aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel')
-rw-r--r--arch/x86/kernel/Makefile8
-rw-r--r--arch/x86/kernel/acpi/boot.c11
-rw-r--r--arch/x86/kernel/amd_iommu.c1
-rw-r--r--arch/x86/kernel/amd_iommu_init.c1
-rw-r--r--arch/x86/kernel/aperture_64.c5
-rw-r--r--arch/x86/kernel/apic.c143
-rw-r--r--arch/x86/kernel/apm_32.c4
-rw-r--r--arch/x86/kernel/asm-offsets_32.c2
-rw-r--r--arch/x86/kernel/asm-offsets_64.c4
-rw-r--r--arch/x86/kernel/bios_uv.c58
-rw-r--r--arch/x86/kernel/check.c161
-rw-r--r--arch/x86/kernel/cpu/Makefile6
-rw-r--r--arch/x86/kernel/cpu/addon_cpuid_features.c8
-rw-r--r--arch/x86/kernel/cpu/amd.c9
-rw-r--r--arch/x86/kernel/cpu/common.c8
-rw-r--r--arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c4
-rw-r--r--arch/x86/kernel/cpu/hypervisor.c58
-rw-r--r--arch/x86/kernel/cpu/intel.c23
-rw-r--r--arch/x86/kernel/cpu/intel_cacheinfo.c17
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_amd_64.c2
-rw-r--r--arch/x86/kernel/cpu/mcheck/mce_intel_64.c2
-rw-r--r--arch/x86/kernel/cpu/mtrr/main.c346
-rw-r--r--arch/x86/kernel/cpu/vmware.c112
-rw-r--r--arch/x86/kernel/crash.c70
-rw-r--r--arch/x86/kernel/ds.c1147
-rw-r--r--arch/x86/kernel/dumpstack.c351
-rw-r--r--arch/x86/kernel/dumpstack.h39
-rw-r--r--arch/x86/kernel/dumpstack_32.c307
-rw-r--r--arch/x86/kernel/dumpstack_64.c289
-rw-r--r--arch/x86/kernel/e820.c16
-rw-r--r--arch/x86/kernel/early-quirks.c1
-rw-r--r--arch/x86/kernel/early_printk.c47
-rw-r--r--arch/x86/kernel/entry_32.S528
-rw-r--r--arch/x86/kernel/entry_64.S1458
-rw-r--r--arch/x86/kernel/es7000_32.c62
-rw-r--r--arch/x86/kernel/ftrace.c390
-rw-r--r--arch/x86/kernel/genapic_64.c4
-rw-r--r--arch/x86/kernel/genx2apic_uv_x.c111
-rw-r--r--arch/x86/kernel/head.c1
-rw-r--r--arch/x86/kernel/head32.c3
-rw-r--r--arch/x86/kernel/head64.c3
-rw-r--r--arch/x86/kernel/hpet.c4
-rw-r--r--arch/x86/kernel/init_task.c1
-rw-r--r--arch/x86/kernel/io_apic.c3
-rw-r--r--arch/x86/kernel/irq_64.c27
-rw-r--r--arch/x86/kernel/irqinit_32.c2
-rw-r--r--arch/x86/kernel/irqinit_64.c66
-rw-r--r--arch/x86/kernel/machine_kexec_32.c104
-rw-r--r--arch/x86/kernel/microcode_amd.c232
-rw-r--r--arch/x86/kernel/microcode_core.c6
-rw-r--r--arch/x86/kernel/microcode_intel.c2
-rw-r--r--arch/x86/kernel/mpparse.c25
-rw-r--r--arch/x86/kernel/nmi.c58
-rw-r--r--arch/x86/kernel/numaq_32.c10
-rw-r--r--arch/x86/kernel/pci-dma.c11
-rw-r--r--arch/x86/kernel/process.c35
-rw-r--r--arch/x86/kernel/process_32.c67
-rw-r--r--arch/x86/kernel/process_64.c58
-rw-r--r--arch/x86/kernel/ptrace.c432
-rw-r--r--arch/x86/kernel/reboot.c126
-rw-r--r--arch/x86/kernel/relocate_kernel_32.S115
-rw-r--r--arch/x86/kernel/setup.c169
-rw-r--r--arch/x86/kernel/sigframe.h42
-rw-r--r--arch/x86/kernel/signal.c (renamed from arch/x86/kernel/signal_32.c)567
-rw-r--r--arch/x86/kernel/signal_64.c516
-rw-r--r--arch/x86/kernel/smp.c31
-rw-r--r--arch/x86/kernel/smpboot.c25
-rw-r--r--arch/x86/kernel/stacktrace.c64
-rw-r--r--arch/x86/kernel/time_32.c2
-rw-r--r--arch/x86/kernel/time_64.c6
-rw-r--r--arch/x86/kernel/tlb_32.c13
-rw-r--r--arch/x86/kernel/tlb_64.c2
-rw-r--r--arch/x86/kernel/tlb_uv.c4
-rw-r--r--arch/x86/kernel/trampoline.c19
-rw-r--r--arch/x86/kernel/traps.c38
-rw-r--r--arch/x86/kernel/tsc.c42
-rw-r--r--arch/x86/kernel/tsc_sync.c8
-rw-r--r--arch/x86/kernel/vmi_32.c119
-rw-r--r--arch/x86/kernel/vmlinux_32.lds.S1
-rw-r--r--arch/x86/kernel/vmlinux_64.lds.S1
-rw-r--r--arch/x86/kernel/vsyscall_64.c12
81 files changed, 4557 insertions, 4328 deletions
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index b62a7667828e..88dd768eab6d 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -12,6 +12,7 @@ CFLAGS_REMOVE_tsc.o = -pg
12CFLAGS_REMOVE_rtc.o = -pg 12CFLAGS_REMOVE_rtc.o = -pg
13CFLAGS_REMOVE_paravirt-spinlocks.o = -pg 13CFLAGS_REMOVE_paravirt-spinlocks.o = -pg
14CFLAGS_REMOVE_ftrace.o = -pg 14CFLAGS_REMOVE_ftrace.o = -pg
15CFLAGS_REMOVE_early_printk.o = -pg
15endif 16endif
16 17
17# 18#
@@ -23,9 +24,9 @@ CFLAGS_vsyscall_64.o := $(PROFILING) -g0 $(nostackp)
23CFLAGS_hpet.o := $(nostackp) 24CFLAGS_hpet.o := $(nostackp)
