPU_IRQ(i))
540 continue; 549 continue;
541 if (tmp_cpu_irq < CPU_IRQ(i)) { 550 if (tmp_cpu_irq < CPU_IRQ(i)) {
542 tmp_cpu_irq = CPU_IRQ(i); 551 tmp_cpu_irq = CPU_IRQ(i);
@@ -545,8 +554,9 @@ tryanothercpu:
545 } 554 }
546 555
547 if (tmp_loaded == -1) { 556 if (tmp_loaded == -1) {
548 /* In the case of small number of heavy interrupt sources, 557 /*
549 * loading some of the cpus too much. We use Ingo's original 558 * In the case of small number of heavy interrupt sources,
559 * loading some of the cpus too much. We use Ingo's original
550 * approach to rotate them around. 560 * approach to rotate them around.
551 */ 561 */
552 if (!first_attempt && imbalance >= useful_load_threshold) { 562 if (!first_attempt && imbalance >= useful_load_threshold) {
@@ -555,13 +565,14 @@ tryanothercpu:
555 } 565 }
556 goto not_worth_the_effort; 566 goto not_worth_the_effort;
557 } 567 }
558 568
559 first_attempt = 0; /* heaviest search */ 569 first_attempt = 0; /* heaviest search */
560 max_cpu_irq = tmp_cpu_irq; /* load */ 570 max_cpu_irq = tmp_cpu_irq; /* load */
561 max_loaded = tmp_loaded; /* processor */ 571 max_loaded = tmp_loaded; /* processor */
562 imbalance = (max_cpu_irq - min_cpu_irq) / 2; 572 imbalance = (max_cpu_irq - min_cpu_irq) / 2;
563 573
564 /* if imbalance is less than approx 10% of max load, then 574 /*
575 * if imbalance is less than approx 10% of max load, then
565 * observe diminishing returns action. - quit 576 * observe diminishing returns action. - quit
566 */ 577 */
567 if (imbalance < (max_cpu_irq >> 3)) 578 if (imbalance < (max_cpu_irq >> 3))
@@ -577,26 +588,25 @@ tryanotherirq:
577 /* Is this an active IRQ? */ 588 /* Is this an active IRQ? */
578 if (!irq_desc[j].action) 589 if (!irq_desc[j].action)
579 continue; 590 continue;
580 if (imbalance <= IRQ_DELTA(max_loaded,j)) 591 if (imbalance <= IRQ_DELTA(max_loaded, j))
581 continue; 592 continue;
582 /* Try to find the IRQ that is closest to the imbalance 593 /* Try to find the IRQ that is closest to the imbalance
583 * without going over. 594 * without going over.
584 */ 595 */
585 if (move_this_load < IRQ_DELTA(max_loaded,j)) { 596 if (move_this_load < IRQ_DELTA(max_loaded, j)) {
586 move_this_load = IRQ_DELTA(max_loaded,j); 597 move_this_load = IRQ_DELTA(max_loaded, j);
587 selected_irq = j; 598 selected_irq = j;
588 } 599 }
589 } 600 }
590 if (selected_irq == -1) { 601 if (selected_irq == -1)
591 goto tryanothercpu; 602 goto tryanothercpu;
592 }
593 603
594 imbalance = move_this_load; 604 imbalance = move_this_load;
595 605
596 /* For physical_balance case, we accumulated both load 606 /* For physical_balance case, we accumulated both load
597 * values in the one of the siblings cpu_irq[], 607 * values in the one of the siblings cpu_irq[],
598 * to use the same code for physical and logical processors 608 * to use the same code for physical and logical processors
599 * as much as possible. 609 * as much as possible.
600 * 610 *
601 * NOTE: the cpu_irq[] array holds the sum of the load for 611 * NOTE: the cpu_irq[] array holds the sum of the load for
602 * sibling A and sibling B in the slot for the lowest numbered 612 * sibling A and sibling B in the slot for the lowest numbered
@@ -625,11 +635,11 @@ tryanotherirq:
625 /* mark for change destination */ 635 /* mark for change destination */
626 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded)); 636 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded));
627 637
628 /* Since we made a change, come back sooner to 638 /* Since we made a change, come back sooner to
629 * check for more variation. 639 * check for more variation.
630 */ 640 */
631 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL, 641 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
632 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA); 642 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
633 return; 643 return;
634 } 644 }
635 goto tryanotherirq; 645 goto tryanotherirq;
@@ -640,7 +650,7 @@ not_worth_the_effort:
640 * upward 650 * upward
641 */ 651 */
642 balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL, 652 balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
643 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA); 653 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
644 return; 654 return;
645} 655}
646 656
@@ -679,13 +689,13 @@ static int __init balanced_irq_init(void)
679 cpumask_t tmp; 689 cpumask_t tmp;
680 690
681 cpus_shift_right(tmp, cpu_online_map, 2); 691 cpus_shift_right(tmp, cpu_online_map, 2);
682 c = &boot_cpu_data; 692 c = &boot_cpu_data;
683 /* When not overwritten by the command line ask subarchitecture. */ 693 /* When not overwritten by the command line ask subarchitecture. */
684 if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH) 694 if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
685 irqbalance_disabled = NO_BALANCE_IRQ; 695 irqbalance_disabled = NO_BALANCE_IRQ;
686 if (irqbalance_disabled) 696 if (irqbalance_disabled)
687 return 0; 697 return 0;
688 698
689 /* disable irqbalance completely if there is only one processor online */ 699 /* disable irqbalance completely if there is only one processor online */
690 if (num_online_cpus() < 2) { 700 if (num_online_cpus() < 2) {
691 irqbalance_disabled = 1; 701 irqbalance_disabled = 1;
@@ -699,16 +709,14 @@ static int __init balanced_irq_init(void)
699 physical_balance = 1; 709 physical_balance = 1;
700 710
701 for_each_online_cpu(i) { 711 for_each_online_cpu(i) {
702 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL); 712 irq_cpu_data[i].irq_delta = kzalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
703 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL); 713 irq_cpu_data[i].last_irq = kzalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
704 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) { 714 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
705 printk(KERN_ERR "balanced_irq_init: out of memory"); 715 printk(KERN_ERR "balanced_irq_init: out of memory");
706 goto failed; 716 goto failed;
707 } 717 }
708 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
709 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
710 } 718 }
711 719
712 printk(KERN_INFO "Starting balanced_irq\n"); 720 printk(KERN_INFO "Starting balanced_irq\n");
713 if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd"))) 721 if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd")))
714 return 0; 722 return 0;
@@ -748,7 +756,7 @@ void send_IPI_self(int vector)
748 /* 756 /*
749 * Send the IPI. The write to APIC_ICR fires this off. 757 * Send the IPI. The write to APIC_ICR fires this off.
750 */ 758 */
751 apic_write_around(APIC_ICR, cfg); 759 apic_write(APIC_ICR, cfg);
752} 760}
753#endif /* !CONFIG_SMP */ 761#endif /* !CONFIG_SMP */
754 762
@@ -801,10 +809,10 @@ static int find_irq_entry(int apic, int pin, int type)
801 int i; 809 int i;
802 810
803 for (i = 0; i < mp_irq_entries; i++) 811 for (i = 0; i < mp_irq_entries; i++)
804 if (mp_irqs[i].mpc_irqtype == type && 812 if (mp_irqs[i].mp_irqtype == type &&
805 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid || 813 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
806 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) && 814 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
807 mp_irqs[i].mpc_dstirq == pin) 815 mp_irqs[i].mp_dstirq == pin)
808 return i; 816 return i;
809 817
810 return -1; 818 return -1;
@@ -818,13 +826,13 @@ static int __init find_isa_irq_pin(int irq, int type)
818 int i; 826 int i;
819 827
820 for (i = 0; i < mp_irq_entries; i++) { 828 for (i = 0; i < mp_irq_entries; i++) {
821 int lbus = mp_irqs[i].mpc_srcbus; 829 int lbus = mp_irqs[i].mp_srcbus;
822 830
823 if (test_bit(lbus, mp_bus_not_pci) && 831 if (test_bit(lbus, mp_bus_not_pci) &&
824 (mp_irqs[i].mpc_irqtype == type) && 832 (mp_irqs[i].mp_irqtype == type) &&
825 (mp_irqs[i].mpc_srcbusirq == irq)) 833 (mp_irqs[i].mp_srcbusirq == irq))
826 834
827 return mp_irqs[i].mpc_dstirq; 835 return mp_irqs[i].mp_dstirq;
828 } 836 }
829 return -1; 837 return -1;
830} 838}
@@ -834,17 +842,17 @@ static int __init find_isa_irq_apic(int irq, int type)
834 int i; 842 int i;
835 843
836 for (i = 0; i < mp_irq_entries; i++) { 844 for (i = 0; i < mp_irq_entries; i++) {
837 int lbus = mp_irqs[i].mpc_srcbus; 845 int lbus = mp_irqs[i].mp_srcbus;
838 846
839 if (test_bit(lbus, mp_bus_not_pci) && 847 if (test_bit(lbus, mp_bus_not_pci) &&
840 (mp_irqs[i].mpc_irqtype == type) && 848 (mp_irqs[i].mp_irqtype == type) &&
841 (mp_irqs[i].mpc_srcbusirq == irq)) 849 (mp_irqs[i].mp_srcbusirq == irq))
842 break; 850 break;
843 } 851 }
844 if (i < mp_irq_entries) { 852 if (i < mp_irq_entries) {
845 int apic; 853 int apic;
846 for(apic = 0; apic < nr_ioapics; apic++) { 854 for (apic = 0; apic < nr_ioapics; apic++) {
847 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic) 855 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
848 return apic; 856 return apic;
849 } 857 }
850 } 858 }
@@ -864,28 +872,28 @@ int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
864 872
865 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, " 873 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
866 "slot:%d, pin:%d.\n", bus, slot, pin); 874 "slot:%d, pin:%d.\n", bus, slot, pin);
867 if (mp_bus_id_to_pci_bus[bus] == -1) { 875 if (test_bit(bus, mp_bus_not_pci)) {
868 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus); 876 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
869 return -1; 877 return -1;
870 } 878 }
871 for (i = 0; i < mp_irq_entries; i++) { 879 for (i = 0; i < mp_irq_entries; i++) {
872 int lbus = mp_irqs[i].mpc_srcbus; 880 int lbus = mp_irqs[i].mp_srcbus;
873 881
874 for (apic = 0; apic < nr_ioapics; apic++) 882 for (apic = 0; apic < nr_ioapics; apic++)
875 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic || 883 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
876 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) 884 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
877 break; 885 break;
878 886
879 if (!test_bit(lbus, mp_bus_not_pci) && 887 if (!test_bit(lbus, mp_bus_not_pci) &&
880 !mp_irqs[i].mpc_irqtype && 888 !mp_irqs[i].mp_irqtype &&
881 (bus == lbus) && 889 (bus == lbus) &&
882 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) { 890 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
883 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq); 891 int irq = pin_2_irq(i, apic, mp_irqs[i].mp_dstirq);
884 892
885 if (!(apic || IO_APIC_IRQ(irq))) 893 if (!(apic || IO_APIC_IRQ(irq)))
886 continue; 894 continue;
887 895
888 if (pin == (mp_irqs[i].mpc_srcbusirq & 3)) 896 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
889 return irq; 897 return irq;
890 /* 898 /*
891 * Use the first all-but-pin matching entry as a 899 * Use the first all-but-pin matching entry as a
@@ -900,7 +908,7 @@ int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
900EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector); 908EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
901 909
902/* 910/*
903 * This function currently is only a helper for the i386 smp boot process where 911 * This function currently is only a helper for the i386 smp boot process where
904 * we need to reprogram the ioredtbls to cater for the cpus which have come online 912 * we need to reprogram the ioredtbls to cater for the cpus which have come online
905 * so mask in all cases should simply be TARGET_CPUS 913 * so mask in all cases should simply be TARGET_CPUS
906 */ 914 */
@@ -952,7 +960,7 @@ static int EISA_ELCR(unsigned int irq)
952 * EISA conforming in the MP table, that means its trigger type must 960 * EISA conforming in the MP table, that means its trigger type must
953 * be read in from the ELCR */ 961 * be read in from the ELCR */
954 962
955#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq)) 963#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
956#define default_EISA_polarity(idx) default_ISA_polarity(idx) 964#define default_EISA_polarity(idx) default_ISA_polarity(idx)
957 965
958/* PCI interrupts are always polarity one level triggered, 966/* PCI interrupts are always polarity one level triggered,
@@ -969,118 +977,115 @@ static int EISA_ELCR(unsigned int irq)
969 977
970static int MPBIOS_polarity(int idx) 978static int MPBIOS_polarity(int idx)
971{ 979{
972 int bus = mp_irqs[idx].mpc_srcbus; 980 int bus = mp_irqs[idx].mp_srcbus;
973 int polarity; 981 int polarity;
974 982
975 /* 983 /*
976 * Determine IRQ line polarity (high active or low active): 984 * Determine IRQ line polarity (high active or low active):
977 */ 985 */
978 switch (mp_irqs[idx].mpc_irqflag & 3) 986 switch (mp_irqs[idx].mp_irqflag & 3) {
987 case 0: /* conforms, ie. bus-type dependent polarity */
979 { 988 {
980 case 0: /* conforms, ie. bus-type dependent polarity */ 989 polarity = test_bit(bus, mp_bus_not_pci)?
981 { 990 default_ISA_polarity(idx):
982 polarity = test_bit(bus, mp_bus_not_pci)? 991 default_PCI_polarity(idx);
983 default_ISA_polarity(idx): 992 break;
984 default_PCI_polarity(idx); 993 }
985 break; 994 case 1: /* high active */
986 } 995 {
987 case 1: /* high active */ 996 polarity = 0;
988 { 997 break;
989 polarity = 0; 998 }
990 break; 999 case 2: /* reserved */
991 } 1000 {
992 case 2: /* reserved */ 1001 printk(KERN_WARNING "broken BIOS!!\n");
993 { 1002 polarity = 1;
994 printk(KERN_WARNING "broken BIOS!!\n"); 1003 break;
995 polarity = 1; 1004 }
996 break; 1005 case 3: /* low active */
997 } 1006 {
998 case 3: /* low active */ 1007 polarity = 1;
999 { 1008 break;
1000 polarity = 1; 1009 }
1001 break; 1010 default: /* invalid */
1002 } 1011 {
1003 default: /* invalid */ 1012 printk(KERN_WARNING "broken BIOS!!\n");
1004 { 1013 polarity = 1;
1005 printk(KERN_WARNING "broken BIOS!!\n"); 1014 break;
1006 polarity = 1; 1015 }
1007 break;
1008 }
1009 } 1016 }
1010 return polarity; 1017 return polarity;
1011} 1018}
1012 1019
1013static int MPBIOS_trigger(int idx) 1020static int MPBIOS_trigger(int idx)
1014{ 1021{
1015 int bus = mp_irqs[idx].mpc_srcbus; 1022 int bus = mp_irqs[idx].mp_srcbus;
1016 int trigger; 1023 int trigger;
1017 1024
1018 /* 1025 /*
1019 * Determine IRQ trigger mode (edge or level sensitive): 1026 * Determine IRQ trigger mode (edge or level sensitive):
1020 */ 1027 */
1021 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3) 1028 switch ((mp_irqs[idx].mp_irqflag>>2) & 3) {
1029 case 0: /* conforms, ie. bus-type dependent */
1022 { 1030 {
1023 case 0: /* conforms, ie. bus-type dependent */ 1031 trigger = test_bit(bus, mp_bus_not_pci)?
1024 { 1032 default_ISA_trigger(idx):
1025 trigger = test_bit(bus, mp_bus_not_pci)? 1033 default_PCI_trigger(idx);
1026 default_ISA_trigger(idx):
1027 default_PCI_trigger(idx);
1028#if defined(CONFIG_EISA) || defined(CONFIG_MCA) 1034#if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1029 switch (mp_bus_id_to_type[bus]) 1035 switch (mp_bus_id_to_type[bus]) {
1030 { 1036 case MP_BUS_ISA: /* ISA pin */
1031 case MP_BUS_ISA: /* ISA pin */ 1037 {
1032 { 1038 /* set before the switch */
1033 /* set before the switch */
1034 break;
1035 }
1036 case MP_BUS_EISA: /* EISA pin */
1037 {
1038 trigger = default_EISA_trigger(idx);
1039 break;
1040 }
1041 case MP_BUS_PCI: /* PCI pin */
1042 {
1043 /* set before the switch */
1044 break;
1045 }
1046 case MP_BUS_MCA: /* MCA pin */
1047 {
1048 trigger = default_MCA_trigger(idx);
1049 break;
1050 }
1051 default:
1052 {
1053 printk(KERN_WARNING "broken BIOS!!\n");
1054 trigger = 1;
1055 break;
1056 }
1057 }
1058#endif
1059 break; 1039 break;
1060 } 1040 }
1061 case 1: /* edge */ 1041 case MP_BUS_EISA: /* EISA pin */
1062 { 1042 {
1063 trigger = 0; 1043 trigger = default_EISA_trigger(idx);
1064 break; 1044 break;
1065 } 1045 }
1066 case 2: /* reserved */ 1046 case MP_BUS_PCI: /* PCI pin */
1067 { 1047 {
1068 printk(KERN_WARNING "broken BIOS!!\n"); 1048 /* set before the switch */
1069 trigger = 1;
1070 break; 1049 break;
1071 } 1050 }
1072 case 3: /* level */ 1051 case MP_BUS_MCA: /* MCA pin */
1073 { 1052 {
1074 trigger = 1; 1053 trigger = default_MCA_trigger(idx);
1075 break; 1054 break;
1076 } 1055 }
1077 default: /* invalid */ 1056 default:
1078 { 1057 {
1079 printk(KERN_WARNING "broken BIOS!!\n"); 1058 printk(KERN_WARNING "broken BIOS!!\n");
1080 trigger = 0; 1059 trigger = 1;
1081 break; 1060 break;
1082 } 1061 }
1083 } 1062 }
1063#endif
1064 break;
1065 }
1066 case 1: /* edge */
1067 {
1068 trigger = 0;
1069 break;
1070 }
1071 case 2: /* reserved */
1072 {
1073 printk(KERN_WARNING "broken BIOS!!\n");
1074 trigger = 1;
1075 break;
1076 }
1077 case 3: /* level */
1078 {
1079 trigger = 1;
1080 break;
1081 }
1082 default: /* invalid */
1083 {
1084 printk(KERN_WARNING "broken BIOS!!\n");
1085 trigger = 0;
1086 break;
1087 }
1088 }
1084 return trigger; 1089 return trigger;
1085} 1090}
1086 1091
@@ -1097,16 +1102,16 @@ static inline int irq_trigger(int idx)
1097static int pin_2_irq(int idx, int apic, int pin) 1102static int pin_2_irq(int idx, int apic, int pin)
1098{ 1103{
1099 int irq, i; 1104 int irq, i;
1100 int bus = mp_irqs[idx].mpc_srcbus; 1105 int bus = mp_irqs[idx].mp_srcbus;
1101 1106
1102 /* 1107 /*
1103 * Debugging check, we are in big trouble if this message pops up! 1108 * Debugging check, we are in big trouble if this message pops up!
1104 */ 1109 */
1105 if (mp_irqs[idx].mpc_dstirq != pin) 1110 if (mp_irqs[idx].mp_dstirq != pin)
1106 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n"); 1111 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1107 1112
1108 if (test_bit(bus, mp_bus_not_pci)) 1113 if (test_bit(bus, mp_bus_not_pci))
1109 irq = mp_irqs[idx].mpc_srcbusirq; 1114 irq = mp_irqs[idx].mp_srcbusirq;
1110 else { 1115 else {
1111 /* 1116 /*
1112 * PCI IRQs are mapped in order 1117 * PCI IRQs are mapped in order
@@ -1148,8 +1153,8 @@ static inline int IO_APIC_irq_trigger(int irq)
1148 1153
1149 for (apic = 0; apic < nr_ioapics; apic++) { 1154 for (apic = 0; apic < nr_ioapics; apic++) {
1150 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) { 1155 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1151 idx = find_irq_entry(apic,pin,mp_INT); 1156 idx = find_irq_entry(apic, pin, mp_INT);
1152 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin))) 1157 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1153 return irq_trigger(idx); 1158 return irq_trigger(idx);
1154 } 1159 }
1155 } 1160 }
@@ -1164,7 +1169,7 @@ static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 }
1164 1169
1165static int __assign_irq_vector(int irq) 1170static int __assign_irq_vector(int irq)
1166{ 1171{
1167 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0; 1172 static int current_vector = FIRST_DEVICE_VECTOR, current_offset;
1168 int vector, offset; 1173 int vector, offset;
1169 1174
1170 BUG_ON((unsigned)irq >= NR_IRQ_VECTORS); 1175 BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
@@ -1176,7 +1181,7 @@ static int __assign_irq_vector(int irq)
1176 offset = current_offset; 1181 offset = current_offset;
1177next: 1182next:
1178 vector += 8; 1183 vector += 8;
1179 if (vector >= FIRST_SYSTEM_VECTOR) { 1184 if (vector >= first_system_vector) {
1180 offset = (offset + 1) % 8; 1185 offset = (offset + 1) % 8;
1181 vector = FIRST_DEVICE_VECTOR + offset; 1186 vector = FIRST_DEVICE_VECTOR + offset;
1182 } 1187 }
@@ -1203,6 +1208,11 @@ static int assign_irq_vector(int irq)
1203 1208
1204 return vector; 1209 return vector;
1205} 1210}
1211
1212void setup_vector_irq(int cpu)
1213{
1214}
1215
1206static struct irq_chip ioapic_chip; 1216static struct irq_chip ioapic_chip;
1207 1217
1208#define IOAPIC_AUTO -1 1218#define IOAPIC_AUTO -1
@@ -1237,25 +1247,25 @@ static void __init setup_IO_APIC_irqs(void)
1237 /* 1247 /*
1238 * add it to the IO-APIC irq-routing table: 1248 * add it to the IO-APIC irq-routing table:
1239 */ 1249 */
1240 memset(&entry,0,sizeof(entry)); 1250 memset(&entry, 0, sizeof(entry));
1241 1251
1242 entry.delivery_mode = INT_DELIVERY_MODE; 1252 entry.delivery_mode = INT_DELIVERY_MODE;
1243 entry.dest_mode = INT_DEST_MODE; 1253 entry.dest_mode = INT_DEST_MODE;
1244 entry.mask = 0; /* enable IRQ */ 1254 entry.mask = 0; /* enable IRQ */
1245 entry.dest.logical.logical_dest = 1255 entry.dest.logical.logical_dest =
1246 cpu_mask_to_apicid(TARGET_CPUS); 1256 cpu_mask_to_apicid(TARGET_CPUS);
1247 1257
1248 idx = find_irq_entry(apic,pin,mp_INT); 1258 idx = find_irq_entry(apic, pin, mp_INT);
1249 if (idx == -1) { 1259 if (idx == -1) {
1250 if (first_notcon) { 1260 if (first_notcon) {
1251 apic_printk(APIC_VERBOSE, KERN_DEBUG 1261 apic_printk(APIC_VERBOSE, KERN_DEBUG
1252 " IO-APIC (apicid-pin) %d-%d", 1262 " IO-APIC (apicid-pin) %d-%d",
1253 mp_ioapics[apic].mpc_apicid, 1263 mp_ioapics[apic].mp_apicid,
1254 pin); 1264 pin);
1255 first_notcon = 0; 1265 first_notcon = 0;
1256 } else 1266 } else
1257 apic_printk(APIC_VERBOSE, ", %d-%d", 1267 apic_printk(APIC_VERBOSE, ", %d-%d",
1258 mp_ioapics[apic].mpc_apicid, pin); 1268 mp_ioapics[apic].mp_apicid, pin);
1259 continue; 1269 continue;
1260 } 1270 }
1261 1271
@@ -1289,7 +1299,7 @@ static void __init setup_IO_APIC_irqs(void)
1289 vector = assign_irq_vector(irq); 1299 vector = assign_irq_vector(irq);
1290 entry.vector = vector; 1300 entry.vector = vector;
1291 ioapic_register_intr(irq, vector, IOAPIC_AUTO); 1301 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1292 1302
1293 if (!apic && (irq < 16)) 1303 if (!apic && (irq < 16))
1294 disable_8259A_irq(irq); 1304 disable_8259A_irq(irq);
1295 } 1305 }
@@ -1302,25 +1312,21 @@ static void __init setup_IO_APIC_irqs(void)
1302} 1312}
1303 1313
1304/* 1314/*
1305 * Set up the 8259A-master output pin: 1315 * Set up the timer pin, possibly with the 8259A-master behind.
1306 */ 1316 */
1307static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector) 1317static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1318 int vector)
1308{ 1319{
1309 struct IO_APIC_route_entry entry; 1320 struct IO_APIC_route_entry entry;
1310 1321
1311 memset(&entry,0,sizeof(entry)); 1322 memset(&entry, 0, sizeof(entry));
1312
1313 disable_8259A_irq(0);
1314
1315 /* mask LVT0 */
1316 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1317 1323
1318 /* 1324 /*
1319 * We use logical delivery to get the timer IRQ 1325 * We use logical delivery to get the timer IRQ
1320 * to the first CPU. 1326 * to the first CPU.
1321 */ 1327 */
1322 entry.dest_mode = INT_DEST_MODE; 1328 entry.dest_mode = INT_DEST_MODE;
1323 entry.mask = 0; /* unmask IRQ now */ 1329 entry.mask = 1; /* mask IRQ now */
1324 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS); 1330 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1325 entry.delivery_mode = INT_DELIVERY_MODE; 1331 entry.delivery_mode = INT_DELIVERY_MODE;
1326 entry.polarity = 0; 1332 entry.polarity = 0;
@@ -1329,17 +1335,14 @@ static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, in
1329 1335
1330 /* 1336 /*
1331 * The timer IRQ doesn't have to know that behind the 1337 * The timer IRQ doesn't have to know that behind the
1332 * scene we have a 8259A-master in AEOI mode ... 1338 * scene we may have a 8259A-master in AEOI mode ...
1333 */ 1339 */
1334 irq_desc[0].chip = &ioapic_chip; 1340 ioapic_register_intr(0, vector, IOAPIC_EDGE);
1335 set_irq_handler(0, handle_edge_irq);
1336 1341
1337 /* 1342 /*
1338 * Add it to the IO-APIC irq-routing table: 1343 * Add it to the IO-APIC irq-routing table:
1339 */ 1344 */
1340 ioapic_write_entry(apic, pin, entry); 1345 ioapic_write_entry(apic, pin, entry);
1341
1342 enable_8259A_irq(0);
1343} 1346}
1344 1347
1345void __init print_IO_APIC(void) 1348void __init print_IO_APIC(void)
@@ -1354,10 +1357,10 @@ void __init print_IO_APIC(void)
1354 if (apic_verbosity == APIC_QUIET) 1357 if (apic_verbosity == APIC_QUIET)
1355 return; 1358 return;
1356 1359
1357 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries); 1360 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1358 for (i = 0; i < nr_ioapics; i++) 1361 for (i = 0; i < nr_ioapics; i++)
1359 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n", 1362 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1360 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]); 1363 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1361 1364
1362 /* 1365 /*
1363 * We are a bit conservative about what we expect. We have to 1366 * We are a bit conservative about what we expect. We have to
@@ -1376,7 +1379,7 @@ void __init print_IO_APIC(void)
1376 reg_03.raw = io_apic_read(apic, 3); 1379 reg_03.raw = io_apic_read(apic, 3);
1377 spin_unlock_irqrestore(&ioapic_lock, flags); 1380 spin_unlock_irqrestore(&ioapic_lock, flags);
1378 1381
1379 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid); 1382 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1380 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw); 1383 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1381 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID); 1384 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1382 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type); 1385 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
@@ -1459,7 +1462,7 @@ void __init print_IO_APIC(void)
1459 1462
1460#if 0 1463#if 0
1461 1464
1462static void print_APIC_bitfield (int base) 1465static void print_APIC_bitfield(int base)
1463{ 1466{
1464 unsigned int v; 1467 unsigned int v;
1465 int i, j; 1468 int i, j;
@@ -1480,7 +1483,7 @@ static void print_APIC_bitfield (int base)
1480 } 1483 }
1481} 1484}
1482 1485
1483void /*__init*/ print_local_APIC(void * dummy) 1486void /*__init*/ print_local_APIC(void *dummy)
1484{ 1487{
1485 unsigned int v, ver, maxlvt; 1488 unsigned int v, ver, maxlvt;
1486 1489
@@ -1489,6 +1492,7 @@ void /*__init*/ print_local_APIC(void * dummy)
1489 1492
1490 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n", 1493 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1491 smp_processor_id(), hard_smp_processor_id()); 1494 smp_processor_id(), hard_smp_processor_id());
1495 v = apic_read(APIC_ID);
1492 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, 1496 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v,
1493 GET_APIC_ID(read_apic_id())); 1497 GET_APIC_ID(read_apic_id()));
1494 v = apic_read(APIC_LVR); 1498 v = apic_read(APIC_LVR);
@@ -1563,9 +1567,9 @@ void /*__init*/ print_local_APIC(void * dummy)
1563 printk("\n"); 1567 printk("\n");
1564} 1568}
1565 1569
1566void print_all_local_APICs (void) 1570void print_all_local_APICs(void)
1567{ 1571{
1568 on_each_cpu(print_local_APIC, NULL, 1, 1); 1572 on_each_cpu(print_local_APIC, NULL, 1);
1569} 1573}
1570 1574
1571void /*__init*/ print_PIC(void) 1575void /*__init*/ print_PIC(void)
@@ -1586,11 +1590,11 @@ void /*__init*/ print_PIC(void)
1586 v = inb(0xa0) << 8 | inb(0x20); 1590 v = inb(0xa0) << 8 | inb(0x20);
1587 printk(KERN_DEBUG "... PIC IRR: %04x\n", v); 1591 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1588 1592
1589 outb(0x0b,0xa0); 1593 outb(0x0b, 0xa0);
1590 outb(0x0b,0x20); 1594 outb(0x0b, 0x20);
1591 v = inb(0xa0) << 8 | inb(0x20); 1595 v = inb(0xa0) << 8 | inb(0x20);
1592 outb(0x0a,0xa0); 1596 outb(0x0a, 0xa0);
1593 outb(0x0a,0x20); 1597 outb(0x0a, 0x20);
1594 1598
1595 spin_unlock_irqrestore(&i8259A_lock, flags); 1599 spin_unlock_irqrestore(&i8259A_lock, flags);
1596 1600
@@ -1626,7 +1630,7 @@ static void __init enable_IO_APIC(void)
1626 spin_unlock_irqrestore(&ioapic_lock, flags); 1630 spin_unlock_irqrestore(&ioapic_lock, flags);
1627 nr_ioapic_registers[apic] = reg_01.bits.entries+1; 1631 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1628 } 1632 }
1629 for(apic = 0; apic < nr_ioapics; apic++) { 1633 for (apic = 0; apic < nr_ioapics; apic++) {
1630 int pin; 1634 int pin;
1631 /* See if any of the pins is in ExtINT mode */ 1635 /* See if any of the pins is in ExtINT mode */
1632 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) { 1636 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
@@ -1716,7 +1720,6 @@ void disable_IO_APIC(void)
1716 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999 1720 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1717 */ 1721 */
1718 1722
1719#ifndef CONFIG_X86_NUMAQ
1720static void __init setup_ioapic_ids_from_mpc(void) 1723static void __init setup_ioapic_ids_from_mpc(void)
1721{ 1724{
1722 union IO_APIC_reg_00 reg_00; 1725 union IO_APIC_reg_00 reg_00;
@@ -1726,6 +1729,11 @@ static void __init setup_ioapic_ids_from_mpc(void)
1726 unsigned char old_id; 1729 unsigned char old_id;
1727 unsigned long flags; 1730 unsigned long flags;
1728 1731
1732#ifdef CONFIG_X86_NUMAQ
1733 if (found_numaq)
1734 return;
1735#endif
1736
1729 /* 1737 /*
1730 * Don't check I/O APIC IDs for xAPIC systems. They have 1738 * Don't check I/O APIC IDs for xAPIC systems. They have
1731 * no meaning without the serial APIC bus. 1739 * no meaning without the serial APIC bus.
@@ -1748,15 +1756,15 @@ static void __init setup_ioapic_ids_from_mpc(void)
1748 spin_lock_irqsave(&ioapic_lock, flags); 1756 spin_lock_irqsave(&ioapic_lock, flags);
1749 reg_00.raw = io_apic_read(apic, 0); 1757 reg_00.raw = io_apic_read(apic, 0);
1750 spin_unlock_irqrestore(&ioapic_lock, flags); 1758 spin_unlock_irqrestore(&ioapic_lock, flags);
1751
1752 old_id = mp_ioapics[apic].mpc_apicid;
1753 1759
1754 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) { 1760 old_id = mp_ioapics[apic].mp_apicid;
1761
1762 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
1755 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n", 1763 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1756 apic, mp_ioapics[apic].mpc_apicid); 1764 apic, mp_ioapics[apic].mp_apicid);
1757 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n", 1765 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1758 reg_00.bits.ID); 1766 reg_00.bits.ID);
1759 mp_ioapics[apic].mpc_apicid = reg_00.bits.ID; 1767 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
1760 } 1768 }
1761 1769
1762 /* 1770 /*
@@ -1765,9 +1773,9 @@ static void __init setup_ioapic_ids_from_mpc(void)
1765 * 'stuck on smp_invalidate_needed IPI wait' messages. 1773 * 'stuck on smp_invalidate_needed IPI wait' messages.
1766 */ 1774 */
1767 if (check_apicid_used(phys_id_present_map, 1775 if (check_apicid_used(phys_id_present_map,
1768 mp_ioapics[apic].mpc_apicid)) { 1776 mp_ioapics[apic].mp_apicid)) {
1769 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n", 1777 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1770 apic, mp_ioapics[apic].mpc_apicid); 1778 apic, mp_ioapics[apic].mp_apicid);
1771 for (i = 0; i < get_physical_broadcast(); i++) 1779 for (i = 0; i < get_physical_broadcast(); i++)
1772 if (!physid_isset(i, phys_id_present_map)) 1780 if (!physid_isset(i, phys_id_present_map))
1773 break; 1781 break;
@@ -1776,13 +1784,13 @@ static void __init setup_ioapic_ids_from_mpc(void)
1776 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n", 1784 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1777 i); 1785 i);
1778 physid_set(i, phys_id_present_map); 1786 physid_set(i, phys_id_present_map);
1779 mp_ioapics[apic].mpc_apicid = i; 1787 mp_ioapics[apic].mp_apicid = i;
1780 } else { 1788 } else {
1781 physid_mask_t tmp; 1789 physid_mask_t tmp;
1782 tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid); 1790 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
1783 apic_printk(APIC_VERBOSE, "Setting %d in the " 1791 apic_printk(APIC_VERBOSE, "Setting %d in the "
1784 "phys_id_present_map\n", 1792 "phys_id_present_map\n",
1785 mp_ioapics[apic].mpc_apicid); 1793 mp_ioapics[apic].mp_apicid);
1786 physids_or(phys_id_present_map, phys_id_present_map, tmp); 1794 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1787 } 1795 }
1788 1796
@@ -1791,21 +1799,21 @@ static void __init setup_ioapic_ids_from_mpc(void)
1791 * We need to adjust the IRQ routing table 1799 * We need to adjust the IRQ routing table
1792 * if the ID changed. 1800 * if the ID changed.
1793 */ 1801 */
1794 if (old_id != mp_ioapics[apic].mpc_apicid) 1802 if (old_id != mp_ioapics[apic].mp_apicid)
1795 for (i = 0; i < mp_irq_entries; i++) 1803 for (i = 0; i < mp_irq_entries; i++)
1796 if (mp_irqs[i].mpc_dstapic == old_id) 1804 if (mp_irqs[i].mp_dstapic == old_id)
1797 mp_irqs[i].mpc_dstapic 1805 mp_irqs[i].mp_dstapic
1798 = mp_ioapics[apic].mpc_apicid; 1806 = mp_ioapics[apic].mp_apicid;
1799 1807
1800 /* 1808 /*
1801 * Read the right value from the MPC table and 1809 * Read the right value from the MPC table and
1802 * write it into the ID register. 1810 * write it into the ID register.
1803 */ 1811 */
1804 apic_printk(APIC_VERBOSE, KERN_INFO 1812 apic_printk(APIC_VERBOSE, KERN_INFO
1805 "...changing IO-APIC physical APIC ID to %d ...", 1813 "...changing IO-APIC physical APIC ID to %d ...",
1806 mp_ioapics[apic].mpc_apicid); 1814 mp_ioapics[apic].mp_apicid);
1807 1815
1808 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid; 1816 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
1809 spin_lock_irqsave(&ioapic_lock, flags); 1817 spin_lock_irqsave(&ioapic_lock, flags);
1810 io_apic_write(apic, 0, reg_00.raw); 1818 io_apic_write(apic, 0, reg_00.raw);
1811 spin_unlock_irqrestore(&ioapic_lock, flags); 1819 spin_unlock_irqrestore(&ioapic_lock, flags);
@@ -1816,15 +1824,12 @@ static void __init setup_ioapic_ids_from_mpc(void)
1816 spin_lock_irqsave(&ioapic_lock, flags); 1824 spin_lock_irqsave(&ioapic_lock, flags);
1817 reg_00.raw = io_apic_read(apic, 0); 1825 reg_00.raw = io_apic_read(apic, 0);
1818 spin_unlock_irqrestore(&ioapic_lock, flags); 1826 spin_unlock_irqrestore(&ioapic_lock, flags);
1819 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid) 1827 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
1820 printk("could not set ID!\n"); 1828 printk("could not set ID!\n");
1821 else 1829 else
1822 apic_printk(APIC_VERBOSE, " ok.\n"); 1830 apic_printk(APIC_VERBOSE, " ok.\n");
1823 } 1831 }
1824} 1832}
1825#else
1826static void __init setup_ioapic_ids_from_mpc(void) { }
1827#endif
1828 1833
1829int no_timer_check __initdata; 1834int no_timer_check __initdata;
1830 1835
@@ -2015,45 +2020,53 @@ static inline void init_IO_APIC_traps(void)
2015 * The local APIC irq-chip implementation: 2020 * The local APIC irq-chip implementation:
2016 */ 2021 */
2017 2022
2018static void ack_apic(unsigned int irq) 2023static void ack_lapic_irq(unsigned int irq)
2019{ 2024{
2020 ack_APIC_irq(); 2025 ack_APIC_irq();
2021} 2026}
2022 2027
2023static void mask_lapic_irq (unsigned int irq) 2028static void mask_lapic_irq(unsigned int irq)
2024{ 2029{
2025 unsigned long v; 2030 unsigned long v;
2026 2031
2027 v = apic_read(APIC_LVT0); 2032 v = apic_read(APIC_LVT0);
2028 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED); 2033 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2029} 2034}
2030 2035
2031static void unmask_lapic_irq (unsigned int irq) 2036static void unmask_lapic_irq(unsigned int irq)
2032{ 2037{
2033 unsigned long v; 2038 unsigned long v;
2034 2039
2035 v = apic_read(APIC_LVT0); 2040 v = apic_read(APIC_LVT0);
2036 apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED); 2041 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2037} 2042}
2038 2043
2039static struct irq_chip lapic_chip __read_mostly = { 2044static struct irq_chip lapic_chip __read_mostly = {
2040 .name = "local-APIC-edge", 2045 .name = "local-APIC",
2041 .mask = mask_lapic_irq, 2046 .mask = mask_lapic_irq,
2042 .unmask = unmask_lapic_irq, 2047 .unmask = unmask_lapic_irq,
2043 .eoi = ack_apic, 2048 .ack = ack_lapic_irq,
2044}; 2049};
2045 2050
2051static void lapic_register_intr(int irq, int vector)
2052{
2053 irq_desc[irq].status &= ~IRQ_LEVEL;
2054 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2055 "edge");
2056 set_intr_gate(vector, interrupt[irq]);
2057}
2058
2046static void __init setup_nmi(void) 2059static void __init setup_nmi(void)
2047{ 2060{
2048 /* 2061 /*
2049 * Dirty trick to enable the NMI watchdog ... 2062 * Dirty trick to enable the NMI watchdog ...
2050 * We put the 8259A master into AEOI mode and 2063 * We put the 8259A master into AEOI mode and
2051 * unmask on all local APICs LVT0 as NMI. 2064 * unmask on all local APICs LVT0 as NMI.
2052 * 2065 *
2053 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire') 2066 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2054 * is from Maciej W. Rozycki - so we do not have to EOI from 2067 * is from Maciej W. Rozycki - so we do not have to EOI from
2055 * the NMI handler or the timer interrupt. 2068 * the NMI handler or the timer interrupt.
2056 */ 2069 */
2057 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ..."); 2070 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2058 2071
2059 enable_NMI_through_LVT0(); 2072 enable_NMI_through_LVT0();
@@ -2129,11 +2142,16 @@ static inline void __init unlock_ExtINT_logic(void)
2129static inline void __init check_timer(void) 2142static inline void __init check_timer(void)
2130{ 2143{
2131 int apic1, pin1, apic2, pin2; 2144 int apic1, pin1, apic2, pin2;
2145 int no_pin1 = 0;
2132 int vector; 2146 int vector;
2147 unsigned int ver;
2133 unsigned long flags; 2148 unsigned long flags;
2134 2149
2135 local_irq_save(flags); 2150 local_irq_save(flags);
2136 2151
2152 ver = apic_read(APIC_LVR);
2153 ver = GET_APIC_VERSION(ver);
2154
2137 /* 2155 /*
2138 * get/set the timer IRQ vector: 2156 * get/set the timer IRQ vector:
2139 */ 2157 */
@@ -2142,34 +2160,54 @@ static inline void __init check_timer(void)
2142 set_intr_gate(vector, interrupt[0]); 2160 set_intr_gate(vector, interrupt[0]);
2143 2161
2144 /* 2162 /*
2145 * Subtle, code in do_timer_interrupt() expects an AEOI 2163 * As IRQ0 is to be enabled in the 8259A, the virtual
2146 * mode for the 8259A whenever interrupts are routed 2164 * wire has to be disabled in the local APIC. Also
2147 * through I/O APICs. Also IRQ0 has to be enabled in 2165 * timer interrupts need to be acknowledged manually in
2148 * the 8259A which implies the virtual wire has to be 2166 * the 8259A for the i82489DX when using the NMI
2149 * disabled in the local APIC. 2167 * watchdog as that APIC treats NMIs as level-triggered.
2168 * The AEOI mode will finish them in the 8259A
2169 * automatically.
2150 */ 2170 */
2151 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT); 2171 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2152 init_8259A(1); 2172 init_8259A(1);
2153 timer_ack = 1; 2173 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2154 if (timer_over_8254 > 0)
2155 enable_8259A_irq(0);
2156 2174
2157 pin1 = find_isa_irq_pin(0, mp_INT); 2175 pin1 = find_isa_irq_pin(0, mp_INT);
2158 apic1 = find_isa_irq_apic(0, mp_INT); 2176 apic1 = find_isa_irq_apic(0, mp_INT);
2159 pin2 = ioapic_i8259.pin; 2177 pin2 = ioapic_i8259.pin;
2160 apic2 = ioapic_i8259.apic; 2178 apic2 = ioapic_i8259.apic;
2161 2179
2162 printk(KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n", 2180 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2163 vector, apic1, pin1, apic2, pin2); 2181 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2182 vector, apic1, pin1, apic2, pin2);
2183
2184 /*
2185 * Some BIOS writers are clueless and report the ExtINTA
2186 * I/O APIC input from the cascaded 8259A as the timer
2187 * interrupt input. So just in case, if only one pin
2188 * was found above, try it both directly and through the
2189 * 8259A.
2190 */
2191 if (pin1 == -1) {
2192 pin1 = pin2;
2193 apic1 = apic2;
2194 no_pin1 = 1;
2195 } else if (pin2 == -1) {
2196 pin2 = pin1;
2197 apic2 = apic1;
2198 }
2164 2199
2165 if (pin1 != -1) { 2200 if (pin1 != -1) {
2166 /* 2201 /*
2167 * Ok, does IRQ0 through the IOAPIC work? 2202 * Ok, does IRQ0 through the IOAPIC work?
2168 */ 2203 */
2204 if (no_pin1) {
2205 add_pin_to_irq(0, apic1, pin1);
2206 setup_timer_IRQ0_pin(apic1, pin1, vector);
2207 }
2169 unmask_IO_APIC_irq(0); 2208 unmask_IO_APIC_irq(0);
2170 if (timer_irq_works()) { 2209 if (timer_irq_works()) {
2171 if (nmi_watchdog == NMI_IO_APIC) { 2210 if (nmi_watchdog == NMI_IO_APIC) {
2172 disable_8259A_irq(0);
2173 setup_nmi(); 2211 setup_nmi();
2174 enable_8259A_irq(0); 2212 enable_8259A_irq(0);
2175 } 2213 }
@@ -2178,81 +2216,97 @@ static inline void __init check_timer(void)
2178 goto out; 2216 goto out;
2179 } 2217 }
2180 clear_IO_APIC_pin(apic1, pin1); 2218 clear_IO_APIC_pin(apic1, pin1);
2181 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to " 2219 if (!no_pin1)
2182 "IO-APIC\n"); 2220 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2183 } 2221 "8254 timer not connected to IO-APIC\n");
2184 2222
2185 printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... "); 2223 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2186 if (pin2 != -1) { 2224 "(IRQ0) through the 8259A ...\n");
2187 printk("\n..... (found pin %d) ...", pin2); 2225 apic_printk(APIC_QUIET, KERN_INFO
2226 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2188 /* 2227 /*
2189 * legacy devices should be connected to IO APIC #0 2228 * legacy devices should be connected to IO APIC #0
2190 */ 2229 */
2191 setup_ExtINT_IRQ0_pin(apic2, pin2, vector); 2230 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2231 setup_timer_IRQ0_pin(apic2, pin2, vector);
2232 unmask_IO_APIC_irq(0);
2233 enable_8259A_irq(0);
2192 if (timer_irq_works()) { 2234 if (timer_irq_works()) {
2193 printk("works.\n"); 2235 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2194 if (pin1 != -1) 2236 timer_through_8259 = 1;
2195 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2196 else
2197 add_pin_to_irq(0, apic2, pin2);
2198 if (nmi_watchdog == NMI_IO_APIC) { 2237 if (nmi_watchdog == NMI_IO_APIC) {
2238 disable_8259A_irq(0);
2199 setup_nmi(); 2239 setup_nmi();
2240 enable_8259A_irq(0);
2200 } 2241 }
2201 goto out; 2242 goto out;
2202 } 2243 }
2203 /* 2244 /*
2204 * Cleanup, just in case ... 2245 * Cleanup, just in case ...
2205 */ 2246 */
2247 disable_8259A_irq(0);
2206 clear_IO_APIC_pin(apic2, pin2); 2248 clear_IO_APIC_pin(apic2, pin2);
2249 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2207 } 2250 }
2208 printk(" failed.\n");
2209 2251
2210 if (nmi_watchdog == NMI_IO_APIC) { 2252 if (nmi_watchdog == NMI_IO_APIC) {
2211 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n"); 2253 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2212 nmi_watchdog = 0; 2254 "through the IO-APIC - disabling NMI Watchdog!\n");
2255 nmi_watchdog = NMI_NONE;
2213 } 2256 }
2257 timer_ack = 0;
2214 2258
2215 printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ..."); 2259 apic_printk(APIC_QUIET, KERN_INFO
2260 "...trying to set up timer as Virtual Wire IRQ...\n");
2216 2261
2217 disable_8259A_irq(0); 2262 lapic_register_intr(0, vector);
2218 set_irq_chip_and_handler_name(0, &lapic_chip, handle_fasteoi_irq, 2263 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2219 "fasteoi");
2220 apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2221 enable_8259A_irq(0); 2264 enable_8259A_irq(0);
2222 2265
2223 if (timer_irq_works()) { 2266 if (timer_irq_works()) {
2224 printk(" works.\n"); 2267 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2225 goto out; 2268 goto out;
2226 } 2269 }
2227 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector); 2270 disable_8259A_irq(0);
2228 printk(" failed.\n"); 2271 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2272 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2229 2273
2230 printk(KERN_INFO "...trying to set up timer as ExtINT IRQ..."); 2274 apic_printk(APIC_QUIET, KERN_INFO
2275 "...trying to set up timer as ExtINT IRQ...\n");
2231 2276
2232 timer_ack = 0;
2233 init_8259A(0); 2277 init_8259A(0);
2234 make_8259A_irq(0); 2278 make_8259A_irq(0);
2235 apic_write_around(APIC_LVT0, APIC_DM_EXTINT); 2279 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2236 2280
2237 unlock_ExtINT_logic(); 2281 unlock_ExtINT_logic();
2238 2282
2239 if (timer_irq_works()) { 2283 if (timer_irq_works()) {
2240 printk(" works.\n"); 2284 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2241 goto out; 2285 goto out;
2242 } 2286 }
2243 printk(" failed :(.\n"); 2287 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2244 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a " 2288 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2245 "report. Then try booting with the 'noapic' option"); 2289 "report. Then try booting with the 'noapic' option.\n");
2246out: 2290out:
2247 local_irq_restore(flags); 2291 local_irq_restore(flags);
2248} 2292}
2249 2293
2250/* 2294/*
2251 * 2295 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2252 * IRQ's that are handled by the PIC in the MPS IOAPIC case. 2296 * to devices. However there may be an I/O APIC pin available for
2253 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ. 2297 * this interrupt regardless. The pin may be left unconnected, but
2254 * Linux doesn't really care, as it's not actually used 2298 * typically it will be reused as an ExtINT cascade interrupt for
2255 * for any interrupt handling anyway. 2299 * the master 8259A. In the MPS case such a pin will normally be
2300 * reported as an ExtINT interrupt in the MP table. With ACPI
2301 * there is no provision for ExtINT interrupts, and in the absence
2302 * of an override it would be treated as an ordinary ISA I/O APIC
2303 * interrupt, that is edge-triggered and unmasked by default. We
2304 * used to do this, but it caused problems on some systems because
2305 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2306 * the same ExtINT cascade interrupt to drive the local APIC of the
2307 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2308 * the I/O APIC in all cases now. No actual device should request
2309 * it anyway. --macro
2256 */ 2310 */
2257#define PIC_IRQS (1 << PIC_CASCADE_IR) 2311#define PIC_IRQS (1 << PIC_CASCADE_IR)
2258 2312
@@ -2261,15 +2315,12 @@ void __init setup_IO_APIC(void)
2261 int i; 2315 int i;
2262 2316
2263 /* Reserve all the system vectors. */ 2317 /* Reserve all the system vectors. */
2264 for (i = FIRST_SYSTEM_VECTOR; i < NR_VECTORS; i++) 2318 for (i = first_system_vector; i < NR_VECTORS; i++)
2265 set_bit(i, used_vectors); 2319 set_bit(i, used_vectors);
2266 2320
2267 enable_IO_APIC(); 2321 enable_IO_APIC();
2268 2322
2269 if (acpi_ioapic) 2323 io_apic_irqs = ~PIC_IRQS;
2270 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
2271 else
2272 io_apic_irqs = ~PIC_IRQS;
2273 2324
2274 printk("ENABLING IO-APIC IRQs\n"); 2325 printk("ENABLING IO-APIC IRQs\n");
2275 2326
@@ -2286,28 +2337,14 @@ void __init setup_IO_APIC(void)
2286 print_IO_APIC(); 2337 print_IO_APIC();
2287} 2338}
2288 2339
2289static int __init setup_disable_8254_timer(char *s)
2290{
2291 timer_over_8254 = -1;
2292 return 1;
2293}
2294static int __init setup_enable_8254_timer(char *s)
2295{
2296 timer_over_8254 = 2;
2297 return 1;
2298}
2299
2300__setup("disable_8254_timer", setup_disable_8254_timer);
2301__setup("enable_8254_timer", setup_enable_8254_timer);
2302
2303/* 2340/*
2304 * Called after all the initialization is done. If we didnt find any 2341 * Called after all the initialization is done. If we didnt find any
2305 * APIC bugs then we can allow the modify fast path 2342 * APIC bugs then we can allow the modify fast path
2306 */ 2343 */
2307 2344
2308static int __init io_apic_bug_finalize(void) 2345static int __init io_apic_bug_finalize(void)
2309{ 2346{
2310 if(sis_apic_bug == -1) 2347 if (sis_apic_bug == -1)
2311 sis_apic_bug = 0; 2348 sis_apic_bug = 0;
2312 return 0; 2349 return 0;
2313} 2350}
@@ -2318,17 +2355,17 @@ struct sysfs_ioapic_data {
2318 struct sys_device dev; 2355 struct sys_device dev;
2319 struct IO_APIC_route_entry entry[0]; 2356 struct IO_APIC_route_entry entry[0];
2320}; 2357};
2321static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS]; 2358static struct sysfs_ioapic_data *mp_ioapic_data[MAX_IO_APICS];
2322 2359
2323static int ioapic_suspend(struct sys_device *dev, pm_message_t state) 2360static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2324{ 2361{
2325 struct IO_APIC_route_entry *entry; 2362 struct IO_APIC_route_entry *entry;
2326 struct sysfs_ioapic_data *data; 2363 struct sysfs_ioapic_data *data;
2327 int i; 2364 int i;
2328 2365
2329 data = container_of(dev, struct sysfs_ioapic_data, dev); 2366 data = container_of(dev, struct sysfs_ioapic_data, dev);
2330 entry = data->entry; 2367 entry = data->entry;
2331 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++) 2368 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
2332 entry[i] = ioapic_read_entry(dev->id, i); 2369 entry[i] = ioapic_read_entry(dev->id, i);
2333 2370
2334 return 0; 2371 return 0;
@@ -2341,18 +2378,18 @@ static int ioapic_resume(struct sys_device *dev)
2341 unsigned long flags; 2378 unsigned long flags;
2342 union IO_APIC_reg_00 reg_00; 2379 union IO_APIC_reg_00 reg_00;
2343 int i; 2380 int i;
2344 2381
2345 data = container_of(dev, struct sysfs_ioapic_data, dev); 2382 data = container_of(dev, struct sysfs_ioapic_data, dev);
2346 entry = data->entry; 2383 entry = data->entry;
2347 2384
2348 spin_lock_irqsave(&ioapic_lock, flags); 2385 spin_lock_irqsave(&ioapic_lock, flags);
2349 reg_00.raw = io_apic_read(dev->id, 0); 2386 reg_00.raw = io_apic_read(dev->id, 0);
2350 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) { 2387 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
2351 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid; 2388 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
2352 io_apic_write(dev->id, 0, reg_00.raw); 2389 io_apic_write(dev->id, 0, reg_00.raw);
2353 } 2390 }
2354 spin_unlock_irqrestore(&ioapic_lock, flags); 2391 spin_unlock_irqrestore(&ioapic_lock, flags);
2355 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++) 2392 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
2356 ioapic_write_entry(dev->id, i, entry[i]); 2393 ioapic_write_entry(dev->id, i, entry[i]);
2357 2394
2358 return 0; 2395 return 0;
@@ -2366,24 +2403,23 @@ static struct sysdev_class ioapic_sysdev_class = {
2366 2403
2367static int __init ioapic_init_sysfs(void) 2404static int __init ioapic_init_sysfs(void)
2368{ 2405{
2369 struct sys_device * dev; 2406 struct sys_device *dev;
2370 int i, size, error = 0; 2407 int i, size, error = 0;
2371 2408
2372 error = sysdev_class_register(&ioapic_sysdev_class); 2409 error = sysdev_class_register(&ioapic_sysdev_class);
2373 if (error) 2410 if (error)
2374 return error; 2411 return error;
2375 2412
2376 for (i = 0; i < nr_ioapics; i++ ) { 2413 for (i = 0; i < nr_ioapics; i++) {
2377 size = sizeof(struct sys_device) + nr_ioapic_registers[i] 2414 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2378 * sizeof(struct IO_APIC_route_entry); 2415 * sizeof(struct IO_APIC_route_entry);
2379 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL); 2416 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
2380 if (!mp_ioapic_data[i]) { 2417 if (!mp_ioapic_data[i]) {
2381 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i); 2418 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2382 continue; 2419 continue;
2383 } 2420 }
2384 memset(mp_ioapic_data[i], 0, size);
2385 dev = &mp_ioapic_data[i]->dev; 2421 dev = &mp_ioapic_data[i]->dev;
2386 dev->id = i; 2422 dev->id = i;
2387 dev->cls = &ioapic_sysdev_class; 2423 dev->cls = &ioapic_sysdev_class;
2388 error = sysdev_register(dev); 2424 error = sysdev_register(dev);
2389 if (error) { 2425 if (error) {
@@ -2458,7 +2494,7 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
2458 msg->address_lo = 2494 msg->address_lo =
2459 MSI_ADDR_BASE_LO | 2495 MSI_ADDR_BASE_LO |
2460 ((INT_DEST_MODE == 0) ? 2496 ((INT_DEST_MODE == 0) ?
2461 MSI_ADDR_DEST_MODE_PHYSICAL: 2497MSI_ADDR_DEST_MODE_PHYSICAL:
2462 MSI_ADDR_DEST_MODE_LOGICAL) | 2498 MSI_ADDR_DEST_MODE_LOGICAL) |
2463 ((INT_DELIVERY_MODE != dest_LowestPrio) ? 2499 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2464 MSI_ADDR_REDIRECTION_CPU: 2500 MSI_ADDR_REDIRECTION_CPU:
@@ -2469,7 +2505,7 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
2469 MSI_DATA_TRIGGER_EDGE | 2505 MSI_DATA_TRIGGER_EDGE |
2470 MSI_DATA_LEVEL_ASSERT | 2506 MSI_DATA_LEVEL_ASSERT |
2471 ((INT_DELIVERY_MODE != dest_LowestPrio) ? 2507 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2472 MSI_DATA_DELIVERY_FIXED: 2508MSI_DATA_DELIVERY_FIXED:
2473 MSI_DATA_DELIVERY_LOWPRI) | 2509 MSI_DATA_DELIVERY_LOWPRI) |
2474 MSI_DATA_VECTOR(vector); 2510 MSI_DATA_VECTOR(vector);
2475 } 2511 }
@@ -2640,12 +2676,12 @@ int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2640#endif /* CONFIG_HT_IRQ */ 2676#endif /* CONFIG_HT_IRQ */
2641 2677
2642/* -------------------------------------------------------------------------- 2678/* --------------------------------------------------------------------------
2643 ACPI-based IOAPIC Configuration 2679 ACPI-based IOAPIC Configuration
2644 -------------------------------------------------------------------------- */ 2680 -------------------------------------------------------------------------- */
2645 2681
2646#ifdef CONFIG_ACPI 2682#ifdef CONFIG_ACPI
2647 2683
2648int __init io_apic_get_unique_id (int ioapic, int apic_id) 2684int __init io_apic_get_unique_id(int ioapic, int apic_id)
2649{ 2685{
2650 union IO_APIC_reg_00 reg_00; 2686 union IO_APIC_reg_00 reg_00;
2651 static physid_mask_t apic_id_map = PHYSID_MASK_NONE; 2687 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
@@ -2654,10 +2690,10 @@ int __init io_apic_get_unique_id (int ioapic, int apic_id)
2654 int i = 0; 2690 int i = 0;
2655 2691
2656 /* 2692 /*
2657 * The P4 platform supports up to 256 APIC IDs on two separate APIC 2693 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2658 * buses (one for LAPICs, one for IOAPICs), where predecessors only 2694 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2659 * supports up to 16 on one shared APIC bus. 2695 * supports up to 16 on one shared APIC bus.
2660 * 2696 *
2661 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full 2697 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2662 * advantage of new APIC bus architecture. 2698 * advantage of new APIC bus architecture.
2663 */ 2699 */
@@ -2676,7 +2712,7 @@ int __init io_apic_get_unique_id (int ioapic, int apic_id)
2676 } 2712 }
2677 2713
2678 /* 2714 /*
2679 * Every APIC in a system must have a unique ID or we get lots of nice 2715 * Every APIC in a system must have a unique ID or we get lots of nice
2680 * 'stuck on smp_invalidate_needed IPI wait' messages. 2716 * 'stuck on smp_invalidate_needed IPI wait' messages.
2681 */ 2717 */
2682 if (check_apicid_used(apic_id_map, apic_id)) { 2718 if (check_apicid_used(apic_id_map, apic_id)) {
@@ -2693,7 +2729,7 @@ int __init io_apic_get_unique_id (int ioapic, int apic_id)
2693 "trying %d\n", ioapic, apic_id, i); 2729 "trying %d\n", ioapic, apic_id, i);
2694 2730
2695 apic_id = i; 2731 apic_id = i;
2696 } 2732 }
2697 2733
2698 tmp = apicid_to_cpu_present(apic_id); 2734 tmp = apicid_to_cpu_present(apic_id);
2699 physids_or(apic_id_map, apic_id_map, tmp); 2735 physids_or(apic_id_map, apic_id_map, tmp);
@@ -2720,7 +2756,7 @@ int __init io_apic_get_unique_id (int ioapic, int apic_id)
2720} 2756}
2721 2757
2722 2758
2723int __init io_apic_get_version (int ioapic) 2759int __init io_apic_get_version(int ioapic)
2724{ 2760{
2725 union IO_APIC_reg_01 reg_01; 2761 union IO_APIC_reg_01 reg_01;
2726 unsigned long flags; 2762 unsigned long flags;
@@ -2733,7 +2769,7 @@ int __init io_apic_get_version (int ioapic)
2733} 2769}
2734 2770
2735 2771
2736int __init io_apic_get_redir_entries (int ioapic) 2772int __init io_apic_get_redir_entries(int ioapic)
2737{ 2773{
2738 union IO_APIC_reg_01 reg_01; 2774 union IO_APIC_reg_01 reg_01;
2739 unsigned long flags; 2775 unsigned long flags;
@@ -2746,7 +2782,7 @@ int __init io_apic_get_redir_entries (int ioapic)
2746} 2782}
2747 2783
2748 2784
2749int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low) 2785int io_apic_set_pci_routing(int ioapic, int pin, int irq, int edge_level, int active_high_low)
2750{ 2786{
2751 struct IO_APIC_route_entry entry; 2787 struct IO_APIC_route_entry entry;
2752 2788
@@ -2762,7 +2798,7 @@ int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int a
2762 * corresponding device driver registers for this IRQ. 2798 * corresponding device driver registers for this IRQ.
2763 */ 2799 */
2764 2800
2765 memset(&entry,0,sizeof(entry)); 2801 memset(&entry, 0, sizeof(entry));
2766 2802
2767 entry.delivery_mode = INT_DELIVERY_MODE; 2803 entry.delivery_mode = INT_DELIVERY_MODE;
2768 entry.dest_mode = INT_DEST_MODE; 2804 entry.dest_mode = INT_DEST_MODE;
@@ -2781,7 +2817,7 @@ int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int a
2781 2817
2782 apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry " 2818 apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2783 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic, 2819 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2784 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq, 2820 mp_ioapics[ioapic].mp_apicid, pin, entry.vector, irq,
2785 edge_level, active_high_low); 2821 edge_level, active_high_low);
2786 2822
2787 ioapic_register_intr(irq, entry.vector, edge_level); 2823 ioapic_register_intr(irq, entry.vector, edge_level);
@@ -2802,8 +2838,8 @@ int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2802 return -1; 2838 return -1;
2803 2839
2804 for (i = 0; i < mp_irq_entries; i++) 2840 for (i = 0; i < mp_irq_entries; i++)
2805 if (mp_irqs[i].mpc_irqtype == mp_INT && 2841 if (mp_irqs[i].mp_irqtype == mp_INT &&
2806 mp_irqs[i].mpc_srcbusirq == bus_irq) 2842 mp_irqs[i].mp_srcbusirq == bus_irq)
2807 break; 2843 break;
2808 if (i >= mp_irq_entries) 2844 if (i >= mp_irq_entries)
2809 return -1; 2845 return -1;
@@ -2836,3 +2872,34 @@ static int __init parse_noapic(char *arg)
2836 return 0; 2872 return 0;
2837} 2873}
2838early_param("noapic", parse_noapic); 2874early_param("noapic", parse_noapic);
2875
2876void __init ioapic_init_mappings(void)
2877{
2878 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2879 int i;
2880
2881 for (i = 0; i < nr_ioapics; i++) {
2882 if (smp_found_config) {
2883 ioapic_phys = mp_ioapics[i].mp_apicaddr;
2884 if (!ioapic_phys) {
2885 printk(KERN_ERR
2886 "WARNING: bogus zero IO-APIC "
2887 "address found in MPTABLE, "
2888 "disabling IO/APIC support!\n");
2889 smp_found_config = 0;
2890 skip_ioapic_setup = 1;
2891 goto fake_ioapic_page;
2892 }
2893 } else {
2894fake_ioapic_page:
2895 ioapic_phys = (unsigned long)
2896 alloc_bootmem_pages(PAGE_SIZE);
2897 ioapic_phys = __pa(ioapic_phys);
2898 }
2899 set_fixmap_nocache(idx, ioapic_phys);
2900 printk(KERN_DEBUG "mapped IOAPIC to %08lx (%08lx)\n",
2901 __fix_to_virt(idx), ioapic_phys);
2902 idx++;
2903 }
2904}
2905
diff --git a/arch/x86/kernel/io_apic_64.c b/arch/x86/kernel/io_apic_64.c
index ef1a8dfcc529..8269434d1707 100644
--- a/arch/x86/kernel/io_apic_64.c
+++ b/arch/x86/kernel/io_apic_64.c
@@ -45,6 +45,7 @@
45#include <asm/proto.h> 45#include <asm/proto.h>
46#include <asm/acpi.h> 46#include <asm/acpi.h>
47#include <asm/dma.h> 47#include <asm/dma.h>
48#include <asm/i8259.h>
48#include <asm/nmi.h> 49#include <asm/nmi.h>
49#include <asm/msidef.h> 50#include <asm/msidef.h>
50#include <asm/hypertransport.h> 51#include <asm/hypertransport.h>
@@ -61,7 +62,7 @@ struct irq_cfg {
61}; 62};
62 63
63/* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */ 64/* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
64struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = { 65static struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
65 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, }, 66 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
66 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, }, 67 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
67 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, }, 68 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
@@ -82,6 +83,10 @@ struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
82 83
83static int assign_irq_vector(int irq, cpumask_t mask); 84static int assign_irq_vector(int irq, cpumask_t mask);
84 85
86int first_system_vector = 0xfe;
87
88char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};
89
85#define __apicdebuginit __init 90#define __apicdebuginit __init
86 91
87int sis_apic_bug; /* not actually supported, dummy for compile */ 92int sis_apic_bug; /* not actually supported, dummy for compile */
@@ -90,7 +95,7 @@ static int no_timer_check;
90 95
91static int disable_timer_pin_1 __initdata; 96static int disable_timer_pin_1 __initdata;
92 97
93int timer_over_8254 __initdata = 1; 98int timer_through_8259 __initdata;
94 99
95/* Where if anywhere is the i8259 connect in external int mode */ 100/* Where if anywhere is the i8259 connect in external int mode */
96static struct { int pin, apic; } ioapic_i8259 = { -1, -1 }; 101static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
@@ -104,15 +109,17 @@ DEFINE_SPINLOCK(vector_lock);
104int nr_ioapic_registers[MAX_IO_APICS]; 109int nr_ioapic_registers[MAX_IO_APICS];
105 110
106/* I/O APIC entries */ 111/* I/O APIC entries */
107struct mpc_config_ioapic mp_ioapics[MAX_IO_APICS]; 112struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
108int nr_ioapics; 113int nr_ioapics;
109 114
110/* MP IRQ source entries */ 115/* MP IRQ source entries */
111struct mpc_config_intsrc mp_irqs[MAX_IRQ_SOURCES]; 116struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
112 117
113/* # of MP IRQ source entries */ 118/* # of MP IRQ source entries */
114int mp_irq_entries; 119int mp_irq_entries;
115 120
121DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
122
116/* 123/*
117 * Rough estimation of how many shared IRQs there are, can 124 * Rough estimation of how many shared IRQs there are, can
118 * be changed anytime. 125 * be changed anytime.
@@ -140,7 +147,7 @@ struct io_apic {
140static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx) 147static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
141{ 148{
142 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx) 149 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
143 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK); 150 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
144} 151}
145 152
146static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg) 153static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
@@ -183,7 +190,7 @@ static bool io_apic_level_ack_pending(unsigned int irq)
183 break; 190 break;
184 reg = io_apic_read(entry->apic, 0x10 + pin*2); 191 reg = io_apic_read(entry->apic, 0x10 + pin*2);
185 /* Is the remote IRR bit set? */ 192 /* Is the remote IRR bit set? */
186 if ((reg >> 14) & 1) { 193 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
187 spin_unlock_irqrestore(&ioapic_lock, flags); 194 spin_unlock_irqrestore(&ioapic_lock, flags);
188 return true; 195 return true;
189 } 196 }
@@ -298,7 +305,7 @@ static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
298 break; 305 break;
299 io_apic_write(apic, 0x11 + pin*2, dest); 306 io_apic_write(apic, 0x11 + pin*2, dest);
300 reg = io_apic_read(apic, 0x10 + pin*2); 307 reg = io_apic_read(apic, 0x10 + pin*2);
301 reg &= ~0x000000ff; 308 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
302 reg |= vector; 309 reg |= vector;
303 io_apic_modify(apic, reg); 310 io_apic_modify(apic, reg);
304 if (!entry->next) 311 if (!entry->next)
@@ -360,16 +367,37 @@ static void add_pin_to_irq(unsigned int irq, int apic, int pin)
360 entry->pin = pin; 367 entry->pin = pin;
361} 368}
362 369
370/*
371 * Reroute an IRQ to a different pin.
372 */
373static void __init replace_pin_at_irq(unsigned int irq,
374 int oldapic, int oldpin,
375 int newapic, int newpin)
376{
377 struct irq_pin_list *entry = irq_2_pin + irq;
378
379 while (1) {
380 if (entry->apic == oldapic && entry->pin == oldpin) {
381 entry->apic = newapic;
382 entry->pin = newpin;
383 }
384 if (!entry->next)
385 break;
386 entry = irq_2_pin + entry->next;
387 }
388}
389
363 390
364#define DO_ACTION(name,R,ACTION, FINAL) \ 391#define DO_ACTION(name,R,ACTION, FINAL) \
365 \ 392 \
366 static void name##_IO_APIC_irq (unsigned int irq) \ 393 static void name##_IO_APIC_irq (unsigned int irq) \
367 __DO_ACTION(R, ACTION, FINAL) 394 __DO_ACTION(R, ACTION, FINAL)
368 395
369DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) ) 396/* mask = 1 */
370 /* mask = 1 */ 397DO_ACTION(__mask, 0, |= IO_APIC_REDIR_MASKED, io_apic_sync(entry->apic))
371DO_ACTION( __unmask, 0, &= 0xfffeffff, ) 398
372 /* mask = 0 */ 399/* mask = 0 */
400DO_ACTION(__unmask, 0, &= ~IO_APIC_REDIR_MASKED, )
373 401
374static void mask_IO_APIC_irq (unsigned int irq) 402static void mask_IO_APIC_irq (unsigned int irq)
375{ 403{
@@ -430,20 +458,6 @@ static int __init disable_timer_pin_setup(char *arg)
430} 458}
431__setup("disable_timer_pin_1", disable_timer_pin_setup); 459__setup("disable_timer_pin_1", disable_timer_pin_setup);
432 460
433static int __init setup_disable_8254_timer(char *s)
434{
435 timer_over_8254 = -1;
436 return 1;
437}
438static int __init setup_enable_8254_timer(char *s)
439{
440 timer_over_8254 = 2;
441 return 1;
442}
443
444__setup("disable_8254_timer", setup_disable_8254_timer);
445__setup("enable_8254_timer", setup_enable_8254_timer);
446
447 461
448/* 462/*
449 * Find the IRQ entry number of a certain pin. 463 * Find the IRQ entry number of a certain pin.
@@ -453,10 +467,10 @@ static int find_irq_entry(int apic, int pin, int type)
453 int i; 467 int i;
454 468
455 for (i = 0; i < mp_irq_entries; i++) 469 for (i = 0; i < mp_irq_entries; i++)
456 if (mp_irqs[i].mpc_irqtype == type && 470 if (mp_irqs[i].mp_irqtype == type &&
457 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid || 471 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
458 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) && 472 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
459 mp_irqs[i].mpc_dstirq == pin) 473 mp_irqs[i].mp_dstirq == pin)
460 return i; 474 return i;
461 475
462 return -1; 476 return -1;
@@ -470,13 +484,13 @@ static int __init find_isa_irq_pin(int irq, int type)
470 int i; 484 int i;
471 485
472 for (i = 0; i < mp_irq_entries; i++) { 486 for (i = 0; i < mp_irq_entries; i++) {
473 int lbus = mp_irqs[i].mpc_srcbus; 487 int lbus = mp_irqs[i].mp_srcbus;
474 488
475 if (test_bit(lbus, mp_bus_not_pci) && 489 if (test_bit(lbus, mp_bus_not_pci) &&
476 (mp_irqs[i].mpc_irqtype == type) && 490 (mp_irqs[i].mp_irqtype == type) &&
477 (mp_irqs[i].mpc_srcbusirq == irq)) 491 (mp_irqs[i].mp_srcbusirq == irq))
478 492
479 return mp_irqs[i].mpc_dstirq; 493 return mp_irqs[i].mp_dstirq;
480 } 494 }
481 return -1; 495 return -1;
482} 496}
@@ -486,17 +500,17 @@ static int __init find_isa_irq_apic(int irq, int type)
486 int i; 500 int i;
487 501
488 for (i = 0; i < mp_irq_entries; i++) { 502 for (i = 0; i < mp_irq_entries; i++) {
489 int lbus = mp_irqs[i].mpc_srcbus; 503 int lbus = mp_irqs[i].mp_srcbus;
490 504
491 if (test_bit(lbus, mp_bus_not_pci) && 505 if (test_bit(lbus, mp_bus_not_pci) &&
492 (mp_irqs[i].mpc_irqtype == type) && 506 (mp_irqs[i].mp_irqtype == type) &&
493 (mp_irqs[i].mpc_srcbusirq == irq)) 507 (mp_irqs[i].mp_srcbusirq == irq))
494 break; 508 break;
495 } 509 }
496 if (i < mp_irq_entries) { 510 if (i < mp_irq_entries) {
497 int apic; 511 int apic;
498 for(apic = 0; apic < nr_ioapics; apic++) { 512 for(apic = 0; apic < nr_ioapics; apic++) {
499 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic) 513 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
500 return apic; 514 return apic;
501 } 515 }
502 } 516 }
@@ -516,28 +530,28 @@ int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
516 530
517 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n", 531 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
518 bus, slot, pin); 532 bus, slot, pin);
519 if (mp_bus_id_to_pci_bus[bus] == -1) { 533 if (test_bit(bus, mp_bus_not_pci)) {
520 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus); 534 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
521 return -1; 535 return -1;
522 } 536 }
523 for (i = 0; i < mp_irq_entries; i++) { 537 for (i = 0; i < mp_irq_entries; i++) {
524 int lbus = mp_irqs[i].mpc_srcbus; 538 int lbus = mp_irqs[i].mp_srcbus;
525 539
526 for (apic = 0; apic < nr_ioapics; apic++) 540 for (apic = 0; apic < nr_ioapics; apic++)
527 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic || 541 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
528 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) 542 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
529 break; 543 break;
530 544
531 if (!test_bit(lbus, mp_bus_not_pci) && 545 if (!test_bit(lbus, mp_bus_not_pci) &&
532 !mp_irqs[i].mpc_irqtype && 546 !mp_irqs[i].mp_irqtype &&
533 (bus == lbus) && 547 (bus == lbus) &&
534 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) { 548 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
535 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq); 549 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
536 550
537 if (!(apic || IO_APIC_IRQ(irq))) 551 if (!(apic || IO_APIC_IRQ(irq)))
538 continue; 552 continue;
539 553
540 if (pin == (mp_irqs[i].mpc_srcbusirq & 3)) 554 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
541 return irq; 555 return irq;
542 /* 556 /*
543 * Use the first all-but-pin matching entry as a 557 * Use the first all-but-pin matching entry as a
@@ -565,13 +579,13 @@ int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
565 579
566static int MPBIOS_polarity(int idx) 580static int MPBIOS_polarity(int idx)
567{ 581{
568 int bus = mp_irqs[idx].mpc_srcbus; 582 int bus = mp_irqs[idx].mp_srcbus;
569 int polarity; 583 int polarity;
570 584
571 /* 585 /*
572 * Determine IRQ line polarity (high active or low active): 586 * Determine IRQ line polarity (high active or low active):
573 */ 587 */
574 switch (mp_irqs[idx].mpc_irqflag & 3) 588 switch (mp_irqs[idx].mp_irqflag & 3)
575 { 589 {
576 case 0: /* conforms, ie. bus-type dependent polarity */ 590 case 0: /* conforms, ie. bus-type dependent polarity */
577 if (test_bit(bus, mp_bus_not_pci)) 591 if (test_bit(bus, mp_bus_not_pci))
@@ -607,13 +621,13 @@ static int MPBIOS_polarity(int idx)
607 621
608static int MPBIOS_trigger(int idx) 622static int MPBIOS_trigger(int idx)
609{ 623{
610 int bus = mp_irqs[idx].mpc_srcbus; 624 int bus = mp_irqs[idx].mp_srcbus;
611 int trigger; 625 int trigger;
612 626
613 /* 627 /*
614 * Determine IRQ trigger mode (edge or level sensitive): 628 * Determine IRQ trigger mode (edge or level sensitive):
615 */ 629 */
616 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3) 630 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
617 { 631 {
618 case 0: /* conforms, ie. bus-type dependent */ 632 case 0: /* conforms, ie. bus-type dependent */
619 if (test_bit(bus, mp_bus_not_pci)) 633 if (test_bit(bus, mp_bus_not_pci))
@@ -660,16 +674,16 @@ static inline int irq_trigger(int idx)
660static int pin_2_irq(int idx, int apic, int pin) 674static int pin_2_irq(int idx, int apic, int pin)
661{ 675{
662 int irq, i; 676 int irq, i;
663 int bus = mp_irqs[idx].mpc_srcbus; 677 int bus = mp_irqs[idx].mp_srcbus;
664 678
665 /* 679 /*
666 * Debugging check, we are in big trouble if this message pops up! 680 * Debugging check, we are in big trouble if this message pops up!
667 */ 681 */
668 if (mp_irqs[idx].mpc_dstirq != pin) 682 if (mp_irqs[idx].mp_dstirq != pin)
669 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n"); 683 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
670 684
671 if (test_bit(bus, mp_bus_not_pci)) { 685 if (test_bit(bus, mp_bus_not_pci)) {
672 irq = mp_irqs[idx].mpc_srcbusirq; 686 irq = mp_irqs[idx].mp_srcbusirq;
673 } else { 687 } else {
674 /* 688 /*
675 * PCI IRQs are mapped in order 689 * PCI IRQs are mapped in order
@@ -718,7 +732,7 @@ static int __assign_irq_vector(int irq, cpumask_t mask)
718 return 0; 732 return 0;
719 } 733 }
720 734
721 for_each_cpu_mask(cpu, mask) { 735 for_each_cpu_mask_nr(cpu, mask) {
722 cpumask_t domain, new_mask; 736 cpumask_t domain, new_mask;
723 int new_cpu; 737 int new_cpu;
724 int vector, offset; 738 int vector, offset;
@@ -730,7 +744,7 @@ static int __assign_irq_vector(int irq, cpumask_t mask)
730 offset = current_offset; 744 offset = current_offset;
731next: 745next:
732 vector += 8; 746 vector += 8;
733 if (vector >= FIRST_SYSTEM_VECTOR) { 747 if (vector >= first_system_vector) {
734 /* If we run out of vectors on large boxen, must share them. */ 748 /* If we run out of vectors on large boxen, must share them. */
735 offset = (offset + 1) % 8; 749 offset = (offset + 1) % 8;
736 vector = FIRST_DEVICE_VECTOR + offset; 750 vector = FIRST_DEVICE_VECTOR + offset;
@@ -739,7 +753,7 @@ next:
739 continue; 753 continue;
740 if (vector == IA32_SYSCALL_VECTOR) 754 if (vector == IA32_SYSCALL_VECTOR)
741 goto next; 755 goto next;
742 for_each_cpu_mask(new_cpu, new_mask) 756 for_each_cpu_mask_nr(new_cpu, new_mask)
743 if (per_cpu(vector_irq, new_cpu)[vector] != -1) 757 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
744 goto next; 758 goto next;
745 /* Found one! */ 759 /* Found one! */
@@ -749,7 +763,7 @@ next:
749 cfg->move_in_progress = 1; 763 cfg->move_in_progress = 1;
750 cfg->old_domain = cfg->domain; 764 cfg->old_domain = cfg->domain;
751 } 765 }
752 for_each_cpu_mask(new_cpu, new_mask) 766 for_each_cpu_mask_nr(new_cpu, new_mask)
753 per_cpu(vector_irq, new_cpu)[vector] = irq; 767 per_cpu(vector_irq, new_cpu)[vector] = irq;
754 cfg->vector = vector; 768 cfg->vector = vector;
755 cfg->domain = domain; 769 cfg->domain = domain;
@@ -781,14 +795,14 @@ static void __clear_irq_vector(int irq)
781 795
782 vector = cfg->vector; 796 vector = cfg->vector;
783 cpus_and(mask, cfg->domain, cpu_online_map); 797 cpus_and(mask, cfg->domain, cpu_online_map);
784 for_each_cpu_mask(cpu, mask) 798 for_each_cpu_mask_nr(cpu, mask)
785 per_cpu(vector_irq, cpu)[vector] = -1; 799 per_cpu(vector_irq, cpu)[vector] = -1;
786 800
787 cfg->vector = 0; 801 cfg->vector = 0;
788 cpus_clear(cfg->domain); 802 cpus_clear(cfg->domain);
789} 803}
790 804
791void __setup_vector_irq(int cpu) 805static void __setup_vector_irq(int cpu)
792{ 806{
793 /* Initialize vector_irq on a new cpu */ 807 /* Initialize vector_irq on a new cpu */
794 /* This function must be called with vector_lock held */ 808 /* This function must be called with vector_lock held */
@@ -811,6 +825,13 @@ void __setup_vector_irq(int cpu)
811 } 825 }
812} 826}
813 827
828void setup_vector_irq(int cpu)
829{
830 spin_lock(&vector_lock);
831 __setup_vector_irq(smp_processor_id());
832 spin_unlock(&vector_lock);
833}
834
814 835
815static struct irq_chip ioapic_chip; 836static struct irq_chip ioapic_chip;
816 837
@@ -846,7 +867,7 @@ static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
846 apic_printk(APIC_VERBOSE,KERN_DEBUG 867 apic_printk(APIC_VERBOSE,KERN_DEBUG
847 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> " 868 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
848 "IRQ %d Mode:%i Active:%i)\n", 869 "IRQ %d Mode:%i Active:%i)\n",
849 apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector, 870 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
850 irq, trigger, polarity); 871 irq, trigger, polarity);
851 872
852 /* 873 /*
@@ -887,10 +908,10 @@ static void __init setup_IO_APIC_irqs(void)
887 idx = find_irq_entry(apic,pin,mp_INT); 908 idx = find_irq_entry(apic,pin,mp_INT);
888 if (idx == -1) { 909 if (idx == -1) {
889 if (first_notcon) { 910 if (first_notcon) {
890 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin); 911 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mp_apicid, pin);
891 first_notcon = 0; 912 first_notcon = 0;
892 } else 913 } else
893 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin); 914 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mp_apicid, pin);
894 continue; 915 continue;
895 } 916 }
896 if (!first_notcon) { 917 if (!first_notcon) {
@@ -911,26 +932,21 @@ static void __init setup_IO_APIC_irqs(void)
911} 932}
912 933
913/* 934/*
914 * Set up the 8259A-master output pin as broadcast to all 935 * Set up the timer pin, possibly with the 8259A-master behind.
915 * CPUs.
916 */ 936 */
917static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector) 937static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
938 int vector)
918{ 939{
919 struct IO_APIC_route_entry entry; 940 struct IO_APIC_route_entry entry;
920 941
921 memset(&entry, 0, sizeof(entry)); 942 memset(&entry, 0, sizeof(entry));
922 943
923 disable_8259A_irq(0);
924
925 /* mask LVT0 */
926 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
927
928 /* 944 /*
929 * We use logical delivery to get the timer IRQ 945 * We use logical delivery to get the timer IRQ
930 * to the first CPU. 946 * to the first CPU.
931 */ 947 */
932 entry.dest_mode = INT_DEST_MODE; 948 entry.dest_mode = INT_DEST_MODE;
933 entry.mask = 0; /* unmask IRQ now */ 949 entry.mask = 1; /* mask IRQ now */
934 entry.dest = cpu_mask_to_apicid(TARGET_CPUS); 950 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
935 entry.delivery_mode = INT_DELIVERY_MODE; 951 entry.delivery_mode = INT_DELIVERY_MODE;
936 entry.polarity = 0; 952 entry.polarity = 0;
@@ -939,7 +955,7 @@ static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, in
939 955
940 /* 956 /*
941 * The timer IRQ doesn't have to know that behind the 957 * The timer IRQ doesn't have to know that behind the
942 * scene we have a 8259A-master in AEOI mode ... 958 * scene we may have a 8259A-master in AEOI mode ...
943 */ 959 */
944 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge"); 960 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
945 961
@@ -947,8 +963,6 @@ static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, in
947 * Add it to the IO-APIC irq-routing table: 963 * Add it to the IO-APIC irq-routing table:
948 */ 964 */
949 ioapic_write_entry(apic, pin, entry); 965 ioapic_write_entry(apic, pin, entry);
950
951 enable_8259A_irq(0);
952} 966}
953 967
954void __apicdebuginit print_IO_APIC(void) 968void __apicdebuginit print_IO_APIC(void)
@@ -965,7 +979,7 @@ void __apicdebuginit print_IO_APIC(void)
965 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries); 979 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
966 for (i = 0; i < nr_ioapics; i++) 980 for (i = 0; i < nr_ioapics; i++)
967 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n", 981 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
968 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]); 982 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
969 983
970 /* 984 /*
971 * We are a bit conservative about what we expect. We have to 985 * We are a bit conservative about what we expect. We have to
@@ -983,7 +997,7 @@ void __apicdebuginit print_IO_APIC(void)
983 spin_unlock_irqrestore(&ioapic_lock, flags); 997 spin_unlock_irqrestore(&ioapic_lock, flags);
984 998
985 printk("\n"); 999 printk("\n");
986 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid); 1000 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
987 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw); 1001 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
988 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID); 1002 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
989 1003
@@ -1077,6 +1091,7 @@ void __apicdebuginit print_local_APIC(void * dummy)
1077 1091
1078 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n", 1092 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1079 smp_processor_id(), hard_smp_processor_id()); 1093 smp_processor_id(), hard_smp_processor_id());
1094 v = apic_read(APIC_ID);
1080 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id())); 1095 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id()));
1081 v = apic_read(APIC_LVR); 1096 v = apic_read(APIC_LVR);
1082 printk(KERN_INFO "... APIC VERSION: %08x\n", v); 1097 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
@@ -1146,7 +1161,7 @@ void __apicdebuginit print_local_APIC(void * dummy)
1146 1161
1147void print_all_local_APICs (void) 1162void print_all_local_APICs (void)
1148{ 1163{
1149 on_each_cpu(print_local_APIC, NULL, 1, 1); 1164 on_each_cpu(print_local_APIC, NULL, 1);
1150} 1165}
1151 1166
1152void __apicdebuginit print_PIC(void) 1167void __apicdebuginit print_PIC(void)
@@ -1358,12 +1373,10 @@ static unsigned int startup_ioapic_irq(unsigned int irq)
1358static int ioapic_retrigger_irq(unsigned int irq) 1373static int ioapic_retrigger_irq(unsigned int irq)
1359{ 1374{
1360 struct irq_cfg *cfg = &irq_cfg[irq]; 1375 struct irq_cfg *cfg = &irq_cfg[irq];
1361 cpumask_t mask;
1362 unsigned long flags; 1376 unsigned long flags;
1363 1377
1364 spin_lock_irqsave(&vector_lock, flags); 1378 spin_lock_irqsave(&vector_lock, flags);
1365 mask = cpumask_of_cpu(first_cpu(cfg->domain)); 1379 send_IPI_mask(cpumask_of_cpu(first_cpu(cfg->domain)), cfg->vector);
1366 send_IPI_mask(mask, cfg->vector);
1367 spin_unlock_irqrestore(&vector_lock, flags); 1380 spin_unlock_irqrestore(&vector_lock, flags);
1368 1381
1369 return 1; 1382 return 1;
@@ -1540,7 +1553,7 @@ static inline void init_IO_APIC_traps(void)
1540 } 1553 }
1541} 1554}
1542 1555
1543static void enable_lapic_irq (unsigned int irq) 1556static void unmask_lapic_irq(unsigned int irq)
1544{ 1557{
1545 unsigned long v; 1558 unsigned long v;
1546 1559
@@ -1548,7 +1561,7 @@ static void enable_lapic_irq (unsigned int irq)
1548 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED); 1561 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1549} 1562}
1550 1563
1551static void disable_lapic_irq (unsigned int irq) 1564static void mask_lapic_irq(unsigned int irq)
1552{ 1565{
1553 unsigned long v; 1566 unsigned long v;
1554 1567
@@ -1561,19 +1574,20 @@ static void ack_lapic_irq (unsigned int irq)
1561 ack_APIC_irq(); 1574 ack_APIC_irq();
1562} 1575}
1563 1576
1564static void end_lapic_irq (unsigned int i) { /* nothing */ } 1577static struct irq_chip lapic_chip __read_mostly = {
1565 1578 .name = "local-APIC",
1566static struct hw_interrupt_type lapic_irq_type __read_mostly = { 1579 .mask = mask_lapic_irq,
1567 .name = "local-APIC", 1580 .unmask = unmask_lapic_irq,
1568 .typename = "local-APIC-edge", 1581 .ack = ack_lapic_irq,
1569 .startup = NULL, /* startup_irq() not used for IRQ0 */
1570 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1571 .enable = enable_lapic_irq,
1572 .disable = disable_lapic_irq,
1573 .ack = ack_lapic_irq,
1574 .end = end_lapic_irq,
1575}; 1582};
1576 1583
1584static void lapic_register_intr(int irq)
1585{
1586 irq_desc[irq].status &= ~IRQ_LEVEL;
1587 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
1588 "edge");
1589}
1590
1577static void __init setup_nmi(void) 1591static void __init setup_nmi(void)
1578{ 1592{
1579 /* 1593 /*
@@ -1659,6 +1673,7 @@ static inline void __init check_timer(void)
1659 struct irq_cfg *cfg = irq_cfg + 0; 1673 struct irq_cfg *cfg = irq_cfg + 0;
1660 int apic1, pin1, apic2, pin2; 1674 int apic1, pin1, apic2, pin2;
1661 unsigned long flags; 1675 unsigned long flags;
1676 int no_pin1 = 0;
1662 1677
1663 local_irq_save(flags); 1678 local_irq_save(flags);
1664 1679
@@ -1669,34 +1684,48 @@ static inline void __init check_timer(void)
1669 assign_irq_vector(0, TARGET_CPUS); 1684 assign_irq_vector(0, TARGET_CPUS);
1670 1685
1671 /* 1686 /*
1672 * Subtle, code in do_timer_interrupt() expects an AEOI 1687 * As IRQ0 is to be enabled in the 8259A, the virtual
1673 * mode for the 8259A whenever interrupts are routed 1688 * wire has to be disabled in the local APIC.
1674 * through I/O APICs. Also IRQ0 has to be enabled in
1675 * the 8259A which implies the virtual wire has to be
1676 * disabled in the local APIC.
1677 */ 1689 */
1678 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT); 1690 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1679 init_8259A(1); 1691 init_8259A(1);
1680 if (timer_over_8254 > 0)
1681 enable_8259A_irq(0);
1682 1692
1683 pin1 = find_isa_irq_pin(0, mp_INT); 1693 pin1 = find_isa_irq_pin(0, mp_INT);
1684 apic1 = find_isa_irq_apic(0, mp_INT); 1694 apic1 = find_isa_irq_apic(0, mp_INT);
1685 pin2 = ioapic_i8259.pin; 1695 pin2 = ioapic_i8259.pin;
1686 apic2 = ioapic_i8259.apic; 1696 apic2 = ioapic_i8259.apic;
1687 1697
1688 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n", 1698 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
1689 cfg->vector, apic1, pin1, apic2, pin2); 1699 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
1700 cfg->vector, apic1, pin1, apic2, pin2);
1701
1702 /*
1703 * Some BIOS writers are clueless and report the ExtINTA
1704 * I/O APIC input from the cascaded 8259A as the timer
1705 * interrupt input. So just in case, if only one pin
1706 * was found above, try it both directly and through the
1707 * 8259A.
1708 */
1709 if (pin1 == -1) {
1710 pin1 = pin2;
1711 apic1 = apic2;
1712 no_pin1 = 1;
1713 } else if (pin2 == -1) {
1714 pin2 = pin1;
1715 apic2 = apic1;
1716 }
1690 1717
1691 if (pin1 != -1) { 1718 if (pin1 != -1) {
1692 /* 1719 /*
1693 * Ok, does IRQ0 through the IOAPIC work? 1720 * Ok, does IRQ0 through the IOAPIC work?
1694 */ 1721 */
1722 if (no_pin1) {
1723 add_pin_to_irq(0, apic1, pin1);
1724 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
1725 }
1695 unmask_IO_APIC_irq(0); 1726 unmask_IO_APIC_irq(0);
1696 if (!no_timer_check && timer_irq_works()) { 1727 if (!no_timer_check && timer_irq_works()) {
1697 nmi_watchdog_default();
1698 if (nmi_watchdog == NMI_IO_APIC) { 1728 if (nmi_watchdog == NMI_IO_APIC) {
1699 disable_8259A_irq(0);
1700 setup_nmi(); 1729 setup_nmi();
1701 enable_8259A_irq(0); 1730 enable_8259A_irq(0);
1702 } 1731 }
@@ -1705,54 +1734,62 @@ static inline void __init check_timer(void)
1705 goto out; 1734 goto out;
1706 } 1735 }
1707 clear_IO_APIC_pin(apic1, pin1); 1736 clear_IO_APIC_pin(apic1, pin1);
1708 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not " 1737 if (!no_pin1)
1709 "connected to IO-APIC\n"); 1738 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
1710 } 1739 "8254 timer not connected to IO-APIC\n");
1711 1740
1712 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) " 1741 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
1713 "through the 8259A ... "); 1742 "(IRQ0) through the 8259A ...\n");
1714 if (pin2 != -1) { 1743 apic_printk(APIC_QUIET, KERN_INFO
1715 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...", 1744 "..... (found apic %d pin %d) ...\n", apic2, pin2);
1716 apic2, pin2);
1717 /* 1745 /*
1718 * legacy devices should be connected to IO APIC #0 1746 * legacy devices should be connected to IO APIC #0
1719 */ 1747 */
1720 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector); 1748 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
1749 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
1750 unmask_IO_APIC_irq(0);
1751 enable_8259A_irq(0);
1721 if (timer_irq_works()) { 1752 if (timer_irq_works()) {
1722 apic_printk(APIC_VERBOSE," works.\n"); 1753 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
1723 nmi_watchdog_default(); 1754 timer_through_8259 = 1;
1724 if (nmi_watchdog == NMI_IO_APIC) { 1755 if (nmi_watchdog == NMI_IO_APIC) {
1756 disable_8259A_irq(0);
1725 setup_nmi(); 1757 setup_nmi();
1758 enable_8259A_irq(0);
1726 } 1759 }
1727 goto out; 1760 goto out;
1728 } 1761 }
1729 /* 1762 /*
1730 * Cleanup, just in case ... 1763 * Cleanup, just in case ...
1731 */ 1764 */
1765 disable_8259A_irq(0);
1732 clear_IO_APIC_pin(apic2, pin2); 1766 clear_IO_APIC_pin(apic2, pin2);
1767 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
1733 } 1768 }
1734 apic_printk(APIC_VERBOSE," failed.\n");
1735 1769
1736 if (nmi_watchdog == NMI_IO_APIC) { 1770 if (nmi_watchdog == NMI_IO_APIC) {
1737 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n"); 1771 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
1738 nmi_watchdog = 0; 1772 "through the IO-APIC - disabling NMI Watchdog!\n");
1773 nmi_watchdog = NMI_NONE;
1739 } 1774 }
1740 1775
1741 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ..."); 1776 apic_printk(APIC_QUIET, KERN_INFO
1777 "...trying to set up timer as Virtual Wire IRQ...\n");
1742 1778
1743 disable_8259A_irq(0); 1779 lapic_register_intr(0);
1744 irq_desc[0].chip = &lapic_irq_type;
1745 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */ 1780 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1746 enable_8259A_irq(0); 1781 enable_8259A_irq(0);
1747 1782
1748 if (timer_irq_works()) { 1783 if (timer_irq_works()) {
1749 apic_printk(APIC_VERBOSE," works.\n"); 1784 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
1750 goto out; 1785 goto out;
1751 } 1786 }
1787 disable_8259A_irq(0);
1752 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector); 1788 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1753 apic_printk(APIC_VERBOSE," failed.\n"); 1789 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
1754 1790
1755 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ..."); 1791 apic_printk(APIC_QUIET, KERN_INFO
1792 "...trying to set up timer as ExtINT IRQ...\n");
1756 1793
1757 init_8259A(0); 1794 init_8259A(0);
1758 make_8259A_irq(0); 1795 make_8259A_irq(0);
@@ -1761,11 +1798,12 @@ static inline void __init check_timer(void)
1761 unlock_ExtINT_logic(); 1798 unlock_ExtINT_logic();
1762 1799
1763 if (timer_irq_works()) { 1800 if (timer_irq_works()) {
1764 apic_printk(APIC_VERBOSE," works.\n"); 1801 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
1765 goto out; 1802 goto out;
1766 } 1803 }
1767 apic_printk(APIC_VERBOSE," failed :(.\n"); 1804 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
1768 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n"); 1805 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
1806 "report. Then try booting with the 'noapic' option.\n");
1769out: 1807out:
1770 local_irq_restore(flags); 1808 local_irq_restore(flags);
1771} 1809}
@@ -1778,11 +1816,21 @@ static int __init notimercheck(char *s)
1778__setup("no_timer_check", notimercheck); 1816__setup("no_timer_check", notimercheck);
1779 1817
1780/* 1818/*
1781 * 1819 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
1782 * IRQs that are handled by the PIC in the MPS IOAPIC case. 1820 * to devices. However there may be an I/O APIC pin available for
1783 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ. 1821 * this interrupt regardless. The pin may be left unconnected, but
1784 * Linux doesn't really care, as it's not actually used 1822 * typically it will be reused as an ExtINT cascade interrupt for
1785 * for any interrupt handling anyway. 1823 * the master 8259A. In the MPS case such a pin will normally be
1824 * reported as an ExtINT interrupt in the MP table. With ACPI
1825 * there is no provision for ExtINT interrupts, and in the absence
1826 * of an override it would be treated as an ordinary ISA I/O APIC
1827 * interrupt, that is edge-triggered and unmasked by default. We
1828 * used to do this, but it caused problems on some systems because
1829 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
1830 * the same ExtINT cascade interrupt to drive the local APIC of the
1831 * bootstrap processor. Therefore we refrain from routing IRQ2 to
1832 * the I/O APIC in all cases now. No actual device should request
1833 * it anyway. --macro
1786 */ 1834 */
1787#define PIC_IRQS (1<<2) 1835#define PIC_IRQS (1<<2)
1788 1836
@@ -1793,10 +1841,7 @@ void __init setup_IO_APIC(void)
1793 * calling enable_IO_APIC() is moved to setup_local_APIC for BP 1841 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
1794 */ 1842 */
1795 1843
1796 if (acpi_ioapic) 1844 io_apic_irqs = ~PIC_IRQS;
1797 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1798 else
1799 io_apic_irqs = ~PIC_IRQS;
1800 1845
1801 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n"); 1846 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1802 1847
@@ -1841,8 +1886,8 @@ static int ioapic_resume(struct sys_device *dev)
1841 1886
1842 spin_lock_irqsave(&ioapic_lock, flags); 1887 spin_lock_irqsave(&ioapic_lock, flags);
1843 reg_00.raw = io_apic_read(dev->id, 0); 1888 reg_00.raw = io_apic_read(dev->id, 0);
1844 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) { 1889 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
1845 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid; 1890 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
1846 io_apic_write(dev->id, 0, reg_00.raw); 1891 io_apic_write(dev->id, 0, reg_00.raw);
1847 } 1892 }
1848 spin_unlock_irqrestore(&ioapic_lock, flags); 1893 spin_unlock_irqrestore(&ioapic_lock, flags);
@@ -2242,8 +2287,8 @@ int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2242 return -1; 2287 return -1;
2243 2288
2244 for (i = 0; i < mp_irq_entries; i++) 2289 for (i = 0; i < mp_irq_entries; i++)
2245 if (mp_irqs[i].mpc_irqtype == mp_INT && 2290 if (mp_irqs[i].mp_irqtype == mp_INT &&
2246 mp_irqs[i].mpc_srcbusirq == bus_irq) 2291 mp_irqs[i].mp_srcbusirq == bus_irq)
2247 break; 2292 break;
2248 if (i >= mp_irq_entries) 2293 if (i >= mp_irq_entries)
2249 return -1; 2294 return -1;
@@ -2336,7 +2381,7 @@ void __init ioapic_init_mappings(void)
2336 ioapic_res = ioapic_setup_resources(); 2381 ioapic_res = ioapic_setup_resources();
2337 for (i = 0; i < nr_ioapics; i++) { 2382 for (i = 0; i < nr_ioapics; i++) {
2338 if (smp_found_config) { 2383 if (smp_found_config) {
2339 ioapic_phys = mp_ioapics[i].mpc_apicaddr; 2384 ioapic_phys = mp_ioapics[i].mp_apicaddr;
2340 } else { 2385 } else {
2341 ioapic_phys = (unsigned long) 2386 ioapic_phys = (unsigned long)
2342 alloc_bootmem_pages(PAGE_SIZE); 2387 alloc_bootmem_pages(PAGE_SIZE);
diff --git a/arch/x86/kernel/io_delay.c b/arch/x86/kernel/io_delay.c
index 5921e5f0a640..1c3a66a67f83 100644
--- a/arch/x86/kernel/io_delay.c
+++ b/arch/x86/kernel/io_delay.c
@@ -103,6 +103,9 @@ void __init io_delay_init(void)
103 103
104static int __init io_delay_param(char *s) 104static int __init io_delay_param(char *s)
105{ 105{
106 if (!s)
107 return -EINVAL;
108
106 if (!strcmp(s, "0x80")) 109 if (!strcmp(s, "0x80"))
107 io_delay_type = CONFIG_IO_DELAY_TYPE_0X80; 110 io_delay_type = CONFIG_IO_DELAY_TYPE_0X80;
108 else if (!strcmp(s, "0xed")) 111 else if (!strcmp(s, "0xed"))
diff --git a/arch/x86/kernel/ipi.c b/arch/x86/kernel/ipi.c
index c0df7b89ca23..3f7537b669d3 100644
--- a/arch/x86/kernel/ipi.c
+++ b/arch/x86/kernel/ipi.c
@@ -8,7 +8,6 @@
8#include <linux/kernel_stat.h> 8#include <linux/kernel_stat.h>
9#include <linux/mc146818rtc.h> 9#include <linux/mc146818rtc.h>
10#include <linux/cache.h> 10#include <linux/cache.h>
11#include <linux/interrupt.h>
12#include <linux/cpu.h> 11#include <linux/cpu.h>
13#include <linux/module.h> 12#include <linux/module.h>
14 13
@@ -71,7 +70,7 @@ void __send_IPI_shortcut(unsigned int shortcut, int vector)
71 /* 70 /*
72 * Send the IPI. The write to APIC_ICR fires this off. 71 * Send the IPI. The write to APIC_ICR fires this off.
73 */ 72 */
74 apic_write_around(APIC_ICR, cfg); 73 apic_write(APIC_ICR, cfg);
75} 74}
76 75
77void send_IPI_self(int vector) 76void send_IPI_self(int vector)
@@ -99,7 +98,7 @@ static inline void __send_IPI_dest_field(unsigned long mask, int vector)
99 * prepare target chip field 98 * prepare target chip field
100 */ 99 */
101 cfg = __prepare_ICR2(mask); 100 cfg = __prepare_ICR2(mask);
102 apic_write_around(APIC_ICR2, cfg); 101 apic_write(APIC_ICR2, cfg);
103 102
104 /* 103 /*
105 * program the ICR 104 * program the ICR
@@ -109,7 +108,7 @@ static inline void __send_IPI_dest_field(unsigned long mask, int vector)
109 /* 108 /*
110 * Send the IPI. The write to APIC_ICR fires this off. 109 * Send the IPI. The write to APIC_ICR fires this off.
111 */ 110 */
112 apic_write_around(APIC_ICR, cfg); 111 apic_write(APIC_ICR, cfg);
113} 112}
114 113
115/* 114/*
diff --git a/arch/x86/kernel/irq_32.c b/arch/x86/kernel/irq_32.c
index 147352df28b9..1cf8c1fcc088 100644
--- a/arch/x86/kernel/irq_32.c
+++ b/arch/x86/kernel/irq_32.c
@@ -48,6 +48,29 @@ void ack_bad_irq(unsigned int irq)
48#endif 48#endif
49} 49}
50 50
51#ifdef CONFIG_DEBUG_STACKOVERFLOW
52/* Debugging check for stack overflow: is there less than 1KB free? */
53static int check_stack_overflow(void)
54{
55 long sp;
56
57 __asm__ __volatile__("andl %%esp,%0" :
58 "=r" (sp) : "0" (THREAD_SIZE - 1));
59
60 return sp < (sizeof(struct thread_info) + STACK_WARN);
61}
62
63static void print_stack_overflow(void)
64{
65 printk(KERN_WARNING "low stack detected by irq handler\n");
66 dump_stack();
67}
68
69#else
70static inline int check_stack_overflow(void) { return 0; }
71static inline void print_stack_overflow(void) { }
72#endif
73
51#ifdef CONFIG_4KSTACKS 74#ifdef CONFIG_4KSTACKS
52/* 75/*
53 * per-CPU IRQ handling contexts (thread information and stack) 76 * per-CPU IRQ handling contexts (thread information and stack)
@@ -59,48 +82,26 @@ union irq_ctx {
59 82
60static union irq_ctx *hardirq_ctx[NR_CPUS] __read_mostly; 83static union irq_ctx *hardirq_ctx[NR_CPUS] __read_mostly;
61static union irq_ctx *softirq_ctx[NR_CPUS] __read_mostly; 84static union irq_ctx *softirq_ctx[NR_CPUS] __read_mostly;
62#endif
63
64/*
65 * do_IRQ handles all normal device IRQ's (the special
66 * SMP cross-CPU interrupts have their own specific
67 * handlers).
68 */
69unsigned int do_IRQ(struct pt_regs *regs)
70{
71 struct pt_regs *old_regs;
72 /* high bit used in ret_from_ code */
73 int irq = ~regs->orig_ax;
74 struct irq_desc *desc = irq_desc + irq;
75#ifdef CONFIG_4KSTACKS
76 union irq_ctx *curctx, *irqctx;
77 u32 *isp;
78#endif
79 85
80 if (unlikely((unsigned)irq >= NR_IRQS)) { 86static char softirq_stack[NR_CPUS * THREAD_SIZE] __page_aligned_bss;
81 printk(KERN_EMERG "%s: cannot handle IRQ %d\n", 87static char hardirq_stack[NR_CPUS * THREAD_SIZE] __page_aligned_bss;
82 __func__, irq);
83 BUG();
84 }
85 88
86 old_regs = set_irq_regs(regs); 89static void call_on_stack(void *func, void *stack)
87 irq_enter(); 90{
88#ifdef CONFIG_DEBUG_STACKOVERFLOW 91 asm volatile("xchgl %%ebx,%%esp \n"
89 /* Debugging check for stack overflow: is there less than 1KB free? */ 92 "call *%%edi \n"
90 { 93 "movl %%ebx,%%esp \n"
91 long sp; 94 : "=b" (stack)
92 95 : "0" (stack),
93 __asm__ __volatile__("andl %%esp,%0" : 96 "D"(func)
94 "=r" (sp) : "0" (THREAD_SIZE - 1)); 97 : "memory", "cc", "edx", "ecx", "eax");
95 if (unlikely(sp < (sizeof(struct thread_info) + STACK_WARN))) { 98}
96 printk("do_IRQ: stack overflow: %ld\n",
97 sp - sizeof(struct thread_info));
98 dump_stack();
99 }
100 }
101#endif
102 99
103#ifdef CONFIG_4KSTACKS 100static inline int
101execute_on_irq_stack(int overflow, struct irq_desc *desc, int irq)
102{
103 union irq_ctx *curctx, *irqctx;
104 u32 *isp, arg1, arg2;
104 105
105 curctx = (union irq_ctx *) current_thread_info(); 106 curctx = (union irq_ctx *) current_thread_info();
106 irqctx = hardirq_ctx[smp_processor_id()]; 107 irqctx = hardirq_ctx[smp_processor_id()];
@@ -111,52 +112,39 @@ unsigned int do_IRQ(struct pt_regs *regs)
111 * handler) we can't do that and just have to keep using the 112 * handler) we can't do that and just have to keep using the
112 * current stack (which is the irq stack already after all) 113 * current stack (which is the irq stack already after all)
113 */ 114 */
114 if (curctx != irqctx) { 115 if (unlikely(curctx == irqctx))
115 int arg1, arg2, bx; 116 return 0;
116
117 /* build the stack frame on the IRQ stack */
118 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
119 irqctx->tinfo.task = curctx->tinfo.task;
120 irqctx->tinfo.previous_esp = current_stack_pointer;
121 117
122 /* 118 /* build the stack frame on the IRQ stack */
123 * Copy the softirq bits in preempt_count so that the 119 isp = (u32 *) ((char*)irqctx + sizeof(*irqctx));
124 * softirq checks work in the hardirq context. 120 irqctx->tinfo.task = curctx->tinfo.task;
125 */ 121 irqctx->tinfo.previous_esp = current_stack_pointer;
126 irqctx->tinfo.preempt_count =
127 (irqctx->tinfo.preempt_count & ~SOFTIRQ_MASK) |
128 (curctx->tinfo.preempt_count & SOFTIRQ_MASK);
129
130 asm volatile(
131 " xchgl %%ebx,%%esp \n"
132 " call *%%edi \n"
133 " movl %%ebx,%%esp \n"
134 : "=a" (arg1), "=d" (arg2), "=b" (bx)
135 : "0" (irq), "1" (desc), "2" (isp),
136 "D" (desc->handle_irq)
137 : "memory", "cc", "ecx"
138 );
139 } else
140#endif
141 desc->handle_irq(irq, desc);
142 122
143 irq_exit(); 123 /*
144 set_irq_regs(old_regs); 124 * Copy the softirq bits in preempt_count so that the
125 * softirq checks work in the hardirq context.
126 */
127 irqctx->tinfo.preempt_count =
128 (irqctx->tinfo.preempt_count & ~SOFTIRQ_MASK) |
129 (curctx->tinfo.preempt_count & SOFTIRQ_MASK);
130
131 if (unlikely(overflow))
132 call_on_stack(print_stack_overflow, isp);
133
134 asm volatile("xchgl %%ebx,%%esp \n"
135 "call *%%edi \n"
136 "movl %%ebx,%%esp \n"
137 : "=a" (arg1), "=d" (arg2), "=b" (isp)
138 : "0" (irq), "1" (desc), "2" (isp),
139 "D" (desc->handle_irq)
140 : "memory", "cc", "ecx");
145 return 1; 141 return 1;
146} 142}
147 143
148#ifdef CONFIG_4KSTACKS
149
150static char softirq_stack[NR_CPUS * THREAD_SIZE]
151 __attribute__((__section__(".bss.page_aligned")));
152
153static char hardirq_stack[NR_CPUS * THREAD_SIZE]
154 __attribute__((__section__(".bss.page_aligned")));
155
156/* 144/*
157 * allocate per-cpu stacks for hardirq and for softirq processing 145 * allocate per-cpu stacks for hardirq and for softirq processing
158 */ 146 */
159void irq_ctx_init(int cpu) 147void __cpuinit irq_ctx_init(int cpu)
160{ 148{
161 union irq_ctx *irqctx; 149 union irq_ctx *irqctx;
162 150
@@ -164,25 +152,25 @@ void irq_ctx_init(int cpu)
164 return; 152 return;
165 153
166 irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE]; 154 irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE];
167 irqctx->tinfo.task = NULL; 155 irqctx->tinfo.task = NULL;
168 irqctx->tinfo.exec_domain = NULL; 156 irqctx->tinfo.exec_domain = NULL;
169 irqctx->tinfo.cpu = cpu; 157 irqctx->tinfo.cpu = cpu;
170 irqctx->tinfo.preempt_count = HARDIRQ_OFFSET; 158 irqctx->tinfo.preempt_count = HARDIRQ_OFFSET;
171 irqctx->tinfo.addr_limit = MAKE_MM_SEG(0); 159 irqctx->tinfo.addr_limit = MAKE_MM_SEG(0);
172 160
173 hardirq_ctx[cpu] = irqctx; 161 hardirq_ctx[cpu] = irqctx;
174 162
175 irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE]; 163 irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE];
176 irqctx->tinfo.task = NULL; 164 irqctx->tinfo.task = NULL;
177 irqctx->tinfo.exec_domain = NULL; 165 irqctx->tinfo.exec_domain = NULL;
178 irqctx->tinfo.cpu = cpu; 166 irqctx->tinfo.cpu = cpu;
179 irqctx->tinfo.preempt_count = 0; 167 irqctx->tinfo.preempt_count = 0;
180 irqctx->tinfo.addr_limit = MAKE_MM_SEG(0); 168 irqctx->tinfo.addr_limit = MAKE_MM_SEG(0);
181 169
182 softirq_ctx[cpu] = irqctx; 170 softirq_ctx[cpu] = irqctx;
183 171
184 printk("CPU %u irqstacks, hard=%p soft=%p\n", 172 printk(KERN_DEBUG "CPU %u irqstacks, hard=%p soft=%p\n",
185 cpu,hardirq_ctx[cpu],softirq_ctx[cpu]); 173 cpu,hardirq_ctx[cpu],softirq_ctx[cpu]);
186} 174}
187 175
188void irq_ctx_exit(int cpu) 176void irq_ctx_exit(int cpu)
@@ -211,25 +199,56 @@ asmlinkage void do_softirq(void)
211 /* build the stack frame on the softirq stack */ 199 /* build the stack frame on the softirq stack */
212 isp = (u32*) ((char*)irqctx + sizeof(*irqctx)); 200 isp = (u32*) ((char*)irqctx + sizeof(*irqctx));
213 201
214 asm volatile( 202 call_on_stack(__do_softirq, isp);
215 " xchgl %%ebx,%%esp \n"
216 " call __do_softirq \n"
217 " movl %%ebx,%%esp \n"
218 : "=b"(isp)
219 : "0"(isp)
220 : "memory", "cc", "edx", "ecx", "eax"
221 );
222 /* 203 /*
223 * Shouldnt happen, we returned above if in_interrupt(): 204 * Shouldnt happen, we returned above if in_interrupt():
224 */ 205 */
225 WARN_ON_ONCE(softirq_count()); 206 WARN_ON_ONCE(softirq_count());
226 } 207 }
227 208
228 local_irq_restore(flags); 209 local_irq_restore(flags);
229} 210}
211
212#else
213static inline int
214execute_on_irq_stack(int overflow, struct irq_desc *desc, int irq) { return 0; }
230#endif 215#endif
231 216
232/* 217/*
218 * do_IRQ handles all normal device IRQ's (the special
219 * SMP cross-CPU interrupts have their own specific
220 * handlers).
221 */
222unsigned int do_IRQ(struct pt_regs *regs)
223{
224 struct pt_regs *old_regs;
225 /* high bit used in ret_from_ code */
226 int overflow, irq = ~regs->orig_ax;
227 struct irq_desc *desc = irq_desc + irq;
228
229 if (unlikely((unsigned)irq >= NR_IRQS)) {
230 printk(KERN_EMERG "%s: cannot handle IRQ %d\n",
231 __func__, irq);
232 BUG();
233 }
234
235 old_regs = set_irq_regs(regs);
236 irq_enter();
237
238 overflow = check_stack_overflow();
239
240 if (!execute_on_irq_stack(overflow, desc, irq)) {
241 if (unlikely(overflow))
242 print_stack_overflow();
243 desc->handle_irq(irq, desc);
244 }
245
246 irq_exit();
247 set_irq_regs(old_regs);
248 return 1;
249}
250
251/*
233 * Interrupt statistics: 252 * Interrupt statistics:
234 */ 253 */
235 254
@@ -313,16 +332,20 @@ skip:
313 per_cpu(irq_stat,j).irq_tlb_count); 332 per_cpu(irq_stat,j).irq_tlb_count);
314 seq_printf(p, " TLB shootdowns\n"); 333 seq_printf(p, " TLB shootdowns\n");
315#endif 334#endif
335#ifdef CONFIG_X86_MCE
316 seq_printf(p, "TRM: "); 336 seq_printf(p, "TRM: ");
317 for_each_online_cpu(j) 337 for_each_online_cpu(j)
318 seq_printf(p, "%10u ", 338 seq_printf(p, "%10u ",
319 per_cpu(irq_stat,j).irq_thermal_count); 339 per_cpu(irq_stat,j).irq_thermal_count);
320 seq_printf(p, " Thermal event interrupts\n"); 340 seq_printf(p, " Thermal event interrupts\n");
341#endif
342#ifdef CONFIG_X86_LOCAL_APIC
321 seq_printf(p, "SPU: "); 343 seq_printf(p, "SPU: ");
322 for_each_online_cpu(j) 344 for_each_online_cpu(j)
323 seq_printf(p, "%10u ", 345 seq_printf(p, "%10u ",
324 per_cpu(irq_stat,j).irq_spurious_count); 346 per_cpu(irq_stat,j).irq_spurious_count);
325 seq_printf(p, " Spurious interrupts\n"); 347 seq_printf(p, " Spurious interrupts\n");
348#endif
326 seq_printf(p, "ERR: %10u\n", atomic_read(&irq_err_count)); 349 seq_printf(p, "ERR: %10u\n", atomic_read(&irq_err_count));
327#if defined(CONFIG_X86_IO_APIC) 350#if defined(CONFIG_X86_IO_APIC)
328 seq_printf(p, "MIS: %10u\n", atomic_read(&irq_mis_count)); 351 seq_printf(p, "MIS: %10u\n", atomic_read(&irq_mis_count));
@@ -331,6 +354,40 @@ skip:
331 return 0; 354 return 0;
332} 355}
333 356
357/*
358 * /proc/stat helpers
359 */
360u64 arch_irq_stat_cpu(unsigned int cpu)
361{
362 u64 sum = nmi_count(cpu);
363
364#ifdef CONFIG_X86_LOCAL_APIC
365 sum += per_cpu(irq_stat, cpu).apic_timer_irqs;
366#endif
367#ifdef CONFIG_SMP
368 sum += per_cpu(irq_stat, cpu).irq_resched_count;
369 sum += per_cpu(irq_stat, cpu).irq_call_count;
370 sum += per_cpu(irq_stat, cpu).irq_tlb_count;
371#endif
372#ifdef CONFIG_X86_MCE
373 sum += per_cpu(irq_stat, cpu).irq_thermal_count;
374#endif
375#ifdef CONFIG_X86_LOCAL_APIC
376 sum += per_cpu(irq_stat, cpu).irq_spurious_count;
377#endif
378 return sum;
379}
380
381u64 arch_irq_stat(void)
382{
383 u64 sum = atomic_read(&irq_err_count);
384
385#ifdef CONFIG_X86_IO_APIC
386 sum += atomic_read(&irq_mis_count);
387#endif
388 return sum;
389}
390
334#ifdef CONFIG_HOTPLUG_CPU 391#ifdef CONFIG_HOTPLUG_CPU
335#include <mach_apic.h> 392#include <mach_apic.h>
336 393
diff --git a/arch/x86/kernel/irq_64.c b/arch/x86/kernel/irq_64.c
index 3aac15466a91..1f78b238d8d2 100644
--- a/arch/x86/kernel/irq_64.c
+++ b/arch/x86/kernel/irq_64.c
@@ -135,6 +135,7 @@ skip:
135 seq_printf(p, "%10u ", cpu_pda(j)->irq_tlb_count); 135 seq_printf(p, "%10u ", cpu_pda(j)->irq_tlb_count);
136 seq_printf(p, " TLB shootdowns\n"); 136 seq_printf(p, " TLB shootdowns\n");
137#endif 137#endif
138#ifdef CONFIG_X86_MCE
138 seq_printf(p, "TRM: "); 139 seq_printf(p, "TRM: ");
139 for_each_online_cpu(j) 140 for_each_online_cpu(j)
140 seq_printf(p, "%10u ", cpu_pda(j)->irq_thermal_count); 141 seq_printf(p, "%10u ", cpu_pda(j)->irq_thermal_count);
@@ -143,6 +144,7 @@ skip:
143 for_each_online_cpu(j) 144 for_each_online_cpu(j)
144 seq_printf(p, "%10u ", cpu_pda(j)->irq_threshold_count); 145 seq_printf(p, "%10u ", cpu_pda(j)->irq_threshold_count);
145 seq_printf(p, " Threshold APIC interrupts\n"); 146 seq_printf(p, " Threshold APIC interrupts\n");
147#endif
146 seq_printf(p, "SPU: "); 148 seq_printf(p, "SPU: ");
147 for_each_online_cpu(j) 149 for_each_online_cpu(j)
148 seq_printf(p, "%10u ", cpu_pda(j)->irq_spurious_count); 150 seq_printf(p, "%10u ", cpu_pda(j)->irq_spurious_count);
@@ -153,6 +155,32 @@ skip:
153} 155}
154 156
155/* 157/*
158 * /proc/stat helpers
159 */
160u64 arch_irq_stat_cpu(unsigned int cpu)
161{
162 u64 sum = cpu_pda(cpu)->__nmi_count;
163
164 sum += cpu_pda(cpu)->apic_timer_irqs;
165#ifdef CONFIG_SMP
166 sum += cpu_pda(cpu)->irq_resched_count;
167 sum += cpu_pda(cpu)->irq_call_count;
168 sum += cpu_pda(cpu)->irq_tlb_count;
169#endif
170#ifdef CONFIG_X86_MCE
171 sum += cpu_pda(cpu)->irq_thermal_count;
172 sum += cpu_pda(cpu)->irq_threshold_count;
173#endif
174 sum += cpu_pda(cpu)->irq_spurious_count;
175 return sum;
176}
177
178u64 arch_irq_stat(void)
179{
180 return atomic_read(&irq_err_count);
181}
182
183/*
156 * do_IRQ handles all normal device IRQ's (the special 184 * do_IRQ handles all normal device IRQ's (the special
157 * SMP cross-CPU interrupts have their own specific 185 * SMP cross-CPU interrupts have their own specific
158 * handlers). 186 * handlers).
diff --git a/arch/x86/kernel/irqinit_32.c b/arch/x86/kernel/irqinit_32.c
new file mode 100644
index 000000000000..d66914287ee1
--- /dev/null
+++ b/arch/x86/kernel/irqinit_32.c
@@ -0,0 +1,114 @@
1#include <linux/errno.h>
2#include <linux/signal.h>
3#include <linux/sched.h>
4#include <linux/ioport.h>
5#include <linux/interrupt.h>
6#include <linux/slab.h>
7#include <linux/random.h>
8#include <linux/init.h>
9#include <linux/kernel_stat.h>
10#include <linux/sysdev.h>
11#include <linux/bitops.h>
12
13#include <asm/atomic.h>
14#include <asm/system.h>
15#include <asm/io.h>
16#include <asm/timer.h>
17#include <asm/pgtable.h>
18#include <asm/delay.h>
19#include <asm/desc.h>
20#include <asm/apic.h>
21#include <asm/arch_hooks.h>
22#include <asm/i8259.h>
23
24
25
26/*
27 * Note that on a 486, we don't want to do a SIGFPE on an irq13
28 * as the irq is unreliable, and exception 16 works correctly
29 * (ie as explained in the intel literature). On a 386, you
30 * can't use exception 16 due to bad IBM design, so we have to
31 * rely on the less exact irq13.
32 *
33 * Careful.. Not only is IRQ13 unreliable, but it is also
34 * leads to races. IBM designers who came up with it should
35 * be shot.
36 */
37
38
39static irqreturn_t math_error_irq(int cpl, void *dev_id)
40{
41 extern void math_error(void __user *);
42 outb(0,0xF0);
43 if (ignore_fpu_irq || !boot_cpu_data.hard_math)
44 return IRQ_NONE;
45 math_error((void __user *)get_irq_regs()->ip);
46 return IRQ_HANDLED;
47}
48
49/*
50 * New motherboards sometimes make IRQ 13 be a PCI interrupt,
51 * so allow interrupt sharing.
52 */
53static struct irqaction fpu_irq = {
54 .handler = math_error_irq,
55 .mask = CPU_MASK_NONE,
56 .name = "fpu",
57};
58
59void __init init_ISA_irqs (void)
60{
61 int i;
62
63#ifdef CONFIG_X86_LOCAL_APIC
64 init_bsp_APIC();
65#endif
66 init_8259A(0);
67
68 /*
69 * 16 old-style INTA-cycle interrupts:
70 */
71 for (i = 0; i < 16; i++) {
72 set_irq_chip_and_handler_name(i, &i8259A_chip,
73 handle_level_irq, "XT");
74 }
75}
76
77/* Overridden in paravirt.c */
78void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
79
80void __init native_init_IRQ(void)
81{
82 int i;
83
84 /* all the set up before the call gates are initialised */
85 pre_intr_init_hook();
86
87 /*
88 * Cover the whole vector space, no vector can escape
89 * us. (some of these will be overridden and become
90 * 'special' SMP interrupts)
91 */
92 for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
93 int vector = FIRST_EXTERNAL_VECTOR + i;
94 if (i >= NR_IRQS)
95 break;
96 /* SYSCALL_VECTOR was reserved in trap_init. */
97 if (!test_bit(vector, used_vectors))
98 set_intr_gate(vector, interrupt[i]);
99 }
100
101 /* setup after call gates are initialised (usually add in
102 * the architecture specific gates)
103 */
104 intr_init_hook();
105
106 /*
107 * External FPU? Set up irq13 if so, for
108 * original braindamaged IBM FERR coupling.
109 */
110 if (boot_cpu_data.hard_math && !cpu_has_fpu)
111 setup_irq(FPU_IRQ, &fpu_irq);
112
113 irq_ctx_init(smp_processor_id());
114}
diff --git a/arch/x86/kernel/irqinit_64.c b/arch/x86/kernel/irqinit_64.c
new file mode 100644
index 000000000000..1f26fd9ec4f4
--- /dev/null
+++ b/arch/x86/kernel/irqinit_64.c
@@ -0,0 +1,222 @@
1#include <linux/linkage.h>
2#include <linux/errno.h>
3#include <linux/signal.h>
4#include <linux/sched.h>
5#include <linux/ioport.h>
6#include <linux/interrupt.h>
7#include <linux/timex.h>
8#include <linux/slab.h>
9#include <linux/random.h>
10#include <linux/init.h>
11#include <linux/kernel_stat.h>
12#include <linux/sysdev.h>
13#include <linux/bitops.h>
14
15#include <asm/acpi.h>
16#include <asm/atomic.h>
17#include <asm/system.h>
18#include <asm/io.h>
19#include <asm/hw_irq.h>
20#include <asm/pgtable.h>
21#include <asm/delay.h>
22#include <asm/desc.h>
23#include <asm/apic.h>
24#include <asm/i8259.h>
25
26/*
27 * Common place to define all x86 IRQ vectors
28 *
29 * This builds up the IRQ handler stubs using some ugly macros in irq.h
30 *
31 * These macros create the low-level assembly IRQ routines that save
32 * register context and call do_IRQ(). do_IRQ() then does all the
33 * operations that are needed to keep the AT (or SMP IOAPIC)
34 * interrupt-controller happy.
35 */
36
37#define IRQ_NAME2(nr) nr##_interrupt(void)
38#define IRQ_NAME(nr) IRQ_NAME2(IRQ##nr)
39
40/*
41 * SMP has a few special interrupts for IPI messages
42 */
43
44#define BUILD_IRQ(nr) \
45 asmlinkage void IRQ_NAME(nr); \
46 asm("\n.text\n.p2align\n" \
47 "IRQ" #nr "_interrupt:\n\t" \
48 "push $~(" #nr ") ; " \
49 "jmp common_interrupt\n" \
50 ".previous");
51
52#define BI(x,y) \
53 BUILD_IRQ(x##y)
54
55#define BUILD_16_IRQS(x) \
56 BI(x,0) BI(x,1) BI(x,2) BI(x,3) \
57 BI(x,4) BI(x,5) BI(x,6) BI(x,7) \
58 BI(x,8) BI(x,9) BI(x,a) BI(x,b) \
59 BI(x,c) BI(x,d) BI(x,e) BI(x,f)
60
61/*
62 * ISA PIC or low IO-APIC triggered (INTA-cycle or APIC) interrupts:
63 * (these are usually mapped to vectors 0x30-0x3f)
64 */
65
66/*
67 * The IO-APIC gives us many more interrupt sources. Most of these
68 * are unused but an SMP system is supposed to have enough memory ...
69 * sometimes (mostly wrt. hw bugs) we get corrupted vectors all
70 * across the spectrum, so we really want to be prepared to get all
71 * of these. Plus, more powerful systems might have more than 64
72 * IO-APIC registers.
73 *
74 * (these are usually mapped into the 0x30-0xff vector range)
75 */
76 BUILD_16_IRQS(0x2) BUILD_16_IRQS(0x3)
77BUILD_16_IRQS(0x4) BUILD_16_IRQS(0x5) BUILD_16_IRQS(0x6) BUILD_16_IRQS(0x7)
78BUILD_16_IRQS(0x8) BUILD_16_IRQS(0x9) BUILD_16_IRQS(0xa) BUILD_16_IRQS(0xb)
79BUILD_16_IRQS(0xc) BUILD_16_IRQS(0xd) BUILD_16_IRQS(0xe) BUILD_16_IRQS(0xf)
80
81#undef BUILD_16_IRQS
82#undef BI
83
84
85#define IRQ(x,y) \
86 IRQ##x##y##_interrupt
87
88#define IRQLIST_16(x) \
89 IRQ(x,0), IRQ(x,1), IRQ(x,2), IRQ(x,3), \
90 IRQ(x,4), IRQ(x,5), IRQ(x,6), IRQ(x,7), \
91 IRQ(x,8), IRQ(x,9), IRQ(x,a), IRQ(x,b), \
92 IRQ(x,c), IRQ(x,d), IRQ(x,e), IRQ(x,f)
93
94/* for the irq vectors */
95static void (*__initdata interrupt[NR_VECTORS - FIRST_EXTERNAL_VECTOR])(void) = {
96 IRQLIST_16(0x2), IRQLIST_16(0x3),
97 IRQLIST_16(0x4), IRQLIST_16(0x5), IRQLIST_16(0x6), IRQLIST_16(0x7),
98 IRQLIST_16(0x8), IRQLIST_16(0x9), IRQLIST_16(0xa), IRQLIST_16(0xb),
99 IRQLIST_16(0xc), IRQLIST_16(0xd), IRQLIST_16(0xe), IRQLIST_16(0xf)
100};
101
102#undef IRQ
103#undef IRQLIST_16
104
105
106
107
108/*
109 * IRQ2 is cascade interrupt to second interrupt controller
110 */
111
112static struct irqaction irq2 = {
113 .handler = no_action,
114 .mask = CPU_MASK_NONE,
115 .name = "cascade",
116};
117DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
118 [0 ... IRQ0_VECTOR - 1] = -1,
119 [IRQ0_VECTOR] = 0,
120 [IRQ1_VECTOR] = 1,
121 [IRQ2_VECTOR] = 2,
122 [IRQ3_VECTOR] = 3,
123 [IRQ4_VECTOR] = 4,
124 [IRQ5_VECTOR] = 5,
125 [IRQ6_VECTOR] = 6,
126 [IRQ7_VECTOR] = 7,
127 [IRQ8_VECTOR] = 8,
128 [IRQ9_VECTOR] = 9,
129 [IRQ10_VECTOR] = 10,
130 [IRQ11_VECTOR] = 11,
131 [IRQ12_VECTOR] = 12,
132 [IRQ13_VECTOR] = 13,
133 [IRQ14_VECTOR] = 14,
134 [IRQ15_VECTOR] = 15,
135 [IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
136};
137
138static void __init init_ISA_irqs (void)
139{
140 int i;
141
142 init_bsp_APIC();
143 init_8259A(0);
144
145 for (i = 0; i < NR_IRQS; i++) {
146 irq_desc[i].status = IRQ_DISABLED;
147 irq_desc[i].action = NULL;
148 irq_desc[i].depth = 1;
149
150 if (i < 16) {
151 /*
152 * 16 old-style INTA-cycle interrupts:
153 */
154 set_irq_chip_and_handler_name(i, &i8259A_chip,
155 handle_level_irq, "XT");
156 } else {
157 /*
158 * 'high' PCI IRQs filled in on demand
159 */
160 irq_desc[i].chip = &no_irq_chip;
161 }
162 }
163}
164
165void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
166
167void __init native_init_IRQ(void)
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
184 /*
185 * The reschedule interrupt is a CPU-to-CPU reschedule-helper
186 * IPI, driven by wakeup.
187 */
188 alloc_intr_gate(RESCHEDULE_VECTOR, reschedule_interrupt);
189
190 /* IPIs for invalidation */
191 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+0, invalidate_interrupt0);
192 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+1, invalidate_interrupt1);
193 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+2, invalidate_interrupt2);
194 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+3, invalidate_interrupt3);
195 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+4, invalidate_interrupt4);
196 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+5, invalidate_interrupt5);
197 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+6, invalidate_interrupt6);
198 alloc_intr_gate(INVALIDATE_TLB_VECTOR_START+7, invalidate_interrupt7);
199
200 /* IPI for generic function call */
201 alloc_intr_gate(CALL_FUNCTION_VECTOR, call_function_interrupt);
202
203 /* IPI for generic single function call */
204 alloc_intr_gate(CALL_FUNCTION_SINGLE_VECTOR,
205 call_function_single_interrupt);
206
207 /* Low priority IPI to cleanup after moving an irq */
208 set_intr_gate(IRQ_MOVE_CLEANUP_VECTOR, irq_move_cleanup_interrupt);
209#endif
210 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
211 alloc_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt);
212
213 /* self generated IPI for local APIC timer */
214 alloc_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
215
216 /* IPI vectors for APIC spurious and error interrupts */
217 alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
218 alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
219
220 if (!acpi_ioapic)
221 setup_irq(2, &irq2);
222}
diff --git a/arch/x86/kernel/kdebugfs.c b/arch/x86/kernel/kdebugfs.c
index c03205991718..f2d43bc75514 100644
--- a/arch/x86/kernel/kdebugfs.c
+++ b/arch/x86/kernel/kdebugfs.c
@@ -12,9 +12,13 @@
12#include <linux/init.h> 12#include <linux/init.h>
13#include <linux/io.h> 13#include <linux/io.h>
14#include <linux/mm.h> 14#include <linux/mm.h>
15#include <linux/module.h>
15 16
16#include <asm/setup.h> 17#include <asm/setup.h>
17 18
19struct dentry *arch_debugfs_dir;
20EXPORT_SYMBOL(arch_debugfs_dir);
21
18#ifdef CONFIG_DEBUG_BOOT_PARAMS 22#ifdef CONFIG_DEBUG_BOOT_PARAMS
19struct setup_data_node { 23struct setup_data_node {
20 u64 paddr; 24 u64 paddr;
@@ -209,6 +213,10 @@ static int __init arch_kdebugfs_init(void)
209{ 213{
210 int error = 0; 214 int error = 0;
211 215
216 arch_debugfs_dir = debugfs_create_dir("x86", NULL);
217 if (!arch_debugfs_dir)
218 return -ENOMEM;
219
212#ifdef CONFIG_DEBUG_BOOT_PARAMS 220#ifdef CONFIG_DEBUG_BOOT_PARAMS
213 error = boot_params_kdebugfs_init(); 221 error = boot_params_kdebugfs_init();
214#endif 222#endif
diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
index b8c6743a13da..6c27679ec6aa 100644
--- a/arch/x86/kernel/kprobes.c
+++ b/arch/x86/kernel/kprobes.c
@@ -431,7 +431,6 @@ static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
431 regs->ip = (unsigned long)p->ainsn.insn; 431 regs->ip = (unsigned long)p->ainsn.insn;
432} 432}
433 433
434/* Called with kretprobe_lock held */
435void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri, 434void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
436 struct pt_regs *regs) 435 struct pt_regs *regs)
437{ 436{
@@ -682,8 +681,7 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
682 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline; 681 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
683 682
684 INIT_HLIST_HEAD(&empty_rp); 683 INIT_HLIST_HEAD(&empty_rp);
685 spin_lock_irqsave(&kretprobe_lock, flags); 684 kretprobe_hash_lock(current, &head, &flags);
686 head = kretprobe_inst_table_head(current);
687 /* fixup registers */ 685 /* fixup registers */
688#ifdef CONFIG_X86_64 686#ifdef CONFIG_X86_64
689 regs->cs = __KERNEL_CS; 687 regs->cs = __KERNEL_CS;
@@ -732,7 +730,7 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
732 730
733 kretprobe_assert(ri, orig_ret_address, trampoline_address); 731 kretprobe_assert(ri, orig_ret_address, trampoline_address);
734 732
735 spin_unlock_irqrestore(&kretprobe_lock, flags); 733 kretprobe_hash_unlock(current, &flags);
736 734
737 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) { 735 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
738 hlist_del(&ri->hlist); 736 hlist_del(&ri->hlist);
@@ -860,7 +858,6 @@ static int __kprobes post_kprobe_handler(struct pt_regs *regs)
860 858
861 resume_execution(cur, regs, kcb); 859 resume_execution(cur, regs, kcb);
862 regs->flags |= kcb->kprobe_saved_flags; 860 regs->flags |= kcb->kprobe_saved_flags;
863 trace_hardirqs_fixup_flags(regs->flags);
864 861
865 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) { 862 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
866 kcb->kprobe_status = KPROBE_HIT_SSDONE; 863 kcb->kprobe_status = KPROBE_HIT_SSDONE;
diff --git a/arch/x86/kernel/kvmclock.c b/arch/x86/kernel/kvmclock.c
index 08a30986d472..d02def06ca91 100644
--- a/arch/x86/kernel/kvmclock.c
+++ b/arch/x86/kernel/kvmclock.c
@@ -18,6 +18,7 @@
18 18
19#include <linux/clocksource.h> 19#include <linux/clocksource.h>
20#include <linux/kvm_para.h> 20#include <linux/kvm_para.h>
21#include <asm/pvclock.h>
21#include <asm/arch_hooks.h> 22#include <asm/arch_hooks.h>
22#include <asm/msr.h> 23#include <asm/msr.h>
23#include <asm/apic.h> 24#include <asm/apic.h>
@@ -36,18 +37,9 @@ static int parse_no_kvmclock(char *arg)
36early_param("no-kvmclock", parse_no_kvmclock); 37early_param("no-kvmclock", parse_no_kvmclock);
37 38
38/* The hypervisor will put information about time periodically here */ 39/* The hypervisor will put information about time periodically here */
39static DEFINE_PER_CPU_SHARED_ALIGNED(struct kvm_vcpu_time_info, hv_clock); 40static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
40#define get_clock(cpu, field) per_cpu(hv_clock, cpu).field 41static struct pvclock_wall_clock wall_clock;
41 42
42static inline u64 kvm_get_delta(u64 last_tsc)
43{
44 int cpu = smp_processor_id();
45 u64 delta = native_read_tsc() - last_tsc;
46 return (delta * get_clock(cpu, tsc_to_system_mul)) >> KVM_SCALE;
47}
48
49static struct kvm_wall_clock wall_clock;
50static cycle_t kvm_clock_read(void);
51/* 43/*
52 * The wallclock is the time of day when we booted. Since then, some time may 44 * The wallclock is the time of day when we booted. Since then, some time may
53 * have elapsed since the hypervisor wrote the data. So we try to account for 45 * have elapsed since the hypervisor wrote the data. So we try to account for
@@ -55,64 +47,37 @@ static cycle_t kvm_clock_read(void);
55 */ 47 */
56static unsigned long kvm_get_wallclock(void) 48static unsigned long kvm_get_wallclock(void)
57{ 49{
58 u32 wc_sec, wc_nsec; 50 struct pvclock_vcpu_time_info *vcpu_time;
59 u64 delta;
60 struct timespec ts; 51 struct timespec ts;
61 int version, nsec;
62 int low, high; 52 int low, high;
63 53
64 low = (int)__pa(&wall_clock); 54 low = (int)__pa(&wall_clock);
65 high = ((u64)__pa(&wall_clock) >> 32); 55 high = ((u64)__pa(&wall_clock) >> 32);
56 native_write_msr(MSR_KVM_WALL_CLOCK, low, high);
66 57
67 delta = kvm_clock_read(); 58 vcpu_time = &get_cpu_var(hv_clock);
59 pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
60 put_cpu_var(hv_clock);
68 61
69 native_write_msr(MSR_KVM_WALL_CLOCK, low, high); 62 return ts.tv_sec;
70 do {
71 version = wall_clock.wc_version;
72 rmb();
73 wc_sec = wall_clock.wc_sec;
74 wc_nsec = wall_clock.wc_nsec;
75 rmb();
76 } while ((wall_clock.wc_version != version) || (version & 1));
77
78 delta = kvm_clock_read() - delta;
79 delta += wc_nsec;
80 nsec = do_div(delta, NSEC_PER_SEC);
81 set_normalized_timespec(&ts, wc_sec + delta, nsec);
82 /*
83 * Of all mechanisms of time adjustment I've tested, this one
84 * was the champion!
85 */
86 return ts.tv_sec + 1;
87} 63}
88 64
89static int kvm_set_wallclock(unsigned long now) 65static int kvm_set_wallclock(unsigned long now)
90{ 66{
91 return 0; 67 return -1;
92} 68}
93 69
94/*
95 * This is our read_clock function. The host puts an tsc timestamp each time
96 * it updates a new time. Without the tsc adjustment, we can have a situation
97 * in which a vcpu starts to run earlier (smaller system_time), but probes
98 * time later (compared to another vcpu), leading to backwards time
99 */
100static cycle_t kvm_clock_read(void) 70static cycle_t kvm_clock_read(void)
101{ 71{
102 u64 last_tsc, now; 72 struct pvclock_vcpu_time_info *src;
103 int cpu; 73 cycle_t ret;
104 74
105 preempt_disable(); 75 src = &get_cpu_var(hv_clock);
106 cpu = smp_processor_id(); 76 ret = pvclock_clocksource_read(src);
107 77 put_cpu_var(hv_clock);
108 last_tsc = get_clock(cpu, tsc_timestamp); 78 return ret;
109 now = get_clock(cpu, system_time);
110
111 now += kvm_get_delta(last_tsc);
112 preempt_enable();
113
114 return now;
115} 79}
80
116static struct clocksource kvm_clock = { 81static struct clocksource kvm_clock = {
117 .name = "kvm-clock", 82 .name = "kvm-clock",
118 .read = kvm_clock_read, 83 .read = kvm_clock_read,
@@ -123,13 +88,14 @@ static struct clocksource kvm_clock = {
123 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 88 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
124}; 89};
125 90
126static int kvm_register_clock(void) 91static int kvm_register_clock(char *txt)
127{ 92{
128 int cpu = smp_processor_id(); 93 int cpu = smp_processor_id();
129 int low, high; 94 int low, high;
130 low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1; 95 low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
131 high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32); 96 high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
132 97 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
98 cpu, high, low, txt);
133 return native_write_msr_safe(MSR_KVM_SYSTEM_TIME, low, high); 99 return native_write_msr_safe(MSR_KVM_SYSTEM_TIME, low, high);
134} 100}
135 101
@@ -140,12 +106,20 @@ static void kvm_setup_secondary_clock(void)
140 * Now that the first cpu already had this clocksource initialized, 106 * Now that the first cpu already had this clocksource initialized,
141 * we shouldn't fail. 107 * we shouldn't fail.
142 */ 108 */
143 WARN_ON(kvm_register_clock()); 109 WARN_ON(kvm_register_clock("secondary cpu clock"));
144 /* ok, done with our trickery, call native */ 110 /* ok, done with our trickery, call native */
145 setup_secondary_APIC_clock(); 111 setup_secondary_APIC_clock();
146} 112}
147#endif 113#endif
148 114
115#ifdef CONFIG_SMP
116static void __init kvm_smp_prepare_boot_cpu(void)
117{
118 WARN_ON(kvm_register_clock("primary cpu clock"));
119 native_smp_prepare_boot_cpu();
120}
121#endif
122
149/* 123/*
150 * After the clock is registered, the host will keep writing to the 124 * After the clock is registered, the host will keep writing to the
151 * registered memory location. If the guest happens to shutdown, this memory 125 * registered memory location. If the guest happens to shutdown, this memory
@@ -174,7 +148,7 @@ void __init kvmclock_init(void)
174 return; 148 return;
175 149
176 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) { 150 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
177 if (kvm_register_clock()) 151 if (kvm_register_clock("boot clock"))
178 return; 152 return;
179 pv_time_ops.get_wallclock = kvm_get_wallclock; 153 pv_time_ops.get_wallclock = kvm_get_wallclock;
180 pv_time_ops.set_wallclock = kvm_set_wallclock; 154 pv_time_ops.set_wallclock = kvm_set_wallclock;
@@ -182,6 +156,9 @@ void __init kvmclock_init(void)
182#ifdef CONFIG_X86_LOCAL_APIC 156#ifdef CONFIG_X86_LOCAL_APIC
183 pv_apic_ops.setup_secondary_clock = kvm_setup_secondary_clock; 157 pv_apic_ops.setup_secondary_clock = kvm_setup_secondary_clock;
184#endif 158#endif
159#ifdef CONFIG_SMP
160 smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
161#endif
185 machine_ops.shutdown = kvm_shutdown; 162 machine_ops.shutdown = kvm_shutdown;
186#ifdef CONFIG_KEXEC 163#ifdef CONFIG_KEXEC
187 machine_ops.crash_shutdown = kvm_crash_shutdown; 164 machine_ops.crash_shutdown = kvm_crash_shutdown;
diff --git a/arch/x86/kernel/ldt.c b/arch/x86/kernel/ldt.c
index 0224c3637c73..b68e21f06f4f 100644
--- a/arch/x86/kernel/ldt.c
+++ b/arch/x86/kernel/ldt.c
@@ -20,9 +20,9 @@
20#include <asm/mmu_context.h> 20#include <asm/mmu_context.h>
21 21
22#ifdef CONFIG_SMP 22#ifdef CONFIG_SMP
23static void flush_ldt(void *null) 23static void flush_ldt(void *current_mm)
24{ 24{
25 if (current->active_mm) 25 if (current->active_mm == current_mm)
26 load_LDT(&current->active_mm->context); 26 load_LDT(&current->active_mm->context);
27} 27}
28#endif 28#endif
@@ -62,13 +62,11 @@ static int alloc_ldt(mm_context_t *pc, int mincount, int reload)
62 62
63 if (reload) { 63 if (reload) {
64#ifdef CONFIG_SMP 64#ifdef CONFIG_SMP
65 cpumask_t mask;
66
67 preempt_disable(); 65 preempt_disable();
68 load_LDT(pc); 66 load_LDT(pc);
69 mask = cpumask_of_cpu(smp_processor_id()); 67 if (!cpus_equal(current->mm->cpu_vm_mask,
70 if (!cpus_equal(current->mm->cpu_vm_mask, mask)) 68 cpumask_of_cpu(smp_processor_id())))
71 smp_call_function(flush_ldt, NULL, 1, 1); 69 smp_call_function(flush_ldt, current->mm, 1);
72 preempt_enable(); 70 preempt_enable();
73#else 71#else
74 load_LDT(pc); 72 load_LDT(pc);
diff --git a/arch/x86/kernel/machine_kexec_32.c b/arch/x86/kernel/machine_kexec_32.c
index d0b234c9fc31..9fe478d98406 100644
--- a/arch/x86/kernel/machine_kexec_32.c
+++ b/arch/x86/kernel/machine_kexec_32.c
@@ -11,6 +11,8 @@
11#include <linux/delay.h> 11#include <linux/delay.h>
12#include <linux/init.h> 12#include <linux/init.h>
13#include <linux/numa.h> 13#include <linux/numa.h>
14#include <linux/ftrace.h>
15
14#include <asm/pgtable.h> 16#include <asm/pgtable.h>
15#include <asm/pgalloc.h> 17#include <asm/pgalloc.h>
16#include <asm/tlbflush.h> 18#include <asm/tlbflush.h>
@@ -20,6 +22,7 @@
20#include <asm/cpufeature.h> 22#include <asm/cpufeature.h>
21#include <asm/desc.h> 23#include <asm/desc.h>
22#include <asm/system.h> 24#include <asm/system.h>
25#include <asm/cacheflush.h>
23 26
24#define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE))) 27#define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE)))
25static u32 kexec_pgd[1024] PAGE_ALIGNED; 28static u32 kexec_pgd[1024] PAGE_ALIGNED;
@@ -39,7 +42,7 @@ static void set_idt(void *newidt, __u16 limit)
39 curidt.address = (unsigned long)newidt; 42 curidt.address = (unsigned long)newidt;
40 43
41 load_idt(&curidt); 44 load_idt(&curidt);
42}; 45}
43 46
44 47
45static void set_gdt(void *newgdt, __u16 limit) 48static void set_gdt(void *newgdt, __u16 limit)
@@ -51,7 +54,7 @@ static void set_gdt(void *newgdt, __u16 limit)
51 curgdt.address = (unsigned long)newgdt; 54 curgdt.address = (unsigned long)newgdt;
52 55
53 load_gdt(&curgdt); 56 load_gdt(&curgdt);
54}; 57}
55 58
56static void load_segments(void) 59static void load_segments(void)
57{ 60{
@@ -83,10 +86,12 @@ static void load_segments(void)
83 * reboot code buffer to allow us to avoid allocations 86 * reboot code buffer to allow us to avoid allocations
84 * later. 87 * later.
85 * 88 *
86 * Currently nothing. 89 * Make control page executable.
87 */ 90 */
88int machine_kexec_prepare(struct kimage *image) 91int machine_kexec_prepare(struct kimage *image)
89{ 92{
93 if (nx_enabled)
94 set_pages_x(image->control_code_page, 1);
90 return 0; 95 return 0;
91} 96}
92 97
@@ -96,25 +101,48 @@ int machine_kexec_prepare(struct kimage *image)
96 */ 101 */
97void machine_kexec_cleanup(struct kimage *image) 102void machine_kexec_cleanup(struct kimage *image)
98{ 103{
104 if (nx_enabled)
105 set_pages_nx(image->control_code_page, 1);
99} 106}
100 107
101/* 108/*
102 * Do not allocate memory (or fail in any way) in machine_kexec(). 109 * Do not allocate memory (or fail in any way) in machine_kexec().
103 * We are past the point of no return, committed to rebooting now. 110 * We are past the point of no return, committed to rebooting now.
104 */ 111 */
105NORET_TYPE void machine_kexec(struct kimage *image) 112void machine_kexec(struct kimage *image)
106{ 113{
107 unsigned long page_list[PAGES_NR]; 114 unsigned long page_list[PAGES_NR];
108 void *control_page; 115 void *control_page;
116 asmlinkage unsigned long
117 (*relocate_kernel_ptr)(unsigned long indirection_page,
118 unsigned long control_page,
119 unsigned long start_address,
120 unsigned int has_pae,
121 unsigned int preserve_context);
122
123 tracer_disable();
109 124
110 /* Interrupts aren't acceptable while we reboot */ 125 /* Interrupts aren't acceptable while we reboot */
111 local_irq_disable(); 126 local_irq_disable();
112 127
128 if (image->preserve_context) {
129#ifdef CONFIG_X86_IO_APIC
130 /* We need to put APICs in legacy mode so that we can
131 * get timer interrupts in second kernel. kexec/kdump
132 * paths already have calls to disable_IO_APIC() in
133 * one form or other. kexec jump path also need
134 * one.
135 */
136 disable_IO_APIC();
137#endif
138 }
139
113 control_page = page_address(image->control_code_page); 140 control_page = page_address(image->control_code_page);
114 memcpy(control_page, relocate_kernel, PAGE_SIZE); 141 memcpy(control_page, relocate_kernel, PAGE_SIZE/2);
115 142
143 relocate_kernel_ptr = control_page;
116 page_list[PA_CONTROL_PAGE] = __pa(control_page); 144 page_list[PA_CONTROL_PAGE] = __pa(control_page);
117 page_list[VA_CONTROL_PAGE] = (unsigned long)relocate_kernel; 145 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
118 page_list[PA_PGD] = __pa(kexec_pgd); 146 page_list[PA_PGD] = __pa(kexec_pgd);
119 page_list[VA_PGD] = (unsigned long)kexec_pgd; 147 page_list[VA_PGD] = (unsigned long)kexec_pgd;
120#ifdef CONFIG_X86_PAE 148#ifdef CONFIG_X86_PAE
@@ -127,6 +155,7 @@ NORET_TYPE void machine_kexec(struct kimage *image)
127 page_list[VA_PTE_0] = (unsigned long)kexec_pte0; 155 page_list[VA_PTE_0] = (unsigned long)kexec_pte0;
128 page_list[PA_PTE_1] = __pa(kexec_pte1); 156 page_list[PA_PTE_1] = __pa(kexec_pte1);
129 page_list[VA_PTE_1] = (unsigned long)kexec_pte1; 157 page_list[VA_PTE_1] = (unsigned long)kexec_pte1;
158 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page) << PAGE_SHIFT);
130 159
131 /* The segment registers are funny things, they have both a 160 /* The segment registers are funny things, they have both a
132 * visible and an invisible part. Whenever the visible part is 161 * visible and an invisible part. Whenever the visible part is
@@ -145,8 +174,10 @@ NORET_TYPE void machine_kexec(struct kimage *image)
145 set_idt(phys_to_virt(0),0); 174 set_idt(phys_to_virt(0),0);
146 175
147 /* now call it */ 176 /* now call it */
148 relocate_kernel((unsigned long)image->head, (unsigned long)page_list, 177 image->start = relocate_kernel_ptr((unsigned long)image->head,
149 image->start, cpu_has_pae); 178 (unsigned long)page_list,
179 image->start, cpu_has_pae,
180 image->preserve_context);
150} 181}
151 182
152void arch_crash_save_vmcoreinfo(void) 183void arch_crash_save_vmcoreinfo(void)
diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c
index 576a03db4511..c43caa3a91f3 100644
--- a/arch/x86/kernel/machine_kexec_64.c
+++ b/arch/x86/kernel/machine_kexec_64.c
@@ -11,6 +11,8 @@
11#include <linux/string.h> 11#include <linux/string.h>
12#include <linux/reboot.h> 12#include <linux/reboot.h>
13#include <linux/numa.h> 13#include <linux/numa.h>
14#include <linux/ftrace.h>
15
14#include <asm/pgtable.h> 16#include <asm/pgtable.h>
15#include <asm/tlbflush.h> 17#include <asm/tlbflush.h>
16#include <asm/mmu_context.h> 18#include <asm/mmu_context.h>
@@ -110,7 +112,7 @@ static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
110{ 112{
111 pgd_t *level4p; 113 pgd_t *level4p;
112 level4p = (pgd_t *)__va(start_pgtable); 114 level4p = (pgd_t *)__va(start_pgtable);
113 return init_level4_page(image, level4p, 0, end_pfn << PAGE_SHIFT); 115 return init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
114} 116}
115 117
116static void set_idt(void *newidt, u16 limit) 118static void set_idt(void *newidt, u16 limit)
@@ -179,11 +181,13 @@ void machine_kexec_cleanup(struct kimage *image)
179 * Do not allocate memory (or fail in any way) in machine_kexec(). 181 * Do not allocate memory (or fail in any way) in machine_kexec().
180 * We are past the point of no return, committed to rebooting now. 182 * We are past the point of no return, committed to rebooting now.
181 */ 183 */
182NORET_TYPE void machine_kexec(struct kimage *image) 184void machine_kexec(struct kimage *image)
183{ 185{
184 unsigned long page_list[PAGES_NR]; 186 unsigned long page_list[PAGES_NR];
185 void *control_page; 187 void *control_page;
186 188
189 tracer_disable();
190
187 /* Interrupts aren't acceptable while we reboot */ 191 /* Interrupts aren't acceptable while we reboot */
188 local_irq_disable(); 192 local_irq_disable();
189 193
diff --git a/arch/x86/kernel/microcode.c b/arch/x86/kernel/microcode.c
index 69729e38b78a..652fa5c38ebe 100644
--- a/arch/x86/kernel/microcode.c
+++ b/arch/x86/kernel/microcode.c
@@ -5,13 +5,14 @@
5 * 2006 Shaohua Li <shaohua.li@intel.com> 5 * 2006 Shaohua Li <shaohua.li@intel.com>
6 * 6 *
7 * This driver allows to upgrade microcode on Intel processors 7 * This driver allows to upgrade microcode on Intel processors
8 * belonging to IA-32 family - PentiumPro, Pentium II, 8 * belonging to IA-32 family - PentiumPro, Pentium II,
9 * Pentium III, Xeon, Pentium 4, etc. 9 * Pentium III, Xeon, Pentium 4, etc.
10 * 10 *
11 * Reference: Section 8.10 of Volume III, Intel Pentium 4 Manual, 11 * Reference: Section 8.11 of Volume 3a, IA-32 Intel? Architecture
12 * Order Number 245472 or free download from: 12 * Software Developer's Manual
13 * 13 * Order Number 253668 or free download from:
14 * http://developer.intel.com/design/pentium4/manuals/245472.htm 14 *
15 * http://developer.intel.com/design/pentium4/manuals/253668.htm
15 * 16 *
16 * For more information, go to http://www.urbanmyth.org/microcode 17 * For more information, go to http://www.urbanmyth.org/microcode
17 * 18 *
@@ -58,12 +59,12 @@
58 * nature of implementation. 59 * nature of implementation.
59 * 1.11 22 Mar 2002 Tigran Aivazian <tigran@veritas.com> 60 * 1.11 22 Mar 2002 Tigran Aivazian <tigran@veritas.com>
60 * Fix the panic when writing zero-length microcode chunk. 61 * Fix the panic when writing zero-length microcode chunk.
61 * 1.12 29 Sep 2003 Nitin Kamble <nitin.a.kamble@intel.com>, 62 * 1.12 29 Sep 2003 Nitin Kamble <nitin.a.kamble@intel.com>,
62 * Jun Nakajima <jun.nakajima@intel.com> 63 * Jun Nakajima <jun.nakajima@intel.com>
63 * Support for the microcode updates in the new format. 64 * Support for the microcode updates in the new format.
64 * 1.13 10 Oct 2003 Tigran Aivazian <tigran@veritas.com> 65 * 1.13 10 Oct 2003 Tigran Aivazian <tigran@veritas.com>
65 * Removed ->read() method and obsoleted MICROCODE_IOCFREE ioctl 66 * Removed ->read() method and obsoleted MICROCODE_IOCFREE ioctl
66 * because we no longer hold a copy of applied microcode 67 * because we no longer hold a copy of applied microcode
67 * in kernel memory. 68 * in kernel memory.
68 * 1.14 25 Jun 2004 Tigran Aivazian <tigran@veritas.com> 69 * 1.14 25 Jun 2004 Tigran Aivazian <tigran@veritas.com>
69 * Fix sigmatch() macro to handle old CPUs with pf == 0. 70 * Fix sigmatch() macro to handle old CPUs with pf == 0.
@@ -75,6 +76,7 @@
75#include <linux/kernel.h> 76#include <linux/kernel.h>
76#include <linux/init.h> 77#include <linux/init.h>
77#include <linux/sched.h> 78#include <linux/sched.h>
79#include <linux/smp_lock.h>
78#include <linux/cpumask.h> 80#include <linux/cpumask.h>
79#include <linux/module.h> 81#include <linux/module.h>
80#include <linux/slab.h> 82#include <linux/slab.h>
@@ -320,11 +322,11 @@ static void apply_microcode(int cpu)
320 return; 322 return;
321 323
322 /* serialize access to the physical write to MSR 0x79 */ 324 /* serialize access to the physical write to MSR 0x79 */
323 spin_lock_irqsave(&microcode_update_lock, flags); 325 spin_lock_irqsave(&microcode_update_lock, flags);
324 326
325 /* write microcode via MSR 0x79 */ 327 /* write microcode via MSR 0x79 */
326 wrmsr(MSR_IA32_UCODE_WRITE, 328 wrmsr(MSR_IA32_UCODE_WRITE,
327 (unsigned long) uci->mc->bits, 329 (unsigned long) uci->mc->bits,
328 (unsigned long) uci->mc->bits >> 16 >> 16); 330 (unsigned long) uci->mc->bits >> 16 >> 16);
329 wrmsr(MSR_IA32_UCODE_REV, 0, 0); 331 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
330 332
@@ -341,7 +343,7 @@ static void apply_microcode(int cpu)
341 return; 343 return;
342 } 344 }
343 printk(KERN_INFO "microcode: CPU%d updated from revision " 345 printk(KERN_INFO "microcode: CPU%d updated from revision "
344 "0x%x to 0x%x, date = %08x \n", 346 "0x%x to 0x%x, date = %08x \n",
345 cpu_num, uci->rev, val[1], uci->mc->hdr.date); 347 cpu_num, uci->rev, val[1], uci->mc->hdr.date);
346 uci->rev = val[1]; 348 uci->rev = val[1];
347} 349}
@@ -422,6 +424,7 @@ out:
422 424
423static int microcode_open (struct inode *unused1, struct file *unused2) 425static int microcode_open (struct inode *unused1, struct file *unused2)
424{ 426{
427 cycle_kernel_lock();
425 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM; 428 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
426} 429}
427 430
@@ -488,7 +491,7 @@ MODULE_ALIAS_MISCDEV(MICROCODE_MINOR);
488#define microcode_dev_exit() do { } while(0) 491#define microcode_dev_exit() do { } while(0)
489#endif 492#endif
490 493
491static long get_next_ucode_from_buffer(void **mc, void *buf, 494static long get_next_ucode_from_buffer(void **mc, const u8 *buf,
492 unsigned long size, long offset) 495 unsigned long size, long offset)
493{ 496{
494 microcode_header_t *mc_header; 497 microcode_header_t *mc_header;
@@ -522,7 +525,7 @@ static int cpu_request_microcode(int cpu)
522 char name[30]; 525 char name[30];
523 struct cpuinfo_x86 *c = &cpu_data(cpu); 526 struct cpuinfo_x86 *c = &cpu_data(cpu);
524 const struct firmware *firmware; 527 const struct firmware *firmware;
525 void *buf; 528 const u8 *buf;
526 unsigned long size; 529 unsigned long size;
527 long offset = 0; 530 long offset = 0;
528 int error; 531 int error;
@@ -534,7 +537,7 @@ static int cpu_request_microcode(int cpu)
534 c->x86, c->x86_model, c->x86_mask); 537 c->x86, c->x86_model, c->x86_mask);
535 error = request_firmware(&firmware, name, &microcode_pdev->dev); 538 error = request_firmware(&firmware, name, &microcode_pdev->dev);
536 if (error) { 539 if (error) {
537 pr_debug("microcode: ucode data file %s load failed\n", name); 540 pr_debug("microcode: data file %s load failed\n", name);
538 return error; 541 return error;
539 } 542 }
540 buf = firmware->data; 543 buf = firmware->data;
@@ -641,7 +644,9 @@ static void microcode_fini_cpu(int cpu)
641 mutex_unlock(&microcode_mutex); 644 mutex_unlock(&microcode_mutex);
642} 645}
643 646
644static ssize_t reload_store(struct sys_device *dev, const char *buf, size_t sz) 647static ssize_t reload_store(struct sys_device *dev,
648 struct sysdev_attribute *attr,
649 const char *buf, size_t sz)
645{ 650{
646 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id; 651 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
647 char *end; 652 char *end;
@@ -652,9 +657,7 @@ static ssize_t reload_store(struct sys_device *dev, const char *buf, size_t sz)
652 if (end == buf) 657 if (end == buf)
653 return -EINVAL; 658 return -EINVAL;
654 if (val == 1) { 659 if (val == 1) {
655 cpumask_t old; 660 cpumask_t old = current->cpus_allowed;
656
657 old = current->cpus_allowed;
658 661
659 get_online_cpus(); 662 get_online_cpus();
660 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); 663 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
@@ -671,14 +674,16 @@ static ssize_t reload_store(struct sys_device *dev, const char *buf, size_t sz)
671 return sz; 674 return sz;
672} 675}
673 676
674static ssize_t version_show(struct sys_device *dev, char *buf) 677static ssize_t version_show(struct sys_device *dev,
678 struct sysdev_attribute *attr, char *buf)
675{ 679{
676 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id; 680 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
677 681
678 return sprintf(buf, "0x%x\n", uci->rev); 682 return sprintf(buf, "0x%x\n", uci->rev);
679} 683}
680 684
681static ssize_t pf_show(struct sys_device *dev, char *buf) 685static ssize_t pf_show(struct sys_device *dev,
686 struct sysdev_attribute *attr, char *buf)
682{ 687{
683 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id; 688 struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
684 689
@@ -805,6 +810,9 @@ static int __init microcode_init (void)
805{ 810{
806 int error; 811 int error;
807 812
813 printk(KERN_INFO
814 "IA-32 Microcode Update Driver: v" MICROCODE_VERSION " <tigran@aivazian.fsnet.co.uk>\n");
815
808 error = microcode_dev_init(); 816 error = microcode_dev_init();
809 if (error) 817 if (error)
810 return error; 818 return error;
@@ -825,9 +833,6 @@ static int __init microcode_init (void)
825 } 833 }
826 834
827 register_hotcpu_notifier(&mc_cpu_notifier); 835 register_hotcpu_notifier(&mc_cpu_notifier);
828
829 printk(KERN_INFO
830 "IA-32 Microcode Update Driver: v" MICROCODE_VERSION " <tigran@aivazian.fsnet.co.uk>\n");
831 return 0; 836 return 0;
832} 837}
833 838
diff --git a/arch/x86/kernel/mmconf-fam10h_64.c b/arch/x86/kernel/mmconf-fam10h_64.c
index edc5fbfe85c0..fdfdc550b366 100644
--- a/arch/x86/kernel/mmconf-fam10h_64.c
+++ b/arch/x86/kernel/mmconf-fam10h_64.c
@@ -12,6 +12,7 @@
12#include <asm/io.h> 12#include <asm/io.h>
13#include <asm/msr.h> 13#include <asm/msr.h>
14#include <asm/acpi.h> 14#include <asm/acpi.h>
15#include <asm/mmconfig.h>
15 16
16#include "../pci/pci.h" 17#include "../pci/pci.h"
17 18
diff --git a/arch/x86/kernel/module_64.c b/arch/x86/kernel/module_64.c
index a888e67f5874..6ba87830d4b1 100644
--- a/arch/x86/kernel/module_64.c
+++ b/arch/x86/kernel/module_64.c
@@ -22,6 +22,7 @@
22#include <linux/fs.h> 22#include <linux/fs.h>
23#include <linux/string.h> 23#include <linux/string.h>
24#include <linux/kernel.h> 24#include <linux/kernel.h>
25#include <linux/mm.h>
25#include <linux/slab.h> 26#include <linux/slab.h>
26#include <linux/bug.h> 27#include <linux/bug.h>
27 28
@@ -150,7 +151,8 @@ int module_finalize(const Elf_Ehdr *hdr,
150 const Elf_Shdr *sechdrs, 151 const Elf_Shdr *sechdrs,
151 struct module *me) 152 struct module *me)
152{ 153{
153 const Elf_Shdr *s, *text = NULL, *alt = NULL, *locks = NULL; 154 const Elf_Shdr *s, *text = NULL, *alt = NULL, *locks = NULL,
155 *para = NULL;
154 char *secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset; 156 char *secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
155 157
156 for (s = sechdrs; s < sechdrs + hdr->e_shnum; s++) { 158 for (s = sechdrs; s < sechdrs + hdr->e_shnum; s++) {
@@ -160,6 +162,8 @@ int module_finalize(const Elf_Ehdr *hdr,
160 alt = s; 162 alt = s;
161 if (!strcmp(".smp_locks", secstrings + s->sh_name)) 163 if (!strcmp(".smp_locks", secstrings + s->sh_name))
162 locks= s; 164 locks= s;
165 if (!strcmp(".parainstructions", secstrings + s->sh_name))
166 para = s;
163 } 167 }
164 168
165 if (alt) { 169 if (alt) {
@@ -175,6 +179,11 @@ int module_finalize(const Elf_Ehdr *hdr,
175 tseg, tseg + text->sh_size); 179 tseg, tseg + text->sh_size);
176 } 180 }
177 181
182 if (para) {
183 void *pseg = (void *)para->sh_addr;
184 apply_paravirt(pseg, pseg + para->sh_size);
185 }
186
178 return module_bug_finalize(hdr, sechdrs, me); 187 return module_bug_finalize(hdr, sechdrs, me);
179} 188}
180 189
diff --git a/arch/x86/kernel/mpparse.c b/arch/x86/kernel/mpparse.c
index 404683b94e79..6ae005ccaed8 100644
--- a/arch/x86/kernel/mpparse.c
+++ b/arch/x86/kernel/mpparse.c
@@ -25,6 +25,9 @@
25#include <asm/proto.h> 25#include <asm/proto.h>
26#include <asm/acpi.h> 26#include <asm/acpi.h>
27#include <asm/bios_ebda.h> 27#include <asm/bios_ebda.h>
28#include <asm/e820.h>
29#include <asm/trampoline.h>
30#include <asm/setup.h>
28 31
29#include <mach_apic.h> 32#include <mach_apic.h>
30#ifdef CONFIG_X86_32 33#ifdef CONFIG_X86_32
@@ -32,28 +35,6 @@
32#include <mach_mpparse.h> 35#include <mach_mpparse.h>
33#endif 36#endif
34 37
35/* Have we found an MP table */
36int smp_found_config;
37
38/*
39 * Various Linux-internal data structures created from the
40 * MP-table.
41 */
42#if defined (CONFIG_MCA) || defined (CONFIG_EISA)
43int mp_bus_id_to_type[MAX_MP_BUSSES];
44#endif
45
46DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
47int mp_bus_id_to_pci_bus[MAX_MP_BUSSES] = {[0 ... MAX_MP_BUSSES - 1] = -1 };
48
49static int mp_current_pci_id;
50
51int pic_mode;
52
53/*
54 * Intel MP BIOS table parsing routines:
55 */
56
57/* 38/*
58 * Checksum an MP configuration block. 39 * Checksum an MP configuration block.
59 */ 40 */
@@ -68,18 +49,6 @@ static int __init mpf_checksum(unsigned char *mp, int len)
68 return sum & 0xFF; 49 return sum & 0xFF;
69} 50}
70 51
71#ifdef CONFIG_X86_NUMAQ
72/*
73 * Have to match translation table entries to main table entries by counter
74 * hence the mpc_record variable .... can't see a less disgusting way of
75 * doing this ....
76 */
77
78static int mpc_record;
79static struct mpc_config_translation *translation_table[MAX_MPC_ENTRY]
80 __cpuinitdata;
81#endif
82
83static void __cpuinit MP_processor_info(struct mpc_config_processor *m) 52static void __cpuinit MP_processor_info(struct mpc_config_processor *m)
84{ 53{
85 int apicid; 54 int apicid;
@@ -89,11 +58,12 @@ static void __cpuinit MP_processor_info(struct mpc_config_processor *m)
89 disabled_cpus++; 58 disabled_cpus++;
90 return; 59 return;
91 } 60 }
92#ifdef CONFIG_X86_NUMAQ 61
93 apicid = mpc_apic_id(m, translation_table[mpc_record]); 62 if (x86_quirks->mpc_apic_id)
94#else 63 apicid = x86_quirks->mpc_apic_id(m);
95 apicid = m->mpc_apicid; 64 else
96#endif 65 apicid = m->mpc_apicid;
66
97 if (m->mpc_cpuflag & CPU_BOOTPROCESSOR) { 67 if (m->mpc_cpuflag & CPU_BOOTPROCESSOR) {
98 bootup_cpu = " (Bootup-CPU)"; 68 bootup_cpu = " (Bootup-CPU)";
99 boot_cpu_physical_apicid = m->mpc_apicid; 69 boot_cpu_physical_apicid = m->mpc_apicid;
@@ -103,18 +73,17 @@ static void __cpuinit MP_processor_info(struct mpc_config_processor *m)
103 generic_processor_info(apicid, m->mpc_apicver); 73 generic_processor_info(apicid, m->mpc_apicver);
104} 74}
105 75
76#ifdef CONFIG_X86_IO_APIC
106static void __init MP_bus_info(struct mpc_config_bus *m) 77static void __init MP_bus_info(struct mpc_config_bus *m)
107{ 78{
108 char str[7]; 79 char str[7];
109
110 memcpy(str, m->mpc_bustype, 6); 80 memcpy(str, m->mpc_bustype, 6);
111 str[6] = 0; 81 str[6] = 0;
112 82
113#ifdef CONFIG_X86_NUMAQ 83 if (x86_quirks->mpc_oem_bus_info)
114 mpc_oem_bus_info(m, str, translation_table[mpc_record]); 84 x86_quirks->mpc_oem_bus_info(m, str);
115#else 85 else
116 Dprintk("Bus #%d is %s\n", m->mpc_busid, str); 86 printk(KERN_INFO "Bus #%d is %s\n", m->mpc_busid, str);
117#endif
118 87
119#if MAX_MP_BUSSES < 256 88#if MAX_MP_BUSSES < 256
120 if (m->mpc_busid >= MAX_MP_BUSSES) { 89 if (m->mpc_busid >= MAX_MP_BUSSES) {
@@ -131,12 +100,10 @@ static void __init MP_bus_info(struct mpc_config_bus *m)
131 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_ISA; 100 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_ISA;
132#endif 101#endif
133 } else if (strncmp(str, BUSTYPE_PCI, sizeof(BUSTYPE_PCI) - 1) == 0) { 102 } else if (strncmp(str, BUSTYPE_PCI, sizeof(BUSTYPE_PCI) - 1) == 0) {
134#ifdef CONFIG_X86_NUMAQ 103 if (x86_quirks->mpc_oem_pci_bus)
135 mpc_oem_pci_bus(m, translation_table[mpc_record]); 104 x86_quirks->mpc_oem_pci_bus(m);
136#endif 105
137 clear_bit(m->mpc_busid, mp_bus_not_pci); 106 clear_bit(m->mpc_busid, mp_bus_not_pci);
138 mp_bus_id_to_pci_bus[m->mpc_busid] = mp_current_pci_id;
139 mp_current_pci_id++;
140#if defined(CONFIG_EISA) || defined (CONFIG_MCA) 107#if defined(CONFIG_EISA) || defined (CONFIG_MCA)
141 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_PCI; 108 mp_bus_id_to_type[m->mpc_busid] = MP_BUS_PCI;
142 } else if (strncmp(str, BUSTYPE_EISA, sizeof(BUSTYPE_EISA) - 1) == 0) { 109 } else if (strncmp(str, BUSTYPE_EISA, sizeof(BUSTYPE_EISA) - 1) == 0) {
@@ -147,6 +114,7 @@ static void __init MP_bus_info(struct mpc_config_bus *m)
147 } else 114 } else
148 printk(KERN_WARNING "Unknown bustype %s - ignoring\n", str); 115 printk(KERN_WARNING "Unknown bustype %s - ignoring\n", str);
149} 116}
117#endif
150 118
151#ifdef CONFIG_X86_IO_APIC 119#ifdef CONFIG_X86_IO_APIC
152 120
@@ -176,117 +144,111 @@ static void __init MP_ioapic_info(struct mpc_config_ioapic *m)
176 if (bad_ioapic(m->mpc_apicaddr)) 144 if (bad_ioapic(m->mpc_apicaddr))
177 return; 145 return;
178 146
179 mp_ioapics[nr_ioapics] = *m; 147 mp_ioapics[nr_ioapics].mp_apicaddr = m->mpc_apicaddr;
148 mp_ioapics[nr_ioapics].mp_apicid = m->mpc_apicid;
149 mp_ioapics[nr_ioapics].mp_type = m->mpc_type;
150 mp_ioapics[nr_ioapics].mp_apicver = m->mpc_apicver;
151 mp_ioapics[nr_ioapics].mp_flags = m->mpc_flags;
180 nr_ioapics++; 152 nr_ioapics++;
181} 153}
182 154
183static void __init MP_intsrc_info(struct mpc_config_intsrc *m) 155static void print_MP_intsrc_info(struct mpc_config_intsrc *m)
184{ 156{
185 mp_irqs[mp_irq_entries] = *m; 157 printk(KERN_CONT "Int: type %d, pol %d, trig %d, bus %02x,"
186 Dprintk("Int: type %d, pol %d, trig %d, bus %d,"
187 " IRQ %02x, APIC ID %x, APIC INT %02x\n", 158 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
188 m->mpc_irqtype, m->mpc_irqflag & 3, 159 m->mpc_irqtype, m->mpc_irqflag & 3,
189 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbus, 160 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbus,
190 m->mpc_srcbusirq, m->mpc_dstapic, m->mpc_dstirq); 161 m->mpc_srcbusirq, m->mpc_dstapic, m->mpc_dstirq);
191 if (++mp_irq_entries == MAX_IRQ_SOURCES)
192 panic("Max # of irq sources exceeded!!\n");
193} 162}
194 163
195#endif 164static void __init print_mp_irq_info(struct mp_config_intsrc *mp_irq)
196
197static void __init MP_lintsrc_info(struct mpc_config_lintsrc *m)
198{ 165{
199 Dprintk("Lint: type %d, pol %d, trig %d, bus %d," 166 printk(KERN_CONT "Int: type %d, pol %d, trig %d, bus %02x,"
200 " IRQ %02x, APIC ID %x, APIC LINT %02x\n", 167 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
201 m->mpc_irqtype, m->mpc_irqflag & 3, 168 mp_irq->mp_irqtype, mp_irq->mp_irqflag & 3,
202 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbusid, 169 (mp_irq->mp_irqflag >> 2) & 3, mp_irq->mp_srcbus,
203 m->mpc_srcbusirq, m->mpc_destapic, m->mpc_destapiclint); 170 mp_irq->mp_srcbusirq, mp_irq->mp_dstapic, mp_irq->mp_dstirq);
204} 171}
205 172
206#ifdef CONFIG_X86_NUMAQ 173static void __init assign_to_mp_irq(struct mpc_config_intsrc *m,
207static void __init MP_translation_info(struct mpc_config_translation *m) 174 struct mp_config_intsrc *mp_irq)
208{ 175{
209 printk(KERN_INFO 176 mp_irq->mp_dstapic = m->mpc_dstapic;
210 "Translation: record %d, type %d, quad %d, global %d, local %d\n", 177 mp_irq->mp_type = m->mpc_type;
211 mpc_record, m->trans_type, m->trans_quad, m->trans_global, 178 mp_irq->mp_irqtype = m->mpc_irqtype;
212 m->trans_local); 179 mp_irq->mp_irqflag = m->mpc_irqflag;
180 mp_irq->mp_srcbus = m->mpc_srcbus;
181 mp_irq->mp_srcbusirq = m->mpc_srcbusirq;
182 mp_irq->mp_dstirq = m->mpc_dstirq;
183}
213 184
214 if (mpc_record >= MAX_MPC_ENTRY) 185static void __init assign_to_mpc_intsrc(struct mp_config_intsrc *mp_irq,
215 printk(KERN_ERR "MAX_MPC_ENTRY exceeded!\n"); 186 struct mpc_config_intsrc *m)
216 else 187{
217 translation_table[mpc_record] = m; /* stash this for later */ 188 m->mpc_dstapic = mp_irq->mp_dstapic;
218 if (m->trans_quad < MAX_NUMNODES && !node_online(m->trans_quad)) 189 m->mpc_type = mp_irq->mp_type;
219 node_set_online(m->trans_quad); 190 m->mpc_irqtype = mp_irq->mp_irqtype;
191 m->mpc_irqflag = mp_irq->mp_irqflag;
192 m->mpc_srcbus = mp_irq->mp_srcbus;
193 m->mpc_srcbusirq = mp_irq->mp_srcbusirq;
194 m->mpc_dstirq = mp_irq->mp_dstirq;
220} 195}
221 196
222/* 197static int __init mp_irq_mpc_intsrc_cmp(struct mp_config_intsrc *mp_irq,
223 * Read/parse the MPC oem tables 198 struct mpc_config_intsrc *m)
224 */ 199{
200 if (mp_irq->mp_dstapic != m->mpc_dstapic)
201 return 1;
202 if (mp_irq->mp_type != m->mpc_type)
203 return 2;
204 if (mp_irq->mp_irqtype != m->mpc_irqtype)
205 return 3;
206 if (mp_irq->mp_irqflag != m->mpc_irqflag)
207 return 4;
208 if (mp_irq->mp_srcbus != m->mpc_srcbus)
209 return 5;
210 if (mp_irq->mp_srcbusirq != m->mpc_srcbusirq)
211 return 6;
212 if (mp_irq->mp_dstirq != m->mpc_dstirq)
213 return 7;
214
215 return 0;
216}
225 217
226static void __init smp_read_mpc_oem(struct mp_config_oemtable *oemtable, 218static void __init MP_intsrc_info(struct mpc_config_intsrc *m)
227 unsigned short oemsize)
228{ 219{
229 int count = sizeof(*oemtable); /* the header size */ 220 int i;
230 unsigned char *oemptr = ((unsigned char *)oemtable) + count; 221
231 222 print_MP_intsrc_info(m);
232 mpc_record = 0; 223
233 printk(KERN_INFO "Found an OEM MPC table at %8p - parsing it ... \n", 224 for (i = 0; i < mp_irq_entries; i++) {
234 oemtable); 225 if (!mp_irq_mpc_intsrc_cmp(&mp_irqs[i], m))
235 if (memcmp(oemtable->oem_signature, MPC_OEM_SIGNATURE, 4)) { 226 return;
236 printk(KERN_WARNING
237 "SMP mpc oemtable: bad signature [%c%c%c%c]!\n",
238 oemtable->oem_signature[0], oemtable->oem_signature[1],
239 oemtable->oem_signature[2], oemtable->oem_signature[3]);
240 return;
241 }
242 if (mpf_checksum((unsigned char *)oemtable, oemtable->oem_length)) {
243 printk(KERN_WARNING "SMP oem mptable: checksum error!\n");
244 return;
245 }
246 while (count < oemtable->oem_length) {
247 switch (*oemptr) {
248 case MP_TRANSLATION:
249 {
250 struct mpc_config_translation *m =
251 (struct mpc_config_translation *)oemptr;
252 MP_translation_info(m);
253 oemptr += sizeof(*m);
254 count += sizeof(*m);
255 ++mpc_record;
256 break;
257 }
258 default:
259 {
260 printk(KERN_WARNING
261 "Unrecognised OEM table entry type! - %d\n",
262 (int)*oemptr);
263 return;
264 }
265 }
266 } 227 }
228
229 assign_to_mp_irq(m, &mp_irqs[mp_irq_entries]);
230 if (++mp_irq_entries == MAX_IRQ_SOURCES)
231 panic("Max # of irq sources exceeded!!\n");
267} 232}
268 233
269static inline void mps_oem_check(struct mp_config_table *mpc, char *oem, 234#endif
270 char *productid) 235
236static void __init MP_lintsrc_info(struct mpc_config_lintsrc *m)
271{ 237{
272 if (strncmp(oem, "IBM NUMA", 8)) 238 printk(KERN_INFO "Lint: type %d, pol %d, trig %d, bus %02x,"
273 printk("Warning! May not be a NUMA-Q system!\n"); 239 " IRQ %02x, APIC ID %x, APIC LINT %02x\n",
274 if (mpc->mpc_oemptr) 240 m->mpc_irqtype, m->mpc_irqflag & 3,
275 smp_read_mpc_oem((struct mp_config_oemtable *)mpc->mpc_oemptr, 241 (m->mpc_irqflag >> 2) & 3, m->mpc_srcbusid,
276 mpc->mpc_oemsize); 242 m->mpc_srcbusirq, m->mpc_destapic, m->mpc_destapiclint);
277} 243}
278#endif /* CONFIG_X86_NUMAQ */
279 244
280/* 245/*
281 * Read/parse the MPC 246 * Read/parse the MPC
282 */ 247 */
283 248
284static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early) 249static int __init smp_check_mpc(struct mp_config_table *mpc, char *oem,
250 char *str)
285{ 251{
286 char str[16];
287 char oem[10];
288 int count = sizeof(*mpc);
289 unsigned char *mpt = ((unsigned char *)mpc) + count;
290 252
291 if (memcmp(mpc->mpc_signature, MPC_SIGNATURE, 4)) { 253 if (memcmp(mpc->mpc_signature, MPC_SIGNATURE, 4)) {
292 printk(KERN_ERR "MPTABLE: bad signature [%c%c%c%c]!\n", 254 printk(KERN_ERR "MPTABLE: bad signature [%c%c%c%c]!\n",
@@ -309,19 +271,41 @@ static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
309 } 271 }
310 memcpy(oem, mpc->mpc_oem, 8); 272 memcpy(oem, mpc->mpc_oem, 8);
311 oem[8] = 0; 273 oem[8] = 0;
312 printk(KERN_INFO "MPTABLE: OEM ID: %s ", oem); 274 printk(KERN_INFO "MPTABLE: OEM ID: %s\n", oem);
313 275
314 memcpy(str, mpc->mpc_productid, 12); 276 memcpy(str, mpc->mpc_productid, 12);
315 str[12] = 0; 277 str[12] = 0;
316 printk("Product ID: %s ", str);
317 278
318#ifdef CONFIG_X86_32 279 printk(KERN_INFO "MPTABLE: Product ID: %s\n", str);
319 mps_oem_check(mpc, oem, str);
320#endif
321 printk(KERN_INFO "MPTABLE: Product ID: %s ", str);
322 280
323 printk(KERN_INFO "MPTABLE: APIC at: 0x%X\n", mpc->mpc_lapic); 281 printk(KERN_INFO "MPTABLE: APIC at: 0x%X\n", mpc->mpc_lapic);
324 282
283 return 1;
284}
285
286static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
287{
288 char str[16];
289 char oem[10];
290
291 int count = sizeof(*mpc);
292 unsigned char *mpt = ((unsigned char *)mpc) + count;
293
294 if (!smp_check_mpc(mpc, oem, str))
295 return 0;
296
297#ifdef CONFIG_X86_32
298 /*
299 * need to make sure summit and es7000's mps_oem_check is safe to be
300 * called early via genericarch 's mps_oem_check
301 */
302 if (early) {
303#ifdef CONFIG_X86_NUMAQ
304 numaq_mps_oem_check(mpc, oem, str);
305#endif
306 } else
307 mps_oem_check(mpc, oem, str);
308#endif
325 /* save the local APIC address, it might be non-default */ 309 /* save the local APIC address, it might be non-default */
326 if (!acpi_lapic) 310 if (!acpi_lapic)
327 mp_lapic_addr = mpc->mpc_lapic; 311 mp_lapic_addr = mpc->mpc_lapic;
@@ -329,12 +313,17 @@ static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
329 if (early) 313 if (early)
330 return 1; 314 return 1;
331 315
316 if (mpc->mpc_oemptr && x86_quirks->smp_read_mpc_oem) {
317 struct mp_config_oemtable *oem_table = (struct mp_config_oemtable *)(unsigned long)mpc->mpc_oemptr;
318 x86_quirks->smp_read_mpc_oem(oem_table, mpc->mpc_oemsize);
319 }
320
332 /* 321 /*
333 * Now process the configuration blocks. 322 * Now process the configuration blocks.
334 */ 323 */
335#ifdef CONFIG_X86_NUMAQ 324 if (x86_quirks->mpc_record)
336 mpc_record = 0; 325 *x86_quirks->mpc_record = 0;
337#endif 326
338 while (count < mpc->mpc_length) { 327 while (count < mpc->mpc_length) {
339 switch (*mpt) { 328 switch (*mpt) {
340 case MP_PROCESSOR: 329 case MP_PROCESSOR:
@@ -352,7 +341,9 @@ static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
352 { 341 {
353 struct mpc_config_bus *m = 342 struct mpc_config_bus *m =
354 (struct mpc_config_bus *)mpt; 343 (struct mpc_config_bus *)mpt;
344#ifdef CONFIG_X86_IO_APIC
355 MP_bus_info(m); 345 MP_bus_info(m);
346#endif
356 mpt += sizeof(*m); 347 mpt += sizeof(*m);
357 count += sizeof(*m); 348 count += sizeof(*m);
358 break; 349 break;
@@ -398,10 +389,14 @@ static int __init smp_read_mpc(struct mp_config_table *mpc, unsigned early)
398 count = mpc->mpc_length; 389 count = mpc->mpc_length;
399 break; 390 break;
400 } 391 }
401#ifdef CONFIG_X86_NUMAQ 392 if (x86_quirks->mpc_record)
402 ++mpc_record; 393 (*x86_quirks->mpc_record)++;
403#endif
404 } 394 }
395
396#ifdef CONFIG_X86_GENERICARCH
397 generic_bigsmp_probe();
398#endif
399
405 setup_apic_routing(); 400 setup_apic_routing();
406 if (!num_processors) 401 if (!num_processors)
407 printk(KERN_ERR "MPTABLE: no processors registered!\n"); 402 printk(KERN_ERR "MPTABLE: no processors registered!\n");
@@ -427,7 +422,7 @@ static void __init construct_default_ioirq_mptable(int mpc_default_type)
427 intsrc.mpc_type = MP_INTSRC; 422 intsrc.mpc_type = MP_INTSRC;
428 intsrc.mpc_irqflag = 0; /* conforming */ 423 intsrc.mpc_irqflag = 0; /* conforming */
429 intsrc.mpc_srcbus = 0; 424 intsrc.mpc_srcbus = 0;
430 intsrc.mpc_dstapic = mp_ioapics[0].mpc_apicid; 425 intsrc.mpc_dstapic = mp_ioapics[0].mp_apicid;
431 426
432 intsrc.mpc_irqtype = mp_INT; 427 intsrc.mpc_irqtype = mp_INT;
433 428
@@ -488,40 +483,11 @@ static void __init construct_default_ioirq_mptable(int mpc_default_type)
488 MP_intsrc_info(&intsrc); 483 MP_intsrc_info(&intsrc);
489} 484}
490 485
491#endif
492 486
493static inline void __init construct_default_ISA_mptable(int mpc_default_type) 487static void construct_ioapic_table(int mpc_default_type)
494{ 488{
495 struct mpc_config_processor processor;
496 struct mpc_config_bus bus;
497#ifdef CONFIG_X86_IO_APIC
498 struct mpc_config_ioapic ioapic; 489 struct mpc_config_ioapic ioapic;
499#endif 490 struct mpc_config_bus bus;
500 struct mpc_config_lintsrc lintsrc;
501 int linttypes[2] = { mp_ExtINT, mp_NMI };
502 int i;
503
504 /*
505 * local APIC has default address
506 */
507 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
508
509 /*
510 * 2 CPUs, numbered 0 & 1.