24CFLAGS_tsc.o := $(nostackp) 25CFLAGS_tsc.o := $(nostackp)
25 26
26obj-y := process_$(BITS).o signal_$(BITS).o entry_$(BITS).o 27obj-y := process_$(BITS).o signal.o entry_$(BITS).o
27obj-y += traps.o irq.o irq_$(BITS).o dumpstack_$(BITS).o 28obj-y += traps.o irq.o irq_$(BITS).o dumpstack_$(BITS).o
28obj-y += time_$(BITS).o ioport.o ldt.o 29obj-y += time_$(BITS).o ioport.o ldt.o dumpstack.o
29obj-y += setup.o i8259.o irqinit_$(BITS).o setup_percpu.o 30obj-y += setup.o i8259.o irqinit_$(BITS).o setup_percpu.o
30obj-$(CONFIG_X86_VISWS) += visws_quirks.o 31obj-$(CONFIG_X86_VISWS) += visws_quirks.o
31obj-$(CONFIG_X86_32) += probe_roms_32.o 32obj-$(CONFIG_X86_32) += probe_roms_32.o
@@ -65,6 +66,7 @@ obj-$(CONFIG_X86_LOCAL_APIC) += apic.o nmi.o
65obj-$(CONFIG_X86_IO_APIC) += io_apic.o 66obj-$(CONFIG_X86_IO_APIC) += io_apic.o
66obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o 67obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o
67obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o 68obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o
69obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
68obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o 70obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o
69obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o 71obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o
70obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o 72obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o
@@ -105,6 +107,8 @@ microcode-$(CONFIG_MICROCODE_INTEL) += microcode_intel.o
105microcode-$(CONFIG_MICROCODE_AMD) += microcode_amd.o 107microcode-$(CONFIG_MICROCODE_AMD) += microcode_amd.o
106obj-$(CONFIG_MICROCODE) += microcode.o 108obj-$(CONFIG_MICROCODE) += microcode.o
107 109
110obj-$(CONFIG_X86_CHECK_BIOS_CORRUPTION) += check.o
111
108### 112###
109# 64 bit specific files 113# 64 bit specific files
110ifeq ($(CONFIG_X86_64),y) 114ifeq ($(CONFIG_X86_64),y)
diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 4c51a2f8fd31..65d0b72777ea 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -1360,6 +1360,17 @@ static void __init acpi_process_madt(void)
1360 disable_acpi(); 1360 disable_acpi();
1361 } 1361 }
1362 } 1362 }
1363
1364 /*
1365 * ACPI supports both logical (e.g. Hyper-Threading) and physical
1366 * processors, where MPS only supports physical.
1367 */
1368 if (acpi_lapic && acpi_ioapic)
1369 printk(KERN_INFO "Using ACPI (MADT) for SMP configuration "
1370 "information\n");
1371 else if (acpi_lapic)
1372 printk(KERN_INFO "Using ACPI for processor (LAPIC) "
1373 "configuration information\n");
1363#endif 1374#endif
1364 return; 1375 return;
1365} 1376}
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 0a60d60ed036..2e2da717b350 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -24,6 +24,7 @@
24#include <linux/iommu-helper.h> 24#include <linux/iommu-helper.h>
25#include <asm/proto.h> 25#include <asm/proto.h>
26#include <asm/iommu.h> 26#include <asm/iommu.h>
27#include <asm/gart.h>
27#include <asm/amd_iommu_types.h> 28#include <asm/amd_iommu_types.h>
28#include <asm/amd_iommu.h> 29#include <asm/amd_iommu.h>
29 30
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index c6cc22815d35..c625800c55ca 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -28,6 +28,7 @@
28#include <asm/amd_iommu_types.h> 28#include <asm/amd_iommu_types.h>
29#include <asm/amd_iommu.h> 29#include <asm/amd_iommu.h>
30#include <asm/iommu.h> 30#include <asm/iommu.h>
31#include <asm/gart.h>
31 32
32/* 33/*
33 * definitions for the ACPI scanning code 34 * definitions for the ACPI scanning code
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index 9a32b37ee2ee..676debfc1702 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -1,8 +1,9 @@
1/* 1/*
2 * Firmware replacement code. 2 * Firmware replacement code.
3 * 3 *
4 * Work around broken BIOSes that don't set an aperture or only set the 4 * Work around broken BIOSes that don't set an aperture, only set the
5 * aperture in the AGP bridge. 5 * aperture in the AGP bridge, or set too small aperture.