511 */
512 processor.mpc_type = MP_PROCESSOR;
513 /* Either an integrated APIC or a discrete 82489DX. */
514 processor.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01;
515 processor.mpc_cpuflag = CPU_ENABLED;
516 processor.mpc_cpufeature = (boot_cpu_data.x86 << 8) |
517 (boot_cpu_data.x86_model << 4) | boot_cpu_data.x86_mask;
518 processor.mpc_featureflag = boot_cpu_data.x86_capability[0];
519 processor.mpc_reserved[0] = 0;
520 processor.mpc_reserved[1] = 0;
521 for (i = 0; i < 2; i++) {
522 processor.mpc_apicid = i;
523 MP_processor_info(&processor);
524 }
525 491
526 bus.mpc_type = MP_BUS; 492 bus.mpc_type = MP_BUS;
527 bus.mpc_busid = 0; 493 bus.mpc_busid = 0;
@@ -550,7 +516,6 @@ static inline void __init construct_default_ISA_mptable(int mpc_default_type)
550 MP_bus_info(&bus); 516 MP_bus_info(&bus);
551 } 517 }
552 518
553#ifdef CONFIG_X86_IO_APIC
554 ioapic.mpc_type = MP_IOAPIC; 519 ioapic.mpc_type = MP_IOAPIC;
555 ioapic.mpc_apicid = 2; 520 ioapic.mpc_apicid = 2;
556 ioapic.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01; 521 ioapic.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01;
@@ -562,7 +527,42 @@ static inline void __init construct_default_ISA_mptable(int mpc_default_type)
562 * We set up most of the low 16 IO-APIC pins according to MPS rules. 527 * We set up most of the low 16 IO-APIC pins according to MPS rules.
563 */ 528 */
564 construct_default_ioirq_mptable(mpc_default_type); 529 construct_default_ioirq_mptable(mpc_default_type);
530}
531#else
532static inline void construct_ioapic_table(int mpc_default_type) { }
565#endif 533#endif
534
535static inline void __init construct_default_ISA_mptable(int mpc_default_type)
536{
537 struct mpc_config_processor processor;
538 struct mpc_config_lintsrc lintsrc;
539 int linttypes[2] = { mp_ExtINT, mp_NMI };
540 int i;
541
542 /*
543 * local APIC has default address
544 */
545 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
546
547 /*
548 * 2 CPUs, numbered 0 & 1.
549 */
550 processor.mpc_type = MP_PROCESSOR;
551 /* Either an integrated APIC or a discrete 82489DX. */
552 processor.mpc_apicver = mpc_default_type > 4 ? 0x10 : 0x01;
553 processor.mpc_cpuflag = CPU_ENABLED;
554 processor.mpc_cpufeature = (boot_cpu_data.x86 << 8) |
555 (boot_cpu_data.x86_model << 4) | boot_cpu_data.x86_mask;
556 processor.mpc_featureflag = boot_cpu_data.x86_capability[0];
557 processor.mpc_reserved[0] = 0;
558 processor.mpc_reserved[1] = 0;
559 for (i = 0; i < 2; i++) {
560 processor.mpc_apicid = i;
561 MP_processor_info(&processor);
562 }
563
564 construct_ioapic_table(mpc_default_type);
565
566 lintsrc.mpc_type = MP_LINTSRC; 566 lintsrc.mpc_type = MP_LINTSRC;
567 lintsrc.mpc_irqflag = 0; /* conforming */ 567 lintsrc.mpc_irqflag = 0; /* conforming */
568 lintsrc.mpc_srcbusid = 0; 568 lintsrc.mpc_srcbusid = 0;
@@ -580,10 +580,14 @@ static struct intel_mp_floating *mpf_found;
580/* 580/*
581 * Scan the memory blocks for an SMP configuration block. 581 * Scan the memory blocks for an SMP configuration block.
582 */ 582 */
583static void __init __get_smp_config(unsigned early) 583static void __init __get_smp_config(unsigned int early)
584{ 584{
585 struct intel_mp_floating *mpf = mpf_found; 585 struct intel_mp_floating *mpf = mpf_found;
586 586
587 if (x86_quirks->mach_get_smp_config) {
588 if (x86_quirks->mach_get_smp_config(early))
589 return;
590 }
587 if (acpi_lapic && early) 591 if (acpi_lapic && early)
588 return; 592 return;
589 /* 593 /*
@@ -600,7 +604,7 @@ static void __init __get_smp_config(unsigned early)
600 604
601 printk(KERN_INFO "Intel MultiProcessor Specification v1.%d\n", 605 printk(KERN_INFO "Intel MultiProcessor Specification v1.%d\n",
602 mpf->mpf_specification); 606 mpf->mpf_specification);
603#ifdef CONFIG_X86_32 607#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86_32)
604 if (mpf->mpf_feature2 & (1 << 7)) { 608 if (mpf->mpf_feature2 & (1 << 7)) {
605 printk(KERN_INFO " IMCR and PIC compatibility mode.\n"); 609 printk(KERN_INFO " IMCR and PIC compatibility mode.\n");
606 pic_mode = 1; 610 pic_mode = 1;
@@ -632,7 +636,9 @@ static void __init __get_smp_config(unsigned early)
632 * override the defaults. 636 * override the defaults.
633 */ 637 */
634 if (!smp_read_mpc(phys_to_virt(mpf->mpf_physptr), early)) { 638 if (!smp_read_mpc(phys_to_virt(mpf->mpf_physptr), early)) {
639#ifdef CONFIG_X86_LOCAL_APIC
635 smp_found_config = 0; 640 smp_found_config = 0;
641#endif
636 printk(KERN_ERR 642 printk(KERN_ERR
637 "BIOS bug, MP table errors detected!...\n"); 643 "BIOS bug, MP table errors detected!...\n");
638 printk(KERN_ERR "... disabling SMP support. " 644 printk(KERN_ERR "... disabling SMP support. "
@@ -689,7 +695,7 @@ static int __init smp_scan_config(unsigned long base, unsigned long length,
689 unsigned int *bp = phys_to_virt(base); 695 unsigned int *bp = phys_to_virt(base);
690 struct intel_mp_floating *mpf; 696 struct intel_mp_floating *mpf;
691 697
692 Dprintk("Scan SMP from %p for %ld bytes.\n", bp, length); 698 printk(KERN_DEBUG "Scan SMP from %p for %ld bytes.\n", bp, length);
693 BUILD_BUG_ON(sizeof(*mpf) != 16); 699 BUILD_BUG_ON(sizeof(*mpf) != 16);
694 700
695 while (length > 0) { 701 while (length > 0) {
@@ -699,15 +705,21 @@ static int __init smp_scan_config(unsigned long base, unsigned long length,
699 !mpf_checksum((unsigned char *)bp, 16) && 705 !mpf_checksum((unsigned char *)bp, 16) &&
700 ((mpf->mpf_specification == 1) 706 ((mpf->mpf_specification == 1)
701 || (mpf->mpf_specification == 4))) { 707 || (mpf->mpf_specification == 4))) {
702 708#ifdef CONFIG_X86_LOCAL_APIC
703 smp_found_config = 1; 709 smp_found_config = 1;
710#endif
704 mpf_found = mpf; 711 mpf_found = mpf;
705#ifdef CONFIG_X86_32 712
706 printk(KERN_INFO "found SMP MP-table at [%p] %08lx\n", 713 printk(KERN_INFO "found SMP MP-table at [%p] %08lx\n",
707 mpf, virt_to_phys(mpf)); 714 mpf, virt_to_phys(mpf));
708 reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE, 715
716 if (!reserve)
717 return 1;
718 reserve_bootmem_generic(virt_to_phys(mpf), PAGE_SIZE,
709 BOOTMEM_DEFAULT); 719 BOOTMEM_DEFAULT);
710 if (mpf->mpf_physptr) { 720 if (mpf->mpf_physptr) {
721 unsigned long size = PAGE_SIZE;
722#ifdef CONFIG_X86_32
711 /* 723 /*
712 * We cannot access to MPC table to compute 724 * We cannot access to MPC table to compute
713 * table size yet, as only few megabytes from 725 * table size yet, as only few megabytes from
@@ -717,24 +729,15 @@ static int __init smp_scan_config(unsigned long base, unsigned long length,
717 * PAGE_SIZE from mpg->mpf_physptr yields BUG() 729 * PAGE_SIZE from mpg->mpf_physptr yields BUG()
718 * in reserve_bootmem. 730 * in reserve_bootmem.
719 */ 731 */
720 unsigned long size = PAGE_SIZE;
721 unsigned long end = max_low_pfn * PAGE_SIZE; 732 unsigned long end = max_low_pfn * PAGE_SIZE;
722 if (mpf->mpf_physptr + size > end) 733 if (mpf->mpf_physptr + size > end)
723 size = end - mpf->mpf_physptr; 734 size = end - mpf->mpf_physptr;
724 reserve_bootmem(mpf->mpf_physptr, size, 735#endif
736 reserve_bootmem_generic(mpf->mpf_physptr, size,
725 BOOTMEM_DEFAULT); 737 BOOTMEM_DEFAULT);
726 } 738 }
727 739
728#else 740 return 1;
729 if (!reserve)
730 return 1;
731
732 reserve_bootmem_generic(virt_to_phys(mpf), PAGE_SIZE);
733 if (mpf->mpf_physptr)
734 reserve_bootmem_generic(mpf->mpf_physptr,
735 PAGE_SIZE);
736#endif
737 return 1;
738 } 741 }
739 bp += 4; 742 bp += 4;
740 length -= 16; 743 length -= 16;
@@ -742,10 +745,14 @@ static int __init smp_scan_config(unsigned long base, unsigned long length,
742 return 0; 745 return 0;
743} 746}
744 747
745static void __init __find_smp_config(unsigned reserve) 748static void __init __find_smp_config(unsigned int reserve)
746{ 749{
747 unsigned int address; 750 unsigned int address;
748 751
752 if (x86_quirks->mach_find_smp_config) {
753 if (x86_quirks->mach_find_smp_config(reserve))
754 return;
755 }
749 /* 756 /*
750 * FIXME: Linux assumes you have 640K of base ram.. 757 * FIXME: Linux assumes you have 640K of base ram..
751 * this continues the error... 758 * this continues the error...
@@ -790,298 +797,294 @@ void __init find_smp_config(void)
790 __find_smp_config(1); 797 __find_smp_config(1);
791} 798}
792 799
793/* -------------------------------------------------------------------------- 800#ifdef CONFIG_X86_IO_APIC
794 ACPI-based MP Configuration 801static u8 __initdata irq_used[MAX_IRQ_SOURCES];
795 -------------------------------------------------------------------------- */
796 802
797/* 803static int __init get_MP_intsrc_index(struct mpc_config_intsrc *m)
798 * Keep this outside and initialized to 0, for !CONFIG_ACPI builds: 804{
799 */ 805 int i;
800int es7000_plat;
801 806
802#ifdef CONFIG_ACPI 807 if (m->mpc_irqtype != mp_INT)
808 return 0;
803 809
804#ifdef CONFIG_X86_IO_APIC 810 if (m->mpc_irqflag != 0x0f)
811 return 0;
805 812
806#define MP_ISA_BUS 0 813 /* not legacy */
807 814
808extern struct mp_ioapic_routing mp_ioapic_routing[MAX_IO_APICS]; 815 for (i = 0; i < mp_irq_entries; i++) {
816 if (mp_irqs[i].mp_irqtype != mp_INT)
817 continue;
809 818
810static int mp_find_ioapic(int gsi) 819 if (mp_irqs[i].mp_irqflag != 0x0f)
811{ 820 continue;
812 int i = 0;
813 821
814 /* Find the IOAPIC that manages this GSI. */ 822 if (mp_irqs[i].mp_srcbus != m->mpc_srcbus)
815 for (i = 0; i < nr_ioapics; i++) { 823 continue;
816 if ((gsi >= mp_ioapic_routing[i].gsi_base) 824 if (mp_irqs[i].mp_srcbusirq != m->mpc_srcbusirq)
817 && (gsi <= mp_ioapic_routing[i].gsi_end)) 825 continue;
818 return i; 826 if (irq_used[i]) {
827 /* already claimed */
828 return -2;
829 }
830 irq_used[i] = 1;
831 return i;
819 } 832 }
820 833
821 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi); 834 /* not found */
822 return -1; 835 return -1;
823} 836}
824 837
825static u8 __init uniq_ioapic_id(u8 id) 838#define SPARE_SLOT_NUM 20
826{ 839
827#ifdef CONFIG_X86_32 840static struct mpc_config_intsrc __initdata *m_spare[SPARE_SLOT_NUM];
828 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
829 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
830 return io_apic_get_unique_id(nr_ioapics, id);
831 else
832 return id;
833#else
834 int i;
835 DECLARE_BITMAP(used, 256);
836 bitmap_zero(used, 256);
837 for (i = 0; i < nr_ioapics; i++) {
838 struct mpc_config_ioapic *ia = &mp_ioapics[i];
839 __set_bit(ia->mpc_apicid, used);
840 }
841 if (!test_bit(id, used))
842 return id;
843 return find_first_zero_bit(used, 256);
844#endif 841#endif
845}
846 842
847void __init mp_register_ioapic(int id, u32 address, u32 gsi_base) 843static int __init replace_intsrc_all(struct mp_config_table *mpc,
844 unsigned long mpc_new_phys,
845 unsigned long mpc_new_length)
848{ 846{
849 int idx = 0; 847#ifdef CONFIG_X86_IO_APIC
850 848 int i;
851 if (bad_ioapic(address)) 849 int nr_m_spare = 0;
852 return; 850#endif
853 851
854 idx = nr_ioapics; 852 int count = sizeof(*mpc);
853 unsigned char *mpt = ((unsigned char *)mpc) + count;
855 854
856 mp_ioapics[idx].mpc_type = MP_IOAPIC; 855 printk(KERN_INFO "mpc_length %x\n", mpc->mpc_length);
857 mp_ioapics[idx].mpc_flags = MPC_APIC_USABLE; 856 while (count < mpc->mpc_length) {
858 mp_ioapics[idx].mpc_apicaddr = address; 857 switch (*mpt) {
858 case MP_PROCESSOR:
859 {
860 struct mpc_config_processor *m =
861 (struct mpc_config_processor *)mpt;
862 mpt += sizeof(*m);
863 count += sizeof(*m);
864 break;
865 }
866 case MP_BUS:
867 {
868 struct mpc_config_bus *m =
869 (struct mpc_config_bus *)mpt;
870 mpt += sizeof(*m);
871 count += sizeof(*m);
872 break;
873 }
874 case MP_IOAPIC:
875 {
876 mpt += sizeof(struct mpc_config_ioapic);
877 count += sizeof(struct mpc_config_ioapic);
878 break;
879 }
880 case MP_INTSRC:
881 {
882#ifdef CONFIG_X86_IO_APIC
883 struct mpc_config_intsrc *m =
884 (struct mpc_config_intsrc *)mpt;
859 885
860 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address); 886 printk(KERN_INFO "OLD ");
861 mp_ioapics[idx].mpc_apicid = uniq_ioapic_id(id); 887 print_MP_intsrc_info(m);
862#ifdef CONFIG_X86_32 888 i = get_MP_intsrc_index(m);
863 mp_ioapics[idx].mpc_apicver = io_apic_get_version(idx); 889 if (i > 0) {
864#else 890 assign_to_mpc_intsrc(&mp_irqs[i], m);
865 mp_ioapics[idx].mpc_apicver = 0; 891 printk(KERN_INFO "NEW ");
892 print_mp_irq_info(&mp_irqs[i]);
893 } else if (!i) {
894 /* legacy, do nothing */
895 } else if (nr_m_spare < SPARE_SLOT_NUM) {
896 /*
897 * not found (-1), or duplicated (-2)
898 * are invalid entries,
899 * we need to use the slot later
900 */
901 m_spare[nr_m_spare] = m;
902 nr_m_spare++;
903 }
866#endif 904#endif
867 /* 905 mpt += sizeof(struct mpc_config_intsrc);
868 * Build basic GSI lookup table to facilitate gsi->io_apic lookups 906 count += sizeof(struct mpc_config_intsrc);
869 * and to prevent reprogramming of IOAPIC pins (PCI GSIs). 907 break;
870 */ 908 }
871 mp_ioapic_routing[idx].apic_id = mp_ioapics[idx].mpc_apicid; 909 case MP_LINTSRC:
872 mp_ioapic_routing[idx].gsi_base = gsi_base; 910 {
873 mp_ioapic_routing[idx].gsi_end = gsi_base + 911 struct mpc_config_lintsrc *m =
874 io_apic_get_redir_entries(idx); 912 (struct mpc_config_lintsrc *)mpt;
875 913 mpt += sizeof(*m);
876 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, " 914 count += sizeof(*m);
877 "GSI %d-%d\n", idx, mp_ioapics[idx].mpc_apicid, 915 break;
878 mp_ioapics[idx].mpc_apicver, mp_ioapics[idx].mpc_apicaddr, 916 }
879 mp_ioapic_routing[idx].gsi_base, mp_ioapic_routing[idx].gsi_end); 917 default:
880 918 /* wrong mptable */
881 nr_ioapics++; 919 printk(KERN_ERR "Your mptable is wrong, contact your HW vendor!\n");
882} 920 printk(KERN_ERR "type %x\n", *mpt);
921 print_hex_dump(KERN_ERR, " ", DUMP_PREFIX_ADDRESS, 16,
922 1, mpc, mpc->mpc_length, 1);
923 goto out;
924 }
925 }
883 926
884void __init mp_override_legacy_irq(u8 bus_irq, u8 polarity, u8 trigger, u32 gsi) 927#ifdef CONFIG_X86_IO_APIC
885{ 928 for (i = 0; i < mp_irq_entries; i++) {
886 struct mpc_config_intsrc intsrc; 929 if (irq_used[i])
887 int ioapic = -1; 930 continue;
888 int pin = -1;
889 931
890 /* 932 if (mp_irqs[i].mp_irqtype != mp_INT)
891 * Convert 'gsi' to 'ioapic.pin'. 933 continue;
892 */
893 ioapic = mp_find_ioapic(gsi);
894 if (ioapic < 0)
895 return;
896 pin = gsi - mp_ioapic_routing[ioapic].gsi_base;
897 934
898 /* 935 if (mp_irqs[i].mp_irqflag != 0x0f)
899 * TBD: This check is for faulty timer entries, where the override 936 continue;
900 * erroneously sets the trigger to level, resulting in a HUGE
901 * increase of timer interrupts!
902 */
903 if ((bus_irq == 0) && (trigger == 3))
904 trigger = 1;
905 937
906 intsrc.mpc_type = MP_INTSRC; 938 if (nr_m_spare > 0) {
907 intsrc.mpc_irqtype = mp_INT; 939 printk(KERN_INFO "*NEW* found ");
908 intsrc.mpc_irqflag = (trigger << 2) | polarity; 940 nr_m_spare--;
909 intsrc.mpc_srcbus = MP_ISA_BUS; 941 assign_to_mpc_intsrc(&mp_irqs[i], m_spare[nr_m_spare]);
910 intsrc.mpc_srcbusirq = bus_irq; /* IRQ */ 942 m_spare[nr_m_spare] = NULL;
911 intsrc.mpc_dstapic = mp_ioapics[ioapic].mpc_apicid; /* APIC ID */ 943 } else {
912 intsrc.mpc_dstirq = pin; /* INTIN# */ 944 struct mpc_config_intsrc *m =
945 (struct mpc_config_intsrc *)mpt;
946 count += sizeof(struct mpc_config_intsrc);
947 if (!mpc_new_phys) {
948 printk(KERN_INFO "No spare slots, try to append...take your risk, new mpc_length %x\n", count);
949 } else {
950 if (count <= mpc_new_length)
951 printk(KERN_INFO "No spare slots, try to append..., new mpc_length %x\n", count);
952 else {
953 printk(KERN_ERR "mpc_new_length %lx is too small\n", mpc_new_length);
954 goto out;
955 }
956 }
957 assign_to_mpc_intsrc(&mp_irqs[i], m);
958 mpc->mpc_length = count;
959 mpt += sizeof(struct mpc_config_intsrc);
960 }
961 print_mp_irq_info(&mp_irqs[i]);
962 }
963#endif
964out:
965 /* update checksum */
966 mpc->mpc_checksum = 0;
967 mpc->mpc_checksum -= mpf_checksum((unsigned char *)mpc,
968 mpc->mpc_length);
913 969
914 MP_intsrc_info(&intsrc); 970 return 0;
915} 971}
916 972
917void __init mp_config_acpi_legacy_irqs(void) 973static int __initdata enable_update_mptable;
918{
919 struct mpc_config_intsrc intsrc;
920 int i = 0;
921 int ioapic = -1;
922 974
923#if defined (CONFIG_MCA) || defined (CONFIG_EISA) 975static int __init update_mptable_setup(char *str)
924 /* 976{
925 * Fabricate the legacy ISA bus (bus #31). 977 enable_update_mptable = 1;
926 */ 978 return 0;
927 mp_bus_id_to_type[MP_ISA_BUS] = MP_BUS_ISA; 979}
928#endif 980early_param("update_mptable", update_mptable_setup);
929 set_bit(MP_ISA_BUS, mp_bus_not_pci);
930 Dprintk("Bus #%d is ISA\n", MP_ISA_BUS);
931 981
932 /* 982static unsigned long __initdata mpc_new_phys;
933 * Older generations of ES7000 have no legacy identity mappings 983static unsigned long mpc_new_length __initdata = 4096;
934 */
935 if (es7000_plat == 1)
936 return;
937 984
938 /* 985/* alloc_mptable or alloc_mptable=4k */
939 * Locate the IOAPIC that manages the ISA IRQs (0-15). 986static int __initdata alloc_mptable;
940 */ 987static int __init parse_alloc_mptable_opt(char *p)
941 ioapic = mp_find_ioapic(0); 988{
942 if (ioapic < 0) 989 enable_update_mptable = 1;
943 return; 990 alloc_mptable = 1;
991 if (!p)
992 return 0;
993 mpc_new_length = memparse(p, &p);
994 return 0;
995}
996early_param("alloc_mptable", parse_alloc_mptable_opt);
944 997
945 intsrc.mpc_type = MP_INTSRC; 998void __init early_reserve_e820_mpc_new(void)
946 intsrc.mpc_irqflag = 0; /* Conforming */ 999{
947 intsrc.mpc_srcbus = MP_ISA_BUS; 1000 if (enable_update_mptable && alloc_mptable) {
948#ifdef CONFIG_X86_IO_APIC 1001 u64 startt = 0;
949 intsrc.mpc_dstapic = mp_ioapics[ioapic].mpc_apicid; 1002#ifdef CONFIG_X86_TRAMPOLINE
1003 startt = TRAMPOLINE_BASE;
950#endif 1004#endif
951 /* 1005 mpc_new_phys = early_reserve_e820(startt, mpc_new_length, 4);
952 * Use the default configuration for the IRQs 0-15. Unless
953 * overridden by (MADT) interrupt source override entries.
954 */
955 for (i = 0; i < 16; i++) {
956 int idx;
957
958 for (idx = 0; idx < mp_irq_entries; idx++) {
959 struct mpc_config_intsrc *irq = mp_irqs + idx;
960
961 /* Do we already have a mapping for this ISA IRQ? */
962 if (irq->mpc_srcbus == MP_ISA_BUS
963 && irq->mpc_srcbusirq == i)
964 break;
965
966 /* Do we already have a mapping for this IOAPIC pin */
967 if ((irq->mpc_dstapic == intsrc.mpc_dstapic) &&
968 (irq->mpc_dstirq == i))
969 break;
970 }
971
972 if (idx != mp_irq_entries) {
973 printk(KERN_DEBUG "ACPI: IRQ%d used by override.\n", i);
974 continue; /* IRQ already used */
975 }
976
977 intsrc.mpc_irqtype = mp_INT;
978 intsrc.mpc_srcbusirq = i; /* Identity mapped */
979 intsrc.mpc_dstirq = i;
980
981 MP_intsrc_info(&intsrc);
982 } 1006 }
983} 1007}
984 1008
985int mp_register_gsi(u32 gsi, int triggering, int polarity) 1009static int __init update_mp_table(void)
986{ 1010{
987 int ioapic; 1011 char str[16];
988 int ioapic_pin; 1012 char oem[10];
989#ifdef CONFIG_X86_32 1013 struct intel_mp_floating *mpf;
990#define MAX_GSI_NUM 4096 1014 struct mp_config_table *mpc;
991#define IRQ_COMPRESSION_START 64 1015 struct mp_config_table *mpc_new;
1016
1017 if (!enable_update_mptable)
1018 return 0;
1019
1020 mpf = mpf_found;
1021 if (!mpf)
1022 return 0;
992 1023
993 static int pci_irq = IRQ_COMPRESSION_START;
994 /* 1024 /*
995 * Mapping between Global System Interrupts, which 1025 * Now see if we need to go further.
996 * represent all possible interrupts, and IRQs
997 * assigned to actual devices.
998 */ 1026 */
999 static int gsi_to_irq[MAX_GSI_NUM]; 1027 if (mpf->mpf_feature1 != 0)
1000#else 1028 return 0;
1001 1029
1002 if (acpi_irq_model != ACPI_IRQ_MODEL_IOAPIC) 1030 if (!mpf->mpf_physptr)
1003 return gsi; 1031 return 0;
1004#endif
1005 1032
1006 /* Don't set up the ACPI SCI because it's already set up */ 1033 mpc = phys_to_virt(mpf->mpf_physptr);
1007 if (acpi_gbl_FADT.sci_interrupt == gsi)
1008 return gsi;
1009 1034
1010 ioapic = mp_find_ioapic(gsi); 1035 if (!smp_check_mpc(mpc, oem, str))
1011 if (ioapic < 0) { 1036 return 0;
1012 printk(KERN_WARNING "No IOAPIC for GSI %u\n", gsi);
1013 return gsi;
1014 }
1015 1037
1016 ioapic_pin = gsi - mp_ioapic_routing[ioapic].gsi_base; 1038 printk(KERN_INFO "mpf: %lx\n", virt_to_phys(mpf));
1039 printk(KERN_INFO "mpf_physptr: %x\n", mpf->mpf_physptr);
1017 1040
1018#ifdef CONFIG_X86_32 1041 if (mpc_new_phys && mpc->mpc_length > mpc_new_length) {
1019 if (ioapic_renumber_irq) 1042 mpc_new_phys = 0;
1020 gsi = ioapic_renumber_irq(ioapic, gsi); 1043 printk(KERN_INFO "mpc_new_length is %ld, please use alloc_mptable=8k\n",
1021#endif 1044 mpc_new_length);
1022
1023 /*
1024 * Avoid pin reprogramming. PRTs typically include entries
1025 * with redundant pin->gsi mappings (but unique PCI devices);
1026 * we only program the IOAPIC on the first.
1027 */
1028 if (ioapic_pin > MP_MAX_IOAPIC_PIN) {
1029 printk(KERN_ERR "Invalid reference to IOAPIC pin "
1030 "%d-%d\n", mp_ioapic_routing[ioapic].apic_id,
1031 ioapic_pin);
1032 return gsi;
1033 } 1045 }
1034 if (test_bit(ioapic_pin, mp_ioapic_routing[ioapic].pin_programmed)) { 1046
1035 Dprintk(KERN_DEBUG "Pin %d-%d already programmed\n", 1047 if (!mpc_new_phys) {
1036 mp_ioapic_routing[ioapic].apic_id, ioapic_pin); 1048 unsigned char old, new;
1037#ifdef CONFIG_X86_32 1049 /* check if we can change the postion */
1038 return (gsi < IRQ_COMPRESSION_START ? gsi : gsi_to_irq[gsi]); 1050 mpc->mpc_checksum = 0;
1039#else 1051 old = mpf_checksum((unsigned char *)mpc, mpc->mpc_length);
1040 return gsi; 1052 mpc->mpc_checksum = 0xff;
1041#endif 1053 new = mpf_checksum((unsigned char *)mpc, mpc->mpc_length);
1054 if (old == new) {
1055 printk(KERN_INFO "mpc is readonly, please try alloc_mptable instead\n");
1056 return 0;
1057 }
1058 printk(KERN_INFO "use in-positon replacing\n");
1059 } else {
1060 mpf->mpf_physptr = mpc_new_phys;
1061 mpc_new = phys_to_virt(mpc_new_phys);
1062 memcpy(mpc_new, mpc, mpc->mpc_length);
1063 mpc = mpc_new;
1064 /* check if we can modify that */
1065 if (mpc_new_phys - mpf->mpf_physptr) {
1066 struct intel_mp_floating *mpf_new;
1067 /* steal 16 bytes from [0, 1k) */
1068 printk(KERN_INFO "mpf new: %x\n", 0x400 - 16);
1069 mpf_new = phys_to_virt(0x400 - 16);
1070 memcpy(mpf_new, mpf, 16);
1071 mpf = mpf_new;
1072 mpf->mpf_physptr = mpc_new_phys;
1073 }
1074 mpf->mpf_checksum = 0;
1075 mpf->mpf_checksum -= mpf_checksum((unsigned char *)mpf, 16);
1076 printk(KERN_INFO "mpf_physptr new: %x\n", mpf->mpf_physptr);
1042 } 1077 }
1043 1078
1044 set_bit(ioapic_pin, mp_ioapic_routing[ioapic].pin_programmed);
1045#ifdef CONFIG_X86_32
1046 /* 1079 /*
1047 * For GSI >= 64, use IRQ compression 1080 * only replace the one with mp_INT and
1081 * MP_IRQ_TRIGGER_LEVEL|MP_IRQ_POLARITY_LOW,
1082 * already in mp_irqs , stored by ... and mp_config_acpi_gsi,
1083 * may need pci=routeirq for all coverage
1048 */ 1084 */
1049 if ((gsi >= IRQ_COMPRESSION_START) 1085 replace_intsrc_all(mpc, mpc_new_phys, mpc_new_length);
1050 && (triggering == ACPI_LEVEL_SENSITIVE)) { 1086
1051 /* 1087 return 0;
1052 * For PCI devices assign IRQs in order, avoiding gaps
1053 * due to unused I/O APIC pins.
1054 */
1055 int irq = gsi;
1056 if (gsi < MAX_GSI_NUM) {
1057 /*
1058 * Retain the VIA chipset work-around (gsi > 15), but
1059 * avoid a problem where the 8254 timer (IRQ0) is setup
1060 * via an override (so it's not on pin 0 of the ioapic),
1061 * and at the same time, the pin 0 interrupt is a PCI
1062 * type. The gsi > 15 test could cause these two pins
1063 * to be shared as IRQ0, and they are not shareable.
1064 * So test for this condition, and if necessary, avoid
1065 * the pin collision.
1066 */
1067 gsi = pci_irq++;
1068 /*
1069 * Don't assign IRQ used by ACPI SCI
1070 */
1071 if (gsi == acpi_gbl_FADT.sci_interrupt)
1072 gsi = pci_irq++;
1073 gsi_to_irq[irq] = gsi;
1074 } else {
1075 printk(KERN_ERR "GSI %u is too high\n", gsi);
1076 return gsi;
1077 }
1078 }
1079#endif
1080 io_apic_set_pci_routing(ioapic, ioapic_pin, gsi,
1081 triggering == ACPI_EDGE_SENSITIVE ? 0 : 1,
1082 polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
1083 return gsi;
1084} 1088}
1085 1089
1086#endif /* CONFIG_X86_IO_APIC */ 1090late_initcall(update_mp_table);
1087#endif /* CONFIG_ACPI */
diff --git a/arch/x86/kernel/msr.c b/arch/x86/kernel/msr.c
index 1f3abe048e93..9fd809552447 100644
--- a/arch/x86/kernel/msr.c
+++ b/arch/x86/kernel/msr.c
@@ -117,12 +117,20 @@ static int msr_open(struct inode *inode, struct file *file)
117{ 117{
118 unsigned int cpu = iminor(file->f_path.dentry->d_inode); 118 unsigned int cpu = iminor(file->f_path.dentry->d_inode);
119 struct cpuinfo_x86 *c = &cpu_data(cpu); 119 struct cpuinfo_x86 *c = &cpu_data(cpu);
120 int ret = 0;
120 121
121 if (cpu >= NR_CPUS || !cpu_online(cpu)) 122 lock_kernel();
122 return -ENXIO; /* No such CPU */ 123 cpu = iminor(file->f_path.dentry->d_inode);
123 if (!cpu_has(c, X86_FEATURE_MSR))
124 return -EIO; /* MSR not supported */
125 124
125 if (cpu >= NR_CPUS || !cpu_online(cpu)) {
126 ret = -ENXIO; /* No such CPU */
127 goto out;
128 }
129 c = &cpu_data(cpu);
130 if (!cpu_has(c, X86_FEATURE_MSR))
131 ret = -EIO; /* MSR not supported */
132out:
133 unlock_kernel();
126 return 0; 134 return 0;
127} 135}
128 136
@@ -141,8 +149,8 @@ static int __cpuinit msr_device_create(int cpu)
141{ 149{
142 struct device *dev; 150 struct device *dev;
143 151
144 dev = device_create(msr_class, NULL, MKDEV(MSR_MAJOR, cpu), 152 dev = device_create_drvdata(msr_class, NULL, MKDEV(MSR_MAJOR, cpu),
145 "msr%d", cpu); 153 NULL, "msr%d", cpu);
146 return IS_ERR(dev) ? PTR_ERR(dev) : 0; 154 return IS_ERR(dev) ? PTR_ERR(dev) : 0;
147} 155}
148 156
diff --git a/arch/x86/kernel/nmi_64.c b/arch/x86/kernel/nmi.c
index 5a29ded994fa..ac6d51222e7d 100644
--- a/arch/x86/kernel/nmi_64.c
+++ b/arch/x86/kernel/nmi.c
@@ -6,10 +6,13 @@
6 * Fixes: 6 * Fixes:
7 * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog. 7 * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog.
8 * Mikael Pettersson : Power Management for local APIC NMI watchdog. 8 * Mikael Pettersson : Power Management for local APIC NMI watchdog.
9 * Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog.
9 * Pavel Machek and 10 * Pavel Machek and
10 * Mikael Pettersson : PM converted to driver model. Disable/enable API. 11 * Mikael Pettersson : PM converted to driver model. Disable/enable API.
11 */ 12 */
12 13
14#include <asm/apic.h>
15
13#include <linux/nmi.h> 16#include <linux/nmi.h>
14#include <linux/mm.h> 17#include <linux/mm.h>
15#include <linux/delay.h> 18#include <linux/delay.h>
@@ -17,20 +20,26 @@
17#include <linux/module.h> 20#include <linux/module.h>
18#include <linux/sysdev.h> 21#include <linux/sysdev.h>
19#include <linux/sysctl.h> 22#include <linux/sysctl.h>
23#include <linux/percpu.h>
20#include <linux/kprobes.h> 24#include <linux/kprobes.h>
21#include <linux/cpumask.h> 25#include <linux/cpumask.h>
26#include <linux/kernel_stat.h>
22#include <linux/kdebug.h> 27#include <linux/kdebug.h>
28#include <linux/smp.h>
23 29
30#include <asm/i8259.h>
31#include <asm/io_apic.h>
24#include <asm/smp.h> 32#include <asm/smp.h>
25#include <asm/nmi.h> 33#include <asm/nmi.h>
26#include <asm/proto.h> 34#include <asm/proto.h>
35#include <asm/timer.h>
36
27#include <asm/mce.h> 37#include <asm/mce.h>
28 38
29#include <mach_traps.h> 39#include <mach_traps.h>
30 40
31int unknown_nmi_panic; 41int unknown_nmi_panic;
32int nmi_watchdog_enabled; 42int nmi_watchdog_enabled;
33int panic_on_unrecovered_nmi;
34 43
35static cpumask_t backtrace_mask = CPU_MASK_NONE; 44static cpumask_t backtrace_mask = CPU_MASK_NONE;
36 45
@@ -41,37 +50,65 @@ static cpumask_t backtrace_mask = CPU_MASK_NONE;
41 * 0: the lapic NMI watchdog is disabled, but can be enabled 50 * 0: the lapic NMI watchdog is disabled, but can be enabled
42 */ 51 */
43atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */ 52atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */
53EXPORT_SYMBOL(nmi_active);
54
55unsigned int nmi_watchdog = NMI_NONE;
56EXPORT_SYMBOL(nmi_watchdog);
57
44static int panic_on_timeout; 58static int panic_on_timeout;
45 59
46unsigned int nmi_watchdog = NMI_DEFAULT;
47static unsigned int nmi_hz = HZ; 60static unsigned int nmi_hz = HZ;
48
49static DEFINE_PER_CPU(short, wd_enabled); 61static DEFINE_PER_CPU(short, wd_enabled);
62static int endflag __initdata;
50 63
51/* Run after command line and cpu_init init, but before all other checks */ 64static inline unsigned int get_nmi_count(int cpu)
52void nmi_watchdog_default(void)
53{ 65{
54 if (nmi_watchdog != NMI_DEFAULT) 66#ifdef CONFIG_X86_64
55 return; 67 return cpu_pda(cpu)->__nmi_count;
56 nmi_watchdog = NMI_NONE; 68#else
69 return nmi_count(cpu);
70#endif
71}
72
73static inline int mce_in_progress(void)
74{
75#if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE)
76 return atomic_read(&mce_entry) > 0;
77#endif
78 return 0;
57} 79}
58 80
59static int endflag __initdata = 0; 81/*
82 * Take the local apic timer and PIT/HPET into account. We don't
83 * know which one is active, when we have highres/dyntick on
84 */
85static inline unsigned int get_timer_irqs(int cpu)
86{
87#ifdef CONFIG_X86_64
88 return read_pda(apic_timer_irqs) + read_pda(irq0_irqs);
89#else
90 return per_cpu(irq_stat, cpu).apic_timer_irqs +
91 per_cpu(irq_stat, cpu).irq0_irqs;
92#endif
93}
60 94
61#ifdef CONFIG_SMP 95#ifdef CONFIG_SMP
62/* The performance counters used by NMI_LOCAL_APIC don't trigger when 96/*
97 * The performance counters used by NMI_LOCAL_APIC don't trigger when
63 * the CPU is idle. To make sure the NMI watchdog really ticks on all 98 * the CPU is idle. To make sure the NMI watchdog really ticks on all
64 * CPUs during the test make them busy. 99 * CPUs during the test make them busy.
65 */ 100 */
66static __init void nmi_cpu_busy(void *data) 101static __init void nmi_cpu_busy(void *data)
67{ 102{
68 local_irq_enable_in_hardirq(); 103 local_irq_enable_in_hardirq();
69 /* Intentionally don't use cpu_relax here. This is 104 /*
70 to make sure that the performance counter really ticks, 105 * Intentionally don't use cpu_relax here. This is
71 even if there is a simulator or similar that catches the 106 * to make sure that the performance counter really ticks,
72 pause instruction. On a real HT machine this is fine because 107 * even if there is a simulator or similar that catches the
73 all other CPUs are busy with "useless" delay loops and don't 108 * pause instruction. On a real HT machine this is fine because
74 care if they get somewhat less cycles. */ 109 * all other CPUs are busy with "useless" delay loops and don't
110 * care if they get somewhat less cycles.
111 */
75 while (endflag == 0) 112 while (endflag == 0)
76 mb(); 113 mb();
77} 114}
@@ -79,40 +116,37 @@ static __init void nmi_cpu_busy(void *data)
79 116
80int __init check_nmi_watchdog(void) 117int __init check_nmi_watchdog(void)
81{ 118{
82 int *prev_nmi_count; 119 unsigned int *prev_nmi_count;
83 int cpu; 120 int cpu;
84 121
85 if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DISABLED)) 122 if (!nmi_watchdog_active() || !atomic_read(&nmi_active))
86 return 0;
87
88 if (!atomic_read(&nmi_active))
89 return 0; 123 return 0;
90 124
91 prev_nmi_count = kmalloc(NR_CPUS * sizeof(int), GFP_KERNEL); 125 prev_nmi_count = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
92 if (!prev_nmi_count) 126 if (!prev_nmi_count)
93 return -1; 127 goto error;
94 128
95 printk(KERN_INFO "Testing NMI watchdog ... "); 129 printk(KERN_INFO "Testing NMI watchdog ... ");
96 130
97#ifdef CONFIG_SMP 131#ifdef CONFIG_SMP
98 if (nmi_watchdog == NMI_LOCAL_APIC) 132 if (nmi_watchdog == NMI_LOCAL_APIC)
99 smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0); 133 smp_call_function(nmi_cpu_busy, (void *)&endflag, 0);
100#endif 134#endif
101 135
102 for (cpu = 0; cpu < NR_CPUS; cpu++) 136 for_each_possible_cpu(cpu)
103 prev_nmi_count[cpu] = cpu_pda(cpu)->__nmi_count; 137 prev_nmi_count[cpu] = get_nmi_count(cpu);
104 local_irq_enable(); 138 local_irq_enable();
105 mdelay((20*1000)/nmi_hz); // wait 20 ticks 139 mdelay((20 * 1000) / nmi_hz); /* wait 20 ticks */
106 140
107 for_each_online_cpu(cpu) { 141 for_each_online_cpu(cpu) {
108 if (!per_cpu(wd_enabled, cpu)) 142 if (!per_cpu(wd_enabled, cpu))
109 continue; 143 continue;
110 if (cpu_pda(cpu)->__nmi_count - prev_nmi_count[cpu] <= 5) { 144 if (get_nmi_count(cpu) - prev_nmi_count[cpu] <= 5) {
111 printk(KERN_WARNING "WARNING: CPU#%d: NMI " 145 printk(KERN_WARNING "WARNING: CPU#%d: NMI "
112 "appears to be stuck (%d->%d)!\n", 146 "appears to be stuck (%d->%d)!\n",
113 cpu, 147 cpu,
114 prev_nmi_count[cpu], 148 prev_nmi_count[cpu],
115 cpu_pda(cpu)->__nmi_count); 149 get_nmi_count(cpu));
116 per_cpu(wd_enabled, cpu) = 0; 150 per_cpu(wd_enabled, cpu) = 0;
117 atomic_dec(&nmi_active); 151 atomic_dec(&nmi_active);
118 } 152 }
@@ -121,24 +155,33 @@ int __init check_nmi_watchdog(void)
121 if (!atomic_read(&nmi_active)) { 155 if (!atomic_read(&nmi_active)) {
122 kfree(prev_nmi_count); 156 kfree(prev_nmi_count);
123 atomic_set(&nmi_active, -1); 157 atomic_set(&nmi_active, -1);
124 return -1; 158 goto error;
125 } 159 }
126 printk("OK.\n"); 160 printk("OK.\n");
127 161
128 /* now that we know it works we can reduce NMI frequency to 162 /*
129 something more reasonable; makes a difference in some configs */ 163 * now that we know it works we can reduce NMI frequency to
164 * something more reasonable; makes a difference in some configs
165 */
130 if (nmi_watchdog == NMI_LOCAL_APIC) 166 if (nmi_watchdog == NMI_LOCAL_APIC)
131 nmi_hz = lapic_adjust_nmi_hz(1); 167 nmi_hz = lapic_adjust_nmi_hz(1);
132 168
133 kfree(prev_nmi_count); 169 kfree(prev_nmi_count);
134 return 0; 170 return 0;
171error:
172 if (nmi_watchdog == NMI_IO_APIC && !timer_through_8259)
173 disable_8259A_irq(0);
174#ifdef CONFIG_X86_32
175 timer_ack = 0;
176#endif
177 return -1;
135} 178}
136 179
137static int __init setup_nmi_watchdog(char *str) 180static int __init setup_nmi_watchdog(char *str)
138{ 181{
139 int nmi; 182 unsigned int nmi;
140 183
141 if (!strncmp(str,"panic",5)) { 184 if (!strncmp(str, "panic", 5)) {
142 panic_on_timeout = 1; 185 panic_on_timeout = 1;
143 str = strchr(str, ','); 186 str = strchr(str, ',');
144 if (!str) 187 if (!str)
@@ -148,15 +191,17 @@ static int __init setup_nmi_watchdog(char *str)
148 191
149 get_option(&str, &nmi); 192 get_option(&str, &nmi);
150 193
151 if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE)) 194 if (nmi >= NMI_INVALID)
152 return 0; 195 return 0;
153 196
154 nmi_watchdog = nmi; 197 nmi_watchdog = nmi;
155 return 1; 198 return 1;
156} 199}
157
158__setup("nmi_watchdog=", setup_nmi_watchdog); 200__setup("nmi_watchdog=", setup_nmi_watchdog);
159 201
202/*
203 * Suspend/resume support
204 */
160#ifdef CONFIG_PM 205#ifdef CONFIG_PM
161 206
162static int nmi_pm_active; /* nmi_active before suspend */ 207static int nmi_pm_active; /* nmi_active before suspend */
@@ -195,7 +240,8 @@ static int __init init_lapic_nmi_sysfs(void)
195{ 240{
196 int error; 241 int error;
197 242
198 /* should really be a BUG_ON but b/c this is an 243 /*
244 * should really be a BUG_ON but b/c this is an
199 * init call, it just doesn't work. -dcz 245 * init call, it just doesn't work. -dcz
200 */ 246 */
201 if (nmi_watchdog != NMI_LOCAL_APIC) 247 if (nmi_watchdog != NMI_LOCAL_APIC)
@@ -209,6 +255,7 @@ static int __init init_lapic_nmi_sysfs(void)
209 error = sysdev_register(&device_lapic_nmi); 255 error = sysdev_register(&device_lapic_nmi);
210 return error; 256 return error;
211} 257}
258
212/* must come after the local APIC's device_initcall() */ 259/* must come after the local APIC's device_initcall() */
213late_initcall(init_lapic_nmi_sysfs); 260late_initcall(init_lapic_nmi_sysfs);
214 261
@@ -225,7 +272,7 @@ static void __acpi_nmi_enable(void *__unused)
225void acpi_nmi_enable(void) 272void acpi_nmi_enable(void)
226{ 273{
227 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) 274 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
228 on_each_cpu(__acpi_nmi_enable, NULL, 0, 1); 275 on_each_cpu(__acpi_nmi_enable, NULL, 1);
229} 276}
230 277
231static void __acpi_nmi_disable(void *__unused) 278static void __acpi_nmi_disable(void *__unused)
@@ -239,7 +286,7 @@ static void __acpi_nmi_disable(void *__unused)
239void acpi_nmi_disable(void) 286void acpi_nmi_disable(void)
240{ 287{
241 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC) 288 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
242 on_each_cpu(__acpi_nmi_disable, NULL, 0, 1); 289 on_each_cpu(__acpi_nmi_disable, NULL, 1);
243} 290}
244 291
245void setup_apic_nmi_watchdog(void *unused) 292void setup_apic_nmi_watchdog(void *unused)
@@ -249,11 +296,12 @@ void setup_apic_nmi_watchdog(void *unused)
249 296
250 /* cheap hack to support suspend/resume */ 297 /* cheap hack to support suspend/resume */
251 /* if cpu0 is not active neither should the other cpus */ 298 /* if cpu0 is not active neither should the other cpus */
252 if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0)) 299 if (smp_processor_id() != 0 && atomic_read(&nmi_active) <= 0)
253 return; 300 return;
254 301
255 switch (nmi_watchdog) { 302 switch (nmi_watchdog) {
256 case NMI_LOCAL_APIC: 303 case NMI_LOCAL_APIC:
304 /* enable it before to avoid race with handler */
257 __get_cpu_var(wd_enabled) = 1; 305 __get_cpu_var(wd_enabled) = 1;
258 if (lapic_watchdog_init(nmi_hz) < 0) { 306 if (lapic_watchdog_init(nmi_hz) < 0) {
259 __get_cpu_var(wd_enabled) = 0; 307 __get_cpu_var(wd_enabled) = 0;
@@ -269,9 +317,8 @@ void setup_apic_nmi_watchdog(void *unused)
269void stop_apic_nmi_watchdog(void *unused) 317void stop_apic_nmi_watchdog(void *unused)
270{ 318{
271 /* only support LOCAL and IO APICs for now */ 319 /* only support LOCAL and IO APICs for now */
272 if ((nmi_watchdog != NMI_LOCAL_APIC) && 320 if (!nmi_watchdog_active())
273 (nmi_watchdog != NMI_IO_APIC)) 321 return;
274 return;
275 if (__get_cpu_var(wd_enabled) == 0) 322 if (__get_cpu_var(wd_enabled) == 0)
276 return; 323 return;
277 if (nmi_watchdog == NMI_LOCAL_APIC) 324 if (nmi_watchdog == NMI_LOCAL_APIC)
@@ -287,6 +334,11 @@ void stop_apic_nmi_watchdog(void *unused)
287 * 334 *
288 * as these watchdog NMI IRQs are generated on every CPU, we only 335 * as these watchdog NMI IRQs are generated on every CPU, we only
289 * have to check the current processor. 336 * have to check the current processor.
337 *
338 * since NMIs don't listen to _any_ locks, we have to be extremely
339 * careful not to rely on unsafe variables. The printk might lock
340 * up though, so we have to break up any console locks first ...
341 * [when there will be more tty-related locks, break them up here too!]
290 */ 342 */
291 343
292static DEFINE_PER_CPU(unsigned, last_irq_sum); 344static DEFINE_PER_CPU(unsigned, last_irq_sum);
@@ -295,11 +347,11 @@ static DEFINE_PER_CPU(int, nmi_touch);
295 347
296void touch_nmi_watchdog(void) 348void touch_nmi_watchdog(void)
297{ 349{
298 if (nmi_watchdog > 0) { 350 if (nmi_watchdog_active()) {
299 unsigned cpu; 351 unsigned cpu;
300 352
301 /* 353 /*
302 * Tell other CPUs to reset their alert counters. We cannot 354 * Tell other CPUs to reset their alert counters. We cannot
303 * do it ourselves because the alert count increase is not 355 * do it ourselves because the alert count increase is not
304 * atomic. 356 * atomic.
305 */ 357 */
@@ -309,6 +361,9 @@ void touch_nmi_watchdog(void)
309 } 361 }
310 } 362 }
311 363
364 /*
365 * Tickle the softlockup detector too:
366 */
312 touch_softlockup_watchdog(); 367 touch_softlockup_watchdog();
313} 368}
314EXPORT_SYMBOL(touch_nmi_watchdog); 369EXPORT_SYMBOL(touch_nmi_watchdog);
@@ -316,7 +371,12 @@ EXPORT_SYMBOL(touch_nmi_watchdog);
316notrace __kprobes int 371notrace __kprobes int
317nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) 372nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
318{ 373{
319 int sum; 374 /*
375 * Since current_thread_info()-> is always on the stack, and we
376 * always switch the stack NMI-atomically, it's safe to use
377 * smp_processor_id().
378 */
379 unsigned int sum;
320 int touched = 0; 380 int touched = 0;
321 int cpu = smp_processor_id(); 381 int cpu = smp_processor_id();
322 int rc = 0; 382 int rc = 0;
@@ -328,7 +388,8 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
328 touched = 1; 388 touched = 1;
329 } 389 }
330 390
331 sum = read_pda(apic_timer_irqs) + read_pda(irq0_irqs); 391 sum = get_timer_irqs(cpu);
392
332 if (__get_cpu_var(nmi_touch)) { 393 if (__get_cpu_var(nmi_touch)) {
333 __get_cpu_var(nmi_touch) = 0; 394 __get_cpu_var(nmi_touch) = 0;
334 touched = 1; 395 touched = 1;
@@ -338,28 +399,29 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
338 static DEFINE_SPINLOCK(lock); /* Serialise the printks */ 399 static DEFINE_SPINLOCK(lock); /* Serialise the printks */
339 400
340 spin_lock(&lock); 401 spin_lock(&lock);
341 printk("NMI backtrace for cpu %d\n", cpu); 402 printk(KERN_WARNING "NMI backtrace for cpu %d\n", cpu);
342 dump_stack(); 403 dump_stack();
343 spin_unlock(&lock); 404 spin_unlock(&lock);
344 cpu_clear(cpu, backtrace_mask); 405 cpu_clear(cpu, backtrace_mask);
345 } 406 }
346 407
347#ifdef CONFIG_X86_MCE 408 /* Could check oops_in_progress here too, but it's safer not to */
348 /* Could check oops_in_progress here too, but it's safer 409 if (mce_in_progress())
349 not too */
350 if (atomic_read(&mce_entry) > 0)
351 touched = 1; 410 touched = 1;
352#endif 411
353 /* if the apic timer isn't firing, this cpu isn't doing much */ 412 /* if the none of the timers isn't firing, this cpu isn't doing much */
354 if (!touched && __get_cpu_var(last_irq_sum) == sum) { 413 if (!touched && __get_cpu_var(last_irq_sum) == sum) {
355 /* 414 /*
356 * Ayiee, looks like this CPU is stuck ... 415 * Ayiee, looks like this CPU is stuck ...
357 * wait a few IRQs (5 seconds) before doing the oops ... 416 * wait a few IRQs (5 seconds) before doing the oops ...
358 */ 417 */
359 local_inc(&__get_cpu_var(alert_counter)); 418 local_inc(&__get_cpu_var(alert_counter));
360 if (local_read(&__get_cpu_var(alert_counter)) == 5*nmi_hz) 419 if (local_read(&__get_cpu_var(alert_counter)) == 5 * nmi_hz)
361 die_nmi("NMI Watchdog detected LOCKUP on CPU %d\n", regs, 420 /*
362 panic_on_timeout); 421 * die_nmi will return ONLY if NOTIFY_STOP happens..
422 */
423 die_nmi("BUG: NMI Watchdog detected LOCKUP",
424 regs, panic_on_timeout);
363 } else { 425 } else {
364 __get_cpu_var(last_irq_sum) = sum; 426 __get_cpu_var(last_irq_sum) = sum;
365 local_set(&__get_cpu_var(alert_counter), 0); 427 local_set(&__get_cpu_var(alert_counter), 0);
@@ -373,7 +435,8 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
373 rc |= lapic_wd_event(nmi_hz); 435 rc |= lapic_wd_event(nmi_hz);
374 break; 436 break;
375 case NMI_IO_APIC: 437 case NMI_IO_APIC:
376 /* don't know how to accurately check for this. 438 /*
439 * don't know how to accurately check for this.
377 * just assume it was a watchdog timer interrupt 440 * just assume it was a watchdog timer interrupt
378 * This matches the old behaviour. 441 * This matches the old behaviour.
379 */ 442 */
@@ -383,31 +446,14 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
383 return rc; 446 return rc;
384} 447}
385 448
386static unsigned ignore_nmis; 449#ifdef CONFIG_SYSCTL
387
388asmlinkage notrace __kprobes void
389do_nmi(struct pt_regs *regs, long error_code)
390{
391 nmi_enter();
392 add_pda(__nmi_count,1);
393 if (!ignore_nmis)
394 default_do_nmi(regs);
395 nmi_exit();
396}
397
398void stop_nmi(void)
399{
400 acpi_nmi_disable();
401 ignore_nmis++;
402}
403 450
404void restart_nmi(void) 451static int __init setup_unknown_nmi_panic(char *str)
405{ 452{
406 ignore_nmis--; 453 unknown_nmi_panic = 1;
407 acpi_nmi_enable(); 454 return 1;
408} 455}
409 456__setup("unknown_nmi_panic", setup_unknown_nmi_panic);
410#ifdef CONFIG_SYSCTL
411 457
412static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu) 458static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu)
413{ 459{
@@ -415,7 +461,7 @@ static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu)
415 char buf[64]; 461 char buf[64];
416 462
417 sprintf(buf, "NMI received for unknown reason %02x\n", reason); 463 sprintf(buf, "NMI received for unknown reason %02x\n", reason);
418 die_nmi(buf, regs, 1); /* Always panic here */ 464 die_nmi(buf, regs, 1); /* Always panic here */
419 return 0; 465 return 0;
420} 466}
421 467
@@ -433,28 +479,26 @@ int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file,
433 if (!!old_state == !!nmi_watchdog_enabled) 479 if (!!old_state == !!nmi_watchdog_enabled)
434 return 0; 480 return 0;
435 481
436 if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) { 482 if (atomic_read(&nmi_active) < 0 || !nmi_watchdog_active()) {
437 printk( KERN_WARNING "NMI watchdog is permanently disabled\n"); 483 printk(KERN_WARNING
484 "NMI watchdog is permanently disabled\n");
438 return -EIO; 485 return -EIO;
439 } 486 }
440 487
441 /* if nmi_watchdog is not set yet, then set it */
442 nmi_watchdog_default();
443
444 if (nmi_watchdog == NMI_LOCAL_APIC) { 488 if (nmi_watchdog == NMI_LOCAL_APIC) {
445 if (nmi_watchdog_enabled) 489 if (nmi_watchdog_enabled)
446 enable_lapic_nmi_watchdog(); 490 enable_lapic_nmi_watchdog();
447 else 491 else
448 disable_lapic_nmi_watchdog(); 492 disable_lapic_nmi_watchdog();
449 } else { 493 } else {
450 printk( KERN_WARNING 494 printk(KERN_WARNING
451 "NMI watchdog doesn't know what hardware to touch\n"); 495 "NMI watchdog doesn't know what hardware to touch\n");
452 return -EIO; 496 return -EIO;
453 } 497 }
454 return 0; 498 return 0;
455} 499}
456 500
457#endif 501#endif /* CONFIG_SYSCTL */
458 502
459int do_nmi_callback(struct pt_regs *regs, int cpu) 503int do_nmi_callback(struct pt_regs *regs, int cpu)
460{ 504{
@@ -477,6 +521,3 @@ void __trigger_all_cpu_backtrace(void)
477 mdelay(1); 521 mdelay(1);
478 } 522 }
479} 523}
480
481EXPORT_SYMBOL(nmi_active);
482EXPORT_SYMBOL(nmi_watchdog);
diff --git a/arch/x86/kernel/nmi_32.c b/arch/x86/kernel/nmi_32.c
deleted file mode 100644
index 84160f74eeb0..000000000000
--- a/arch/x86/kernel/nmi_32.c
+++ /dev/null
@@ -1,467 +0,0 @@
1/*
2 * NMI watchdog support on APIC systems
3 *
4 * Started by Ingo Molnar <mingo@redhat.com>
5 *
6 * Fixes:
7 * Mikael Pettersson : AMD K7 support for local APIC NMI watchdog.
8 * Mikael Pettersson : Power Management for local APIC NMI watchdog.
9 * Mikael Pettersson : Pentium 4 support for local APIC NMI watchdog.
10 * Pavel Machek and
11 * Mikael Pettersson : PM converted to driver model. Disable/enable API.
12 */
13
14#include <linux/delay.h>
15#include <linux/interrupt.h>
16#include <linux/module.h>
17#include <linux/nmi.h>
18#include <linux/sysdev.h>
19#include <linux/sysctl.h>
20#include <linux/percpu.h>
21#include <linux/kprobes.h>
22#include <linux/cpumask.h>
23#include <linux/kernel_stat.h>
24#include <linux/kdebug.h>
25#include <linux/slab.h>
26
27#include <asm/smp.h>
28#include <asm/nmi.h>
29
30#include "mach_traps.h"
31
32int unknown_nmi_panic;
33int nmi_watchdog_enabled;
34
35static cpumask_t backtrace_mask = CPU_MASK_NONE;
36
37/* nmi_active:
38 * >0: the lapic NMI watchdog is active, but can be disabled
39 * <0: the lapic NMI watchdog has not been set up, and cannot
40 * be enabled
41 * 0: the lapic NMI watchdog is disabled, but can be enabled
42 */
43atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */
44
45unsigned int nmi_watchdog = NMI_DEFAULT;
46static unsigned int nmi_hz = HZ;
47
48static DEFINE_PER_CPU(short, wd_enabled);
49
50static int endflag __initdata = 0;
51
52#ifdef CONFIG_SMP
53/* The performance counters used by NMI_LOCAL_APIC don't trigger when
54 * the CPU is idle. To make sure the NMI watchdog really ticks on all
55 * CPUs during the test make them busy.
56 */
57static __init void nmi_cpu_busy(void *data)
58{
59 local_irq_enable_in_hardirq();
60 /* Intentionally don't use cpu_relax here. This is
61 to make sure that the performance counter really ticks,
62 even if there is a simulator or similar that catches the
63 pause instruction. On a real HT machine this is fine because
64 all other CPUs are busy with "useless" delay loops and don't
65 care if they get somewhat less cycles. */
66 while (endflag == 0)
67 mb();
68}
69#endif
70
71int __init check_nmi_watchdog(void)
72{
73 unsigned int *prev_nmi_count;
74 int cpu;
75
76 if ((nmi_watchdog == NMI_NONE) || (nmi_watchdog == NMI_DISABLED))
77 return 0;
78
79 if (!atomic_read(&nmi_active))
80 return 0;
81
82 prev_nmi_count = kmalloc(NR_CPUS * sizeof(int), GFP_KERNEL);
83 if (!prev_nmi_count)
84 return -1;
85
86 printk(KERN_INFO "Testing NMI watchdog ... ");
87
88#ifdef CONFIG_SMP
89 if (nmi_watchdog == NMI_LOCAL_APIC)
90 smp_call_function(nmi_cpu_busy, (void *)&endflag, 0, 0);
91#endif
92
93 for_each_possible_cpu(cpu)
94 prev_nmi_count[cpu] = nmi_count(cpu);
95 local_irq_enable();
96 mdelay((20*1000)/nmi_hz); // wait 20 ticks
97
98 for_each_possible_cpu(cpu) {
99#ifdef CONFIG_SMP
100 /* Check cpu_callin_map here because that is set
101 after the timer is started. */
102 if (!cpu_isset(cpu, cpu_callin_map))
103 continue;
104#endif
105 if (!per_cpu(wd_enabled, cpu))
106 continue;
107 if (nmi_count(cpu) - prev_nmi_count[cpu] <= 5) {
108 printk(KERN_WARNING "WARNING: CPU#%d: NMI "
109 "appears to be stuck (%d->%d)!\n",
110 cpu,
111 prev_nmi_count[cpu],
112 nmi_count(cpu));
113 per_cpu(wd_enabled, cpu) = 0;
114 atomic_dec(&nmi_active);
115 }
116 }
117 endflag = 1;
118 if (!atomic_read(&nmi_active)) {
119 kfree(prev_nmi_count);
120 atomic_set(&nmi_active, -1);
121 return -1;
122 }
123 printk("OK.\n");
124
125 /* now that we know it works we can reduce NMI frequency to
126 something more reasonable; makes a difference in some configs */
127 if (nmi_watchdog == NMI_LOCAL_APIC)
128 nmi_hz = lapic_adjust_nmi_hz(1);
129
130 kfree(prev_nmi_count);
131 return 0;
132}
133
134static int __init setup_nmi_watchdog(char *str)
135{
136 int nmi;
137
138 get_option(&str, &nmi);
139
140 if ((nmi >= NMI_INVALID) || (nmi < NMI_NONE))
141 return 0;
142
143 nmi_watchdog = nmi;
144 return 1;
145}
146
147__setup("nmi_watchdog=", setup_nmi_watchdog);
148
149
150/* Suspend/resume support */
151
152#ifdef CONFIG_PM
153
154static int nmi_pm_active; /* nmi_active before suspend */
155
156static int lapic_nmi_suspend(struct sys_device *dev, pm_message_t state)
157{
158 /* only CPU0 goes here, other CPUs should be offline */
159 nmi_pm_active = atomic_read(&nmi_active);
160 stop_apic_nmi_watchdog(NULL);
161 BUG_ON(atomic_read(&nmi_active) != 0);
162 return 0;
163}
164
165static int lapic_nmi_resume(struct sys_device *dev)
166{
167 /* only CPU0 goes here, other CPUs should be offline */
168 if (nmi_pm_active > 0) {
169 setup_apic_nmi_watchdog(NULL);
170 touch_nmi_watchdog();
171 }
172 return 0;
173}
174
175
176static struct sysdev_class nmi_sysclass = {
177 .name = "lapic_nmi",
178 .resume = lapic_nmi_resume,
179 .suspend = lapic_nmi_suspend,
180};
181
182static struct sys_device device_lapic_nmi = {
183 .id = 0,
184 .cls = &nmi_sysclass,
185};
186
187static int __init init_lapic_nmi_sysfs(void)
188{
189 int error;
190
191 /* should really be a BUG_ON but b/c this is an
192 * init call, it just doesn't work. -dcz
193 */
194 if (nmi_watchdog != NMI_LOCAL_APIC)
195 return 0;
196
197 if (atomic_read(&nmi_active) < 0)
198 return 0;
199
200 error = sysdev_class_register(&nmi_sysclass);
201 if (!error)
202 error = sysdev_register(&device_lapic_nmi);
203 return error;
204}
205/* must come after the local APIC's device_initcall() */
206late_initcall(init_lapic_nmi_sysfs);
207
208#endif /* CONFIG_PM */
209
210static void __acpi_nmi_enable(void *__unused)
211{
212 apic_write_around(APIC_LVT0, APIC_DM_NMI);
213}
214
215/*
216 * Enable timer based NMIs on all CPUs:
217 */
218void acpi_nmi_enable(void)
219{
220 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
221 on_each_cpu(__acpi_nmi_enable, NULL, 0, 1);
222}
223
224static void __acpi_nmi_disable(void *__unused)
225{
226 apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED);
227}
228
229/*
230 * Disable timer based NMIs on all CPUs:
231 */
232void acpi_nmi_disable(void)
233{
234 if (atomic_read(&nmi_active) && nmi_watchdog == NMI_IO_APIC)
235 on_each_cpu(__acpi_nmi_disable, NULL, 0, 1);
236}
237
238void setup_apic_nmi_watchdog(void *unused)
239{
240 if (__get_cpu_var(wd_enabled))
241 return;
242
243 /* cheap hack to support suspend/resume */
244 /* if cpu0 is not active neither should the other cpus */
245 if ((smp_processor_id() != 0) && (atomic_read(&nmi_active) <= 0))
246 return;
247
248 switch (nmi_watchdog) {
249 case NMI_LOCAL_APIC:
250 __get_cpu_var(wd_enabled) = 1; /* enable it before to avoid race with handler */
251 if (lapic_watchdog_init(nmi_hz) < 0) {
252 __get_cpu_var(wd_enabled) = 0;
253 return;
254 }
255 /* FALL THROUGH */
256 case NMI_IO_APIC:
257 __get_cpu_var(wd_enabled) = 1;
258 atomic_inc(&nmi_active);
259 }
260}
261
262void stop_apic_nmi_watchdog(void *unused)
263{
264 /* only support LOCAL and IO APICs for now */
265 if ((nmi_watchdog != NMI_LOCAL_APIC) &&
266 (nmi_watchdog != NMI_IO_APIC))
267 return;
268 if (__get_cpu_var(wd_enabled) == 0)
269 return;
270 if (nmi_watchdog == NMI_LOCAL_APIC)
271 lapic_watchdog_stop();
272 __get_cpu_var(wd_enabled) = 0;
273 atomic_dec(&nmi_active);
274}
275
276/*
277 * the best way to detect whether a CPU has a 'hard lockup' problem
278 * is to check it's local APIC timer IRQ counts. If they are not
279 * changing then that CPU has some problem.
280 *
281 * as these watchdog NMI IRQs are generated on every CPU, we only
282 * have to check the current processor.
283 *
284 * since NMIs don't listen to _any_ locks, we have to be extremely
285 * careful not to rely on unsafe variables. The printk might lock
286 * up though, so we have to break up any console locks first ...
287 * [when there will be more tty-related locks, break them up
288 * here too!]
289 */
290
291static unsigned int
292 last_irq_sums [NR_CPUS],
293 alert_counter [NR_CPUS];
294
295void touch_nmi_watchdog(void)
296{
297 if (nmi_watchdog > 0) {
298 unsigned cpu;
299
300 /*
301 * Just reset the alert counters, (other CPUs might be
302 * spinning on locks we hold):
303 */
304 for_each_present_cpu(cpu) {
305 if (alert_counter[cpu])
306 alert_counter[cpu] = 0;
307 }
308 }
309
310 /*
311 * Tickle the softlockup detector too:
312 */
313 touch_softlockup_watchdog();
314}
315EXPORT_SYMBOL(touch_nmi_watchdog);
316
317extern void die_nmi(struct pt_regs *, const char *msg);
318
319notrace __kprobes int
320nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
321{
322
323 /*
324 * Since current_thread_info()-> is always on the stack, and we
325 * always switch the stack NMI-atomically, it's safe to use
326 * smp_processor_id().
327 */
328 unsigned int sum;
329 int touched = 0;
330 int cpu = smp_processor_id();
331 int rc = 0;
332
333 /* check for other users first */
334 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
335 == NOTIFY_STOP) {
336 rc = 1;
337 touched = 1;
338 }
339
340 if (cpu_isset(cpu, backtrace_mask)) {
341 static DEFINE_SPINLOCK(lock); /* Serialise the printks */
342
343 spin_lock(&lock);
344 printk("NMI backtrace for cpu %d\n", cpu);
345 dump_stack();
346 spin_unlock(&lock);
347 cpu_clear(cpu, backtrace_mask);
348 }
349
350 /*
351 * Take the local apic timer and PIT/HPET into account. We don't
352 * know which one is active, when we have highres/dyntick on
353 */
354 sum = per_cpu(irq_stat, cpu).apic_timer_irqs +
355 per_cpu(irq_stat, cpu).irq0_irqs;
356
357 /* if the none of the timers isn't firing, this cpu isn't doing much */
358 if (!touched && last_irq_sums[cpu] == sum) {
359 /*
360 * Ayiee, looks like this CPU is stuck ...
361 * wait a few IRQs (5 seconds) before doing the oops ...
362 */
363 alert_counter[cpu]++;
364 if (alert_counter[cpu] == 5*nmi_hz)
365 /*
366 * die_nmi will return ONLY if NOTIFY_STOP happens..
367 */
368 die_nmi(regs, "BUG: NMI Watchdog detected LOCKUP");
369 } else {
370 last_irq_sums[cpu] = sum;
371 alert_counter[cpu] = 0;
372 }
373 /* see if the nmi watchdog went off */
374 if (!__get_cpu_var(wd_enabled))
375 return rc;
376 switch (nmi_watchdog) {
377 case NMI_LOCAL_APIC:
378 rc |= lapic_wd_event(nmi_hz);
379 break;
380 case NMI_IO_APIC:
381 /* don't know how to accurately check for this.
382 * just assume it was a watchdog timer interrupt
383 * This matches the old behaviour.
384 */
385 rc = 1;
386 break;
387 }
388 return rc;
389}
390
391#ifdef CONFIG_SYSCTL
392
393static int unknown_nmi_panic_callback(struct pt_regs *regs, int cpu)
394{
395 unsigned char reason = get_nmi_reason();
396 char buf[64];
397
398 sprintf(buf, "NMI received for unknown reason %02x\n", reason);
399 die_nmi(regs, buf);
400 return 0;
401}
402
403/*
404 * proc handler for /proc/sys/kernel/nmi
405 */
406int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file,
407 void __user *buffer, size_t *length, loff_t *ppos)
408{
409 int old_state;
410
411 nmi_watchdog_enabled = (atomic_read(&nmi_active) > 0) ? 1 : 0;
412 old_state = nmi_watchdog_enabled;
413 proc_dointvec(table, write, file, buffer, length, ppos);
414 if (!!old_state == !!nmi_watchdog_enabled)
415 return 0;
416
417 if (atomic_read(&nmi_active) < 0 || nmi_watchdog == NMI_DISABLED) {
418 printk( KERN_WARNING "NMI watchdog is permanently disabled\n");
419 return -EIO;
420 }
421
422 if (nmi_watchdog == NMI_DEFAULT) {
423 if (lapic_watchdog_ok())
424 nmi_watchdog = NMI_LOCAL_APIC;
425 else
426 nmi_watchdog = NMI_IO_APIC;
427 }
428
429 if (nmi_watchdog == NMI_LOCAL_APIC) {
430 if (nmi_watchdog_enabled)
431 enable_lapic_nmi_watchdog();
432 else
433 disable_lapic_nmi_watchdog();
434 } else {
435 printk( KERN_WARNING
436 "NMI watchdog doesn't know what hardware to touch\n");
437 return -EIO;
438 }
439 return 0;
440}
441
442#endif
443
444int do_nmi_callback(struct pt_regs *regs, int cpu)
445{
446#ifdef CONFIG_SYSCTL
447 if (unknown_nmi_panic)
448 return unknown_nmi_panic_callback(regs, cpu);
449#endif
450 return 0;
451}
452
453void __trigger_all_cpu_backtrace(void)
454{
455 int i;
456
457 backtrace_mask = cpu_online_map;
458 /* Wait for up to 10 seconds for all CPUs to do the backtrace */
459 for (i = 0; i < 10 * 1000; i++) {
460 if (cpus_empty(backtrace_mask))
461 break;
462 mdelay(1);
463 }
464}
465
466EXPORT_SYMBOL(nmi_active);
467EXPORT_SYMBOL(nmi_watchdog);
diff --git a/arch/x86/kernel/numaq_32.c b/arch/x86/kernel/numaq_32.c
index e65281b1634b..b8c45610b20a 100644
--- a/arch/x86/kernel/numaq_32.c
+++ b/arch/x86/kernel/numaq_32.c
@@ -31,6 +31,9 @@
31#include <asm/numaq.h> 31#include <asm/numaq.h>
32#include <asm/topology.h> 32#include <asm/topology.h>
33#include <asm/processor.h> 33#include <asm/processor.h>
34#include <asm/mpspec.h>
35#include <asm/e820.h>
36#include <asm/setup.h>
34 37
35#define MB_TO_PAGES(addr) ((addr) << (20 - PAGE_SHIFT)) 38#define MB_TO_PAGES(addr) ((addr) << (20 - PAGE_SHIFT))
36 39
@@ -58,6 +61,8 @@ static void __init smp_dump_qct(void)
58 node_end_pfn[node] = MB_TO_PAGES( 61 node_end_pfn[node] = MB_TO_PAGES(
59 eq->hi_shrd_mem_start + eq->hi_shrd_mem_size); 62 eq->hi_shrd_mem_start + eq->hi_shrd_mem_size);
60 63
64 e820_register_active_regions(node, node_start_pfn[node],
65 node_end_pfn[node]);
61 memory_present(node, 66 memory_present(node,
62 node_start_pfn[node], node_end_pfn[node]); 67 node_start_pfn[node], node_end_pfn[node]);
63 node_remap_size[node] = node_memmap_size_bytes(node, 68 node_remap_size[node] = node_memmap_size_bytes(node,
@@ -67,23 +72,209 @@ static void __init smp_dump_qct(void)
67 } 72 }
68} 73}
69 74
70/*
71 * Unlike Summit, we don't really care to let the NUMA-Q
72 * fall back to flat mode. Don't compile for NUMA-Q
73 * unless you really need it!
74 */
75int __init get_memcfg_numaq(void)
76{
77 smp_dump_qct();
78 return 1;
79}
80 75
81static int __init numaq_tsc_disable(void) 76void __init numaq_tsc_disable(void)
82{ 77{
78 if (!found_numaq)
79 return;
80
83 if (num_online_nodes() > 1) { 81 if (num_online_nodes() > 1) {
84 printk(KERN_DEBUG "NUMAQ: disabling TSC\n"); 82 printk(KERN_DEBUG "NUMAQ: disabling TSC\n");
85 setup_clear_cpu_cap(X86_FEATURE_TSC); 83 setup_clear_cpu_cap(X86_FEATURE_TSC);
86 } 84 }
85}
86
87static int __init numaq_pre_time_init(void)
88{
89 numaq_tsc_disable();
87 return 0; 90 return 0;
88} 91}
89arch_initcall(numaq_tsc_disable); 92
93int found_numaq;
94/*
95 * Have to match translation table entries to main table entries by counter
96 * hence the mpc_record variable .... can't see a less disgusting way of
97 * doing this ....
98 */
99struct mpc_config_translation {
100 unsigned char mpc_type;
101 unsigned char trans_len;
102 unsigned char trans_type;
103 unsigned char trans_quad;
104 unsigned char trans_global;
105 unsigned char trans_local;
106 unsigned short trans_reserved;
107};
108
109/* x86_quirks member */
110static int mpc_record;
111static struct mpc_config_translation *translation_table[MAX_MPC_ENTRY]
112 __cpuinitdata;
113
114static inline int generate_logical_apicid(int quad, int phys_apicid)
115{
116 return (quad << 4) + (phys_apicid ? phys_apicid << 1 : 1);
117}
118
119/* x86_quirks member */
120static int mpc_apic_id(struct mpc_config_processor *m)
121{
122 int quad = translation_table[mpc_record]->trans_quad;
123 int logical_apicid = generate_logical_apicid(quad, m->mpc_apicid);
124
125 printk(KERN_DEBUG "Processor #%d %u:%u APIC version %d (quad %d, apic %d)\n",
126 m->mpc_apicid,
127 (m->mpc_cpufeature & CPU_FAMILY_MASK) >> 8,
128 (m->mpc_cpufeature & CPU_MODEL_MASK) >> 4,
129 m->mpc_apicver, quad, logical_apicid);
130 return logical_apicid;
131}
132
133int mp_bus_id_to_node[MAX_MP_BUSSES];
134
135int mp_bus_id_to_local[MAX_MP_BUSSES];
136
137/* x86_quirks member */
138static void mpc_oem_bus_info(struct mpc_config_bus *m, char *name)
139{
140 int quad = translation_table[mpc_record]->trans_quad;
141 int local = translation_table[mpc_record]->trans_local;
142
143 mp_bus_id_to_node[m->mpc_busid] = quad;
144 mp_bus_id_to_local[m->mpc_busid] = local;
145 printk(KERN_INFO "Bus #%d is %s (node %d)\n",
146 m->mpc_busid, name, quad);
147}
148
149int quad_local_to_mp_bus_id [NR_CPUS/4][4];
150
151/* x86_quirks member */
152static void mpc_oem_pci_bus(struct mpc_config_bus *m)
153{
154 int quad = translation_table[mpc_record]->trans_quad;
155 int local = translation_table[mpc_record]->trans_local;
156
157 quad_local_to_mp_bus_id[quad][local] = m->mpc_busid;
158}
159
160static void __init MP_translation_info(struct mpc_config_translation *m)
161{
162 printk(KERN_INFO
163 "Translation: record %d, type %d, quad %d, global %d, local %d\n",
164 mpc_record, m->trans_type, m->trans_quad, m->trans_global,
165 m->trans_local);
166
167 if (mpc_record >= MAX_MPC_ENTRY)
168 printk(KERN_ERR "MAX_MPC_ENTRY exceeded!\n");
169 else
170 translation_table[mpc_record] = m; /* stash this for later */
171 if (m->trans_quad < MAX_NUMNODES && !node_online(m->trans_quad))
172 node_set_online(m->trans_quad);
173}
174
175static int __init mpf_checksum(unsigned char *mp, int len)
176{
177 int sum = 0;
178
179 while (len--)
180 sum += *mp++;
181
182 return sum & 0xFF;
183}
184
185/*
186 * Read/parse the MPC oem tables
187 */
188
189static void __init smp_read_mpc_oem(struct mp_config_oemtable *oemtable,
190 unsigned short oemsize)
191{
192 int count = sizeof(*oemtable); /* the header size */
193 unsigned char *oemptr = ((unsigned char *)oemtable) + count;
194
195 mpc_record = 0;
196 printk(KERN_INFO "Found an OEM MPC table at %8p - parsing it ... \n",
197 oemtable);
198 if (memcmp(oemtable->oem_signature, MPC_OEM_SIGNATURE, 4)) {
199 printk(KERN_WARNING
200 "SMP mpc oemtable: bad signature [%c%c%c%c]!\n",
201 oemtable->oem_signature[0], oemtable->oem_signature[1],
202 oemtable->oem_signature[2], oemtable->oem_signature[3]);
203 return;
204 }
205 if (mpf_checksum((unsigned char *)oemtable, oemtable->oem_length)) {
206 printk(KERN_WARNING "SMP oem mptable: checksum error!\n");
207 return;
208 }
209 while (count < oemtable->oem_length) {
210 switch (*oemptr) {
211 case MP_TRANSLATION:
212 {
213 struct mpc_config_translation *m =
214 (struct mpc_config_translation *)oemptr;
215 MP_translation_info(m);
216 oemptr += sizeof(*m);
217 count += sizeof(*m);
218 ++mpc_record;
219 break;
220 }
221 default:
222 {
223 printk(KERN_WARNING
224 "Unrecognised OEM table entry type! - %d\n",
225 (int)*oemptr);
226 return;
227 }
228 }
229 }
230}
231
232static struct x86_quirks numaq_x86_quirks __initdata = {
233 .arch_pre_time_init = numaq_pre_time_init,
234 .arch_time_init = NULL,
235 .arch_pre_intr_init = NULL,
236 .arch_memory_setup = NULL,
237 .arch_intr_init = NULL,
238 .arch_trap_init = NULL,
239 .mach_get_smp_config = NULL,
240 .mach_find_smp_config = NULL,
241 .mpc_record = &mpc_record,
242 .mpc_apic_id = mpc_apic_id,
243 .mpc_oem_bus_info = mpc_oem_bus_info,
244 .mpc_oem_pci_bus = mpc_oem_pci_bus,
245 .smp_read_mpc_oem = smp_read_mpc_oem,
246};
247
248void numaq_mps_oem_check(struct mp_config_table *mpc, char *oem,
249 char *productid)
250{
251 if (strncmp(oem, "IBM NUMA", 8))
252 printk("Warning! Not a NUMA-Q system!\n");
253 else
254 found_numaq = 1;
255}
256
257static __init void early_check_numaq(void)
258{
259 /*
260 * Find possible boot-time SMP configuration:
261 */
262 early_find_smp_config();
263 /*
264 * get boot-time SMP configuration:
265 */
266 if (smp_found_config)
267 early_get_smp_config();
268
269 if (found_numaq)
270 x86_quirks = &numaq_x86_quirks;
271}
272
273int __init get_memcfg_numaq(void)
274{
275 early_check_numaq();
276 if (!found_numaq)
277 return 0;
278 smp_dump_qct();
279 return 1;
280}
diff --git a/arch/x86/kernel/paravirt.c b/arch/x86/kernel/paravirt.c
index 74f0c5ea2a03..94da4d52d798 100644
--- a/arch/x86/kernel/paravirt.c
+++ b/arch/x86/kernel/paravirt.c
@@ -29,7 +29,9 @@
29#include <asm/desc.h> 29#include <asm/desc.h>
30#include <asm/setup.h> 30#include <asm/setup.h>
31#include <asm/arch_hooks.h> 31#include <asm/arch_hooks.h>
32#include <asm/pgtable.h>
32#include <asm/time.h> 33#include <asm/time.h>
34#include <asm/pgalloc.h>
33#include <asm/irq.h> 35#include <asm/irq.h>
34#include <asm/delay.h> 36#include <asm/delay.h>
35#include <asm/fixmap.h> 37#include <asm/fixmap.h>
@@ -122,6 +124,7 @@ static void *get_call_destination(u8 type)
122 .pv_irq_ops = pv_irq_ops, 124 .pv_irq_ops = pv_irq_ops,
123 .pv_apic_ops = pv_apic_ops, 125 .pv_apic_ops = pv_apic_ops,
124 .pv_mmu_ops = pv_mmu_ops, 126 .pv_mmu_ops = pv_mmu_ops,
127 .pv_lock_ops = pv_lock_ops,
125 }; 128 };
126 return *((void **)&tmpl + type); 129 return *((void **)&tmpl + type);
127} 130}
@@ -139,7 +142,9 @@ unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
139 /* If the operation is a nop, then nop the callsite */ 142 /* If the operation is a nop, then nop the callsite */
140 ret = paravirt_patch_nop(); 143 ret = paravirt_patch_nop();
141 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) || 144 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
142 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_syscall_ret)) 145 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
146 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
147 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
143 /* If operation requires a jmp, then jmp */ 148 /* If operation requires a jmp, then jmp */
144 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len); 149 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
145 else 150 else
@@ -190,7 +195,9 @@ static void native_flush_tlb_single(unsigned long addr)
190 195
191/* These are in entry.S */ 196/* These are in entry.S */
192extern void native_iret(void); 197extern void native_iret(void);
193extern void native_irq_enable_syscall_ret(void); 198extern void native_irq_enable_sysexit(void);
199extern void native_usergs_sysret32(void);
200extern void native_usergs_sysret64(void);
194 201
195static int __init print_banner(void) 202static int __init print_banner(void)
196{ 203{
@@ -261,6 +268,17 @@ enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
261 return __get_cpu_var(paravirt_lazy_mode); 268 return __get_cpu_var(paravirt_lazy_mode);
262} 269}
263 270
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
264struct pv_info pv_info = { 282struct pv_info pv_info = {
265 .name = "bare hardware", 283 .name = "bare hardware",
266 .paravirt_enabled = 0, 284 .paravirt_enabled = 0,
@@ -280,7 +298,7 @@ struct pv_time_ops pv_time_ops = {
280 .get_wallclock = native_get_wallclock, 298 .get_wallclock = native_get_wallclock,
281 .set_wallclock = native_set_wallclock, 299 .set_wallclock = native_set_wallclock,
282 .sched_clock = native_sched_clock, 300 .sched_clock = native_sched_clock,
283 .get_cpu_khz = native_calculate_cpu_khz, 301 .get_tsc_khz = native_calibrate_tsc,
284}; 302};
285 303
286struct pv_irq_ops pv_irq_ops = { 304struct pv_irq_ops pv_irq_ops = {
@@ -291,6 +309,9 @@ struct pv_irq_ops pv_irq_ops = {
291 .irq_enable = native_irq_enable, 309 .irq_enable = native_irq_enable,
292 .safe_halt = native_safe_halt, 310 .safe_halt = native_safe_halt,
293 .halt = native_halt, 311 .halt = native_halt,
312#ifdef CONFIG_X86_64
313 .adjust_exception_frame = paravirt_nop,
314#endif
294}; 315};
295 316
296struct pv_cpu_ops pv_cpu_ops = { 317struct pv_cpu_ops pv_cpu_ops = {
@@ -321,12 +342,23 @@ struct pv_cpu_ops pv_cpu_ops = {
321 .store_idt = native_store_idt, 342 .store_idt = native_store_idt,
322 .store_tr = native_store_tr, 343 .store_tr = native_store_tr,
323 .load_tls = native_load_tls, 344 .load_tls = native_load_tls,
345#ifdef CONFIG_X86_64
346 .load_gs_index = native_load_gs_index,
347#endif
324 .write_ldt_entry = native_write_ldt_entry, 348 .write_ldt_entry = native_write_ldt_entry,
325 .write_gdt_entry = native_write_gdt_entry, 349 .write_gdt_entry = native_write_gdt_entry,
326 .write_idt_entry = native_write_idt_entry, 350 .write_idt_entry = native_write_idt_entry,
327 .load_sp0 = native_load_sp0, 351 .load_sp0 = native_load_sp0,
328 352
329 .irq_enable_syscall_ret = native_irq_enable_syscall_ret, 353#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
354 .irq_enable_sysexit = native_irq_enable_sysexit,
355#endif
356#ifdef CONFIG_X86_64
357#ifdef CONFIG_IA32_EMULATION
358 .usergs_sysret32 = native_usergs_sysret32,
359#endif
360 .usergs_sysret64 = native_usergs_sysret64,
361#endif
330 .iret = native_iret, 362 .iret = native_iret,
331 .swapgs = native_swapgs, 363 .swapgs = native_swapgs,
332 364
@@ -342,7 +374,6 @@ struct pv_cpu_ops pv_cpu_ops = {
342struct pv_apic_ops pv_apic_ops = { 374struct pv_apic_ops pv_apic_ops = {
343#ifdef CONFIG_X86_LOCAL_APIC 375#ifdef CONFIG_X86_LOCAL_APIC
344 .apic_write = native_apic_write, 376 .apic_write = native_apic_write,
345 .apic_write_atomic = native_apic_write_atomic,
346 .apic_read = native_apic_read, 377 .apic_read = native_apic_read,
347 .setup_boot_clock = setup_boot_APIC_clock, 378 .setup_boot_clock = setup_boot_APIC_clock,
348 .setup_secondary_clock = setup_secondary_APIC_clock, 379 .setup_secondary_clock = setup_secondary_APIC_clock,
@@ -354,6 +385,9 @@ struct pv_mmu_ops pv_mmu_ops = {
354#ifndef CONFIG_X86_64 385#ifndef CONFIG_X86_64
355 .pagetable_setup_start = native_pagetable_setup_start, 386 .pagetable_setup_start = native_pagetable_setup_start,
356 .pagetable_setup_done = native_pagetable_setup_done, 387 .pagetable_setup_done = native_pagetable_setup_done,
388#else
389 .pagetable_setup_start = paravirt_nop,
390 .pagetable_setup_done = paravirt_nop,
357#endif 391#endif
358 392
359 .read_cr2 = native_read_cr2, 393 .read_cr2 = native_read_cr2,
@@ -366,6 +400,9 @@ struct pv_mmu_ops pv_mmu_ops = {
366 .flush_tlb_single = native_flush_tlb_single, 400 .flush_tlb_single = native_flush_tlb_single,
367 .flush_tlb_others = native_flush_tlb_others, 401 .flush_tlb_others = native_flush_tlb_others,
368 402
403 .pgd_alloc = __paravirt_pgd_alloc,
404 .pgd_free = paravirt_nop,
405
369 .alloc_pte = paravirt_nop, 406 .alloc_pte = paravirt_nop,
370 .alloc_pmd = paravirt_nop, 407 .alloc_pmd = paravirt_nop,
371 .alloc_pmd_clone = paravirt_nop, 408 .alloc_pmd_clone = paravirt_nop,
@@ -380,6 +417,9 @@ struct pv_mmu_ops pv_mmu_ops = {
380 .pte_update = paravirt_nop, 417 .pte_update = paravirt_nop,
381 .pte_update_defer = paravirt_nop, 418 .pte_update_defer = paravirt_nop,
382 419
420 .ptep_modify_prot_start = __ptep_modify_prot_start,
421 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
422
383#ifdef CONFIG_HIGHPTE 423#ifdef CONFIG_HIGHPTE
384 .kmap_atomic_pte = kmap_atomic, 424 .kmap_atomic_pte = kmap_atomic,
385#endif 425#endif
@@ -403,6 +443,7 @@ struct pv_mmu_ops pv_mmu_ops = {
403#endif /* PAGETABLE_LEVELS >= 3 */ 443#endif /* PAGETABLE_LEVELS >= 3 */
404 444
405 .pte_val = native_pte_val, 445 .pte_val = native_pte_val,
446 .pte_flags = native_pte_flags,
406 .pgd_val = native_pgd_val, 447 .pgd_val = native_pgd_val,
407 448
408 .make_pte = native_make_pte, 449 .make_pte = native_make_pte,
@@ -416,7 +457,21 @@ struct pv_mmu_ops pv_mmu_ops = {
416 .enter = paravirt_nop, 457 .enter = paravirt_nop,
417 .leave = paravirt_nop, 458 .leave = paravirt_nop,
418 }, 459 },
460
461 .set_fixmap = native_set_fixmap,
462};
463
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
419}; 473};
474EXPORT_SYMBOL_GPL(pv_lock_ops);
420 475
421EXPORT_SYMBOL_GPL(pv_time_ops); 476EXPORT_SYMBOL_GPL(pv_time_ops);
422EXPORT_SYMBOL (pv_cpu_ops); 477EXPORT_SYMBOL (pv_cpu_ops);
diff --git a/arch/x86/kernel/paravirt_patch_32.c b/arch/x86/kernel/paravirt_patch_32.c
index 82fc5fcab4f4..58262218781b 100644
--- a/arch/x86/kernel/paravirt_patch_32.c
+++ b/arch/x86/kernel/paravirt_patch_32.c
@@ -5,7 +5,7 @@ DEF_NATIVE(pv_irq_ops, irq_enable, "sti");
5DEF_NATIVE(pv_irq_ops, restore_fl, "push %eax; popf"); 5DEF_NATIVE(pv_irq_ops, restore_fl, "push %eax; popf");
6DEF_NATIVE(pv_irq_ops, save_fl, "pushf; pop %eax"); 6DEF_NATIVE(pv_irq_ops, save_fl, "pushf; pop %eax");
7DEF_NATIVE(pv_cpu_ops, iret, "iret"); 7DEF_NATIVE(pv_cpu_ops, iret, "iret");
8DEF_NATIVE(pv_cpu_ops, irq_enable_syscall_ret, "sti; sysexit"); 8DEF_NATIVE(pv_cpu_ops, irq_enable_sysexit, "sti; sysexit");
9DEF_NATIVE(pv_mmu_ops, read_cr2, "mov %cr2, %eax"); 9DEF_NATIVE(pv_mmu_ops, read_cr2, "mov %cr2, %eax");
10DEF_NATIVE(pv_mmu_ops, write_cr3, "mov %eax, %cr3"); 10DEF_NATIVE(pv_mmu_ops, write_cr3, "mov %eax, %cr3");
11DEF_NATIVE(pv_mmu_ops, read_cr3, "mov %cr3, %eax"); 11DEF_NATIVE(pv_mmu_ops, read_cr3, "mov %cr3, %eax");
@@ -29,7 +29,7 @@ unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
29 PATCH_SITE(pv_irq_ops, restore_fl); 29 PATCH_SITE(pv_irq_ops, restore_fl);
30 PATCH_SITE(pv_irq_ops, save_fl); 30 PATCH_SITE(pv_irq_ops, save_fl);
31 PATCH_SITE(pv_cpu_ops, iret); 31 PATCH_SITE(pv_cpu_ops, iret);
32 PATCH_SITE(pv_cpu_ops, irq_enable_syscall_ret); 32 PATCH_SITE(pv_cpu_ops, irq_enable_sysexit);
33 PATCH_SITE(pv_mmu_ops, read_cr2); 33 PATCH_SITE(pv_mmu_ops, read_cr2);
34 PATCH_SITE(pv_mmu_ops, read_cr3); 34 PATCH_SITE(pv_mmu_ops, read_cr3);
35 PATCH_SITE(pv_mmu_ops, write_cr3); 35 PATCH_SITE(pv_mmu_ops, write_cr3);
diff --git a/arch/x86/kernel/paravirt_patch_64.c b/arch/x86/kernel/paravirt_patch_64.c
index 7d904e138d7e..061d01df9ae6 100644
--- a/arch/x86/kernel/paravirt_patch_64.c
+++ b/arch/x86/kernel/paravirt_patch_64.c
@@ -14,8 +14,9 @@ DEF_NATIVE(pv_mmu_ops, flush_tlb_single, "invlpg (%rdi)");
14DEF_NATIVE(pv_cpu_ops, clts, "clts"); 14DEF_NATIVE(pv_cpu_ops, clts, "clts");
15DEF_NATIVE(pv_cpu_ops, wbinvd, "wbinvd"); 15DEF_NATIVE(pv_cpu_ops, wbinvd, "wbinvd");
16 16
17/* the three commands give us more control to how to return from a syscall */ 17DEF_NATIVE(pv_cpu_ops, irq_enable_sysexit, "swapgs; sti; sysexit");
18DEF_NATIVE(pv_cpu_ops, irq_enable_syscall_ret, "movq %gs:" __stringify(pda_oldrsp) ", %rsp; swapgs; sysretq;"); 18DEF_NATIVE(pv_cpu_ops, usergs_sysret64, "swapgs; sysretq");
19DEF_NATIVE(pv_cpu_ops, usergs_sysret32, "swapgs; sysretl");
19DEF_NATIVE(pv_cpu_ops, swapgs, "swapgs"); 20DEF_NATIVE(pv_cpu_ops, swapgs, "swapgs");
20 21
21unsigned native_patch(u8 type, u16 clobbers, void *ibuf, 22unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
@@ -35,7 +36,9 @@ unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
35 PATCH_SITE(pv_irq_ops, irq_enable); 36 PATCH_SITE(pv_irq_ops, irq_enable);
36 PATCH_SITE(pv_irq_ops, irq_disable); 37 PATCH_SITE(pv_irq_ops, irq_disable);
37 PATCH_SITE(pv_cpu_ops, iret); 38 PATCH_SITE(pv_cpu_ops, iret);
38 PATCH_SITE(pv_cpu_ops, irq_enable_syscall_ret); 39 PATCH_SITE(pv_cpu_ops, irq_enable_sysexit);
40 PATCH_SITE(pv_cpu_ops, usergs_sysret32);
41 PATCH_SITE(pv_cpu_ops, usergs_sysret64);
39 PATCH_SITE(pv_cpu_ops, swapgs); 42 PATCH_SITE(pv_cpu_ops, swapgs);
40 PATCH_SITE(pv_mmu_ops, read_cr2); 43 PATCH_SITE(pv_mmu_ops, read_cr2);
41 PATCH_SITE(pv_mmu_ops, read_cr3); 44 PATCH_SITE(pv_mmu_ops, read_cr3);
diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index e28ec497e142..b67a4b1d4eae 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -29,6 +29,7 @@
29#include <linux/mm.h> 29#include <linux/mm.h>
30#include <linux/spinlock.h> 30#include <linux/spinlock.h>
31#include <linux/string.h> 31#include <linux/string.h>
32#include <linux/crash_dump.h>
32#include <linux/dma-mapping.h> 33#include <linux/dma-mapping.h>
33#include <linux/bitops.h> 34#include <linux/bitops.h>
34#include <linux/pci_ids.h> 35#include <linux/pci_ids.h>
@@ -36,7 +37,8 @@
36#include <linux/delay.h> 37#include <linux/delay.h>
37#include <linux/scatterlist.h> 38#include <linux/scatterlist.h>
38#include <linux/iommu-helper.h> 39#include <linux/iommu-helper.h>
39#include <asm/gart.h> 40
41#include <asm/iommu.h>
40#include <asm/calgary.h> 42#include <asm/calgary.h>
41#include <asm/tce.h> 43#include <asm/tce.h>
42#include <asm/pci-direct.h> 44#include <asm/pci-direct.h>
@@ -167,6 +169,8 @@ static void calgary_dump_error_regs(struct iommu_table *tbl);
167static void calioc2_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev); 169static void calioc2_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev);
168static void calioc2_tce_cache_blast(struct iommu_table *tbl); 170static void calioc2_tce_cache_blast(struct iommu_table *tbl);
169static void calioc2_dump_error_regs(struct iommu_table *tbl); 171static void calioc2_dump_error_regs(struct iommu_table *tbl);
172static void calgary_init_bitmap_from_tce_table(struct iommu_table *tbl);
173static void get_tce_space_from_tar(void);
170 174
171static struct cal_chipset_ops calgary_chip_ops = { 175static struct cal_chipset_ops calgary_chip_ops = {
172 .handle_quirks = calgary_handle_quirks, 176 .handle_quirks = calgary_handle_quirks,
@@ -410,22 +414,6 @@ static void calgary_unmap_sg(struct device *dev,
410 } 414 }
411} 415}
412 416
413static int calgary_nontranslate_map_sg(struct device* dev,
414 struct scatterlist *sg, int nelems, int direction)
415{
416 struct scatterlist *s;
417 int i;
418
419 for_each_sg(sg, s, nelems, i) {
420 struct page *p = sg_page(s);
421
422 BUG_ON(!p);
423 s->dma_address = virt_to_bus(sg_virt(s));
424 s->dma_length = s->length;
425 }
426 return nelems;
427}
428
429static int calgary_map_sg(struct device *dev, struct scatterlist *sg, 417static int calgary_map_sg(struct device *dev, struct scatterlist *sg,
430 int nelems, int direction) 418 int nelems, int direction)
431{ 419{
@@ -436,9 +424,6 @@ static int calgary_map_sg(struct device *dev, struct scatterlist *sg,
436 unsigned long entry; 424 unsigned long entry;
437 int i; 425 int i;
438 426
439 if (!translation_enabled(tbl))
440 return calgary_nontranslate_map_sg(dev, sg, nelems, direction);
441
442 for_each_sg(sg, s, nelems, i) { 427 for_each_sg(sg, s, nelems, i) {
443 BUG_ON(!sg_page(s)); 428 BUG_ON(!sg_page(s));
444 429
@@ -474,7 +459,6 @@ error:
474static dma_addr_t calgary_map_single(struct device *dev, phys_addr_t paddr, 459static dma_addr_t calgary_map_single(struct device *dev, phys_addr_t paddr,
475 size_t size, int direction) 460 size_t size, int direction)
476{ 461{
477 dma_addr_t dma_handle = bad_dma_address;
478 void *vaddr = phys_to_virt(paddr); 462 void *vaddr = phys_to_virt(paddr);
479 unsigned long uaddr; 463 unsigned long uaddr;
480 unsigned int npages; 464 unsigned int npages;
@@ -483,12 +467,7 @@ static dma_addr_t calgary_map_single(struct device *dev, phys_addr_t paddr,
483 uaddr = (unsigned long)vaddr; 467 uaddr = (unsigned long)vaddr;
484 npages = num_dma_pages(uaddr, size); 468 npages = num_dma_pages(uaddr, size);
485 469
486 if (translation_enabled(tbl)) 470 return iommu_alloc(dev, tbl, vaddr, npages, direction);
487 dma_handle = iommu_alloc(dev, tbl, vaddr, npages, direction);
488 else
489 dma_handle = virt_to_bus(vaddr);
490
491 return dma_handle;
492} 471}
493 472
494static void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle, 473static void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle,
@@ -497,9 +476,6 @@ static void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle,
497 struct iommu_table *tbl = find_iommu_table(dev); 476 struct iommu_table *tbl = find_iommu_table(dev);
498 unsigned int npages; 477 unsigned int npages;
499 478
500 if (!translation_enabled(tbl))
501 return;
502
503 npages = num_dma_pages(dma_handle, size); 479 npages = num_dma_pages(dma_handle, size);
504 iommu_free(tbl, dma_handle, npages); 480 iommu_free(tbl, dma_handle, npages);
505} 481}
@@ -522,18 +498,12 @@ static void* calgary_alloc_coherent(struct device *dev, size_t size,
522 goto error; 498 goto error;
523 memset(ret, 0, size); 499 memset(ret, 0, size);
524 500
525 if (translation_enabled(tbl)) { 501 /* set up tces to cover the allocated range */
526 /* set up tces to cover the allocated range */ 502 mapping = iommu_alloc(dev, tbl, ret, npages, DMA_BIDIRECTIONAL);
527 mapping = iommu_alloc(dev, tbl, ret, npages, DMA_BIDIRECTIONAL); 503 if (mapping == bad_dma_address)
528 if (mapping == bad_dma_address) 504 goto free;
529 goto free; 505 *dma_handle = mapping;
530
531 *dma_handle = mapping;
532 } else /* non translated slot */
533 *dma_handle = virt_to_bus(ret);
534
535 return ret; 506 return ret;
536
537free: 507free:
538 free_pages((unsigned long)ret, get_order(size)); 508 free_pages((unsigned long)ret, get_order(size));
539 ret = NULL; 509 ret = NULL;
@@ -541,7 +511,7 @@ error:
541 return ret; 511 return ret;
542} 512}
543 513
544static const struct dma_mapping_ops calgary_dma_ops = { 514static struct dma_mapping_ops calgary_dma_ops = {
545 .alloc_coherent = calgary_alloc_coherent, 515 .alloc_coherent = calgary_alloc_coherent,
546 .map_single = calgary_map_single, 516 .map_single = calgary_map_single,
547 .unmap_single = calgary_unmap_single, 517 .unmap_single = calgary_unmap_single,
@@ -830,7 +800,11 @@ static int __init calgary_setup_tar(struct pci_dev *dev, void __iomem *bbar)
830 800
831 tbl = pci_iommu(dev->bus); 801 tbl = pci_iommu(dev->bus);
832 tbl->it_base = (unsigned long)bus_info[dev->bus->number].tce_space; 802 tbl->it_base = (unsigned long)bus_info[dev->bus->number].tce_space;
833 tce_free(tbl, 0, tbl->it_size); 803
804 if (is_kdump_kernel())
805 calgary_init_bitmap_from_tce_table(tbl);
806 else
807 tce_free(tbl, 0, tbl->it_size);
834 808
835 if (is_calgary(dev->device)) 809 if (is_calgary(dev->device))
836 tbl->chip_ops = &calgary_chip_ops; 810 tbl->chip_ops = &calgary_chip_ops;
@@ -1209,6 +1183,10 @@ static int __init calgary_init(void)
1209 if (ret) 1183 if (ret)
1210 return ret; 1184 return ret;
1211 1185
1186 /* Purely for kdump kernel case */
1187 if (is_kdump_kernel())
1188 get_tce_space_from_tar();
1189
1212 do { 1190 do {
1213 dev = pci_get_device(PCI_VENDOR_ID_IBM, PCI_ANY_ID, dev); 1191 dev = pci_get_device(PCI_VENDOR_ID_IBM, PCI_ANY_ID, dev);
1214 if (!dev) 1192 if (!dev)
@@ -1230,6 +1208,16 @@ static int __init calgary_init(void)
1230 goto error; 1208 goto error;
1231 } while (1); 1209 } while (1);
1232 1210
1211 dev = NULL;
1212 for_each_pci_dev(dev) {
1213 struct iommu_table *tbl;
1214
1215 tbl = find_iommu_table(&dev->dev);
1216
1217 if (translation_enabled(tbl))
1218 dev->dev.archdata.dma_ops = &calgary_dma_ops;
1219 }
1220
1233 return ret; 1221 return ret;
1234 1222
1235error: 1223error:
@@ -1251,6 +1239,7 @@ error:
1251 calgary_disable_translation(dev); 1239 calgary_disable_translation(dev);
1252 calgary_free_bus(dev); 1240 calgary_free_bus(dev);
1253 pci_dev_put(dev); /* Undo calgary_init_one()'s pci_dev_get() */ 1241 pci_dev_put(dev); /* Undo calgary_init_one()'s pci_dev_get() */
1242 dev->dev.archdata.dma_ops = NULL;
1254 } while (1); 1243 } while (1);
1255 1244
1256 return ret; 1245 return ret;
@@ -1339,6 +1328,61 @@ static int __init calgary_bus_has_devices(int bus, unsigned short pci_dev)
1339 return (val != 0xffffffff); 1328 return (val != 0xffffffff);
1340} 1329}
1341 1330
1331/*
1332 * calgary_init_bitmap_from_tce_table():
1333 * Funtion for kdump case. In the second/kdump kernel initialize
1334 * the bitmap based on the tce table entries obtained from first kernel
1335 */
1336static void calgary_init_bitmap_from_tce_table(struct iommu_table *tbl)
1337{
1338 u64 *tp;
1339 unsigned int index;
1340 tp = ((u64 *)tbl->it_base);
1341 for (index = 0 ; index < tbl->it_size; index++) {
1342 if (*tp != 0x0)
1343 set_bit(index, tbl->it_map);
1344 tp++;
1345 }
1346}
1347
1348/*
1349 * get_tce_space_from_tar():
1350 * Function for kdump case. Get the tce tables from first kernel
1351 * by reading the contents of the base adress register of calgary iommu
1352 */
1353static void get_tce_space_from_tar()
1354{
1355 int bus;
1356 void __iomem *target;
1357 unsigned long tce_space;
1358
1359 for (bus = 0; bus < MAX_PHB_BUS_NUM; bus++) {
1360 struct calgary_bus_info *info = &bus_info[bus];
1361 unsigned short pci_device;
1362 u32 val;
1363
1364 val = read_pci_config(bus, 0, 0, 0);
1365 pci_device = (val & 0xFFFF0000) >> 16;
1366
1367 if (!is_cal_pci_dev(pci_device))
1368 continue;
1369 if (info->translation_disabled)
1370 continue;
1371
1372 if (calgary_bus_has_devices(bus, pci_device) ||
1373 translate_empty_slots) {
1374 target = calgary_reg(bus_info[bus].bbar,
1375 tar_offset(bus));
1376 tce_space = be64_to_cpu(readq(target));
1377 tce_space = tce_space & TAR_SW_BITS;
1378
1379 tce_space = tce_space & (~specified_table_size);
1380 info->tce_space = (u64 *)__va(tce_space);
1381 }
1382 }
1383 return;
1384}
1385
1342void __init detect_calgary(void) 1386void __init detect_calgary(void)
1343{ 1387{
1344 int bus; 1388 int bus;
@@ -1394,7 +1438,8 @@ void __init detect_calgary(void)
1394 return; 1438 return;
1395 } 1439 }
1396 1440
1397 specified_table_size = determine_tce_table_size(end_pfn * PAGE_SIZE); 1441 specified_table_size = determine_tce_table_size((is_kdump_kernel() ?
1442 saved_max_pfn : max_pfn) * PAGE_SIZE);
1398 1443
1399 for (bus = 0; bus < MAX_PHB_BUS_NUM; bus++) { 1444 for (bus = 0; bus < MAX_PHB_BUS_NUM; bus++) {
1400 struct calgary_bus_info *info = &bus_info[bus]; 1445 struct calgary_bus_info *info = &bus_info[bus];
@@ -1412,10 +1457,16 @@ void __init detect_calgary(void)
1412 1457
1413 if (calgary_bus_has_devices(bus, pci_device) || 1458 if (calgary_bus_has_devices(bus, pci_device) ||
1414 translate_empty_slots) { 1459 translate_empty_slots) {
1415 tbl = alloc_tce_table(); 1460 /*
1416 if (!tbl) 1461 * If it is kdump kernel, find and use tce tables
1417 goto cleanup; 1462 * from first kernel, else allocate tce tables here
1418 info->tce_space = tbl; 1463 */
1464 if (!is_kdump_kernel()) {
1465 tbl = alloc_tce_table();
1466 if (!tbl)
1467 goto cleanup;
1468 info->tce_space = tbl;
1469 }
1419 calgary_found = 1; 1470 calgary_found = 1;
1420 } 1471 }
1421 } 1472 }
@@ -1430,6 +1481,10 @@ void __init detect_calgary(void)
1430 printk(KERN_INFO "PCI-DMA: Calgary TCE table spec is %d, " 1481 printk(KERN_INFO "PCI-DMA: Calgary TCE table spec is %d, "
1431 "CONFIG_IOMMU_DEBUG is %s.\n", specified_table_size, 1482 "CONFIG_IOMMU_DEBUG is %s.\n", specified_table_size,
1432 debugging ? "enabled" : "disabled"); 1483 debugging ? "enabled" : "disabled");
1484
1485 /* swiotlb for devices that aren't behind the Calgary. */
1486 if (max_pfn > MAX_DMA32_PFN)
1487 swiotlb = 1;
1433 } 1488 }
1434 return; 1489 return;
1435 1490
@@ -1446,7 +1501,7 @@ int __init calgary_iommu_init(void)
1446{ 1501{
1447 int ret; 1502 int ret;
1448 1503
1449 if (no_iommu || swiotlb) 1504 if (no_iommu || (swiotlb && !calgary_detected))
1450 return -ENODEV; 1505 return -ENODEV;
1451 1506
1452 if (!calgary_detected) 1507 if (!calgary_detected)
@@ -1459,15 +1514,14 @@ int __init calgary_iommu_init(void)
1459 if (ret) { 1514 if (ret) {
1460 printk(KERN_ERR "PCI-DMA: Calgary init failed %d, " 1515 printk(KERN_ERR "PCI-DMA: Calgary init failed %d, "
1461 "falling back to no_iommu\n", ret); 1516 "falling back to no_iommu\n", ret);
1462 if (end_pfn > MAX_DMA32_PFN)
1463 printk(KERN_ERR "WARNING more than 4GB of memory, "
1464 "32bit PCI may malfunction.\n");
1465 return ret; 1517 return ret;
1466 } 1518 }
1467 1519
1468 force_iommu = 1; 1520 force_iommu = 1;
1469 bad_dma_address = 0x0; 1521 bad_dma_address = 0x0;
1470 dma_ops = &calgary_dma_ops; 1522 /* dma_ops is set to swiotlb or nommu */
1523 if (!dma_ops)
1524 dma_ops = &nommu_dma_ops;
1471 1525
1472 return 0; 1526 return 0;
1473} 1527}
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index dc00a1331ace..8dbffb846de9 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -5,13 +5,13 @@
5 5
6#include <asm/proto.h> 6#include <asm/proto.h>
7#include <asm/dma.h> 7#include <asm/dma.h>
8#include <asm/gart.h> 8#include <asm/iommu.h>
9#include <asm/calgary.h> 9#include <asm/calgary.h>
10#include <asm/amd_iommu.h>
10 11
11int forbid_dac __read_mostly; 12static int forbid_dac __read_mostly;
12EXPORT_SYMBOL(forbid_dac);
13 13
14const struct dma_mapping_ops *dma_ops; 14struct dma_mapping_ops *dma_ops;
15EXPORT_SYMBOL(dma_ops); 15EXPORT_SYMBOL(dma_ops);
16 16
17static int iommu_sac_force __read_mostly; 17static int iommu_sac_force __read_mostly;
@@ -74,13 +74,17 @@ early_param("dma32_size", parse_dma32_size_opt);
74void __init dma32_reserve_bootmem(void) 74void __init dma32_reserve_bootmem(void)
75{ 75{
76 unsigned long size, align; 76 unsigned long size, align;
77 if (end_pfn <= MAX_DMA32_PFN) 77 if (max_pfn <= MAX_DMA32_PFN)
78 return; 78 return;
79 79
80 /*
81 * check aperture_64.c allocate_aperture() for reason about
82 * using 512M as goal
83 */
80 align = 64ULL<<20; 84 align = 64ULL<<20;
81 size = round_up(dma32_bootmem_size, align); 85 size = round_up(dma32_bootmem_size, align);
82 dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align, 86 dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align,
83 __pa(MAX_DMA_ADDRESS)); 87 512ULL<<20);
84 if (dma32_bootmem_ptr) 88 if (dma32_bootmem_ptr)
85 dma32_bootmem_size = size; 89 dma32_bootmem_size = size;
86 else 90 else
@@ -88,17 +92,14 @@ void __init dma32_reserve_bootmem(void)
88} 92}
89static void __init dma32_free_bootmem(void) 93static void __init dma32_free_bootmem(void)
90{ 94{
91 int node;
92 95
93 if (end_pfn <= MAX_DMA32_PFN) 96 if (max_pfn <= MAX_DMA32_PFN)
94 return; 97 return;
95 98
96 if (!dma32_bootmem_ptr) 99 if (!dma32_bootmem_ptr)
97 return; 100 return;
98 101
99 for_each_online_node(node) 102 free_bootmem(__pa(dma32_bootmem_ptr), dma32_bootmem_size);
100 free_bootmem_node(NODE_DATA(node), __pa(dma32_bootmem_ptr),
101 dma32_bootmem_size);
102 103
103 dma32_bootmem_ptr = NULL; 104 dma32_bootmem_ptr = NULL;
104 dma32_bootmem_size = 0; 105 dma32_bootmem_size = 0;
@@ -112,19 +113,15 @@ void __init pci_iommu_alloc(void)
112 * The order of these functions is important for 113 * The order of these functions is important for
113 * fall-back/fail-over reasons 114 * fall-back/fail-over reasons
114 */ 115 */
115#ifdef CONFIG_GART_IOMMU
116 gart_iommu_hole_init(); 116 gart_iommu_hole_init();
117#endif
118 117
119#ifdef CONFIG_CALGARY_IOMMU
120 detect_calgary(); 118 detect_calgary();
121#endif
122 119
123 detect_intel_iommu(); 120 detect_intel_iommu();
124 121
125#ifdef CONFIG_SWIOTLB 122 amd_iommu_detect();
123
126 pci_swiotlb_init(); 124 pci_swiotlb_init();
127#endif
128} 125}
129#endif 126#endif
130 127
@@ -180,9 +177,7 @@ static __init int iommu_setup(char *p)
180 swiotlb = 1; 177 swiotlb = 1;
181#endif 178#endif
182 179
183#ifdef CONFIG_GART_IOMMU
184 gart_parse_options(p); 180 gart_parse_options(p);
185#endif
186 181
187#ifdef CONFIG_CALGARY_IOMMU 182#ifdef CONFIG_CALGARY_IOMMU
188 if (!strncmp(p, "calgary", 7)) 183 if (!strncmp(p, "calgary", 7))
@@ -197,136 +192,19 @@ static __init int iommu_setup(char *p)
197} 192}
198early_param("iommu", iommu_setup); 193early_param("iommu", iommu_setup);
199 194
200#ifdef CONFIG_X86_32
201int dma_declare_coherent_memory(struct device *dev, dma_addr_t bus_addr,
202 dma_addr_t device_addr, size_t size, int flags)
203{
204 void __iomem *mem_base = NULL;
205 int pages = size >> PAGE_SHIFT;
206 int bitmap_size = BITS_TO_LONGS(pages) * sizeof(long);
207
208 if ((flags & (DMA_MEMORY_MAP | DMA_MEMORY_IO)) == 0)
209 goto out;
210 if (!size)
211 goto out;
212 if (dev->dma_mem)
213 goto out;
214
215 /* FIXME: this routine just ignores DMA_MEMORY_INCLUDES_CHILDREN */
216
217 mem_base = ioremap(bus_addr, size);
218 if (!mem_base)
219 goto out;
220
221 dev->dma_mem = kzalloc(sizeof(struct dma_coherent_mem), GFP_KERNEL);
222 if (!dev->dma_mem)
223 goto out;
224 dev->dma_mem->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
225 if (!dev->dma_mem->bitmap)
226 goto free1_out;
227
228 dev->dma_mem->virt_base = mem_base;
229 dev->dma_mem->device_base = device_addr;
230 dev->dma_mem->size = pages;
231 dev->dma_mem->flags = flags;
232
233 if (flags & DMA_MEMORY_MAP)
234 return DMA_MEMORY_MAP;
235
236 return DMA_MEMORY_IO;
237
238 free1_out:
239 kfree(dev->dma_mem);
240 out:
241 if (mem_base)
242 iounmap(mem_base);
243 return 0;
244}
245EXPORT_SYMBOL(dma_declare_coherent_memory);
246
247void dma_release_declared_memory(struct device *dev)
248{
249 struct dma_coherent_mem *mem = dev->dma_mem;
250
251 if (!mem)
252 return;
253 dev->dma_mem = NULL;
254 iounmap(mem->virt_base);
255 kfree(mem->bitmap);
256 kfree(mem);
257}
258EXPORT_SYMBOL(dma_release_declared_memory);
259
260void *dma_mark_declared_memory_occupied(struct device *dev,
261 dma_addr_t device_addr, size_t size)
262{
263 struct dma_coherent_mem *mem = dev->dma_mem;
264 int pos, err;
265 int pages = (size + (device_addr & ~PAGE_MASK) + PAGE_SIZE - 1);
266
267 pages >>= PAGE_SHIFT;
268
269 if (!mem)
270 return ERR_PTR(-EINVAL);
271
272 pos = (device_addr - mem->device_base) >> PAGE_SHIFT;
273 err = bitmap_allocate_region(mem->bitmap, pos, get_order(pages));
274 if (err != 0)
275 return ERR_PTR(err);
276 return mem->virt_base + (pos << PAGE_SHIFT);
277}
278EXPORT_SYMBOL(dma_mark_declared_memory_occupied);
279
280static int dma_alloc_from_coherent_mem(struct device *dev, ssize_t size,
281 dma_addr_t *dma_handle, void **ret)
282{
283 struct dma_coherent_mem *mem = dev ? dev->dma_mem : NULL;
284 int order = get_order(size);
285
286 if (mem) {
287 int page = bitmap_find_free_region(mem->bitmap, mem->size,
288 order);
289 if (page >= 0) {
290 *dma_handle = mem->device_base + (page << PAGE_SHIFT);
291 *ret = mem->virt_base + (page << PAGE_SHIFT);
292 memset(*ret, 0, size);
293 }
294 if (mem->flags & DMA_MEMORY_EXCLUSIVE)
295 *ret = NULL;
296 }
297 return (mem != NULL);
298}
299
300static int dma_release_coherent(struct device *dev, int order, void *vaddr)
301{
302 struct dma_coherent_mem *mem = dev ? dev->dma_mem : NULL;
303
304 if (mem && vaddr >= mem->virt_base && vaddr <
305 (mem->virt_base + (mem->size << PAGE_SHIFT))) {
306 int page = (vaddr - mem->virt_base) >> PAGE_SHIFT;
307
308 bitmap_release_region(mem->bitmap, page, order);
309 return 1;
310 }
311 return 0;
312}
313#else
314#define dma_alloc_from_coherent_mem(dev, size, handle, ret) (0)
315#define dma_release_coherent(dev, order, vaddr) (0)
316#endif /* CONFIG_X86_32 */
317
318int dma_supported(struct device *dev, u64 mask) 195int dma_supported(struct device *dev, u64 mask)
319{ 196{
197 struct dma_mapping_ops *ops = get_dma_ops(dev);
198
320#ifdef CONFIG_PCI 199#ifdef CONFIG_PCI
321 if (mask > 0xffffffff && forbid_dac > 0) { 200 if (mask > 0xffffffff && forbid_dac > 0) {
322 printk(KERN_INFO "PCI: Disallowing DAC for device %s\n", 201 dev_info(dev, "PCI: Disallowing DAC for device\n");
323 dev->bus_id);
324 return 0; 202 return 0;
325 } 203 }
326#endif 204#endif
327 205
328 if (dma_ops->dma_supported) 206 if (ops->dma_supported)
329 return dma_ops->dma_supported(dev, mask); 207 return ops->dma_supported(dev, mask);
330 208
331 /* Copied from i386. Doesn't make much sense, because it will 209 /* Copied from i386. Doesn't make much sense, because it will
332 only work for pci_alloc_coherent. 210 only work for pci_alloc_coherent.
@@ -347,8 +225,7 @@ int dma_supported(struct device *dev, u64 mask)
347 type. Normally this doesn't make any difference, but gives 225 type. Normally this doesn't make any difference, but gives
348 more gentle handling of IOMMU overflow. */ 226 more gentle handling of IOMMU overflow. */
349 if (iommu_sac_force && (mask >= DMA_40BIT_MASK)) { 227 if (iommu_sac_force && (mask >= DMA_40BIT_MASK)) {
350 printk(KERN_INFO "%s: Force SAC with mask %Lx\n", 228 dev_info(dev, "Force SAC with mask %Lx\n", mask);
351 dev->bus_id, mask);
352 return 0; 229 return 0;
353 } 230 }
354 231
@@ -357,7 +234,7 @@ int dma_supported(struct device *dev, u64 mask)
357EXPORT_SYMBOL(dma_supported); 234EXPORT_SYMBOL(dma_supported);
358 235
359/* Allocate DMA memory on node near device */ 236/* Allocate DMA memory on node near device */
360noinline struct page * 237static noinline struct page *
361dma_alloc_pages(struct device *dev, gfp_t gfp, unsigned order) 238dma_alloc_pages(struct device *dev, gfp_t gfp, unsigned order)
362{ 239{
363 int node; 240 int node;
@@ -374,6 +251,7 @@ void *
374dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle, 251dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
375 gfp_t gfp) 252 gfp_t gfp)
376{ 253{
254 struct dma_mapping_ops *ops = get_dma_ops(dev);
377 void *memory = NULL; 255 void *memory = NULL;
378 struct page *page; 256 struct page *page;
379 unsigned long dma_mask = 0; 257 unsigned long dma_mask = 0;
@@ -383,7 +261,7 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
383 /* ignore region specifiers */ 261 /* ignore region specifiers */
384 gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32); 262 gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
385 263
386 if (dma_alloc_from_coherent_mem(dev, size, dma_handle, &memory)) 264 if (dma_alloc_from_coherent(dev, size, dma_handle, &memory))
387 return memory; 265 return memory;
388 266
389 if (!dev) { 267 if (!dev) {
@@ -442,8 +320,8 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
442 /* Let low level make its own zone decisions */ 320 /* Let low level make its own zone decisions */
443 gfp &= ~(GFP_DMA32|GFP_DMA); 321 gfp &= ~(GFP_DMA32|GFP_DMA);
444 322
445 if (dma_ops->alloc_coherent) 323 if (ops->alloc_coherent)
446 return dma_ops->alloc_coherent(dev, size, 324 return ops->alloc_coherent(dev, size,
447 dma_handle, gfp); 325 dma_handle, gfp);
448 return NULL; 326 return NULL;
449 } 327 }
@@ -455,14 +333,14 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
455 } 333 }
456 } 334 }
457 335
458 if (dma_ops->alloc_coherent) { 336 if (ops->alloc_coherent) {
459 free_pages((unsigned long)memory, get_order(size)); 337 free_pages((unsigned long)memory, get_order(size));
460 gfp &= ~(GFP_DMA|GFP_DMA32); 338 gfp &= ~(GFP_DMA|GFP_DMA32);
461 return dma_ops->alloc_coherent(dev, size, dma_handle, gfp); 339 return ops->alloc_coherent(dev, size, dma_handle, gfp);
462 } 340 }
463 341
464 if (dma_ops->map_simple) { 342 if (ops->map_simple) {
465 *dma_handle = dma_ops->map_simple(dev, virt_to_phys(memory), 343 *dma_handle = ops->map_simple(dev, virt_to_phys(memory),
466 size, 344 size,
467 PCI_DMA_BIDIRECTIONAL); 345 PCI_DMA_BIDIRECTIONAL);
468 if (*dma_handle != bad_dma_address) 346 if (*dma_handle != bad_dma_address)
@@ -484,27 +362,27 @@ EXPORT_SYMBOL(dma_alloc_coherent);
484void dma_free_coherent(struct device *dev, size_t size, 362void dma_free_coherent(struct device *dev, size_t size,
485 void *vaddr, dma_addr_t bus) 363 void *vaddr, dma_addr_t bus)
486{ 364{
365 struct dma_mapping_ops *ops = get_dma_ops(dev);
366
487 int order = get_order(size); 367 int order = get_order(size);
488 WARN_ON(irqs_disabled()); /* for portability */ 368 WARN_ON(irqs_disabled()); /* for portability */
489 if (dma_release_coherent(dev, order, vaddr)) 369 if (dma_release_from_coherent(dev, order, vaddr))
490 return; 370 return;
491 if (dma_ops->unmap_single) 371 if (ops->unmap_single)
492 dma_ops->unmap_single(dev, bus, size, 0); 372 ops->unmap_single(dev, bus, size, 0);
493 free_pages((unsigned long)vaddr, order); 373 free_pages((unsigned long)vaddr, order);
494} 374}
495EXPORT_SYMBOL(dma_free_coherent); 375EXPORT_SYMBOL(dma_free_coherent);
496 376
497static int __init pci_iommu_init(void) 377static int __init pci_iommu_init(void)
498{ 378{
499#ifdef CONFIG_CALGARY_IOMMU
500 calgary_iommu_init(); 379 calgary_iommu_init();
501#endif
502 380
503 intel_iommu_init(); 381 intel_iommu_init();
504 382
505#ifdef CONFIG_GART_IOMMU 383 amd_iommu_init();
384
506 gart_iommu_init(); 385 gart_iommu_init();
507#endif
508 386
509 no_iommu_init(); 387 no_iommu_init();
510 return 0; 388 return 0;
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index aa8ec928caa8..49285f8fd4d5 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -32,6 +32,7 @@
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>
35#include <asm/iommu.h>
35#include <asm/gart.h> 36#include <asm/gart.h>
36#include <asm/cacheflush.h> 37#include <asm/cacheflush.h>
37#include <asm/swiotlb.h> 38#include <asm/swiotlb.h>
@@ -66,9 +67,6 @@ static u32 gart_unmapped_entry;
66 (((x) & 0xfffff000) | (((x) >> 32) << 4) | GPTE_VALID | GPTE_COHERENT) 67 (((x) & 0xfffff000) | (((x) >> 32) << 4) | GPTE_VALID | GPTE_COHERENT)
67#define GPTE_DECODE(x) (((x) & 0xfffff000) | (((u64)(x) & 0xff0) << 28)) 68#define GPTE_DECODE(x) (((x) & 0xfffff000) | (((u64)(x) & 0xff0) << 28))
68 69
69#define to_pages(addr, size) \
70 (round_up(((addr) & ~PAGE_MASK) + (size), PAGE_SIZE) >> PAGE_SHIFT)
71
72#define EMERGENCY_PAGES 32 /* = 128KB */ 70#define EMERGENCY_PAGES 32 /* = 128KB */
73 71
74#ifdef CONFIG_AGP 72#ifdef CONFIG_AGP
@@ -104,7 +102,6 @@ static unsigned long alloc_iommu(struct device *dev, int size)
104 size, base_index, boundary_size, 0); 102 size, base_index, boundary_size, 0);
105 } 103 }
106 if (offset != -1) { 104 if (offset != -1) {
107 set_bit_string(iommu_gart_bitmap, offset, size);
108 next_bit = offset+size; 105 next_bit = offset+size;
109 if (next_bit >= iommu_pages) { 106 if (next_bit >= iommu_pages) {
110 next_bit = 0; 107 next_bit = 0;
@@ -198,9 +195,7 @@ static void iommu_full(struct device *dev, size_t size, int dir)
198 * out. Hopefully no network devices use single mappings that big. 195 * out. Hopefully no network devices use single mappings that big.
199 */ 196 */
200 197
201 printk(KERN_ERR 198 dev_err(dev, "PCI-DMA: Out of IOMMU space for %lu bytes\n", size);
202 "PCI-DMA: Out of IOMMU space for %lu bytes at device %s\n",
203 size, dev->bus_id);
204 199
205 if (size > PAGE_SIZE*EMERGENCY_PAGES) { 200 if (size > PAGE_SIZE*EMERGENCY_PAGES) {
206 if (dir == PCI_DMA_FROMDEVICE || dir == PCI_DMA_BIDIRECTIONAL) 201 if (dir == PCI_DMA_FROMDEVICE || dir == PCI_DMA_BIDIRECTIONAL)
@@ -243,7 +238,7 @@ nonforced_iommu(struct device *dev, unsigned long addr, size_t size)
243static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem, 238static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem,
244 size_t size, int dir) 239 size_t size, int dir)
245{ 240{
246 unsigned long npages = to_pages(phys_mem, size); 241 unsigned long npages = iommu_num_pages(phys_mem, size);
247 unsigned long iommu_page = alloc_iommu(dev, npages); 242 unsigned long iommu_page = alloc_iommu(dev, npages);
248 int i; 243 int i;
249 244
@@ -306,7 +301,7 @@ static void gart_unmap_single(struct device *dev, dma_addr_t dma_addr,
306 return; 301 return;
307 302
308 iommu_page = (dma_addr - iommu_bus_base)>>PAGE_SHIFT; 303 iommu_page = (dma_addr - iommu_bus_base)>>PAGE_SHIFT;
309 npages = to_pages(dma_addr, size); 304 npages = iommu_num_pages(dma_addr, size);
310 for (i = 0; i < npages; i++) { 305 for (i = 0; i < npages; i++) {
311 iommu_gatt_base[iommu_page + i] = gart_unmapped_entry; 306 iommu_gatt_base[iommu_page + i] = gart_unmapped_entry;
312 CLEAR_LEAK(iommu_page + i); 307 CLEAR_LEAK(iommu_page + i);
@@ -389,7 +384,7 @@ static int __dma_map_cont(struct device *dev, struct scatterlist *start,
389 } 384 }
390 385
391 addr = phys_addr; 386 addr = phys_addr;
392 pages = to_pages(s->offset, s->length); 387 pages = iommu_num_pages(s->offset, s->length);
393 while (pages--) { 388 while (pages--) {
394 iommu_gatt_base[iommu_page] = GPTE_ENCODE(addr); 389 iommu_gatt_base[iommu_page] = GPTE_ENCODE(addr);
395 SET_LEAK(iommu_page); 390 SET_LEAK(iommu_page);
@@ -472,7 +467,7 @@ gart_map_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
472 467
473 seg_size += s->length; 468 seg_size += s->length;
474 need = nextneed; 469 need = nextneed;
475 pages += to_pages(s->offset, s->length); 470 pages += iommu_num_pages(s->offset, s->length);
476 ps = s; 471 ps = s;
477 } 472 }
478 if (dma_map_cont(dev, start_sg, i - start, sgmap, pages, need) < 0) 473 if (dma_map_cont(dev, start_sg, i - start, sgmap, pages, need) < 0)
@@ -534,8 +529,8 @@ static __init unsigned read_aperture(struct pci_dev *dev, u32 *size)
534 unsigned aper_size = 0, aper_base_32, aper_order; 529 unsigned aper_size = 0, aper_base_32, aper_order;
535 u64 aper_base; 530 u64 aper_base;
536 531
537 pci_read_config_dword(dev, 0x94, &aper_base_32); 532 pci_read_config_dword(dev, AMD64_GARTAPERTUREBASE, &aper_base_32);
538 pci_read_config_dword(dev, 0x90, &aper_order); 533 pci_read_config_dword(dev, AMD64_GARTAPERTURECTL, &aper_order);
539 aper_order = (aper_order >> 1) & 7; 534 aper_order = (aper_order >> 1) & 7;
540 535
541 aper_base = aper_base_32 & 0x7fff; 536 aper_base = aper_base_32 & 0x7fff;
@@ -549,14 +544,63 @@ static __init unsigned read_aperture(struct pci_dev *dev, u32 *size)
549 return aper_base; 544 return aper_base;
550} 545}
551 546
547static void enable_gart_translations(void)
548{
549 int i;
550
551 for (i = 0; i < num_k8_northbridges; i++) {
552 struct pci_dev *dev = k8_northbridges[i];
553
554 enable_gart_translation(dev, __pa(agp_gatt_table));
555 }
556}
557
558/*
559 * If fix_up_north_bridges is set, the north bridges have to be fixed up on
560 * resume in the same way as they are handled in gart_iommu_hole_init().
561 */
562static bool fix_up_north_bridges;
563static u32 aperture_order;
564static u32 aperture_alloc;
565
566void set_up_gart_resume(u32 aper_order, u32 aper_alloc)
567{
568 fix_up_north_bridges = true;
569 aperture_order = aper_order;
570 aperture_alloc = aper_alloc;
571}
572
552static int gart_resume(struct sys_device *dev) 573static int gart_resume(struct sys_device *dev)
553{ 574{
575 printk(KERN_INFO "PCI-DMA: Resuming GART IOMMU\n");
576
577 if (fix_up_north_bridges) {
578 int i;
579
580 printk(KERN_INFO "PCI-DMA: Restoring GART aperture settings\n");
581
582 for (i = 0; i < num_k8_northbridges; i++) {
583 struct pci_dev *dev = k8_northbridges[i];
584
585 /*
586 * Don't enable translations just yet. That is the next
587 * step. Restore the pre-suspend aperture settings.
588 */
589 pci_write_config_dword(dev, AMD64_GARTAPERTURECTL,
590 aperture_order << 1);
591 pci_write_config_dword(dev, AMD64_GARTAPERTUREBASE,
592 aperture_alloc >> 25);
593 }
594 }
595
596 enable_gart_translations();
597
554 return 0; 598 return 0;
555} 599}
556 600
557static int gart_suspend(struct sys_device *dev, pm_message_t state) 601static int gart_suspend(struct sys_device *dev, pm_message_t state)
558{ 602{
559 return -EINVAL; 603 return 0;
560} 604}
561 605
562static struct sysdev_class gart_sysdev_class = { 606static struct sysdev_class gart_sysdev_class = {
@@ -582,6 +626,7 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
582 struct pci_dev *dev; 626 struct pci_dev *dev;
583 void *gatt; 627 void *gatt;
584 int i, error; 628 int i, error;
629 unsigned long start_pfn, end_pfn;
585 630
586 printk(KERN_INFO "PCI-DMA: Disabling AGP.\n"); 631 printk(KERN_INFO "PCI-DMA: Disabling AGP.\n");
587 aper_size = aper_base = info->aper_size = 0; 632 aper_size = aper_base = info->aper_size = 0;
@@ -614,31 +659,25 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
614 memset(gatt, 0, gatt_size); 659 memset(gatt, 0, gatt_size);
615 agp_gatt_table = gatt; 660 agp_gatt_table = gatt;
616 661
617 for (i = 0; i < num_k8_northbridges; i++) { 662 enable_gart_translations();
618 u32 gatt_reg;
619 u32 ctl;
620
621 dev = k8_northbridges[i];
622 gatt_reg = __pa(gatt) >> 12;
623 gatt_reg <<= 4;
624 pci_write_config_dword(dev, 0x98, gatt_reg);
625 pci_read_config_dword(dev, 0x90, &ctl);
626
627 ctl |= 1;
628 ctl &= ~((1<<4) | (1<<5));
629
630 pci_write_config_dword(dev, 0x90, ctl);
631 }
632 663
633 error = sysdev_class_register(&gart_sysdev_class); 664 error = sysdev_class_register(&gart_sysdev_class);
634 if (!error) 665 if (!error)
635 error = sysdev_register(&device_gart); 666 error = sysdev_register(&device_gart);
636 if (error) 667 if (error)
637 panic("Could not register gart_sysdev -- would corrupt data on next suspend"); 668 panic("Could not register gart_sysdev -- would corrupt data on next suspend");
669
638 flush_gart(); 670 flush_gart();
639 671
640 printk(KERN_INFO "PCI-DMA: aperture base @ %x size %u KB\n", 672 printk(KERN_INFO "PCI-DMA: aperture base @ %x size %u KB\n",
641 aper_base, aper_size>>10); 673 aper_base, aper_size>>10);
674
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 }
642 return 0; 681 return 0;
643 682
644 nommu: 683 nommu:
@@ -650,8 +689,7 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
650 689
651extern int agp_amd64_init(void); 690extern int agp_amd64_init(void);
652 691
653static const struct dma_mapping_ops gart_dma_ops = { 692static struct dma_mapping_ops gart_dma_ops = {
654 .mapping_error = NULL,
655 .map_single = gart_map_single, 693 .map_single = gart_map_single,
656 .map_simple = gart_map_simple, 694 .map_simple = gart_map_simple,
657 .unmap_single = gart_unmap_single, 695 .unmap_single = gart_unmap_single,
@@ -677,11 +715,11 @@ void gart_iommu_shutdown(void)
677 u32 ctl; 715 u32 ctl;
678 716
679 dev = k8_northbridges[i]; 717 dev = k8_northbridges[i];
680 pci_read_config_dword(dev, 0x90, &ctl); 718 pci_read_config_dword(dev, AMD64_GARTAPERTURECTL, &ctl);
681 719
682 ctl &= ~1; 720 ctl &= ~GARTEN;
683 721
684 pci_write_config_dword(dev, 0x90, ctl); 722 pci_write_config_dword(dev, AMD64_GARTAPERTURECTL, ctl);
685 } 723 }
686} 724}
687 725
@@ -716,10 +754,10 @@ void __init gart_iommu_init(void)
716 return; 754 return;
717 755
718 if (no_iommu || 756 if (no_iommu ||
719 (!force_iommu && end_pfn <= MAX_DMA32_PFN) || 757 (!force_iommu && max_pfn <= MAX_DMA32_PFN) ||
720 !gart_iommu_aperture || 758 !gart_iommu_aperture ||
721 (no_agp && init_k8_gatt(&info) < 0)) { 759 (no_agp && init_k8_gatt(&info) < 0)) {
722 if (end_pfn > MAX_DMA32_PFN) { 760 if (max_pfn > MAX_DMA32_PFN) {
723 printk(KERN_WARNING "More than 4GB of memory " 761 printk(KERN_WARNING "More than 4GB of memory "
724 "but GART IOMMU not available.\n" 762 "but GART IOMMU not available.\n"
725 KERN_WARNING "falling back to iommu=soft.\n"); 763 KERN_WARNING "falling back to iommu=soft.\n");
@@ -788,10 +826,10 @@ void __init gart_iommu_init(void)
788 wbinvd(); 826 wbinvd();
789 827
790 /* 828 /*
791 * Try to workaround a bug (thanks to BenH) 829 * Try to workaround a bug (thanks to BenH):
792 * Set unmapped entries to a scratch page instead of 0. 830 * Set unmapped entries to a scratch page instead of 0.
793 * Any prefetches that hit unmapped entries won't get an bus abort 831 * Any prefetches that hit unmapped entries won't get an bus abort
794 * then. 832 * then. (P2P bridge may be prefetching on DMA reads).
795 */ 833 */
796 scratch = get_zeroed_page(GFP_KERNEL); 834 scratch = get_zeroed_page(GFP_KERNEL);
797 if (!scratch) 835 if (!scratch)
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index aec43d56f49c..3f91f71cdc3e 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -7,7 +7,7 @@
7#include <linux/dma-mapping.h> 7#include <linux/dma-mapping.h>
8#include <linux/scatterlist.h> 8#include <linux/scatterlist.h>
9 9
10#include <asm/gart.h> 10#include <asm/iommu.h>
11#include <asm/processor.h> 11#include <asm/processor.h>
12#include <asm/dma.h> 12#include <asm/dma.h>
13 13
@@ -72,21 +72,9 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg,
72 return nents; 72 return nents;
73} 73}
74 74
75/* Make sure we keep the same behaviour */ 75struct dma_mapping_ops nommu_dma_ops = {
76static int nommu_mapping_error(dma_addr_t dma_addr)
77{
78#ifdef CONFIG_X86_32
79 return 0;
80#else
81 return (dma_addr == bad_dma_address);
82#endif
83}
84
85
86const struct dma_mapping_ops nommu_dma_ops = {
87 .map_single = nommu_map_single, 76 .map_single = nommu_map_single,
88 .map_sg = nommu_map_sg, 77 .map_sg = nommu_map_sg,
89 .mapping_error = nommu_mapping_error,
90 .is_phys = 1, 78 .is_phys = 1,
91}; 79};
92 80
diff --git a/arch/x86/kernel/pci-swiotlb_64.c b/arch/x86/kernel/pci-swiotlb_64.c
index 490da7f4b8d0..c4ce0332759e 100644
--- a/arch/x86/kernel/pci-swiotlb_64.c
+++ b/arch/x86/kernel/pci-swiotlb_64.c
@@ -5,7 +5,7 @@
5#include <linux/module.h> 5#include <linux/module.h>
6#include <linux/dma-mapping.h> 6#include <linux/dma-mapping.h>
7 7
8#include <asm/gart.h> 8#include <asm/iommu.h>
9#include <asm/swiotlb.h> 9#include <asm/swiotlb.h>
10#include <asm/dma.h> 10#include <asm/dma.h>
11 11
@@ -18,7 +18,7 @@ swiotlb_map_single_phys(struct device *hwdev, phys_addr_t paddr, size_t size,
18 return swiotlb_map_single(hwdev, phys_to_virt(paddr), size, direction); 18 return swiotlb_map_single(hwdev, phys_to_virt(paddr), size, direction);
19} 19}
20 20
21const struct dma_mapping_ops swiotlb_dma_ops = { 21struct dma_mapping_ops swiotlb_dma_ops = {
22 .mapping_error = swiotlb_dma_mapping_error, 22 .mapping_error = swiotlb_dma_mapping_error,
23 .alloc_coherent = swiotlb_alloc_coherent, 23 .alloc_coherent = swiotlb_alloc_coherent,
24 .free_coherent = swiotlb_free_coherent, 24 .free_coherent = swiotlb_free_coherent,
@@ -38,7 +38,7 @@ const struct dma_mapping_ops swiotlb_dma_ops = {
38void __init pci_swiotlb_init(void) 38void __init pci_swiotlb_init(void)
39{ 39{
40 /* don't initialize swiotlb if iommu=off (no_iommu=1) */ 40 /* don't initialize swiotlb if iommu=off (no_iommu=1) */
41 if (!iommu_detected && !no_iommu && end_pfn > MAX_DMA32_PFN) 41 if (!iommu_detected && !no_iommu && max_pfn > MAX_DMA32_PFN)
42 swiotlb = 1; 42 swiotlb = 1;
43 if (swiotlb_force) 43 if (swiotlb_force)
44 swiotlb = 1; 44 swiotlb = 1;
diff --git a/arch/x86/kernel/probe_roms_32.c b/arch/x86/kernel/probe_roms_32.c
new file mode 100644
index 000000000000..675a48c404a5
--- /dev/null
+++ b/arch/x86/kernel/probe_roms_32.c
@@ -0,0 +1,166 @@
1#include <linux/sched.h>
2#include <linux/mm.h>
3#include <linux/uaccess.h>
4#include <linux/mmzone.h>
5#include <linux/ioport.h>
6#include <linux/seq_file.h>
7#include <linux/console.h>
8#include <linux/init.h>
9#include <linux/edd.h>
10#include <linux/dmi.h>
11#include <linux/pfn.h>
12#include <linux/pci.h>
13#include <asm/pci-direct.h>
14
15
16#include <asm/e820.h>
17#include <asm/mmzone.h>
18#include <asm/setup.h>
19#include <asm/sections.h>
20#include <asm/io.h>
21#include <setup_arch.h>
22
23static struct resource system_rom_resource = {
24 .name = "System ROM",
25 .start = 0xf0000,
26 .end = 0xfffff,
27 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
28};
29
30static struct resource extension_rom_resource = {
31 .name = "Extension ROM",
32 .start = 0xe0000,
33 .end = 0xeffff,
34 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
35};
36
37static struct resource adapter_rom_resources[] = { {
38 .name = "Adapter ROM",
39 .start = 0xc8000,
40 .end = 0,
41 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
42}, {
43 .name = "Adapter ROM",
44 .start = 0,
45 .end = 0,
46 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
47}, {
48 .name = "Adapter ROM",
49 .start = 0,
50 .end = 0,
51 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
52}, {
53 .name = "Adapter ROM",
54 .start = 0,
55 .end = 0,
56 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
57}, {
58 .name = "Adapter ROM",
59 .start = 0,
60 .end = 0,
61 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
62}, {
63 .name = "Adapter ROM",
64 .start = 0,
65 .end = 0,
66 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
67} };
68
69static struct resource video_rom_resource = {
70 .name = "Video ROM",
71 .start = 0xc0000,
72 .end = 0xc7fff,
73 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
74};
75
76#define ROMSIGNATURE 0xaa55
77
78static int __init romsignature(const unsigned char *rom)
79{
80 const unsigned short * const ptr = (const unsigned short *)rom;
81 unsigned short sig;
82
83 return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
84}
85
86static int __init romchecksum(const unsigned char *rom, unsigned long length)
87{
88 unsigned char sum, c;
89
90 for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
91 sum += c;
92 return !length && !sum;
93}
94
95void __init probe_roms(void)
96{
97 const unsigned char *rom;
98 unsigned long start, length, upper;
99 unsigned char c;
100 int i;
101
102 /* video rom */
103 upper = adapter_rom_resources[0].start;
104 for (start = video_rom_resource.start; start < upper; start += 2048) {
105 rom = isa_bus_to_virt(start);
106 if (!romsignature(rom))
107 continue;
108
109 video_rom_resource.start = start;
110
111 if (probe_kernel_address(rom + 2, c) != 0)
112 continue;
113
114 /* 0 < length <= 0x7f * 512, historically */
115 length = c * 512;
116
117 /* if checksum okay, trust length byte */
118 if (length && romchecksum(rom, length))
119 video_rom_resource.end = start + length - 1;
120
121 request_resource(&iomem_resource, &video_rom_resource);
122 break;
123 }
124
125 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
126 if (start < upper)
127 start = upper;
128
129 /* system rom */
130 request_resource(&iomem_resource, &system_rom_resource);
131 upper = system_rom_resource.start;
132
133 /* check for extension rom (ignore length byte!) */
134 rom = isa_bus_to_virt(extension_rom_resource.start);
135 if (romsignature(rom)) {
136 length = extension_rom_resource.end - extension_rom_resource.start + 1;
137 if (romchecksum(rom, length)) {
138 request_resource(&iomem_resource, &extension_rom_resource);
139 upper = extension_rom_resource.start;
140 }
141 }
142
143 /* check for adapter roms on 2k boundaries */
144 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
145 rom = isa_bus_to_virt(start);
146 if (!romsignature(rom))
147 continue;
148
149 if (probe_kernel_address(rom + 2, c) != 0)
150 continue;
151
152 /* 0 < length <= 0x7f * 512, historically */
153 length = c * 512;
154
155 /* but accept any length that fits if checksum okay */
156 if (!length || start + length > upper || !romchecksum(rom, length))
157 continue;
158
159 adapter_rom_resources[i].start = start;
160 adapter_rom_resources[i].end = start + length - 1;
161 request_resource(&iomem_resource, &adapter_rom_resources[i]);
162
163 start = adapter_rom_resources[i++].end & ~2047UL;
164 }
165}
166
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ba370dc8685b..7fc4d5b0a6a0 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -6,8 +6,16 @@
6#include <linux/sched.h> 6#include <linux/sched.h>
7#include <linux/module.h> 7#include <linux/module.h>
8#include <linux/pm.h> 8#include <linux/pm.h>
9#include <linux/clockchips.h>
10#include <asm/system.h>
11
12unsigned long idle_halt;
13EXPORT_SYMBOL(idle_halt);
14unsigned long idle_nomwait;
15EXPORT_SYMBOL(idle_nomwait);
9 16
10struct kmem_cache *task_xstate_cachep; 17struct kmem_cache *task_xstate_cachep;
18static int force_mwait __cpuinitdata;
11 19
12int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 20int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
13{ 21{
@@ -45,6 +53,76 @@ void arch_task_cache_init(void)
45 SLAB_PANIC, NULL); 53 SLAB_PANIC, NULL);
46} 54}
47 55
56/*
57 * Idle related variables and functions
58 */
59unsigned long boot_option_idle_override = 0;
60EXPORT_SYMBOL(boot_option_idle_override);
61
62/*
63 * Powermanagement idle function, if any..
64 */
65void (*pm_idle)(void);
66EXPORT_SYMBOL(pm_idle);
67
68#ifdef CONFIG_X86_32
69/*
70 * This halt magic was a workaround for ancient floppy DMA
71 * wreckage. It should be safe to remove.
72 */
73static int hlt_counter;
74void disable_hlt(void)
75{
76 hlt_counter++;
77}
78EXPORT_SYMBOL(disable_hlt);
79
80void enable_hlt(void)
81{
82 hlt_counter--;
83}
84EXPORT_SYMBOL(enable_hlt);
85
86static inline int hlt_use_halt(void)
87{
88 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
89}
90#else
91static inline int hlt_use_halt(void)
92{
93 return 1;
94}
95#endif
96
97/*
98 * We use this if we don't have any better
99 * idle routine..
100 */
101void default_idle(void)
102{
103 if (hlt_use_halt()) {
104 current_thread_info()->status &= ~TS_POLLING;
105 /*
106 * TS_POLLING-cleared state must be visible before we
107 * test NEED_RESCHED:
108 */
109 smp_mb();
110
111 if (!need_resched())
112 safe_halt(); /* enables interrupts racelessly */
113 else
114 local_irq_enable();
115 current_thread_info()->status |= TS_POLLING;
116 } else {
117 local_irq_enable();
118 /* loop is done by the caller */
119 cpu_relax();
120 }
121}
122#ifdef CONFIG_APM_MODULE
123EXPORT_SYMBOL(default_idle);
124#endif
125
48static void do_nothing(void *unused) 126static void do_nothing(void *unused)
49{ 127{
50} 128}
@@ -61,7 +139,7 @@ void cpu_idle_wait(void)
61{ 139{
62 smp_mb(); 140 smp_mb();
63 /* kick all the CPUs so that they exit out of pm_idle */ 141 /* kick all the CPUs so that they exit out of pm_idle */
64 smp_call_function(do_nothing, NULL, 0, 1); 142 smp_call_function(do_nothing, NULL, 1);
65} 143}
66EXPORT_SYMBOL_GPL(cpu_idle_wait); 144EXPORT_SYMBOL_GPL(cpu_idle_wait);
67 145
@@ -122,54 +200,163 @@ static void poll_idle(void)
122 * 200 *
123 * idle=mwait overrides this decision and forces the usage of mwait. 201 * idle=mwait overrides this decision and forces the usage of mwait.
124 */ 202 */
203static int __cpuinitdata force_mwait;
204
205#define MWAIT_INFO 0x05
206#define MWAIT_ECX_EXTENDED_INFO 0x01
207#define MWAIT_EDX_C1 0xf0
208
125static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c) 209static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
126{ 210{
211 u32 eax, ebx, ecx, edx;
212
127 if (force_mwait) 213 if (force_mwait)
128 return 1; 214 return 1;
129 215
130 if (c->x86_vendor == X86_VENDOR_AMD) { 216 if (c->cpuid_level < MWAIT_INFO)
131 switch(c->x86) { 217 return 0;
132 case 0x10: 218
133 case 0x11: 219 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
134 return 0; 220 /* Check, whether EDX has extended info about MWAIT */
135 } 221 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
136 } 222 return 1;
223
224 /*
225 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
226 * C1 supports MWAIT
227 */
228 return (edx & MWAIT_EDX_C1);
229}
230
231/*
232 * Check for AMD CPUs, which have potentially C1E support
233 */
234static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
235{
236 if (c->x86_vendor != X86_VENDOR_AMD)
237 return 0;
238
239 if (c->x86 < 0x0F)
240 return 0;
241
242 /* Family 0x0f models < rev F do not have C1E */
243 if (c->x86 == 0x0f && c->x86_model < 0x40)
244 return 0;
245
137 return 1; 246 return 1;
138} 247}
139 248
140void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) 249/*
250 * 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
252 * way as C3 power states (local apic timer and TSC stop)
253 */
254static void c1e_idle(void)
141{ 255{
142 static int selected; 256 static cpumask_t c1e_mask = CPU_MASK_NONE;
257 static int c1e_detected;
143 258
144 if (selected) 259 if (need_resched())
145 return; 260 return;
261
262 if (!c1e_detected) {
263 u32 lo, hi;
264
265 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
266 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
267 c1e_detected = 1;
268 mark_tsc_unstable("TSC halt in C1E");
269 printk(KERN_INFO "System has C1E enabled\n");
270 }
271 }
272
273 if (c1e_detected) {
274 int cpu = smp_processor_id();
275
276 if (!cpu_isset(cpu, c1e_mask)) {
277 cpu_set(cpu, c1e_mask);
278 /*
279 * Force broadcast so ACPI can not interfere. Needs
280 * to run with interrupts enabled as it uses
281 * smp_function_call.
282 */
283 local_irq_enable();
284 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
285 &cpu);
286 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
287 cpu);
288 local_irq_disable();
289 }
290 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
291
292 default_idle();
293
294 /*
295 * The switch back from broadcast mode needs to be
296 * called with interrupts disabled.
297 */
298 local_irq_disable();
299 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
300 local_irq_enable();
301 } else
302 default_idle();
303}
304
305void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
306{
146#ifdef CONFIG_X86_SMP 307#ifdef CONFIG_X86_SMP
147 if (pm_idle == poll_idle && smp_num_siblings > 1) { 308 if (pm_idle == poll_idle && smp_num_siblings > 1) {
148 printk(KERN_WARNING "WARNING: polling idle and HT enabled," 309 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
149 " performance may degrade.\n"); 310 " performance may degrade.\n");
150 } 311 }
151#endif 312#endif
313 if (pm_idle)
314 return;
315
152 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) { 316 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
153 /* 317 /*
154 * Skip, if setup has overridden idle.
155 * One CPU supports mwait => All CPUs supports mwait 318 * One CPU supports mwait => All CPUs supports mwait
156 */ 319 */
157 if (!pm_idle) { 320 printk(KERN_INFO "using mwait in idle threads.\n");
158 printk(KERN_INFO "using mwait in idle threads.\n"); 321 pm_idle = mwait_idle;
159 pm_idle = mwait_idle; 322 } else if (check_c1e_idle(c)) {
160 } 323 printk(KERN_INFO "using C1E aware idle routine\n");
161 } 324 pm_idle = c1e_idle;
162 selected = 1; 325 } else
326 pm_idle = default_idle;
163} 327}
164 328
165static int __init idle_setup(char *str) 329static int __init idle_setup(char *str)
166{ 330{
331 if (!str)
332 return -EINVAL;
333
167 if (!strcmp(str, "poll")) { 334 if (!strcmp(str, "poll")) {
168 printk("using polling idle threads.\n"); 335 printk("using polling idle threads.\n");
169 pm_idle = poll_idle; 336 pm_idle = poll_idle;
170 } else if (!strcmp(str, "mwait")) 337 } else if (!strcmp(str, "mwait"))
171 force_mwait = 1; 338 force_mwait = 1;
172 else 339 else if (!strcmp(str, "halt")) {
340 /*
341 * When the boot option of idle=halt is added, halt is
342 * forced to be used for CPU idle. In such case CPU C2/C3
343 * won't be used again.
344 * To continue to load the CPU idle driver, don't touch
345 * the boot_option_idle_override.
346 */
347 pm_idle = default_idle;
348 idle_halt = 1;
349 return 0;
350 } else if (!strcmp(str, "nomwait")) {
351 /*
352 * If the boot option of "idle=nomwait" is added,
353 * it means that mwait will be disabled for CPU C2/C3
354 * states. In such case it won't touch the variable
355 * of boot_option_idle_override.
356 */
357 idle_nomwait = 1;
358 return 0;
359 } else
173 return -1; 360 return -1;
174 361
175 boot_option_idle_override = 1; 362 boot_option_idle_override = 1;
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 6d5483356e74..53bc653ed5ca 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -58,11 +58,6 @@
58 58
59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
60 60
61static int hlt_counter;
62
63unsigned long boot_option_idle_override = 0;
64EXPORT_SYMBOL(boot_option_idle_override);
65
66DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; 61DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
67EXPORT_PER_CPU_SYMBOL(current_task); 62EXPORT_PER_CPU_SYMBOL(current_task);
68 63
@@ -77,57 +72,24 @@ unsigned long thread_saved_pc(struct task_struct *tsk)
77 return ((unsigned long *)tsk->thread.sp)[3]; 72 return ((unsigned long *)tsk->thread.sp)[3];
78} 73}
79 74
80/* 75#ifdef CONFIG_HOTPLUG_CPU
81 * Powermanagement idle function, if any.. 76#include <asm/nmi.h>
82 */
83void (*pm_idle)(void);
84EXPORT_SYMBOL(pm_idle);
85 77
86void disable_hlt(void) 78static void cpu_exit_clear(void)
87{ 79{
88 hlt_counter++; 80 int cpu = raw_smp_processor_id();
89}
90 81
91EXPORT_SYMBOL(disable_hlt); 82 idle_task_exit();
92 83
93void enable_hlt(void) 84 cpu_uninit();
94{ 85 irq_ctx_exit(cpu);
95 hlt_counter--;
96}
97 86
98EXPORT_SYMBOL(enable_hlt); 87 cpu_clear(cpu, cpu_callout_map);
99 88 cpu_clear(cpu, cpu_callin_map);
100/*
101 * We use this if we don't have any better
102 * idle routine..
103 */
104void default_idle(void)
105{
106 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
107 current_thread_info()->status &= ~TS_POLLING;
108 /*
109 * TS_POLLING-cleared state must be visible before we
110 * test NEED_RESCHED:
111 */
112 smp_mb();
113 89
114 if (!need_resched()) 90 numa_remove_cpu(cpu);
115 safe_halt(); /* enables interrupts racelessly */
116 else
117 local_irq_enable();
118 current_thread_info()->status |= TS_POLLING;
119 } else {
120 local_irq_enable();
121 /* loop is done by the caller */
122 cpu_relax();
123 }
124} 91}
125#ifdef CONFIG_APM_MODULE
126EXPORT_SYMBOL(default_idle);
127#endif
128 92
129#ifdef CONFIG_HOTPLUG_CPU
130#include <asm/nmi.h>
131/* We don't actually take CPU down, just spin without interrupts. */ 93/* We don't actually take CPU down, just spin without interrupts. */
132static inline void play_dead(void) 94static inline void play_dead(void)
133{ 95{
@@ -166,26 +128,24 @@ void cpu_idle(void)
166 128
167 /* endless idle loop with no priority at all */ 129 /* endless idle loop with no priority at all */
168 while (1) { 130 while (1) {
169 tick_nohz_stop_sched_tick(); 131 tick_nohz_stop_sched_tick(1);
170 while (!need_resched()) { 132 while (!need_resched()) {
171 void (*idle)(void);
172 133
173 check_pgt_cache(); 134 check_pgt_cache();
174 rmb(); 135 rmb();
175 idle = pm_idle;
176 136
177 if (rcu_pending(cpu)) 137 if (rcu_pending(cpu))
178 rcu_check_callbacks(cpu, 0); 138 rcu_check_callbacks(cpu, 0);
179 139
180 if (!idle)
181 idle = default_idle;
182
183 if (cpu_is_offline(cpu)) 140 if (cpu_is_offline(cpu))
184 play_dead(); 141 play_dead();
185 142
186 local_irq_disable(); 143 local_irq_disable();
187 __get_cpu_var(irq_stat).idle_timestamp = jiffies; 144 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
188 idle(); 145 /* Don't trace irqs off for idle */
146 stop_critical_timings();
147 pm_idle();
148 start_critical_timings();
189 } 149 }
190 tick_nohz_restart_sched_tick(); 150 tick_nohz_restart_sched_tick();
191 preempt_enable_no_resched(); 151 preempt_enable_no_resched();
@@ -333,6 +293,7 @@ void flush_thread(void)
333 /* 293 /*
334 * Forget coprocessor state.. 294 * Forget coprocessor state..
335 */ 295 */
296 tsk->fpu_counter = 0;
336 clear_fpu(tsk); 297 clear_fpu(tsk);
337 clear_used_math(); 298 clear_used_math();
338} 299}
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index ac54ff56df80..3fb62a7d9a16 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -56,15 +56,6 @@ asmlinkage extern void ret_from_fork(void);
56 56
57unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED; 57unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
58 58
59unsigned long boot_option_idle_override = 0;
60EXPORT_SYMBOL(boot_option_idle_override);
61
62/*
63 * Powermanagement idle function, if any..
64 */
65void (*pm_idle)(void);
66EXPORT_SYMBOL(pm_idle);
67
68static ATOMIC_NOTIFIER_HEAD(idle_notifier); 59static ATOMIC_NOTIFIER_HEAD(idle_notifier);
69 60
70void idle_notifier_register(struct notifier_block *n) 61void idle_notifier_register(struct notifier_block *n)
@@ -94,25 +85,6 @@ void exit_idle(void)
94 __exit_idle(); 85 __exit_idle();
95} 86}
96 87
97/*
98 * We use this if we don't have any better
99 * idle routine..
100 */
101void default_idle(void)
102{
103 current_thread_info()->status &= ~TS_POLLING;
104 /*
105 * TS_POLLING-cleared state must be visible before we
106 * test NEED_RESCHED:
107 */
108 smp_mb();
109 if (!need_resched())
110 safe_halt(); /* enables interrupts racelessly */
111 else
112 local_irq_enable();
113 current_thread_info()->status |= TS_POLLING;
114}
115
116#ifdef CONFIG_HOTPLUG_CPU 88#ifdef CONFIG_HOTPLUG_CPU
117DECLARE_PER_CPU(int, cpu_state); 89DECLARE_PER_CPU(int, cpu_state);
118 90
@@ -148,14 +120,11 @@ void cpu_idle(void)
148 current_thread_info()->status |= TS_POLLING; 120 current_thread_info()->status |= TS_POLLING;
149 /* endless idle loop with no priority at all */ 121 /* endless idle loop with no priority at all */
150 while (1) { 122 while (1) {
151 tick_nohz_stop_sched_tick(); 123 tick_nohz_stop_sched_tick(1);
152 while (!need_resched()) { 124 while (!need_resched()) {
153 void (*idle)(void);
154 125
155 rmb(); 126 rmb();
156 idle = pm_idle; 127
157 if (!idle)
158 idle = default_idle;
159 if (cpu_is_offline(smp_processor_id())) 128 if (cpu_is_offline(smp_processor_id()))
160 play_dead(); 129 play_dead();
161 /* 130 /*
@@ -165,7 +134,10 @@ void cpu_idle(void)
165 */ 134 */
166 local_irq_disable(); 135 local_irq_disable();
167 enter_idle(); 136 enter_idle();
168 idle(); 137 /* Don't trace irqs off for idle */
138 stop_critical_timings();
139 pm_idle();
140 start_critical_timings();
169 /* In many cases the interrupt that ended idle 141 /* In many cases the interrupt that ended idle
170 has already called exit_idle. But some idle 142 has already called exit_idle. But some idle
171 loops can be woken up without interrupt. */ 143 loops can be woken up without interrupt. */
@@ -294,6 +266,7 @@ void flush_thread(void)
294 /* 266 /*
295 * Forget coprocessor state.. 267 * Forget coprocessor state..
296 */ 268 */
269 tsk->fpu_counter = 0;
297 clear_fpu(tsk); 270 clear_fpu(tsk);
298 clear_used_math(); 271 clear_used_math();
299} 272}
@@ -365,10 +338,10 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
365 p->thread.fs = me->thread.fs; 338 p->thread.fs = me->thread.fs;
366 p->thread.gs = me->thread.gs; 339 p->thread.gs = me->thread.gs;
367 340
368 asm("mov %%gs,%0" : "=m" (p->thread.gsindex)); 341 savesegment(gs, p->thread.gsindex);
369 asm("mov %%fs,%0" : "=m" (p->thread.fsindex)); 342 savesegment(fs, p->thread.fsindex);
370 asm("mov %%es,%0" : "=m" (p->thread.es)); 343 savesegment(es, p->thread.es);
371 asm("mov %%ds,%0" : "=m" (p->thread.ds)); 344 savesegment(ds, p->thread.ds);
372 345
373 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { 346 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
374 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); 347 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
@@ -407,7 +380,9 @@ out:
407void 380void
408start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 381start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
409{ 382{
410 asm volatile("movl %0, %%fs; movl %0, %%es; movl %0, %%ds" :: "r"(0)); 383 loadsegment(fs, 0);
384 loadsegment(es, 0);
385 loadsegment(ds, 0);
411 load_gs_index(0); 386 load_gs_index(0);
412 regs->ip = new_ip; 387 regs->ip = new_ip;
413 regs->sp = new_sp; 388 regs->sp = new_sp;
@@ -562,10 +537,11 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
562struct task_struct * 537struct task_struct *
563__switch_to(struct task_struct *prev_p, struct task_struct *next_p) 538__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
564{ 539{
565 struct thread_struct *prev = &prev_p->thread, 540 struct thread_struct *prev = &prev_p->thread;
566 *next = &next_p->thread; 541 struct thread_struct *next = &next_p->thread;
567 int cpu = smp_processor_id(); 542 int cpu = smp_processor_id();
568 struct tss_struct *tss = &per_cpu(init_tss, cpu); 543 struct tss_struct *tss = &per_cpu(init_tss, cpu);
544 unsigned fsindex, gsindex;
569 545
570 /* we're going to use this soon, after a few expensive things */ 546 /* we're going to use this soon, after a few expensive things */
571 if (next_p->fpu_counter>5) 547 if (next_p->fpu_counter>5)
@@ -580,52 +556,64 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
580 * Switch DS and ES. 556 * Switch DS and ES.
581 * This won't pick up thread selector changes, but I guess that is ok. 557 * This won't pick up thread selector changes, but I guess that is ok.
582 */ 558 */
583 asm volatile("mov %%es,%0" : "=m" (prev->es)); 559 savesegment(es, prev->es);
584 if (unlikely(next->es | prev->es)) 560 if (unlikely(next->es | prev->es))
585 loadsegment(es, next->es); 561 loadsegment(es, next->es);
586 562
587 asm volatile ("mov %%ds,%0" : "=m" (prev->ds)); 563 savesegment(ds, prev->ds);
588 if (unlikely(next->ds | prev->ds)) 564 if (unlikely(next->ds | prev->ds))
589 loadsegment(ds, next->ds); 565 loadsegment(ds, next->ds);
590 566
567
568 /* We must save %fs and %gs before load_TLS() because
569 * %fs and %gs may be cleared by load_TLS().
570 *
571 * (e.g. xen_load_tls())
572 */
573 savesegment(fs, fsindex);
574 savesegment(gs, gsindex);
575
591 load_TLS(next, cpu); 576 load_TLS(next, cpu);
592 577
578 /*
579 * Leave lazy mode, flushing any hypercalls made here.
580 * This must be done before restoring TLS segments so
581 * the GDT and LDT are properly updated, and must be
582 * done before math_state_restore, so the TS bit is up
583 * to date.
584 */
585 arch_leave_lazy_cpu_mode();
586
593 /* 587 /*
594 * Switch FS and GS. 588 * Switch FS and GS.
589 *
590 * Segment register != 0 always requires a reload. Also
591 * reload when it has changed. When prev process used 64bit
592 * base always reload to avoid an information leak.
595 */ 593 */
596 { 594 if (unlikely(fsindex | next->fsindex | prev->fs)) {
597 unsigned fsindex; 595 loadsegment(fs, next->fsindex);
598 asm volatile("movl %%fs,%0" : "=r" (fsindex)); 596 /*
599 /* segment register != 0 always requires a reload. 597 * Check if the user used a selector != 0; if yes
600 also reload when it has changed. 598 * clear 64bit base, since overloaded base is always
601 when prev process used 64bit base always reload 599 * mapped to the Null selector
602 to avoid an information leak. */ 600 */
603 if (unlikely(fsindex | next->fsindex | prev->fs)) { 601 if (fsindex)
604 loadsegment(fs, next->fsindex);
605 /* check if the user used a selector != 0
606 * if yes clear 64bit base, since overloaded base
607 * is always mapped to the Null selector
608 */
609 if (fsindex)
610 prev->fs = 0; 602 prev->fs = 0;
611 }
612 /* when next process has a 64bit base use it */
613 if (next->fs)
614 wrmsrl(MSR_FS_BASE, next->fs);
615 prev->fsindex = fsindex;
616 } 603 }
617 { 604 /* when next process has a 64bit base use it */
618 unsigned gsindex; 605 if (next->fs)
619 asm volatile("movl %%gs,%0" : "=r" (gsindex)); 606 wrmsrl(MSR_FS_BASE, next->fs);
620 if (unlikely(gsindex | next->gsindex | prev->gs)) { 607 prev->fsindex = fsindex;
621 load_gs_index(next->gsindex); 608
622 if (gsindex) 609 if (unlikely(gsindex | next->gsindex | prev->gs)) {
610 load_gs_index(next->gsindex);
611 if (gsindex)
623 prev->gs = 0; 612 prev->gs = 0;
624 }
625 if (next->gs)
626 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
627 prev->gsindex = gsindex;
628 } 613 }
614 if (next->gs)
615 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
616 prev->gsindex = gsindex;
629 617
630 /* Must be after DS reload */ 618 /* Must be after DS reload */
631 unlazy_fpu(prev_p); 619 unlazy_fpu(prev_p);
@@ -638,7 +626,8 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
638 write_pda(pcurrent, next_p); 626 write_pda(pcurrent, next_p);
639 627
640 write_pda(kernelstack, 628 write_pda(kernelstack,
641 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET); 629 (unsigned long)task_stack_page(next_p) +
630 THREAD_SIZE - PDA_STACKOFFSET);
642#ifdef CONFIG_CC_STACKPROTECTOR 631#ifdef CONFIG_CC_STACKPROTECTOR
643 write_pda(stack_canary, next_p->stack_canary); 632 write_pda(stack_canary, next_p->stack_canary);
644 /* 633 /*
@@ -797,7 +786,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
797 set_32bit_tls(task, FS_TLS, addr); 786 set_32bit_tls(task, FS_TLS, addr);
798 if (doit) { 787 if (doit) {
799 load_TLS(&task->thread, cpu); 788 load_TLS(&task->thread, cpu);
800 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL)); 789 loadsegment(fs, FS_TLS_SEL);
801 } 790 }
802 task->thread.fsindex = FS_TLS_SEL; 791 task->thread.fsindex = FS_TLS_SEL;
803 task->thread.fs = 0; 792 task->thread.fs = 0;
@@ -807,7 +796,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
807 if (doit) { 796 if (doit) {
808 /* set the selector to 0 to not confuse 797 /* set the selector to 0 to not confuse
809 __switch_to */ 798 __switch_to */
810 asm volatile("movl %0,%%fs" :: "r" (0)); 799 loadsegment(fs, 0);
811 ret = checking_wrmsrl(MSR_FS_BASE, addr); 800 ret = checking_wrmsrl(MSR_FS_BASE, addr);
812 } 801 }
813 } 802 }
@@ -830,7 +819,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
830 if (task->thread.gsindex == GS_TLS_SEL) 819 if (task->thread.gsindex == GS_TLS_SEL)
831 base = read_32bit_tls(task, GS_TLS); 820 base = read_32bit_tls(task, GS_TLS);
832 else if (doit) { 821 else if (doit) {
833 asm("movl %%gs,%0" : "=r" (gsindex)); 822 savesegment(gs, gsindex);
834 if (gsindex) 823 if (gsindex)
835 rdmsrl(MSR_KERNEL_GS_BASE, base); 824 rdmsrl(MSR_KERNEL_GS_BASE, base);
836 else 825 else
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index a7835f282936..e37dccce85db 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -943,13 +943,13 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
943 return copy_regset_to_user(child, &user_x86_32_view, 943 return copy_regset_to_user(child, &user_x86_32_view,
944 REGSET_XFP, 944 REGSET_XFP,
945 0, sizeof(struct user_fxsr_struct), 945 0, sizeof(struct user_fxsr_struct),
946 datap); 946 datap) ? -EIO : 0;
947 947
948 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */ 948 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
949 return copy_regset_from_user(child, &user_x86_32_view, 949 return copy_regset_from_user(child, &user_x86_32_view,
950 REGSET_XFP, 950 REGSET_XFP,
951 0, sizeof(struct user_fxsr_struct), 951 0, sizeof(struct user_fxsr_struct),
952 datap); 952 datap) ? -EIO : 0;
953#endif 953#endif
954 954
955#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION 955#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
@@ -1357,8 +1357,6 @@ const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1357#endif 1357#endif
1358} 1358}
1359 1359
1360#ifdef CONFIG_X86_32
1361
1362void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code) 1360void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1363{ 1361{
1364 struct siginfo info; 1362 struct siginfo info;
@@ -1377,89 +1375,10 @@ void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1377 force_sig_info(SIGTRAP, &info, tsk); 1375 force_sig_info(SIGTRAP, &info, tsk);
1378} 1376}
1379 1377
1380/* notification of system call entry/exit
1381 * - triggered by current->work.syscall_trace
1382 */
1383int do_syscall_trace(struct pt_regs *regs, int entryexit)
1384{
1385 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
1386 /*
1387 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
1388 * interception
1389 */
1390 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
1391 int ret = 0;
1392
1393 /* do the secure computing check first */
1394 if (!entryexit)
1395 secure_computing(regs->orig_ax);
1396
1397 if (unlikely(current->audit_context)) {
1398 if (entryexit)
1399 audit_syscall_exit(AUDITSC_RESULT(regs->ax),
1400 regs->ax);
1401 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
1402 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
1403 * not used, entry.S will call us only on syscall exit, not
1404 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
1405 * calling send_sigtrap() on syscall entry.
1406 *
1407 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
1408 * is_singlestep is false, despite his name, so we will still do
1409 * the correct thing.
1410 */
1411 else if (is_singlestep)
1412 goto out;
1413 }
1414
1415 if (!(current->ptrace & PT_PTRACED))
1416 goto out;
1417
1418 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
1419 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
1420 * here. We have to check this and return */
1421 if (is_sysemu && entryexit)
1422 return 0;
1423
1424 /* Fake a debug trap */
1425 if (is_singlestep)
1426 send_sigtrap(current, regs, 0);
1427
1428 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
1429 goto out;
1430
1431 /* the 0x80 provides a way for the tracing parent to distinguish
1432 between a syscall stop and SIGTRAP delivery */
1433 /* Note that the debugger could change the result of test_thread_flag!*/
1434 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1435
1436 /*
1437 * this isn't the same as continuing with a signal, but it will do
1438 * for normal use. strace only continues with a signal if the
1439 * stopping signal is not SIGTRAP. -brl
1440 */
1441 if (current->exit_code) {
1442 send_sig(current->exit_code, current, 1);
1443 current->exit_code = 0;
1444 }
1445 ret = is_sysemu;
1446out:
1447 if (unlikely(current->audit_context) && !entryexit)
1448 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
1449 regs->bx, regs->cx, regs->dx, regs->si);
1450 if (ret == 0)
1451 return 0;
1452
1453 regs->orig_ax = -1; /* force skip of syscall restarting */
1454 if (unlikely(current->audit_context))
1455 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1456 return 1;
1457}
1458
1459#else /* CONFIG_X86_64 */
1460
1461static void syscall_trace(struct pt_regs *regs) 1378static void syscall_trace(struct pt_regs *regs)
1462{ 1379{
1380 if (!(current->ptrace & PT_PTRACED))
1381 return;
1463 1382
1464#if 0 1383#if 0
1465 printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n", 1384 printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
@@ -1481,39 +1400,81 @@ static void syscall_trace(struct pt_regs *regs)
1481 } 1400 }
1482} 1401}
1483 1402
1484asmlinkage void syscall_trace_enter(struct pt_regs *regs) 1403#ifdef CONFIG_X86_32
1404# define IS_IA32 1
1405#elif defined CONFIG_IA32_EMULATION
1406# define IS_IA32 test_thread_flag(TIF_IA32)
1407#else
1408# define IS_IA32 0
1409#endif
1410
1411/*
1412 * We must return the syscall number to actually look up in the table.
1413 * This can be -1L to skip running any syscall at all.
1414 */
1415asmregparm long syscall_trace_enter(struct pt_regs *regs)
1485{ 1416{
1417 long ret = 0;
1418
1419 /*
1420 * If we stepped into a sysenter/syscall insn, it trapped in
1421 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1422 * If user-mode had set TF itself, then it's still clear from
1423 * do_debug() and we need to set it again to restore the user
1424 * state. If we entered on the slow path, TF was already set.
1425 */
1426 if (test_thread_flag(TIF_SINGLESTEP))
1427 regs->flags |= X86_EFLAGS_TF;
1428
1486 /* do the secure computing check first */ 1429 /* do the secure computing check first */
1487 secure_computing(regs->orig_ax); 1430 secure_computing(regs->orig_ax);
1488 1431
1489 if (test_thread_flag(TIF_SYSCALL_TRACE) 1432 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1490 && (current->ptrace & PT_PTRACED)) 1433 ret = -1L;
1434
1435 if (ret || test_thread_flag(TIF_SYSCALL_TRACE))
1491 syscall_trace(regs); 1436 syscall_trace(regs);
1492 1437
1493 if (unlikely(current->audit_context)) { 1438 if (unlikely(current->audit_context)) {
1494 if (test_thread_flag(TIF_IA32)) { 1439 if (IS_IA32)
1495 audit_syscall_entry(AUDIT_ARCH_I386, 1440 audit_syscall_entry(AUDIT_ARCH_I386,
1496 regs->orig_ax, 1441 regs->orig_ax,
1497 regs->bx, regs->cx, 1442 regs->bx, regs->cx,
1498 regs->dx, regs->si); 1443 regs->dx, regs->si);
1499 } else { 1444#ifdef CONFIG_X86_64
1445 else
1500 audit_syscall_entry(AUDIT_ARCH_X86_64, 1446 audit_syscall_entry(AUDIT_ARCH_X86_64,
1501 regs->orig_ax, 1447 regs->orig_ax,
1502 regs->di, regs->si, 1448 regs->di, regs->si,
1503 regs->dx, regs->r10); 1449 regs->dx, regs->r10);
1504 } 1450#endif
1505 } 1451 }
1452
1453 return ret ?: regs->orig_ax;
1506} 1454}
1507 1455
1508asmlinkage void syscall_trace_leave(struct pt_regs *regs) 1456asmregparm void syscall_trace_leave(struct pt_regs *regs)
1509{ 1457{
1510 if (unlikely(current->audit_context)) 1458 if (unlikely(current->audit_context))
1511 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax); 1459 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1512 1460
1513 if ((test_thread_flag(TIF_SYSCALL_TRACE) 1461 if (test_thread_flag(TIF_SYSCALL_TRACE))
1514 || test_thread_flag(TIF_SINGLESTEP))
1515 && (current->ptrace & PT_PTRACED))
1516 syscall_trace(regs); 1462 syscall_trace(regs);
1517}
1518 1463
1519#endif /* CONFIG_X86_32 */ 1464 /*
1465 * If TIF_SYSCALL_EMU is set, we only get here because of
1466 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1467 * We already reported this syscall instruction in
1468 * syscall_trace_enter(), so don't do any more now.
1469 */
1470 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1471 return;
1472
1473 /*
1474 * If we are single-stepping, synthesize a trap to follow the
1475 * system call instruction.
1476 */
1477 if (test_thread_flag(TIF_SINGLESTEP) &&
1478 (current->ptrace & PT_PTRACED))
1479 send_sigtrap(current, regs, 0);
1480}
diff --git a/arch/x86/kernel/pvclock.c b/arch/x86/kernel/pvclock.c
new file mode 100644
index 000000000000..05fbe9a0325a
--- /dev/null
+++ b/arch/x86/kernel/pvclock.c
@@ -0,0 +1,141 @@
1/* paravirtual clock -- common code used by kvm/xen
2
3 This program is free software; you can redistribute it and/or modify
4 it under the terms of the GNU General Public License as published by
5 the Free Software Foundation; either version 2 of the License, or
6 (at your option) any later version.
7
8 This program is distributed in the hope that it will be useful,
9 but WITHOUT ANY WARRANTY; without even the implied warranty of
10 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 GNU General Public License for more details.
12
13 You should have received a copy of the GNU General Public License
14 along with this program; if not, write to the Free Software
15 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
16*/
17
18#include <linux/kernel.h>
19#include <linux/percpu.h>
20#include <asm/pvclock.h>
21
22/*
23 * These are perodically updated
24 * xen: magic shared_info page
25 * kvm: gpa registered via msr
26 * and then copied here.
27 */
28struct pvclock_shadow_time {
29 u64 tsc_timestamp; /* TSC at last update of time vals. */
30 u64 system_timestamp; /* Time, in nanosecs, since boot. */
31 u32 tsc_to_nsec_mul;
32 int tsc_shift;
33 u32 version;
34};
35
36/*
37 * Scale a 64-bit delta by scaling and multiplying by a 32-bit fraction,
38 * yielding a 64-bit result.
39 */
40static inline u64 scale_delta(u64 delta, u32 mul_frac, int shift)
41{
42 u64 product;
43#ifdef __i386__
44 u32 tmp1, tmp2;
45#endif
46
47 if (shift < 0)
48 delta >>= -shift;
49 else
50 delta <<= shift;
51
52#ifdef __i386__
53 __asm__ (
54 "mul %5 ; "
55 "mov %4,%%eax ; "
56 "mov %%edx,%4 ; "
57 "mul %5 ; "
58 "xor %5,%5 ; "
59 "add %4,%%eax ; "
60 "adc %5,%%edx ; "
61 : "=A" (product), "=r" (tmp1), "=r" (tmp2)
62 : "a" ((u32)delta), "1" ((u32)(delta >> 32)), "2" (mul_frac) );
63#elif __x86_64__
64 __asm__ (
65 "mul %%rdx ; shrd $32,%%rdx,%%rax"
66 : "=a" (product) : "0" (delta), "d" ((u64)mul_frac) );
67#else
68#error implement me!
69#endif
70
71 return product;
72}
73
74static u64 pvclock_get_nsec_offset(struct pvclock_shadow_time *shadow)
75{
76 u64 delta = native_read_tsc() - shadow->tsc_timestamp;
77 return scale_delta(delta, shadow->tsc_to_nsec_mul, shadow->tsc_shift);
78}
79
80/*
81 * Reads a consistent set of time-base values from hypervisor,
82 * into a shadow data area.
83 */
84static unsigned pvclock_get_time_values(struct pvclock_shadow_time *dst,
85 struct pvclock_vcpu_time_info *src)
86{
87 do {
88 dst->version = src->version;
89 rmb(); /* fetch version before data */
90 dst->tsc_timestamp = src->tsc_timestamp;
91 dst->system_timestamp = src->system_time;
92 dst->tsc_to_nsec_mul = src->tsc_to_system_mul;
93 dst->tsc_shift = src->tsc_shift;
94 rmb(); /* test version after fetching data */
95 } while ((src->version & 1) || (dst->version != src->version));
96
97 return dst->version;
98}
99
100cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
101{
102 struct pvclock_shadow_time shadow;
103 unsigned version;
104 cycle_t ret, offset;
105
106 do {
107 version = pvclock_get_time_values(&shadow, src);
108 barrier();
109 offset = pvclock_get_nsec_offset(&shadow);
110 ret = shadow.system_timestamp + offset;
111 barrier();
112 } while (version != src->version);
113
114 return ret;
115}
116
117void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
118 struct pvclock_vcpu_time_info *vcpu_time,
119 struct timespec *ts)
120{
121 u32 version;
122 u64 delta;
123 struct timespec now;
124
125 /* get wallclock at system boot */
126 do {
127 version = wall_clock->version;
128 rmb(); /* fetch version before time */
129 now.tv_sec = wall_clock->sec;
130 now.tv_nsec = wall_clock->nsec;
131 rmb(); /* fetch time before checking version */
132 } while ((wall_clock->version & 1) || (version != wall_clock->version));
133
134 delta = pvclock_clocksource_read(vcpu_time); /* time since system boot */
135 delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec;
136
137 now.tv_nsec = do_div(delta, NSEC_PER_SEC);
138 now.tv_sec = delta;
139
140 set_normalized_timespec(ts, now.tv_sec, now.tv_nsec);
141}
diff --git a/arch/x86/kernel/quirks.c b/arch/x86/kernel/quirks.c
index f327abafe3e6..d13858818100 100644
--- a/arch/x86/kernel/quirks.c
+++ b/arch/x86/kernel/quirks.c
@@ -65,6 +65,7 @@ static enum {
65 ICH_FORCE_HPET_RESUME, 65 ICH_FORCE_HPET_RESUME,
66 VT8237_FORCE_HPET_RESUME, 66 VT8237_FORCE_HPET_RESUME,
67 NVIDIA_FORCE_HPET_RESUME, 67 NVIDIA_FORCE_HPET_RESUME,
68 ATI_FORCE_HPET_RESUME,
68} force_hpet_resume_type; 69} force_hpet_resume_type;
69 70
70static void __iomem *rcba_base; 71static void __iomem *rcba_base;
@@ -176,6 +177,12 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ICH9_7,
176 177
177static struct pci_dev *cached_dev; 178static struct pci_dev *cached_dev;
178 179
180static void hpet_print_force_info(void)
181{
182 printk(KERN_INFO "HPET not enabled in BIOS. "
183 "You might try hpet=force boot option\n");
184}
185
179static void old_ich_force_hpet_resume(void) 186static void old_ich_force_hpet_resume(void)
180{ 187{
181 u32 val; 188 u32 val;
@@ -255,6 +262,8 @@ static void old_ich_force_enable_hpet_user(struct pci_dev *dev)
255{ 262{
256 if (hpet_force_user) 263 if (hpet_force_user)
257 old_ich_force_enable_hpet(dev); 264 old_ich_force_enable_hpet(dev);
265 else
266 hpet_print_force_info();
258} 267}
259 268
260DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ESB_1, 269DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ESB_1,
@@ -294,8 +303,13 @@ static void vt8237_force_enable_hpet(struct pci_dev *dev)
294{ 303{
295 u32 uninitialized_var(val); 304 u32 uninitialized_var(val);
296 305
297 if (!hpet_force_user || hpet_address || force_hpet_address) 306 if (hpet_address || force_hpet_address)
307 return;
308
309 if (!hpet_force_user) {
310 hpet_print_force_info();
298 return; 311 return;
312 }
299 313
300 pci_read_config_dword(dev, 0x68, &val); 314 pci_read_config_dword(dev, 0x68, &val);
301 /* 315 /*
@@ -334,6 +348,36 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_8235,
334DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_8237, 348DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_8237,
335 vt8237_force_enable_hpet); 349 vt8237_force_enable_hpet);
336 350
351static void ati_force_hpet_resume(void)
352{
353 pci_write_config_dword(cached_dev, 0x14, 0xfed00000);
354 printk(KERN_DEBUG "Force enabled HPET at resume\n");
355}
356
357static void ati_force_enable_hpet(struct pci_dev *dev)
358{
359 u32 uninitialized_var(val);
360
361 if (hpet_address || force_hpet_address)
362 return;
363
364 if (!hpet_force_user) {
365 hpet_print_force_info();
366 return;
367 }
368
369 pci_write_config_dword(dev, 0x14, 0xfed00000);
370 pci_read_config_dword(dev, 0x14, &val);
371 force_hpet_address = val;
372 force_hpet_resume_type = ATI_FORCE_HPET_RESUME;
373 dev_printk(KERN_DEBUG, &dev->dev, "Force enabled HPET at 0x%lx\n",
374 force_hpet_address);
375 cached_dev = dev;
376 return;
377}
378DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_ATI, PCI_DEVICE_ID_ATI_IXP400_SMBUS,
379 ati_force_enable_hpet);
380
337/* 381/*
338 * Undocumented chipset feature taken from LinuxBIOS. 382 * Undocumented chipset feature taken from LinuxBIOS.