6 *
6 * If all fails map the aperture over some low memory. This is cheaper than 7 * If all fails map the aperture over some low memory. This is cheaper than
7 * doing bounce buffering. The memory is lost. This is done at early boot 8 * doing bounce buffering. The memory is lost. This is done at early boot
8 * because only the bootmem allocator can allocate 32+MB. 9 * because only the bootmem allocator can allocate 32+MB.
diff --git a/arch/x86/kernel/apic.c b/arch/x86/kernel/apic.c
index 16f94879b525..b5229affb953 100644
--- a/arch/x86/kernel/apic.c
+++ b/arch/x86/kernel/apic.c
@@ -30,6 +30,7 @@
30#include <linux/module.h> 30#include <linux/module.h>
31#include <linux/dmi.h> 31#include <linux/dmi.h>
32#include <linux/dmar.h> 32#include <linux/dmar.h>
33#include <linux/ftrace.h>
33 34
34#include <asm/atomic.h> 35#include <asm/atomic.h>
35#include <asm/smp.h> 36#include <asm/smp.h>
@@ -441,6 +442,7 @@ static void lapic_timer_setup(enum clock_event_mode mode,
441 v = apic_read(APIC_LVTT); 442 v = apic_read(APIC_LVTT);
442 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); 443 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
443 apic_write(APIC_LVTT, v); 444 apic_write(APIC_LVTT, v);
445 apic_write(APIC_TMICT, 0xffffffff);
444 break; 446 break;
445 case CLOCK_EVT_MODE_RESUME: 447 case CLOCK_EVT_MODE_RESUME:
446 /* Nothing to do here */ 448 /* Nothing to do here */
@@ -559,13 +561,13 @@ static int __init calibrate_by_pmtimer(long deltapm, long *delta)
559 } else { 561 } else {
560 res = (((u64)deltapm) * mult) >> 22; 562 res = (((u64)deltapm) * mult) >> 22;
561 do_div(res, 1000000); 563 do_div(res, 1000000);
562 printk(KERN_WARNING "APIC calibration not consistent " 564 pr_warning("APIC calibration not consistent "
563 "with PM Timer: %ldms instead of 100ms\n", 565 "with PM Timer: %ldms instead of 100ms\n",
564 (long)res); 566 (long)res);
565 /* Correct the lapic counter value */ 567 /* Correct the lapic counter value */
566 res = (((u64)(*delta)) * pm_100ms); 568 res = (((u64)(*delta)) * pm_100ms);
567 do_div(res, deltapm); 569 do_div(res, deltapm);
568 printk(KERN_INFO "APIC delta adjusted to PM-Timer: " 570 pr_info("APIC delta adjusted to PM-Timer: "
569 "%lu (%ld)\n", (unsigned long)res, *delta); 571 "%lu (%ld)\n", (unsigned long)res, *delta);
570 *delta = (long)res; 572 *delta = (long)res;
571 } 573 }
@@ -645,8 +647,7 @@ static int __init calibrate_APIC_clock(void)
645 */ 647 */
646 if (calibration_result < (1000000 / HZ)) { 648 if (calibration_result < (1000000 / HZ)) {
647 local_irq_enable(); 649 local_irq_enable();
648 printk(KERN_WARNING 650 pr_warning("APIC frequency too slow, disabling apic timer\n");
649 "APIC frequency too slow, disabling apic timer\n");
650 return -1; 651 return -1;
651 } 652 }
652 653
@@ -672,13 +673,9 @@ static int __init calibrate_APIC_clock(void)
672 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) 673 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
673 cpu_relax(); 674 cpu_relax();
674 675
675 local_irq_disable();
676
677 /* Stop the lapic timer */ 676 /* Stop the lapic timer */
678 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt); 677 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
679 678
680 local_irq_enable();
681
682 /* Jiffies delta */ 679 /* Jiffies delta */
683 deltaj = lapic_cal_j2 - lapic_cal_j1; 680 deltaj = lapic_cal_j2 - lapic_cal_j1;
684 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj); 681 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
@@ -692,8 +689,7 @@ static int __init calibrate_APIC_clock(void)
692 local_irq_enable(); 689 local_irq_enable();
693 690
694 if (levt->features & CLOCK_EVT_FEAT_DUMMY) { 691 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
695 printk(KERN_WARNING 692 pr_warning("APIC timer disabled due to verification failure.\n");
696 "APIC timer disabled due to verification failure.\n");
697 return -1; 693 return -1;
698 } 694 }
699 695
@@ -714,7 +710,7 @@ void __init setup_boot_APIC_clock(void)
714 * broadcast mechanism is used. On UP systems simply ignore it. 710 * broadcast mechanism is used. On UP systems simply ignore it.
715 */ 711 */
716 if (disable_apic_timer) { 712 if (disable_apic_timer) {
717 printk(KERN_INFO "Disabling APIC timer\n"); 713 pr_info("Disabling APIC timer\n");
718 /* No broadcast on UP ! */ 714 /* No broadcast on UP ! */
719 if (num_possible_cpus() > 1) { 715 if (num_possible_cpus() > 1) {
720 lapic_clockevent.mult = 1; 716 lapic_clockevent.mult = 1;
@@ -741,7 +737,7 @@ void __init setup_boot_APIC_clock(void)
741 if (nmi_watchdog != NMI_IO_APIC) 737 if (nmi_watchdog != NMI_IO_APIC)
742 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; 738 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
743 else 739 else
744 printk(KERN_WARNING "APIC timer registered as dummy," 740 pr_warning("APIC timer registered as dummy,"
745 " due to nmi_watchdog=%d!\n", nmi_watchdog); 741 " due to nmi_watchdog=%d!\n", nmi_watchdog);
746 742
747 /* Setup the lapic or request the broadcast */ 743 /* Setup the lapic or request the broadcast */
@@ -773,8 +769,7 @@ static void local_apic_timer_interrupt(void)
773 * spurious. 769 * spurious.