339 */ 383 */
@@ -347,9 +391,14 @@ static void nvidia_force_enable_hpet(struct pci_dev *dev)
347{ 391{
348 u32 uninitialized_var(val); 392 u32 uninitialized_var(val);
349 393
350 if (!hpet_force_user || hpet_address || force_hpet_address) 394 if (hpet_address || force_hpet_address)
351 return; 395 return;
352 396
397 if (!hpet_force_user) {
398 hpet_print_force_info();
399 return;
400 }
401
353 pci_write_config_dword(dev, 0x44, 0xfed00001); 402 pci_write_config_dword(dev, 0x44, 0xfed00001);
354 pci_read_config_dword(dev, 0x44, &val); 403 pci_read_config_dword(dev, 0x44, &val);
355 force_hpet_address = val & 0xfffffffe; 404 force_hpet_address = val & 0xfffffffe;
@@ -401,6 +450,9 @@ void force_hpet_resume(void)
401 case NVIDIA_FORCE_HPET_RESUME: 450 case NVIDIA_FORCE_HPET_RESUME:
402 nvidia_force_hpet_resume(); 451 nvidia_force_hpet_resume();
403 return; 452 return;
453 case ATI_FORCE_HPET_RESUME:
454 ati_force_hpet_resume();
455 return;
404 default: 456 default:
405 break; 457 break;
406 } 458 }
diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c
index f6be7d5f82f8..724adfc63cb9 100644
--- a/arch/x86/kernel/reboot.c
+++ b/arch/x86/kernel/reboot.c
@@ -27,7 +27,7 @@
27void (*pm_power_off)(void); 27void (*pm_power_off)(void);
28EXPORT_SYMBOL(pm_power_off); 28EXPORT_SYMBOL(pm_power_off);
29 29
30static long no_idt[3]; 30static const struct desc_ptr no_idt = {};
31static int reboot_mode; 31static int reboot_mode;
32enum reboot_type reboot_type = BOOT_KBD; 32enum reboot_type reboot_type = BOOT_KBD;
33int reboot_force; 33int reboot_force;
@@ -177,6 +177,14 @@ static struct dmi_system_id __initdata reboot_dmi_table[] = {
177 DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 2400"), 177 DMI_MATCH(DMI_PRODUCT_NAME, "PowerEdge 2400"),
178 }, 178 },
179 }, 179 },
180 { /* Handle problems with rebooting on Dell T5400's */
181 .callback = set_bios_reboot,
182 .ident = "Dell Precision T5400",
183 .matches = {
184 DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."),
185 DMI_MATCH(DMI_PRODUCT_NAME, "Precision WorkStation T5400"),
186 },
187 },
180 { /* Handle problems with rebooting on HP laptops */ 188 { /* Handle problems with rebooting on HP laptops */
181 .callback = set_bios_reboot, 189 .callback = set_bios_reboot,
182 .ident = "HP Compaq Laptop", 190 .ident = "HP Compaq Laptop",
@@ -201,15 +209,15 @@ core_initcall(reboot_init);
201 controller to pulse the CPU reset line, which is more thorough, but 209 controller to pulse the CPU reset line, which is more thorough, but
202 doesn't work with at least one type of 486 motherboard. It is easy 210 doesn't work with at least one type of 486 motherboard. It is easy
203 to stop this code working; hence the copious comments. */ 211 to stop this code working; hence the copious comments. */
204static unsigned long long 212static const unsigned long long
205real_mode_gdt_entries [3] = 213real_mode_gdt_entries [3] =
206{ 214{
207 0x0000000000000000ULL, /* Null descriptor */ 215 0x0000000000000000ULL, /* Null descriptor */
208 0x00009a000000ffffULL, /* 16-bit real-mode 64k code at 0x00000000 */ 216 0x00009b000000ffffULL, /* 16-bit real-mode 64k code at 0x00000000 */
209 0x000092000100ffffULL /* 16-bit real-mode 64k data at 0x00000100 */ 217 0x000093000100ffffULL /* 16-bit real-mode 64k data at 0x00000100 */
210}; 218};
211 219
212static struct desc_ptr 220static const struct desc_ptr
213real_mode_gdt = { sizeof (real_mode_gdt_entries) - 1, (long)real_mode_gdt_entries }, 221real_mode_gdt = { sizeof (real_mode_gdt_entries) - 1, (long)real_mode_gdt_entries },
214real_mode_idt = { 0x3ff, 0 }; 222real_mode_idt = { 0x3ff, 0 };
215 223
@@ -231,7 +239,7 @@ real_mode_idt = { 0x3ff, 0 };
231 239
232 More could be done here to set up the registers as if a CPU reset had 240 More could be done here to set up the registers as if a CPU reset had
233 occurred; hopefully real BIOSs don't assume much. */ 241 occurred; hopefully real BIOSs don't assume much. */
234static unsigned char real_mode_switch [] = 242static const unsigned char real_mode_switch [] =
235{ 243{
236 0x66, 0x0f, 0x20, 0xc0, /* movl %cr0,%eax */ 244 0x66, 0x0f, 0x20, 0xc0, /* movl %cr0,%eax */
237 0x66, 0x83, 0xe0, 0x11, /* andl $0x00000011,%eax */ 245 0x66, 0x83, 0xe0, 0x11, /* andl $0x00000011,%eax */
@@ -245,7 +253,7 @@ static unsigned char real_mode_switch [] =
245 0x24, 0x10, /* f: andb $0x10,al */ 253 0x24, 0x10, /* f: andb $0x10,al */
246 0x66, 0x0f, 0x22, 0xc0 /* movl %eax,%cr0 */ 254 0x66, 0x0f, 0x22, 0xc0 /* movl %eax,%cr0 */
247}; 255};
248static unsigned char jump_to_bios [] = 256static const unsigned char jump_to_bios [] =
249{ 257{
250 0xea, 0x00, 0x00, 0xff, 0xff /* ljmp $0xffff,$0x0000 */ 258 0xea, 0x00, 0x00, 0xff, 0xff /* ljmp $0xffff,$0x0000 */
251}; 259};
@@ -255,7 +263,7 @@ static unsigned char jump_to_bios [] =
255 * specified by the code and length parameters. 263 * specified by the code and length parameters.
256 * We assume that length will aways be less that 100! 264 * We assume that length will aways be less that 100!
257 */ 265 */
258void machine_real_restart(unsigned char *code, int length) 266void machine_real_restart(const unsigned char *code, int length)
259{ 267{
260 local_irq_disable(); 268 local_irq_disable();
261 269
@@ -368,7 +376,7 @@ static void native_machine_emergency_restart(void)
368 } 376 }
369 377
370 case BOOT_TRIPLE: 378 case BOOT_TRIPLE:
371 load_idt((const struct desc_ptr *)&no_idt); 379 load_idt(&no_idt);
372 __asm__ __volatile__("int3"); 380 __asm__ __volatile__("int3");
373 381
374 reboot_type = BOOT_KBD; 382 reboot_type = BOOT_KBD;
@@ -403,10 +411,9 @@ void native_machine_shutdown(void)
403{ 411{
404 /* Stop the cpus and apics */ 412 /* Stop the cpus and apics */
405#ifdef CONFIG_SMP 413#ifdef CONFIG_SMP
406 int reboot_cpu_id;
407 414
408 /* The boot cpu is always logical cpu 0 */ 415 /* The boot cpu is always logical cpu 0 */
409 reboot_cpu_id = 0; 416 int reboot_cpu_id = 0;
410 417
411#ifdef CONFIG_X86_32 418#ifdef CONFIG_X86_32
412 /* See if there has been given a command line override */ 419 /* See if there has been given a command line override */
diff --git a/arch/x86/kernel/reboot_fixups_32.c b/arch/x86/kernel/reboot_fixups_32.c
index dec0b5ec25c2..61a837743fe5 100644
--- a/arch/x86/kernel/reboot_fixups_32.c
+++ b/arch/x86/kernel/reboot_fixups_32.c
@@ -49,7 +49,7 @@ struct device_fixup {
49 void (*reboot_fixup)(struct pci_dev *); 49 void (*reboot_fixup)(struct pci_dev *);
50}; 50};
51 51
52static struct device_fixup fixups_table[] = { 52static const struct device_fixup fixups_table[] = {
53{ PCI_VENDOR_ID_CYRIX, PCI_DEVICE_ID_CYRIX_5530_LEGACY, cs5530a_warm_reset }, 53{ PCI_VENDOR_ID_CYRIX, PCI_DEVICE_ID_CYRIX_5530_LEGACY, cs5530a_warm_reset },
54{ PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_CS5536_ISA, cs5536_warm_reset }, 54{ PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_CS5536_ISA, cs5536_warm_reset },
55{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SC1100_BRIDGE, cs5530a_warm_reset }, 55{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SC1100_BRIDGE, cs5530a_warm_reset },
@@ -64,7 +64,7 @@ static struct device_fixup fixups_table[] = {
64 */ 64 */
65void mach_reboot_fixups(void) 65void mach_reboot_fixups(void)
66{ 66{
67 struct device_fixup *cur; 67 const struct device_fixup *cur;
68 struct pci_dev *dev; 68 struct pci_dev *dev;
69 int i; 69 int i;
70 70
diff --git a/arch/x86/kernel/relocate_kernel_32.S b/arch/x86/kernel/relocate_kernel_32.S
index c30fe25d470d..703310a99023 100644
--- a/arch/x86/kernel/relocate_kernel_32.S
+++ b/arch/x86/kernel/relocate_kernel_32.S
@@ -20,11 +20,44 @@
20#define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY) 20#define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
21#define PAE_PGD_ATTR (_PAGE_PRESENT) 21#define PAE_PGD_ATTR (_PAGE_PRESENT)
22 22
23/* control_page + PAGE_SIZE/2 ~ control_page + PAGE_SIZE * 3/4 are
24 * used to save some data for jumping back
25 */
26#define DATA(offset) (PAGE_SIZE/2+(offset))
27
28/* Minimal CPU state */
29#define ESP DATA(0x0)
30#define CR0 DATA(0x4)
31#define CR3 DATA(0x8)
32#define CR4 DATA(0xc)
33
34/* other data */
35#define CP_VA_CONTROL_PAGE DATA(0x10)
36#define CP_PA_PGD DATA(0x14)
37#define CP_PA_SWAP_PAGE DATA(0x18)
38#define CP_PA_BACKUP_PAGES_MAP DATA(0x1c)
39
23 .text 40 .text
24 .align PAGE_SIZE 41 .align PAGE_SIZE
25 .globl relocate_kernel 42 .globl relocate_kernel
26relocate_kernel: 43relocate_kernel:
27 movl 8(%esp), %ebp /* list of pages */ 44 /* Save the CPU context, used for jumping back */
45
46 pushl %ebx
47 pushl %esi
48 pushl %edi
49 pushl %ebp
50 pushf
51
52 movl 20+8(%esp), %ebp /* list of pages */
53 movl PTR(VA_CONTROL_PAGE)(%ebp), %edi
54 movl %esp, ESP(%edi)
55 movl %cr0, %eax
56 movl %eax, CR0(%edi)
57 movl %cr3, %eax
58 movl %eax, CR3(%edi)
59 movl %cr4, %eax
60 movl %eax, CR4(%edi)
28 61
29#ifdef CONFIG_X86_PAE 62#ifdef CONFIG_X86_PAE
30 /* map the control page at its virtual address */ 63 /* map the control page at its virtual address */
@@ -138,15 +171,25 @@ relocate_kernel:
138 171
139relocate_new_kernel: 172relocate_new_kernel:
140 /* read the arguments and say goodbye to the stack */ 173 /* read the arguments and say goodbye to the stack */
141 movl 4(%esp), %ebx /* page_list */ 174 movl 20+4(%esp), %ebx /* page_list */
142 movl 8(%esp), %ebp /* list of pages */ 175 movl 20+8(%esp), %ebp /* list of pages */
143 movl 12(%esp), %edx /* start address */ 176 movl 20+12(%esp), %edx /* start address */
144 movl 16(%esp), %ecx /* cpu_has_pae */ 177 movl 20+16(%esp), %ecx /* cpu_has_pae */
178 movl 20+20(%esp), %esi /* preserve_context */
145 179
146 /* zero out flags, and disable interrupts */ 180 /* zero out flags, and disable interrupts */
147 pushl $0 181 pushl $0
148 popfl 182 popfl
149 183
184 /* save some information for jumping back */
185 movl PTR(VA_CONTROL_PAGE)(%ebp), %edi
186 movl %edi, CP_VA_CONTROL_PAGE(%edi)
187 movl PTR(PA_PGD)(%ebp), %eax
188 movl %eax, CP_PA_PGD(%edi)
189 movl PTR(PA_SWAP_PAGE)(%ebp), %eax
190 movl %eax, CP_PA_SWAP_PAGE(%edi)
191 movl %ebx, CP_PA_BACKUP_PAGES_MAP(%edi)
192
150 /* get physical address of control page now */ 193 /* get physical address of control page now */
151 /* this is impossible after page table switch */ 194 /* this is impossible after page table switch */
152 movl PTR(PA_CONTROL_PAGE)(%ebp), %edi 195 movl PTR(PA_CONTROL_PAGE)(%ebp), %edi
@@ -197,8 +240,90 @@ identity_mapped:
197 xorl %eax, %eax 240 xorl %eax, %eax
198 movl %eax, %cr3 241 movl %eax, %cr3
199 242
243 movl CP_PA_SWAP_PAGE(%edi), %eax
244 pushl %eax
245 pushl %ebx
246 call swap_pages
247 addl $8, %esp
248
249 /* To be certain of avoiding problems with self-modifying code
250 * I need to execute a serializing instruction here.
251 * So I flush the TLB, it's handy, and not processor dependent.
252 */
253 xorl %eax, %eax
254 movl %eax, %cr3
255
256 /* set all of the registers to known values */
257 /* leave %esp alone */
258
259 testl %esi, %esi
260 jnz 1f
261 xorl %edi, %edi
262 xorl %eax, %eax
263 xorl %ebx, %ebx
264 xorl %ecx, %ecx
265 xorl %edx, %edx
266 xorl %esi, %esi
267 xorl %ebp, %ebp
268 ret
2691:
270 popl %edx
271 movl CP_PA_SWAP_PAGE(%edi), %esp
272 addl $PAGE_SIZE, %esp
2732:
274 call *%edx
275
276 /* get the re-entry point of the peer system */
277 movl 0(%esp), %ebp
278 call 1f
2791:
280 popl %ebx
281 subl $(1b - relocate_kernel), %ebx
282 movl CP_VA_CONTROL_PAGE(%ebx), %edi
283 lea PAGE_SIZE(%ebx), %esp
284 movl CP_PA_SWAP_PAGE(%ebx), %eax
285 movl CP_PA_BACKUP_PAGES_MAP(%ebx), %edx
286 pushl %eax
287 pushl %edx
288 call swap_pages
289 addl $8, %esp
290 movl CP_PA_PGD(%ebx), %eax
291 movl %eax, %cr3
292 movl %cr0, %eax
293 orl $(1<<31), %eax
294 movl %eax, %cr0
295 lea PAGE_SIZE(%edi), %esp
296 movl %edi, %eax
297 addl $(virtual_mapped - relocate_kernel), %eax
298 pushl %eax
299 ret
300
301virtual_mapped:
302 movl CR4(%edi), %eax
303 movl %eax, %cr4
304 movl CR3(%edi), %eax
305 movl %eax, %cr3
306 movl CR0(%edi), %eax
307 movl %eax, %cr0
308 movl ESP(%edi), %esp
309 movl %ebp, %eax
310
311 popf
312 popl %ebp
313 popl %edi
314 popl %esi
315 popl %ebx
316 ret
317
200 /* Do the copies */ 318 /* Do the copies */
201 movl %ebx, %ecx 319swap_pages:
320 movl 8(%esp), %edx
321 movl 4(%esp), %ecx
322 pushl %ebp
323 pushl %ebx
324 pushl %edi
325 pushl %esi
326 movl %ecx, %ebx
202 jmp 1f 327 jmp 1f
203 328
2040: /* top, read another word from the indirection page */ 3290: /* top, read another word from the indirection page */
@@ -226,27 +351,28 @@ identity_mapped:
226 movl %ecx, %esi /* For every source page do a copy */ 351 movl %ecx, %esi /* For every source page do a copy */
227 andl $0xfffff000, %esi 352 andl $0xfffff000, %esi
228 353
354 movl %edi, %eax
355 movl %esi, %ebp
356
357 movl %edx, %edi
229 movl $1024, %ecx 358 movl $1024, %ecx
230 rep ; movsl 359 rep ; movsl
231 jmp 0b
232 360
2333: 361 movl %ebp, %edi
234 362 movl %eax, %esi
235 /* To be certain of avoiding problems with self-modifying code 363 movl $1024, %ecx
236 * I need to execute a serializing instruction here. 364 rep ; movsl
237 * So I flush the TLB, it's handy, and not processor dependent.
238 */
239 xorl %eax, %eax
240 movl %eax, %cr3
241 365
242 /* set all of the registers to known values */ 366 movl %eax, %edi
243 /* leave %esp alone */ 367 movl %edx, %esi
368 movl $1024, %ecx
369 rep ; movsl
244 370
245 xorl %eax, %eax 371 lea PAGE_SIZE(%ebp), %esi
246 xorl %ebx, %ebx 372 jmp 0b
247 xorl %ecx, %ecx 3733:
248 xorl %edx, %edx 374 popl %esi
249 xorl %esi, %esi 375 popl %edi
250 xorl %edi, %edi 376 popl %ebx
251 xorl %ebp, %ebp 377 popl %ebp
252 ret 378 ret
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 6f80b852a196..b520dae02bf4 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1,139 +1,885 @@
1#include <linux/kernel.h> 1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/screen_info.h>
28#include <linux/ioport.h>
29#include <linux/acpi.h>
30#include <linux/apm_bios.h>
31#include <linux/initrd.h>
32#include <linux/bootmem.h>
33#include <linux/seq_file.h>
34#include <linux/console.h>
35#include <linux/mca.h>
36#include <linux/root_dev.h>
37#include <linux/highmem.h>
2#include <linux/module.h> 38#include <linux/module.h>
39#include <linux/efi.h>
3#include <linux/init.h> 40#include <linux/init.h>
4#include <linux/bootmem.h> 41#include <linux/edd.h>
42#include <linux/iscsi_ibft.h>
43#include <linux/nodemask.h>
44#include <linux/kexec.h>
45#include <linux/dmi.h>
46#include <linux/pfn.h>
47#include <linux/pci.h>
48#include <asm/pci-direct.h>
49#include <linux/init_ohci1394_dma.h>
50#include <linux/kvm_para.h>
51
52#include <linux/errno.h>
53#include <linux/kernel.h>
54#include <linux/stddef.h>
55#include <linux/unistd.h>
56#include <linux/ptrace.h>
57#include <linux/slab.h>
58#include <linux/user.h>
59#include <linux/delay.h>
60
61#include <linux/kallsyms.h>
62#include <linux/cpufreq.h>
63#include <linux/dma-mapping.h>
64#include <linux/ctype.h>
65#include <linux/uaccess.h>
66
5#include <linux/percpu.h> 67#include <linux/percpu.h>
6#include <asm/smp.h> 68#include <linux/crash_dump.h>
7#include <asm/percpu.h> 69
70#include <video/edid.h>
71
72#include <asm/mtrr.h>
73#include <asm/apic.h>
74#include <asm/e820.h>
75#include <asm/mpspec.h>
76#include <asm/setup.h>
77#include <asm/arch_hooks.h>
78#include <asm/efi.h>
8#include <asm/sections.h> 79#include <asm/sections.h>
80#include <asm/dmi.h>
81#include <asm/io_apic.h>
82#include <asm/ist.h>
83#include <asm/vmi.h>
84#include <setup_arch.h>
85#include <asm/bios_ebda.h>
86#include <asm/cacheflush.h>
9#include <asm/processor.h> 87#include <asm/processor.h>
10#include <asm/setup.h> 88#include <asm/bugs.h>
89
90#include <asm/system.h>
91#include <asm/vsyscall.h>
92#include <asm/smp.h>
93#include <asm/desc.h>
94#include <asm/dma.h>
95#include <asm/iommu.h>
96#include <asm/mmu_context.h>
97#include <asm/proto.h>
98
99#include <mach_apic.h>
100#include <asm/paravirt.h>
101
102#include <asm/percpu.h>
11#include <asm/topology.h> 103#include <asm/topology.h>
12#include <asm/mpspec.h>
13#include <asm/apicdef.h> 104#include <asm/apicdef.h>
105#ifdef CONFIG_X86_64
106#include <asm/numa_64.h>
107#endif
14 108
15#ifdef CONFIG_X86_LOCAL_APIC 109#ifndef ARCH_SETUP
16unsigned int num_processors; 110#define ARCH_SETUP
17unsigned disabled_cpus __cpuinitdata; 111#endif
18/* Processor that is doing the boot up */
19unsigned int boot_cpu_physical_apicid = -1U;
20EXPORT_SYMBOL(boot_cpu_physical_apicid);
21 112
22DEFINE_PER_CPU(u16, x86_cpu_to_apicid) = BAD_APICID; 113#ifndef CONFIG_DEBUG_BOOT_PARAMS
23EXPORT_PER_CPU_SYMBOL(x86_cpu_to_apicid); 114struct boot_params __initdata boot_params;
115#else
116struct boot_params boot_params;
117#endif
24 118
25/* Bitmask of physically existing CPUs */ 119/*
26physid_mask_t phys_cpu_present_map; 120 * Machine setup..
121 */
122static struct resource data_resource = {
123 .name = "Kernel data",
124 .start = 0,
125 .end = 0,
126 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
127};
128
129static struct resource code_resource = {
130 .name = "Kernel code",
131 .start = 0,
132 .end = 0,
133 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
134};
135
136static struct resource bss_resource = {
137 .name = "Kernel bss",
138 .start = 0,
139 .end = 0,
140 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
141};
142
143
144#ifdef CONFIG_X86_32
145/* This value is set up by the early boot code to point to the value
146 immediately after the boot time page tables. It contains a *physical*
147 address, and must not be in the .bss segment! */
148unsigned long init_pg_tables_start __initdata = ~0UL;
149unsigned long init_pg_tables_end __initdata = ~0UL;
150
151static struct resource video_ram_resource = {
152 .name = "Video RAM area",
153 .start = 0xa0000,
154 .end = 0xbffff,
155 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
156};
157
158/* cpu data as detected by the assembly code in head.S */
159struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
160/* common cpu data for all cpus */
161struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
162EXPORT_SYMBOL(boot_cpu_data);
163static void set_mca_bus(int x)
164{
165#ifdef CONFIG_MCA
166 MCA_bus = x;
167#endif
168}
169
170unsigned int def_to_bigsmp;
171
172/* for MCA, but anyone else can use it if they want */
173unsigned int machine_id;
174unsigned int machine_submodel_id;
175unsigned int BIOS_revision;
176
177struct apm_info apm_info;
178EXPORT_SYMBOL(apm_info);
179
180#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
181 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
182struct ist_info ist_info;
183EXPORT_SYMBOL(ist_info);
184#else
185struct ist_info ist_info;
27#endif 186#endif
28 187
29#if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP) 188#else
189struct cpuinfo_x86 boot_cpu_data __read_mostly;
190EXPORT_SYMBOL(boot_cpu_data);
191#endif
192
193
194#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
195unsigned long mmu_cr4_features;
196#else
197unsigned long mmu_cr4_features = X86_CR4_PAE;
198#endif
199
200/* Boot loader ID as an integer, for the benefit of proc_dointvec */
201int bootloader_type;
202
30/* 203/*
31 * Copy data used in early init routines from the initial arrays to the 204 * Early DMI memory
32 * per cpu data areas. These arrays then become expendable and the
33 * *_early_ptr's are zeroed indicating that the static arrays are gone.
34 */ 205 */
35static void __init setup_per_cpu_maps(void) 206int dmi_alloc_index;
207char dmi_alloc_data[DMI_MAX_DATA];
208
209/*
210 * Setup options
211 */
212struct screen_info screen_info;
213EXPORT_SYMBOL(screen_info);
214struct edid_info edid_info;
215EXPORT_SYMBOL_GPL(edid_info);
216
217extern int root_mountflags;
218
219unsigned long saved_video_mode;
220
221#define RAMDISK_IMAGE_START_MASK 0x07FF
222#define RAMDISK_PROMPT_FLAG 0x8000
223#define RAMDISK_LOAD_FLAG 0x4000
224
225static char __initdata command_line[COMMAND_LINE_SIZE];
226
227#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
228struct edd edd;
229#ifdef CONFIG_EDD_MODULE
230EXPORT_SYMBOL(edd);
231#endif
232/**
233 * copy_edd() - Copy the BIOS EDD information
234 * from boot_params into a safe place.
235 *
236 */
237static inline void copy_edd(void)
238{
239 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
240 sizeof(edd.mbr_signature));
241 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
242 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
243 edd.edd_info_nr = boot_params.eddbuf_entries;
244}
245#else
246static inline void copy_edd(void)
247{
248}
249#endif
250
251#ifdef CONFIG_BLK_DEV_INITRD
252
253#ifdef CONFIG_X86_32
254
255#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
256static void __init relocate_initrd(void)
36{ 257{
37 int cpu;
38 258
39 for_each_possible_cpu(cpu) { 259 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
40 per_cpu(x86_cpu_to_apicid, cpu) = x86_cpu_to_apicid_init[cpu]; 260 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
41 per_cpu(x86_bios_cpu_apicid, cpu) = 261 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
42 x86_bios_cpu_apicid_init[cpu]; 262 u64 ramdisk_here;
43#ifdef CONFIG_NUMA 263 unsigned long slop, clen, mapaddr;
44 per_cpu(x86_cpu_to_node_map, cpu) = 264 char *p, *q;
45 x86_cpu_to_node_map_init[cpu]; 265
266 /* We need to move the initrd down into lowmem */
267 ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
268 PAGE_SIZE);
269
270 if (ramdisk_here == -1ULL)
271 panic("Cannot find place for new RAMDISK of size %lld\n",
272 ramdisk_size);
273
274 /* Note: this includes all the lowmem currently occupied by
275 the initrd, we rely on that fact to keep the data intact. */
276 reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
277 "NEW RAMDISK");
278 initrd_start = ramdisk_here + PAGE_OFFSET;
279 initrd_end = initrd_start + ramdisk_size;
280 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
281 ramdisk_here, ramdisk_here + ramdisk_size);
282
283 q = (char *)initrd_start;
284
285 /* Copy any lowmem portion of the initrd */
286 if (ramdisk_image < end_of_lowmem) {
287 clen = end_of_lowmem - ramdisk_image;
288 p = (char *)__va(ramdisk_image);
289 memcpy(q, p, clen);
290 q += clen;
291 ramdisk_image += clen;
292 ramdisk_size -= clen;
293 }
294
295 /* Copy the highmem portion of the initrd */
296 while (ramdisk_size) {
297 slop = ramdisk_image & ~PAGE_MASK;
298 clen = ramdisk_size;
299 if (clen > MAX_MAP_CHUNK-slop)
300 clen = MAX_MAP_CHUNK-slop;
301 mapaddr = ramdisk_image & PAGE_MASK;
302 p = early_ioremap(mapaddr, clen+slop);
303 memcpy(q, p+slop, clen);
304 early_iounmap(p, clen+slop);
305 q += clen;
306 ramdisk_image += clen;
307 ramdisk_size -= clen;
308 }
309 /* high pages is not converted by early_res_to_bootmem */
310 ramdisk_image = boot_params.hdr.ramdisk_image;
311 ramdisk_size = boot_params.hdr.ramdisk_size;
312 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
313 " %08llx - %08llx\n",
314 ramdisk_image, ramdisk_image + ramdisk_size - 1,
315 ramdisk_here, ramdisk_here + ramdisk_size - 1);
316}
46#endif 317#endif
318
319static void __init reserve_initrd(void)
320{
321 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
322 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
323 u64 ramdisk_end = ramdisk_image + ramdisk_size;
324 u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
325
326 if (!boot_params.hdr.type_of_loader ||
327 !ramdisk_image || !ramdisk_size)
328 return; /* No initrd provided by bootloader */
329
330 initrd_start = 0;
331
332 if (ramdisk_size >= (end_of_lowmem>>1)) {
333 free_early(ramdisk_image, ramdisk_end);
334 printk(KERN_ERR "initrd too large to handle, "
335 "disabling initrd\n");
336 return;
337 }
338
339 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
340 ramdisk_end);
341
342
343 if (ramdisk_end <= end_of_lowmem) {
344 /* All in lowmem, easy case */
345 /*
346 * don't need to reserve again, already reserved early
347 * in i386_start_kernel
348 */
349 initrd_start = ramdisk_image + PAGE_OFFSET;
350 initrd_end = initrd_start + ramdisk_size;
351 return;
47 } 352 }
48 353
49 /* indicate the early static arrays will soon be gone */ 354#ifdef CONFIG_X86_32
50 x86_cpu_to_apicid_early_ptr = NULL; 355 relocate_initrd();
51 x86_bios_cpu_apicid_early_ptr = NULL; 356#else
52#ifdef CONFIG_NUMA 357 printk(KERN_ERR "initrd extends beyond end of memory "
53 x86_cpu_to_node_map_early_ptr = NULL; 358 "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
359 ramdisk_end, end_of_lowmem);
360 initrd_start = 0;
54#endif 361#endif
362 free_early(ramdisk_image, ramdisk_end);
55} 363}
364#else
365static void __init reserve_initrd(void)
366{
367}
368#endif /* CONFIG_BLK_DEV_INITRD */
369
370static void __init parse_setup_data(void)
371{
372 struct setup_data *data;
373 u64 pa_data;
374
375 if (boot_params.hdr.version < 0x0209)
376 return;
377 pa_data = boot_params.hdr.setup_data;
378 while (pa_data) {
379 data = early_ioremap(pa_data, PAGE_SIZE);
380 switch (data->type) {
381 case SETUP_E820_EXT:
382 parse_e820_ext(data, pa_data);
383 break;
384 default:
385 break;
386 }
387 pa_data = data->next;
388 early_iounmap(data, PAGE_SIZE);
389 }
390}
391
392static void __init e820_reserve_setup_data(void)
393{
394 struct setup_data *data;
395 u64 pa_data;
396 int found = 0;
397
398 if (boot_params.hdr.version < 0x0209)
399 return;
400 pa_data = boot_params.hdr.setup_data;
401 while (pa_data) {
402 data = early_ioremap(pa_data, sizeof(*data));
403 e820_update_range(pa_data, sizeof(*data)+data->len,
404 E820_RAM, E820_RESERVED_KERN);
405 found = 1;
406 pa_data = data->next;
407 early_iounmap(data, sizeof(*data));
408 }
409 if (!found)
410 return;
56 411
57#ifdef CONFIG_HAVE_CPUMASK_OF_CPU_MAP 412 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
58cpumask_t *cpumask_of_cpu_map __read_mostly; 413 memcpy(&e820_saved, &e820, sizeof(struct e820map));
59EXPORT_SYMBOL(cpumask_of_cpu_map); 414 printk(KERN_INFO "extended physical RAM map:\n");
415 e820_print_map("reserve setup_data");
416}
60 417
61/* requires nr_cpu_ids to be initialized */ 418static void __init reserve_early_setup_data(void)
62static void __init setup_cpumask_of_cpu(void)
63{ 419{
64 int i; 420 struct setup_data *data;
421 u64 pa_data;
422 char buf[32];
423
424 if (boot_params.hdr.version < 0x0209)
425 return;
426 pa_data = boot_params.hdr.setup_data;
427 while (pa_data) {
428 data = early_ioremap(pa_data, sizeof(*data));
429 sprintf(buf, "setup data %x", data->type);
430 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
431 pa_data = data->next;
432 early_iounmap(data, sizeof(*data));
433 }
434}
435
436/*
437 * --------- Crashkernel reservation ------------------------------
438 */
439
440#ifdef CONFIG_KEXEC
441
442/**
443 * Reserve @size bytes of crashkernel memory at any suitable offset.
444 *
445 * @size: Size of the crashkernel memory to reserve.
446 * Returns the base address on success, and -1ULL on failure.
447 */
448unsigned long long find_and_reserve_crashkernel(unsigned long long size)
449{
450 const unsigned long long alignment = 16<<20; /* 16M */
451 unsigned long long start = 0LL;
452
453 while (1) {
454 int ret;
455
456 start = find_e820_area(start, ULONG_MAX, size, alignment);
457 if (start == -1ULL)
458 return start;
459
460 /* try to reserve it */
461 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
462 if (ret >= 0)
463 return start;
65 464
66 /* alloc_bootmem zeroes memory */ 465 start += alignment;
67 cpumask_of_cpu_map = alloc_bootmem_low(sizeof(cpumask_t) * nr_cpu_ids); 466 }
68 for (i = 0; i < nr_cpu_ids; i++) 467}
69 cpu_set(i, cpumask_of_cpu_map[i]); 468
469static inline unsigned long long get_total_mem(void)
470{
471 unsigned long long total;
472
473 total = max_low_pfn - min_low_pfn;
474#ifdef CONFIG_HIGHMEM
475 total += highend_pfn - highstart_pfn;
476#endif
477
478 return total << PAGE_SHIFT;
479}
480
481static void __init reserve_crashkernel(void)
482{
483 unsigned long long total_mem;
484 unsigned long long crash_size, crash_base;
485 int ret;
486
487 total_mem = get_total_mem();
488
489 ret = parse_crashkernel(boot_command_line, total_mem,
490 &crash_size, &crash_base);
491 if (ret != 0 || crash_size <= 0)
492 return;
493
494 /* 0 means: find the address automatically */
495 if (crash_base <= 0) {
496 crash_base = find_and_reserve_crashkernel(crash_size);
497 if (crash_base == -1ULL) {
498 pr_info("crashkernel reservation failed. "
499 "No suitable area found.\n");
500 return;
501 }
502 } else {
503 ret = reserve_bootmem_generic(crash_base, crash_size,
504 BOOTMEM_EXCLUSIVE);
505 if (ret < 0) {
506 pr_info("crashkernel reservation failed - "
507 "memory is in use\n");
508 return;
509 }
510 }
511
512 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
513 "for crashkernel (System RAM: %ldMB)\n",
514 (unsigned long)(crash_size >> 20),
515 (unsigned long)(crash_base >> 20),
516 (unsigned long)(total_mem >> 20));
517
518 crashk_res.start = crash_base;
519 crashk_res.end = crash_base + crash_size - 1;
520 insert_resource(&iomem_resource, &crashk_res);
70} 521}
71#else 522#else
72static inline void setup_cpumask_of_cpu(void) { } 523static void __init reserve_crashkernel(void)
524{
525}
73#endif 526#endif
74 527
75#ifdef CONFIG_X86_32 528static struct resource standard_io_resources[] = {
76/* 529 { .name = "dma1", .start = 0x00, .end = 0x1f,
77 * Great future not-so-futuristic plan: make i386 and x86_64 do it 530 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
78 * the same way 531 { .name = "pic1", .start = 0x20, .end = 0x21,
532 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
533 { .name = "timer0", .start = 0x40, .end = 0x43,
534 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
535 { .name = "timer1", .start = 0x50, .end = 0x53,
536 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
537 { .name = "keyboard", .start = 0x60, .end = 0x60,
538 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
539 { .name = "keyboard", .start = 0x64, .end = 0x64,
540 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
541 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
542 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
543 { .name = "pic2", .start = 0xa0, .end = 0xa1,
544 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
545 { .name = "dma2", .start = 0xc0, .end = 0xdf,
546 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
547 { .name = "fpu", .start = 0xf0, .end = 0xff,
548 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
549};
550
551static void __init reserve_standard_io_resources(void)
552{
553 int i;
554
555 /* request I/O space for devices used on all i[345]86 PCs */
556 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
557 request_resource(&ioport_resource, &standard_io_resources[i]);
558
559}
560
561#ifdef CONFIG_PROC_VMCORE
562/* elfcorehdr= specifies the location of elf core header
563 * stored by the crashed kernel. This option will be passed
564 * by kexec loader to the capture kernel.
79 */ 565 */
80unsigned long __per_cpu_offset[NR_CPUS] __read_mostly; 566static int __init setup_elfcorehdr(char *arg)
81EXPORT_SYMBOL(__per_cpu_offset); 567{
568 char *end;
569 if (!arg)
570 return -EINVAL;
571 elfcorehdr_addr = memparse(arg, &end);
572 return end > arg ? 0 : -EINVAL;
573}
574early_param("elfcorehdr", setup_elfcorehdr);
82#endif 575#endif
83 576
577static struct x86_quirks default_x86_quirks __initdata;
578
579struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
580
581/*
582 * 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
584 * for initialization. Note, the efi init code path is determined by the
585 * global efi_enabled. This allows the same kernel image to be used on existing
586 * systems (with a traditional BIOS) as well as on EFI systems.
587 */
84/* 588/*
85 * Great future plan: 589 * setup_arch - architecture-specific boot-time initializations
86 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data. 590 *
87 * Always point %gs to its beginning 591 * Note: On x86_64, fixmaps are ready for use even before this is called.
88 */ 592 */
89void __init setup_per_cpu_areas(void) 593
594void __init setup_arch(char **cmdline_p)
90{ 595{
91 int i, highest_cpu = 0; 596#ifdef CONFIG_X86_32
92 unsigned long size; 597 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
598 visws_early_detect();
599 pre_setup_arch_hook();
600#else
601 printk(KERN_INFO "Command line: %s\n", boot_command_line);
602#endif
93 603
94#ifdef CONFIG_HOTPLUG_CPU 604 early_cpu_init();
95 prefill_possible_map(); 605 early_ioremap_init();
606
607 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
608 screen_info = boot_params.screen_info;
609 edid_info = boot_params.edid_info;
610#ifdef CONFIG_X86_32
611 apm_info.bios = boot_params.apm_bios_info;
612 ist_info = boot_params.ist_info;
613 if (boot_params.sys_desc_table.length != 0) {
614 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
615 machine_id = boot_params.sys_desc_table.table[0];
616 machine_submodel_id = boot_params.sys_desc_table.table[1];
617 BIOS_revision = boot_params.sys_desc_table.table[2];
618 }
619#endif
620 saved_video_mode = boot_params.hdr.vid_mode;
621 bootloader_type = boot_params.hdr.type_of_loader;
622
623#ifdef CONFIG_BLK_DEV_RAM
624 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
625 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
626 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
627#endif
628#ifdef CONFIG_EFI
629 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
630#ifdef CONFIG_X86_32
631 "EL32",
632#else
633 "EL64",
96#endif 634#endif
635 4)) {
636 efi_enabled = 1;
637 efi_reserve_early();
638 }
639#endif
640
641 ARCH_SETUP
642
643 setup_memory_map();
644 parse_setup_data();
645 /* update the e820_saved too */
646 e820_reserve_setup_data();
97 647
98 /* Copy section for each CPU (we discard the original) */ 648 copy_edd();
99 size = PERCPU_ENOUGH_ROOM;
100 printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n",
101 size);
102 649
103 for_each_possible_cpu(i) { 650 if (!boot_params.hdr.root_flags)
104 char *ptr; 651 root_mountflags &= ~MS_RDONLY;
105#ifndef CONFIG_NEED_MULTIPLE_NODES 652 init_mm.start_code = (unsigned long) _text;
106 ptr = alloc_bootmem_pages(size); 653 init_mm.end_code = (unsigned long) _etext;
654 init_mm.end_data = (unsigned long) _edata;
655#ifdef CONFIG_X86_32
656 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
107#else 657#else
108 int node = early_cpu_to_node(i); 658 init_mm.brk = (unsigned long) &_end;
109 if (!node_online(node) || !NODE_DATA(node)) {
110 ptr = alloc_bootmem_pages(size);
111 printk(KERN_INFO
112 "cpu %d has no node or node-local memory\n", i);
113 }
114 else
115 ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
116#endif 659#endif
117 if (!ptr) 660
118 panic("Cannot allocate cpu data for CPU %d\n", i); 661 code_resource.start = virt_to_phys(_text);
119#ifdef CONFIG_X86_64 662 code_resource.end = virt_to_phys(_etext)-1;
120 cpu_pda(i)->data_offset = ptr - __per_cpu_start; 663 data_resource.start = virt_to_phys(_etext);
664 data_resource.end = virt_to_phys(_edata)-1;
665 bss_resource.start = virt_to_phys(&__bss_start);
666 bss_resource.end = virt_to_phys(&__bss_stop)-1;
667
668 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
669 *cmdline_p = command_line;
670
671 parse_early_param();
672
673 /* after early param, so could get panic from serial */
674 reserve_early_setup_data();
675
676 if (acpi_mps_check()) {
677#ifdef CONFIG_X86_LOCAL_APIC
678 disable_apic = 1;
679#endif
680 setup_clear_cpu_cap(X86_FEATURE_APIC);
681 }
682
683#ifdef CONFIG_PCI
684 if (pci_early_dump_regs)
685 early_dump_pci_devices();
686#endif
687
688 finish_e820_parsing();
689
690#ifdef CONFIG_X86_32
691 probe_roms();
692#endif
693
694 /* after parse_early_param, so could debug it */
695 insert_resource(&iomem_resource, &code_resource);
696 insert_resource(&iomem_resource, &data_resource);
697 insert_resource(&iomem_resource, &bss_resource);
698
699 if (efi_enabled)
700 efi_init();
701
702#ifdef CONFIG_X86_32
703 if (ppro_with_ram_bug()) {
704 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
705 E820_RESERVED);
706 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
707 printk(KERN_INFO "fixed physical RAM map:\n");
708 e820_print_map("bad_ppro");
709 }
710#else
711 early_gart_iommu_check();
712#endif
713
714 /*
715 * partially used pages are not usable - thus
716 * we are rounding upwards:
717 */
718 max_pfn = e820_end_of_ram_pfn();
719
720 /* preallocate 4k for mptable mpc */
721 early_reserve_e820_mpc_new();
722 /* update e820 for memory not covered by WB MTRRs */
723 mtrr_bp_init();
724 if (mtrr_trim_uncached_memory(max_pfn))
725 max_pfn = e820_end_of_ram_pfn();
726
727#ifdef CONFIG_X86_32
728 /* max_low_pfn get updated here */
729 find_low_pfn_range();
121#else 730#else
122 __per_cpu_offset[i] = ptr - __per_cpu_start; 731 num_physpages = max_pfn;
732
733 check_efer();
734
735 /* How many end-of-memory variables you have, grandma! */
736 /* need this before calling reserve_initrd */
737 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
738 max_low_pfn = e820_end_of_low_ram_pfn();
739 else
740 max_low_pfn = max_pfn;
741
742 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
123#endif 743#endif
124 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
125 744
126 highest_cpu = i; 745 /* max_pfn_mapped is updated here */
746 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
747 max_pfn_mapped = max_low_pfn_mapped;
748
749#ifdef CONFIG_X86_64
750 if (max_pfn > max_low_pfn) {
751 max_pfn_mapped = init_memory_mapping(1UL<<32,
752 max_pfn<<PAGE_SHIFT);
753 /* can we preseve max_low_pfn ?*/
754 max_low_pfn = max_pfn;
127 } 755 }
756#endif
128 757
129 nr_cpu_ids = highest_cpu + 1; 758 /*
130 printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d\n", NR_CPUS, nr_cpu_ids); 759 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
760 */
131 761
132 /* Setup percpu data maps */ 762#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
133 setup_per_cpu_maps(); 763 if (init_ohci1394_dma_early)
764 init_ohci1394_dma_on_all_controllers();
765#endif
134 766
135 /* Setup cpumask_of_cpu map */ 767 reserve_initrd();
136 setup_cpumask_of_cpu(); 768
137} 769#ifdef CONFIG_X86_64
770 vsmp_init();
771#endif
772
773 dmi_scan_machine();
774
775 io_delay_init();
776
777 /*
778 * Parse the ACPI tables for possible boot-time SMP configuration.
779 */
780 acpi_boot_table_init();
781
782#ifdef CONFIG_ACPI_NUMA
783 /*
784 * Parse SRAT to discover nodes.
785 */
786 acpi_numa_init();
787#endif
788
789 initmem_init(0, max_pfn);
790
791#ifdef CONFIG_X86_64
792 dma32_reserve_bootmem();
793#endif
138 794
795#ifdef CONFIG_ACPI_SLEEP
796 /*
797 * Reserve low memory region for sleep support.
798 */
799 acpi_reserve_bootmem();
139#endif 800#endif
801#ifdef CONFIG_X86_FIND_SMP_CONFIG
802 /*
803 * Find and reserve possible boot-time SMP configuration:
804 */
805 find_smp_config();
806#endif
807 reserve_crashkernel();
808
809 reserve_ibft_region();
810
811#ifdef CONFIG_KVM_CLOCK
812 kvmclock_init();
813#endif
814
815#if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
816 /*
817 * Must be after max_low_pfn is determined, and before kernel
818 * pagetables are setup.
819 */
820 vmi_init();
821#endif
822
823 paravirt_pagetable_setup_start(swapper_pg_dir);
824 paging_init();
825 paravirt_pagetable_setup_done(swapper_pg_dir);
826 paravirt_post_allocator_init();
827
828#ifdef CONFIG_X86_64
829 map_vsyscall();
830#endif
831
832#ifdef CONFIG_X86_GENERICARCH
833 generic_apic_probe();
834#endif
835
836 early_quirks();
837
838 /*
839 * Read APIC and some other early information from ACPI tables.
840 */
841 acpi_boot_init();
842
843#if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
844 /*
845 * get boot-time SMP configuration:
846 */
847 if (smp_found_config)
848 get_smp_config();
849#endif
850
851 prefill_possible_map();
852#ifdef CONFIG_X86_64
853 init_cpu_to_node();
854#endif
855
856 init_apic_mappings();
857 ioapic_init_mappings();
858
859#if defined(CONFIG_SMP) && defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
860 if (def_to_bigsmp)
861 printk(KERN_WARNING "More than 8 CPUs detected and "
862 "CONFIG_X86_PC cannot handle it.\nUse "
863 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
864#endif
865 kvm_guest_init();
866
867 e820_reserve_resources();
868 e820_mark_nosave_regions(max_low_pfn);
869
870#ifdef CONFIG_X86_32
871 request_resource(&iomem_resource, &video_ram_resource);
872#endif
873 reserve_standard_io_resources();
874
875 e820_setup_gap();
876
877#ifdef CONFIG_VT
878#if defined(CONFIG_VGA_CONSOLE)
879 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
880 conswitchp = &vga_con;
881#elif defined(CONFIG_DUMMY_CONSOLE)
882 conswitchp = &dummy_con;
883#endif
884#endif
885}
diff --git a/arch/x86/kernel/setup64.c b/arch/x86/kernel/setup64.c
deleted file mode 100644
index aee0e8200777..000000000000
--- a/arch/x86/kernel/setup64.c
+++ /dev/null
@@ -1,287 +0,0 @@
1/*
2 * X86-64 specific CPU setup.
3 * Copyright (C) 1995 Linus Torvalds
4 * Copyright 2001, 2002, 2003 SuSE Labs / Andi Kleen.
5 * See setup.c for older changelog.
6 */
7#include <linux/init.h>
8#include <linux/kernel.h>
9#include <linux/sched.h>
10#include <linux/string.h>
11#include <linux/bootmem.h>
12#include <linux/bitops.h>
13#include <linux/module.h>
14#include <linux/kgdb.h>
15#include <asm/pda.h>
16#include <asm/pgtable.h>
17#include <asm/processor.h>
18#include <asm/desc.h>
19#include <asm/atomic.h>
20#include <asm/mmu_context.h>
21#include <asm/smp.h>
22#include <asm/i387.h>
23#include <asm/percpu.h>
24#include <asm/proto.h>
25#include <asm/sections.h>
26#include <asm/setup.h>
27#include <asm/genapic.h>
28
29#ifndef CONFIG_DEBUG_BOOT_PARAMS
30struct boot_params __initdata boot_params;
31#else
32struct boot_params boot_params;
33#endif
34
35cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
36
37struct x8664_pda *_cpu_pda[NR_CPUS] __read_mostly;
38EXPORT_SYMBOL(_cpu_pda);
39struct x8664_pda boot_cpu_pda[NR_CPUS] __cacheline_aligned;
40
41struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
42
43char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
44
45unsigned long __supported_pte_mask __read_mostly = ~0UL;
46EXPORT_SYMBOL_GPL(__supported_pte_mask);
47
48static int do_not_nx __cpuinitdata = 0;
49
50/* noexec=on|off
51Control non executable mappings for 64bit processes.
52
53on Enable(default)
54off Disable
55*/
56static int __init nonx_setup(char *str)
57{
58 if (!str)
59 return -EINVAL;
60 if (!strncmp(str, "on", 2)) {
61 __supported_pte_mask |= _PAGE_NX;
62 do_not_nx = 0;
63 } else if (!strncmp(str, "off", 3)) {
64 do_not_nx = 1;
65 __supported_pte_mask &= ~_PAGE_NX;
66 }
67 return 0;
68}
69early_param("noexec", nonx_setup);
70
71int force_personality32 = 0;
72
73/* noexec32=on|off
74Control non executable heap for 32bit processes.
75To control the stack too use noexec=off
76
77on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
78off PROT_READ implies PROT_EXEC
79*/
80static int __init nonx32_setup(char *str)
81{
82 if (!strcmp(str, "on"))
83 force_personality32 &= ~READ_IMPLIES_EXEC;
84 else if (!strcmp(str, "off"))
85 force_personality32 |= READ_IMPLIES_EXEC;
86 return 1;
87}
88__setup("noexec32=", nonx32_setup);
89
90void pda_init(int cpu)
91{
92 struct x8664_pda *pda = cpu_pda(cpu);
93
94 /* Setup up data that may be needed in __get_free_pages early */
95 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
96 /* Memory clobbers used to order PDA accessed */
97 mb();
98 wrmsrl(MSR_GS_BASE, pda);
99 mb();
100
101 pda->cpunumber = cpu;
102 pda->irqcount = -1;
103 pda->kernelstack =
104 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
105 pda->active_mm = &init_mm;
106 pda->mmu_state = 0;
107
108 if (cpu == 0) {
109 /* others are initialized in smpboot.c */
110 pda->pcurrent = &init_task;
111 pda->irqstackptr = boot_cpu_stack;
112 } else {
113 pda->irqstackptr = (char *)
114 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
115 if (!pda->irqstackptr)
116 panic("cannot allocate irqstack for cpu %d", cpu);
117 }
118
119
120 pda->irqstackptr += IRQSTACKSIZE-64;
121}
122
123char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
124__attribute__((section(".bss.page_aligned")));
125
126extern asmlinkage void ignore_sysret(void);
127
128/* May not be marked __init: used by software suspend */
129void syscall_init(void)
130{
131 /*
132 * LSTAR and STAR live in a bit strange symbiosis.
133 * They both write to the same internal register. STAR allows to set CS/DS
134 * but only a 32bit target. LSTAR sets the 64bit rip.
135 */
136 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
137 wrmsrl(MSR_LSTAR, system_call);
138 wrmsrl(MSR_CSTAR, ignore_sysret);
139
140#ifdef CONFIG_IA32_EMULATION
141 syscall32_cpu_init ();
142#endif
143
144 /* Flags to clear on syscall */
145 wrmsrl(MSR_SYSCALL_MASK,
146 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
147}
148
149void __cpuinit check_efer(void)
150{
151 unsigned long efer;
152
153 rdmsrl(MSR_EFER, efer);
154 if (!(efer & EFER_NX) || do_not_nx) {
155 __supported_pte_mask &= ~_PAGE_NX;
156 }
157}
158
159unsigned long kernel_eflags;
160
161/*
162 * Copies of the original ist values from the tss are only accessed during
163 * debugging, no special alignment required.
164 */
165DEFINE_PER_CPU(struct orig_ist, orig_ist);
166
167/*
168 * cpu_init() initializes state that is per-CPU. Some data is already
169 * initialized (naturally) in the bootstrap process, such as the GDT
170 * and IDT. We reload them nevertheless, this function acts as a
171 * 'CPU state barrier', nothing should get across.
172 * A lot of state is already set up in PDA init.
173 */
174void __cpuinit cpu_init (void)
175{
176 int cpu = stack_smp_processor_id();
177 struct tss_struct *t = &per_cpu(init_tss, cpu);
178 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
179 unsigned long v;
180 char *estacks = NULL;
181 struct task_struct *me;
182 int i;
183
184 /* CPU 0 is initialised in head64.c */
185 if (cpu != 0) {
186 pda_init(cpu);
187 } else
188 estacks = boot_exception_stacks;
189
190 me = current;
191
192 if (cpu_test_and_set(cpu, cpu_initialized))
193 panic("CPU#%d already initialized!\n", cpu);
194
195 printk("Initializing CPU#%d\n", cpu);
196
197 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
198
199 /*
200 * Initialize the per-CPU GDT with the boot GDT,
201 * and set up the GDT descriptor:
202 */
203 if (cpu)
204 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
205
206 cpu_gdt_descr[cpu].size = GDT_SIZE;
207 load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
208 load_idt((const struct desc_ptr *)&idt_descr);
209
210 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
211 syscall_init();
212
213 wrmsrl(MSR_FS_BASE, 0);
214 wrmsrl(MSR_KERNEL_GS_BASE, 0);
215 barrier();
216
217 check_efer();
218
219 /*
220 * set up and load the per-CPU TSS
221 */
222 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
223 static const unsigned int order[N_EXCEPTION_STACKS] = {
224 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
225 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
226 };
227 if (cpu) {
228 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
229 if (!estacks)
230 panic("Cannot allocate exception stack %ld %d\n",
231 v, cpu);
232 }
233 estacks += PAGE_SIZE << order[v];
234 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
235 }
236
237 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
238 /*
239 * <= is required because the CPU will access up to
240 * 8 bits beyond the end of the IO permission bitmap.
241 */
242 for (i = 0; i <= IO_BITMAP_LONGS; i++)
243 t->io_bitmap[i] = ~0UL;
244
245 atomic_inc(&init_mm.mm_count);
246 me->active_mm = &init_mm;
247 if (me->mm)
248 BUG();
249 enter_lazy_tlb(&init_mm, me);
250
251 set_tss_desc(cpu, t);
252 load_TR_desc();
253 load_LDT(&init_mm.context);
254
255#ifdef CONFIG_KGDB
256 /*
257 * If the kgdb is connected no debug regs should be altered. This
258 * is only applicable when KGDB and a KGDB I/O module are built
259 * into the kernel and you are using early debugging with
260 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
261 */
262 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
263 arch_kgdb_ops.correct_hw_break();
264 else {
265#endif
266 /*
267 * Clear all 6 debug registers:
268 */
269
270 set_debugreg(0UL, 0);
271 set_debugreg(0UL, 1);
272 set_debugreg(0UL, 2);
273 set_debugreg(0UL, 3);
274 set_debugreg(0UL, 6);
275 set_debugreg(0UL, 7);
276#ifdef CONFIG_KGDB
277 /* If the kgdb is connected no debug regs should be altered. */
278 }
279#endif
280
281 fpu_init();
282
283 raw_local_save_flags(kernel_eflags);
284
285 if (is_uv_system())
286 uv_cpu_init();
287}
diff --git a/arch/x86/kernel/setup_32.c b/arch/x86/kernel/setup_32.c
deleted file mode 100644
index 2c5f8b213e86..000000000000
--- a/arch/x86/kernel/setup_32.c
+++ /dev/null
@@ -1,958 +0,0 @@
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/screen_info.h>
28#include <linux/ioport.h>
29#include <linux/acpi.h>
30#include <linux/apm_bios.h>
31#include <linux/initrd.h>
32#include <linux/bootmem.h>
33#include <linux/seq_file.h>
34#include <linux/console.h>
35#include <linux/mca.h>
36#include <linux/root_dev.h>
37#include <linux/highmem.h>
38#include <linux/module.h>
39#include <linux/efi.h>
40#include <linux/init.h>
41#include <linux/edd.h>
42#include <linux/iscsi_ibft.h>
43#include <linux/nodemask.h>
44#include <linux/kexec.h>
45#include <linux/crash_dump.h>
46#include <linux/dmi.h>
47#include <linux/pfn.h>
48#include <linux/pci.h>
49#include <linux/init_ohci1394_dma.h>
50#include <linux/kvm_para.h>
51
52#include <video/edid.h>
53
54#include <asm/mtrr.h>
55#include <asm/apic.h>
56#include <asm/e820.h>
57#include <asm/mpspec.h>
58#include <asm/mmzone.h>
59#include <asm/setup.h>
60#include <asm/arch_hooks.h>
61#include <asm/sections.h>
62#include <asm/io_apic.h>
63#include <asm/ist.h>
64#include <asm/io.h>
65#include <asm/vmi.h>
66#include <setup_arch.h>
67#include <asm/bios_ebda.h>
68#include <asm/cacheflush.h>
69#include <asm/processor.h>
70
71/* This value is set up by the early boot code to point to the value
72 immediately after the boot time page tables. It contains a *physical*
73 address, and must not be in the .bss segment! */
74unsigned long init_pg_tables_end __initdata = ~0UL;
75
76/*
77 * Machine setup..
78 */
79static struct resource data_resource = {
80 .name = "Kernel data",
81 .start = 0,
82 .end = 0,
83 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
84};
85
86static struct resource code_resource = {
87 .name = "Kernel code",
88 .start = 0,
89 .end = 0,
90 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
91};
92
93static struct resource bss_resource = {
94 .name = "Kernel bss",
95 .start = 0,
96 .end = 0,
97 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
98};
99
100static struct resource video_ram_resource = {
101 .name = "Video RAM area",
102 .start = 0xa0000,
103 .end = 0xbffff,
104 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
105};
106
107static struct resource standard_io_resources[] = { {
108 .name = "dma1",
109 .start = 0x0000,
110 .end = 0x001f,
111 .flags = IORESOURCE_BUSY | IORESOURCE_IO
112}, {
113 .name = "pic1",
114 .start = 0x0020,
115 .end = 0x0021,
116 .flags = IORESOURCE_BUSY | IORESOURCE_IO
117}, {
118 .name = "timer0",
119 .start = 0x0040,
120 .end = 0x0043,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO
122}, {
123 .name = "timer1",
124 .start = 0x0050,
125 .end = 0x0053,
126 .flags = IORESOURCE_BUSY | IORESOURCE_IO
127}, {
128 .name = "keyboard",
129 .start = 0x0060,
130 .end = 0x0060,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO
132}, {
133 .name = "keyboard",
134 .start = 0x0064,
135 .end = 0x0064,
136 .flags = IORESOURCE_BUSY | IORESOURCE_IO
137}, {
138 .name = "dma page reg",
139 .start = 0x0080,
140 .end = 0x008f,
141 .flags = IORESOURCE_BUSY | IORESOURCE_IO
142}, {
143 .name = "pic2",
144 .start = 0x00a0,
145 .end = 0x00a1,
146 .flags = IORESOURCE_BUSY | IORESOURCE_IO
147}, {
148 .name = "dma2",
149 .start = 0x00c0,
150 .end = 0x00df,
151 .flags = IORESOURCE_BUSY | IORESOURCE_IO
152}, {
153 .name = "fpu",
154 .start = 0x00f0,
155 .end = 0x00ff,
156 .flags = IORESOURCE_BUSY | IORESOURCE_IO
157} };
158
159/* cpu data as detected by the assembly code in head.S */
160struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
161/* common cpu data for all cpus */
162struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
163EXPORT_SYMBOL(boot_cpu_data);
164
165unsigned int def_to_bigsmp;
166
167#ifndef CONFIG_X86_PAE
168unsigned long mmu_cr4_features;
169#else
170unsigned long mmu_cr4_features = X86_CR4_PAE;
171#endif
172
173/* for MCA, but anyone else can use it if they want */
174unsigned int machine_id;
175unsigned int machine_submodel_id;
176unsigned int BIOS_revision;
177
178/* Boot loader ID as an integer, for the benefit of proc_dointvec */
179int bootloader_type;
180
181/* user-defined highmem size */
182static unsigned int highmem_pages = -1;
183
184/*
185 * Setup options
186 */
187struct screen_info screen_info;
188EXPORT_SYMBOL(screen_info);
189struct apm_info apm_info;
190EXPORT_SYMBOL(apm_info);
191struct edid_info edid_info;
192EXPORT_SYMBOL_GPL(edid_info);
193struct ist_info ist_info;
194#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
195 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
196EXPORT_SYMBOL(ist_info);
197#endif
198
199extern void early_cpu_init(void);
200extern int root_mountflags;
201
202unsigned long saved_video_mode;
203
204#define RAMDISK_IMAGE_START_MASK 0x07FF
205#define RAMDISK_PROMPT_FLAG 0x8000
206#define RAMDISK_LOAD_FLAG 0x4000
207
208static char __initdata command_line[COMMAND_LINE_SIZE];
209
210#ifndef CONFIG_DEBUG_BOOT_PARAMS
211struct boot_params __initdata boot_params;
212#else
213struct boot_params boot_params;
214#endif
215
216#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
217struct edd edd;
218#ifdef CONFIG_EDD_MODULE
219EXPORT_SYMBOL(edd);
220#endif
221/**
222 * copy_edd() - Copy the BIOS EDD information
223 * from boot_params into a safe place.
224 *
225 */
226static inline void copy_edd(void)
227{
228 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
229 sizeof(edd.mbr_signature));
230 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
231 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
232 edd.edd_info_nr = boot_params.eddbuf_entries;
233}
234#else
235static inline void copy_edd(void)
236{
237}
238#endif
239
240int __initdata user_defined_memmap;
241
242/*
243 * "mem=nopentium" disables the 4MB page tables.
244 * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
245 * to <mem>, overriding the bios size.
246 * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
247 * <start> to <start>+<mem>, overriding the bios size.
248 *
249 * HPA tells me bootloaders need to parse mem=, so no new
250 * option should be mem= [also see Documentation/i386/boot.txt]
251 */
252static int __init parse_mem(char *arg)
253{
254 if (!arg)
255 return -EINVAL;
256
257 if (strcmp(arg, "nopentium") == 0) {
258 setup_clear_cpu_cap(X86_FEATURE_PSE);
259 } else {
260 /* If the user specifies memory size, we
261 * limit the BIOS-provided memory map to
262 * that size. exactmap can be used to specify
263 * the exact map. mem=number can be used to
264 * trim the existing memory map.
265 */
266 unsigned long long mem_size;
267
268 mem_size = memparse(arg, &arg);
269 limit_regions(mem_size);
270 user_defined_memmap = 1;
271 }
272 return 0;
273}
274early_param("mem", parse_mem);
275
276#ifdef CONFIG_PROC_VMCORE
277/* elfcorehdr= specifies the location of elf core header
278 * stored by the crashed kernel.
279 */
280static int __init parse_elfcorehdr(char *arg)
281{
282 if (!arg)
283 return -EINVAL;
284
285 elfcorehdr_addr = memparse(arg, &arg);
286 return 0;
287}
288early_param("elfcorehdr", parse_elfcorehdr);
289#endif /* CONFIG_PROC_VMCORE */
290
291/*
292 * highmem=size forces highmem to be exactly 'size' bytes.
293 * This works even on boxes that have no highmem otherwise.
294 * This also works to reduce highmem size on bigger boxes.
295 */
296static int __init parse_highmem(char *arg)
297{
298 if (!arg)
299 return -EINVAL;
300
301 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
302 return 0;
303}
304early_param("highmem", parse_highmem);
305
306/*
307 * vmalloc=size forces the vmalloc area to be exactly 'size'
308 * bytes. This can be used to increase (or decrease) the
309 * vmalloc area - the default is 128m.
310 */
311static int __init parse_vmalloc(char *arg)
312{
313 if (!arg)
314 return -EINVAL;
315
316 __VMALLOC_RESERVE = memparse(arg, &arg);
317 return 0;
318}
319early_param("vmalloc", parse_vmalloc);
320
321/*
322 * reservetop=size reserves a hole at the top of the kernel address space which
323 * a hypervisor can load into later. Needed for dynamically loaded hypervisors,
324 * so relocating the fixmap can be done before paging initialization.
325 */
326static int __init parse_reservetop(char *arg)
327{
328 unsigned long address;
329
330 if (!arg)
331 return -EINVAL;
332
333 address = memparse(arg, &arg);
334 reserve_top_address(address);
335 return 0;
336}
337early_param("reservetop", parse_reservetop);
338
339/*
340 * Determine low and high memory ranges:
341 */
342unsigned long __init find_max_low_pfn(void)
343{
344 unsigned long max_low_pfn;
345
346 max_low_pfn = max_pfn;
347 if (max_low_pfn > MAXMEM_PFN) {
348 if (highmem_pages == -1)
349 highmem_pages = max_pfn - MAXMEM_PFN;
350 if (highmem_pages + MAXMEM_PFN < max_pfn)
351 max_pfn = MAXMEM_PFN + highmem_pages;
352 if (highmem_pages + MAXMEM_PFN > max_pfn) {
353 printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
354 highmem_pages = 0;
355 }
356 max_low_pfn = MAXMEM_PFN;
357#ifndef CONFIG_HIGHMEM
358 /* Maximum memory usable is what is directly addressable */
359 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
360 MAXMEM>>20);
361 if (max_pfn > MAX_NONPAE_PFN)
362 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
363 else
364 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
365 max_pfn = MAXMEM_PFN;
366#else /* !CONFIG_HIGHMEM */
367#ifndef CONFIG_HIGHMEM64G
368 if (max_pfn > MAX_NONPAE_PFN) {
369 max_pfn = MAX_NONPAE_PFN;
370 printk(KERN_WARNING "Warning only 4GB will be used.\n");
371 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
372 }
373#endif /* !CONFIG_HIGHMEM64G */
374#endif /* !CONFIG_HIGHMEM */
375 } else {
376 if (highmem_pages == -1)
377 highmem_pages = 0;
378#ifdef CONFIG_HIGHMEM
379 if (highmem_pages >= max_pfn) {
380 printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
381 highmem_pages = 0;
382 }
383 if (highmem_pages) {
384 if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
385 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
386 highmem_pages = 0;
387 }
388 max_low_pfn -= highmem_pages;
389 }
390#else
391 if (highmem_pages)
392 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
393#endif
394 }
395 return max_low_pfn;
396}
397
398#define BIOS_LOWMEM_KILOBYTES 0x413
399
400/*
401 * The BIOS places the EBDA/XBDA at the top of conventional
402 * memory, and usually decreases the reported amount of
403 * conventional memory (int 0x12) too. This also contains a
404 * workaround for Dell systems that neglect to reserve EBDA.
405 * The same workaround also avoids a problem with the AMD768MPX
406 * chipset: reserve a page before VGA to prevent PCI prefetch
407 * into it (errata #56). Usually the page is reserved anyways,
408 * unless you have no PS/2 mouse plugged in.
409 */
410static void __init reserve_ebda_region(void)
411{
412 unsigned int lowmem, ebda_addr;
413
414 /* To determine the position of the EBDA and the */
415 /* end of conventional memory, we need to look at */
416 /* the BIOS data area. In a paravirtual environment */
417 /* that area is absent. We'll just have to assume */
418 /* that the paravirt case can handle memory setup */
419 /* correctly, without our help. */
420 if (paravirt_enabled())
421 return;
422
423 /* end of low (conventional) memory */
424 lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
425 lowmem <<= 10;
426
427 /* start of EBDA area */
428 ebda_addr = get_bios_ebda();
429
430 /* Fixup: bios puts an EBDA in the top 64K segment */
431 /* of conventional memory, but does not adjust lowmem. */
432 if ((lowmem - ebda_addr) <= 0x10000)
433 lowmem = ebda_addr;
434
435 /* Fixup: bios does not report an EBDA at all. */
436 /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
437 if ((ebda_addr == 0) && (lowmem >= 0x9f000))
438 lowmem = 0x9f000;
439
440 /* Paranoia: should never happen, but... */
441 if ((lowmem == 0) || (lowmem >= 0x100000))
442 lowmem = 0x9f000;
443
444 /* reserve all memory between lowmem and the 1MB mark */
445 reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT);
446}
447
448#ifndef CONFIG_NEED_MULTIPLE_NODES
449static void __init setup_bootmem_allocator(void);
450static unsigned long __init setup_memory(void)
451{
452 /*
453 * partially used pages are not usable - thus
454 * we are rounding upwards:
455 */
456 min_low_pfn = PFN_UP(init_pg_tables_end);
457
458 max_low_pfn = find_max_low_pfn();
459
460#ifdef CONFIG_HIGHMEM
461 highstart_pfn = highend_pfn = max_pfn;
462 if (max_pfn > max_low_pfn) {
463 highstart_pfn = max_low_pfn;
464 }
465 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
466 pages_to_mb(highend_pfn - highstart_pfn));
467 num_physpages = highend_pfn;
468 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
469#else
470 num_physpages = max_low_pfn;
471 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
472#endif
473#ifdef CONFIG_FLATMEM
474 max_mapnr = num_physpages;
475#endif
476 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
477 pages_to_mb(max_low_pfn));
478
479 setup_bootmem_allocator();
480
481 return max_low_pfn;
482}
483
484static void __init zone_sizes_init(void)
485{
486 unsigned long max_zone_pfns[MAX_NR_ZONES];
487 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
488 max_zone_pfns[ZONE_DMA] =
489 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
490 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
491#ifdef CONFIG_HIGHMEM
492 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
493 add_active_range(0, 0, highend_pfn);
494#else
495 add_active_range(0, 0, max_low_pfn);
496#endif
497
498 free_area_init_nodes(max_zone_pfns);
499}
500#else
501extern unsigned long __init setup_memory(void);
502extern void zone_sizes_init(void);
503#endif /* !CONFIG_NEED_MULTIPLE_NODES */
504
505static inline unsigned long long get_total_mem(void)
506{
507 unsigned long long total;
508
509 total = max_low_pfn - min_low_pfn;
510#ifdef CONFIG_HIGHMEM
511 total += highend_pfn - highstart_pfn;
512#endif
513
514 return total << PAGE_SHIFT;
515}
516
517#ifdef CONFIG_KEXEC
518static void __init reserve_crashkernel(void)
519{
520 unsigned long long total_mem;
521 unsigned long long crash_size, crash_base;
522 int ret;
523
524 total_mem = get_total_mem();
525
526 ret = parse_crashkernel(boot_command_line, total_mem,
527 &crash_size, &crash_base);
528 if (ret == 0 && crash_size > 0) {
529 if (crash_base > 0) {
530 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
531 "for crashkernel (System RAM: %ldMB)\n",
532 (unsigned long)(crash_size >> 20),
533 (unsigned long)(crash_base >> 20),
534 (unsigned long)(total_mem >> 20));
535 crashk_res.start = crash_base;
536 crashk_res.end = crash_base + crash_size - 1;
537 reserve_bootmem(crash_base, crash_size,
538 BOOTMEM_DEFAULT);
539 } else
540 printk(KERN_INFO "crashkernel reservation failed - "
541 "you have to specify a base address\n");
542 }
543}
544#else
545static inline void __init reserve_crashkernel(void)
546{}
547#endif
548
549#ifdef CONFIG_BLK_DEV_INITRD
550
551static bool do_relocate_initrd = false;
552
553static void __init reserve_initrd(void)
554{
555 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
556 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
557 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
558 unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
559 unsigned long ramdisk_here;
560
561 initrd_start = 0;
562
563 if (!boot_params.hdr.type_of_loader ||
564 !ramdisk_image || !ramdisk_size)
565 return; /* No initrd provided by bootloader */
566
567 if (ramdisk_end < ramdisk_image) {
568 printk(KERN_ERR "initrd wraps around end of memory, "
569 "disabling initrd\n");
570 return;
571 }
572 if (ramdisk_size >= end_of_lowmem/2) {
573 printk(KERN_ERR "initrd too large to handle, "
574 "disabling initrd\n");
575 return;
576 }
577 if (ramdisk_end <= end_of_lowmem) {
578 /* All in lowmem, easy case */
579 reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT);
580 initrd_start = ramdisk_image + PAGE_OFFSET;
581 initrd_end = initrd_start+ramdisk_size;
582 return;
583 }
584
585 /* We need to move the initrd down into lowmem */
586 ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK;
587
588 /* Note: this includes all the lowmem currently occupied by
589 the initrd, we rely on that fact to keep the data intact. */
590 reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT);
591 initrd_start = ramdisk_here + PAGE_OFFSET;
592 initrd_end = initrd_start + ramdisk_size;
593
594 do_relocate_initrd = true;
595}
596
597#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
598
599static void __init relocate_initrd(void)
600{
601 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
602 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
603 unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
604 unsigned long ramdisk_here;
605 unsigned long slop, clen, mapaddr;
606 char *p, *q;
607
608 if (!do_relocate_initrd)
609 return;
610
611 ramdisk_here = initrd_start - PAGE_OFFSET;
612
613 q = (char *)initrd_start;
614
615 /* Copy any lowmem portion of the initrd */
616 if (ramdisk_image < end_of_lowmem) {
617 clen = end_of_lowmem - ramdisk_image;
618 p = (char *)__va(ramdisk_image);
619 memcpy(q, p, clen);
620 q += clen;
621 ramdisk_image += clen;
622 ramdisk_size -= clen;
623 }
624
625 /* Copy the highmem portion of the initrd */
626 while (ramdisk_size) {
627 slop = ramdisk_image & ~PAGE_MASK;
628 clen = ramdisk_size;
629 if (clen > MAX_MAP_CHUNK-slop)
630 clen = MAX_MAP_CHUNK-slop;
631 mapaddr = ramdisk_image & PAGE_MASK;
632 p = early_ioremap(mapaddr, clen+slop);
633 memcpy(q, p+slop, clen);
634 early_iounmap(p, clen+slop);
635 q += clen;
636 ramdisk_image += clen;
637 ramdisk_size -= clen;
638 }
639}
640
641#endif /* CONFIG_BLK_DEV_INITRD */
642
643void __init setup_bootmem_allocator(void)
644{
645 unsigned long bootmap_size;
646 /*
647 * Initialize the boot-time allocator (with low memory only):
648 */
649 bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
650
651 register_bootmem_low_pages(max_low_pfn);
652
653 /*
654 * Reserve the bootmem bitmap itself as well. We do this in two
655 * steps (first step was init_bootmem()) because this catches
656 * the (very unlikely) case of us accidentally initializing the
657 * bootmem allocator with an invalid RAM area.
658 */
659 reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
660 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
661 BOOTMEM_DEFAULT);
662
663 /*
664 * reserve physical page 0 - it's a special BIOS page on many boxes,
665 * enabling clean reboots, SMP operation, laptop functions.
666 */
667 reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
668
669 /* reserve EBDA region */
670 reserve_ebda_region();
671
672#ifdef CONFIG_SMP
673 /*
674 * But first pinch a few for the stack/trampoline stuff
675 * FIXME: Don't need the extra page at 4K, but need to fix
676 * trampoline before removing it. (see the GDT stuff)
677 */
678 reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
679#endif
680#ifdef CONFIG_ACPI_SLEEP
681 /*
682 * Reserve low memory region for sleep support.
683 */
684 acpi_reserve_bootmem();
685#endif
686#ifdef CONFIG_X86_FIND_SMP_CONFIG
687 /*
688 * Find and reserve possible boot-time SMP configuration:
689 */
690 find_smp_config();
691#endif
692#ifdef CONFIG_BLK_DEV_INITRD
693 reserve_initrd();
694#endif
695 numa_kva_reserve();
696 reserve_crashkernel();
697
698 reserve_ibft_region();
699}
700
701/*
702 * The node 0 pgdat is initialized before all of these because
703 * it's needed for bootmem. node>0 pgdats have their virtual
704 * space allocated before the pagetables are in place to access
705 * them, so they can't be cleared then.
706 *
707 * This should all compile down to nothing when NUMA is off.
708 */
709static void __init remapped_pgdat_init(void)
710{
711 int nid;
712
713 for_each_online_node(nid) {
714 if (nid != 0)
715 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
716 }
717}
718
719#ifdef CONFIG_MCA
720static void set_mca_bus(int x)
721{
722 MCA_bus = x;
723}
724#else
725static void set_mca_bus(int x) { }
726#endif
727
728/* Overridden in paravirt.c if CONFIG_PARAVIRT */
729char * __init __attribute__((weak)) memory_setup(void)
730{
731 return machine_specific_memory_setup();
732}
733
734#ifdef CONFIG_NUMA
735/*
736 * In the golden day, when everything among i386 and x86_64 will be
737 * integrated, this will not live here
738 */
739void *x86_cpu_to_node_map_early_ptr;
740int x86_cpu_to_node_map_init[NR_CPUS] = {
741 [0 ... NR_CPUS-1] = NUMA_NO_NODE
742};
743DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE;
744#endif
745
746/*
747 * Determine if we were loaded by an EFI loader. If so, then we have also been
748 * passed the efi memmap, systab, etc., so we should use these data structures
749 * for initialization. Note, the efi init code path is determined by the
750 * global efi_enabled. This allows the same kernel image to be used on existing
751 * systems (with a traditional BIOS) as well as on EFI systems.
752 */
753void __init setup_arch(char **cmdline_p)
754{
755 unsigned long max_low_pfn;
756
757 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
758 pre_setup_arch_hook();
759 early_cpu_init();
760 early_ioremap_init();
761
762#ifdef CONFIG_EFI
763 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
764 "EL32", 4))
765 efi_enabled = 1;
766#endif
767
768 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
769 screen_info = boot_params.screen_info;
770 edid_info = boot_params.edid_info;
771 apm_info.bios = boot_params.apm_bios_info;
772 ist_info = boot_params.ist_info;
773 saved_video_mode = boot_params.hdr.vid_mode;
774 if( boot_params.sys_desc_table.length != 0 ) {
775 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
776 machine_id = boot_params.sys_desc_table.table[0];
777 machine_submodel_id = boot_params.sys_desc_table.table[1];
778 BIOS_revision = boot_params.sys_desc_table.table[2];
779 }
780 bootloader_type = boot_params.hdr.type_of_loader;
781
782#ifdef CONFIG_BLK_DEV_RAM
783 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
784 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
785 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
786#endif
787 ARCH_SETUP
788
789 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
790 print_memory_map(memory_setup());
791
792 copy_edd();
793
794 if (!boot_params.hdr.root_flags)
795 root_mountflags &= ~MS_RDONLY;
796 init_mm.start_code = (unsigned long) _text;
797 init_mm.end_code = (unsigned long) _etext;
798 init_mm.end_data = (unsigned long) _edata;
799 init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
800
801 code_resource.start = virt_to_phys(_text);
802 code_resource.end = virt_to_phys(_etext)-1;
803 data_resource.start = virt_to_phys(_etext);
804 data_resource.end = virt_to_phys(_edata)-1;
805 bss_resource.start = virt_to_phys(&__bss_start);
806 bss_resource.end = virt_to_phys(&__bss_stop)-1;
807
808 parse_early_param();
809
810 if (user_defined_memmap) {
811 printk(KERN_INFO "user-defined physical RAM map:\n");
812 print_memory_map("user");
813 }
814
815 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
816 *cmdline_p = command_line;
817
818 if (efi_enabled)
819 efi_init();
820
821 /* update e820 for memory not covered by WB MTRRs */
822 propagate_e820_map();
823 mtrr_bp_init();
824 if (mtrr_trim_uncached_memory(max_pfn))
825 propagate_e820_map();
826
827 max_low_pfn = setup_memory();
828
829#ifdef CONFIG_KVM_CLOCK
830 kvmclock_init();
831#endif
832
833#ifdef CONFIG_VMI
834 /*
835 * Must be after max_low_pfn is determined, and before kernel
836 * pagetables are setup.
837 */
838 vmi_init();
839#endif
840 kvm_guest_init();
841
842 /*
843 * NOTE: before this point _nobody_ is allowed to allocate
844 * any memory using the bootmem allocator. Although the
845 * allocator is now initialised only the first 8Mb of the kernel
846 * virtual address space has been mapped. All allocations before
847 * paging_init() has completed must use the alloc_bootmem_low_pages()
848 * variant (which allocates DMA'able memory) and care must be taken
849 * not to exceed the 8Mb limit.
850 */
851
852#ifdef CONFIG_SMP
853 smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
854#endif
855 paging_init();
856
857 /*
858 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
859 */
860
861#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
862 if (init_ohci1394_dma_early)
863 init_ohci1394_dma_on_all_controllers();
864#endif
865
866 remapped_pgdat_init();
867 sparse_init();
868 zone_sizes_init();
869
870 /*
871 * NOTE: at this point the bootmem allocator is fully available.
872 */
873
874#ifdef CONFIG_BLK_DEV_INITRD
875 relocate_initrd();
876#endif
877
878 paravirt_post_allocator_init();
879
880 dmi_scan_machine();
881
882 io_delay_init();
883
884#ifdef CONFIG_X86_SMP
885 /*
886 * setup to use the early static init tables during kernel startup
887 * X86_SMP will exclude sub-arches that don't deal well with it.
888 */
889 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
890 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
891#ifdef CONFIG_NUMA
892 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
893#endif
894#endif
895
896#ifdef CONFIG_X86_GENERICARCH
897 generic_apic_probe();
898#endif
899
900#ifdef CONFIG_ACPI
901 /*
902 * Parse the ACPI tables for possible boot-time SMP configuration.
903 */
904 acpi_boot_table_init();
905#endif
906
907 early_quirks();
908
909#ifdef CONFIG_ACPI
910 acpi_boot_init();
911
912#if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
913 if (def_to_bigsmp)
914 printk(KERN_WARNING "More than 8 CPUs detected and "
915 "CONFIG_X86_PC cannot handle it.\nUse "
916 "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
917#endif
918#endif
919#ifdef CONFIG_X86_LOCAL_APIC
920 if (smp_found_config)
921 get_smp_config();
922#endif
923
924 e820_register_memory();
925 e820_mark_nosave_regions();
926
927#ifdef CONFIG_VT
928#if defined(CONFIG_VGA_CONSOLE)
929 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
930 conswitchp = &vga_con;
931#elif defined(CONFIG_DUMMY_CONSOLE)
932 conswitchp = &dummy_con;
933#endif
934#endif
935}
936
937/*
938 * Request address space for all standard resources
939 *
940 * This is called just before pcibios_init(), which is also a
941 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
942 */
943static int __init request_standard_resources(void)
944{
945 int i;
946
947 printk(KERN_INFO "Setting up standard PCI resources\n");
948 init_iomem_resources(&code_resource, &data_resource, &bss_resource);
949
950 request_resource(&iomem_resource, &video_ram_resource);
951
952 /* request I/O space for devices used on all i[345]86 PCs */
953 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
954 request_resource(&ioport_resource, &standard_io_resources[i]);
955 return 0;
956}
957
958subsys_initcall(request_standard_resources);
diff --git a/arch/x86/kernel/setup_64.c b/arch/x86/kernel/setup_64.c
deleted file mode 100644
index 6dff1286ad8a..000000000000
--- a/arch/x86/kernel/setup_64.c
+++ /dev/null
@@ -1,1194 +0,0 @@
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 */
4
5/*
6 * This file handles the architecture-dependent parts of initialization
7 */
8
9#include <linux/errno.h>
10#include <linux/sched.h>
11#include <linux/kernel.h>
12#include <linux/mm.h>
13#include <linux/stddef.h>
14#include <linux/unistd.h>
15#include <linux/ptrace.h>
16#include <linux/slab.h>
17#include <linux/user.h>
18#include <linux/screen_info.h>
19#include <linux/ioport.h>
20#include <linux/delay.h>
21#include <linux/init.h>
22#include <linux/initrd.h>
23#include <linux/highmem.h>
24#include <linux/bootmem.h>
25#include <linux/module.h>
26#include <asm/processor.h>
27#include <linux/console.h>
28#include <linux/seq_file.h>
29#include <linux/crash_dump.h>
30#include <linux/root_dev.h>
31#include <linux/pci.h>
32#include <asm/pci-direct.h>
33#include <linux/efi.h>
34#include <linux/acpi.h>
35#include <linux/kallsyms.h>
36#include <linux/edd.h>
37#include <linux/iscsi_ibft.h>
38#include <linux/mmzone.h>
39#include <linux/kexec.h>
40#include <linux/cpufreq.h>
41#include <linux/dmi.h>
42#include <linux/dma-mapping.h>
43#include <linux/ctype.h>
44#include <linux/sort.h>
45#include <linux/uaccess.h>
46#include <linux/init_ohci1394_dma.h>
47#include <linux/kvm_para.h>
48
49#include <asm/mtrr.h>
50#include <asm/uaccess.h>
51#include <asm/system.h>
52#include <asm/vsyscall.h>
53#include <asm/io.h>
54#include <asm/smp.h>
55#include <asm/msr.h>
56#include <asm/desc.h>
57#include <video/edid.h>
58#include <asm/e820.h>
59#include <asm/dma.h>
60#include <asm/gart.h>
61#include <asm/mpspec.h>
62#include <asm/mmu_context.h>
63#include <asm/proto.h>
64#include <asm/setup.h>
65#include <asm/numa.h>
66#include <asm/sections.h>
67#include <asm/dmi.h>
68#include <asm/cacheflush.h>
69#include <asm/mce.h>
70#include <asm/ds.h>
71#include <asm/topology.h>
72#include <asm/trampoline.h>
73#include <asm/pat.h>
74
75#include <mach_apic.h>
76#ifdef CONFIG_PARAVIRT
77#include <asm/paravirt.h>
78#else
79#define ARCH_SETUP
80#endif
81
82/*
83 * Machine setup..
84 */
85
86struct cpuinfo_x86 boot_cpu_data __read_mostly;
87EXPORT_SYMBOL(boot_cpu_data);
88
89__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
90
91unsigned long mmu_cr4_features;
92
93/* Boot loader ID as an integer, for the benefit of proc_dointvec */
94int bootloader_type;
95
96unsigned long saved_video_mode;
97
98int force_mwait __cpuinitdata;
99
100/*
101 * Early DMI memory
102 */
103int dmi_alloc_index;
104char dmi_alloc_data[DMI_MAX_DATA];
105
106/*
107 * Setup options
108 */
109struct screen_info screen_info;
110EXPORT_SYMBOL(screen_info);
111struct sys_desc_table_struct {
112 unsigned short length;
113 unsigned char table[0];
114};
115
116struct edid_info edid_info;
117EXPORT_SYMBOL_GPL(edid_info);
118
119extern int root_mountflags;
120
121char __initdata command_line[COMMAND_LINE_SIZE];
122
123static struct resource standard_io_resources[] = {
124 { .name = "dma1", .start = 0x00, .end = 0x1f,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "pic1", .start = 0x20, .end = 0x21,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "timer0", .start = 0x40, .end = 0x43,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "timer1", .start = 0x50, .end = 0x53,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "keyboard", .start = 0x60, .end = 0x60,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
134 { .name = "keyboard", .start = 0x64, .end = 0x64,
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
136 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
137 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
138 { .name = "pic2", .start = 0xa0, .end = 0xa1,
139 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
140 { .name = "dma2", .start = 0xc0, .end = 0xdf,
141 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
142 { .name = "fpu", .start = 0xf0, .end = 0xff,
143 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
144};
145
146#define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
147
148static struct resource data_resource = {
149 .name = "Kernel data",
150 .start = 0,
151 .end = 0,
152 .flags = IORESOURCE_RAM,
153};
154static struct resource code_resource = {
155 .name = "Kernel code",
156 .start = 0,
157 .end = 0,
158 .flags = IORESOURCE_RAM,
159};
160static struct resource bss_resource = {
161 .name = "Kernel bss",
162 .start = 0,
163 .end = 0,
164 .flags = IORESOURCE_RAM,
165};
166
167static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
168
169#ifdef CONFIG_PROC_VMCORE
170/* elfcorehdr= specifies the location of elf core header
171 * stored by the crashed kernel. This option will be passed
172 * by kexec loader to the capture kernel.
173 */
174static int __init setup_elfcorehdr(char *arg)
175{
176 char *end;
177 if (!arg)
178 return -EINVAL;
179 elfcorehdr_addr = memparse(arg, &end);
180 return end > arg ? 0 : -EINVAL;
181}
182early_param("elfcorehdr", setup_elfcorehdr);
183#endif
184
185#ifndef CONFIG_NUMA
186static void __init
187contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
188{
189 unsigned long bootmap_size, bootmap;
190
191 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
192 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
193 PAGE_SIZE);
194 if (bootmap == -1L)
195 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
196 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
197 e820_register_active_regions(0, start_pfn, end_pfn);
198 free_bootmem_with_active_regions(0, end_pfn);
199 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
200 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
201}
202#endif
203
204#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
205struct edd edd;
206#ifdef CONFIG_EDD_MODULE
207EXPORT_SYMBOL(edd);
208#endif
209/**
210 * copy_edd() - Copy the BIOS EDD information
211 * from boot_params into a safe place.
212 *
213 */
214static inline void copy_edd(void)
215{
216 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
217 sizeof(edd.mbr_signature));
218 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
219 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
220 edd.edd_info_nr = boot_params.eddbuf_entries;
221}
222#else
223static inline void copy_edd(void)
224{
225}
226#endif
227
228#ifdef CONFIG_KEXEC
229static void __init reserve_crashkernel(void)
230{
231 unsigned long long total_mem;
232 unsigned long long crash_size, crash_base;
233 int ret;
234
235 total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
236
237 ret = parse_crashkernel(boot_command_line, total_mem,
238 &crash_size, &crash_base);
239 if (ret == 0 && crash_size) {
240 if (crash_base <= 0) {
241 printk(KERN_INFO "crashkernel reservation failed - "
242 "you have to specify a base address\n");
243 return;
244 }
245
246 if (reserve_bootmem(crash_base, crash_size,
247 BOOTMEM_EXCLUSIVE) < 0) {
248 printk(KERN_INFO "crashkernel reservation failed - "
249 "memory is in use\n");
250 return;
251 }
252
253 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
254 "for crashkernel (System RAM: %ldMB)\n",
255 (unsigned long)(crash_size >> 20),
256 (unsigned long)(crash_base >> 20),
257 (unsigned long)(total_mem >> 20));
258 crashk_res.start = crash_base;
259 crashk_res.end = crash_base + crash_size - 1;
260 insert_resource(&iomem_resource, &crashk_res);
261 }
262}
263#else
264static inline void __init reserve_crashkernel(void)
265{}
266#endif
267
268/* Overridden in paravirt.c if CONFIG_PARAVIRT */
269void __attribute__((weak)) __init memory_setup(void)
270{
271 machine_specific_memory_setup();
272}
273
274static void __init parse_setup_data(void)
275{
276 struct setup_data *data;
277 unsigned long pa_data;
278
279 if (boot_params.hdr.version < 0x0209)
280 return;
281 pa_data = boot_params.hdr.setup_data;
282 while (pa_data) {
283 data = early_ioremap(pa_data, PAGE_SIZE);
284 switch (data->type) {
285 default:
286 break;
287 }
288#ifndef CONFIG_DEBUG_BOOT_PARAMS
289 free_early(pa_data, pa_data+sizeof(*data)+data->len);
290#endif
291 pa_data = data->next;
292 early_iounmap(data, PAGE_SIZE);
293 }
294}
295
296#ifdef CONFIG_PCI_MMCONFIG
297extern void __cpuinit fam10h_check_enable_mmcfg(void);
298extern void __init check_enable_amd_mmconf_dmi(void);
299#else
300void __cpuinit fam10h_check_enable_mmcfg(void)
301{
302}
303void __init check_enable_amd_mmconf_dmi(void)
304{
305}
306#endif
307
308/*
309 * setup_arch - architecture-specific boot-time initializations
310 *
311 * Note: On x86_64, fixmaps are ready for use even before this is called.
312 */
313void __init setup_arch(char **cmdline_p)
314{
315 unsigned i;
316
317 printk(KERN_INFO "Command line: %s\n", boot_command_line);
318
319 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
320 screen_info = boot_params.screen_info;
321 edid_info = boot_params.edid_info;
322 saved_video_mode = boot_params.hdr.vid_mode;
323 bootloader_type = boot_params.hdr.type_of_loader;
324
325#ifdef CONFIG_BLK_DEV_RAM
326 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
327 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
328 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
329#endif
330#ifdef CONFIG_EFI
331 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
332 "EL64", 4))
333 efi_enabled = 1;
334#endif
335
336 ARCH_SETUP
337
338 memory_setup();
339 copy_edd();
340
341 if (!boot_params.hdr.root_flags)
342 root_mountflags &= ~MS_RDONLY;
343 init_mm.start_code = (unsigned long) &_text;
344 init_mm.end_code = (unsigned long) &_etext;
345 init_mm.end_data = (unsigned long) &_edata;
346 init_mm.brk = (unsigned long) &_end;
347
348 code_resource.start = virt_to_phys(&_text);
349 code_resource.end = virt_to_phys(&_etext)-1;
350 data_resource.start = virt_to_phys(&_etext);
351 data_resource.end = virt_to_phys(&_edata)-1;
352 bss_resource.start = virt_to_phys(&__bss_start);
353 bss_resource.end = virt_to_phys(&__bss_stop)-1;
354
355 early_identify_cpu(&boot_cpu_data);
356
357 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
358 *cmdline_p = command_line;
359
360 parse_setup_data();
361
362 parse_early_param();
363
364#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
365 if (init_ohci1394_dma_early)
366 init_ohci1394_dma_on_all_controllers();
367#endif
368
369 finish_e820_parsing();
370
371 /* after parse_early_param, so could debug it */
372 insert_resource(&iomem_resource, &code_resource);
373 insert_resource(&iomem_resource, &data_resource);
374 insert_resource(&iomem_resource, &bss_resource);
375
376 early_gart_iommu_check();
377
378 e820_register_active_regions(0, 0, -1UL);
379 /*
380 * partially used pages are not usable - thus
381 * we are rounding upwards:
382 */
383 end_pfn = e820_end_of_ram();
384 /* update e820 for memory not covered by WB MTRRs */
385 mtrr_bp_init();
386 if (mtrr_trim_uncached_memory(end_pfn)) {
387 e820_register_active_regions(0, 0, -1UL);
388 end_pfn = e820_end_of_ram();
389 }
390
391 num_physpages = end_pfn;
392
393 check_efer();
394
395 max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
396 if (efi_enabled)
397 efi_init();
398
399 vsmp_init();
400
401 dmi_scan_machine();
402
403 io_delay_init();
404
405#ifdef CONFIG_KVM_CLOCK
406 kvmclock_init();
407#endif
408
409#ifdef CONFIG_SMP
410 /* setup to use the early static init tables during kernel startup */
411 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
412 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
413#ifdef CONFIG_NUMA
414 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
415#endif
416#endif
417
418#ifdef CONFIG_ACPI
419 /*
420 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
421 * Call this early for SRAT node setup.
422 */
423 acpi_boot_table_init();
424#endif
425
426 /* How many end-of-memory variables you have, grandma! */
427 max_low_pfn = end_pfn;
428 max_pfn = end_pfn;
429 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
430
431 /* Remove active ranges so rediscovery with NUMA-awareness happens */
432 remove_all_active_ranges();
433
434#ifdef CONFIG_ACPI_NUMA
435 /*
436 * Parse SRAT to discover nodes.
437 */
438 acpi_numa_init();
439#endif
440
441#ifdef CONFIG_NUMA
442 numa_initmem_init(0, end_pfn);
443#else
444 contig_initmem_init(0, end_pfn);
445#endif
446
447 dma32_reserve_bootmem();
448
449#ifdef CONFIG_ACPI_SLEEP
450 /*
451 * Reserve low memory region for sleep support.
452 */
453 acpi_reserve_bootmem();
454#endif
455
456 if (efi_enabled)
457 efi_reserve_bootmem();
458
459 /*
460 * Find and reserve possible boot-time SMP configuration:
461 */
462 find_smp_config();
463#ifdef CONFIG_BLK_DEV_INITRD
464 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
465 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
466 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
467 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
468 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
469
470 if (ramdisk_end <= end_of_mem) {
471 /*
472 * don't need to reserve again, already reserved early
473 * in x86_64_start_kernel, and early_res_to_bootmem
474 * convert that to reserved in bootmem
475 */
476 initrd_start = ramdisk_image + PAGE_OFFSET;
477 initrd_end = initrd_start+ramdisk_size;
478 } else {
479 free_bootmem(ramdisk_image, ramdisk_size);
480 printk(KERN_ERR "initrd extends beyond end of memory "
481 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
482 ramdisk_end, end_of_mem);
483 initrd_start = 0;
484 }
485 }
486#endif
487 reserve_crashkernel();
488
489 reserve_ibft_region();
490
491 paging_init();
492 map_vsyscall();
493
494 early_quirks();
495
496#ifdef CONFIG_ACPI
497 /*
498 * Read APIC and some other early information from ACPI tables.