774 */ 770 */
775 if (!evt->event_handler) { 771 if (!evt->event_handler) {
776 printk(KERN_WARNING 772 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
777 "Spurious LAPIC timer interrupt on cpu %d\n", cpu);
778 /* Switch it off */ 773 /* Switch it off */
779 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt); 774 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
780 return; 775 return;
@@ -783,11 +778,7 @@ static void local_apic_timer_interrupt(void)
783 /* 778 /*
784 * the NMI deadlock-detector uses this. 779 * the NMI deadlock-detector uses this.
785 */ 780 */
786#ifdef CONFIG_X86_64 781 inc_irq_stat(apic_timer_irqs);
787 add_pda(apic_timer_irqs, 1);
788#else
789 per_cpu(irq_stat, cpu).apic_timer_irqs++;
790#endif
791 782
792 evt->event_handler(evt); 783 evt->event_handler(evt);
793} 784}
@@ -800,7 +791,7 @@ static void local_apic_timer_interrupt(void)
800 * [ if a single-CPU system runs an SMP kernel then we call the local 791 * [ if a single-CPU system runs an SMP kernel then we call the local
801 * interrupt as well. Thus we cannot inline the local irq ... ] 792 * interrupt as well. Thus we cannot inline the local irq ... ]
802 */ 793 */
803void smp_apic_timer_interrupt(struct pt_regs *regs) 794void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
804{ 795{
805 struct pt_regs *old_regs = set_irq_regs(regs); 796 struct pt_regs *old_regs = set_irq_regs(regs);
806 797
@@ -814,9 +805,7 @@ void smp_apic_timer_interrupt(struct pt_regs *regs)
814 * Besides, if we don't timer interrupts ignore the global 805 * Besides, if we don't timer interrupts ignore the global
815 * interrupt lock, which is the WrongThing (tm) to do. 806 * interrupt lock, which is the WrongThing (tm) to do.
816 */ 807 */
817#ifdef CONFIG_X86_64
818 exit_idle(); 808 exit_idle();
819#endif
820 irq_enter(); 809 irq_enter();
821 local_apic_timer_interrupt(); 810 local_apic_timer_interrupt();
822 irq_exit(); 811 irq_exit();
@@ -1093,7 +1082,7 @@ static void __cpuinit lapic_setup_esr(void)
1093 unsigned int oldvalue, value, maxlvt; 1082 unsigned int oldvalue, value, maxlvt;
1094 1083
1095 if (!lapic_is_integrated()) { 1084 if (!lapic_is_integrated()) {
1096 printk(KERN_INFO "No ESR for 82489DX.\n"); 1085 pr_info("No ESR for 82489DX.\n");
1097 return; 1086 return;
1098 } 1087 }
1099 1088
@@ -1104,7 +1093,7 @@ static void __cpuinit lapic_setup_esr(void)
1104 * ESR disabled - we can't do anything useful with the 1093 * ESR disabled - we can't do anything useful with the
1105 * errors anyway - mbligh 1094 * errors anyway - mbligh
1106 */ 1095 */
1107 printk(KERN_INFO "Leaving ESR disabled.\n"); 1096 pr_info("Leaving ESR disabled.\n");
1108 return; 1097 return;
1109 } 1098 }
1110 1099
@@ -1298,7 +1287,7 @@ void check_x2apic(void)
1298 rdmsr(MSR_IA32_APICBASE, msr, msr2); 1287 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1299 1288
1300 if (msr & X2APIC_ENABLE) { 1289 if (msr & X2APIC_ENABLE) {
1301 printk("x2apic enabled by BIOS, switching to x2apic ops\n"); 1290 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1302 x2apic_preenabled = x2apic = 1; 1291 x2apic_preenabled = x2apic = 1;
1303 apic_ops = &x2apic_ops; 1292 apic_ops = &x2apic_ops;
1304 } 1293 }
@@ -1310,7 +1299,7 @@ void enable_x2apic(void)
1310 1299
1311 rdmsr(MSR_IA32_APICBASE, msr, msr2); 1300 rdmsr(MSR_IA32_APICBASE, msr, msr2);
1312 if (!(msr & X2APIC_ENABLE)) { 1301 if (!(msr & X2APIC_ENABLE)) {
1313 printk("Enabling x2apic\n"); 1302 pr_info("Enabling x2apic\n");
1314 wrmsr(MSR_IA32_APICBASE, msr | X2APIC_ENABLE, 0); 1303 wrmsr(MSR_IA32_APICBASE, msr | X2APIC_ENABLE, 0);
1315 } 1304 }
1316} 1305}
@@ -1325,9 +1314,8 @@ void __init enable_IR_x2apic(void)
1325 return; 1314 return;
1326 1315
1327 if (!x2apic_preenabled && disable_x2apic) { 1316 if (!x2apic_preenabled && disable_x2apic) {
1328 printk(KERN_INFO 1317 pr_info("Skipped enabling x2apic and Interrupt-remapping "
1329 "Skipped enabling x2apic and Interrupt-remapping " 1318 "because of nox2apic\n");
1330 "because of nox2apic\n");
1331 return; 1319 return;
1332 } 1320 }
1333 1321
@@ -1335,22 +1323,19 @@ void __init enable_IR_x2apic(void)
1335 panic("Bios already enabled x2apic, can't enforce nox2apic"); 1323 panic("Bios already enabled x2apic, can't enforce nox2apic");
1336 1324
1337 if (!x2apic_preenabled && skip_ioapic_setup) { 1325 if (!x2apic_preenabled && skip_ioapic_setup) {
1338 printk(KERN_INFO 1326 pr_info("Skipped enabling x2apic and Interrupt-remapping "
1339 "Skipped enabling x2apic and Interrupt-remapping " 1327 "because of skipping io-apic setup\n");
1340 "because of skipping io-apic setup\n");
1341 return; 1328 return;
1342 } 1329 }
1343 1330
1344 ret = dmar_table_init(); 1331 ret = dmar_table_init();
1345 if (ret) { 1332 if (ret) {
1346 printk(KERN_INFO 1333 pr_info("dmar_table_init() failed with %d:\n", ret);
1347 "dmar_table_init() failed with %d:\n", ret);
1348 1334
1349 if (x2apic_preenabled) 1335 if (x2apic_preenabled)
1350 panic("x2apic enabled by bios. But IR enabling failed"); 1336 panic("x2apic enabled by bios. But IR enabling failed");
1351 else 1337 else
1352 printk(KERN_INFO 1338 pr_info("Not enabling x2apic,Intr-remapping\n");
1353 "Not enabling x2apic,Intr-remapping\n");
1354 return; 1339 return;
1355 } 1340 }
1356 1341
@@ -1359,7 +1344,7 @@ void __init enable_IR_x2apic(void)
1359 1344
1360 ret = save_mask_IO_APIC_setup(); 1345 ret = save_mask_IO_APIC_setup();
1361 if (ret) { 1346 if (ret) {
1362 printk(KERN_INFO "Saving IO-APIC state failed: %d\n", ret); 1347 pr_info("Saving IO-APIC state failed: %d\n", ret);
1363 goto end; 1348 goto end;
1364 } 1349 }
1365 1350
@@ -1394,14 +1379,11 @@ end:
1394 1379
1395 if (!ret) { 1380 if (!ret) {
1396 if (!x2apic_preenabled) 1381 if (!x2apic_preenabled)
1397 printk(KERN_INFO 1382 pr_info("Enabled x2apic and interrupt-remapping\n");
1398 "Enabled x2apic and interrupt-remapping\n");
1399 else 1383 else
1400 printk(KERN_INFO 1384 pr_info("Enabled Interrupt-remapping\n");
1401 "Enabled Interrupt-remapping\n");
1402 } else 1385 } else
1403 printk(KERN_ERR 1386 pr_err("Failed to enable Interrupt-remapping and x2apic\n");
1404 "Failed to enable Interrupt-remapping and x2apic\n");
1405#else 1387#else
1406 if (!cpu_has_x2apic) 1388 if (!cpu_has_x2apic)
1407 return; 1389 return;
@@ -1410,8 +1392,8 @@ end:
1410 panic("x2apic enabled prior OS handover," 1392 panic("x2apic enabled prior OS handover,"
1411 " enable CONFIG_INTR_REMAP"); 1393 " enable CONFIG_INTR_REMAP");
1412 1394
1413 printk(KERN_INFO "Enable CONFIG_INTR_REMAP for enabling intr-remapping " 1395 pr_info("Enable CONFIG_INTR_REMAP for enabling intr-remapping "
1414 " and x2apic\n"); 1396 " and x2apic\n");
1415#endif 1397#endif
1416 1398
1417 return; 1399 return;
@@ -1428,7 +1410,7 @@ end:
1428static int __init detect_init_APIC(void) 1410static int __init detect_init_APIC(void)
1429{ 1411{
1430 if (!cpu_has_apic) { 1412 if (!cpu_has_apic) {
1431 printk(KERN_INFO "No local APIC present\n"); 1413 pr_info("No local APIC present\n");
1432 return -1; 1414 return -1;
1433 } 1415 }
1434 1416
@@ -1469,8 +1451,8 @@ static int __init detect_init_APIC(void)
1469 * "lapic" specified. 1451 * "lapic" specified.