499 */
500 acpi_boot_init();
501#endif
502
503 init_cpu_to_node();
504
505 /*
506 * get boot-time SMP configuration:
507 */
508 if (smp_found_config)
509 get_smp_config();
510 init_apic_mappings();
511 ioapic_init_mappings();
512
513 kvm_guest_init();
514
515 /*
516 * We trust e820 completely. No explicit ROM probing in memory.
517 */
518 e820_reserve_resources();
519 e820_mark_nosave_regions();
520
521 /* request I/O space for devices used on all i[345]86 PCs */
522 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
523 request_resource(&ioport_resource, &standard_io_resources[i]);
524
525 e820_setup_gap();
526
527#ifdef CONFIG_VT
528#if defined(CONFIG_VGA_CONSOLE)
529 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
530 conswitchp = &vga_con;
531#elif defined(CONFIG_DUMMY_CONSOLE)
532 conswitchp = &dummy_con;
533#endif
534#endif
535
536 /* do this before identify_cpu for boot cpu */
537 check_enable_amd_mmconf_dmi();
538}
539
540static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
541{
542 unsigned int *v;
543
544 if (c->extended_cpuid_level < 0x80000004)
545 return 0;
546
547 v = (unsigned int *) c->x86_model_id;
548 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
549 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
550 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
551 c->x86_model_id[48] = 0;
552 return 1;
553}
554
555
556static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
557{
558 unsigned int n, dummy, eax, ebx, ecx, edx;
559
560 n = c->extended_cpuid_level;
561
562 if (n >= 0x80000005) {
563 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
564 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
565 "D cache %dK (%d bytes/line)\n",
566 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
567 c->x86_cache_size = (ecx>>24) + (edx>>24);
568 /* On K8 L1 TLB is inclusive, so don't count it */
569 c->x86_tlbsize = 0;
570 }
571
572 if (n >= 0x80000006) {
573 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
574 ecx = cpuid_ecx(0x80000006);
575 c->x86_cache_size = ecx >> 16;
576 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
577
578 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
579 c->x86_cache_size, ecx & 0xFF);
580 }
581 if (n >= 0x80000008) {
582 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
583 c->x86_virt_bits = (eax >> 8) & 0xff;
584 c->x86_phys_bits = eax & 0xff;
585 }
586}
587
588#ifdef CONFIG_NUMA
589static int __cpuinit nearby_node(int apicid)
590{
591 int i, node;
592
593 for (i = apicid - 1; i >= 0; i--) {
594 node = apicid_to_node[i];
595 if (node != NUMA_NO_NODE && node_online(node))
596 return node;
597 }
598 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
599 node = apicid_to_node[i];
600 if (node != NUMA_NO_NODE && node_online(node))
601 return node;
602 }
603 return first_node(node_online_map); /* Shouldn't happen */
604}
605#endif
606
607/*
608 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
609 * Assumes number of cores is a power of two.
610 */
611static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
612{
613#ifdef CONFIG_SMP
614 unsigned bits;
615#ifdef CONFIG_NUMA
616 int cpu = smp_processor_id();
617 int node = 0;
618 unsigned apicid = hard_smp_processor_id();
619#endif
620 bits = c->x86_coreid_bits;
621
622 /* Low order bits define the core id (index of core in socket) */
623 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
624 /* Convert the initial APIC ID into the socket ID */
625 c->phys_proc_id = c->initial_apicid >> bits;
626
627#ifdef CONFIG_NUMA
628 node = c->phys_proc_id;
629 if (apicid_to_node[apicid] != NUMA_NO_NODE)
630 node = apicid_to_node[apicid];
631 if (!node_online(node)) {
632 /* Two possibilities here:
633 - The CPU is missing memory and no node was created.
634 In that case try picking one from a nearby CPU
635 - The APIC IDs differ from the HyperTransport node IDs
636 which the K8 northbridge parsing fills in.
637 Assume they are all increased by a constant offset,
638 but in the same order as the HT nodeids.
639 If that doesn't result in a usable node fall back to the
640 path for the previous case. */
641
642 int ht_nodeid = c->initial_apicid;
643
644 if (ht_nodeid >= 0 &&
645 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
646 node = apicid_to_node[ht_nodeid];
647 /* Pick a nearby node */
648 if (!node_online(node))
649 node = nearby_node(apicid);
650 }
651 numa_set_node(cpu, node);
652
653 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
654#endif
655#endif
656}
657
658static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
659{
660#ifdef CONFIG_SMP
661 unsigned bits, ecx;
662
663 /* Multi core CPU? */
664 if (c->extended_cpuid_level < 0x80000008)
665 return;
666
667 ecx = cpuid_ecx(0x80000008);
668
669 c->x86_max_cores = (ecx & 0xff) + 1;
670
671 /* CPU telling us the core id bits shift? */
672 bits = (ecx >> 12) & 0xF;
673
674 /* Otherwise recompute */
675 if (bits == 0) {
676 while ((1 << bits) < c->x86_max_cores)
677 bits++;
678 }
679
680 c->x86_coreid_bits = bits;
681
682#endif
683}
684
685#define ENABLE_C1E_MASK 0x18000000
686#define CPUID_PROCESSOR_SIGNATURE 1
687#define CPUID_XFAM 0x0ff00000
688#define CPUID_XFAM_K8 0x00000000
689#define CPUID_XFAM_10H 0x00100000
690#define CPUID_XFAM_11H 0x00200000
691#define CPUID_XMOD 0x000f0000
692#define CPUID_XMOD_REV_F 0x00040000
693
694/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
695static __cpuinit int amd_apic_timer_broken(void)
696{
697 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
698
699 switch (eax & CPUID_XFAM) {
700 case CPUID_XFAM_K8:
701 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
702 break;
703 case CPUID_XFAM_10H:
704 case CPUID_XFAM_11H:
705 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
706 if (lo & ENABLE_C1E_MASK)
707 return 1;
708 break;
709 default:
710 /* err on the side of caution */
711 return 1;
712 }
713 return 0;
714}
715
716static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
717{
718 early_init_amd_mc(c);
719
720 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
721 if (c->x86_power & (1<<8))
722 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
723}
724
725static void __cpuinit init_amd(struct cpuinfo_x86 *c)
726{
727 unsigned level;
728
729#ifdef CONFIG_SMP
730 unsigned long value;
731
732 /*
733 * Disable TLB flush filter by setting HWCR.FFDIS on K8
734 * bit 6 of msr C001_0015
735 *
736 * Errata 63 for SH-B3 steppings
737 * Errata 122 for all steppings (F+ have it disabled by default)
738 */
739 if (c->x86 == 15) {
740 rdmsrl(MSR_K8_HWCR, value);
741 value |= 1 << 6;
742 wrmsrl(MSR_K8_HWCR, value);
743 }
744#endif
745
746 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
747 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
748 clear_cpu_cap(c, 0*32+31);
749
750 /* On C+ stepping K8 rep microcode works well for copy/memset */
751 level = cpuid_eax(1);
752 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
753 level >= 0x0f58))
754 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
755 if (c->x86 == 0x10 || c->x86 == 0x11)
756 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
757
758 /* Enable workaround for FXSAVE leak */
759 if (c->x86 >= 6)
760 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
761
762 level = get_model_name(c);
763 if (!level) {
764 switch (c->x86) {
765 case 15:
766 /* Should distinguish Models here, but this is only
767 a fallback anyways. */
768 strcpy(c->x86_model_id, "Hammer");
769 break;
770 }
771 }
772 display_cacheinfo(c);
773
774 /* Multi core CPU? */
775 if (c->extended_cpuid_level >= 0x80000008)
776 amd_detect_cmp(c);
777
778 if (c->extended_cpuid_level >= 0x80000006 &&
779 (cpuid_edx(0x80000006) & 0xf000))
780 num_cache_leaves = 4;
781 else
782 num_cache_leaves = 3;
783
784 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
785 set_cpu_cap(c, X86_FEATURE_K8);
786
787 /* MFENCE stops RDTSC speculation */
788 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
789
790 if (c->x86 == 0x10)
791 fam10h_check_enable_mmcfg();
792
793 if (amd_apic_timer_broken())
794 disable_apic_timer = 1;
795
796 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
797 unsigned long long tseg;
798
799 /*
800 * Split up direct mapping around the TSEG SMM area.
801 * Don't do it for gbpages because there seems very little
802 * benefit in doing so.
803 */
804 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
805 (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
806 set_memory_4k((unsigned long)__va(tseg), 1);
807 }
808}
809
810void __cpuinit detect_ht(struct cpuinfo_x86 *c)
811{
812#ifdef CONFIG_SMP
813 u32 eax, ebx, ecx, edx;
814 int index_msb, core_bits;
815
816 cpuid(1, &eax, &ebx, &ecx, &edx);
817
818
819 if (!cpu_has(c, X86_FEATURE_HT))
820 return;
821 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
822 goto out;
823
824 smp_num_siblings = (ebx & 0xff0000) >> 16;
825
826 if (smp_num_siblings == 1) {
827 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
828 } else if (smp_num_siblings > 1) {
829
830 if (smp_num_siblings > NR_CPUS) {
831 printk(KERN_WARNING "CPU: Unsupported number of "
832 "siblings %d", smp_num_siblings);
833 smp_num_siblings = 1;
834 return;
835 }
836
837 index_msb = get_count_order(smp_num_siblings);
838 c->phys_proc_id = phys_pkg_id(index_msb);
839
840 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
841
842 index_msb = get_count_order(smp_num_siblings);
843
844 core_bits = get_count_order(c->x86_max_cores);
845
846 c->cpu_core_id = phys_pkg_id(index_msb) &
847 ((1 << core_bits) - 1);
848 }
849out:
850 if ((c->x86_max_cores * smp_num_siblings) > 1) {
851 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
852 c->phys_proc_id);
853 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
854 c->cpu_core_id);
855 }
856
857#endif
858}
859
860/*
861 * find out the number of processor cores on the die
862 */
863static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
864{
865 unsigned int eax, t;
866
867 if (c->cpuid_level < 4)
868 return 1;
869
870 cpuid_count(4, 0, &eax, &t, &t, &t);
871
872 if (eax & 0x1f)
873 return ((eax >> 26) + 1);
874 else
875 return 1;
876}
877
878static void __cpuinit srat_detect_node(void)
879{
880#ifdef CONFIG_NUMA
881 unsigned node;
882 int cpu = smp_processor_id();
883 int apicid = hard_smp_processor_id();
884
885 /* Don't do the funky fallback heuristics the AMD version employs
886 for now. */
887 node = apicid_to_node[apicid];
888 if (node == NUMA_NO_NODE || !node_online(node))
889 node = first_node(node_online_map);
890 numa_set_node(cpu, node);
891
892 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
893#endif
894}
895
896static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
897{
898 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
899 (c->x86 == 0x6 && c->x86_model >= 0x0e))
900 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
901}
902
903static void __cpuinit init_intel(struct cpuinfo_x86 *c)
904{
905 /* Cache sizes */
906 unsigned n;
907
908 init_intel_cacheinfo(c);
909 if (c->cpuid_level > 9) {
910 unsigned eax = cpuid_eax(10);
911 /* Check for version and the number of counters */
912 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
913 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
914 }
915
916 if (cpu_has_ds) {
917 unsigned int l1, l2;
918 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
919 if (!(l1 & (1<<11)))
920 set_cpu_cap(c, X86_FEATURE_BTS);
921 if (!(l1 & (1<<12)))
922 set_cpu_cap(c, X86_FEATURE_PEBS);
923 }
924
925
926 if (cpu_has_bts)
927 ds_init_intel(c);
928
929 n = c->extended_cpuid_level;
930 if (n >= 0x80000008) {
931 unsigned eax = cpuid_eax(0x80000008);
932 c->x86_virt_bits = (eax >> 8) & 0xff;
933 c->x86_phys_bits = eax & 0xff;
934 /* CPUID workaround for Intel 0F34 CPU */
935 if (c->x86_vendor == X86_VENDOR_INTEL &&
936 c->x86 == 0xF && c->x86_model == 0x3 &&
937 c->x86_mask == 0x4)
938 c->x86_phys_bits = 36;
939 }
940
941 if (c->x86 == 15)
942 c->x86_cache_alignment = c->x86_clflush_size * 2;
943 if (c->x86 == 6)
944 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
945 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
946 c->x86_max_cores = intel_num_cpu_cores(c);
947
948 srat_detect_node();
949}
950
951static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
952{
953 if (c->x86 == 0x6 && c->x86_model >= 0xf)
954 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
955}
956
957static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
958{
959 /* Cache sizes */
960 unsigned n;
961
962 n = c->extended_cpuid_level;
963 if (n >= 0x80000008) {
964 unsigned eax = cpuid_eax(0x80000008);
965 c->x86_virt_bits = (eax >> 8) & 0xff;
966 c->x86_phys_bits = eax & 0xff;
967 }
968
969 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
970 c->x86_cache_alignment = c->x86_clflush_size * 2;
971 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
972 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
973 }
974 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
975}
976
977static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
978{
979 char *v = c->x86_vendor_id;
980
981 if (!strcmp(v, "AuthenticAMD"))
982 c->x86_vendor = X86_VENDOR_AMD;
983 else if (!strcmp(v, "GenuineIntel"))
984 c->x86_vendor = X86_VENDOR_INTEL;
985 else if (!strcmp(v, "CentaurHauls"))
986 c->x86_vendor = X86_VENDOR_CENTAUR;
987 else
988 c->x86_vendor = X86_VENDOR_UNKNOWN;
989}
990
991/* Do some early cpuid on the boot CPU to get some parameter that are
992 needed before check_bugs. Everything advanced is in identify_cpu
993 below. */
994static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
995{
996 u32 tfms, xlvl;
997
998 c->loops_per_jiffy = loops_per_jiffy;
999 c->x86_cache_size = -1;
1000 c->x86_vendor = X86_VENDOR_UNKNOWN;
1001 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1002 c->x86_vendor_id[0] = '\0'; /* Unset */
1003 c->x86_model_id[0] = '\0'; /* Unset */
1004 c->x86_clflush_size = 64;
1005 c->x86_cache_alignment = c->x86_clflush_size;
1006 c->x86_max_cores = 1;
1007 c->x86_coreid_bits = 0;
1008 c->extended_cpuid_level = 0;
1009 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1010
1011 /* Get vendor name */
1012 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1013 (unsigned int *)&c->x86_vendor_id[0],
1014 (unsigned int *)&c->x86_vendor_id[8],
1015 (unsigned int *)&c->x86_vendor_id[4]);
1016
1017 get_cpu_vendor(c);
1018
1019 /* Initialize the standard set of capabilities */
1020 /* Note that the vendor-specific code below might override */
1021
1022 /* Intel-defined flags: level 0x00000001 */
1023 if (c->cpuid_level >= 0x00000001) {
1024 __u32 misc;
1025 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1026 &c->x86_capability[0]);
1027 c->x86 = (tfms >> 8) & 0xf;
1028 c->x86_model = (tfms >> 4) & 0xf;
1029 c->x86_mask = tfms & 0xf;
1030 if (c->x86 == 0xf)
1031 c->x86 += (tfms >> 20) & 0xff;
1032 if (c->x86 >= 0x6)
1033 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1034 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1035 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1036 } else {
1037 /* Have CPUID level 0 only - unheard of */
1038 c->x86 = 4;
1039 }
1040
1041 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
1042#ifdef CONFIG_SMP
1043 c->phys_proc_id = c->initial_apicid;
1044#endif
1045 /* AMD-defined flags: level 0x80000001 */
1046 xlvl = cpuid_eax(0x80000000);
1047 c->extended_cpuid_level = xlvl;
1048 if ((xlvl & 0xffff0000) == 0x80000000) {
1049 if (xlvl >= 0x80000001) {
1050 c->x86_capability[1] = cpuid_edx(0x80000001);
1051 c->x86_capability[6] = cpuid_ecx(0x80000001);
1052 }
1053 if (xlvl >= 0x80000004)
1054 get_model_name(c); /* Default name */
1055 }
1056
1057 /* Transmeta-defined flags: level 0x80860001 */
1058 xlvl = cpuid_eax(0x80860000);
1059 if ((xlvl & 0xffff0000) == 0x80860000) {
1060 /* Don't set x86_cpuid_level here for now to not confuse. */
1061 if (xlvl >= 0x80860001)
1062 c->x86_capability[2] = cpuid_edx(0x80860001);
1063 }
1064
1065 c->extended_cpuid_level = cpuid_eax(0x80000000);
1066 if (c->extended_cpuid_level >= 0x80000007)
1067 c->x86_power = cpuid_edx(0x80000007);
1068
1069 switch (c->x86_vendor) {
1070 case X86_VENDOR_AMD:
1071 early_init_amd(c);
1072 break;
1073 case X86_VENDOR_INTEL:
1074 early_init_intel(c);
1075 break;
1076 case X86_VENDOR_CENTAUR:
1077 early_init_centaur(c);
1078 break;
1079 }
1080
1081 validate_pat_support(c);
1082}
1083
1084/*
1085 * This does the hard work of actually picking apart the CPU stuff...
1086 */
1087void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1088{
1089 int i;
1090
1091 early_identify_cpu(c);
1092
1093 init_scattered_cpuid_features(c);
1094
1095 c->apicid = phys_pkg_id(0);
1096
1097 /*
1098 * Vendor-specific initialization. In this section we
1099 * canonicalize the feature flags, meaning if there are
1100 * features a certain CPU supports which CPUID doesn't
1101 * tell us, CPUID claiming incorrect flags, or other bugs,
1102 * we handle them here.
1103 *
1104 * At the end of this section, c->x86_capability better
1105 * indicate the features this CPU genuinely supports!
1106 */
1107 switch (c->x86_vendor) {
1108 case X86_VENDOR_AMD:
1109 init_amd(c);
1110 break;
1111
1112 case X86_VENDOR_INTEL:
1113 init_intel(c);
1114 break;
1115
1116 case X86_VENDOR_CENTAUR:
1117 init_centaur(c);
1118 break;
1119
1120 case X86_VENDOR_UNKNOWN:
1121 default:
1122 display_cacheinfo(c);
1123 break;
1124 }
1125
1126 detect_ht(c);
1127
1128 /*
1129 * On SMP, boot_cpu_data holds the common feature set between
1130 * all CPUs; so make sure that we indicate which features are
1131 * common between the CPUs. The first time this routine gets
1132 * executed, c == &boot_cpu_data.
1133 */
1134 if (c != &boot_cpu_data) {
1135 /* AND the already accumulated flags with these */
1136 for (i = 0; i < NCAPINTS; i++)
1137 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1138 }
1139
1140 /* Clear all flags overriden by options */
1141 for (i = 0; i < NCAPINTS; i++)
1142 c->x86_capability[i] &= ~cleared_cpu_caps[i];
1143
1144#ifdef CONFIG_X86_MCE
1145 mcheck_init(c);
1146#endif
1147 select_idle_routine(c);
1148
1149#ifdef CONFIG_NUMA
1150 numa_add_cpu(smp_processor_id());
1151#endif
1152
1153}
1154
1155void __cpuinit identify_boot_cpu(void)
1156{
1157 identify_cpu(&boot_cpu_data);
1158}
1159
1160void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
1161{
1162 BUG_ON(c == &boot_cpu_data);
1163 identify_cpu(c);
1164 mtrr_ap_init();
1165}
1166
1167static __init int setup_noclflush(char *arg)
1168{
1169 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1170 return 1;
1171}
1172__setup("noclflush", setup_noclflush);
1173
1174void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1175{
1176 if (c->x86_model_id[0])
1177 printk(KERN_CONT "%s", c->x86_model_id);
1178
1179 if (c->x86_mask || c->cpuid_level >= 0)
1180 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1181 else
1182 printk(KERN_CONT "\n");
1183}
1184
1185static __init int setup_disablecpuid(char *arg)
1186{
1187 int bit;
1188 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1189 setup_clear_cpu_cap(bit);
1190 else
1191 return 0;
1192 return 1;
1193}
1194__setup("clearcpuid=", setup_disablecpuid);
diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
new file mode 100644
index 000000000000..76e305e064f9
--- /dev/null
+++ b/arch/x86/kernel/setup_percpu.c
@@ -0,0 +1,378 @@
1#include <linux/kernel.h>
2#include <linux/module.h>
3#include <linux/init.h>
4#include <linux/bootmem.h>
5#include <linux/percpu.h>
6#include <linux/kexec.h>
7#include <linux/crash_dump.h>
8#include <asm/smp.h>
9#include <asm/percpu.h>
10#include <asm/sections.h>
11#include <asm/processor.h>
12#include <asm/setup.h>
13#include <asm/topology.h>
14#include <asm/mpspec.h>
15#include <asm/apicdef.h>
16#include <asm/highmem.h>
17
18#ifdef CONFIG_X86_LOCAL_APIC
19unsigned int num_processors;
20unsigned disabled_cpus __cpuinitdata;
21/* Processor that is doing the boot up */
22unsigned int boot_cpu_physical_apicid = -1U;
23unsigned int max_physical_apicid;
24EXPORT_SYMBOL(boot_cpu_physical_apicid);
25
26/* Bitmask of physically existing CPUs */
27physid_mask_t phys_cpu_present_map;
28#endif
29
30/* map cpu index to physical APIC ID */
31DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
32DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
33EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
34EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
35
36#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
37#define X86_64_NUMA 1
38
39/* map cpu index to node index */
40DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
41EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
42
43/* which logical CPUs are on which nodes */
44cpumask_t *node_to_cpumask_map;
45EXPORT_SYMBOL(node_to_cpumask_map);
46
47/* setup node_to_cpumask_map */
48static void __init setup_node_to_cpumask_map(void);
49
50#else
51static inline void setup_node_to_cpumask_map(void) { }
52#endif
53
54#if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP)
55/*
56 * Copy data used in early init routines from the initial arrays to the
57 * per cpu data areas. These arrays then become expendable and the
58 * *_early_ptr's are zeroed indicating that the static arrays are gone.
59 */
60static void __init setup_per_cpu_maps(void)
61{
62 int cpu;
63
64 for_each_possible_cpu(cpu) {
65 per_cpu(x86_cpu_to_apicid, cpu) =
66 early_per_cpu_map(x86_cpu_to_apicid, cpu);
67 per_cpu(x86_bios_cpu_apicid, cpu) =
68 early_per_cpu_map(x86_bios_cpu_apicid, cpu);
69#ifdef X86_64_NUMA
70 per_cpu(x86_cpu_to_node_map, cpu) =
71 early_per_cpu_map(x86_cpu_to_node_map, cpu);
72#endif
73 }
74
75 /* indicate the early static arrays will soon be gone */
76 early_per_cpu_ptr(x86_cpu_to_apicid) = NULL;
77 early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL;
78#ifdef X86_64_NUMA
79 early_per_cpu_ptr(x86_cpu_to_node_map) = NULL;
80#endif
81}
82
83#ifdef CONFIG_X86_32
84/*
85 * Great future not-so-futuristic plan: make i386 and x86_64 do it
86 * the same way
87 */
88unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
89EXPORT_SYMBOL(__per_cpu_offset);
90static inline void setup_cpu_pda_map(void) { }
91
92#elif !defined(CONFIG_SMP)
93static inline void setup_cpu_pda_map(void) { }
94
95#else /* CONFIG_SMP && CONFIG_X86_64 */
96
97/*
98 * Allocate cpu_pda pointer table and array via alloc_bootmem.
99 */
100static void __init setup_cpu_pda_map(void)
101{
102 char *pda;
103 struct x8664_pda **new_cpu_pda;
104 unsigned long size;
105 int cpu;
106
107 size = roundup(sizeof(struct x8664_pda), cache_line_size());
108
109 /* allocate cpu_pda array and pointer table */
110 {
111 unsigned long tsize = nr_cpu_ids * sizeof(void *);
112 unsigned long asize = size * (nr_cpu_ids - 1);
113
114 tsize = roundup(tsize, cache_line_size());
115 new_cpu_pda = alloc_bootmem(tsize + asize);
116 pda = (char *)new_cpu_pda + tsize;
117 }
118
119 /* initialize pointer table to static pda's */
120 for_each_possible_cpu(cpu) {
121 if (cpu == 0) {
122 /* leave boot cpu pda in place */
123 new_cpu_pda[0] = cpu_pda(0);
124 continue;
125 }
126 new_cpu_pda[cpu] = (struct x8664_pda *)pda;
127 new_cpu_pda[cpu]->in_bootmem = 1;
128 pda += size;
129 }
130
131 /* point to new pointer table */
132 _cpu_pda = new_cpu_pda;
133}
134#endif
135
136/*
137 * Great future plan:
138 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
139 * Always point %gs to its beginning
140 */
141void __init setup_per_cpu_areas(void)
142{
143 ssize_t size = PERCPU_ENOUGH_ROOM;
144 char *ptr;
145 int cpu;
146
147 /* Setup cpu_pda map */
148 setup_cpu_pda_map();
149
150 /* Copy section for each CPU (we discard the original) */
151 size = PERCPU_ENOUGH_ROOM;
152 printk(KERN_INFO "PERCPU: Allocating %zd bytes of per cpu data\n",
153 size);
154
155 for_each_possible_cpu(cpu) {
156#ifndef CONFIG_NEED_MULTIPLE_NODES
157 ptr = alloc_bootmem_pages(size);
158#else
159 int node = early_cpu_to_node(cpu);
160 if (!node_online(node) || !NODE_DATA(node)) {
161 ptr = alloc_bootmem_pages(size);
162 printk(KERN_INFO
163 "cpu %d has no node %d or node-local memory\n",
164 cpu, node);
165 }
166 else
167 ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
168#endif
169 per_cpu_offset(cpu) = ptr - __per_cpu_start;
170 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
171
172 }
173
174 printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n",
175 NR_CPUS, nr_cpu_ids, nr_node_ids);
176
177 /* Setup percpu data maps */
178 setup_per_cpu_maps();
179
180 /* Setup node to cpumask map */
181 setup_node_to_cpumask_map();
182}
183
184#endif
185
186#ifdef X86_64_NUMA
187
188/*
189 * Allocate node_to_cpumask_map based on number of available nodes
190 * Requires node_possible_map to be valid.
191 *
192 * Note: node_to_cpumask() is not valid until after this is done.
193 */
194static void __init setup_node_to_cpumask_map(void)
195{
196 unsigned int node, num = 0;
197 cpumask_t *map;
198
199 /* setup nr_node_ids if not done yet */
200 if (nr_node_ids == MAX_NUMNODES) {
201 for_each_node_mask(node, node_possible_map)
202 num = node;
203 nr_node_ids = num + 1;
204 }
205
206 /* allocate the map */
207 map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t));
208
209 pr_debug(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n",
210 map, nr_node_ids);
211
212 /* node_to_cpumask() will now work */
213 node_to_cpumask_map = map;
214}
215
216void __cpuinit numa_set_node(int cpu, int node)
217{
218 int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
219
220 if (cpu_pda(cpu) && node != NUMA_NO_NODE)
221 cpu_pda(cpu)->nodenumber = node;
222
223 if (cpu_to_node_map)
224 cpu_to_node_map[cpu] = node;
225
226 else if (per_cpu_offset(cpu))
227 per_cpu(x86_cpu_to_node_map, cpu) = node;
228
229 else
230 pr_debug("Setting node for non-present cpu %d\n", cpu);
231}
232
233void __cpuinit numa_clear_node(int cpu)
234{
235 numa_set_node(cpu, NUMA_NO_NODE);
236}
237
238#ifndef CONFIG_DEBUG_PER_CPU_MAPS
239
240void __cpuinit numa_add_cpu(int cpu)
241{
242 cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
243}
244
245void __cpuinit numa_remove_cpu(int cpu)
246{
247 cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]);
248}
249
250#else /* CONFIG_DEBUG_PER_CPU_MAPS */
251
252/*
253 * --------- debug versions of the numa functions ---------
254 */
255static void __cpuinit numa_set_cpumask(int cpu, int enable)
256{
257 int node = cpu_to_node(cpu);
258 cpumask_t *mask;
259 char buf[64];
260
261 if (node_to_cpumask_map == NULL) {
262 printk(KERN_ERR "node_to_cpumask_map NULL\n");
263 dump_stack();
264 return;
265 }
266
267 mask = &node_to_cpumask_map[node];
268 if (enable)
269 cpu_set(cpu, *mask);
270 else
271 cpu_clear(cpu, *mask);
272
273 cpulist_scnprintf(buf, sizeof(buf), *mask);
274 printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
275 enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf);
276 }
277
278void __cpuinit numa_add_cpu(int cpu)
279{
280 numa_set_cpumask(cpu, 1);
281}
282
283void __cpuinit numa_remove_cpu(int cpu)
284{
285 numa_set_cpumask(cpu, 0);
286}
287
288int cpu_to_node(int cpu)
289{
290 if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
291 printk(KERN_WARNING
292 "cpu_to_node(%d): usage too early!\n", cpu);
293 dump_stack();
294 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
295 }
296 return per_cpu(x86_cpu_to_node_map, cpu);
297}
298EXPORT_SYMBOL(cpu_to_node);
299
300/*
301 * Same function as cpu_to_node() but used if called before the
302 * per_cpu areas are setup.
303 */
304int early_cpu_to_node(int cpu)
305{
306 if (early_per_cpu_ptr(x86_cpu_to_node_map))
307 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
308
309 if (!per_cpu_offset(cpu)) {
310 printk(KERN_WARNING
311 "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
312 dump_stack();
313 return NUMA_NO_NODE;
314 }
315 return per_cpu(x86_cpu_to_node_map, cpu);
316}
317
318
319/* empty cpumask */
320static const cpumask_t cpu_mask_none;
321
322/*
323 * Returns a pointer to the bitmask of CPUs on Node 'node'.
324 */
325const cpumask_t *_node_to_cpumask_ptr(int node)
326{
327 if (node_to_cpumask_map == NULL) {
328 printk(KERN_WARNING
329 "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n",
330 node);
331 dump_stack();
332 return (const cpumask_t *)&cpu_online_map;
333 }
334 if (node >= nr_node_ids) {
335 printk(KERN_WARNING
336 "_node_to_cpumask_ptr(%d): node > nr_node_ids(%d)\n",
337 node, nr_node_ids);
338 dump_stack();
339 return &cpu_mask_none;
340 }
341 return &node_to_cpumask_map[node];
342}
343EXPORT_SYMBOL(_node_to_cpumask_ptr);
344
345/*
346 * Returns a bitmask of CPUs on Node 'node'.
347 *
348 * Side note: this function creates the returned cpumask on the stack
349 * so with a high NR_CPUS count, excessive stack space is used. The
350 * node_to_cpumask_ptr function should be used whenever possible.
351 */
352cpumask_t node_to_cpumask(int node)
353{
354 if (node_to_cpumask_map == NULL) {
355 printk(KERN_WARNING
356 "node_to_cpumask(%d): no node_to_cpumask_map!\n", node);
357 dump_stack();
358 return cpu_online_map;
359 }
360 if (node >= nr_node_ids) {
361 printk(KERN_WARNING
362 "node_to_cpumask(%d): node > nr_node_ids(%d)\n",
363 node, nr_node_ids);
364 dump_stack();
365 return cpu_mask_none;
366 }
367 return node_to_cpumask_map[node];
368}
369EXPORT_SYMBOL(node_to_cpumask);
370
371/*
372 * --------- end of debug versions of the numa functions ---------
373 */
374
375#endif /* CONFIG_DEBUG_PER_CPU_MAPS */
376
377#endif /* X86_64_NUMA */
378
diff --git a/arch/x86/kernel/signal_32.c b/arch/x86/kernel/signal_32.c
index d92373630963..6fb5bcdd8933 100644
--- a/arch/x86/kernel/signal_32.c
+++ b/arch/x86/kernel/signal_32.c
@@ -212,7 +212,7 @@ asmlinkage unsigned long sys_sigreturn(unsigned long __unused)
212 212
213badframe: 213badframe:
214 if (show_unhandled_signals && printk_ratelimit()) { 214 if (show_unhandled_signals && printk_ratelimit()) {
215 printk(KERN_INFO "%s%s[%d] bad frame in sigreturn frame:" 215 printk("%s%s[%d] bad frame in sigreturn frame:"
216 "%p ip:%lx sp:%lx oeax:%lx", 216 "%p ip:%lx sp:%lx oeax:%lx",
217 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG, 217 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
218 current->comm, task_pid_nr(current), frame, regs->ip, 218 current->comm, task_pid_nr(current), frame, regs->ip,
@@ -657,18 +657,9 @@ static void do_signal(struct pt_regs *regs)
657void 657void
658do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags) 658do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags)
659{ 659{
660 /* Pending single-step? */
661 if (thread_info_flags & _TIF_SINGLESTEP) {
662 regs->flags |= X86_EFLAGS_TF;
663 clear_thread_flag(TIF_SINGLESTEP);
664 }
665
666 /* deal with pending signal delivery */ 660 /* deal with pending signal delivery */
667 if (thread_info_flags & _TIF_SIGPENDING) 661 if (thread_info_flags & _TIF_SIGPENDING)
668 do_signal(regs); 662 do_signal(regs);
669 663
670 if (thread_info_flags & _TIF_HRTICK_RESCHED)
671 hrtick_resched();
672
673 clear_thread_flag(TIF_IRET); 664 clear_thread_flag(TIF_IRET);
674} 665}
diff --git a/arch/x86/kernel/signal_64.c b/arch/x86/kernel/signal_64.c
index e53b267662e7..b45ef8ddd651 100644
--- a/arch/x86/kernel/signal_64.c
+++ b/arch/x86/kernel/signal_64.c
@@ -53,6 +53,59 @@ sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
53 return do_sigaltstack(uss, uoss, regs->sp); 53 return do_sigaltstack(uss, uoss, regs->sp);
54} 54}
55 55
56/*
57 * Signal frame handlers.
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}
88
89/*
90 * This restores directly out of user space. Exceptions are handled.
91 */
92static inline int restore_i387(struct _fpstate __user *buf)
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 return restore_fpu_checking((__force struct i387_fxsave_struct *)buf);
108}
56 109
57/* 110/*
58 * Do a signal return; undo the signal stack. 111 * Do a signal return; undo the signal stack.
@@ -487,12 +540,6 @@ static void do_signal(struct pt_regs *regs)
487void do_notify_resume(struct pt_regs *regs, void *unused, 540void do_notify_resume(struct pt_regs *regs, void *unused,
488 __u32 thread_info_flags) 541 __u32 thread_info_flags)
489{ 542{
490 /* Pending single-step? */
491 if (thread_info_flags & _TIF_SINGLESTEP) {
492 regs->flags |= X86_EFLAGS_TF;
493 clear_thread_flag(TIF_SINGLESTEP);
494 }
495
496#ifdef CONFIG_X86_MCE 543#ifdef CONFIG_X86_MCE
497 /* notify userspace of pending MCEs */ 544 /* notify userspace of pending MCEs */
498 if (thread_info_flags & _TIF_MCE_NOTIFY) 545 if (thread_info_flags & _TIF_MCE_NOTIFY)
@@ -502,9 +549,6 @@ void do_notify_resume(struct pt_regs *regs, void *unused,
502 /* deal with pending signal delivery */ 549 /* deal with pending signal delivery */
503 if (thread_info_flags & _TIF_SIGPENDING) 550 if (thread_info_flags & _TIF_SIGPENDING)
504 do_signal(regs); 551 do_signal(regs);
505
506 if (thread_info_flags & _TIF_HRTICK_RESCHED)
507 hrtick_resched();
508} 552}
509 553
510void signal_fault(struct pt_regs *regs, void __user *frame, char *where) 554void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
diff --git a/arch/x86/kernel/smp.c b/arch/x86/kernel/smp.c
index 0cb7aadc87cd..361b7a4c640c 100644
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -121,132 +121,23 @@ static void native_smp_send_reschedule(int cpu)
121 send_IPI_mask(cpumask_of_cpu(cpu), RESCHEDULE_VECTOR); 121 send_IPI_mask(cpumask_of_cpu(cpu), RESCHEDULE_VECTOR);
122} 122}
123 123
124/* 124void native_send_call_func_single_ipi(int cpu)
125 * Structure and data for smp_call_function(). This is designed to minimise
126 * static memory requirements. It also looks cleaner.
127 */
128static DEFINE_SPINLOCK(call_lock);
129
130struct call_data_struct {
131 void (*func) (void *info);
132 void *info;
133 atomic_t started;
134 atomic_t finished;
135 int wait;
136};
137
138void lock_ipi_call_lock(void)
139{ 125{
140 spin_lock_irq(&call_lock); 126 send_IPI_mask(cpumask_of_cpu(cpu), CALL_FUNCTION_SINGLE_VECTOR);
141}
142
143void unlock_ipi_call_lock(void)
144{
145 spin_unlock_irq(&call_lock);
146}
147
148static struct call_data_struct *call_data;
149
150static void __smp_call_function(void (*func) (void *info), void *info,
151 int nonatomic, int wait)
152{
153 struct call_data_struct data;
154 int cpus = num_online_cpus() - 1;
155
156 if (!cpus)
157 return;
158
159 data.func = func;
160 data.info = info;
161 atomic_set(&data.started, 0);
162 data.wait = wait;
163 if (wait)
164 atomic_set(&data.finished, 0);
165
166 call_data = &data;
167 mb();
168
169 /* Send a message to all other CPUs and wait for them to respond */
170 send_IPI_allbutself(CALL_FUNCTION_VECTOR);
171
172 /* Wait for response */
173 while (atomic_read(&data.started) != cpus)
174 cpu_relax();
175
176 if (wait)
177 while (atomic_read(&data.finished) != cpus)
178 cpu_relax();
179} 127}
180 128
181 129void native_send_call_func_ipi(cpumask_t mask)
182/**
183 * smp_call_function_mask(): Run a function on a set of other CPUs.
184 * @mask: The set of cpus to run on. Must not include the current cpu.
185 * @func: The function to run. This must be fast and non-blocking.
186 * @info: An arbitrary pointer to pass to the function.
187 * @wait: If true, wait (atomically) until function has completed on other CPUs.
188 *
189 * Returns 0 on success, else a negative status code.
190 *
191 * If @wait is true, then returns once @func has returned; otherwise
192 * it returns just before the target cpu calls @func.
193 *
194 * You must not call this function with disabled interrupts or from a
195 * hardware interrupt handler or from a bottom half handler.
196 */
197static int
198native_smp_call_function_mask(cpumask_t mask,
199 void (*func)(void *), void *info,
200 int wait)
201{ 130{
202 struct call_data_struct data;
203 cpumask_t allbutself; 131 cpumask_t allbutself;
204 int cpus;
205
206 /* Can deadlock when called with interrupts disabled */
207 WARN_ON(irqs_disabled());
208
209 /* Holding any lock stops cpus from going down. */
210 spin_lock(&call_lock);
211 132
212 allbutself = cpu_online_map; 133 allbutself = cpu_online_map;
213 cpu_clear(smp_processor_id(), allbutself); 134 cpu_clear(smp_processor_id(), allbutself);
214 135
215 cpus_and(mask, mask, allbutself);
216 cpus = cpus_weight(mask);
217
218 if (!cpus) {
219 spin_unlock(&call_lock);
220 return 0;
221 }
222
223 data.func = func;
224 data.info = info;
225 atomic_set(&data.started, 0);
226 data.wait = wait;
227 if (wait)
228 atomic_set(&data.finished, 0);
229
230 call_data = &data;
231 wmb();
232
233 /* Send a message to other CPUs */
234 if (cpus_equal(mask, allbutself) && 136 if (cpus_equal(mask, allbutself) &&
235 cpus_equal(cpu_online_map, cpu_callout_map)) 137 cpus_equal(cpu_online_map, cpu_callout_map))
236 send_IPI_allbutself(CALL_FUNCTION_VECTOR); 138 send_IPI_allbutself(CALL_FUNCTION_VECTOR);
237 else 139 else
238 send_IPI_mask(mask, CALL_FUNCTION_VECTOR); 140 send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
239
240 /* Wait for response */
241 while (atomic_read(&data.started) != cpus)
242 cpu_relax();
243
244 if (wait)
245 while (atomic_read(&data.finished) != cpus)
246 cpu_relax();
247 spin_unlock(&call_lock);
248
249 return 0;
250} 141}
251 142
252static void stop_this_cpu(void *dummy) 143static void stop_this_cpu(void *dummy)
@@ -268,18 +159,13 @@ static void stop_this_cpu(void *dummy)
268 159
269static void native_smp_send_stop(void) 160static void native_smp_send_stop(void)
270{ 161{
271 int nolock;
272 unsigned long flags; 162 unsigned long flags;
273 163
274 if (reboot_force) 164 if (reboot_force)
275 return; 165 return;
276 166
277 /* Don't deadlock on the call lock in panic */ 167 smp_call_function(stop_this_cpu, NULL, 0);
278 nolock = !spin_trylock(&call_lock);
279 local_irq_save(flags); 168 local_irq_save(flags);
280 __smp_call_function(stop_this_cpu, NULL, 0, 0);
281 if (!nolock)
282 spin_unlock(&call_lock);
283 disable_local_APIC(); 169 disable_local_APIC();
284 local_irq_restore(flags); 170 local_irq_restore(flags);
285} 171}
@@ -301,33 +187,28 @@ void smp_reschedule_interrupt(struct pt_regs *regs)
301 187
302void smp_call_function_interrupt(struct pt_regs *regs) 188void smp_call_function_interrupt(struct pt_regs *regs)
303{ 189{
304 void (*func) (void *info) = call_data->func;
305 void *info = call_data->info;
306 int wait = call_data->wait;
307
308 ack_APIC_irq(); 190 ack_APIC_irq();
309 /*
310 * Notify initiating CPU that I've grabbed the data and am
311 * about to execute the function
312 */
313 mb();
314 atomic_inc(&call_data->started);
315 /*
316 * At this point the info structure may be out of scope unless wait==1
317 */
318 irq_enter(); 191 irq_enter();
319 (*func)(info); 192 generic_smp_call_function_interrupt();
320#ifdef CONFIG_X86_32 193#ifdef CONFIG_X86_32
321 __get_cpu_var(irq_stat).irq_call_count++; 194 __get_cpu_var(irq_stat).irq_call_count++;
322#else 195#else
323 add_pda(irq_call_count, 1); 196 add_pda(irq_call_count, 1);
324#endif 197#endif
325 irq_exit(); 198 irq_exit();
199}
326 200
327 if (wait) { 201void smp_call_function_single_interrupt(struct pt_regs *regs)
328 mb(); 202{
329 atomic_inc(&call_data->finished); 203 ack_APIC_irq();
330 } 204 irq_enter();
205 generic_smp_call_function_single_interrupt();
206#ifdef CONFIG_X86_32
207 __get_cpu_var(irq_stat).irq_call_count++;
208#else
209 add_pda(irq_call_count, 1);
210#endif
211 irq_exit();
331} 212}
332 213
333struct smp_ops smp_ops = { 214struct smp_ops smp_ops = {
@@ -338,7 +219,8 @@ struct smp_ops smp_ops = {
338 219
339 .smp_send_stop = native_smp_send_stop, 220 .smp_send_stop = native_smp_send_stop,
340 .smp_send_reschedule = native_smp_send_reschedule, 221 .smp_send_reschedule = native_smp_send_reschedule,
341 .smp_call_function_mask = native_smp_call_function_mask, 222
223 .send_call_func_ipi = native_send_call_func_ipi,
224 .send_call_func_single_ipi = native_send_call_func_single_ipi,
342}; 225};
343EXPORT_SYMBOL_GPL(smp_ops); 226EXPORT_SYMBOL_GPL(smp_ops);
344
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 56078d61c793..332512767f4f 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -59,7 +59,6 @@
59#include <asm/pgtable.h> 59#include <asm/pgtable.h>
60#include <asm/tlbflush.h> 60#include <asm/tlbflush.h>
61#include <asm/mtrr.h> 61#include <asm/mtrr.h>
62#include <asm/nmi.h>
63#include <asm/vmi.h> 62#include <asm/vmi.h>
64#include <asm/genapic.h> 63#include <asm/genapic.h>
65#include <linux/mc146818rtc.h> 64#include <linux/mc146818rtc.h>
@@ -68,22 +67,6 @@
68#include <mach_wakecpu.h> 67#include <mach_wakecpu.h>
69#include <smpboot_hooks.h> 68#include <smpboot_hooks.h>
70 69
71/*
72 * FIXME: For x86_64, those are defined in other files. But moving them here,
73 * would make the setup areas dependent on smp, which is a loss. When we
74 * integrate apic between arches, we can probably do a better job, but
75 * right now, they'll stay here -- glommer
76 */
77
78/* which logical CPU number maps to which CPU (physical APIC ID) */
79u16 x86_cpu_to_apicid_init[NR_CPUS] __initdata =
80 { [0 ... NR_CPUS-1] = BAD_APICID };
81void *x86_cpu_to_apicid_early_ptr;
82
83u16 x86_bios_cpu_apicid_init[NR_CPUS] __initdata
84 = { [0 ... NR_CPUS-1] = BAD_APICID };
85void *x86_bios_cpu_apicid_early_ptr;
86
87#ifdef CONFIG_X86_32 70#ifdef CONFIG_X86_32
88u8 apicid_2_node[MAX_APICID]; 71u8 apicid_2_node[MAX_APICID];
89static int low_mappings; 72static int low_mappings;
@@ -198,13 +181,12 @@ static void map_cpu_to_logical_apicid(void)
198 map_cpu_to_node(cpu, node); 181 map_cpu_to_node(cpu, node);
199} 182}
200 183
201static void unmap_cpu_to_logical_apicid(int cpu) 184void numa_remove_cpu(int cpu)
202{ 185{
203 cpu_2_logical_apicid[cpu] = BAD_APICID; 186 cpu_2_logical_apicid[cpu] = BAD_APICID;
204 unmap_cpu_to_node(cpu); 187 unmap_cpu_to_node(cpu);
205} 188}
206#else 189#else
207#define unmap_cpu_to_logical_apicid(cpu) do {} while (0)
208#define map_cpu_to_logical_apicid() do {} while (0) 190#define map_cpu_to_logical_apicid() do {} while (0)
209#endif 191#endif
210 192
@@ -234,7 +216,7 @@ static void __cpuinit smp_callin(void)
234 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__, 216 panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
235 phys_id, cpuid); 217 phys_id, cpuid);
236 } 218 }
237 Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id); 219 pr_debug("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);
238 220
239 /* 221 /*
240 * STARTUP IPIs are fragile beasts as they might sometimes 222 * STARTUP IPIs are fragile beasts as they might sometimes
@@ -269,7 +251,7 @@ static void __cpuinit smp_callin(void)
269 * boards) 251 * boards)
270 */ 252 */
271 253
272 Dprintk("CALLIN, before setup_local_APIC().\n"); 254 pr_debug("CALLIN, before setup_local_APIC().\n");
273 smp_callin_clear_local_apic(); 255 smp_callin_clear_local_apic();
274 setup_local_APIC(); 256 setup_local_APIC();
275 end_local_APIC_setup(); 257 end_local_APIC_setup();
@@ -284,7 +266,7 @@ static void __cpuinit smp_callin(void)
284 local_irq_enable(); 266 local_irq_enable();
285 calibrate_delay(); 267 calibrate_delay();
286 local_irq_disable(); 268 local_irq_disable();
287 Dprintk("Stack at about %p\n", &cpuid); 269 pr_debug("Stack at about %p\n", &cpuid);
288 270
289 /* 271 /*
290 * Save our processor parameters 272 * Save our processor parameters
@@ -345,19 +327,12 @@ static void __cpuinit start_secondary(void *unused)
345 * lock helps us to not include this cpu in a currently in progress 327 * lock helps us to not include this cpu in a currently in progress
346 * smp_call_function(). 328 * smp_call_function().
347 */ 329 */
348 lock_ipi_call_lock(); 330 ipi_call_lock_irq();
349#ifdef CONFIG_X86_64 331#ifdef CONFIG_X86_IO_APIC
350 spin_lock(&vector_lock); 332 setup_vector_irq(smp_processor_id());
351
352 /* Setup the per cpu irq handling data structures */
353 __setup_vector_irq(smp_processor_id());
354 /*
355 * Allow the master to continue.
356 */
357 spin_unlock(&vector_lock);
358#endif 333#endif
359 cpu_set(smp_processor_id(), cpu_online_map); 334 cpu_set(smp_processor_id(), cpu_online_map);
360 unlock_ipi_call_lock(); 335 ipi_call_unlock_irq();
361 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; 336 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
362 337
363 setup_secondary_clock(); 338 setup_secondary_clock();
@@ -366,31 +341,8 @@ static void __cpuinit start_secondary(void *unused)
366 cpu_idle(); 341 cpu_idle();
367} 342}
368 343
369#ifdef CONFIG_X86_32
370/*
371 * Everything has been set up for the secondary
372 * CPUs - they just need to reload everything
373 * from the task structure
374 * This function must not return.
375 */
376void __devinit initialize_secondary(void)
377{
378 /*
379 * We don't actually need to load the full TSS,
380 * basically just the stack pointer and the ip.
381 */
382
383 asm volatile(
384 "movl %0,%%esp\n\t"
385 "jmp *%1"
386 :
387 :"m" (current->thread.sp), "m" (current->thread.ip));
388}
389#endif
390
391static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c) 344static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
392{ 345{
393#ifdef CONFIG_X86_32
394 /* 346 /*
395 * Mask B, Pentium, but not Pentium MMX 347 * Mask B, Pentium, but not Pentium MMX
396 */ 348 */
@@ -440,7 +392,6 @@ static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
440 392
441valid_k7: 393valid_k7:
442 ; 394 ;
443#endif
444} 395}
445 396
446static void __cpuinit smp_checks(void) 397static void __cpuinit smp_checks(void)
@@ -487,7 +438,7 @@ void __cpuinit set_cpu_sibling_map(int cpu)
487 cpu_set(cpu, cpu_sibling_setup_map); 438 cpu_set(cpu, cpu_sibling_setup_map);
488 439
489 if (smp_num_siblings > 1) { 440 if (smp_num_siblings > 1) {
490 for_each_cpu_mask(i, cpu_sibling_setup_map) { 441 for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
491 if (c->phys_proc_id == cpu_data(i).phys_proc_id && 442 if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
492 c->cpu_core_id == cpu_data(i).cpu_core_id) { 443 c->cpu_core_id == cpu_data(i).cpu_core_id) {
493 cpu_set(i, per_cpu(cpu_sibling_map, cpu)); 444 cpu_set(i, per_cpu(cpu_sibling_map, cpu));
@@ -510,7 +461,7 @@ void __cpuinit set_cpu_sibling_map(int cpu)
510 return; 461 return;
511 } 462 }
512 463
513 for_each_cpu_mask(i, cpu_sibling_setup_map) { 464 for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
514 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID && 465 if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
515 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) { 466 per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
516 cpu_set(i, c->llc_shared_map); 467 cpu_set(i, c->llc_shared_map);
@@ -555,23 +506,6 @@ cpumask_t cpu_coregroup_map(int cpu)
555 return c->llc_shared_map; 506 return c->llc_shared_map;
556} 507}
557 508
558#ifdef CONFIG_X86_32
559/*
560 * We are called very early to get the low memory for the
561 * SMP bootup trampoline page.
562 */
563void __init smp_alloc_memory(void)
564{
565 trampoline_base = alloc_bootmem_low_pages(PAGE_SIZE);
566 /*
567 * Has to be in very low memory so we can execute
568 * real-mode AP code.
569 */
570 if (__pa(trampoline_base) >= 0x9F000)
571 BUG();
572}
573#endif
574
575static void impress_friends(void) 509static void impress_friends(void)
576{ 510{
577 int cpu; 511 int cpu;
@@ -579,7 +513,7 @@ static void impress_friends(void)
579 /* 513 /*
580 * Allow the user to impress friends. 514 * Allow the user to impress friends.
581 */ 515 */
582 Dprintk("Before bogomips.\n"); 516 pr_debug("Before bogomips.\n");
583 for_each_possible_cpu(cpu) 517 for_each_possible_cpu(cpu)
584 if (cpu_isset(cpu, cpu_callout_map)) 518 if (cpu_isset(cpu, cpu_callout_map))
585 bogosum += cpu_data(cpu).loops_per_jiffy; 519 bogosum += cpu_data(cpu).loops_per_jiffy;
@@ -589,7 +523,7 @@ static void impress_friends(void)
589 bogosum/(500000/HZ), 523 bogosum/(500000/HZ),
590 (bogosum/(5000/HZ))%100); 524 (bogosum/(5000/HZ))%100);
591 525
592 Dprintk("Before bogocount - setting activated=1.\n"); 526 pr_debug("Before bogocount - setting activated=1.\n");
593} 527}
594 528
595static inline void __inquire_remote_apic(int apicid) 529static inline void __inquire_remote_apic(int apicid)
@@ -612,8 +546,8 @@ static inline void __inquire_remote_apic(int apicid)
612 printk(KERN_CONT 546 printk(KERN_CONT
613 "a previous APIC delivery may have failed\n"); 547 "a previous APIC delivery may have failed\n");
614 548
615 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(apicid)); 549 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(apicid));
616 apic_write_around(APIC_ICR, APIC_DM_REMRD | regs[i]); 550 apic_write(APIC_ICR, APIC_DM_REMRD | regs[i]);
617 551
618 timeout = 0; 552 timeout = 0;
619 do { 553 do {
@@ -645,29 +579,24 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
645 int maxlvt; 579 int maxlvt;
646 580
647 /* Target chip */ 581 /* Target chip */
648 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid)); 582 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(logical_apicid));
649 583
650 /* Boot on the stack */ 584 /* Boot on the stack */
651 /* Kick the second */ 585 /* Kick the second */
652 apic_write_around(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL); 586 apic_write(APIC_ICR, APIC_DM_NMI | APIC_DEST_LOGICAL);
653 587
654 Dprintk("Waiting for send to finish...\n"); 588 pr_debug("Waiting for send to finish...\n");
655 send_status = safe_apic_wait_icr_idle(); 589 send_status = safe_apic_wait_icr_idle();
656 590
657 /* 591 /*
658 * Give the other CPU some time to accept the IPI. 592 * Give the other CPU some time to accept the IPI.
659 */ 593 */
660 udelay(200); 594 udelay(200);
661 /*
662 * Due to the Pentium erratum 3AP.
663 */
664 maxlvt = lapic_get_maxlvt(); 595 maxlvt = lapic_get_maxlvt();
665 if (maxlvt > 3) { 596 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
666 apic_read_around(APIC_SPIV);
667 apic_write(APIC_ESR, 0); 597 apic_write(APIC_ESR, 0);
668 }
669 accept_status = (apic_read(APIC_ESR) & 0xEF); 598 accept_status = (apic_read(APIC_ESR) & 0xEF);
670 Dprintk("NMI sent.\n"); 599 pr_debug("NMI sent.\n");
671 600
672 if (send_status) 601 if (send_status)
673 printk(KERN_ERR "APIC never delivered???\n"); 602 printk(KERN_ERR "APIC never delivered???\n");
@@ -691,42 +620,44 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
691 return send_status; 620 return send_status;
692 } 621 }
693 622
623 maxlvt = lapic_get_maxlvt();
624
694 /* 625 /*
695 * Be paranoid about clearing APIC errors. 626 * Be paranoid about clearing APIC errors.
696 */ 627 */
697 if (APIC_INTEGRATED(apic_version[phys_apicid])) { 628 if (APIC_INTEGRATED(apic_version[phys_apicid])) {
698 apic_read_around(APIC_SPIV); 629 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
699 apic_write(APIC_ESR, 0); 630 apic_write(APIC_ESR, 0);
700 apic_read(APIC_ESR); 631 apic_read(APIC_ESR);
701 } 632 }
702 633
703 Dprintk("Asserting INIT.\n"); 634 pr_debug("Asserting INIT.\n");
704 635
705 /* 636 /*
706 * Turn INIT on target chip 637 * Turn INIT on target chip
707 */ 638 */
708 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid)); 639 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
709 640
710 /* 641 /*
711 * Send IPI 642 * Send IPI
712 */ 643 */
713 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_INT_ASSERT 644 apic_write(APIC_ICR,
714 | APIC_DM_INIT); 645 APIC_INT_LEVELTRIG | APIC_INT_ASSERT | APIC_DM_INIT);
715 646
716 Dprintk("Waiting for send to finish...\n"); 647 pr_debug("Waiting for send to finish...\n");
717 send_status = safe_apic_wait_icr_idle(); 648 send_status = safe_apic_wait_icr_idle();
718 649
719 mdelay(10); 650 mdelay(10);
720 651
721 Dprintk("Deasserting INIT.\n"); 652 pr_debug("Deasserting INIT.\n");
722 653
723 /* Target chip */ 654 /* Target chip */
724 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid)); 655 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
725 656
726 /* Send IPI */ 657 /* Send IPI */
727 apic_write_around(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT); 658 apic_write(APIC_ICR, APIC_INT_LEVELTRIG | APIC_DM_INIT);
728 659
729 Dprintk("Waiting for send to finish...\n"); 660 pr_debug("Waiting for send to finish...\n");
730 send_status = safe_apic_wait_icr_idle(); 661 send_status = safe_apic_wait_icr_idle();
731 662
732 mb(); 663 mb();
@@ -748,64 +679,52 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
748 * target processor state. 679 * target processor state.
749 */ 680 */
750 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary, 681 startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
751#ifdef CONFIG_X86_64
752 (unsigned long)init_rsp);
753#else
754 (unsigned long)stack_start.sp); 682 (unsigned long)stack_start.sp);
755#endif
756 683
757 /* 684 /*
758 * Run STARTUP IPI loop. 685 * Run STARTUP IPI loop.
759 */ 686 */
760 Dprintk("#startup loops: %d.\n", num_starts); 687 pr_debug("#startup loops: %d.\n", num_starts);
761
762 maxlvt = lapic_get_maxlvt();
763 688
764 for (j = 1; j <= num_starts; j++) { 689 for (j = 1; j <= num_starts; j++) {
765 Dprintk("Sending STARTUP #%d.\n", j); 690 pr_debug("Sending STARTUP #%d.\n", j);
766 apic_read_around(APIC_SPIV); 691 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
767 apic_write(APIC_ESR, 0); 692 apic_write(APIC_ESR, 0);
768 apic_read(APIC_ESR); 693 apic_read(APIC_ESR);
769 Dprintk("After apic_write.\n"); 694 pr_debug("After apic_write.\n");
770 695
771 /* 696 /*
772 * STARTUP IPI 697 * STARTUP IPI
773 */ 698 */
774 699
775 /* Target chip */ 700 /* Target chip */
776 apic_write_around(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid)); 701 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(phys_apicid));
777 702
778 /* Boot on the stack */ 703 /* Boot on the stack */
779 /* Kick the second */ 704 /* Kick the second */
780 apic_write_around(APIC_ICR, APIC_DM_STARTUP 705 apic_write(APIC_ICR, APIC_DM_STARTUP | (start_eip >> 12));
781 | (start_eip >> 12));
782 706
783 /* 707 /*
784 * Give the other CPU some time to accept the IPI. 708 * Give the other CPU some time to accept the IPI.
785 */ 709 */
786 udelay(300); 710 udelay(300);
787 711
788 Dprintk("Startup point 1.\n"); 712 pr_debug("Startup point 1.\n");
789 713
790 Dprintk("Waiting for send to finish...\n"); 714 pr_debug("Waiting for send to finish...\n");
791 send_status = safe_apic_wait_icr_idle(); 715 send_status = safe_apic_wait_icr_idle();
792 716
793 /* 717 /*
794 * Give the other CPU some time to accept the IPI. 718 * Give the other CPU some time to accept the IPI.
795 */ 719 */
796 udelay(200); 720 udelay(200);
797 /* 721 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
798 * Due to the Pentium erratum 3AP.
799 */
800 if (maxlvt > 3) {
801 apic_read_around(APIC_SPIV);
802 apic_write(APIC_ESR, 0); 722 apic_write(APIC_ESR, 0);
803 }
804 accept_status = (apic_read(APIC_ESR) & 0xEF); 723 accept_status = (apic_read(APIC_ESR) & 0xEF);
805 if (send_status || accept_status) 724 if (send_status || accept_status)
806 break; 725 break;
807 } 726 }
808 Dprintk("After Startup.\n"); 727 pr_debug("After Startup.\n");
809 728
810 if (send_status) 729 if (send_status)
811 printk(KERN_ERR "APIC never delivered???\n"); 730 printk(KERN_ERR "APIC never delivered???\n");
@@ -832,6 +751,45 @@ static void __cpuinit do_fork_idle(struct work_struct *work)
832 complete(&c_idle->done); 751 complete(&c_idle->done);
833} 752}
834 753
754#ifdef CONFIG_X86_64
755/*
756 * Allocate node local memory for the AP pda.
757 *
758 * Must be called after the _cpu_pda pointer table is initialized.
759 */
760int __cpuinit get_local_pda(int cpu)
761{
762 struct x8664_pda *oldpda, *newpda;
763 unsigned long size = sizeof(struct x8664_pda);
764 int node = cpu_to_node(cpu);
765
766 if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
767 return 0;
768
769 oldpda = cpu_pda(cpu);
770 newpda = kmalloc_node(size, GFP_ATOMIC, node);
771 if (!newpda) {
772 printk(KERN_ERR "Could not allocate node local PDA "
773 "for CPU %d on node %d\n", cpu, node);
774
775 if (oldpda)
776 return 0; /* have a usable pda */
777 else
778 return -1;
779 }
780
781 if (oldpda) {
782 memcpy(newpda, oldpda, size);
783 if (!after_bootmem)
784 free_bootmem((unsigned long)oldpda, size);
785 }
786
787 newpda->in_bootmem = 0;
788 cpu_pda(cpu) = newpda;
789 return 0;
790}
791#endif /* CONFIG_X86_64 */
792
835static int __cpuinit do_boot_cpu(int apicid, int cpu) 793static int __cpuinit do_boot_cpu(int apicid, int cpu)
836/* 794/*
837 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad 795 * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
@@ -848,28 +806,14 @@ static int __cpuinit do_boot_cpu(int apicid, int cpu)
848 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done), 806 .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
849 }; 807 };
850 INIT_WORK(&c_idle.work, do_fork_idle); 808 INIT_WORK(&c_idle.work, do_fork_idle);
851#ifdef CONFIG_X86_64
852 /* allocate memory for gdts of secondary cpus. Hotplug is considered */
853 if (!cpu_gdt_descr[cpu].address &&
854 !(cpu_gdt_descr[cpu].address = get_zeroed_page(GFP_KERNEL))) {
855 printk(KERN_ERR "Failed to allocate GDT for CPU %d\n", cpu);
856 return -1;
857 }
858 809
810#ifdef CONFIG_X86_64
859 /* Allocate node local memory for AP pdas */ 811 /* Allocate node local memory for AP pdas */
860 if (cpu_pda(cpu) == &boot_cpu_pda[cpu]) { 812 if (cpu > 0) {
861 struct x8664_pda *newpda, *pda; 813 boot_error = get_local_pda(cpu);
862 int node = cpu_to_node(cpu); 814 if (boot_error)
863 pda = cpu_pda(cpu); 815 goto restore_state;
864 newpda = kmalloc_node(sizeof(struct x8664_pda), GFP_ATOMIC, 816 /* if can't get pda memory, can't start cpu */
865 node);
866 if (newpda) {
867 memcpy(newpda, pda, sizeof(struct x8664_pda));
868 cpu_pda(cpu) = newpda;
869 } else
870 printk(KERN_ERR
871 "Could not allocate node local PDA for CPU %d on node %d\n",
872 cpu, node);
873 } 817 }
874#endif 818#endif
875 819
@@ -905,18 +849,15 @@ do_rest:
905#ifdef CONFIG_X86_32 849#ifdef CONFIG_X86_32
906 per_cpu(current_task, cpu) = c_idle.idle; 850 per_cpu(current_task, cpu) = c_idle.idle;
907 init_gdt(cpu); 851 init_gdt(cpu);
908 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
909 c_idle.idle->thread.ip = (unsigned long) start_secondary;
910 /* Stack for startup_32 can be just as for start_secondary onwards */ 852 /* Stack for startup_32 can be just as for start_secondary onwards */
911 stack_start.sp = (void *) c_idle.idle->thread.sp;
912 irq_ctx_init(cpu); 853 irq_ctx_init(cpu);
913#else 854#else
914 cpu_pda(cpu)->pcurrent = c_idle.idle; 855 cpu_pda(cpu)->pcurrent = c_idle.idle;
915 init_rsp = c_idle.idle->thread.sp;
916 load_sp0(&per_cpu(init_tss, cpu), &c_idle.idle->thread);
917 initial_code = (unsigned long)start_secondary;
918 clear_tsk_thread_flag(c_idle.idle, TIF_FORK); 856 clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
919#endif 857#endif
858 early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
859 initial_code = (unsigned long)start_secondary;
860 stack_start.sp = (void *) c_idle.idle->thread.sp;
920 861
921 /* start_ip had better be page-aligned! */ 862 /* start_ip had better be page-aligned! */
922 start_ip = setup_trampoline(); 863 start_ip = setup_trampoline();
@@ -934,7 +875,7 @@ do_rest:
934 875
935 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) { 876 if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
936 877
937 Dprintk("Setting warm reset code and vector.\n"); 878 pr_debug("Setting warm reset code and vector.\n");
938 879
939 store_NMI_vector(&nmi_high, &nmi_low); 880 store_NMI_vector(&nmi_high, &nmi_low);
940 881
@@ -955,9 +896,9 @@ do_rest:
955 /* 896 /*
956 * allow APs to start initializing. 897 * allow APs to start initializing.
957 */ 898 */
958 Dprintk("Before Callout %d.\n", cpu); 899 pr_debug("Before Callout %d.\n", cpu);
959 cpu_set(cpu, cpu_callout_map); 900 cpu_set(cpu, cpu_callout_map);
960 Dprintk("After Callout %d.\n", cpu); 901 pr_debug("After Callout %d.\n", cpu);
961 902
962 /* 903 /*
963 * Wait 5s total for a response 904 * Wait 5s total for a response
@@ -970,10 +911,10 @@ do_rest:
970 911
971 if (cpu_isset(cpu, cpu_callin_map)) { 912 if (cpu_isset(cpu, cpu_callin_map)) {
972 /* number CPUs logically, starting from 1 (BSP is 0) */ 913 /* number CPUs logically, starting from 1 (BSP is 0) */
973 Dprintk("OK.\n"); 914 pr_debug("OK.\n");
974 printk(KERN_INFO "CPU%d: ", cpu); 915 printk(KERN_INFO "CPU%d: ", cpu);
975 print_cpu_info(&cpu_data(cpu)); 916 print_cpu_info(&cpu_data(cpu));
976 Dprintk("CPU has booted.\n"); 917 pr_debug("CPU has booted.\n");
977 } else { 918 } else {
978 boot_error = 1; 919 boot_error = 1;
979 if (*((volatile unsigned char *)trampoline_base) 920 if (*((volatile unsigned char *)trampoline_base)
@@ -987,16 +928,14 @@ do_rest:
987 inquire_remote_apic(apicid); 928 inquire_remote_apic(apicid);
988 } 929 }
989 } 930 }
990
991 if (boot_error) {
992 /* Try to put things back the way they were before ... */
993 unmap_cpu_to_logical_apicid(cpu);
994#ifdef CONFIG_X86_64 931#ifdef CONFIG_X86_64
995 clear_node_cpumask(cpu); /* was set by numa_add_cpu */ 932restore_state:
996#endif 933#endif
934 if (boot_error) {
935 /* Try to put things back the way they were before ... */
936 numa_remove_cpu(cpu); /* was set by numa_add_cpu */
997 cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */ 937 cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */
998 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */ 938 cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
999 cpu_clear(cpu, cpu_possible_map);
1000 cpu_clear(cpu, cpu_present_map); 939 cpu_clear(cpu, cpu_present_map);
1001 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID; 940 per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
1002 } 941 }
@@ -1020,7 +959,7 @@ int __cpuinit native_cpu_up(unsigned int cpu)
1020 959
1021 WARN_ON(irqs_disabled()); 960 WARN_ON(irqs_disabled());
1022 961
1023 Dprintk("++++++++++++++++++++=_---CPU UP %u\n", cpu); 962 pr_debug("++++++++++++++++++++=_---CPU UP %u\n", cpu);
1024 963
1025 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid || 964 if (apicid == BAD_APICID || apicid == boot_cpu_physical_apicid ||
1026 !physid_isset(apicid, phys_cpu_present_map)) { 965 !physid_isset(apicid, phys_cpu_present_map)) {
@@ -1032,7 +971,7 @@ int __cpuinit native_cpu_up(unsigned int cpu)
1032 * Already booted CPU? 971 * Already booted CPU?
1033 */ 972 */
1034 if (cpu_isset(cpu, cpu_callin_map)) { 973 if (cpu_isset(cpu, cpu_callin_map)) {
1035 Dprintk("do_boot_cpu %d Already started\n", cpu); 974 pr_debug("do_boot_cpu %d Already started\n", cpu);
1036 return -ENOSYS; 975 return -ENOSYS;
1037 } 976 }
1038 977
@@ -1059,7 +998,7 @@ int __cpuinit native_cpu_up(unsigned int cpu)
1059 err = do_boot_cpu(apicid, cpu); 998 err = do_boot_cpu(apicid, cpu);
1060#endif 999#endif
1061 if (err) { 1000 if (err) {
1062 Dprintk("do_boot_cpu failed %d\n", err); 1001 pr_debug("do_boot_cpu failed %d\n", err);
1063 return -EIO; 1002 return -EIO;
1064 } 1003 }
1065 1004
@@ -1088,14 +1027,12 @@ static __init void disable_smp(void)
1088{ 1027{
1089 cpu_present_map = cpumask_of_cpu(0); 1028 cpu_present_map = cpumask_of_cpu(0);
1090 cpu_possible_map = cpumask_of_cpu(0); 1029 cpu_possible_map = cpumask_of_cpu(0);
1091#ifdef CONFIG_X86_32
1092 smpboot_clear_io_apic_irqs(); 1030 smpboot_clear_io_apic_irqs();
1093#endif 1031
1094 if (smp_found_config) 1032 if (smp_found_config)
1095 phys_cpu_present_map = 1033 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1096 physid_mask_of_physid(boot_cpu_physical_apicid);
1097 else 1034 else
1098 phys_cpu_present_map = physid_mask_of_physid(0); 1035 physid_set_mask_of_physid(0, &phys_cpu_present_map);
1099 map_cpu_to_logical_apicid(); 1036 map_cpu_to_logical_apicid();
1100 cpu_set(0, per_cpu(cpu_sibling_map, 0)); 1037 cpu_set(0, per_cpu(cpu_sibling_map, 0));
1101 cpu_set(0, per_cpu(cpu_core_map, 0)); 1038 cpu_set(0, per_cpu(cpu_core_map, 0));
@@ -1158,12 +1095,12 @@ static int __init smp_sanity_check(unsigned max_cpus)
1158 * If SMP should be disabled, then really disable it! 1095 * If SMP should be disabled, then really disable it!
1159 */ 1096 */
1160 if (!max_cpus) { 1097 if (!max_cpus) {
1161 printk(KERN_INFO "SMP mode deactivated," 1098 printk(KERN_INFO "SMP mode deactivated.\n");
1162 "forcing use of dummy APIC emulation.\n");
1163 smpboot_clear_io_apic(); 1099 smpboot_clear_io_apic();
1164#ifdef CONFIG_X86_32 1100
1101 localise_nmi_watchdog();
1102
1165 connect_bsp_APIC(); 1103 connect_bsp_APIC();
1166#endif
1167 setup_local_APIC(); 1104 setup_local_APIC();
1168 end_local_APIC_setup(); 1105 end_local_APIC_setup();
1169 return -1; 1106 return -1;
@@ -1191,7 +1128,6 @@ static void __init smp_cpu_index_default(void)
1191void __init native_smp_prepare_cpus(unsigned int max_cpus) 1128void __init native_smp_prepare_cpus(unsigned int max_cpus)
1192{ 1129{
1193 preempt_disable(); 1130 preempt_disable();
1194 nmi_watchdog_default();
1195 smp_cpu_index_default(); 1131 smp_cpu_index_default();
1196 current_cpu_data = boot_cpu_data; 1132 current_cpu_data = boot_cpu_data;
1197 cpu_callin_map = cpumask_of_cpu(0); 1133 cpu_callin_map = cpumask_of_cpu(0);
@@ -1218,9 +1154,8 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
1218 } 1154 }
1219 preempt_enable(); 1155 preempt_enable();
1220 1156
1221#ifdef CONFIG_X86_32
1222 connect_bsp_APIC(); 1157 connect_bsp_APIC();
1223#endif 1158
1224 /* 1159 /*
1225 * Switch from PIC to APIC mode. 1160 * Switch from PIC to APIC mode.
1226 */ 1161 */
@@ -1258,8 +1193,8 @@ void __init native_smp_prepare_boot_cpu(void)
1258 int me = smp_processor_id(); 1193 int me = smp_processor_id();
1259#ifdef CONFIG_X86_32 1194#ifdef CONFIG_X86_32
1260 init_gdt(me); 1195 init_gdt(me);
1261 switch_to_new_gdt();
1262#endif 1196#endif
1197 switch_to_new_gdt();
1263 /* already set me in cpu_online_map in boot_cpu_init() */ 1198 /* already set me in cpu_online_map in boot_cpu_init() */
1264 cpu_set(me, cpu_callout_map); 1199 cpu_set(me, cpu_callout_map);
1265 per_cpu(cpu_state, me) = CPU_ONLINE; 1200 per_cpu(cpu_state, me) = CPU_ONLINE;
@@ -1267,7 +1202,7 @@ void __init native_smp_prepare_boot_cpu(void)
1267 1202
1268void __init native_smp_cpus_done(unsigned int max_cpus) 1203void __init native_smp_cpus_done(unsigned int max_cpus)
1269{ 1204{
1270 Dprintk("Boot done.\n"); 1205 pr_debug("Boot done.\n");
1271 1206
1272 impress_friends(); 1207 impress_friends();
1273 smp_checks(); 1208 smp_checks();
@@ -1279,29 +1214,12 @@ void __init native_smp_cpus_done(unsigned int max_cpus)
1279 1214
1280#ifdef CONFIG_HOTPLUG_CPU 1215#ifdef CONFIG_HOTPLUG_CPU
1281 1216
1282# ifdef CONFIG_X86_32
1283void cpu_exit_clear(void)
1284{
1285 int cpu = raw_smp_processor_id();
1286
1287 idle_task_exit();
1288
1289 cpu_uninit();
1290 irq_ctx_exit(cpu);
1291
1292 cpu_clear(cpu, cpu_callout_map);
1293 cpu_clear(cpu, cpu_callin_map);
1294
1295 unmap_cpu_to_logical_apicid(cpu);
1296}
1297# endif /* CONFIG_X86_32 */
1298
1299static void remove_siblinginfo(int cpu) 1217static void remove_siblinginfo(int cpu)
1300{ 1218{
1301 int sibling; 1219 int sibling;
1302 struct cpuinfo_x86 *c = &cpu_data(cpu); 1220 struct cpuinfo_x86 *c = &cpu_data(cpu);
1303 1221
1304 for_each_cpu_mask(sibling, per_cpu(cpu_core_map, cpu)) { 1222 for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
1305 cpu_clear(cpu, per_cpu(cpu_core_map, sibling)); 1223 cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
1306 /*/ 1224 /*/
1307 * last thread sibling in this cpu core going down 1225 * last thread sibling in this cpu core going down
@@ -1310,7 +1228,7 @@ static void remove_siblinginfo(int cpu)
1310 cpu_data(sibling).booted_cores--; 1228 cpu_data(sibling).booted_cores--;
1311 } 1229 }
1312 1230
1313 for_each_cpu_mask(sibling, per_cpu(cpu_sibling_map, cpu)) 1231 for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
1314 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling)); 1232 cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
1315 cpus_clear(per_cpu(cpu_sibling_map, cpu)); 1233 cpus_clear(per_cpu(cpu_sibling_map, cpu));
1316 cpus_clear(per_cpu(cpu_core_map, cpu)); 1234 cpus_clear(per_cpu(cpu_core_map, cpu));
@@ -1349,12 +1267,20 @@ __init void prefill_possible_map(void)
1349 int i; 1267 int i;
1350 int possible; 1268 int possible;
1351 1269
1270 /* no processor from mptable or madt */
1271 if (!num_processors)
1272 num_processors = 1;
1273
1274#ifdef CONFIG_HOTPLUG_CPU
1352 if (additional_cpus == -1) { 1275 if (additional_cpus == -1) {
1353 if (disabled_cpus > 0) 1276 if (disabled_cpus > 0)
1354 additional_cpus = disabled_cpus; 1277 additional_cpus = disabled_cpus;
1355 else 1278 else
1356 additional_cpus = 0; 1279 additional_cpus = 0;
1357 } 1280 }
1281#else
1282 additional_cpus = 0;
1283#endif
1358 possible = num_processors + additional_cpus; 1284 possible = num_processors + additional_cpus;
1359 if (possible > NR_CPUS) 1285 if (possible > NR_CPUS)
1360 possible = NR_CPUS; 1286 possible = NR_CPUS;
@@ -1364,18 +1290,18 @@ __init void prefill_possible_map(void)
1364 1290
1365 for (i = 0; i < possible; i++) 1291 for (i = 0; i < possible; i++)
1366 cpu_set(i, cpu_possible_map); 1292 cpu_set(i, cpu_possible_map);
1293
1294 nr_cpu_ids = possible;
1367} 1295}
1368 1296
1369static void __ref remove_cpu_from_maps(int cpu) 1297static void __ref remove_cpu_from_maps(int cpu)
1370{ 1298{
1371 cpu_clear(cpu, cpu_online_map); 1299 cpu_clear(cpu, cpu_online_map);
1372#ifdef CONFIG_X86_64
1373 cpu_clear(cpu, cpu_callout_map); 1300 cpu_clear(cpu, cpu_callout_map);
1374 cpu_clear(cpu, cpu_callin_map); 1301 cpu_clear(cpu, cpu_callin_map);
1375 /* was set by cpu_init() */ 1302 /* was set by cpu_init() */
1376 clear_bit(cpu, (unsigned long *)&cpu_initialized); 1303 cpu_clear(cpu, cpu_initialized);
1377 clear_node_cpumask(cpu); 1304 numa_remove_cpu(cpu);
1378#endif
1379} 1305}
1380 1306
1381int __cpu_disable(void) 1307int __cpu_disable(void)
@@ -1453,7 +1379,8 @@ static int __init parse_maxcpus(char *arg)
1453{ 1379{
1454 extern unsigned int maxcpus; 1380 extern unsigned int maxcpus;
1455 1381
1456 maxcpus = simple_strtoul(arg, NULL, 0); 1382 if (arg)
1383 maxcpus = simple_strtoul(arg, NULL, 0);
1457 return 0; 1384 return 0;
1458} 1385}
1459early_param("maxcpus", parse_maxcpus); 1386early_param("maxcpus", parse_maxcpus);
diff --git a/arch/x86/kernel/smpcommon.c b/arch/x86/kernel/smpcommon.c
index 3449064d141a..99941b37eca0 100644
--- a/arch/x86/kernel/smpcommon.c
+++ b/arch/x86/kernel/smpcommon.c
@@ -25,59 +25,3 @@ __cpuinit void init_gdt(int cpu)
25 per_cpu(cpu_number, cpu) = cpu; 25 per_cpu(cpu_number, cpu) = cpu;
26} 26}
27#endif 27#endif
28
29/**
30 * smp_call_function(): Run a function on all other CPUs.
31 * @func: The function to run. This must be fast and non-blocking.
32 * @info: An arbitrary pointer to pass to the function.
33 * @nonatomic: Unused.
34 * @wait: If true, wait (atomically) until function has completed on other CPUs.
35 *
36 * Returns 0 on success, else a negative status code.
37 *
38 * If @wait is true, then returns once @func has returned; otherwise
39 * it returns just before the target cpu calls @func.
40 *
41 * You must not call this function with disabled interrupts or from a
42 * hardware interrupt handler or from a bottom half handler.
43 */
44int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
45 int wait)
46{
47 return smp_call_function_mask(cpu_online_map, func, info, wait);
48}
49EXPORT_SYMBOL(smp_call_function);
50
51/**
52 * smp_call_function_single - Run a function on a specific CPU
53 * @cpu: The target CPU. Cannot be the calling CPU.
54 * @func: The function to run. This must be fast and non-blocking.
55 * @info: An arbitrary pointer to pass to the function.
56 * @nonatomic: Unused.
57 * @wait: If true, wait until function has completed on other CPUs.
58 *
59 * Returns 0 on success, else a negative status code.
60 *
61 * If @wait is true, then returns once @func has returned; otherwise
62 * it returns just before the target cpu calls @func.
63 */
64int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
65 int nonatomic, int wait)
66{
67 /* prevent preemption and reschedule on another processor */
68 int ret;
69 int me = get_cpu();
70 if (cpu == me) {
71 local_irq_disable();
72 func(info);
73 local_irq_enable();
74 put_cpu();
75 return 0;
76 }
77
78 ret = smp_call_function_mask(cpumask_of_cpu(cpu), func, info, wait);
79
80 put_cpu();
81 return ret;
82}
83EXPORT_SYMBOL(smp_call_function_single);
diff --git a/arch/x86/kernel/smpcommon_32.c b/arch/x86/kernel/smpcommon_32.c
deleted file mode 100644
index 8b137891791f..000000000000
--- a/arch/x86/kernel/smpcommon_32.c
+++ /dev/null
@@ -1 +0,0 @@
1
diff --git a/arch/x86/kernel/srat_32.c b/arch/x86/kernel/srat_32.c
deleted file mode 100644
index 70e4a374b4e8..000000000000
--- a/arch/x86/kernel/srat_32.c
+++ /dev/null
@@ -1,358 +0,0 @@
1/*
2 * Some of the code in this file has been gleaned from the 64 bit
3 * discontigmem support code base.
4 *
5 * Copyright (C) 2002, IBM Corp.
6 *
7 * All rights reserved.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
17 * NON INFRINGEMENT. See the GNU General Public License for more
18 * details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 *
24 * Send feedback to Pat Gaughen <gone@us.ibm.com>
25 */
26#include <linux/mm.h>
27#include <linux/bootmem.h>
28#include <linux/mmzone.h>
29#include <linux/acpi.h>
30#include <linux/nodemask.h>
31#include <asm/srat.h>
32#include <asm/topology.h>
33#include <asm/smp.h>
34
35/*
36 * proximity macros and definitions
37 */
38#define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
39#define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
40#define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
41#define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
42/* bitmap length; _PXM is at most 255 */
43#define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
44static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
45
46#define MAX_CHUNKS_PER_NODE 3
47#define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
48struct node_memory_chunk_s {
49 unsigned long start_pfn;
50 unsigned long end_pfn;
51 u8 pxm; // proximity domain of node
52 u8 nid; // which cnode contains this chunk?
53 u8 bank; // which mem bank on this node
54};
55static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
56
57static int num_memory_chunks; /* total number of memory chunks */
58static u8 __initdata apicid_to_pxm[MAX_APICID];
59
60/* Identify CPU proximity domains */
61static void __init parse_cpu_affinity_structure(char *p)
62{
63 struct acpi_srat_cpu_affinity *cpu_affinity =
64 (struct acpi_srat_cpu_affinity *) p;
65
66 if ((cpu_affinity->flags & ACPI_SRAT_CPU_ENABLED) == 0)
67 return; /* empty entry */
68
69 /* mark this node as "seen" in node bitmap */
70 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain_lo);
71
72 apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain_lo;
73
74 printk("CPU 0x%02X in proximity domain 0x%02X\n",
75 cpu_affinity->apic_id, cpu_affinity->proximity_domain_lo);
76}
77
78/*
79 * Identify memory proximity domains and hot-remove capabilities.
80 * Fill node memory chunk list structure.
81 */
82static void __init parse_memory_affinity_structure (char *sratp)
83{
84 unsigned long long paddr, size;
85 unsigned long start_pfn, end_pfn;
86 u8 pxm;
87 struct node_memory_chunk_s *p, *q, *pend;
88 struct acpi_srat_mem_affinity *memory_affinity =
89 (struct acpi_srat_mem_affinity *) sratp;
90
91 if ((memory_affinity->flags & ACPI_SRAT_MEM_ENABLED) == 0)
92 return; /* empty entry */
93
94 pxm = memory_affinity->proximity_domain & 0xff;
95
96 /* mark this node as "seen" in node bitmap */
97 BMAP_SET(pxm_bitmap, pxm);
98
99 /* calculate info for memory chunk structure */
100 paddr = memory_affinity->base_address;
101 size = memory_affinity->length;
102
103 start_pfn = paddr >> PAGE_SHIFT;
104 end_pfn = (paddr + size) >> PAGE_SHIFT;
105
106
107 if (num_memory_chunks >= MAXCHUNKS) {
108 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
109 size/(1024*1024), paddr);
110 return;
111 }
112
113 /* Insertion sort based on base address */
114 pend = &node_memory_chunk[num_memory_chunks];
115 for (p = &node_memory_chunk[0]; p < pend; p++) {
116 if (start_pfn < p->start_pfn)
117 break;
118 }
119 if (p < pend) {
120 for (q = pend; q >= p; q--)
121 *(q + 1) = *q;
122 }
123 p->start_pfn = start_pfn;
124 p->end_pfn = end_pfn;
125 p->pxm = pxm;
126
127 num_memory_chunks++;
128
129 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
130 start_pfn, end_pfn,
131 memory_affinity->memory_type,
132 pxm,
133 ((memory_affinity->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
134 "enabled and removable" : "enabled" ) );
135}
136
137/*
138 * The SRAT table always lists ascending addresses, so can always
139 * assume that the first "start" address that you see is the real
140 * start of the node, and that the current "end" address is after
141 * the previous one.
142 */
143static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
144{
145 /*
146 * Only add present memory as told by the e820.
147 * There is no guarantee from the SRAT that the memory it
148 * enumerates is present at boot time because it represents
149 * *possible* memory hotplug areas the same as normal RAM.
150 */
151 if (memory_chunk->start_pfn >= max_pfn) {
152 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
153 memory_chunk->start_pfn, memory_chunk->end_pfn);
154 return;
155 }
156 if (memory_chunk->nid != nid)
157 return;
158
159 if (!node_has_online_mem(nid))
160 node_start_pfn[nid] = memory_chunk->start_pfn;
161
162 if (node_start_pfn[nid] > memory_chunk->start_pfn)
163 node_start_pfn[nid] = memory_chunk->start_pfn;
164
165 if (node_end_pfn[nid] < memory_chunk->end_pfn)
166 node_end_pfn[nid] = memory_chunk->end_pfn;
167}
168
169/* Parse the ACPI Static Resource Affinity Table */
170static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
171{
172 u8 *start, *end, *p;
173 int i, j, nid;
174
175 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
176 p = start;
177 end = (u8 *)sratp + sratp->header.length;
178
179 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
180 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
181
182 num_memory_chunks = 0;
183 while (p < end) {
184 switch (*p) {
185 case ACPI_SRAT_TYPE_CPU_AFFINITY:
186 parse_cpu_affinity_structure(p);
187 break;
188 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
189 parse_memory_affinity_structure(p);
190 break;
191 default:
192 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
193 break;
194 }
195 p += p[1];
196 if (p[1] == 0) {
197 printk("acpi20_parse_srat: Entry length value is zero;"
198 " can't parse any further!\n");
199 break;
200 }
201 }
202
203 if (num_memory_chunks == 0) {
204 printk("could not finy any ACPI SRAT memory areas.\n");
205 goto out_fail;
206 }
207
208 /* Calculate total number of nodes in system from PXM bitmap and create
209 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
210 * to specify the range of _PXM values.)
211 */
212 /*
213 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
214 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
215 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
216 * approaches MAX_PXM_DOMAINS for i386.
217 */
218 nodes_clear(node_online_map);
219 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
220 if (BMAP_TEST(pxm_bitmap, i)) {
221 int nid = acpi_map_pxm_to_node(i);
222 node_set_online(nid);
223 }
224 }
225 BUG_ON(num_online_nodes() == 0);
226
227 /* set cnode id in memory chunk structure */
228 for (i = 0; i < num_memory_chunks; i++)
229 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
230
231 printk("pxm bitmap: ");
232 for (i = 0; i < sizeof(pxm_bitmap); i++) {
233 printk("%02X ", pxm_bitmap[i]);
234 }
235 printk("\n");
236 printk("Number of logical nodes in system = %d\n", num_online_nodes());
237 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
238
239 for (i = 0; i < MAX_APICID; i++)
240 apicid_2_node[i] = pxm_to_node(apicid_to_pxm[i]);
241
242 for (j = 0; j < num_memory_chunks; j++){
243 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
244 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
245 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
246 node_read_chunk(chunk->nid, chunk);
247 add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn);
248 }
249
250 for_each_online_node(nid) {
251 unsigned long start = node_start_pfn[nid];
252 unsigned long end = node_end_pfn[nid];
253
254 memory_present(nid, start, end);
255 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
256 }
257 return 1;
258out_fail:
259 return 0;
260}
261
262struct acpi_static_rsdt {
263 struct acpi_table_rsdt table;
264 u32 padding[7]; /* Allow for 7 more table entries */
265};
266
267int __init get_memcfg_from_srat(void)
268{
269 struct acpi_table_header *header = NULL;
270 struct acpi_table_rsdp *rsdp = NULL;
271 struct acpi_table_rsdt *rsdt = NULL;
272 acpi_native_uint rsdp_address = 0;
273 struct acpi_static_rsdt saved_rsdt;
274 int tables = 0;
275 int i = 0;
276
277 rsdp_address = acpi_os_get_root_pointer();
278 if (!rsdp_address) {
279 printk("%s: System description tables not found\n",
280 __func__);
281 goto out_err;
282 }
283
284 printk("%s: assigning address to rsdp\n", __func__);
285 rsdp = (struct acpi_table_rsdp *)(u32)rsdp_address;
286 if (!rsdp) {
287 printk("%s: Didn't find ACPI root!\n", __func__);
288 goto out_err;
289 }
290
291 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
292 rsdp->oem_id);
293
294 if (strncmp(rsdp->signature, ACPI_SIG_RSDP,strlen(ACPI_SIG_RSDP))) {
295 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __func__);
296 goto out_err;
297 }
298
299 rsdt = (struct acpi_table_rsdt *)
300 early_ioremap(rsdp->rsdt_physical_address, sizeof(struct acpi_table_rsdt));
301
302 if (!rsdt) {
303 printk(KERN_WARNING
304 "%s: ACPI: Invalid root system description tables (RSDT)\n",
305 __func__);
306 goto out_err;
307 }
308
309 header = &rsdt->header;
310
311 if (strncmp(header->signature, ACPI_SIG_RSDT, strlen(ACPI_SIG_RSDT))) {
312 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
313 goto out_err;
314 }
315
316 /*
317 * The number of tables is computed by taking the
318 * size of all entries (header size minus total
319 * size of RSDT) divided by the size of each entry
320 * (4-byte table pointers).
321 */
322 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
323
324 if (!tables)
325 goto out_err;
326
327 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
328
329 if (saved_rsdt.table.header.length > sizeof(saved_rsdt)) {
330 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
331 saved_rsdt.table.header.length);
332 goto out_err;
333 }
334
335 printk("Begin SRAT table scan....\n");
336
337 for (i = 0; i < tables; i++) {
338 /* Map in header, then map in full table length. */
339 header = (struct acpi_table_header *)
340 early_ioremap(saved_rsdt.table.table_offset_entry[i], sizeof(struct acpi_table_header));
341 if (!header)
342 break;
343 header = (struct acpi_table_header *)
344 early_ioremap(saved_rsdt.table.table_offset_entry[i], header->length);
345 if (!header)
346 break;
347
348 if (strncmp((char *) &header->signature, ACPI_SIG_SRAT, 4))
349 continue;
350
351 /* we've found the srat table. don't need to look at any more tables */
352 return acpi20_parse_srat((struct acpi_table_srat *)header);
353 }
354out_err:
355 remove_all_active_ranges();
356 printk("failed to get NUMA memory information from SRAT table\n");
357 return 0;
358}
diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
index c28c342c162f..a03e7f6d90c3 100644
--- a/arch/x86/kernel/stacktrace.c
+++ b/arch/x86/kernel/stacktrace.c
@@ -74,6 +74,7 @@ void save_stack_trace(struct stack_trace *trace)
74 if (trace->nr_entries < trace->max_entries) 74 if (trace->nr_entries < trace->max_entries)
75 trace->entries[trace->nr_entries++] = ULONG_MAX; 75 trace->entries[trace->nr_entries++] = ULONG_MAX;
76} 76}
77EXPORT_SYMBOL_GPL(save_stack_trace);
77 78
78void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace) 79void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
79{ 80{
@@ -81,3 +82,4 @@ void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
81 if (trace->nr_entries < trace->max_entries) 82 if (trace->nr_entries < trace->max_entries)
82 trace->entries[trace->nr_entries++] = ULONG_MAX; 83 trace->entries[trace->nr_entries++] = ULONG_MAX;
83} 84}
85EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
diff --git a/arch/x86/kernel/step.c b/arch/x86/kernel/step.c
index 92c20fee6781..e8b9863ef8c4 100644
--- a/arch/x86/kernel/step.c
+++ b/arch/x86/kernel/step.c
@@ -105,6 +105,20 @@ static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
105static int enable_single_step(struct task_struct *child) 105static int enable_single_step(struct task_struct *child)
106{ 106{
107 struct pt_regs *regs = task_pt_regs(child); 107 struct pt_regs *regs = task_pt_regs(child);
108 unsigned long oflags;
109
110 /*
111 * If we stepped into a sysenter/syscall insn, it trapped in
112 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
113 * If user-mode had set TF itself, then it's still clear from
114 * do_debug() and we need to set it again to restore the user
115 * state so we don't wrongly set TIF_FORCED_TF below.
116 * If enable_single_step() was used last and that is what
117 * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
118 * already set and our bookkeeping is fine.
119 */
120 if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
121 regs->flags |= X86_EFLAGS_TF;
108 122
109 /* 123 /*
110 * Always set TIF_SINGLESTEP - this guarantees that 124 * Always set TIF_SINGLESTEP - this guarantees that
@@ -113,11 +127,7 @@ static int enable_single_step(struct task_struct *child)
113 */ 127 */
114 set_tsk_thread_flag(child, TIF_SINGLESTEP); 128 set_tsk_thread_flag(child, TIF_SINGLESTEP);
115 129
116 /* 130 oflags = regs->flags;
117 * If TF was already set, don't do anything else
118 */
119 if (regs->flags & X86_EFLAGS_TF)
120 return 0;
121 131
122 /* Set TF on the kernel stack.. */ 132 /* Set TF on the kernel stack.. */
123 regs->flags |= X86_EFLAGS_TF; 133 regs->flags |= X86_EFLAGS_TF;
@@ -126,9 +136,22 @@ static int enable_single_step(struct task_struct *child)
126 * ..but if TF is changed by the instruction we will trace, 136 * ..but if TF is changed by the instruction we will trace,
127 * don't mark it as being "us" that set it, so that we 137 * don't mark it as being "us" that set it, so that we
128 * won't clear it by hand later. 138 * won't clear it by hand later.
139 *
140 * Note that if we don't actually execute the popf because
141 * of a signal arriving right now or suchlike, we will lose
142 * track of the fact that it really was "us" that set it.
129 */ 143 */
130 if (is_setting_trap_flag(child, regs)) 144 if (is_setting_trap_flag(child, regs)) {
145 clear_tsk_thread_flag(child, TIF_FORCED_TF);
131 return 0; 146 return 0;
147 }
148
149 /*
150 * If TF was already set, check whether it was us who set it.
151 * If not, we should never attempt a block step.
152 */
153 if (oflags & X86_EFLAGS_TF)
154 return test_tsk_thread_flag(child, TIF_FORCED_TF);
132 155
133 set_tsk_thread_flag(child, TIF_FORCED_TF); 156 set_tsk_thread_flag(child, TIF_FORCED_TF);
134 157
diff --git a/arch/x86/kernel/summit_32.c b/arch/x86/kernel/summit_32.c
index ae751094eba9..d67ce5f044ba 100644
--- a/arch/x86/kernel/summit_32.c
+++ b/arch/x86/kernel/summit_32.c
@@ -36,7 +36,9 @@ static 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;
37static struct rio_detail *rio_devs[MAX_NUMNODES*4] __initdata; 37static struct rio_detail *rio_devs[MAX_NUMNODES*4] __initdata;
38 38
39#ifndef CONFIG_X86_NUMAQ
39static int mp_bus_id_to_node[MAX_MP_BUSSES] __initdata; 40static int mp_bus_id_to_node[MAX_MP_BUSSES] __initdata;
41#endif
40 42
41static int __init setup_pci_node_map_for_wpeg(int wpeg_num, int last_bus) 43static int __init setup_pci_node_map_for_wpeg(int wpeg_num, int last_bus)
42{ 44{
diff --git a/arch/x86/kernel/sys_i386_32.c b/arch/x86/kernel/sys_i386_32.c
index d2ab52cc1d6b..7066cb855a60 100644
--- a/arch/x86/kernel/sys_i386_32.c
+++ b/arch/x86/kernel/sys_i386_32.c
@@ -19,8 +19,8 @@
19#include <linux/utsname.h> 19#include <linux/utsname.h>
20#include <linux/ipc.h> 20#include <linux/ipc.h>
21 21
22#include <asm/uaccess.h> 22#include <linux/uaccess.h>
23#include <asm/unistd.h> 23#include <linux/unistd.h>
24 24
25asmlinkage long sys_mmap2(unsigned long addr, unsigned long len, 25asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
26 unsigned long prot, unsigned long flags, 26 unsigned long prot, unsigned long flags,
@@ -103,7 +103,7 @@ asmlinkage int old_select(struct sel_arg_struct __user *arg)
103 * 103 *
104 * This is really horribly ugly. 104 * This is really horribly ugly.
105 */ 105 */
106asmlinkage int sys_ipc (uint call, int first, int second, 106asmlinkage int sys_ipc(uint call, int first, int second,
107 int third, void __user *ptr, long fifth) 107 int third, void __user *ptr, long fifth)
108{ 108{
109 int version, ret; 109 int version, ret;
@@ -113,24 +113,24 @@ asmlinkage int sys_ipc (uint call, int first, int second,
113 113
114 switch (call) { 114 switch (call) {
115 case SEMOP: 115 case SEMOP:
116 return sys_semtimedop (first, (struct sembuf __user *)ptr, second, NULL); 116 return sys_semtimedop(first, (struct sembuf __user *)ptr, second, NULL);
117 case SEMTIMEDOP: 117 case SEMTIMEDOP:
118 return sys_semtimedop(first, (struct sembuf __user *)ptr, second, 118 return sys_semtimedop(first, (struct sembuf __user *)ptr, second,
119 (const struct timespec __user *)fifth); 119 (const struct timespec __user *)fifth);
120 120
121 case SEMGET: 121 case SEMGET:
122 return sys_semget (first, second, third); 122 return sys_semget(first, second, third);
123 case SEMCTL: { 123 case SEMCTL: {
124 union semun fourth; 124 union semun fourth;
125 if (!ptr) 125 if (!ptr)
126 return -EINVAL; 126 return -EINVAL;
127 if (get_user(fourth.__pad, (void __user * __user *) ptr)) 127 if (get_user(fourth.__pad, (void __user * __user *) ptr))
128 return -EFAULT; 128 return -EFAULT;
129 return sys_semctl (first, second, third, fourth); 129 return sys_semctl(first, second, third, fourth);
130 } 130 }
131 131
132 case MSGSND: 132 case MSGSND:
133 return sys_msgsnd (first, (struct msgbuf __user *) ptr, 133 return sys_msgsnd(first, (struct msgbuf __user *) ptr,
134 second, third); 134 second, third);
135 case MSGRCV: 135 case MSGRCV:
136 switch (version) { 136 switch (version) {
@@ -138,45 +138,45 @@ asmlinkage int sys_ipc (uint call, int first, int second,
138 struct ipc_kludge tmp; 138 struct ipc_kludge tmp;
139 if (!ptr) 139 if (!ptr)
140 return -EINVAL; 140 return -EINVAL;
141 141
142 if (copy_from_user(&tmp, 142 if (copy_from_user(&tmp,
143 (struct ipc_kludge __user *) ptr, 143 (struct ipc_kludge __user *) ptr,
144 sizeof (tmp))) 144 sizeof(tmp)))
145 return -EFAULT; 145 return -EFAULT;
146 return sys_msgrcv (first, tmp.msgp, second, 146 return sys_msgrcv(first, tmp.msgp, second,
147 tmp.msgtyp, third); 147 tmp.msgtyp, third);
148 } 148 }
149 default: 149 default:
150 return sys_msgrcv (first, 150 return sys_msgrcv(first,
151 (struct msgbuf __user *) ptr, 151 (struct msgbuf __user *) ptr,
152 second, fifth, third); 152 second, fifth, third);
153 } 153 }
154 case MSGGET: 154 case MSGGET:
155 return sys_msgget ((key_t) first, second); 155 return sys_msgget((key_t) first, second);
156 case MSGCTL: 156 case MSGCTL:
157 return sys_msgctl (first, second, (struct msqid_ds __user *) ptr); 157 return sys_msgctl(first, second, (struct msqid_ds __user *) ptr);
158 158
159 case SHMAT: 159 case SHMAT:
160 switch (version) { 160 switch (version) {
161 default: { 161 default: {
162 ulong raddr; 162 ulong raddr;
163 ret = do_shmat (first, (char __user *) ptr, second, &raddr); 163 ret = do_shmat(first, (char __user *) ptr, second, &raddr);
164 if (ret) 164 if (ret)
165 return ret; 165 return ret;
166 return put_user (raddr, (ulong __user *) third); 166 return put_user(raddr, (ulong __user *) third);
167 } 167 }
168 case 1: /* iBCS2 emulator entry point */ 168 case 1: /* iBCS2 emulator entry point */
169 if (!segment_eq(get_fs(), get_ds())) 169 if (!segment_eq(get_fs(), get_ds()))
170 return -EINVAL; 170 return -EINVAL;
171 /* The "(ulong *) third" is valid _only_ because of the kernel segment thing */ 171 /* The "(ulong *) third" is valid _only_ because of the kernel segment thing */
172 return do_shmat (first, (char __user *) ptr, second, (ulong *) third); 172 return do_shmat(first, (char __user *) ptr, second, (ulong *) third);
173 } 173 }
174 case SHMDT: 174 case SHMDT:
175 return sys_shmdt ((char __user *)ptr); 175 return sys_shmdt((char __user *)ptr);
176 case SHMGET: 176 case SHMGET:
177 return sys_shmget (first, second, third); 177 return sys_shmget(first, second, third);
178 case SHMCTL: 178 case SHMCTL:
179 return sys_shmctl (first, second, 179 return sys_shmctl(first, second,
180 (struct shmid_ds __user *) ptr); 180 (struct shmid_ds __user *) ptr);
181 default: 181 default:
182 return -ENOSYS; 182 return -ENOSYS;
@@ -186,28 +186,28 @@ asmlinkage int sys_ipc (uint call, int first, int second,
186/* 186/*
187 * Old cruft 187 * Old cruft
188 */ 188 */
189asmlinkage int sys_uname(struct old_utsname __user * name) 189asmlinkage int sys_uname(struct old_utsname __user *name)
190{ 190{
191 int err; 191 int err;
192 if (!name) 192 if (!name)
193 return -EFAULT; 193 return -EFAULT;
194 down_read(&uts_sem); 194 down_read(&uts_sem);
195 err = copy_to_user(name, utsname(), sizeof (*name)); 195 err = copy_to_user(name, utsname(), sizeof(*name));
196 up_read(&uts_sem); 196 up_read(&uts_sem);
197 return err?-EFAULT:0; 197 return err? -EFAULT:0;
198} 198}
199 199
200asmlinkage int sys_olduname(struct oldold_utsname __user * name) 200asmlinkage int sys_olduname(struct oldold_utsname __user *name)
201{ 201{
202 int error; 202 int error;
203 203
204 if (!name) 204 if (!name)
205 return -EFAULT; 205 return -EFAULT;
206 if (!access_ok(VERIFY_WRITE,name,sizeof(struct oldold_utsname))) 206 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
207 return -EFAULT; 207 return -EFAULT;
208 208
209 down_read(&uts_sem); 209 down_read(&uts_sem);
210 210
211 error = __copy_to_user(&name->sysname, &utsname()->sysname, 211 error = __copy_to_user(&name->sysname, &utsname()->sysname,
212 __OLD_UTS_LEN); 212 __OLD_UTS_LEN);
213 error |= __put_user(0, name->sysname + __OLD_UTS_LEN); 213 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
@@ -223,9 +223,9 @@ asmlinkage int sys_olduname(struct oldold_utsname __user * name)
223 error |= __copy_to_user(&name->machine, &utsname()->machine, 223 error |= __copy_to_user(&name->machine, &utsname()->machine,
224 __OLD_UTS_LEN); 224 __OLD_UTS_LEN);
225 error |= __put_user(0, name->machine + __OLD_UTS_LEN); 225 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
226 226
227 up_read(&uts_sem); 227 up_read(&uts_sem);
228 228
229 error = error ? -EFAULT : 0; 229 error = error ? -EFAULT : 0;
230 230
231 return error; 231 return error;
@@ -241,6 +241,6 @@ int kernel_execve(const char *filename, char *const argv[], char *const envp[])
241 long __res; 241 long __res;
242 asm volatile ("push %%ebx ; movl %2,%%ebx ; int $0x80 ; pop %%ebx" 242 asm volatile ("push %%ebx ; movl %2,%%ebx ; int $0x80 ; pop %%ebx"
243 : "=a" (__res) 243 : "=a" (__res)
244 : "0" (__NR_execve),"ri" (filename),"c" (argv), "d" (envp) : "memory"); 244 : "0" (__NR_execve), "ri" (filename), "c" (argv), "d" (envp) : "memory");
245 return __res; 245 return __res;
246} 246}
diff --git a/arch/x86/kernel/syscall_table_32.S b/arch/x86/kernel/syscall_table_32.S
index adff5562f5fd..d44395ff34c3 100644
--- a/arch/x86/kernel/syscall_table_32.S
+++ b/arch/x86/kernel/syscall_table_32.S
@@ -326,3 +326,9 @@ ENTRY(sys_call_table)
326 .long sys_fallocate 326 .long sys_fallocate
327 .long sys_timerfd_settime /* 325 */ 327 .long sys_timerfd_settime /* 325 */
328 .long sys_timerfd_gettime 328 .long sys_timerfd_gettime
329 .long sys_signalfd4
330 .long sys_eventfd2
331 .long sys_epoll_create1
332 .long sys_dup3 /* 330 */
333 .long sys_pipe2
334 .long sys_inotify_init1
diff --git a/arch/x86/kernel/time_32.c b/arch/x86/kernel/time_32.c
index 2ff21f398934..ffe3c664afc0 100644
--- a/arch/x86/kernel/time_32.c
+++ b/arch/x86/kernel/time_32.c
@@ -39,9 +39,6 @@
39 39
40#include "do_timer.h" 40#include "do_timer.h"
41 41
42unsigned int cpu_khz; /* Detected as we calibrate the TSC */
43EXPORT_SYMBOL(cpu_khz);
44
45int timer_ack; 42int timer_ack;
46 43
47unsigned long profile_pc(struct pt_regs *regs) 44unsigned long profile_pc(struct pt_regs *regs)
@@ -84,8 +81,7 @@ irqreturn_t timer_interrupt(int irq, void *dev_id)
84 if (timer_ack) { 81 if (timer_ack) {
85 /* 82 /*
86 * Subtle, when I/O APICs are used we have to ack timer IRQ 83 * Subtle, when I/O APICs are used we have to ack timer IRQ
87 * manually to reset the IRR bit for do_slow_gettimeoffset(). 84 * manually to deassert NMI lines for the watchdog if run
88 * This will also deassert NMI lines for the watchdog if run
89 * on an 82489DX-based system. 85 * on an 82489DX-based system.
90 */ 86 */
91 spin_lock(&i8259A_lock); 87 spin_lock(&i8259A_lock);
@@ -133,6 +129,7 @@ void __init hpet_time_init(void)
133 */ 129 */
134void __init time_init(void) 130void __init time_init(void)
135{ 131{
132 pre_time_init_hook();
136 tsc_init(); 133 tsc_init();
137 late_time_init = choose_time_init(); 134 late_time_init = choose_time_init();
138} 135}
diff --git a/arch/x86/kernel/time_64.c b/arch/x86/kernel/time_64.c
index c737849e2ef7..e3d49c553af2 100644
--- a/arch/x86/kernel/time_64.c
+++ b/arch/x86/kernel/time_64.c
@@ -56,7 +56,7 @@ static irqreturn_t timer_event_interrupt(int irq, void *dev_id)
56/* calibrate_cpu is used on systems with fixed rate TSCs to determine 56/* calibrate_cpu is used on systems with fixed rate TSCs to determine
57 * processor frequency */ 57 * processor frequency */
58#define TICK_COUNT 100000000 58#define TICK_COUNT 100000000
59unsigned long __init native_calculate_cpu_khz(void) 59unsigned long __init calibrate_cpu(void)
60{ 60{
61 int tsc_start, tsc_now; 61 int tsc_start, tsc_now;
62 int i, no_ctr_free; 62 int i, no_ctr_free;
@@ -116,23 +116,11 @@ void __init hpet_time_init(void)
116 116
117void __init time_init(void) 117void __init time_init(void)
118{ 118{
119 tsc_calibrate(); 119 tsc_init();
120
121 cpu_khz = tsc_khz;
122 if (cpu_has(&boot_cpu_data, X86_FEATURE_CONSTANT_TSC) &&
123 (boot_cpu_data.x86_vendor == X86_VENDOR_AMD))
124 cpu_khz = calculate_cpu_khz();
125
126 if (unsynchronized_tsc())
127 mark_tsc_unstable("TSCs unsynchronized");
128
129 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP)) 120 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
130 vgetcpu_mode = VGETCPU_RDTSCP; 121 vgetcpu_mode = VGETCPU_RDTSCP;
131 else 122 else
132 vgetcpu_mode = VGETCPU_LSL; 123 vgetcpu_mode = VGETCPU_LSL;
133 124
134 printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n",
135 cpu_khz / 1000, cpu_khz % 1000);
136 init_tsc_clocksource();
137 late_time_init = choose_time_init(); 125 late_time_init = choose_time_init();
138} 126}
diff --git a/arch/x86/kernel/tlb_32.c b/arch/x86/kernel/tlb_32.c
index 9bb2363851af..fec1ecedc9b7 100644
--- a/arch/x86/kernel/tlb_32.c
+++ b/arch/x86/kernel/tlb_32.c
@@ -238,6 +238,6 @@ static void do_flush_tlb_all(void *info)
238 238
239void flush_tlb_all(void) 239void flush_tlb_all(void)
240{ 240{
241 on_each_cpu(do_flush_tlb_all, NULL, 1, 1); 241 on_each_cpu(do_flush_tlb_all, NULL, 1);
242} 242}
243 243
diff --git a/arch/x86/kernel/tlb_64.c b/arch/x86/kernel/tlb_64.c
index a1f07d793202..dcbf7a1159ea 100644
--- a/arch/x86/kernel/tlb_64.c
+++ b/arch/x86/kernel/tlb_64.c
@@ -15,6 +15,8 @@
15#include <asm/proto.h> 15#include <asm/proto.h>
16#include <asm/apicdef.h> 16#include <asm/apicdef.h>
17#include <asm/idle.h> 17#include <asm/idle.h>
18#include <asm/uv/uv_hub.h>
19#include <asm/uv/uv_bau.h>
18 20
19#include <mach_ipi.h> 21#include <mach_ipi.h>
20/* 22/*
@@ -162,6 +164,9 @@ void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm,
162 union smp_flush_state *f; 164 union smp_flush_state *f;
163 cpumask_t cpumask = *cpumaskp; 165 cpumask_t cpumask = *cpumaskp;
164 166
167 if (is_uv_system() && uv_flush_tlb_others(&cpumask, mm, va))
168 return;
169
165 /* Caller has disabled preemption */ 170 /* Caller has disabled preemption */
166 sender = smp_processor_id() % NUM_INVALIDATE_TLB_VECTORS; 171 sender = smp_processor_id() % NUM_INVALIDATE_TLB_VECTORS;
167 f = &per_cpu(flush_state, sender); 172 f = &per_cpu(flush_state, sender);
@@ -270,5 +275,5 @@ static void do_flush_tlb_all(void *info)
270 275
271void flush_tlb_all(void) 276void flush_tlb_all(void)
272{ 277{
273 on_each_cpu(do_flush_tlb_all, NULL, 1, 1); 278 on_each_cpu(do_flush_tlb_all, NULL, 1);
274} 279}
diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
new file mode 100644
index 000000000000..d0fbb7712ab0
--- /dev/null
+++ b/arch/x86/kernel/tlb_uv.c
@@ -0,0 +1,792 @@
1/*
2 * SGI UltraViolet TLB flush routines.
3 *
4 * (c) 2008 Cliff Wickman <cpw@sgi.com>, SGI.
5 *
6 * This code is released under the GNU General Public License version 2 or
7 * later.
8 */
9#include <linux/mc146818rtc.h>
10#include <linux/proc_fs.h>
11#include <linux/kernel.h>
12
13#include <asm/mmu_context.h>
14#include <asm/uv/uv_mmrs.h>
15#include <asm/uv/uv_hub.h>
16#include <asm/uv/uv_bau.h>
17#include <asm/genapic.h>
18#include <asm/idle.h>
19#include <asm/tsc.h>
20
21#include <mach_apic.h>
22
23static struct bau_control **uv_bau_table_bases __read_mostly;
24static int uv_bau_retry_limit __read_mostly;
25
26/* position of pnode (which is nasid>>1): */
27static int uv_nshift __read_mostly;
28
29static unsigned long uv_mmask __read_mostly;
30
31static DEFINE_PER_CPU(struct ptc_stats, ptcstats);
32static DEFINE_PER_CPU(struct bau_control, bau_control);
33
34/*
35 * Free a software acknowledge hardware resource by clearing its Pending
36 * bit. This will return a reply to the sender.
37 * If the message has timed out, a reply has already been sent by the
38 * hardware but the resource has not been released. In that case our
39 * clear of the Timeout bit (as well) will free the resource. No reply will
40 * be sent (the hardware will only do one reply per message).
41 */
42static void uv_reply_to_message(int resource,
43 struct bau_payload_queue_entry *msg,
44 struct bau_msg_status *msp)
45{
46 unsigned long dw;
47
48 dw = (1 << (resource + UV_SW_ACK_NPENDING)) | (1 << resource);
49 msg->replied_to = 1;
50 msg->sw_ack_vector = 0;
51 if (msp)
52 msp->seen_by.bits = 0;
53 uv_write_local_mmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, dw);
54}
55
56/*
57 * Do all the things a cpu should do for a TLB shootdown message.
58 * Other cpu's may come here at the same time for this message.
59 */
60static void uv_bau_process_message(struct bau_payload_queue_entry *msg,
61 int msg_slot, int sw_ack_slot)
62{
63 unsigned long this_cpu_mask;
64 struct bau_msg_status *msp;
65 int cpu;
66
67 msp = __get_cpu_var(bau_control).msg_statuses + msg_slot;
68 cpu = uv_blade_processor_id();
69 msg->number_of_cpus =
70 uv_blade_nr_online_cpus(uv_node_to_blade_id(numa_node_id()));
71 this_cpu_mask = 1UL << cpu;
72 if (msp->seen_by.bits & this_cpu_mask)
73 return;
74 atomic_or_long(&msp->seen_by.bits, this_cpu_mask);
75
76 if (msg->replied_to == 1)
77 return;
78
79 if (msg->address == TLB_FLUSH_ALL) {
80 local_flush_tlb();
81 __get_cpu_var(ptcstats).alltlb++;
82 } else {
83 __flush_tlb_one(msg->address);
84 __get_cpu_var(ptcstats).onetlb++;
85 }
86
87 __get_cpu_var(ptcstats).requestee++;
88
89 atomic_inc_short(&msg->acknowledge_count);
90 if (msg->number_of_cpus == msg->acknowledge_count)
91 uv_reply_to_message(sw_ack_slot, msg, msp);
92}
93
94/*
95 * Examine the payload queue on one distribution node to see
96 * which messages have not been seen, and which cpu(s) have not seen them.
97 *
98 * Returns the number of cpu's that have not responded.
99 */
100static int uv_examine_destination(struct bau_control *bau_tablesp, int sender)
101{
102 struct bau_payload_queue_entry *msg;
103 struct bau_msg_status *msp;
104 int count = 0;
105 int i;
106 int j;
107
108 for (msg = bau_tablesp->va_queue_first, i = 0; i < DEST_Q_SIZE;
109 msg++, i++) {
110 if ((msg->sending_cpu == sender) && (!msg->replied_to)) {
111 msp = bau_tablesp->msg_statuses + i;
112 printk(KERN_DEBUG
113 "blade %d: address:%#lx %d of %d, not cpu(s): ",
114 i, msg->address, msg->acknowledge_count,
115 msg->number_of_cpus);
116 for (j = 0; j < msg->number_of_cpus; j++) {
117 if (!((1L << j) & msp->seen_by.bits)) {
118 count++;
119 printk("%d ", j);
120 }
121 }
122 printk("\n");
123 }
124 }
125 return count;
126}
127
128/*
129 * Examine the payload queue on all the distribution nodes to see
130 * which messages have not been seen, and which cpu(s) have not seen them.
131 *
132 * Returns the number of cpu's that have not responded.
133 */
134static int uv_examine_destinations(struct bau_target_nodemask *distribution)
135{
136 int sender;
137 int i;
138 int count = 0;
139
140 sender = smp_processor_id();
141 for (i = 0; i < sizeof(struct bau_target_nodemask) * BITSPERBYTE; i++) {
142 if (!bau_node_isset(i, distribution))
143 continue;
144 count += uv_examine_destination(uv_bau_table_bases[i], sender);
145 }
146 return count;
147}
148
149/*
150 * wait for completion of a broadcast message
151 *
152 * return COMPLETE, RETRY or GIVEUP
153 */
154static int uv_wait_completion(struct bau_desc *bau_desc,
155 unsigned long mmr_offset, int right_shift)
156{
157 int exams = 0;
158 long destination_timeouts = 0;
159 long source_timeouts = 0;
160 unsigned long descriptor_status;
161
162 while ((descriptor_status = (((unsigned long)
163 uv_read_local_mmr(mmr_offset) >>
164 right_shift) & UV_ACT_STATUS_MASK)) !=
165 DESC_STATUS_IDLE) {
166 if (descriptor_status == DESC_STATUS_SOURCE_TIMEOUT) {
167 source_timeouts++;
168 if (source_timeouts > SOURCE_TIMEOUT_LIMIT)
169 source_timeouts = 0;
170 __get_cpu_var(ptcstats).s_retry++;
171 return FLUSH_RETRY;
172 }
173 /*
174 * spin here looking for progress at the destinations
175 */
176 if (descriptor_status == DESC_STATUS_DESTINATION_TIMEOUT) {
177 destination_timeouts++;
178 if (destination_timeouts > DESTINATION_TIMEOUT_LIMIT) {
179 /*
180 * returns number of cpus not responding
181 */
182 if (uv_examine_destinations
183 (&bau_desc->distribution) == 0) {
184 __get_cpu_var(ptcstats).d_retry++;
185 return FLUSH_RETRY;
186 }
187 exams++;
188 if (exams >= uv_bau_retry_limit) {
189 printk(KERN_DEBUG
190 "uv_flush_tlb_others");
191 printk("giving up on cpu %d\n",
192 smp_processor_id());
193 return FLUSH_GIVEUP;
194 }
195 /*
196 * delays can hang the simulator
197 udelay(1000);
198 */
199 destination_timeouts = 0;
200 }
201 }
202 }
203 return FLUSH_COMPLETE;
204}
205
206/**
207 * uv_flush_send_and_wait
208 *
209 * Send a broadcast and wait for a broadcast message to complete.
210 *
211 * The cpumaskp mask contains the cpus the broadcast was sent to.
212 *
213 * Returns 1 if all remote flushing was done. The mask is zeroed.
214 * Returns 0 if some remote flushing remains to be done. The mask is left
215 * unchanged.
216 */
217int uv_flush_send_and_wait(int cpu, int this_blade, struct bau_desc *bau_desc,
218 cpumask_t *cpumaskp)
219{
220 int completion_status = 0;
221 int right_shift;
222 int tries = 0;
223 int blade;
224 int bit;
225 unsigned long mmr_offset;
226 unsigned long index;
227 cycles_t time1;
228 cycles_t time2;
229
230 if (cpu < UV_CPUS_PER_ACT_STATUS) {
231 mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0;
232 right_shift = cpu * UV_ACT_STATUS_SIZE;
233 } else {
234 mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1;
235 right_shift =
236 ((cpu - UV_CPUS_PER_ACT_STATUS) * UV_ACT_STATUS_SIZE);
237 }
238 time1 = get_cycles();
239 do {
240 tries++;
241 index = (1UL << UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT) |
242 cpu;
243 uv_write_local_mmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index);
244 completion_status = uv_wait_completion(bau_desc, mmr_offset,
245 right_shift);
246 } while (completion_status == FLUSH_RETRY);
247 time2 = get_cycles();
248 __get_cpu_var(ptcstats).sflush += (time2 - time1);
249 if (tries > 1)
250 __get_cpu_var(ptcstats).retriesok++;
251
252 if (completion_status == FLUSH_GIVEUP) {
253 /*
254 * Cause the caller to do an IPI-style TLB shootdown on
255 * the cpu's, all of which are still in the mask.
256 */
257 __get_cpu_var(ptcstats).ptc_i++;
258 return 0;
259 }
260
261 /*
262 * Success, so clear the remote cpu's from the mask so we don't
263 * use the IPI method of shootdown on them.
264 */
265 for_each_cpu_mask(bit, *cpumaskp) {
266 blade = uv_cpu_to_blade_id(bit);
267 if (blade == this_blade)
268 continue;
269 cpu_clear(bit, *cpumaskp);
270 }
271 if (!cpus_empty(*cpumaskp))
272 return 0;
273 return 1;
274}
275
276/**
277 * uv_flush_tlb_others - globally purge translation cache of a virtual
278 * address or all TLB's
279 * @cpumaskp: mask of all cpu's in which the address is to be removed
280 * @mm: mm_struct containing virtual address range
281 * @va: virtual address to be removed (or TLB_FLUSH_ALL for all TLB's on cpu)
282 *
283 * This is the entry point for initiating any UV global TLB shootdown.
284 *
285 * Purges the translation caches of all specified processors of the given
286 * virtual address, or purges all TLB's on specified processors.
287 *
288 * The caller has derived the cpumaskp from the mm_struct and has subtracted
289 * the local cpu from the mask. This function is called only if there
290 * are bits set in the mask. (e.g. flush_tlb_page())
291 *
292 * The cpumaskp is converted into a nodemask of the nodes containing
293 * the cpus.
294 *
295 * Returns 1 if all remote flushing was done.
296 * Returns 0 if some remote flushing remains to be done.
297 */
298int uv_flush_tlb_others(cpumask_t *cpumaskp, struct mm_struct *mm,
299 unsigned long va)
300{
301 int i;
302 int bit;
303 int blade;
304 int cpu;
305 int this_blade;
306 int locals = 0;
307 struct bau_desc *bau_desc;
308
309 cpu = uv_blade_processor_id();
310 this_blade = uv_numa_blade_id();
311 bau_desc = __get_cpu_var(bau_control).descriptor_base;
312 bau_desc += UV_ITEMS_PER_DESCRIPTOR * cpu;
313
314 bau_nodes_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE);
315
316 i = 0;
317 for_each_cpu_mask(bit, *cpumaskp) {
318 blade = uv_cpu_to_blade_id(bit);
319 BUG_ON(blade > (UV_DISTRIBUTION_SIZE - 1));
320 if (blade == this_blade) {
321 locals++;
322 continue;
323 }
324 bau_node_set(blade, &bau_desc->distribution);
325 i++;
326 }
327 if (i == 0) {
328 /*
329 * no off_node flushing; return status for local node
330 */
331 if (locals)
332 return 0;
333 else
334 return 1;
335 }
336 __get_cpu_var(ptcstats).requestor++;
337 __get_cpu_var(ptcstats).ntargeted += i;
338
339 bau_desc->payload.address = va;
340 bau_desc->payload.sending_cpu = smp_processor_id();
341
342 return uv_flush_send_and_wait(cpu, this_blade, bau_desc, cpumaskp);
343}
344
345/*
346 * The BAU message interrupt comes here. (registered by set_intr_gate)
347 * See entry_64.S
348 *
349 * We received a broadcast assist message.
350 *
351 * Interrupts may have been disabled; this interrupt could represent
352 * the receipt of several messages.
353 *
354 * All cores/threads on this node get this interrupt.
355 * The last one to see it does the s/w ack.
356 * (the resource will not be freed until noninterruptable cpus see this
357 * interrupt; hardware will timeout the s/w ack and reply ERROR)
358 */
359void uv_bau_message_interrupt(struct pt_regs *regs)
360{
361 struct bau_payload_queue_entry *va_queue_first;
362 struct bau_payload_queue_entry *va_queue_last;
363 struct bau_payload_queue_entry *msg;
364 struct pt_regs *old_regs = set_irq_regs(regs);
365 cycles_t time1;
366 cycles_t time2;
367 int msg_slot;
368 int sw_ack_slot;
369 int fw;
370 int count = 0;
371 unsigned long local_pnode;
372
373 ack_APIC_irq();
374 exit_idle();
375 irq_enter();
376
377 time1 = get_cycles();
378
379 local_pnode = uv_blade_to_pnode(uv_numa_blade_id());
380
381 va_queue_first = __get_cpu_var(bau_control).va_queue_first;
382 va_queue_last = __get_cpu_var(bau_control).va_queue_last;
383
384 msg = __get_cpu_var(bau_control).bau_msg_head;
385 while (msg->sw_ack_vector) {
386 count++;
387 fw = msg->sw_ack_vector;
388 msg_slot = msg - va_queue_first;
389 sw_ack_slot = ffs(fw) - 1;
390
391 uv_bau_process_message(msg, msg_slot, sw_ack_slot);
392
393 msg++;
394 if (msg > va_queue_last)
395 msg = va_queue_first;
396 __get_cpu_var(bau_control).bau_msg_head = msg;
397 }
398 if (!count)
399 __get_cpu_var(ptcstats).nomsg++;
400 else if (count > 1)
401 __get_cpu_var(ptcstats).multmsg++;
402
403 time2 = get_cycles();
404 __get_cpu_var(ptcstats).dflush += (time2 - time1);
405
406 irq_exit();
407 set_irq_regs(old_regs);
408}
409
410static void uv_enable_timeouts(void)
411{
412 int i;
413 int blade;
414 int last_blade;
415 int pnode;
416 int cur_cpu = 0;
417 unsigned long apicid;
418
419 last_blade = -1;
420 for_each_online_node(i) {
421 blade = uv_node_to_blade_id(i);
422 if (blade == last_blade)
423 continue;
424 last_blade = blade;
425 apicid = per_cpu(x86_cpu_to_apicid, cur_cpu);
426 pnode = uv_blade_to_pnode(blade);
427 cur_cpu += uv_blade_nr_possible_cpus(i);
428 }
429}
430
431static void *uv_ptc_seq_start(struct seq_file *file, loff_t *offset)
432{
433 if (*offset < num_possible_cpus())
434 return offset;
435 return NULL;
436}
437
438static void *uv_ptc_seq_next(struct seq_file *file, void *data, loff_t *offset)
439{
440 (*offset)++;
441 if (*offset < num_possible_cpus())
442 return offset;
443 return NULL;
444}
445
446static void uv_ptc_seq_stop(struct seq_file *file, void *data)
447{
448}
449
450/*
451 * Display the statistics thru /proc
452 * data points to the cpu number
453 */
454static int uv_ptc_seq_show(struct seq_file *file, void *data)
455{
456 struct ptc_stats *stat;
457 int cpu;
458
459 cpu = *(loff_t *)data;
460
461 if (!cpu) {
462 seq_printf(file,
463 "# cpu requestor requestee one all sretry dretry ptc_i ");
464 seq_printf(file,
465 "sw_ack sflush dflush sok dnomsg dmult starget\n");
466 }
467 if (cpu < num_possible_cpus() && cpu_online(cpu)) {
468 stat = &per_cpu(ptcstats, cpu);
469 seq_printf(file, "cpu %d %ld %ld %ld %ld %ld %ld %ld ",
470 cpu, stat->requestor,
471 stat->requestee, stat->onetlb, stat->alltlb,
472 stat->s_retry, stat->d_retry, stat->ptc_i);
473 seq_printf(file, "%lx %ld %ld %ld %ld %ld %ld\n",
474 uv_read_global_mmr64(uv_blade_to_pnode
475 (uv_cpu_to_blade_id(cpu)),
476 UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE),
477 stat->sflush, stat->dflush,
478 stat->retriesok, stat->nomsg,
479 stat->multmsg, stat->ntargeted);
480 }
481
482 return 0;
483}
484
485/*
486 * 0: display meaning of the statistics
487 * >0: retry limit
488 */
489static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user,
490 size_t count, loff_t *data)
491{
492 long newmode;
493 char optstr[64];
494
495 if (count == 0 || count > sizeof(optstr))
496 return -EINVAL;
497 if (copy_from_user(optstr, user, count))
498 return -EFAULT;
499 optstr[count - 1] = '\0';
500 if (strict_strtoul(optstr, 10, &newmode) < 0) {
501 printk(KERN_DEBUG "%s is invalid\n", optstr);
502 return -EINVAL;
503 }
504
505 if (newmode == 0) {
506 printk(KERN_DEBUG "# cpu: cpu number\n");
507 printk(KERN_DEBUG
508 "requestor: times this cpu was the flush requestor\n");
509 printk(KERN_DEBUG
510 "requestee: times this cpu was requested to flush its TLBs\n");
511 printk(KERN_DEBUG
512 "one: times requested to flush a single address\n");
513 printk(KERN_DEBUG
514 "all: times requested to flush all TLB's\n");
515 printk(KERN_DEBUG
516 "sretry: number of retries of source-side timeouts\n");
517 printk(KERN_DEBUG
518 "dretry: number of retries of destination-side timeouts\n");
519 printk(KERN_DEBUG
520 "ptc_i: times UV fell through to IPI-style flushes\n");
521 printk(KERN_DEBUG
522 "sw_ack: image of UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE\n");
523 printk(KERN_DEBUG
524 "sflush_us: cycles spent in uv_flush_tlb_others()\n");
525 printk(KERN_DEBUG
526 "dflush_us: cycles spent in handling flush requests\n");
527 printk(KERN_DEBUG "sok: successes on retry\n");
528 printk(KERN_DEBUG "dnomsg: interrupts with no message\n");
529 printk(KERN_DEBUG
530 "dmult: interrupts with multiple messages\n");
531 printk(KERN_DEBUG "starget: nodes targeted\n");
532 } else {
533 uv_bau_retry_limit = newmode;
534 printk(KERN_DEBUG "timeout retry limit:%d\n",
535 uv_bau_retry_limit);
536 }
537
538 return count;
539}
540
541static const struct seq_operations uv_ptc_seq_ops = {
542 .start = uv_ptc_seq_start,
543 .next = uv_ptc_seq_next,
544 .stop = uv_ptc_seq_stop,
545 .show = uv_ptc_seq_show
546};
547
548static int uv_ptc_proc_open(struct inode *inode, struct file *file)
549{
550 return seq_open(file, &uv_ptc_seq_ops);
551}
552
553static const struct file_operations proc_uv_ptc_operations = {
554 .open = uv_ptc_proc_open,
555 .read = seq_read,
556 .write = uv_ptc_proc_write,
557 .llseek = seq_lseek,
558 .release = seq_release,
559};
560
561static int __init uv_ptc_init(void)
562{
563 struct proc_dir_entry *proc_uv_ptc;
564
565 if (!is_uv_system())
566 return 0;
567
568 if (!proc_mkdir("sgi_uv", NULL))
569 return -EINVAL;
570
571 proc_uv_ptc = create_proc_entry(UV_PTC_BASENAME, 0444, NULL);
572 if (!proc_uv_ptc) {
573 printk(KERN_ERR "unable to create %s proc entry\n",
574 UV_PTC_BASENAME);
575 remove_proc_entry("sgi_uv", NULL);
576 return -EINVAL;
577 }
578 proc_uv_ptc->proc_fops = &proc_uv_ptc_operations;
579 return 0;
580}
581
582/*
583 * begin the initialization of the per-blade control structures
584 */
585static struct bau_control * __init uv_table_bases_init(int blade, int node)
586{
587 int i;
588 int *ip;
589 struct bau_msg_status *msp;
590 struct bau_control *bau_tabp;
591
592 bau_tabp =
593 kmalloc_node(sizeof(struct bau_control), GFP_KERNEL, node);
594 BUG_ON(!bau_tabp);
595
596 bau_tabp->msg_statuses =
597 kmalloc_node(sizeof(struct bau_msg_status) *
598 DEST_Q_SIZE, GFP_KERNEL, node);
599 BUG_ON(!bau_tabp->msg_statuses);
600
601 for (i = 0, msp = bau_tabp->msg_statuses; i < DEST_Q_SIZE; i++, msp++)
602 bau_cpubits_clear(&msp->seen_by, (int)
603 uv_blade_nr_possible_cpus(blade));
604
605 bau_tabp->watching =
606 kmalloc_node(sizeof(int) * DEST_NUM_RESOURCES, GFP_KERNEL, node);
607 BUG_ON(!bau_tabp->watching);
608
609 for (i = 0, ip = bau_tabp->watching; i < DEST_Q_SIZE; i++, ip++)
610 *ip = 0;
611
612 uv_bau_table_bases[blade] = bau_tabp;
613
614 return bau_tabp;
615}
616
617/*
618 * finish the initialization of the per-blade control structures
619 */
620static void __init
621uv_table_bases_finish(int blade, int node, int cur_cpu,
622 struct bau_control *bau_tablesp,
623 struct bau_desc *adp)
624{
625 struct bau_control *bcp;
626 int i;
627
628 for (i = cur_cpu; i < cur_cpu + uv_blade_nr_possible_cpus(blade); i++) {
629 bcp = (struct bau_control *)&per_cpu(bau_control, i);
630
631 bcp->bau_msg_head = bau_tablesp->va_queue_first;
632 bcp->va_queue_first = bau_tablesp->va_queue_first;
633 bcp->va_queue_last = bau_tablesp->va_queue_last;
634 bcp->watching = bau_tablesp->watching;
635 bcp->msg_statuses = bau_tablesp->msg_statuses;
636 bcp->descriptor_base = adp;
637 }
638}
639
640/*
641 * initialize the sending side's sending buffers
642 */
643static struct bau_desc * __init
644uv_activation_descriptor_init(int node, int pnode)
645{
646 int i;
647 unsigned long pa;
648 unsigned long m;
649 unsigned long n;
650 unsigned long mmr_image;
651 struct bau_desc *adp;
652 struct bau_desc *ad2;
653
654 adp = (struct bau_desc *)
655 kmalloc_node(16384, GFP_KERNEL, node);
656 BUG_ON(!adp);
657
658 pa = __pa((unsigned long)adp);
659 n = pa >> uv_nshift;
660 m = pa & uv_mmask;
661
662 mmr_image = uv_read_global_mmr64(pnode, UVH_LB_BAU_SB_DESCRIPTOR_BASE);
663 if (mmr_image) {
664 uv_write_global_mmr64(pnode, (unsigned long)
665 UVH_LB_BAU_SB_DESCRIPTOR_BASE,
666 (n << UV_DESC_BASE_PNODE_SHIFT | m));
667 }
668
669 for (i = 0, ad2 = adp; i < UV_ACTIVATION_DESCRIPTOR_SIZE; i++, ad2++) {
670 memset(ad2, 0, sizeof(struct bau_desc));
671 ad2->header.sw_ack_flag = 1;
672 ad2->header.base_dest_nodeid =
673 uv_blade_to_pnode(uv_cpu_to_blade_id(0));
674 ad2->header.command = UV_NET_ENDPOINT_INTD;
675 ad2->header.int_both = 1;
676 /*
677 * all others need to be set to zero:
678 * fairness chaining multilevel count replied_to
679 */
680 }
681 return adp;
682}
683
684/*
685 * initialize the destination side's receiving buffers
686 */
687static struct bau_payload_queue_entry * __init
688uv_payload_queue_init(int node, int pnode, struct bau_control *bau_tablesp)
689{
690 struct bau_payload_queue_entry *pqp;
691 char *cp;
692
693 pqp = (struct bau_payload_queue_entry *) kmalloc_node(
694 (DEST_Q_SIZE + 1) * sizeof(struct bau_payload_queue_entry),
695 GFP_KERNEL, node);
696 BUG_ON(!pqp);
697
698 cp = (char *)pqp + 31;
699 pqp = (struct bau_payload_queue_entry *)(((unsigned long)cp >> 5) << 5);
700 bau_tablesp->va_queue_first = pqp;
701 uv_write_global_mmr64(pnode,
702 UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST,
703 ((unsigned long)pnode <<
704 UV_PAYLOADQ_PNODE_SHIFT) |
705 uv_physnodeaddr(pqp));
706 uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL,
707 uv_physnodeaddr(pqp));
708 bau_tablesp->va_queue_last = pqp + (DEST_Q_SIZE - 1);
709 uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST,
710 (unsigned long)
711 uv_physnodeaddr(bau_tablesp->va_queue_last));
712 memset(pqp, 0, sizeof(struct bau_payload_queue_entry) * DEST_Q_SIZE);
713
714 return pqp;
715}
716
717/*
718 * Initialization of each UV blade's structures
719 */
720static int __init uv_init_blade(int blade, int node, int cur_cpu)
721{
722 int pnode;
723 unsigned long pa;
724 unsigned long apicid;
725 struct bau_desc *adp;
726 struct bau_payload_queue_entry *pqp;
727 struct bau_control *bau_tablesp;
728
729 bau_tablesp = uv_table_bases_init(blade, node);
730 pnode = uv_blade_to_pnode(blade);
731 adp = uv_activation_descriptor_init(node, pnode);
732 pqp = uv_payload_queue_init(node, pnode, bau_tablesp);
733 uv_table_bases_finish(blade, node, cur_cpu, bau_tablesp, adp);
734 /*
735 * the below initialization can't be in firmware because the
736 * messaging IRQ will be determined by the OS
737 */
738 apicid = per_cpu(x86_cpu_to_apicid, cur_cpu);
739 pa = uv_read_global_mmr64(pnode, UVH_BAU_DATA_CONFIG);
740 if ((pa & 0xff) != UV_BAU_MESSAGE) {
741 uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG,
742 ((apicid << 32) | UV_BAU_MESSAGE));
743 }
744 return 0;
745}
746
747/*
748 * Initialization of BAU-related structures
749 */
750static int __init uv_bau_init(void)
751{
752 int blade;
753 int node;
754 int nblades;
755 int last_blade;
756 int cur_cpu = 0;
757
758 if (!is_uv_system())
759 return 0;
760
761 uv_bau_retry_limit = 1;
762 uv_nshift = uv_hub_info->n_val;
763 uv_mmask = (1UL << uv_hub_info->n_val) - 1;
764 nblades = 0;
765 last_blade = -1;
766 for_each_online_node(node) {
767 blade = uv_node_to_blade_id(node);
768 if (blade == last_blade)
769 continue;
770 last_blade = blade;
771 nblades++;
772 }
773 uv_bau_table_bases = (struct bau_control **)
774 kmalloc(nblades * sizeof(struct bau_control *), GFP_KERNEL);
775 BUG_ON(!uv_bau_table_bases);
776
777 last_blade = -1;
778 for_each_online_node(node) {
779 blade = uv_node_to_blade_id(node);
780 if (blade == last_blade)
781 continue;
782 last_blade = blade;
783 uv_init_blade(blade, node, cur_cpu);
784 cur_cpu += uv_blade_nr_possible_cpus(blade);
785 }
786 set_intr_gate(UV_BAU_MESSAGE, uv_bau_message_intr1);
787 uv_enable_timeouts();
788
789 return 0;
790}
791__initcall(uv_bau_init);
792__initcall(uv_ptc_init);
diff --git a/arch/x86/kernel/trampoline.c b/arch/x86/kernel/trampoline.c
index abbf199adebb..1106fac6024d 100644
--- a/arch/x86/kernel/trampoline.c
+++ b/arch/x86/kernel/trampoline.c
@@ -2,7 +2,7 @@
2 2
3#include <asm/trampoline.h> 3#include <asm/trampoline.h>
4 4
5/* ready for x86_64, no harm for x86, since it will overwrite after alloc */ 5/* ready for x86_64 and x86 */
6unsigned char *trampoline_base = __va(TRAMPOLINE_BASE); 6unsigned char *trampoline_base = __va(TRAMPOLINE_BASE);
7 7
8/* 8/*
diff --git a/arch/x86/kernel/traps_32.c b/arch/x86/kernel/traps_32.c
index 08d752de4eee..03df8e45e5a1 100644
--- a/arch/x86/kernel/traps_32.c
+++ b/arch/x86/kernel/traps_32.c
@@ -1,5 +1,6 @@
1/* 1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds 2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
3 * 4 *
4 * Pentium III FXSR, SSE support 5 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000 6 * Gareth Hughes <gareth@valinux.com>, May 2000
@@ -57,11 +58,10 @@
57#include <asm/nmi.h> 58#include <asm/nmi.h>
58#include <asm/smp.h> 59#include <asm/smp.h>
59#include <asm/io.h> 60#include <asm/io.h>
61#include <asm/traps.h>
60 62
61#include "mach_traps.h" 63#include "mach_traps.h"
62 64
63int panic_on_unrecovered_nmi;
64
65DECLARE_BITMAP(used_vectors, NR_VECTORS); 65DECLARE_BITMAP(used_vectors, NR_VECTORS);
66EXPORT_SYMBOL_GPL(used_vectors); 66EXPORT_SYMBOL_GPL(used_vectors);
67 67
@@ -78,39 +78,22 @@ char ignore_fpu_irq;
78gate_desc idt_table[256] 78gate_desc idt_table[256]
79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, }; 79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
80 80
81asmlinkage void divide_error(void); 81int panic_on_unrecovered_nmi;
82asmlinkage void debug(void);
83asmlinkage void nmi(void);
84asmlinkage void int3(void);
85asmlinkage void overflow(void);
86asmlinkage void bounds(void);
87asmlinkage void invalid_op(void);
88asmlinkage void device_not_available(void);
89asmlinkage void coprocessor_segment_overrun(void);
90asmlinkage void invalid_TSS(void);
91asmlinkage void segment_not_present(void);
92asmlinkage void stack_segment(void);
93asmlinkage void general_protection(void);
94asmlinkage void page_fault(void);
95asmlinkage void coprocessor_error(void);
96asmlinkage void simd_coprocessor_error(void);
97asmlinkage void alignment_check(void);
98asmlinkage void spurious_interrupt_bug(void);
99asmlinkage void machine_check(void);
100
101int kstack_depth_to_print = 24; 82int kstack_depth_to_print = 24;
102static unsigned int code_bytes = 64; 83static unsigned int code_bytes = 64;
84static int ignore_nmis;
85static int die_counter;
103 86
104void printk_address(unsigned long address, int reliable) 87void printk_address(unsigned long address, int reliable)
105{ 88{
106#ifdef CONFIG_KALLSYMS 89#ifdef CONFIG_KALLSYMS
107 char namebuf[KSYM_NAME_LEN];
108 unsigned long offset = 0; 90 unsigned long offset = 0;
109 unsigned long symsize; 91 unsigned long symsize;
110 const char *symname; 92 const char *symname;
111 char reliab[4] = "";
112 char *delim = ":";
113 char *modname; 93 char *modname;
94 char *delim = ":";
95 char namebuf[KSYM_NAME_LEN];
96 char reliab[4] = "";
114 97
115 symname = kallsyms_lookup(address, &symsize, &offset, 98 symname = kallsyms_lookup(address, &symsize, &offset,
116 &modname, namebuf); 99 &modname, namebuf);
@@ -130,22 +113,23 @@ void printk_address(unsigned long address, int reliable)
130#endif 113#endif
131} 114}
132 115
133static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size) 116static inline int valid_stack_ptr(struct thread_info *tinfo,
117 void *p, unsigned int size)
134{ 118{
135 return p > (void *)tinfo && 119 void *t = tinfo;
136 p <= (void *)tinfo + THREAD_SIZE - size; 120 return p > t && p <= t + THREAD_SIZE - size;
137} 121}
138 122
139/* The form of the top of the frame on the stack */ 123/* The form of the top of the frame on the stack */
140struct stack_frame { 124struct stack_frame {
141 struct stack_frame *next_frame; 125 struct stack_frame *next_frame;
142 unsigned long return_address; 126 unsigned long return_address;
143}; 127};
144 128
145static inline unsigned long 129static inline unsigned long
146print_context_stack(struct thread_info *tinfo, 130print_context_stack(struct thread_info *tinfo,
147 unsigned long *stack, unsigned long bp, 131 unsigned long *stack, unsigned long bp,
148 const struct stacktrace_ops *ops, void *data) 132 const struct stacktrace_ops *ops, void *data)
149{ 133{
150 struct stack_frame *frame = (struct stack_frame *)bp; 134 struct stack_frame *frame = (struct stack_frame *)bp;
151 135
@@ -167,8 +151,6 @@ print_context_stack(struct thread_info *tinfo,
167 return bp; 151 return bp;
168} 152}
169 153
170#define MSG(msg) ops->warning(data, msg)
171
172void dump_trace(struct task_struct *task, struct pt_regs *regs, 154void dump_trace(struct task_struct *task, struct pt_regs *regs,
173 unsigned long *stack, unsigned long bp, 155 unsigned long *stack, unsigned long bp,
174 const struct stacktrace_ops *ops, void *data) 156 const struct stacktrace_ops *ops, void *data)
@@ -178,7 +160,6 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
178 160
179 if (!stack) { 161 if (!stack) {
180 unsigned long dummy; 162 unsigned long dummy;
181
182 stack = &dummy; 163 stack = &dummy;
183 if (task != current) 164 if (task != current)
184 stack = (unsigned long *)task->thread.sp; 165 stack = (unsigned long *)task->thread.sp;
@@ -196,7 +177,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
196 } 177 }
197#endif 178#endif
198 179
199 while (1) { 180 for (;;) {
200 struct thread_info *context; 181 struct thread_info *context;
201 182
202 context = (struct thread_info *) 183 context = (struct thread_info *)
@@ -248,15 +229,15 @@ static void print_trace_address(void *data, unsigned long addr, int reliable)
248} 229}
249 230
250static const struct stacktrace_ops print_trace_ops = { 231static const struct stacktrace_ops print_trace_ops = {
251 .warning = print_trace_warning, 232 .warning = print_trace_warning,
252 .warning_symbol = print_trace_warning_symbol, 233 .warning_symbol = print_trace_warning_symbol,
253 .stack = print_trace_stack, 234 .stack = print_trace_stack,
254 .address = print_trace_address, 235 .address = print_trace_address,
255}; 236};
256 237
257static void 238static void
258show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs, 239show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
259 unsigned long *stack, unsigned long bp, char *log_lvl) 240 unsigned long *stack, unsigned long bp, char *log_lvl)
260{ 241{
261 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl); 242 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
262 printk("%s =======================\n", log_lvl); 243 printk("%s =======================\n", log_lvl);
@@ -351,15 +332,14 @@ void show_registers(struct pt_regs *regs)
351 printk(KERN_EMERG "Code: "); 332 printk(KERN_EMERG "Code: ");
352 333
353 ip = (u8 *)regs->ip - code_prologue; 334 ip = (u8 *)regs->ip - code_prologue;
354 if (ip < (u8 *)PAGE_OFFSET || 335 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
355 probe_kernel_address(ip, c)) {
356 /* try starting at EIP */ 336 /* try starting at EIP */
357 ip = (u8 *)regs->ip; 337 ip = (u8 *)regs->ip;
358 code_len = code_len - code_prologue + 1; 338 code_len = code_len - code_prologue + 1;
359 } 339 }
360 for (i = 0; i < code_len; i++, ip++) { 340 for (i = 0; i < code_len; i++, ip++) {
361 if (ip < (u8 *)PAGE_OFFSET || 341 if (ip < (u8 *)PAGE_OFFSET ||
362 probe_kernel_address(ip, c)) { 342 probe_kernel_address(ip, c)) {
363 printk(" Bad EIP value."); 343 printk(" Bad EIP value.");
364 break; 344 break;
365 } 345 }
@@ -384,7 +364,53 @@ int is_valid_bugaddr(unsigned long ip)
384 return ud2 == 0x0b0f; 364 return ud2 == 0x0b0f;
385} 365}
386 366
387static int die_counter; 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}
388 414
389int __kprobes __die(const char *str, struct pt_regs *regs, long err) 415int __kprobes __die(const char *str, struct pt_regs *regs, long err)
390{ 416{
@@ -402,26 +428,22 @@ int __kprobes __die(const char *str, struct pt_regs *regs, long err)
402 printk("DEBUG_PAGEALLOC"); 428 printk("DEBUG_PAGEALLOC");
403#endif 429#endif
404 printk("\n"); 430 printk("\n");
405
406 if (notify_die(DIE_OOPS, str, regs, err, 431 if (notify_die(DIE_OOPS, str, regs, err,
407 current->thread.trap_no, SIGSEGV) != NOTIFY_STOP) { 432 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
408 433 return 1;
409 show_registers(regs);
410 /* Executive summary in case the oops scrolled away */
411 sp = (unsigned long) (&regs->sp);
412 savesegment(ss, ss);
413 if (user_mode(regs)) {
414 sp = regs->sp;
415 ss = regs->ss & 0xffff;
416 }
417 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
418 print_symbol("%s", regs->ip);
419 printk(" SS:ESP %04x:%08lx\n", ss, sp);
420 434
421 return 0; 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;
422 } 442 }
423 443 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
424 return 1; 444 print_symbol("%s", regs->ip);
445 printk(" SS:ESP %04x:%08lx\n", ss, sp);
446 return 0;
425} 447}
426 448
427/* 449/*
@@ -430,31 +452,9 @@ int __kprobes __die(const char *str, struct pt_regs *regs, long err)
430 */ 452 */
431void die(const char *str, struct pt_regs *regs, long err) 453void die(const char *str, struct pt_regs *regs, long err)
432{ 454{
433 static struct { 455 unsigned long flags = oops_begin();
434 raw_spinlock_t lock;
435 u32 lock_owner;
436 int lock_owner_depth;
437 } die = {
438 .lock = __RAW_SPIN_LOCK_UNLOCKED,
439 .lock_owner = -1,
440 .lock_owner_depth = 0
441 };
442 unsigned long flags;
443
444 oops_enter();
445
446 if (die.lock_owner != raw_smp_processor_id()) {
447 console_verbose();
448 raw_local_irq_save(flags);
449 __raw_spin_lock(&die.lock);
450 die.lock_owner = smp_processor_id();
451 die.lock_owner_depth = 0;
452 bust_spinlocks(1);
453 } else {
454 raw_local_irq_save(flags);
455 }
456 456
457 if (++die.lock_owner_depth < 3) { 457 if (die_nest_count < 3) {
458 report_bug(regs->ip, regs); 458 report_bug(regs->ip, regs);
459 459
460 if (__die(str, regs, err)) 460 if (__die(str, regs, err))
@@ -463,26 +463,7 @@ void die(const char *str, struct pt_regs *regs, long err)
463 printk(KERN_EMERG "Recursive die() failure, output suppressed\n"); 463 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
464 } 464 }
465 465
466 bust_spinlocks(0); 466 oops_end(flags, regs, SIGSEGV);
467 die.lock_owner = -1;
468 add_taint(TAINT_DIE);
469 __raw_spin_unlock(&die.lock);
470 raw_local_irq_restore(flags);
471
472 if (!regs)
473 return;
474
475 if (kexec_should_crash(current))
476 crash_kexec(regs);
477
478 if (in_interrupt())
479 panic("Fatal exception in interrupt");
480
481 if (panic_on_oops)
482 panic("Fatal exception");
483
484 oops_exit();
485 do_exit(SIGSEGV);
486} 467}
487 468
488static inline void 469static inline void
@@ -546,7 +527,7 @@ void do_##name(struct pt_regs *regs, long error_code) \
546{ \ 527{ \
547 trace_hardirqs_fixup(); \ 528 trace_hardirqs_fixup(); \
548 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 529 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
549 == NOTIFY_STOP) \ 530 == NOTIFY_STOP) \
550 return; \ 531 return; \
551 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \ 532 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
552} 533}
@@ -562,7 +543,7 @@ void do_##name(struct pt_regs *regs, long error_code) \
562 info.si_code = sicode; \ 543 info.si_code = sicode; \
563 info.si_addr = (void __user *)siaddr; \ 544 info.si_addr = (void __user *)siaddr; \
564 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 545 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
565 == NOTIFY_STOP) \ 546 == NOTIFY_STOP) \
566 return; \ 547 return; \
567 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \ 548 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
568} 549}
@@ -571,7 +552,7 @@ void do_##name(struct pt_regs *regs, long error_code) \
571void do_##name(struct pt_regs *regs, long error_code) \ 552void do_##name(struct pt_regs *regs, long error_code) \
572{ \ 553{ \
573 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 554 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
574 == NOTIFY_STOP) \ 555 == NOTIFY_STOP) \
575 return; \ 556 return; \
576 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \ 557 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
577} 558}
@@ -586,27 +567,29 @@ void do_##name(struct pt_regs *regs, long error_code) \
586 info.si_addr = (void __user *)siaddr; \ 567 info.si_addr = (void __user *)siaddr; \
587 trace_hardirqs_fixup(); \ 568 trace_hardirqs_fixup(); \
588 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 569 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
589 == NOTIFY_STOP) \ 570 == NOTIFY_STOP) \
590 return; \ 571 return; \
591 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \ 572 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
592} 573}
593 574
594DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip) 575DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
595#ifndef CONFIG_KPROBES 576#ifndef CONFIG_KPROBES
596DO_VM86_ERROR(3, SIGTRAP, "int3", int3) 577DO_VM86_ERROR(3, SIGTRAP, "int3", int3)
597#endif 578#endif
598DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow) 579DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow)
599DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds) 580DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds)
600DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0) 581DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
601DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) 582DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
602DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) 583DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
603DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) 584DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
604DO_ERROR(12, SIGBUS, "stack segment", stack_segment) 585DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
605DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0) 586DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
606DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1) 587DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1)
607 588
608void __kprobes do_general_protection(struct pt_regs *regs, long error_code) 589void __kprobes
590do_general_protection(struct pt_regs *regs, long error_code)
609{ 591{
592 struct task_struct *tsk;
610 struct thread_struct *thread; 593 struct thread_struct *thread;
611 struct tss_struct *tss; 594 struct tss_struct *tss;
612 int cpu; 595 int cpu;
@@ -647,23 +630,24 @@ void __kprobes do_general_protection(struct pt_regs *regs, long error_code)
647 if (regs->flags & X86_VM_MASK) 630 if (regs->flags & X86_VM_MASK)
648 goto gp_in_vm86; 631 goto gp_in_vm86;
649 632
633 tsk = current;
650 if (!user_mode(regs)) 634 if (!user_mode(regs))
651 goto gp_in_kernel; 635 goto gp_in_kernel;
652 636
653 current->thread.error_code = error_code; 637 tsk->thread.error_code = error_code;
654 current->thread.trap_no = 13; 638 tsk->thread.trap_no = 13;
655 639
656 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) && 640 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
657 printk_ratelimit()) { 641 printk_ratelimit()) {
658 printk(KERN_INFO 642 printk(KERN_INFO
659 "%s[%d] general protection ip:%lx sp:%lx error:%lx", 643 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
660 current->comm, task_pid_nr(current), 644 tsk->comm, task_pid_nr(tsk),
661 regs->ip, regs->sp, error_code); 645 regs->ip, regs->sp, error_code);
662 print_vma_addr(" in ", regs->ip); 646 print_vma_addr(" in ", regs->ip);
663 printk("\n"); 647 printk("\n");
664 } 648 }
665 649
666 force_sig(SIGSEGV, current); 650 force_sig(SIGSEGV, tsk);
667 return; 651 return;
668 652
669gp_in_vm86: 653gp_in_vm86:
@@ -672,14 +656,15 @@ gp_in_vm86:
672 return; 656 return;
673 657
674gp_in_kernel: 658gp_in_kernel:
675 if (!fixup_exception(regs)) { 659 if (fixup_exception(regs))
676 current->thread.error_code = error_code; 660 return;
677 current->thread.trap_no = 13; 661
678 if (notify_die(DIE_GPF, "general protection fault", regs, 662 tsk->thread.error_code = error_code;
663 tsk->thread.trap_no = 13;
664 if (notify_die(DIE_GPF, "general protection fault", regs,
679 error_code, 13, SIGSEGV) == NOTIFY_STOP) 665 error_code, 13, SIGSEGV) == NOTIFY_STOP)
680 return; 666 return;
681 die("general protection fault", regs, error_code); 667 die("general protection fault", regs, error_code);
682 }
683} 668}
684 669
685static notrace __kprobes void 670static notrace __kprobes void
@@ -756,9 +741,9 @@ unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
756 741
757static DEFINE_SPINLOCK(nmi_print_lock); 742static DEFINE_SPINLOCK(nmi_print_lock);
758 743
759void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg) 744void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
760{ 745{
761 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) == NOTIFY_STOP) 746 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
762 return; 747 return;
763 748
764 spin_lock(&nmi_print_lock); 749 spin_lock(&nmi_print_lock);
@@ -767,10 +752,12 @@ void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg)
767 * to get a message out: 752 * to get a message out:
768 */ 753 */
769 bust_spinlocks(1); 754 bust_spinlocks(1);
770 printk(KERN_EMERG "%s", msg); 755 printk(KERN_EMERG "%s", str);
771 printk(" on CPU%d, ip %08lx, registers:\n", 756 printk(" on CPU%d, ip %08lx, registers:\n",
772 smp_processor_id(), regs->ip); 757 smp_processor_id(), regs->ip);
773 show_registers(regs); 758 show_registers(regs);
759 if (do_panic)
760 panic("Non maskable interrupt");
774 console_silent(); 761 console_silent();
775 spin_unlock(&nmi_print_lock); 762 spin_unlock(&nmi_print_lock);
776 bust_spinlocks(0); 763 bust_spinlocks(0);
@@ -790,14 +777,17 @@ void notrace __kprobes die_nmi(struct pt_regs *regs, const char *msg)
790static notrace __kprobes void default_do_nmi(struct pt_regs *regs) 777static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
791{ 778{
792 unsigned char reason = 0; 779 unsigned char reason = 0;
780 int cpu;
793 781
794 /* Only the BSP gets external NMIs from the system: */ 782 cpu = smp_processor_id();
795 if (!smp_processor_id()) 783
784 /* Only the BSP gets external NMIs from the system. */
785 if (!cpu)
796 reason = get_nmi_reason(); 786 reason = get_nmi_reason();
797 787
798 if (!(reason & 0xc0)) { 788 if (!(reason & 0xc0)) {
799 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) 789 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
800 == NOTIFY_STOP) 790 == NOTIFY_STOP)
801 return; 791 return;
802#ifdef CONFIG_X86_LOCAL_APIC 792#ifdef CONFIG_X86_LOCAL_APIC
803 /* 793 /*
@@ -806,7 +796,7 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
806 */ 796 */
807 if (nmi_watchdog_tick(regs, reason)) 797 if (nmi_watchdog_tick(regs, reason))
808 return; 798 return;
809 if (!do_nmi_callback(regs, smp_processor_id())) 799 if (!do_nmi_callback(regs, cpu))
810 unknown_nmi_error(reason, regs); 800 unknown_nmi_error(reason, regs);
811#else 801#else
812 unknown_nmi_error(reason, regs); 802 unknown_nmi_error(reason, regs);
@@ -816,6 +806,8 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
816 } 806 }
817 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) 807 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
818 return; 808 return;
809
810 /* AK: following checks seem to be broken on modern chipsets. FIXME */
819 if (reason & 0x80) 811 if (reason & 0x80)
820 mem_parity_error(reason, regs); 812 mem_parity_error(reason, regs);
821 if (reason & 0x40) 813 if (reason & 0x40)
@@ -827,8 +819,6 @@ static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
827 reassert_nmi(); 819 reassert_nmi();
828} 820}
829 821
830static int ignore_nmis;
831
832notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code) 822notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code)
833{ 823{
834 int cpu; 824 int cpu;
@@ -913,7 +903,7 @@ void __kprobes do_debug(struct pt_regs *regs, long error_code)
913 tsk->thread.debugctlmsr = 0; 903 tsk->thread.debugctlmsr = 0;
914 904
915 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, 905 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
916 SIGTRAP) == NOTIFY_STOP) 906 SIGTRAP) == NOTIFY_STOP)
917 return; 907 return;
918 /* It's safe to allow irq's after DR6 has been saved */ 908 /* It's safe to allow irq's after DR6 has been saved */
919 if (regs->flags & X86_EFLAGS_IF) 909 if (regs->flags & X86_EFLAGS_IF)
@@ -974,9 +964,8 @@ clear_TF_reenable:
974void math_error(void __user *ip) 964void math_error(void __user *ip)
975{ 965{
976 struct task_struct *task; 966 struct task_struct *task;
977 unsigned short cwd;
978 unsigned short swd;
979 siginfo_t info; 967 siginfo_t info;
968 unsigned short cwd, swd;
980 969
981 /* 970 /*
982 * Save the info for the exception handler and clear the error. 971 * Save the info for the exception handler and clear the error.
@@ -995,7 +984,7 @@ void math_error(void __user *ip)
995 * C1 reg you need in case of a stack fault, 0x040 is the stack 984 * C1 reg you need in case of a stack fault, 0x040 is the stack
996 * fault bit. We should only be taking one exception at a time, 985 * fault bit. We should only be taking one exception at a time,
997 * so if this combination doesn't produce any single exception, 986 * so if this combination doesn't produce any single exception,
998 * then we have a bad program that isn't syncronizing its FPU usage 987 * then we have a bad program that isn't synchronizing its FPU usage
999 * and it will suffer the consequences since we won't be able to 988 * and it will suffer the consequences since we won't be able to
1000 * fully reproduce the context of the exception 989 * fully reproduce the context of the exception
1001 */ 990 */
@@ -1004,7 +993,7 @@ void math_error(void __user *ip)
1004 switch (swd & ~cwd & 0x3f) { 993 switch (swd & ~cwd & 0x3f) {
1005 case 0x000: /* No unmasked exception */ 994 case 0x000: /* No unmasked exception */
1006 return; 995 return;
1007 default: /* Multiple exceptions */ 996 default: /* Multiple exceptions */
1008 break; 997 break;
1009 case 0x001: /* Invalid Op */ 998 case 0x001: /* Invalid Op */
1010 /* 999 /*
@@ -1040,8 +1029,8 @@ void do_coprocessor_error(struct pt_regs *regs, long error_code)
1040static void simd_math_error(void __user *ip) 1029static void simd_math_error(void __user *ip)
1041{ 1030{
1042 struct task_struct *task; 1031 struct task_struct *task;
1043 unsigned short mxcsr;
1044 siginfo_t info; 1032 siginfo_t info;
1033 unsigned short mxcsr;
1045 1034
1046 /* 1035 /*
1047 * Save the info for the exception handler and clear the error. 1036 * Save the info for the exception handler and clear the error.
@@ -1117,7 +1106,7 @@ void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
1117 1106
1118unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp) 1107unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
1119{ 1108{
1120 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt; 1109 struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
1121 unsigned long base = (kesp - uesp) & -THREAD_SIZE; 1110 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1122 unsigned long new_kesp = kesp - base; 1111 unsigned long new_kesp = kesp - base;
1123 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT; 1112 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
@@ -1196,19 +1185,16 @@ void __init trap_init(void)
1196 early_iounmap(p, 4); 1185 early_iounmap(p, 4);
1197#endif 1186#endif
1198 1187
1199#ifdef CONFIG_X86_LOCAL_APIC 1188 set_trap_gate(0, &divide_error);
1200 init_apic_mappings(); 1189 set_intr_gate(1, &debug);
1201#endif 1190 set_intr_gate(2, &nmi);
1202 set_trap_gate(0, &divide_error); 1191 set_system_intr_gate(3, &int3); /* int3 can be called from all */
1203 set_intr_gate(1, &debug); 1192 set_system_gate(4, &overflow); /* int4 can be called from all */
1204 set_intr_gate(2, &nmi); 1193 set_trap_gate(5, &bounds);
1205 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */ 1194 set_trap_gate(6, &invalid_op);
1206 set_system_gate(4, &overflow); 1195 set_trap_gate(7, &device_not_available);
1207 set_trap_gate(5, &bounds); 1196 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1208 set_trap_gate(6, &invalid_op); 1197 set_trap_gate(9, &coprocessor_segment_overrun);
1209 set_trap_gate(7, &device_not_available);
1210 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
1211 set_trap_gate(9, &coprocessor_segment_overrun);
1212 set_trap_gate(10, &invalid_TSS); 1198 set_trap_gate(10, &invalid_TSS);
1213 set_trap_gate(11, &segment_not_present); 1199 set_trap_gate(11, &segment_not_present);
1214 set_trap_gate(12, &stack_segment); 1200 set_trap_gate(12, &stack_segment);
diff --git a/arch/x86/kernel/traps_64.c b/arch/x86/kernel/traps_64.c
index adff76ea97c4..3f18d73f420c 100644
--- a/arch/x86/kernel/traps_64.c
+++ b/arch/x86/kernel/traps_64.c
@@ -10,73 +10,56 @@
10 * 'Traps.c' handles hardware traps and faults after we have saved some 10 * 'Traps.c' handles hardware traps and faults after we have saved some
11 * state in 'entry.S'. 11 * state in 'entry.S'.
12 */ 12 */
13#include <linux/sched.h> 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>
14#include <linux/kernel.h> 21#include <linux/kernel.h>
22#include <linux/module.h>
23#include <linux/ptrace.h>
15#include <linux/string.h> 24#include <linux/string.h>
25#include <linux/unwind.h>
26#include <linux/delay.h>
16#include <linux/errno.h> 27#include <linux/errno.h>
17#include <linux/ptrace.h> 28#include <linux/kexec.h>
29#include <linux/sched.h>
18#include <linux/timer.h> 30#include <linux/timer.h>
19#include <linux/mm.h>
20#include <linux/init.h> 31#include <linux/init.h>
21#include <linux/delay.h>
22#include <linux/spinlock.h>
23#include <linux/interrupt.h>
24#include <linux/kallsyms.h>
25#include <linux/module.h>
26#include <linux/moduleparam.h>
27#include <linux/nmi.h>
28#include <linux/kprobes.h>
29#include <linux/kexec.h>
30#include <linux/unwind.h>
31#include <linux/uaccess.h>
32#include <linux/bug.h> 32#include <linux/bug.h>
33#include <linux/kdebug.h> 33#include <linux/nmi.h>
34#include <linux/utsname.h> 34#include <linux/mm.h>
35
36#include <mach_traps.h>
37 35
38#if defined(CONFIG_EDAC) 36#if defined(CONFIG_EDAC)
39#include <linux/edac.h> 37#include <linux/edac.h>
40#endif 38#endif
41 39
42#include <asm/system.h> 40#include <asm/stacktrace.h>
43#include <asm/io.h> 41#include <asm/processor.h>
44#include <asm/atomic.h>
45#include <asm/debugreg.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> 46#include <asm/desc.h>
47#include <asm/i387.h> 47#include <asm/i387.h>
48#include <asm/processor.h> 48#include <asm/nmi.h>
49#include <asm/unwind.h>
50#include <asm/smp.h> 49#include <asm/smp.h>
50#include <asm/io.h>
51#include <asm/pgalloc.h> 51#include <asm/pgalloc.h>
52#include <asm/pda.h>
53#include <asm/proto.h> 52#include <asm/proto.h>
54#include <asm/nmi.h> 53#include <asm/pda.h>
55#include <asm/stacktrace.h> 54#include <asm/traps.h>
56 55
57asmlinkage void divide_error(void); 56#include <mach_traps.h>
58asmlinkage void debug(void);
59asmlinkage void nmi(void);
60asmlinkage void int3(void);
61asmlinkage void overflow(void);
62asmlinkage void bounds(void);
63asmlinkage void invalid_op(void);
64asmlinkage void device_not_available(void);
65asmlinkage void double_fault(void);
66asmlinkage void coprocessor_segment_overrun(void);
67asmlinkage void invalid_TSS(void);
68asmlinkage void segment_not_present(void);
69asmlinkage void stack_segment(void);
70asmlinkage void general_protection(void);
71asmlinkage void page_fault(void);
72asmlinkage void coprocessor_error(void);
73asmlinkage void simd_coprocessor_error(void);
74asmlinkage void reserved(void);
75asmlinkage void alignment_check(void);
76asmlinkage void machine_check(void);
77asmlinkage void spurious_interrupt_bug(void);
78 57
58int panic_on_unrecovered_nmi;
59int kstack_depth_to_print = 12;
79static unsigned int code_bytes = 64; 60static unsigned int code_bytes = 64;
61static int ignore_nmis;
62static int die_counter;
80 63
81static inline void conditional_sti(struct pt_regs *regs) 64static inline void conditional_sti(struct pt_regs *regs)
82{ 65{
@@ -100,34 +83,9 @@ static inline void preempt_conditional_cli(struct pt_regs *regs)
100 dec_preempt_count(); 83 dec_preempt_count();
101} 84}
102 85
103int kstack_depth_to_print = 12;
104
105void printk_address(unsigned long address, int reliable) 86void printk_address(unsigned long address, int reliable)
106{ 87{
107#ifdef CONFIG_KALLSYMS 88 printk(" [<%016lx>] %s%pS\n", address, reliable ? "": "? ", (void *) address);
108 unsigned long offset = 0, symsize;
109 const char *symname;
110 char *modname;
111 char *delim = ":";
112 char namebuf[KSYM_NAME_LEN];
113 char reliab[4] = "";
114
115 symname = kallsyms_lookup(address, &symsize, &offset,
116 &modname, namebuf);
117 if (!symname) {
118 printk(" [<%016lx>]\n", address);
119 return;
120 }
121 if (!reliable)
122 strcpy(reliab, "? ");
123
124 if (!modname)
125 modname = delim = "";
126 printk(" [<%016lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
127 address, reliab, delim, modname, delim, symname, offset, symsize);
128#else
129 printk(" [<%016lx>]\n", address);
130#endif
131} 89}
132 90
133static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack, 91static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
@@ -204,8 +162,6 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
204 return NULL; 162 return NULL;
205} 163}
206 164
207#define MSG(txt) ops->warning(data, txt)
208
209/* 165/*
210 * x86-64 can have up to three kernel stacks: 166 * x86-64 can have up to three kernel stacks:
211 * process stack 167 * process stack
@@ -232,11 +188,11 @@ struct stack_frame {
232 unsigned long return_address; 188 unsigned long return_address;
233}; 189};
234 190
235 191static inline unsigned long
236static inline unsigned long print_context_stack(struct thread_info *tinfo, 192print_context_stack(struct thread_info *tinfo,
237 unsigned long *stack, unsigned long bp, 193 unsigned long *stack, unsigned long bp,
238 const struct stacktrace_ops *ops, void *data, 194 const struct stacktrace_ops *ops, void *data,
239 unsigned long *end) 195 unsigned long *end)
240{ 196{
241 struct stack_frame *frame = (struct stack_frame *)bp; 197 struct stack_frame *frame = (struct stack_frame *)bp;
242 198
@@ -258,7 +214,7 @@ static inline unsigned long print_context_stack(struct thread_info *tinfo,
258 return bp; 214 return bp;
259} 215}
260 216
261void dump_trace(struct task_struct *tsk, struct pt_regs *regs, 217void dump_trace(struct task_struct *task, struct pt_regs *regs,
262 unsigned long *stack, unsigned long bp, 218 unsigned long *stack, unsigned long bp,
263 const struct stacktrace_ops *ops, void *data) 219 const struct stacktrace_ops *ops, void *data)
264{ 220{
@@ -267,36 +223,34 @@ void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
267 unsigned used = 0; 223 unsigned used = 0;
268 struct thread_info *tinfo; 224 struct thread_info *tinfo;
269 225
270 if (!tsk) 226 if (!task)
271 tsk = current; 227 task = current;
272 tinfo = task_thread_info(tsk);
273 228
274 if (!stack) { 229 if (!stack) {
275 unsigned long dummy; 230 unsigned long dummy;
276 stack = &dummy; 231 stack = &dummy;
277 if (tsk && tsk != current) 232 if (task && task != current)
278 stack = (unsigned long *)tsk->thread.sp; 233 stack = (unsigned long *)task->thread.sp;
279 } 234 }
280 235
281#ifdef CONFIG_FRAME_POINTER 236#ifdef CONFIG_FRAME_POINTER
282 if (!bp) { 237 if (!bp) {
283 if (tsk == current) { 238 if (task == current) {
284 /* Grab bp right from our regs */ 239 /* Grab bp right from our regs */
285 asm("movq %%rbp, %0" : "=r" (bp):); 240 asm("movq %%rbp, %0" : "=r" (bp) :);
286 } else { 241 } else {
287 /* bp is the last reg pushed by switch_to */ 242 /* bp is the last reg pushed by switch_to */
288 bp = *(unsigned long *) tsk->thread.sp; 243 bp = *(unsigned long *) task->thread.sp;
289 } 244 }
290 } 245 }
291#endif 246#endif
292 247
293
294
295 /* 248 /*
296 * Print function call entries in all stacks, starting at the 249 * Print function call entries in all stacks, starting at the
297 * current stack address. If the stacks consist of nested 250 * current stack address. If the stacks consist of nested
298 * exceptions 251 * exceptions
299 */ 252 */
253 tinfo = task_thread_info(task);
300 for (;;) { 254 for (;;) {
301 char *id; 255 char *id;
302 unsigned long *estack_end; 256 unsigned long *estack_end;
@@ -381,18 +335,24 @@ static const struct stacktrace_ops print_trace_ops = {
381 .address = print_trace_address, 335 .address = print_trace_address,
382}; 336};
383 337
384void 338static void
385show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack, 339show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
386 unsigned long bp) 340 unsigned long *stack, unsigned long bp, char *log_lvl)
387{ 341{
388 printk("\nCall Trace:\n"); 342 printk("\nCall Trace:\n");
389 dump_trace(tsk, regs, stack, bp, &print_trace_ops, NULL); 343 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
390 printk("\n"); 344 printk("\n");
391} 345}
392 346
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
393static void 353static void
394_show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *sp, 354show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
395 unsigned long bp) 355 unsigned long *sp, unsigned long bp, char *log_lvl)
396{ 356{
397 unsigned long *stack; 357 unsigned long *stack;
398 int i; 358 int i;
@@ -404,14 +364,14 @@ _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *sp,
404 // back trace for this cpu. 364 // back trace for this cpu.
405 365
406 if (sp == NULL) { 366 if (sp == NULL) {
407 if (tsk) 367 if (task)
408 sp = (unsigned long *)tsk->thread.sp; 368 sp = (unsigned long *)task->thread.sp;
409 else 369 else
410 sp = (unsigned long *)&sp; 370 sp = (unsigned long *)&sp;
411 } 371 }
412 372
413 stack = sp; 373 stack = sp;
414 for(i=0; i < kstack_depth_to_print; i++) { 374 for (i = 0; i < kstack_depth_to_print; i++) {
415 if (stack >= irqstack && stack <= irqstack_end) { 375 if (stack >= irqstack && stack <= irqstack_end) {
416 if (stack == irqstack_end) { 376 if (stack == irqstack_end) {
417 stack = (unsigned long *) (irqstack_end[-1]); 377 stack = (unsigned long *) (irqstack_end[-1]);
@@ -426,12 +386,12 @@ _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *sp,
426 printk(" %016lx", *stack++); 386 printk(" %016lx", *stack++);
427 touch_nmi_watchdog(); 387 touch_nmi_watchdog();
428 } 388 }
429 show_trace(tsk, regs, sp, bp); 389 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
430} 390}
431 391
432void show_stack(struct task_struct *tsk, unsigned long * sp) 392void show_stack(struct task_struct *task, unsigned long *sp)
433{ 393{
434 _show_stack(tsk, NULL, sp, 0); 394 show_stack_log_lvl(task, NULL, sp, 0, "");
435} 395}
436 396
437/* 397/*
@@ -439,8 +399,8 @@ void show_stack(struct task_struct *tsk, unsigned long * sp)
439 */ 399 */
440void dump_stack(void) 400void dump_stack(void)
441{ 401{
442 unsigned long dummy;
443 unsigned long bp = 0; 402 unsigned long bp = 0;
403 unsigned long stack;
444 404
445#ifdef CONFIG_FRAME_POINTER 405#ifdef CONFIG_FRAME_POINTER
446 if (!bp) 406 if (!bp)
@@ -452,7 +412,7 @@ void dump_stack(void)
452 init_utsname()->release, 412 init_utsname()->release,
453 (int)strcspn(init_utsname()->version, " "), 413 (int)strcspn(init_utsname()->version, " "),
454 init_utsname()->version); 414 init_utsname()->version);
455 show_trace(NULL, NULL, &dummy, bp); 415 show_trace(NULL, NULL, &stack, bp);
456} 416}
457 417
458EXPORT_SYMBOL(dump_stack); 418EXPORT_SYMBOL(dump_stack);
@@ -463,12 +423,8 @@ void show_registers(struct pt_regs *regs)
463 unsigned long sp; 423 unsigned long sp;
464 const int cpu = smp_processor_id(); 424 const int cpu = smp_processor_id();
465 struct task_struct *cur = cpu_pda(cpu)->pcurrent; 425 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
466 u8 *ip;
467 unsigned int code_prologue = code_bytes * 43 / 64;
468 unsigned int code_len = code_bytes;
469 426
470 sp = regs->sp; 427 sp = regs->sp;
471 ip = (u8 *) regs->ip - code_prologue;
472 printk("CPU %d ", cpu); 428 printk("CPU %d ", cpu);
473 __show_regs(regs); 429 __show_regs(regs);
474 printk("Process %s (pid: %d, threadinfo %p, task %p)\n", 430 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
@@ -479,15 +435,22 @@ void show_registers(struct pt_regs *regs)
479 * time of the fault.. 435 * time of the fault..
480 */ 436 */
481 if (!user_mode(regs)) { 437 if (!user_mode(regs)) {
438 unsigned int code_prologue = code_bytes * 43 / 64;
439 unsigned int code_len = code_bytes;
482 unsigned char c; 440 unsigned char c;
441 u8 *ip;
442
483 printk("Stack: "); 443 printk("Stack: ");
484 _show_stack(NULL, regs, (unsigned long *)sp, regs->bp); 444 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
445 regs->bp, "");
485 printk("\n"); 446 printk("\n");
486 447
487 printk(KERN_EMERG "Code: "); 448 printk(KERN_EMERG "Code: ");
449
450 ip = (u8 *)regs->ip - code_prologue;
488 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) { 451 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
489 /* try starting at RIP */ 452 /* try starting at RIP */
490 ip = (u8 *) regs->ip; 453 ip = (u8 *)regs->ip;
491 code_len = code_len - code_prologue + 1; 454 code_len = code_len - code_prologue + 1;
492 } 455 }
493 for (i = 0; i < code_len; i++, ip++) { 456 for (i = 0; i < code_len; i++, ip++) {
@@ -503,7 +466,7 @@ void show_registers(struct pt_regs *regs)
503 } 466 }
504 } 467 }
505 printk("\n"); 468 printk("\n");
506} 469}
507 470
508int is_valid_bugaddr(unsigned long ip) 471int is_valid_bugaddr(unsigned long ip)
509{ 472{
@@ -543,7 +506,7 @@ unsigned __kprobes long oops_begin(void)
543} 506}
544 507
545void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr) 508void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
546{ 509{
547 die_owner = -1; 510 die_owner = -1;
548 bust_spinlocks(0); 511 bust_spinlocks(0);
549 die_nest_count--; 512 die_nest_count--;
@@ -561,10 +524,9 @@ void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
561 do_exit(signr); 524 do_exit(signr);
562} 525}
563 526
564int __kprobes __die(const char * str, struct pt_regs * regs, long err) 527int __kprobes __die(const char *str, struct pt_regs *regs, long err)
565{ 528{
566 static int die_counter; 529 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff, ++die_counter);
567 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
568#ifdef CONFIG_PREEMPT 530#ifdef CONFIG_PREEMPT
569 printk("PREEMPT "); 531 printk("PREEMPT ");
570#endif 532#endif
@@ -575,8 +537,10 @@ int __kprobes __die(const char * str, struct pt_regs * regs, long err)
575 printk("DEBUG_PAGEALLOC"); 537 printk("DEBUG_PAGEALLOC");
576#endif 538#endif
577 printk("\n"); 539 printk("\n");
578 if (notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV) == NOTIFY_STOP) 540 if (notify_die(DIE_OOPS, str, regs, err,
541 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
579 return 1; 542 return 1;
543
580 show_registers(regs); 544 show_registers(regs);
581 add_taint(TAINT_DIE); 545 add_taint(TAINT_DIE);
582 /* Executive summary in case the oops scrolled away */ 546 /* Executive summary in case the oops scrolled away */
@@ -588,7 +552,7 @@ int __kprobes __die(const char * str, struct pt_regs * regs, long err)
588 return 0; 552 return 0;
589} 553}
590 554
591void die(const char * str, struct pt_regs * regs, long err) 555void die(const char *str, struct pt_regs *regs, long err)
592{ 556{
593 unsigned long flags = oops_begin(); 557 unsigned long flags = oops_begin();
594 558
@@ -605,8 +569,7 @@ die_nmi(char *str, struct pt_regs *regs, int do_panic)
605{ 569{
606 unsigned long flags; 570 unsigned long flags;
607 571
608 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == 572 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
609 NOTIFY_STOP)
610 return; 573 return;
611 574
612 flags = oops_begin(); 575 flags = oops_begin();
@@ -614,7 +577,9 @@ die_nmi(char *str, struct pt_regs *regs, int do_panic)
614 * We are in trouble anyway, lets at least try 577 * We are in trouble anyway, lets at least try
615 * to get a message out. 578 * to get a message out.
616 */ 579 */
617 printk(str, smp_processor_id()); 580 printk(KERN_EMERG "%s", str);
581 printk(" on CPU%d, ip %08lx, registers:\n",
582 smp_processor_id(), regs->ip);
618 show_registers(regs); 583 show_registers(regs);
619 if (kexec_should_crash(current)) 584 if (kexec_should_crash(current))
620 crash_kexec(regs); 585 crash_kexec(regs);
@@ -626,44 +591,44 @@ die_nmi(char *str, struct pt_regs *regs, int do_panic)
626 do_exit(SIGBUS); 591 do_exit(SIGBUS);
627} 592}
628 593
629static void __kprobes do_trap(int trapnr, int signr, char *str, 594static void __kprobes
630 struct pt_regs * regs, long error_code, 595do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
631 siginfo_t *info) 596 long error_code, siginfo_t *info)
632{ 597{
633 struct task_struct *tsk = current; 598 struct task_struct *tsk = current;
634 599
635 if (user_mode(regs)) { 600 if (!user_mode(regs))
636 /* 601 goto kernel_trap;
637 * We want error_code and trap_no set for userspace
638 * faults and kernelspace faults which result in
639 * die(), but not kernelspace faults which are fixed
640 * up. die() gives the process no chance to handle
641 * the signal and notice the kernel fault information,
642 * so that won't result in polluting the information
643 * about previously queued, but not yet delivered,
644 * faults. See also do_general_protection below.
645 */
646 tsk->thread.error_code = error_code;
647 tsk->thread.trap_no = trapnr;
648
649 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
650 printk_ratelimit()) {
651 printk(KERN_INFO
652 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
653 tsk->comm, tsk->pid, str,
654 regs->ip, regs->sp, error_code);
655 print_vma_addr(" in ", regs->ip);
656 printk("\n");
657 }
658 602
659 if (info) 603 /*
660 force_sig_info(signr, info, tsk); 604 * We want error_code and trap_no set for userspace faults and
661 else 605 * kernelspace faults which result in die(), but not
662 force_sig(signr, tsk); 606 * kernelspace faults which are fixed up. die() gives the
663 return; 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");
664 } 623 }
665 624
625 if (info)
626 force_sig_info(signr, info, tsk);
627 else
628 force_sig(signr, tsk);
629 return;
666 630
631kernel_trap:
667 if (!fixup_exception(regs)) { 632 if (!fixup_exception(regs)) {
668 tsk->thread.error_code = error_code; 633 tsk->thread.error_code = error_code;
669 tsk->thread.trap_no = trapnr; 634 tsk->thread.trap_no = trapnr;
@@ -673,41 +638,39 @@ static void __kprobes do_trap(int trapnr, int signr, char *str,
673} 638}
674 639
675#define DO_ERROR(trapnr, signr, str, name) \ 640#define DO_ERROR(trapnr, signr, str, name) \
676asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ 641asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
677{ \ 642{ \
678 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 643 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
679 == NOTIFY_STOP) \ 644 == NOTIFY_STOP) \
680 return; \ 645 return; \
681 conditional_sti(regs); \ 646 conditional_sti(regs); \
682 do_trap(trapnr, signr, str, regs, error_code, NULL); \ 647 do_trap(trapnr, signr, str, regs, error_code, NULL); \
683} 648}
684 649
685#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ 650#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
686asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ 651asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
687{ \ 652{ \
688 siginfo_t info; \ 653 siginfo_t info; \
689 info.si_signo = signr; \ 654 info.si_signo = signr; \
690 info.si_errno = 0; \ 655 info.si_errno = 0; \
691 info.si_code = sicode; \ 656 info.si_code = sicode; \
692 info.si_addr = (void __user *)siaddr; \ 657 info.si_addr = (void __user *)siaddr; \
693 trace_hardirqs_fixup(); \ 658 trace_hardirqs_fixup(); \
694 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ 659 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
695 == NOTIFY_STOP) \ 660 == NOTIFY_STOP) \
696 return; \ 661 return; \
697 conditional_sti(regs); \ 662 conditional_sti(regs); \
698 do_trap(trapnr, signr, str, regs, error_code, &info); \ 663 do_trap(trapnr, signr, str, regs, error_code, &info); \
699} 664}
700 665
701DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip) 666DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
702DO_ERROR( 4, SIGSEGV, "overflow", overflow) 667DO_ERROR(4, SIGSEGV, "overflow", overflow)
703DO_ERROR( 5, SIGSEGV, "bounds", bounds) 668DO_ERROR(5, SIGSEGV, "bounds", bounds)
704DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip) 669DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
705DO_ERROR( 7, SIGSEGV, "device not available", device_not_available) 670DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
706DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
707DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) 671DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
708DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) 672DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
709DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0) 673DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
710DO_ERROR(18, SIGSEGV, "reserved", reserved)
711 674
712/* Runs on IST stack */ 675/* Runs on IST stack */
713asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code) 676asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
@@ -737,31 +700,34 @@ asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
737 die(str, regs, error_code); 700 die(str, regs, error_code);
738} 701}
739 702
740asmlinkage void __kprobes do_general_protection(struct pt_regs * regs, 703asmlinkage void __kprobes
741 long error_code) 704do_general_protection(struct pt_regs *regs, long error_code)
742{ 705{
743 struct task_struct *tsk = current; 706 struct task_struct *tsk;
744 707
745 conditional_sti(regs); 708 conditional_sti(regs);
746 709
747 if (user_mode(regs)) { 710 tsk = current;
748 tsk->thread.error_code = error_code; 711 if (!user_mode(regs))
749 tsk->thread.trap_no = 13; 712 goto gp_in_kernel;
750
751 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
752 printk_ratelimit()) {
753 printk(KERN_INFO
754 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
755 tsk->comm, tsk->pid,
756 regs->ip, regs->sp, error_code);
757 print_vma_addr(" in ", regs->ip);
758 printk("\n");
759 }
760 713
761 force_sig(SIGSEGV, tsk); 714 tsk->thread.error_code = error_code;
762 return; 715 tsk->thread.trap_no = 13;
763 } 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;
764 729
730gp_in_kernel:
765 if (fixup_exception(regs)) 731 if (fixup_exception(regs))
766 return; 732 return;
767 733
@@ -774,14 +740,14 @@ asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
774} 740}
775 741
776static notrace __kprobes void 742static notrace __kprobes void
777mem_parity_error(unsigned char reason, struct pt_regs * regs) 743mem_parity_error(unsigned char reason, struct pt_regs *regs)
778{ 744{
779 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n", 745 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
780 reason); 746 reason);
781 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n"); 747 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
782 748
783#if defined(CONFIG_EDAC) 749#if defined(CONFIG_EDAC)
784 if(edac_handler_set()) { 750 if (edac_handler_set()) {
785 edac_atomic_assert_error(); 751 edac_atomic_assert_error();
786 return; 752 return;
787 } 753 }
@@ -798,7 +764,7 @@ mem_parity_error(unsigned char reason, struct pt_regs * regs)
798} 764}
799 765
800static notrace __kprobes void 766static notrace __kprobes void
801io_check_error(unsigned char reason, struct pt_regs * regs) 767io_check_error(unsigned char reason, struct pt_regs *regs)
802{ 768{
803 printk("NMI: IOCK error (debug interrupt?)\n"); 769 printk("NMI: IOCK error (debug interrupt?)\n");
804 show_registers(regs); 770 show_registers(regs);
@@ -828,14 +794,14 @@ unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
828 794
829/* Runs on IST stack. This code must keep interrupts off all the time. 795/* Runs on IST stack. This code must keep interrupts off all the time.
830 Nested NMIs are prevented by the CPU. */ 796 Nested NMIs are prevented by the CPU. */
831asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs) 797asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
832{ 798{
833 unsigned char reason = 0; 799 unsigned char reason = 0;
834 int cpu; 800 int cpu;
835 801
836 cpu = smp_processor_id(); 802 cpu = smp_processor_id();
837 803
838 /* Only the BSP gets external NMIs from the system. */ 804 /* Only the BSP gets external NMIs from the system. */
839 if (!cpu) 805 if (!cpu)
840 reason = get_nmi_reason(); 806 reason = get_nmi_reason();
841 807
@@ -847,32 +813,57 @@ asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
847 * Ok, so this is none of the documented NMI sources, 813 * Ok, so this is none of the documented NMI sources,
848 * so it must be the NMI watchdog. 814 * so it must be the NMI watchdog.