1470 */ 1452 */
1471 if (!force_enable_local_apic) { 1453 if (!force_enable_local_apic) {
1472 printk(KERN_INFO "Local APIC disabled by BIOS -- " 1454 pr_info("Local APIC disabled by BIOS -- "
1473 "you can enable it with \"lapic\"\n"); 1455 "you can enable it with \"lapic\"\n");
1474 return -1; 1456 return -1;
1475 } 1457 }
1476 /* 1458 /*
@@ -1480,8 +1462,7 @@ static int __init detect_init_APIC(void)
1480 */ 1462 */
1481 rdmsr(MSR_IA32_APICBASE, l, h); 1463 rdmsr(MSR_IA32_APICBASE, l, h);
1482 if (!(l & MSR_IA32_APICBASE_ENABLE)) { 1464 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1483 printk(KERN_INFO 1465 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1484 "Local APIC disabled by BIOS -- reenabling.\n");
1485 l &= ~MSR_IA32_APICBASE_BASE; 1466 l &= ~MSR_IA32_APICBASE_BASE;
1486 l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE; 1467 l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE;
1487 wrmsr(MSR_IA32_APICBASE, l, h); 1468 wrmsr(MSR_IA32_APICBASE, l, h);
@@ -1494,7 +1475,7 @@ static int __init detect_init_APIC(void)
1494 */ 1475 */
1495 features = cpuid_edx(1); 1476 features = cpuid_edx(1);
1496 if (!(features & (1 << X86_FEATURE_APIC))) { 1477 if (!(features & (1 << X86_FEATURE_APIC))) {
1497 printk(KERN_WARNING "Could not enable APIC!\n"); 1478 pr_warning("Could not enable APIC!\n");
1498 return -1; 1479 return -1;
1499 } 1480 }
1500 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); 1481 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
@@ -1505,14 +1486,14 @@ static int __init detect_init_APIC(void)
1505 if (l & MSR_IA32_APICBASE_ENABLE) 1486 if (l & MSR_IA32_APICBASE_ENABLE)
1506 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE; 1487 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1507 1488
1508 printk(KERN_INFO "Found and enabled local APIC!\n"); 1489 pr_info("Found and enabled local APIC!\n");
1509 1490
1510 apic_pm_activate(); 1491 apic_pm_activate();
1511 1492
1512 return 0; 1493 return 0;
1513 1494
1514no_apic: 1495no_apic:
1515 printk(KERN_INFO "No local APIC present or hardware disabled\n"); 1496 pr_info("No local APIC present or hardware disabled\n");
1516 return -1; 1497 return -1;
1517} 1498}
1518#endif 1499#endif
@@ -1588,12 +1569,12 @@ int __init APIC_init_uniprocessor(void)
1588{ 1569{
1589#ifdef CONFIG_X86_64 1570#ifdef CONFIG_X86_64
1590 if (disable_apic) { 1571 if (disable_apic) {
1591 printk(KERN_INFO "Apic disabled\n"); 1572 pr_info("Apic disabled\n");
1592 return -1; 1573 return -1;
1593 } 1574 }
1594 if (!cpu_has_apic) { 1575 if (!cpu_has_apic) {
1595 disable_apic = 1; 1576 disable_apic = 1;
1596 printk(KERN_INFO "Apic disabled by BIOS\n"); 1577 pr_info("Apic disabled by BIOS\n");
1597 return -1; 1578 return -1;
1598 } 1579 }
1599#else 1580#else
@@ -1605,8 +1586,8 @@ int __init APIC_init_uniprocessor(void)
1605 */ 1586 */
1606 if (!cpu_has_apic && 1587 if (!cpu_has_apic &&
1607 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) { 1588 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1608 printk(KERN_ERR "BIOS bug, local APIC 0x%x not detected!...\n", 1589 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1609 boot_cpu_physical_apicid); 1590 boot_cpu_physical_apicid);
1610 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); 1591 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1611 return -1; 1592 return -1;
1612 } 1593 }
@@ -1682,9 +1663,7 @@ void smp_spurious_interrupt(struct pt_regs *regs)
1682{ 1663{
1683 u32 v; 1664 u32 v;
1684 1665
1685#ifdef CONFIG_X86_64
1686 exit_idle(); 1666 exit_idle();
1687#endif
1688 irq_enter(); 1667 irq_enter();
1689 /* 1668 /*
1690 * Check if this really is a spurious interrupt and ACK it 1669 * Check if this really is a spurious interrupt and ACK it
@@ -1695,14 +1674,11 @@ void smp_spurious_interrupt(struct pt_regs *regs)
1695 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f))) 1674 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1696 ack_APIC_irq(); 1675 ack_APIC_irq();
1697 1676
1698#ifdef CONFIG_X86_64 1677 inc_irq_stat(irq_spurious_count);
1699 add_pda(irq_spurious_count, 1); 1678
1700#else
1701 /* see sw-dev-man vol 3, chapter 7.4.13.5 */ 1679 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1702 printk(KERN_INFO "spurious APIC interrupt on CPU#%d, " 1680 pr_info("spurious APIC interrupt on CPU#%d, "
1703 "should never happen.\n", smp_processor_id()); 1681 "should never happen.\n", smp_processor_id());
1704 __get_cpu_var(irq_stat).irq_spurious_count++;
1705#endif
1706 irq_exit(); 1682 irq_exit();
1707} 1683}
1708 1684