849 */ 815 */
850 if (nmi_watchdog_tick(regs,reason)) 816 if (nmi_watchdog_tick(regs, reason))
851 return; 817 return;
852 if (!do_nmi_callback(regs,cpu)) 818 if (!do_nmi_callback(regs, cpu))
853 unknown_nmi_error(reason, regs); 819 unknown_nmi_error(reason, regs);
854 820
855 return; 821 return;
856 } 822 }
857 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) 823 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
858 return; 824 return;
859 825
860 /* AK: following checks seem to be broken on modern chipsets. FIXME */ 826 /* AK: following checks seem to be broken on modern chipsets. FIXME */
861
862 if (reason & 0x80) 827 if (reason & 0x80)
863 mem_parity_error(reason, regs); 828 mem_parity_error(reason, regs);
864 if (reason & 0x40) 829 if (reason & 0x40)
865 io_check_error(reason, regs); 830 io_check_error(reason, regs);
866} 831}
867 832
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
868/* runs on IST stack. */ 858/* runs on IST stack. */
869asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code) 859asmlinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
870{ 860{
871 trace_hardirqs_fixup(); 861 trace_hardirqs_fixup();
872 862
873 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) { 863 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
864 == NOTIFY_STOP)
874 return; 865 return;
875 } 866
876 preempt_conditional_sti(regs); 867 preempt_conditional_sti(regs);
877 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL); 868 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
878 preempt_conditional_cli(regs); 869 preempt_conditional_cli(regs);
@@ -903,8 +894,8 @@ asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
903asmlinkage void __kprobes do_debug(struct pt_regs * regs, 894asmlinkage void __kprobes do_debug(struct pt_regs * regs,
904 unsigned long error_code) 895 unsigned long error_code)
905{ 896{
906 unsigned long condition;
907 struct task_struct *tsk = current; 897 struct task_struct *tsk = current;
898 unsigned long condition;
908 siginfo_t info; 899 siginfo_t info;
909 900
910 trace_hardirqs_fixup(); 901 trace_hardirqs_fixup();
@@ -925,21 +916,19 @@ asmlinkage void __kprobes do_debug(struct pt_regs * regs,
925 916
926 /* Mask out spurious debug traps due to lazy DR7 setting */ 917 /* Mask out spurious debug traps due to lazy DR7 setting */
927 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { 918 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
928 if (!tsk->thread.debugreg7) { 919 if (!tsk->thread.debugreg7)
929 goto clear_dr7; 920 goto clear_dr7;
930 }
931 } 921 }
932 922
933 tsk->thread.debugreg6 = condition; 923 tsk->thread.debugreg6 = condition;
934 924
935
936 /* 925 /*
937 * Single-stepping through TF: make sure we ignore any events in 926 * Single-stepping through TF: make sure we ignore any events in
938 * kernel space (but re-enable TF when returning to user mode). 927 * kernel space (but re-enable TF when returning to user mode).
939 */ 928 */
940 if (condition & DR_STEP) { 929 if (condition & DR_STEP) {
941 if (!user_mode(regs)) 930 if (!user_mode(regs))
942 goto clear_TF_reenable; 931 goto clear_TF_reenable;
943 } 932 }
944 933
945 /* Ok, finally something we can handle */ 934 /* Ok, finally something we can handle */
@@ -952,7 +941,7 @@ asmlinkage void __kprobes do_debug(struct pt_regs * regs,
952 force_sig_info(SIGTRAP, &info, tsk); 941 force_sig_info(SIGTRAP, &info, tsk);
953 942
954clear_dr7: 943clear_dr7:
955 set_debugreg(0UL, 7); 944 set_debugreg(0, 7);
956 preempt_conditional_cli(regs); 945 preempt_conditional_cli(regs);
957 return; 946 return;
958 947
@@ -960,6 +949,7 @@ clear_TF_reenable:
960 set_tsk_thread_flag(tsk, TIF_SINGLESTEP); 949 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
961 regs->flags &= ~X86_EFLAGS_TF; 950 regs->flags &= ~X86_EFLAGS_TF;
962 preempt_conditional_cli(regs); 951 preempt_conditional_cli(regs);
952 return;
963} 953}
964 954
965static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr) 955static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
@@ -982,7 +972,7 @@ static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
982asmlinkage void do_coprocessor_error(struct pt_regs *regs) 972asmlinkage void do_coprocessor_error(struct pt_regs *regs)
983{ 973{
984 void __user *ip = (void __user *)(regs->ip); 974 void __user *ip = (void __user *)(regs->ip);
985 struct task_struct * task; 975 struct task_struct *task;
986 siginfo_t info; 976 siginfo_t info;
987 unsigned short cwd, swd; 977 unsigned short cwd, swd;
988 978
@@ -1015,30 +1005,30 @@ asmlinkage void do_coprocessor_error(struct pt_regs *regs)
1015 cwd = get_fpu_cwd(task); 1005 cwd = get_fpu_cwd(task);
1016 swd = get_fpu_swd(task); 1006 swd = get_fpu_swd(task);
1017 switch (swd & ~cwd & 0x3f) { 1007 switch (swd & ~cwd & 0x3f) {
1018 case 0x000: 1008 case 0x000: /* No unmasked exception */
1019 default: 1009 default: /* Multiple exceptions */
1020 break; 1010 break;
1021 case 0x001: /* Invalid Op */ 1011 case 0x001: /* Invalid Op */
1022 /* 1012 /*
1023 * swd & 0x240 == 0x040: Stack Underflow 1013 * swd & 0x240 == 0x040: Stack Underflow
1024 * swd & 0x240 == 0x240: Stack Overflow 1014 * swd & 0x240 == 0x240: Stack Overflow
1025 * User must clear the SF bit (0x40) if set 1015 * User must clear the SF bit (0x40) if set
1026 */ 1016 */
1027 info.si_code = FPE_FLTINV; 1017 info.si_code = FPE_FLTINV;
1028 break; 1018 break;
1029 case 0x002: /* Denormalize */ 1019 case 0x002: /* Denormalize */
1030 case 0x010: /* Underflow */ 1020 case 0x010: /* Underflow */
1031 info.si_code = FPE_FLTUND; 1021 info.si_code = FPE_FLTUND;
1032 break; 1022 break;
1033 case 0x004: /* Zero Divide */ 1023 case 0x004: /* Zero Divide */
1034 info.si_code = FPE_FLTDIV; 1024 info.si_code = FPE_FLTDIV;
1035 break; 1025 break;
1036 case 0x008: /* Overflow */ 1026 case 0x008: /* Overflow */
1037 info.si_code = FPE_FLTOVF; 1027 info.si_code = FPE_FLTOVF;
1038 break; 1028 break;
1039 case 0x020: /* Precision */ 1029 case 0x020: /* Precision */
1040 info.si_code = FPE_FLTRES; 1030 info.si_code = FPE_FLTRES;
1041 break; 1031 break;
1042 } 1032 }
1043 force_sig_info(SIGFPE, &info, task); 1033 force_sig_info(SIGFPE, &info, task);
1044} 1034}
@@ -1051,7 +1041,7 @@ asmlinkage void bad_intr(void)
1051asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs) 1041asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1052{ 1042{
1053 void __user *ip = (void __user *)(regs->ip); 1043 void __user *ip = (void __user *)(regs->ip);
1054 struct task_struct * task; 1044 struct task_struct *task;
1055 siginfo_t info; 1045 siginfo_t info;
1056 unsigned short mxcsr; 1046 unsigned short mxcsr;
1057 1047
@@ -1079,25 +1069,25 @@ asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1079 */ 1069 */
1080 mxcsr = get_fpu_mxcsr(task); 1070 mxcsr = get_fpu_mxcsr(task);
1081 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { 1071 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1082 case 0x000: 1072 case 0x000:
1083 default: 1073 default:
1084 break; 1074 break;
1085 case 0x001: /* Invalid Op */ 1075 case 0x001: /* Invalid Op */
1086 info.si_code = FPE_FLTINV; 1076 info.si_code = FPE_FLTINV;
1087 break; 1077 break;
1088 case 0x002: /* Denormalize */ 1078 case 0x002: /* Denormalize */
1089 case 0x010: /* Underflow */ 1079 case 0x010: /* Underflow */
1090 info.si_code = FPE_FLTUND; 1080 info.si_code = FPE_FLTUND;
1091 break; 1081 break;
1092 case 0x004: /* Zero Divide */ 1082 case 0x004: /* Zero Divide */
1093 info.si_code = FPE_FLTDIV; 1083 info.si_code = FPE_FLTDIV;
1094 break; 1084 break;
1095 case 0x008: /* Overflow */ 1085 case 0x008: /* Overflow */
1096 info.si_code = FPE_FLTOVF; 1086 info.si_code = FPE_FLTOVF;
1097 break; 1087 break;
1098 case 0x020: /* Precision */ 1088 case 0x020: /* Precision */
1099 info.si_code = FPE_FLTRES; 1089 info.si_code = FPE_FLTRES;
1100 break; 1090 break;
1101 } 1091 }
1102 force_sig_info(SIGFPE, &info, task); 1092 force_sig_info(SIGFPE, &info, task);
1103} 1093}
@@ -1115,7 +1105,7 @@ asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1115} 1105}
1116 1106
1117/* 1107/*
1118 * 'math_state_restore()' saves the current math information in the 1108 * 'math_state_restore()' saves the current math information in the
1119 * old math state array, and gets the new ones from the current task 1109 * old math state array, and gets the new ones from the current task
1120 * 1110 *
1121 * Careful.. There are problems with IBM-designed IRQ13 behaviour. 1111 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
@@ -1140,7 +1130,7 @@ asmlinkage void math_state_restore(void)
1140 local_irq_disable(); 1130 local_irq_disable();
1141 } 1131 }
1142 1132
1143 clts(); /* Allow maths ops (or we recurse) */ 1133 clts(); /* Allow maths ops (or we recurse) */
1144 restore_fpu_checking(&me->thread.xstate->fxsave); 1134 restore_fpu_checking(&me->thread.xstate->fxsave);
1145 task_thread_info(me)->status |= TS_USEDFPU; 1135 task_thread_info(me)->status |= TS_USEDFPU;
1146 me->fpu_counter++; 1136 me->fpu_counter++;
@@ -1149,64 +1139,61 @@ EXPORT_SYMBOL_GPL(math_state_restore);
1149 1139
1150void __init trap_init(void) 1140void __init trap_init(void)
1151{ 1141{
1152 set_intr_gate(0,&divide_error); 1142 set_intr_gate(0, &divide_error);
1153 set_intr_gate_ist(1,&debug,DEBUG_STACK); 1143 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1154 set_intr_gate_ist(2,&nmi,NMI_STACK); 1144 set_intr_gate_ist(2, &nmi, NMI_STACK);
1155 set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */ 1145 set_system_gate_ist(3, &int3, DEBUG_STACK); /* int3 can be called from all */
1156 set_system_gate(4,&overflow); /* int4 can be called from all */ 1146 set_system_gate(4, &overflow); /* int4 can be called from all */
1157 set_intr_gate(5,&bounds); 1147 set_intr_gate(5, &bounds);
1158 set_intr_gate(6,&invalid_op); 1148 set_intr_gate(6, &invalid_op);
1159 set_intr_gate(7,&device_not_available); 1149 set_intr_gate(7, &device_not_available);
1160 set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK); 1150 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1161 set_intr_gate(9,&coprocessor_segment_overrun); 1151 set_intr_gate(9, &coprocessor_segment_overrun);
1162 set_intr_gate(10,&invalid_TSS); 1152 set_intr_gate(10, &invalid_TSS);
1163 set_intr_gate(11,&segment_not_present); 1153 set_intr_gate(11, &segment_not_present);
1164 set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK); 1154 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1165 set_intr_gate(13,&general_protection); 1155 set_intr_gate(13, &general_protection);
1166 set_intr_gate(14,&page_fault); 1156 set_intr_gate(14, &page_fault);
1167 set_intr_gate(15,&spurious_interrupt_bug); 1157 set_intr_gate(15, &spurious_interrupt_bug);
1168 set_intr_gate(16,&coprocessor_error); 1158 set_intr_gate(16, &coprocessor_error);
1169 set_intr_gate(17,&alignment_check); 1159 set_intr_gate(17, &alignment_check);
1170#ifdef CONFIG_X86_MCE 1160#ifdef CONFIG_X86_MCE
1171 set_intr_gate_ist(18,&machine_check, MCE_STACK); 1161 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1172#endif 1162#endif
1173 set_intr_gate(19,&simd_coprocessor_error); 1163 set_intr_gate(19, &simd_coprocessor_error);
1174 1164
1175#ifdef CONFIG_IA32_EMULATION 1165#ifdef CONFIG_IA32_EMULATION
1176 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall); 1166 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1177#endif 1167#endif
1178
1179 /* 1168 /*
1180 * initialize the per thread extended state: 1169 * initialize the per thread extended state:
1181 */ 1170 */
1182 init_thread_xstate(); 1171 init_thread_xstate();
1183 /* 1172 /*
1184 * Should be a barrier for any external CPU state. 1173 * Should be a barrier for any external CPU state:
1185 */ 1174 */
1186 cpu_init(); 1175 cpu_init();
1187} 1176}
1188 1177
1189
1190static int __init oops_setup(char *s) 1178static int __init oops_setup(char *s)
1191{ 1179{
1192 if (!s) 1180 if (!s)
1193 return -EINVAL; 1181 return -EINVAL;
1194 if (!strcmp(s, "panic")) 1182 if (!strcmp(s, "panic"))
1195 panic_on_oops = 1; 1183 panic_on_oops = 1;
1196 return 0; 1184 return 0;
1197} 1185}
1198early_param("oops", oops_setup); 1186early_param("oops", oops_setup);
1199 1187
1200static int __init kstack_setup(char *s) 1188static int __init kstack_setup(char *s)
1201{ 1189{
1202 if (!s) 1190 if (!s)
1203 return -EINVAL; 1191 return -EINVAL;
1204 kstack_depth_to_print = simple_strtoul(s,NULL,0); 1192 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1205 return 0; 1193 return 0;
1206} 1194}
1207early_param("kstack", kstack_setup); 1195early_param("kstack", kstack_setup);
1208 1196
1209
1210static int __init code_bytes_setup(char *s) 1197static int __init code_bytes_setup(char *s)
1211{ 1198{
1212 code_bytes = simple_strtoul(s, NULL, 0); 1199 code_bytes = simple_strtoul(s, NULL, 0);
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
new file mode 100644
index 000000000000..7603c0553909
--- /dev/null
+++ b/arch/x86/kernel/tsc.c
@@ -0,0 +1,535 @@
1#include <linux/kernel.h>
2#include <linux/sched.h>
3#include <linux/init.h>
4#include <linux/module.h>
5#include <linux/timer.h>
6#include <linux/acpi_pmtmr.h>
7#include <linux/cpufreq.h>
8#include <linux/dmi.h>
9#include <linux/delay.h>
10#include <linux/clocksource.h>
11#include <linux/percpu.h>
12
13#include <asm/hpet.h>
14#include <asm/timer.h>
15#include <asm/vgtod.h>
16#include <asm/time.h>
17#include <asm/delay.h>
18
19unsigned int cpu_khz; /* TSC clocks / usec, not used here */
20EXPORT_SYMBOL(cpu_khz);
21unsigned int tsc_khz;
22EXPORT_SYMBOL(tsc_khz);
23
24/*
25 * TSC can be unstable due to cpufreq or due to unsynced TSCs
26 */
27static int tsc_unstable;
28
29/* native_sched_clock() is called before tsc_init(), so
30 we must start with the TSC soft disabled to prevent
31 erroneous rdtsc usage on !cpu_has_tsc processors */
32static int tsc_disabled = -1;
33
34/*
35 * Scheduler clock - returns current time in nanosec units.
36 */
37u64 native_sched_clock(void)
38{
39 u64 this_offset;
40
41 /*
42 * Fall back to jiffies if there's no TSC available:
43 * ( But note that we still use it if the TSC is marked
44 * unstable. We do this because unlike Time Of Day,
45 * the scheduler clock tolerates small errors and it's
46 * very important for it to be as fast as the platform
47 * can achive it. )
48 */
49 if (unlikely(tsc_disabled)) {
50 /* No locking but a rare wrong value is not a big deal: */
51 return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
52 }
53
54 /* read the Time Stamp Counter: */
55 rdtscll(this_offset);
56
57 /* return the value in ns */
58 return cycles_2_ns(this_offset);
59}
60
61/* We need to define a real function for sched_clock, to override the
62 weak default version */
63#ifdef CONFIG_PARAVIRT
64unsigned long long sched_clock(void)
65{
66 return paravirt_sched_clock();
67}
68#else
69unsigned long long
70sched_clock(void) __attribute__((alias("native_sched_clock")));
71#endif
72
73int check_tsc_unstable(void)
74{
75 return tsc_unstable;
76}
77EXPORT_SYMBOL_GPL(check_tsc_unstable);
78
79#ifdef CONFIG_X86_TSC
80int __init notsc_setup(char *str)
81{
82 printk(KERN_WARNING "notsc: Kernel compiled with CONFIG_X86_TSC, "
83 "cannot disable TSC completely.\n");
84 tsc_disabled = 1;
85 return 1;
86}
87#else
88/*
89 * disable flag for tsc. Takes effect by clearing the TSC cpu flag
90 * in cpu/common.c
91 */
92int __init notsc_setup(char *str)
93{
94 setup_clear_cpu_cap(X86_FEATURE_TSC);
95 return 1;
96}
97#endif
98
99__setup("notsc", notsc_setup);
100
101#define MAX_RETRIES 5
102#define SMI_TRESHOLD 50000
103
104/*
105 * Read TSC and the reference counters. Take care of SMI disturbance
106 */
107static u64 __init tsc_read_refs(u64 *pm, u64 *hpet)
108{
109 u64 t1, t2;
110 int i;
111
112 for (i = 0; i < MAX_RETRIES; i++) {
113 t1 = get_cycles();
114 if (hpet)
115 *hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF;
116 else
117 *pm = acpi_pm_read_early();
118 t2 = get_cycles();
119 if ((t2 - t1) < SMI_TRESHOLD)
120 return t2;
121 }
122 return ULLONG_MAX;
123}
124
125/**
126 * native_calibrate_tsc - calibrate the tsc on boot
127 */
128unsigned long native_calibrate_tsc(void)
129{
130 unsigned long flags;
131 u64 tsc1, tsc2, tr1, tr2, delta, pm1, pm2, hpet1, hpet2;
132 int hpet = is_hpet_enabled();
133 unsigned int tsc_khz_val = 0;
134
135 local_irq_save(flags);
136
137 tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL);
138
139 outb((inb(0x61) & ~0x02) | 0x01, 0x61);
140
141 outb(0xb0, 0x43);
142 outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
143 outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42);
144 tr1 = get_cycles();
145 while ((inb(0x61) & 0x20) == 0);
146 tr2 = get_cycles();
147
148 tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL);
149
150 local_irq_restore(flags);
151
152 /*
153 * Preset the result with the raw and inaccurate PIT
154 * calibration value
155 */
156 delta = (tr2 - tr1);
157 do_div(delta, 50);
158 tsc_khz_val = delta;
159
160 /* hpet or pmtimer available ? */
161 if (!hpet && !pm1 && !pm2) {
162 printk(KERN_INFO "TSC calibrated against PIT\n");
163 goto out;
164 }
165
166 /* Check, whether the sampling was disturbed by an SMI */
167 if (tsc1 == ULLONG_MAX || tsc2 == ULLONG_MAX) {
168 printk(KERN_WARNING "TSC calibration disturbed by SMI, "
169 "using PIT calibration result\n");
170 goto out;
171 }
172
173 tsc2 = (tsc2 - tsc1) * 1000000LL;
174
175 if (hpet) {
176 printk(KERN_INFO "TSC calibrated against HPET\n");
177 if (hpet2 < hpet1)
178 hpet2 += 0x100000000ULL;
179 hpet2 -= hpet1;
180 tsc1 = ((u64)hpet2 * hpet_readl(HPET_PERIOD));
181 do_div(tsc1, 1000000);
182 } else {
183 printk(KERN_INFO "TSC calibrated against PM_TIMER\n");
184 if (pm2 < pm1)
185 pm2 += (u64)ACPI_PM_OVRRUN;
186 pm2 -= pm1;
187 tsc1 = pm2 * 1000000000LL;
188 do_div(tsc1, PMTMR_TICKS_PER_SEC);
189 }
190
191 do_div(tsc2, tsc1);
192 tsc_khz_val = tsc2;
193
194out:
195 return tsc_khz_val;
196}
197
198
199#ifdef CONFIG_X86_32
200/* Only called from the Powernow K7 cpu freq driver */
201int recalibrate_cpu_khz(void)
202{
203#ifndef CONFIG_SMP
204 unsigned long cpu_khz_old = cpu_khz;
205
206 if (cpu_has_tsc) {
207 tsc_khz = calibrate_tsc();
208 cpu_khz = tsc_khz;
209 cpu_data(0).loops_per_jiffy =
210 cpufreq_scale(cpu_data(0).loops_per_jiffy,
211 cpu_khz_old, cpu_khz);
212 return 0;
213 } else
214 return -ENODEV;
215#else
216 return -ENODEV;
217#endif
218}
219
220EXPORT_SYMBOL(recalibrate_cpu_khz);
221
222#endif /* CONFIG_X86_32 */
223
224/* Accelerators for sched_clock()
225 * convert from cycles(64bits) => nanoseconds (64bits)
226 * basic equation:
227 * ns = cycles / (freq / ns_per_sec)
228 * ns = cycles * (ns_per_sec / freq)
229 * ns = cycles * (10^9 / (cpu_khz * 10^3))
230 * ns = cycles * (10^6 / cpu_khz)
231 *
232 * Then we use scaling math (suggested by george@mvista.com) to get:
233 * ns = cycles * (10^6 * SC / cpu_khz) / SC
234 * ns = cycles * cyc2ns_scale / SC
235 *
236 * And since SC is a constant power of two, we can convert the div
237 * into a shift.
238 *
239 * We can use khz divisor instead of mhz to keep a better precision, since
240 * cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
241 * (mathieu.desnoyers@polymtl.ca)
242 *
243 * -johnstul@us.ibm.com "math is hard, lets go shopping!"
244 */
245
246DEFINE_PER_CPU(unsigned long, cyc2ns);
247
248static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
249{
250 unsigned long long tsc_now, ns_now;
251 unsigned long flags, *scale;
252
253 local_irq_save(flags);
254 sched_clock_idle_sleep_event();
255
256 scale = &per_cpu(cyc2ns, cpu);
257
258 rdtscll(tsc_now);
259 ns_now = __cycles_2_ns(tsc_now);
260
261 if (cpu_khz)
262 *scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz;
263
264 sched_clock_idle_wakeup_event(0);
265 local_irq_restore(flags);
266}
267
268#ifdef CONFIG_CPU_FREQ
269
270/* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
271 * changes.
272 *
273 * RED-PEN: On SMP we assume all CPUs run with the same frequency. It's
274 * not that important because current Opteron setups do not support
275 * scaling on SMP anyroads.
276 *
277 * Should fix up last_tsc too. Currently gettimeofday in the
278 * first tick after the change will be slightly wrong.
279 */
280
281static unsigned int ref_freq;
282static unsigned long loops_per_jiffy_ref;
283static unsigned long tsc_khz_ref;
284
285static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
286 void *data)
287{
288 struct cpufreq_freqs *freq = data;
289 unsigned long *lpj, dummy;
290
291 if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC))
292 return 0;
293
294 lpj = &dummy;
295 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
296#ifdef CONFIG_SMP
297 lpj = &cpu_data(freq->cpu).loops_per_jiffy;
298#else
299 lpj = &boot_cpu_data.loops_per_jiffy;
300#endif
301
302 if (!ref_freq) {
303 ref_freq = freq->old;
304 loops_per_jiffy_ref = *lpj;
305 tsc_khz_ref = tsc_khz;
306 }
307 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
308 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
309 (val == CPUFREQ_RESUMECHANGE)) {
310 *lpj = cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
311
312 tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
313 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
314 mark_tsc_unstable("cpufreq changes");
315 }
316
317 set_cyc2ns_scale(tsc_khz_ref, freq->cpu);
318
319 return 0;
320}
321
322static struct notifier_block time_cpufreq_notifier_block = {
323 .notifier_call = time_cpufreq_notifier
324};
325
326static int __init cpufreq_tsc(void)
327{
328 cpufreq_register_notifier(&time_cpufreq_notifier_block,
329 CPUFREQ_TRANSITION_NOTIFIER);
330 return 0;
331}
332
333core_initcall(cpufreq_tsc);
334
335#endif /* CONFIG_CPU_FREQ */
336
337/* clocksource code */
338
339static struct clocksource clocksource_tsc;
340
341/*
342 * We compare the TSC to the cycle_last value in the clocksource
343 * structure to avoid a nasty time-warp. This can be observed in a
344 * very small window right after one CPU updated cycle_last under
345 * xtime/vsyscall_gtod lock and the other CPU reads a TSC value which
346 * is smaller than the cycle_last reference value due to a TSC which
347 * is slighty behind. This delta is nowhere else observable, but in
348 * that case it results in a forward time jump in the range of hours
349 * due to the unsigned delta calculation of the time keeping core
350 * code, which is necessary to support wrapping clocksources like pm
351 * timer.
352 */
353static cycle_t read_tsc(void)
354{
355 cycle_t ret = (cycle_t)get_cycles();
356
357 return ret >= clocksource_tsc.cycle_last ?
358 ret : clocksource_tsc.cycle_last;
359}
360
361#ifdef CONFIG_X86_64
362static cycle_t __vsyscall_fn vread_tsc(void)
363{
364 cycle_t ret = (cycle_t)vget_cycles();
365
366 return ret >= __vsyscall_gtod_data.clock.cycle_last ?
367 ret : __vsyscall_gtod_data.clock.cycle_last;
368}
369#endif
370
371static struct clocksource clocksource_tsc = {
372 .name = "tsc",
373 .rating = 300,
374 .read = read_tsc,
375 .mask = CLOCKSOURCE_MASK(64),
376 .shift = 22,
377 .flags = CLOCK_SOURCE_IS_CONTINUOUS |
378 CLOCK_SOURCE_MUST_VERIFY,
379#ifdef CONFIG_X86_64
380 .vread = vread_tsc,
381#endif
382};
383
384void mark_tsc_unstable(char *reason)
385{
386 if (!tsc_unstable) {
387 tsc_unstable = 1;
388 printk("Marking TSC unstable due to %s\n", reason);
389 /* Change only the rating, when not registered */
390 if (clocksource_tsc.mult)
391 clocksource_change_rating(&clocksource_tsc, 0);
392 else
393 clocksource_tsc.rating = 0;
394 }
395}
396
397EXPORT_SYMBOL_GPL(mark_tsc_unstable);
398
399static int __init dmi_mark_tsc_unstable(const struct dmi_system_id *d)
400{
401 printk(KERN_NOTICE "%s detected: marking TSC unstable.\n",
402 d->ident);
403 tsc_unstable = 1;
404 return 0;
405}
406
407/* List of systems that have known TSC problems */
408static struct dmi_system_id __initdata bad_tsc_dmi_table[] = {
409 {
410 .callback = dmi_mark_tsc_unstable,
411 .ident = "IBM Thinkpad 380XD",
412 .matches = {
413 DMI_MATCH(DMI_BOARD_VENDOR, "IBM"),
414 DMI_MATCH(DMI_BOARD_NAME, "2635FA0"),
415 },
416 },
417 {}
418};
419
420/*
421 * Geode_LX - the OLPC CPU has a possibly a very reliable TSC
422 */
423#ifdef CONFIG_MGEODE_LX
424/* RTSC counts during suspend */
425#define RTSC_SUSP 0x100
426
427static void __init check_geode_tsc_reliable(void)
428{
429 unsigned long res_low, res_high;
430
431 rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
432 if (res_low & RTSC_SUSP)
433 clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
434}
435#else
436static inline void check_geode_tsc_reliable(void) { }
437#endif
438
439/*
440 * Make an educated guess if the TSC is trustworthy and synchronized
441 * over all CPUs.
442 */
443__cpuinit int unsynchronized_tsc(void)
444{
445 if (!cpu_has_tsc || tsc_unstable)
446 return 1;
447
448#ifdef CONFIG_SMP
449 if (apic_is_clustered_box())
450 return 1;
451#endif
452
453 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
454 return 0;
455 /*
456 * Intel systems are normally all synchronized.
457 * Exceptions must mark TSC as unstable:
458 */
459 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
460 /* assume multi socket systems are not synchronized: */
461 if (num_possible_cpus() > 1)
462 tsc_unstable = 1;
463 }
464
465 return tsc_unstable;
466}
467
468static void __init init_tsc_clocksource(void)
469{
470 clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
471 clocksource_tsc.shift);
472 /* lower the rating if we already know its unstable: */
473 if (check_tsc_unstable()) {
474 clocksource_tsc.rating = 0;
475 clocksource_tsc.flags &= ~CLOCK_SOURCE_IS_CONTINUOUS;
476 }
477 clocksource_register(&clocksource_tsc);
478}
479
480void __init tsc_init(void)
481{
482 u64 lpj;
483 int cpu;
484
485 if (!cpu_has_tsc)
486 return;
487
488 tsc_khz = calibrate_tsc();
489 cpu_khz = tsc_khz;
490
491 if (!tsc_khz) {
492 mark_tsc_unstable("could not calculate TSC khz");
493 return;
494 }
495
496#ifdef CONFIG_X86_64
497 if (cpu_has(&boot_cpu_data, X86_FEATURE_CONSTANT_TSC) &&
498 (boot_cpu_data.x86_vendor == X86_VENDOR_AMD))
499 cpu_khz = calibrate_cpu();
500#endif
501
502 lpj = ((u64)tsc_khz * 1000);
503 do_div(lpj, HZ);
504 lpj_fine = lpj;
505
506 printk("Detected %lu.%03lu MHz processor.\n",
507 (unsigned long)cpu_khz / 1000,
508 (unsigned long)cpu_khz % 1000);
509
510 /*
511 * Secondary CPUs do not run through tsc_init(), so set up
512 * all the scale factors for all CPUs, assuming the same
513 * speed as the bootup CPU. (cpufreq notifiers will fix this
514 * up if their speed diverges)
515 */
516 for_each_possible_cpu(cpu)
517 set_cyc2ns_scale(cpu_khz, cpu);
518
519 if (tsc_disabled > 0)
520 return;
521
522 /* now allow native_sched_clock() to use rdtsc */
523 tsc_disabled = 0;
524
525 use_tsc_delay();
526 /* Check and install the TSC clocksource */
527 dmi_check_system(bad_tsc_dmi_table);
528
529 if (unsynchronized_tsc())
530 mark_tsc_unstable("TSCs unsynchronized");
531
532 check_geode_tsc_reliable();
533 init_tsc_clocksource();
534}
535
diff --git a/arch/x86/kernel/tsc_32.c b/arch/x86/kernel/tsc_32.c
deleted file mode 100644
index 068759db63dd..000000000000
--- a/arch/x86/kernel/tsc_32.c
+++ /dev/null
@@ -1,453 +0,0 @@
1#include <linux/sched.h>
2#include <linux/clocksource.h>
3#include <linux/workqueue.h>
4#include <linux/cpufreq.h>
5#include <linux/jiffies.h>
6#include <linux/init.h>
7#include <linux/dmi.h>
8#include <linux/percpu.h>
9
10#include <asm/delay.h>
11#include <asm/tsc.h>
12#include <asm/io.h>
13#include <asm/timer.h>
14
15#include "mach_timer.h"
16
17static int tsc_disabled;
18
19/*
20 * On some systems the TSC frequency does not
21 * change with the cpu frequency. So we need
22 * an extra value to store the TSC freq
23 */
24unsigned int tsc_khz;
25EXPORT_SYMBOL_GPL(tsc_khz);
26
27#ifdef CONFIG_X86_TSC
28static int __init tsc_setup(char *str)
29{
30 printk(KERN_WARNING "notsc: Kernel compiled with CONFIG_X86_TSC, "
31 "cannot disable TSC completely.\n");
32 tsc_disabled = 1;
33 return 1;
34}
35#else
36/*
37 * disable flag for tsc. Takes effect by clearing the TSC cpu flag
38 * in cpu/common.c
39 */
40static int __init tsc_setup(char *str)
41{
42 setup_clear_cpu_cap(X86_FEATURE_TSC);
43 return 1;
44}
45#endif
46
47__setup("notsc", tsc_setup);
48
49/*
50 * code to mark and check if the TSC is unstable
51 * due to cpufreq or due to unsynced TSCs
52 */
53static int tsc_unstable;
54
55int check_tsc_unstable(void)
56{
57 return tsc_unstable;
58}
59EXPORT_SYMBOL_GPL(check_tsc_unstable);
60
61/* Accelerators for sched_clock()
62 * convert from cycles(64bits) => nanoseconds (64bits)
63 * basic equation:
64 * ns = cycles / (freq / ns_per_sec)
65 * ns = cycles * (ns_per_sec / freq)
66 * ns = cycles * (10^9 / (cpu_khz * 10^3))
67 * ns = cycles * (10^6 / cpu_khz)
68 *
69 * Then we use scaling math (suggested by george@mvista.com) to get:
70 * ns = cycles * (10^6 * SC / cpu_khz) / SC
71 * ns = cycles * cyc2ns_scale / SC
72 *
73 * And since SC is a constant power of two, we can convert the div
74 * into a shift.
75 *
76 * We can use khz divisor instead of mhz to keep a better precision, since
77 * cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
78 * (mathieu.desnoyers@polymtl.ca)
79 *
80 * -johnstul@us.ibm.com "math is hard, lets go shopping!"
81 */
82
83DEFINE_PER_CPU(unsigned long, cyc2ns);
84
85static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
86{
87 unsigned long long tsc_now, ns_now;
88 unsigned long flags, *scale;
89
90 local_irq_save(flags);
91 sched_clock_idle_sleep_event();
92
93 scale = &per_cpu(cyc2ns, cpu);
94
95 rdtscll(tsc_now);
96 ns_now = __cycles_2_ns(tsc_now);
97
98 if (cpu_khz)
99 *scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz;
100
101 /*
102 * Start smoothly with the new frequency:
103 */
104 sched_clock_idle_wakeup_event(0);
105 local_irq_restore(flags);
106}
107
108/*
109 * Scheduler clock - returns current time in nanosec units.
110 */
111unsigned long long native_sched_clock(void)
112{
113 unsigned long long this_offset;
114
115 /*
116 * Fall back to jiffies if there's no TSC available:
117 * ( But note that we still use it if the TSC is marked
118 * unstable. We do this because unlike Time Of Day,
119 * the scheduler clock tolerates small errors and it's
120 * very important for it to be as fast as the platform
121 * can achive it. )
122 */
123 if (unlikely(tsc_disabled))
124 /* No locking but a rare wrong value is not a big deal: */
125 return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
126
127 /* read the Time Stamp Counter: */
128 rdtscll(this_offset);
129
130 /* return the value in ns */
131 return cycles_2_ns(this_offset);
132}
133
134/* We need to define a real function for sched_clock, to override the
135 weak default version */
136#ifdef CONFIG_PARAVIRT
137unsigned long long sched_clock(void)
138{
139 return paravirt_sched_clock();
140}
141#else
142unsigned long long sched_clock(void)
143 __attribute__((alias("native_sched_clock")));
144#endif
145
146unsigned long native_calculate_cpu_khz(void)
147{
148 unsigned long long start, end;
149 unsigned long count;
150 u64 delta64 = (u64)ULLONG_MAX;
151 int i;
152 unsigned long flags;
153
154 local_irq_save(flags);
155
156 /* run 3 times to ensure the cache is warm and to get an accurate reading */
157 for (i = 0; i < 3; i++) {
158 mach_prepare_counter();
159 rdtscll(start);
160 mach_countup(&count);
161 rdtscll(end);
162
163 /*
164 * Error: ECTCNEVERSET
165 * The CTC wasn't reliable: we got a hit on the very first read,
166 * or the CPU was so fast/slow that the quotient wouldn't fit in
167 * 32 bits..
168 */
169 if (count <= 1)
170 continue;
171
172 /* cpu freq too slow: */
173 if ((end - start) <= CALIBRATE_TIME_MSEC)
174 continue;
175
176 /*
177 * We want the minimum time of all runs in case one of them
178 * is inaccurate due to SMI or other delay
179 */
180 delta64 = min(delta64, (end - start));
181 }
182
183 /* cpu freq too fast (or every run was bad): */
184 if (delta64 > (1ULL<<32))
185 goto err;
186
187 delta64 += CALIBRATE_TIME_MSEC/2; /* round for do_div */
188 do_div(delta64,CALIBRATE_TIME_MSEC);
189
190 local_irq_restore(flags);
191 return (unsigned long)delta64;
192err:
193 local_irq_restore(flags);
194 return 0;
195}
196
197int recalibrate_cpu_khz(void)
198{
199#ifndef CONFIG_SMP
200 unsigned long cpu_khz_old = cpu_khz;
201
202 if (cpu_has_tsc) {
203 cpu_khz = calculate_cpu_khz();
204 tsc_khz = cpu_khz;
205 cpu_data(0).loops_per_jiffy =
206 cpufreq_scale(cpu_data(0).loops_per_jiffy,
207 cpu_khz_old, cpu_khz);
208 return 0;
209 } else
210 return -ENODEV;
211#else
212 return -ENODEV;
213#endif
214}
215
216EXPORT_SYMBOL(recalibrate_cpu_khz);
217
218#ifdef CONFIG_CPU_FREQ
219
220/*
221 * if the CPU frequency is scaled, TSC-based delays will need a different
222 * loops_per_jiffy value to function properly.
223 */
224static unsigned int ref_freq;
225static unsigned long loops_per_jiffy_ref;
226static unsigned long cpu_khz_ref;
227
228static int
229time_cpufreq_notifier(struct notifier_block *nb, unsigned long val, void *data)
230{
231 struct cpufreq_freqs *freq = data;
232
233 if (!ref_freq) {
234 if (!freq->old){
235 ref_freq = freq->new;
236 return 0;
237 }
238 ref_freq = freq->old;
239 loops_per_jiffy_ref = cpu_data(freq->cpu).loops_per_jiffy;
240 cpu_khz_ref = cpu_khz;
241 }
242
243 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
244 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
245 (val == CPUFREQ_RESUMECHANGE)) {
246 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
247 cpu_data(freq->cpu).loops_per_jiffy =
248 cpufreq_scale(loops_per_jiffy_ref,
249 ref_freq, freq->new);
250
251 if (cpu_khz) {
252
253 if (num_online_cpus() == 1)
254 cpu_khz = cpufreq_scale(cpu_khz_ref,
255 ref_freq, freq->new);
256 if (!(freq->flags & CPUFREQ_CONST_LOOPS)) {
257 tsc_khz = cpu_khz;
258 set_cyc2ns_scale(cpu_khz, freq->cpu);
259 /*
260 * TSC based sched_clock turns
261 * to junk w/ cpufreq
262 */
263 mark_tsc_unstable("cpufreq changes");
264 }
265 }
266 }
267
268 return 0;
269}
270
271static struct notifier_block time_cpufreq_notifier_block = {
272 .notifier_call = time_cpufreq_notifier
273};
274
275static int __init cpufreq_tsc(void)
276{
277 return cpufreq_register_notifier(&time_cpufreq_notifier_block,
278 CPUFREQ_TRANSITION_NOTIFIER);
279}
280core_initcall(cpufreq_tsc);
281
282#endif
283
284/* clock source code */
285
286static unsigned long current_tsc_khz;
287static struct clocksource clocksource_tsc;
288
289/*
290 * We compare the TSC to the cycle_last value in the clocksource
291 * structure to avoid a nasty time-warp issue. This can be observed in
292 * a very small window right after one CPU updated cycle_last under
293 * xtime lock and the other CPU reads a TSC value which is smaller
294 * than the cycle_last reference value due to a TSC which is slighty
295 * behind. This delta is nowhere else observable, but in that case it
296 * results in a forward time jump in the range of hours due to the
297 * unsigned delta calculation of the time keeping core code, which is
298 * necessary to support wrapping clocksources like pm timer.
299 */
300static cycle_t read_tsc(void)
301{
302 cycle_t ret;
303
304 rdtscll(ret);
305
306 return ret >= clocksource_tsc.cycle_last ?
307 ret : clocksource_tsc.cycle_last;
308}
309
310static struct clocksource clocksource_tsc = {
311 .name = "tsc",
312 .rating = 300,
313 .read = read_tsc,
314 .mask = CLOCKSOURCE_MASK(64),
315 .mult = 0, /* to be set */
316 .shift = 22,
317 .flags = CLOCK_SOURCE_IS_CONTINUOUS |
318 CLOCK_SOURCE_MUST_VERIFY,
319};
320
321void mark_tsc_unstable(char *reason)
322{
323 if (!tsc_unstable) {
324 tsc_unstable = 1;
325 printk("Marking TSC unstable due to: %s.\n", reason);
326 /* Can be called before registration */
327 if (clocksource_tsc.mult)
328 clocksource_change_rating(&clocksource_tsc, 0);
329 else
330 clocksource_tsc.rating = 0;
331 }
332}
333EXPORT_SYMBOL_GPL(mark_tsc_unstable);
334
335static int __init dmi_mark_tsc_unstable(const struct dmi_system_id *d)
336{
337 printk(KERN_NOTICE "%s detected: marking TSC unstable.\n",
338 d->ident);
339 tsc_unstable = 1;
340 return 0;
341}
342
343/* List of systems that have known TSC problems */
344static struct dmi_system_id __initdata bad_tsc_dmi_table[] = {
345 {
346 .callback = dmi_mark_tsc_unstable,
347 .ident = "IBM Thinkpad 380XD",
348 .matches = {
349 DMI_MATCH(DMI_BOARD_VENDOR, "IBM"),
350 DMI_MATCH(DMI_BOARD_NAME, "2635FA0"),
351 },
352 },
353 {}
354};
355
356/*
357 * Make an educated guess if the TSC is trustworthy and synchronized
358 * over all CPUs.
359 */
360__cpuinit int unsynchronized_tsc(void)
361{
362 if (!cpu_has_tsc || tsc_unstable)
363 return 1;
364
365 /* Anything with constant TSC should be synchronized */
366 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
367 return 0;
368
369 /*
370 * Intel systems are normally all synchronized.
371 * Exceptions must mark TSC as unstable:
372 */
373 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
374 /* assume multi socket systems are not synchronized: */
375 if (num_possible_cpus() > 1)
376 tsc_unstable = 1;
377 }
378 return tsc_unstable;
379}
380
381/*
382 * Geode_LX - the OLPC CPU has a possibly a very reliable TSC
383 */
384#ifdef CONFIG_MGEODE_LX
385/* RTSC counts during suspend */
386#define RTSC_SUSP 0x100
387
388static void __init check_geode_tsc_reliable(void)
389{
390 unsigned long res_low, res_high;
391
392 rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
393 if (res_low & RTSC_SUSP)
394 clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
395}
396#else
397static inline void check_geode_tsc_reliable(void) { }
398#endif
399
400
401void __init tsc_init(void)
402{
403 int cpu;
404
405 if (!cpu_has_tsc || tsc_disabled) {
406 /* Disable the TSC in case of !cpu_has_tsc */
407 tsc_disabled = 1;
408 return;
409 }
410
411 cpu_khz = calculate_cpu_khz();
412 tsc_khz = cpu_khz;
413
414 if (!cpu_khz) {
415 mark_tsc_unstable("could not calculate TSC khz");
416 /*
417 * We need to disable the TSC completely in this case
418 * to prevent sched_clock() from using it.
419 */
420 tsc_disabled = 1;
421 return;
422 }
423
424 printk("Detected %lu.%03lu MHz processor.\n",
425 (unsigned long)cpu_khz / 1000,
426 (unsigned long)cpu_khz % 1000);
427
428 /*
429 * Secondary CPUs do not run through tsc_init(), so set up
430 * all the scale factors for all CPUs, assuming the same
431 * speed as the bootup CPU. (cpufreq notifiers will fix this
432 * up if their speed diverges)
433 */
434 for_each_possible_cpu(cpu)
435 set_cyc2ns_scale(cpu_khz, cpu);
436
437 use_tsc_delay();
438
439 /* Check and install the TSC clocksource */
440 dmi_check_system(bad_tsc_dmi_table);
441
442 unsynchronized_tsc();
443 check_geode_tsc_reliable();
444 current_tsc_khz = tsc_khz;
445 clocksource_tsc.mult = clocksource_khz2mult(current_tsc_khz,
446 clocksource_tsc.shift);
447 /* lower the rating if we already know its unstable: */
448 if (check_tsc_unstable()) {
449 clocksource_tsc.rating = 0;
450 clocksource_tsc.flags &= ~CLOCK_SOURCE_IS_CONTINUOUS;
451 }
452 clocksource_register(&clocksource_tsc);
453}
diff --git a/arch/x86/kernel/tsc_64.c b/arch/x86/kernel/tsc_64.c
deleted file mode 100644
index 1784b8077a12..000000000000
--- a/arch/x86/kernel/tsc_64.c
+++ /dev/null
@@ -1,357 +0,0 @@
1#include <linux/kernel.h>
2#include <linux/sched.h>
3#include <linux/interrupt.h>
4#include <linux/init.h>
5#include <linux/clocksource.h>
6#include <linux/time.h>
7#include <linux/acpi.h>
8#include <linux/cpufreq.h>
9#include <linux/acpi_pmtmr.h>
10
11#include <asm/hpet.h>
12#include <asm/timex.h>
13#include <asm/timer.h>
14#include <asm/vgtod.h>
15
16static int notsc __initdata = 0;
17
18unsigned int cpu_khz; /* TSC clocks / usec, not used here */
19EXPORT_SYMBOL(cpu_khz);
20unsigned int tsc_khz;
21EXPORT_SYMBOL(tsc_khz);
22
23/* Accelerators for sched_clock()
24 * convert from cycles(64bits) => nanoseconds (64bits)
25 * basic equation:
26 * ns = cycles / (freq / ns_per_sec)
27 * ns = cycles * (ns_per_sec / freq)
28 * ns = cycles * (10^9 / (cpu_khz * 10^3))
29 * ns = cycles * (10^6 / cpu_khz)
30 *
31 * Then we use scaling math (suggested by george@mvista.com) to get:
32 * ns = cycles * (10^6 * SC / cpu_khz) / SC
33 * ns = cycles * cyc2ns_scale / SC
34 *
35 * And since SC is a constant power of two, we can convert the div
36 * into a shift.
37 *
38 * We can use khz divisor instead of mhz to keep a better precision, since
39 * cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
40 * (mathieu.desnoyers@polymtl.ca)
41 *
42 * -johnstul@us.ibm.com "math is hard, lets go shopping!"
43 */
44DEFINE_PER_CPU(unsigned long, cyc2ns);
45
46static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
47{
48 unsigned long long tsc_now, ns_now;
49 unsigned long flags, *scale;
50
51 local_irq_save(flags);
52 sched_clock_idle_sleep_event();
53
54 scale = &per_cpu(cyc2ns, cpu);
55
56 rdtscll(tsc_now);
57 ns_now = __cycles_2_ns(tsc_now);
58
59 if (cpu_khz)
60 *scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz;
61
62 sched_clock_idle_wakeup_event(0);
63 local_irq_restore(flags);
64}
65
66unsigned long long native_sched_clock(void)
67{
68 unsigned long a = 0;
69
70 /* Could do CPU core sync here. Opteron can execute rdtsc speculatively,
71 * which means it is not completely exact and may not be monotonous
72 * between CPUs. But the errors should be too small to matter for
73 * scheduling purposes.
74 */
75
76 rdtscll(a);
77 return cycles_2_ns(a);
78}
79
80/* We need to define a real function for sched_clock, to override the
81 weak default version */
82#ifdef CONFIG_PARAVIRT
83unsigned long long sched_clock(void)
84{
85 return paravirt_sched_clock();
86}
87#else
88unsigned long long
89sched_clock(void) __attribute__((alias("native_sched_clock")));
90#endif
91
92
93static int tsc_unstable;
94
95int check_tsc_unstable(void)
96{
97 return tsc_unstable;
98}
99EXPORT_SYMBOL_GPL(check_tsc_unstable);
100
101#ifdef CONFIG_CPU_FREQ
102
103/* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
104 * changes.
105 *
106 * RED-PEN: On SMP we assume all CPUs run with the same frequency. It's
107 * not that important because current Opteron setups do not support
108 * scaling on SMP anyroads.
109 *
110 * Should fix up last_tsc too. Currently gettimeofday in the
111 * first tick after the change will be slightly wrong.
112 */
113
114static unsigned int ref_freq;
115static unsigned long loops_per_jiffy_ref;
116static unsigned long tsc_khz_ref;
117
118static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
119 void *data)
120{
121 struct cpufreq_freqs *freq = data;
122 unsigned long *lpj, dummy;
123
124 if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC))
125 return 0;
126
127 lpj = &dummy;
128 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
129#ifdef CONFIG_SMP
130 lpj = &cpu_data(freq->cpu).loops_per_jiffy;
131#else
132 lpj = &boot_cpu_data.loops_per_jiffy;
133#endif
134
135 if (!ref_freq) {
136 ref_freq = freq->old;
137 loops_per_jiffy_ref = *lpj;
138 tsc_khz_ref = tsc_khz;
139 }
140 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
141 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
142 (val == CPUFREQ_RESUMECHANGE)) {
143 *lpj =
144 cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
145
146 tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
147 if (!(freq->flags & CPUFREQ_CONST_LOOPS))
148 mark_tsc_unstable("cpufreq changes");
149 }
150
151 set_cyc2ns_scale(tsc_khz_ref, freq->cpu);
152
153 return 0;
154}
155
156static struct notifier_block time_cpufreq_notifier_block = {
157 .notifier_call = time_cpufreq_notifier
158};
159
160static int __init cpufreq_tsc(void)
161{
162 cpufreq_register_notifier(&time_cpufreq_notifier_block,
163 CPUFREQ_TRANSITION_NOTIFIER);
164 return 0;
165}
166
167core_initcall(cpufreq_tsc);
168
169#endif
170
171#define MAX_RETRIES 5
172#define SMI_TRESHOLD 50000
173
174/*
175 * Read TSC and the reference counters. Take care of SMI disturbance
176 */
177static unsigned long __init tsc_read_refs(unsigned long *pm,
178 unsigned long *hpet)
179{
180 unsigned long t1, t2;
181 int i;
182
183 for (i = 0; i < MAX_RETRIES; i++) {
184 t1 = get_cycles();
185 if (hpet)
186 *hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF;
187 else
188 *pm = acpi_pm_read_early();
189 t2 = get_cycles();
190 if ((t2 - t1) < SMI_TRESHOLD)
191 return t2;
192 }
193 return ULONG_MAX;
194}
195
196/**
197 * tsc_calibrate - calibrate the tsc on boot
198 */
199void __init tsc_calibrate(void)
200{
201 unsigned long flags, tsc1, tsc2, tr1, tr2, pm1, pm2, hpet1, hpet2;
202 int hpet = is_hpet_enabled(), cpu;
203
204 local_irq_save(flags);
205
206 tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL);
207
208 outb((inb(0x61) & ~0x02) | 0x01, 0x61);
209
210 outb(0xb0, 0x43);
211 outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
212 outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42);
213 tr1 = get_cycles();
214 while ((inb(0x61) & 0x20) == 0);
215 tr2 = get_cycles();
216
217 tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL);
218
219 local_irq_restore(flags);
220
221 /*
222 * Preset the result with the raw and inaccurate PIT
223 * calibration value
224 */
225 tsc_khz = (tr2 - tr1) / 50;
226
227 /* hpet or pmtimer available ? */
228 if (!hpet && !pm1 && !pm2) {
229 printk(KERN_INFO "TSC calibrated against PIT\n");
230 goto out;
231 }
232
233 /* Check, whether the sampling was disturbed by an SMI */
234 if (tsc1 == ULONG_MAX || tsc2 == ULONG_MAX) {
235 printk(KERN_WARNING "TSC calibration disturbed by SMI, "
236 "using PIT calibration result\n");
237 goto out;
238 }
239
240 tsc2 = (tsc2 - tsc1) * 1000000L;
241
242 if (hpet) {
243 printk(KERN_INFO "TSC calibrated against HPET\n");
244 if (hpet2 < hpet1)
245 hpet2 += 0x100000000;
246 hpet2 -= hpet1;
247 tsc1 = (hpet2 * hpet_readl(HPET_PERIOD)) / 1000000;
248 } else {
249 printk(KERN_INFO "TSC calibrated against PM_TIMER\n");
250 if (pm2 < pm1)
251 pm2 += ACPI_PM_OVRRUN;
252 pm2 -= pm1;
253 tsc1 = (pm2 * 1000000000) / PMTMR_TICKS_PER_SEC;
254 }
255
256 tsc_khz = tsc2 / tsc1;
257
258out:
259 for_each_possible_cpu(cpu)
260 set_cyc2ns_scale(tsc_khz, cpu);
261}
262
263/*
264 * Make an educated guess if the TSC is trustworthy and synchronized
265 * over all CPUs.
266 */
267__cpuinit int unsynchronized_tsc(void)
268{
269 if (tsc_unstable)
270 return 1;
271
272#ifdef CONFIG_SMP
273 if (apic_is_clustered_box())
274 return 1;
275#endif
276
277 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
278 return 0;
279
280 /* Assume multi socket systems are not synchronized */
281 return num_present_cpus() > 1;
282}
283
284int __init notsc_setup(char *s)
285{
286 notsc = 1;
287 return 1;
288}
289
290__setup("notsc", notsc_setup);
291
292static struct clocksource clocksource_tsc;
293
294/*
295 * We compare the TSC to the cycle_last value in the clocksource
296 * structure to avoid a nasty time-warp. This can be observed in a
297 * very small window right after one CPU updated cycle_last under
298 * xtime/vsyscall_gtod lock and the other CPU reads a TSC value which
299 * is smaller than the cycle_last reference value due to a TSC which
300 * is slighty behind. This delta is nowhere else observable, but in
301 * that case it results in a forward time jump in the range of hours
302 * due to the unsigned delta calculation of the time keeping core
303 * code, which is necessary to support wrapping clocksources like pm
304 * timer.
305 */
306static cycle_t read_tsc(void)
307{
308 cycle_t ret = (cycle_t)get_cycles();
309
310 return ret >= clocksource_tsc.cycle_last ?
311 ret : clocksource_tsc.cycle_last;
312}
313
314static cycle_t __vsyscall_fn vread_tsc(void)
315{
316 cycle_t ret = (cycle_t)vget_cycles();
317
318 return ret >= __vsyscall_gtod_data.clock.cycle_last ?
319 ret : __vsyscall_gtod_data.clock.cycle_last;
320}
321
322static struct clocksource clocksource_tsc = {
323 .name = "tsc",
324 .rating = 300,
325 .read = read_tsc,
326 .mask = CLOCKSOURCE_MASK(64),
327 .shift = 22,
328 .flags = CLOCK_SOURCE_IS_CONTINUOUS |
329 CLOCK_SOURCE_MUST_VERIFY,
330 .vread = vread_tsc,
331};
332
333void mark_tsc_unstable(char *reason)
334{
335 if (!tsc_unstable) {
336 tsc_unstable = 1;
337 printk("Marking TSC unstable due to %s\n", reason);
338 /* Change only the rating, when not registered */
339 if (clocksource_tsc.mult)
340 clocksource_change_rating(&clocksource_tsc, 0);
341 else
342 clocksource_tsc.rating = 0;
343 }
344}
345EXPORT_SYMBOL_GPL(mark_tsc_unstable);
346
347void __init init_tsc_clocksource(void)
348{
349 if (!notsc) {
350 clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
351 clocksource_tsc.shift);
352 if (check_tsc_unstable())
353 clocksource_tsc.rating = 0;
354
355 clocksource_register(&clocksource_tsc);
356 }
357}
diff --git a/arch/x86/kernel/visws_quirks.c b/arch/x86/kernel/visws_quirks.c
new file mode 100644
index 000000000000..41e01b145c48
--- /dev/null
+++ b/arch/x86/kernel/visws_quirks.c
@@ -0,0 +1,707 @@
1/*
2 * SGI Visual Workstation support and quirks, unmaintained.
3 *
4 * Split out from setup.c by davej@suse.de
5 *
6 * Copyright (C) 1999 Bent Hagemark, Ingo Molnar
7 *
8 * SGI Visual Workstation interrupt controller
9 *
10 * The Cobalt system ASIC in the Visual Workstation contains a "Cobalt" APIC
11 * which serves as the main interrupt controller in the system. Non-legacy
12 * hardware in the system uses this controller directly. Legacy devices
13 * are connected to the PIIX4 which in turn has its 8259(s) connected to
14 * a of the Cobalt APIC entry.
15 *
16 * 09/02/2000 - Updated for 2.4 by jbarnes@sgi.com
17 *
18 * 25/11/2002 - Updated for 2.5 by Andrey Panin <pazke@orbita1.ru>
19 */
20#include <linux/interrupt.h>
21#include <linux/module.h>
22#include <linux/init.h>
23#include <linux/smp.h>
24
25#include <asm/visws/cobalt.h>
26#include <asm/visws/piix4.h>
27#include <asm/arch_hooks.h>
28#include <asm/fixmap.h>
29#include <asm/reboot.h>
30#include <asm/setup.h>
31#include <asm/e820.h>
32#include <asm/smp.h>
33#include <asm/io.h>
34
35#include <mach_ipi.h>
36
37#include "mach_apic.h"
38
39#include <linux/init.h>
40#include <linux/smp.h>
41
42#include <linux/kernel_stat.h>
43#include <linux/interrupt.h>
44#include <linux/init.h>
45
46#include <asm/io.h>
47#include <asm/apic.h>
48#include <asm/i8259.h>
49#include <asm/irq_vectors.h>
50#include <asm/visws/cobalt.h>
51#include <asm/visws/lithium.h>
52#include <asm/visws/piix4.h>
53
54#include <linux/sched.h>
55#include <linux/kernel.h>
56#include <linux/init.h>
57#include <linux/pci.h>
58#include <linux/pci_ids.h>
59
60extern int no_broadcast;
61
62#include <asm/io.h>
63#include <asm/apic.h>
64#include <asm/arch_hooks.h>
65#include <asm/visws/cobalt.h>
66#include <asm/visws/lithium.h>
67
68char visws_board_type = -1;
69char visws_board_rev = -1;
70
71int is_visws_box(void)
72{
73 return visws_board_type >= 0;
74}
75
76static int __init visws_time_init(void)
77{
78 printk(KERN_INFO "Starting Cobalt Timer system clock\n");
79
80 /* Set the countdown value */
81 co_cpu_write(CO_CPU_TIMEVAL, CO_TIME_HZ/HZ);
82
83 /* Start the timer */
84 co_cpu_write(CO_CPU_CTRL, co_cpu_read(CO_CPU_CTRL) | CO_CTRL_TIMERUN);
85
86 /* Enable (unmask) the timer interrupt */
87 co_cpu_write(CO_CPU_CTRL, co_cpu_read(CO_CPU_CTRL) & ~CO_CTRL_TIMEMASK);
88
89 /*
90 * Zero return means the generic timer setup code will set up
91 * the standard vector:
92 */
93 return 0;
94}
95
96static int __init visws_pre_intr_init(void)
97{
98 init_VISWS_APIC_irqs();
99
100 /*
101 * We dont want ISA irqs to be set up by the generic code:
102 */
103 return 1;
104}
105
106/* Quirk for machine specific memory setup. */
107
108#define MB (1024 * 1024)
109
110unsigned long sgivwfb_mem_phys;
111unsigned long sgivwfb_mem_size;
112EXPORT_SYMBOL(sgivwfb_mem_phys);
113EXPORT_SYMBOL(sgivwfb_mem_size);
114
115long long mem_size __initdata = 0;
116
117static char * __init visws_memory_setup(void)
118{
119 long long gfx_mem_size = 8 * MB;
120
121 mem_size = boot_params.alt_mem_k;
122
123 if (!mem_size) {
124 printk(KERN_WARNING "Bootloader didn't set memory size, upgrade it !\n");
125 mem_size = 128 * MB;
126 }
127
128 /*
129 * this hardcodes the graphics memory to 8 MB
130 * it really should be sized dynamically (or at least
131 * set as a boot param)
132 */
133 if (!sgivwfb_mem_size) {
134 printk(KERN_WARNING "Defaulting to 8 MB framebuffer size\n");
135 sgivwfb_mem_size = 8 * MB;
136 }
137
138 /*
139 * Trim to nearest MB
140 */
141 sgivwfb_mem_size &= ~((1 << 20) - 1);
142 sgivwfb_mem_phys = mem_size - gfx_mem_size;
143
144 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
145 e820_add_region(HIGH_MEMORY, mem_size - sgivwfb_mem_size - HIGH_MEMORY, E820_RAM);
146 e820_add_region(sgivwfb_mem_phys, sgivwfb_mem_size, E820_RESERVED);
147
148 return "PROM";
149}
150
151static void visws_machine_emergency_restart(void)
152{
153 /*
154 * Visual Workstations restart after this
155 * register is poked on the PIIX4
156 */
157 outb(PIIX4_RESET_VAL, PIIX4_RESET_PORT);
158}
159
160static void visws_machine_power_off(void)
161{
162 unsigned short pm_status;
163/* extern unsigned int pci_bus0; */
164
165 while ((pm_status = inw(PMSTS_PORT)) & 0x100)
166 outw(pm_status, PMSTS_PORT);
167
168 outw(PM_SUSPEND_ENABLE, PMCNTRL_PORT);
169
170 mdelay(10);
171
172#define PCI_CONF1_ADDRESS(bus, devfn, reg) \
173 (0x80000000 | (bus << 16) | (devfn << 8) | (reg & ~3))
174
175/* outl(PCI_CONF1_ADDRESS(pci_bus0, SPECIAL_DEV, SPECIAL_REG), 0xCF8); */
176 outl(PIIX_SPECIAL_STOP, 0xCFC);
177}
178
179static int __init visws_get_smp_config(unsigned int early)
180{
181 /*
182 * Prevent MP-table parsing by the generic code:
183 */
184 return 1;
185}
186
187extern unsigned int __cpuinitdata maxcpus;
188
189/*
190 * The Visual Workstation is Intel MP compliant in the hardware
191 * sense, but it doesn't have a BIOS(-configuration table).
192 * No problem for Linux.
193 */
194
195static void __init MP_processor_info(struct mpc_config_processor *m)
196{
197 int ver, logical_apicid;
198 physid_mask_t apic_cpus;
199
200 if (!(m->mpc_cpuflag & CPU_ENABLED))
201 return;
202
203 logical_apicid = m->mpc_apicid;
204 printk(KERN_INFO "%sCPU #%d %u:%u APIC version %d\n",
205 m->mpc_cpuflag & CPU_BOOTPROCESSOR ? "Bootup " : "",
206 m->mpc_apicid,
207 (m->mpc_cpufeature & CPU_FAMILY_MASK) >> 8,
208 (m->mpc_cpufeature & CPU_MODEL_MASK) >> 4,
209 m->mpc_apicver);
210
211 if (m->mpc_cpuflag & CPU_BOOTPROCESSOR)
212 boot_cpu_physical_apicid = m->mpc_apicid;
213
214 ver = m->mpc_apicver;
215 if ((ver >= 0x14 && m->mpc_apicid >= 0xff) || m->mpc_apicid >= 0xf) {
216 printk(KERN_ERR "Processor #%d INVALID. (Max ID: %d).\n",
217 m->mpc_apicid, MAX_APICS);
218 return;
219 }
220
221 apic_cpus = apicid_to_cpu_present(m->mpc_apicid);
222 physids_or(phys_cpu_present_map, phys_cpu_present_map, apic_cpus);
223 /*
224 * Validate version
225 */
226 if (ver == 0x0) {
227 printk(KERN_ERR "BIOS bug, APIC version is 0 for CPU#%d! "
228 "fixing up to 0x10. (tell your hw vendor)\n",
229 m->mpc_apicid);
230 ver = 0x10;
231 }
232 apic_version[m->mpc_apicid] = ver;
233}
234
235static int __init visws_find_smp_config(unsigned int reserve)
236{
237 struct mpc_config_processor *mp = phys_to_virt(CO_CPU_TAB_PHYS);
238 unsigned short ncpus = readw(phys_to_virt(CO_CPU_NUM_PHYS));
239
240 if (ncpus > CO_CPU_MAX) {
241 printk(KERN_WARNING "find_visws_smp: got cpu count of %d at %p\n",
242 ncpus, mp);
243
244 ncpus = CO_CPU_MAX;
245 }
246
247 if (ncpus > maxcpus)
248 ncpus = maxcpus;
249
250#ifdef CONFIG_X86_LOCAL_APIC
251 smp_found_config = 1;
252#endif
253 while (ncpus--)
254 MP_processor_info(mp++);
255
256 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
257
258 return 1;
259}
260
261static int visws_trap_init(void);
262
263static struct x86_quirks visws_x86_quirks __initdata = {
264 .arch_time_init = visws_time_init,
265 .arch_pre_intr_init = visws_pre_intr_init,
266 .arch_memory_setup = visws_memory_setup,
267 .arch_intr_init = NULL,
268 .arch_trap_init = visws_trap_init,
269 .mach_get_smp_config = visws_get_smp_config,
270 .mach_find_smp_config = visws_find_smp_config,
271};
272
273void __init visws_early_detect(void)
274{
275 int raw;
276
277 visws_board_type = (char)(inb_p(PIIX_GPI_BD_REG) & PIIX_GPI_BD_REG)
278 >> PIIX_GPI_BD_SHIFT;
279
280 if (visws_board_type < 0)
281 return;
282
283 /*
284 * Install special quirks for timer, interrupt and memory setup:
285 * Fall back to generic behavior for traps:
286 * Override generic MP-table parsing:
287 */
288 x86_quirks = &visws_x86_quirks;
289
290 /*
291 * Install reboot quirks:
292 */
293 pm_power_off = visws_machine_power_off;
294 machine_ops.emergency_restart = visws_machine_emergency_restart;
295
296 /*
297 * Do not use broadcast IPIs:
298 */
299 no_broadcast = 0;
300
301#ifdef CONFIG_X86_IO_APIC
302 /*
303 * Turn off IO-APIC detection and initialization:
304 */
305 skip_ioapic_setup = 1;
306#endif
307
308 /*
309 * Get Board rev.
310 * First, we have to initialize the 307 part to allow us access
311 * to the GPIO registers. Let's map them at 0x0fc0 which is right
312 * after the PIIX4 PM section.
313 */
314 outb_p(SIO_DEV_SEL, SIO_INDEX);
315 outb_p(SIO_GP_DEV, SIO_DATA); /* Talk to GPIO regs. */
316
317 outb_p(SIO_DEV_MSB, SIO_INDEX);
318 outb_p(SIO_GP_MSB, SIO_DATA); /* MSB of GPIO base address */
319
320 outb_p(SIO_DEV_LSB, SIO_INDEX);
321 outb_p(SIO_GP_LSB, SIO_DATA); /* LSB of GPIO base address */
322
323 outb_p(SIO_DEV_ENB, SIO_INDEX);
324 outb_p(1, SIO_DATA); /* Enable GPIO registers. */
325
326 /*
327 * Now, we have to map the power management section to write
328 * a bit which enables access to the GPIO registers.
329 * What lunatic came up with this shit?
330 */
331 outb_p(SIO_DEV_SEL, SIO_INDEX);
332 outb_p(SIO_PM_DEV, SIO_DATA); /* Talk to GPIO regs. */
333
334 outb_p(SIO_DEV_MSB, SIO_INDEX);
335 outb_p(SIO_PM_MSB, SIO_DATA); /* MSB of PM base address */
336
337 outb_p(SIO_DEV_LSB, SIO_INDEX);
338 outb_p(SIO_PM_LSB, SIO_DATA); /* LSB of PM base address */
339
340 outb_p(SIO_DEV_ENB, SIO_INDEX);
341 outb_p(1, SIO_DATA); /* Enable PM registers. */
342
343 /*
344 * Now, write the PM register which enables the GPIO registers.
345 */
346 outb_p(SIO_PM_FER2, SIO_PM_INDEX);
347 outb_p(SIO_PM_GP_EN, SIO_PM_DATA);
348
349 /*
350 * Now, initialize the GPIO registers.
351 * We want them all to be inputs which is the
352 * power on default, so let's leave them alone.
353 * So, let's just read the board rev!
354 */
355 raw = inb_p(SIO_GP_DATA1);
356 raw &= 0x7f; /* 7 bits of valid board revision ID. */
357
358 if (visws_board_type == VISWS_320) {
359 if (raw < 0x6) {
360 visws_board_rev = 4;
361 } else if (raw < 0xc) {
362 visws_board_rev = 5;
363 } else {
364 visws_board_rev = 6;
365 }
366 } else if (visws_board_type == VISWS_540) {
367 visws_board_rev = 2;
368 } else {
369 visws_board_rev = raw;
370 }
371
372 printk(KERN_INFO "Silicon Graphics Visual Workstation %s (rev %d) detected\n",
373 (visws_board_type == VISWS_320 ? "320" :
374 (visws_board_type == VISWS_540 ? "540" :
375 "unknown")), visws_board_rev);
376}
377
378#define A01234 (LI_INTA_0 | LI_INTA_1 | LI_INTA_2 | LI_INTA_3 | LI_INTA_4)
379#define BCD (LI_INTB | LI_INTC | LI_INTD)
380#define ALLDEVS (A01234 | BCD)
381
382static __init void lithium_init(void)
383{
384 set_fixmap(FIX_LI_PCIA, LI_PCI_A_PHYS);
385 set_fixmap(FIX_LI_PCIB, LI_PCI_B_PHYS);
386
387 if ((li_pcia_read16(PCI_VENDOR_ID) != PCI_VENDOR_ID_SGI) ||
388 (li_pcia_read16(PCI_DEVICE_ID) != PCI_DEVICE_ID_SGI_LITHIUM)) {
389 printk(KERN_EMERG "Lithium hostbridge %c not found\n", 'A');
390/* panic("This machine is not SGI Visual Workstation 320/540"); */
391 }
392
393 if ((li_pcib_read16(PCI_VENDOR_ID) != PCI_VENDOR_ID_SGI) ||
394 (li_pcib_read16(PCI_DEVICE_ID) != PCI_DEVICE_ID_SGI_LITHIUM)) {
395 printk(KERN_EMERG "Lithium hostbridge %c not found\n", 'B');
396/* panic("This machine is not SGI Visual Workstation 320/540"); */
397 }
398
399 li_pcia_write16(LI_PCI_INTEN, ALLDEVS);
400 li_pcib_write16(LI_PCI_INTEN, ALLDEVS);
401}
402
403static __init void cobalt_init(void)
404{
405 /*
406 * On normal SMP PC this is used only with SMP, but we have to
407 * use it and set it up here to start the Cobalt clock
408 */
409 set_fixmap(FIX_APIC_BASE, APIC_DEFAULT_PHYS_BASE);
410 setup_local_APIC();
411 printk(KERN_INFO "Local APIC Version %#x, ID %#x\n",
412 (unsigned int)apic_read(APIC_LVR),
413 (unsigned int)apic_read(APIC_ID));
414
415 set_fixmap(FIX_CO_CPU, CO_CPU_PHYS);
416 set_fixmap(FIX_CO_APIC, CO_APIC_PHYS);
417 printk(KERN_INFO "Cobalt Revision %#lx, APIC ID %#lx\n",
418 co_cpu_read(CO_CPU_REV), co_apic_read(CO_APIC_ID));
419
420 /* Enable Cobalt APIC being careful to NOT change the ID! */
421 co_apic_write(CO_APIC_ID, co_apic_read(CO_APIC_ID) | CO_APIC_ENABLE);
422
423 printk(KERN_INFO "Cobalt APIC enabled: ID reg %#lx\n",
424 co_apic_read(CO_APIC_ID));
425}
426
427static int __init visws_trap_init(void)
428{
429 lithium_init();
430 cobalt_init();
431
432 return 1;
433}
434
435/*
436 * IRQ controller / APIC support:
437 */
438
439static DEFINE_SPINLOCK(cobalt_lock);
440
441/*
442 * Set the given Cobalt APIC Redirection Table entry to point
443 * to the given IDT vector/index.
444 */
445static inline void co_apic_set(int entry, int irq)
446{
447 co_apic_write(CO_APIC_LO(entry), CO_APIC_LEVEL | (irq + FIRST_EXTERNAL_VECTOR));
448 co_apic_write(CO_APIC_HI(entry), 0);
449}
450
451/*
452 * Cobalt (IO)-APIC functions to handle PCI devices.
453 */
454static inline int co_apic_ide0_hack(void)
455{
456 extern char visws_board_type;
457 extern char visws_board_rev;
458
459 if (visws_board_type == VISWS_320 && visws_board_rev == 5)
460 return 5;
461 return CO_APIC_IDE0;
462}
463
464static int is_co_apic(unsigned int irq)
465{
466 if (IS_CO_APIC(irq))
467 return CO_APIC(irq);
468
469 switch (irq) {
470 case 0: return CO_APIC_CPU;
471 case CO_IRQ_IDE0: return co_apic_ide0_hack();
472 case CO_IRQ_IDE1: return CO_APIC_IDE1;
473 default: return -1;
474 }
475}
476
477
478/*
479 * This is the SGI Cobalt (IO-)APIC:
480 */
481
482static void enable_cobalt_irq(unsigned int irq)
483{
484 co_apic_set(is_co_apic(irq), irq);
485}
486
487static void disable_cobalt_irq(unsigned int irq)
488{
489 int entry = is_co_apic(irq);
490
491 co_apic_write(CO_APIC_LO(entry), CO_APIC_MASK);
492 co_apic_read(CO_APIC_LO(entry));
493}
494
495/*
496 * "irq" really just serves to identify the device. Here is where we
497 * map this to the Cobalt APIC entry where it's physically wired.
498 * This is called via request_irq -> setup_irq -> irq_desc->startup()
499 */
500static unsigned int startup_cobalt_irq(unsigned int irq)
501{
502 unsigned long flags;
503
504 spin_lock_irqsave(&cobalt_lock, flags);
505 if ((irq_desc[irq].status & (IRQ_DISABLED | IRQ_INPROGRESS | IRQ_WAITING)))
506 irq_desc[irq].status &= ~(IRQ_DISABLED | IRQ_INPROGRESS | IRQ_WAITING);
507 enable_cobalt_irq(irq);
508 spin_unlock_irqrestore(&cobalt_lock, flags);
509 return 0;
510}
511
512static void ack_cobalt_irq(unsigned int irq)
513{
514 unsigned long flags;
515
516 spin_lock_irqsave(&cobalt_lock, flags);
517 disable_cobalt_irq(irq);
518 apic_write(APIC_EOI, APIC_EIO_ACK);
519 spin_unlock_irqrestore(&cobalt_lock, flags);
520}
521
522static void end_cobalt_irq(unsigned int irq)
523{
524 unsigned long flags;
525
526 spin_lock_irqsave(&cobalt_lock, flags);
527 if (!(irq_desc[irq].status & (IRQ_DISABLED | IRQ_INPROGRESS)))
528 enable_cobalt_irq(irq);
529 spin_unlock_irqrestore(&cobalt_lock, flags);
530}
531
532static struct irq_chip cobalt_irq_type = {
533 .typename = "Cobalt-APIC",
534 .startup = startup_cobalt_irq,
535 .shutdown = disable_cobalt_irq,
536 .enable = enable_cobalt_irq,
537 .disable = disable_cobalt_irq,
538 .ack = ack_cobalt_irq,
539 .end = end_cobalt_irq,
540};
541
542
543/*
544 * This is the PIIX4-based 8259 that is wired up indirectly to Cobalt
545 * -- not the manner expected by the code in i8259.c.
546 *
547 * there is a 'master' physical interrupt source that gets sent to
548 * the CPU. But in the chipset there are various 'virtual' interrupts
549 * waiting to be handled. We represent this to Linux through a 'master'
550 * interrupt controller type, and through a special virtual interrupt-
551 * controller. Device drivers only see the virtual interrupt sources.
552 */
553static unsigned int startup_piix4_master_irq(unsigned int irq)
554{
555 init_8259A(0);
556
557 return startup_cobalt_irq(irq);
558}
559
560static void end_piix4_master_irq(unsigned int irq)
561{
562 unsigned long flags;
563
564 spin_lock_irqsave(&cobalt_lock, flags);
565 enable_cobalt_irq(irq);
566 spin_unlock_irqrestore(&cobalt_lock, flags);
567}
568
569static struct irq_chip piix4_master_irq_type = {
570 .typename = "PIIX4-master",
571 .startup = startup_piix4_master_irq,
572 .ack = ack_cobalt_irq,
573 .end = end_piix4_master_irq,
574};
575
576
577static struct irq_chip piix4_virtual_irq_type = {
578 .typename = "PIIX4-virtual",
579 .shutdown = disable_8259A_irq,
580 .enable = enable_8259A_irq,
581 .disable = disable_8259A_irq,
582};
583
584
585/*
586 * PIIX4-8259 master/virtual functions to handle interrupt requests
587 * from legacy devices: floppy, parallel, serial, rtc.
588 *
589 * None of these get Cobalt APIC entries, neither do they have IDT
590 * entries. These interrupts are purely virtual and distributed from
591 * the 'master' interrupt source: CO_IRQ_8259.
592 *
593 * When the 8259 interrupts its handler figures out which of these
594 * devices is interrupting and dispatches to its handler.
595 *
596 * CAREFUL: devices see the 'virtual' interrupt only. Thus disable/
597 * enable_irq gets the right irq. This 'master' irq is never directly
598 * manipulated by any driver.
599 */
600static irqreturn_t piix4_master_intr(int irq, void *dev_id)
601{
602 int realirq;
603 irq_desc_t *desc;
604 unsigned long flags;
605
606 spin_lock_irqsave(&i8259A_lock, flags);
607
608 /* Find out what's interrupting in the PIIX4 master 8259 */
609 outb(0x0c, 0x20); /* OCW3 Poll command */
610 realirq = inb(0x20);
611
612 /*
613 * Bit 7 == 0 means invalid/spurious
614 */
615 if (unlikely(!(realirq & 0x80)))
616 goto out_unlock;
617
618 realirq &= 7;
619
620 if (unlikely(realirq == 2)) {
621 outb(0x0c, 0xa0);
622 realirq = inb(0xa0);
623
624 if (unlikely(!(realirq & 0x80)))
625 goto out_unlock;
626
627 realirq = (realirq & 7) + 8;
628 }
629
630 /* mask and ack interrupt */
631 cached_irq_mask |= 1 << realirq;
632 if (unlikely(realirq > 7)) {
633 inb(0xa1);
634 outb(cached_slave_mask, 0xa1);
635 outb(0x60 + (realirq & 7), 0xa0);
636 outb(0x60 + 2, 0x20);
637 } else {
638 inb(0x21);
639 outb(cached_master_mask, 0x21);
640 outb(0x60 + realirq, 0x20);
641 }
642
643 spin_unlock_irqrestore(&i8259A_lock, flags);
644
645 desc = irq_desc + realirq;
646
647 /*
648 * handle this 'virtual interrupt' as a Cobalt one now.
649 */
650 kstat_cpu(smp_processor_id()).irqs[realirq]++;
651
652 if (likely(desc->action != NULL))
653 handle_IRQ_event(realirq, desc->action);
654
655 if (!(desc->status & IRQ_DISABLED))
656 enable_8259A_irq(realirq);
657
658 return IRQ_HANDLED;
659
660out_unlock:
661 spin_unlock_irqrestore(&i8259A_lock, flags);
662 return IRQ_NONE;
663}
664
665static struct irqaction master_action = {
666 .handler = piix4_master_intr,
667 .name = "PIIX4-8259",
668};
669
670static struct irqaction cascade_action = {
671 .handler = no_action,
672 .name = "cascade",
673};
674
675
676void init_VISWS_APIC_irqs(void)
677{
678 int i;
679
680 for (i = 0; i < CO_IRQ_APIC0 + CO_APIC_LAST + 1; i++) {
681 irq_desc[i].status = IRQ_DISABLED;
682 irq_desc[i].action = 0;
683 irq_desc[i].depth = 1;
684
685 if (i == 0) {
686 irq_desc[i].chip = &cobalt_irq_type;
687 }
688 else if (i == CO_IRQ_IDE0) {
689 irq_desc[i].chip = &cobalt_irq_type;
690 }
691 else if (i == CO_IRQ_IDE1) {
692 irq_desc[i].chip = &cobalt_irq_type;
693 }
694 else if (i == CO_IRQ_8259) {
695 irq_desc[i].chip = &piix4_master_irq_type;
696 }
697 else if (i < CO_IRQ_APIC0) {
698 irq_desc[i].chip = &piix4_virtual_irq_type;
699 }
700 else if (IS_CO_APIC(i)) {
701 irq_desc[i].chip = &cobalt_irq_type;
702 }
703 }
704
705 setup_irq(CO_IRQ_8259, &master_action);
706 setup_irq(2, &cascade_action);
707}
diff --git a/arch/x86/kernel/vmi_32.c b/arch/x86/kernel/vmi_32.c
index 956f38927aa7..0a1b1a9d922d 100644
--- a/arch/x86/kernel/vmi_32.c
+++ b/arch/x86/kernel/vmi_32.c
@@ -151,7 +151,7 @@ static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
151 insns, ip); 151 insns, ip);
152 case PARAVIRT_PATCH(pv_cpu_ops.iret): 152 case PARAVIRT_PATCH(pv_cpu_ops.iret):
153 return patch_internal(VMI_CALL_IRET, len, insns, ip); 153 return patch_internal(VMI_CALL_IRET, len, insns, ip);
154 case PARAVIRT_PATCH(pv_cpu_ops.irq_enable_syscall_ret): 154 case PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit):
155 return patch_internal(VMI_CALL_SYSEXIT, len, insns, ip); 155 return patch_internal(VMI_CALL_SYSEXIT, len, insns, ip);
156 default: 156 default:
157 break; 157 break;
@@ -896,7 +896,7 @@ static inline int __init activate_vmi(void)
896 * the backend. They are performance critical anyway, so requiring 896 * the backend. They are performance critical anyway, so requiring
897 * a patch is not a big problem. 897 * a patch is not a big problem.
898 */ 898 */
899 pv_cpu_ops.irq_enable_syscall_ret = (void *)0xfeedbab0; 899 pv_cpu_ops.irq_enable_sysexit = (void *)0xfeedbab0;
900 pv_cpu_ops.iret = (void *)0xbadbab0; 900 pv_cpu_ops.iret = (void *)0xbadbab0;
901 901
902#ifdef CONFIG_SMP 902#ifdef CONFIG_SMP
@@ -906,7 +906,6 @@ static inline int __init activate_vmi(void)
906#ifdef CONFIG_X86_LOCAL_APIC 906#ifdef CONFIG_X86_LOCAL_APIC
907 para_fill(pv_apic_ops.apic_read, APICRead); 907 para_fill(pv_apic_ops.apic_read, APICRead);
908 para_fill(pv_apic_ops.apic_write, APICWrite); 908 para_fill(pv_apic_ops.apic_write, APICWrite);
909 para_fill(pv_apic_ops.apic_write_atomic, APICWrite);
910#endif 909#endif
911 910
912 /* 911 /*
@@ -932,7 +931,7 @@ static inline int __init activate_vmi(void)
932 pv_apic_ops.setup_secondary_clock = vmi_time_ap_init; 931 pv_apic_ops.setup_secondary_clock = vmi_time_ap_init;
933#endif 932#endif
934 pv_time_ops.sched_clock = vmi_sched_clock; 933 pv_time_ops.sched_clock = vmi_sched_clock;
935 pv_time_ops.get_cpu_khz = vmi_cpu_khz; 934 pv_time_ops.get_tsc_khz = vmi_tsc_khz;
936 935
937 /* We have true wallclock functions; disable CMOS clock sync */ 936 /* We have true wallclock functions; disable CMOS clock sync */
938 no_sync_cmos_clock = 1; 937 no_sync_cmos_clock = 1;
diff --git a/arch/x86/kernel/vmiclock_32.c b/arch/x86/kernel/vmiclock_32.c
index a2b030780aa9..6953859fe289 100644
--- a/arch/x86/kernel/vmiclock_32.c
+++ b/arch/x86/kernel/vmiclock_32.c
@@ -33,8 +33,7 @@
33#include <asm/apic.h> 33#include <asm/apic.h>
34#include <asm/timer.h> 34#include <asm/timer.h>
35#include <asm/i8253.h> 35#include <asm/i8253.h>
36 36#include <asm/irq_vectors.h>
37#include <irq_vectors.h>
38 37
39#define VMI_ONESHOT (VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL | vmi_get_alarm_wiring()) 38#define VMI_ONESHOT (VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
40#define VMI_PERIODIC (VMI_ALARM_IS_PERIODIC | VMI_CYCLES_REAL | vmi_get_alarm_wiring()) 39#define VMI_PERIODIC (VMI_ALARM_IS_PERIODIC | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
@@ -70,8 +69,8 @@ unsigned long long vmi_sched_clock(void)
70 return cycles_2_ns(vmi_timer_ops.get_cycle_counter(VMI_CYCLES_AVAILABLE)); 69 return cycles_2_ns(vmi_timer_ops.get_cycle_counter(VMI_CYCLES_AVAILABLE));
71} 70}
72 71
73/* paravirt_ops.get_cpu_khz = vmi_cpu_khz */ 72/* paravirt_ops.get_tsc_khz = vmi_tsc_khz */
74unsigned long vmi_cpu_khz(void) 73unsigned long vmi_tsc_khz(void)
75{ 74{
76 unsigned long long khz; 75 unsigned long long khz;
77 khz = vmi_timer_ops.get_cycle_frequency(); 76 khz = vmi_timer_ops.get_cycle_frequency();
diff --git a/arch/x86/kernel/vmlinux_32.lds.S b/arch/x86/kernel/vmlinux_32.lds.S
index ce5ed083a1e9..cdb2363697d2 100644
--- a/arch/x86/kernel/vmlinux_32.lds.S
+++ b/arch/x86/kernel/vmlinux_32.lds.S
@@ -49,23 +49,14 @@ SECTIONS
49 _etext = .; /* End of text section */ 49 _etext = .; /* End of text section */
50 } :text = 0x9090 50 } :text = 0x9090
51 51
52 NOTES :text :note
53
52 . = ALIGN(16); /* Exception table */ 54 . = ALIGN(16); /* Exception table */
53 __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) { 55 __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) {
54 __start___ex_table = .; 56 __start___ex_table = .;
55 *(__ex_table) 57 *(__ex_table)
56 __stop___ex_table = .; 58 __stop___ex_table = .;
57 } 59 } :text = 0x9090
58
59 NOTES :text :note
60
61 BUG_TABLE :text
62
63 . = ALIGN(4);
64 .tracedata : AT(ADDR(.tracedata) - LOAD_OFFSET) {
65 __tracedata_start = .;
66 *(.tracedata)
67 __tracedata_end = .;
68 }
69 60
70 RODATA 61 RODATA
71 62
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index fad3674b06a5..63e5c1a22e88 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -19,7 +19,7 @@ PHDRS {
19 data PT_LOAD FLAGS(7); /* RWE */ 19 data PT_LOAD FLAGS(7); /* RWE */
20 user PT_LOAD FLAGS(7); /* RWE */ 20 user PT_LOAD FLAGS(7); /* RWE */
21 data.init PT_LOAD FLAGS(7); /* RWE */ 21 data.init PT_LOAD FLAGS(7); /* RWE */
22 note PT_NOTE FLAGS(4); /* R__ */ 22 note PT_NOTE FLAGS(0); /* ___ */
23} 23}
24SECTIONS 24SECTIONS
25{ 25{
@@ -40,26 +40,17 @@ SECTIONS
40 _etext = .; /* End of text section */ 40 _etext = .; /* End of text section */
41 } :text = 0x9090 41 } :text = 0x9090
42 42
43 NOTES :text :note
44
43 . = ALIGN(16); /* Exception table */ 45 . = ALIGN(16); /* Exception table */
44 __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) { 46 __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) {
45 __start___ex_table = .; 47 __start___ex_table = .;
46 *(__ex_table) 48 *(__ex_table)
47 __stop___ex_table = .; 49 __stop___ex_table = .;
48 } 50 } :text = 0x9090
49
50 NOTES :text :note
51
52 BUG_TABLE :text
53 51
54 RODATA 52 RODATA
55 53
56 . = ALIGN(4);
57 .tracedata : AT(ADDR(.tracedata) - LOAD_OFFSET) {
58 __tracedata_start = .;
59 *(.tracedata)
60 __tracedata_end = .;
61 }
62
63 . = ALIGN(PAGE_SIZE); /* Align data segment to page size boundary */ 54 . = ALIGN(PAGE_SIZE); /* Align data segment to page size boundary */
64 /* Data */ 55 /* Data */
65 .data : AT(ADDR(.data) - LOAD_OFFSET) { 56 .data : AT(ADDR(.data) - LOAD_OFFSET) {
@@ -177,6 +168,7 @@ SECTIONS
177 *(.con_initcall.init) 168 *(.con_initcall.init)
178 } 169 }
179 __con_initcall_end = .; 170 __con_initcall_end = .;
171 . = ALIGN(16);
180 __x86cpuvendor_start = .; 172 __x86cpuvendor_start = .;
181 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 173 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) {
182 *(.x86cpuvendor.init) 174 *(.x86cpuvendor.init)
diff --git a/arch/x86/kernel/vsmp_64.c b/arch/x86/kernel/vsmp_64.c
index ba8c0b75ab0a..0c029e8959c7 100644
--- a/arch/x86/kernel/vsmp_64.c
+++ b/arch/x86/kernel/vsmp_64.c
@@ -15,9 +15,12 @@
15#include <linux/init.h> 15#include <linux/init.h>
16#include <linux/pci_ids.h> 16#include <linux/pci_ids.h>
17#include <linux/pci_regs.h> 17#include <linux/pci_regs.h>
18
19#include <asm/apic.h>
18#include <asm/pci-direct.h> 20#include <asm/pci-direct.h>
19#include <asm/io.h> 21#include <asm/io.h>
20#include <asm/paravirt.h> 22#include <asm/paravirt.h>
23#include <asm/setup.h>
21 24
22#if defined CONFIG_PCI && defined CONFIG_PARAVIRT 25#if defined CONFIG_PCI && defined CONFIG_PARAVIRT
23/* 26/*
diff --git a/arch/x86/kernel/vsyscall_64.c b/arch/x86/kernel/vsyscall_64.c
index 61efa2f7d564..0b8b6690a86d 100644
--- a/arch/x86/kernel/vsyscall_64.c
+++ b/arch/x86/kernel/vsyscall_64.c
@@ -42,7 +42,8 @@
42#include <asm/topology.h> 42#include <asm/topology.h>
43#include <asm/vgtod.h> 43#include <asm/vgtod.h>
44 44
45#define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr))) 45#define __vsyscall(nr) \
46 __attribute__ ((unused, __section__(".vsyscall_" #nr))) notrace
46#define __syscall_clobber "r11","cx","memory" 47#define __syscall_clobber "r11","cx","memory"
47 48
48/* 49/*
@@ -249,7 +250,7 @@ static ctl_table kernel_root_table2[] = {
249 doesn't violate that. We'll find out if it does. */ 250 doesn't violate that. We'll find out if it does. */
250static void __cpuinit vsyscall_set_cpu(int cpu) 251static void __cpuinit vsyscall_set_cpu(int cpu)
251{ 252{
252 unsigned long *d; 253 unsigned long d;
253 unsigned long node = 0; 254 unsigned long node = 0;
254#ifdef CONFIG_NUMA 255#ifdef CONFIG_NUMA
255 node = cpu_to_node(cpu); 256 node = cpu_to_node(cpu);
@@ -260,11 +261,11 @@ static void __cpuinit vsyscall_set_cpu(int cpu)
260 /* Store cpu number in limit so that it can be loaded quickly 261 /* Store cpu number in limit so that it can be loaded quickly
261 in user space in vgetcpu. 262 in user space in vgetcpu.
262 12 bits for the CPU and 8 bits for the node. */ 263 12 bits for the CPU and 8 bits for the node. */
263 d = (unsigned long *)(get_cpu_gdt_table(cpu) + GDT_ENTRY_PER_CPU); 264 d = 0x0f40000000000ULL;
264 *d = 0x0f40000000000ULL; 265 d |= cpu;
265 *d |= cpu; 266 d |= (node & 0xf) << 12;
266 *d |= (node & 0xf) << 12; 267 d |= (node >> 4) << 48;
267 *d |= (node >> 4) << 48; 268 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
268} 269}
269 270
270static void __cpuinit cpu_vsyscall_init(void *arg) 271static void __cpuinit cpu_vsyscall_init(void *arg)
@@ -278,7 +279,7 @@ cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
278{ 279{
279 long cpu = (long)arg; 280 long cpu = (long)arg;
280 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) 281 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
281 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 0, 1); 282 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
282 return NOTIFY_DONE; 283 return NOTIFY_DONE;
283} 284}
284 285
@@ -301,7 +302,7 @@ static int __init vsyscall_init(void)
301#ifdef CONFIG_SYSCTL 302#ifdef CONFIG_SYSCTL
302 register_sysctl_table(kernel_root_table2); 303 register_sysctl_table(kernel_root_table2);
303#endif 304#endif
304 on_each_cpu(cpu_vsyscall_init, NULL, 0, 1); 305 on_each_cpu(cpu_vsyscall_init, NULL, 1);
305 hotcpu_notifier(cpu_vsyscall_notifier, 0); 306 hotcpu_notifier(cpu_vsyscall_notifier, 0);
306 return 0; 307 return 0;
307} 308}
diff --git a/arch/x86/kernel/x8664_ksyms_64.c b/arch/x86/kernel/x8664_ksyms_64.c
index f6c05d0410fb..b545f371b5f5 100644
--- a/arch/x86/kernel/x8664_ksyms_64.c
+++ b/arch/x86/kernel/x8664_ksyms_64.c
@@ -2,13 +2,20 @@
2 All C exports should go in the respective C files. */ 2 All C exports should go in the respective C files. */
3 3
4#include <linux/module.h> 4#include <linux/module.h>
5#include <net/checksum.h>
6#include <linux/smp.h> 5#include <linux/smp.h>
7 6
7#include <net/checksum.h>
8
8#include <asm/processor.h> 9#include <asm/processor.h>
9#include <asm/uaccess.h>
10#include <asm/pgtable.h> 10#include <asm/pgtable.h>
11#include <asm/uaccess.h>
11#include <asm/desc.h> 12#include <asm/desc.h>
13#include <asm/ftrace.h>
14
15#ifdef CONFIG_FTRACE
16/* mcount is defined in assembly */
17EXPORT_SYMBOL(mcount);
18#endif
12 19
13EXPORT_SYMBOL(kernel_thread); 20EXPORT_SYMBOL(kernel_thread);
14 21
@@ -53,8 +60,3 @@ EXPORT_SYMBOL(init_level4_pgt);
53EXPORT_SYMBOL(load_gs_index); 60EXPORT_SYMBOL(load_gs_index);
54 61
55EXPORT_SYMBOL(_proxy_pda); 62EXPORT_SYMBOL(_proxy_pda);
56
57#ifdef CONFIG_PARAVIRT
58/* Virtualized guests may want to use it */
59EXPORT_SYMBOL_GPL(cpu_gdt_descr);
60#endif
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-04 13:52:55 -0500