@@ -1713,9 +1689,7 @@ void smp_error_interrupt(struct pt_regs *regs)
1713{ 1689{
1714 u32 v, v1; 1690 u32 v, v1;
1715 1691
1716#ifdef CONFIG_X86_64
1717 exit_idle(); 1692 exit_idle();
1718#endif
1719 irq_enter(); 1693 irq_enter();
1720 /* First tickle the hardware, only then report what went on. -- REW */ 1694 /* First tickle the hardware, only then report what went on. -- REW */
1721 v = apic_read(APIC_ESR); 1695 v = apic_read(APIC_ESR);
@@ -1724,17 +1698,18 @@ void smp_error_interrupt(struct pt_regs *regs)
1724 ack_APIC_irq(); 1698 ack_APIC_irq();
1725 atomic_inc(&irq_err_count); 1699 atomic_inc(&irq_err_count);
1726 1700
1727 /* Here is what the APIC error bits mean: 1701 /*
1728 0: Send CS error 1702 * Here is what the APIC error bits mean:
1729 1: Receive CS error 1703 * 0: Send CS error
1730 2: Send accept error 1704 * 1: Receive CS error
1731 3: Receive accept error 1705 * 2: Send accept error
1732 4: Reserved 1706 * 3: Receive accept error
1733 5: Send illegal vector 1707 * 4: Reserved
1734 6: Received illegal vector 1708 * 5: Send illegal vector
1735 7: Illegal register address 1709 * 6: Received illegal vector
1736 */ 1710 * 7: Illegal register address
1737 printk(KERN_DEBUG "APIC error on CPU%d: %02x(%02x)\n", 1711 */
1712 pr_debug("APIC error on CPU%d: %02x(%02x)\n",
1738 smp_processor_id(), v , v1); 1713 smp_processor_id(), v , v1);
1739 irq_exit(); 1714 irq_exit();
1740} 1715}
@@ -1838,15 +1813,15 @@ void __cpuinit generic_processor_info(int apicid, int version)
1838 * Validate version 1813 * Validate version
1839 */ 1814 */
1840 if (version == 0x0) { 1815 if (version == 0x0) {
1841 printk(KERN_WARNING "BIOS bug, APIC version is 0 for CPU#%d! " 1816 pr_warning("BIOS bug, APIC version is 0 for CPU#%d! "
1842 "fixing up to 0x10. (tell your hw vendor)\n", 1817 "fixing up to 0x10. (tell your hw vendor)\n",
1843 version); 1818 version);
1844 version = 0x10; 1819 version = 0x10;
1845 } 1820 }
1846 apic_version[apicid] = version; 1821 apic_version[apicid] = version;
1847 1822
1848 if (num_processors >= NR_CPUS) { 1823 if (num_processors >= NR_CPUS) {
1849 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached." 1824 pr_warning("WARNING: NR_CPUS limit of %i reached."
1850 " Processor ignored.\n", NR_CPUS); 1825 " Processor ignored.\n", NR_CPUS);
1851 return; 1826 return;
1852 } 1827 }
@@ -2209,7 +2184,7 @@ static int __init apic_set_verbosity(char *arg)
2209 else if (strcmp("verbose", arg) == 0) 2184 else if (strcmp("verbose", arg) == 0)
2210 apic_verbosity = APIC_VERBOSE; 2185 apic_verbosity = APIC_VERBOSE;
2211 else { 2186 else {
2212 printk(KERN_WARNING "APIC Verbosity level %s not recognised" 2187 pr_warning("APIC Verbosity level %s not recognised"
2213 " use apic=verbose or apic=debug\n", arg); 2188 " use apic=verbose or apic=debug\n", arg);
2214 return -EINVAL; 2189 return -EINVAL;
2215 } 2190 }
diff --git a/arch/x86/kernel/apm_32.c b/arch/x86/kernel/apm_32.c
index 5145a6e72bbb..3a26525a3f31 100644
--- a/arch/x86/kernel/apm_32.c
+++ b/arch/x86/kernel/apm_32.c
@@ -391,11 +391,7 @@ static int power_off;
391#else 391#else
392static int power_off = 1; 392static int power_off = 1;
393#endif 393#endif
394#ifdef CONFIG_APM_REAL_MODE_POWER_OFF
395static int realmode_power_off = 1;
396#else
397static int realmode_power_off; 394static int realmode_power_off;
398#endif
399#ifdef CONFIG_APM_ALLOW_INTS 395#ifdef CONFIG_APM_ALLOW_INTS
400static int allow_ints = 1; 396static int allow_ints = 1;
401#else 397#else
diff --git a/arch/x86/kernel/asm-offsets_32.c b/arch/x86/kernel/asm-offsets_32.c
index 6649d09ad88f..ee4df08feee6 100644
--- a/arch/x86/kernel/asm-offsets_32.c
+++ b/arch/x86/kernel/asm-offsets_32.c
@@ -11,7 +11,7 @@
11#include <linux/suspend.h> 11#include <linux/suspend.h>
12#include <linux/kbuild.h> 12#include <linux/kbuild.h>
13#include <asm/ucontext.h> 13#include <asm/ucontext.h>
14#include "sigframe.h" 14#include <asm/sigframe.h>
15#include <asm/pgtable.h> 15#include <asm/pgtable.h>
16#include <asm/fixmap.h> 16#include <asm/fixmap.h>
17#include <asm/processor.h> 17#include <asm/processor.h>
diff --git a/arch/x86/kernel/asm-offsets_64.c b/arch/x86/kernel/asm-offsets_64.c
index 7fcf63d22f8b..1d41d3f1edbc 100644
--- a/arch/x86/kernel/asm-offsets_64.c
+++ b/arch/x86/kernel/asm-offsets_64.c
@@ -20,6 +20,8 @@
20 20
21#include <xen/interface/xen.h> 21#include <xen/interface/xen.h>
22 22
23#include <asm/sigframe.h>
24
23#define __NO_STUBS 1 25#define __NO_STUBS 1
24#undef __SYSCALL 26#undef __SYSCALL
25#undef _ASM_X86_UNISTD_64_H 27#undef _ASM_X86_UNISTD_64_H
@@ -87,7 +89,7 @@ int main(void)
87 BLANK(); 89 BLANK();
88#undef ENTRY 90#undef ENTRY
89 DEFINE(IA32_RT_SIGFRAME_sigcontext, 91 DEFINE(IA32_RT_SIGFRAME_sigcontext,
90 offsetof (struct rt_sigframe32, uc.uc_mcontext)); 92 offsetof (struct rt_sigframe_ia32, uc.uc_mcontext));
91 BLANK(); 93 BLANK();
92#endif 94#endif
93 DEFINE(pbe_address, offsetof(struct pbe, address)); 95 DEFINE(pbe_address, offsetof(struct pbe, address));
diff --git a/arch/x86/kernel/bios_uv.c b/arch/x86/kernel/bios_uv.c
index f0dfe6f17e7e..2a0a2a3cac26 100644
--- a/arch/x86/kernel/bios_uv.c
+++ b/arch/x86/kernel/bios_uv.c
@@ -69,10 +69,10 @@ s64 uv_bios_call_reentrant(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3,
69 69
70long sn_partition_id; 70long sn_partition_id;
71EXPORT_SYMBOL_GPL(sn_partition_id); 71EXPORT_SYMBOL_GPL(sn_partition_id);
72long uv_coherency_id; 72long sn_coherency_id;
73EXPORT_SYMBOL_GPL(uv_coherency_id); 73EXPORT_SYMBOL_GPL(sn_coherency_id);
74long uv_region_size; 74long sn_region_size;
75EXPORT_SYMBOL_GPL(uv_region_size); 75EXPORT_SYMBOL_GPL(sn_region_size);
76int uv_type; 76int uv_type;
77 77
78 78
@@ -100,6 +100,56 @@ s64 uv_bios_get_sn_info(int fc, int *uvtype, long *partid, long *coher,
100 return ret; 100 return ret;
101} 101}
102 102
103int
104uv_bios_mq_watchlist_alloc(int blade, unsigned long addr, unsigned int mq_size,
105 unsigned long *intr_mmr_offset)
106{
107 union uv_watchlist_u size_blade;
108 u64 watchlist;
109 s64 ret;
110
111 size_blade.size = mq_size;
112 size_blade.blade = blade;
113
114 /*
115 * bios returns watchlist number or negative error number.
116 */
117 ret = (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_ALLOC, addr,
118 size_blade.val, (u64)intr_mmr_offset,
119 (u64)&watchlist, 0);
120 if (ret < BIOS_STATUS_SUCCESS)
121 return ret;
122
123 return watchlist;
124}
125EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_alloc);
126
127int
128uv_bios_mq_watchlist_free(int blade, int watchlist_num)
129{
130 return (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_FREE,
131 blade, watchlist_num, 0, 0, 0);
132}
133EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_free);
134
135s64
136uv_bios_change_memprotect(u64 paddr, u64 len, enum uv_memprotect perms)
137{
138 return uv_bios_call_irqsave(UV_BIOS_MEMPROTECT, paddr, len,
139 perms, 0, 0);
140}
141EXPORT_SYMBOL_GPL(uv_bios_change_memprotect);
142
143s64
144uv_bios_reserved_page_pa(u64 buf, u64 *cookie, u64 *addr, u64 *len)
145{
146 s64 ret;
147
148 ret = uv_bios_call_irqsave(UV_BIOS_GET_PARTITION_ADDR, (u64)cookie,
149 (u64)addr, buf, (u64)len, 0);
150 return ret;
151}
152EXPORT_SYMBOL_GPL(uv_bios_reserved_page_pa);
103 153
104s64 uv_bios_freq_base(u64 clock_type, u64 *ticks_per_second) 154s64 uv_bios_freq_base(u64 clock_type, u64 *ticks_per_second)
105{ 155{
diff --git a/arch/x86/kernel/check.c b/arch/x86/kernel/check.c
new file mode 100644
index 000000000000..2ac0ab71412a
--- /dev/null
+++ b/arch/x86/kernel/check.c
@@ -0,0 +1,161 @@
1#include <linux/module.h>
2#include <linux/sched.h>
3#include <linux/kthread.h>
4#include <linux/workqueue.h>
5#include <asm/e820.h>
6#include <asm/proto.h>
7
8/*
9 * Some BIOSes seem to corrupt the low 64k of memory during events
10 * like suspend/resume and unplugging an HDMI cable. Reserve all
11 * remaining free memory in that area and fill it with a distinct
12 * pattern.
13 */
14#define MAX_SCAN_AREAS 8
15
16static int __read_mostly memory_corruption_check = -1;
17
18static unsigned __read_mostly corruption_check_size = 64*1024;
19static unsigned __read_mostly corruption_check_period = 60; /* seconds */
20
21static struct e820entry scan_areas[MAX_SCAN_AREAS];
22static int num_scan_areas;
23
24
25static __init int set_corruption_check(char *arg)
26{
27 char *end;
28
29 memory_corruption_check = simple_strtol(arg, &end, 10);
30
31 return (*end == 0) ? 0 : -EINVAL;
32}
33early_param("memory_corruption_check", set_corruption_check);
34
35static __init int set_corruption_check_period(char *arg)
36{
37 char *end;
38
39 corruption_check_period = simple_strtoul(arg, &end, 10);
40
41 return (*end == 0) ? 0 : -EINVAL;
42}
43early_param("memory_corruption_check_period", set_corruption_check_period);
44
45static __init int set_corruption_check_size(char *arg)
46{
47 char *end;
48 unsigned size;
49
50 size = memparse(arg, &end);
51
52 if (*end == '\0')
53 corruption_check_size = size;
54
55 return (size == corruption_check_size) ? 0 : -EINVAL;
56}
57early_param("memory_corruption_check_size", set_corruption_check_size);
58
59
60void __init setup_bios_corruption_check(void)
61{
62 u64 addr = PAGE_SIZE; /* assume first page is reserved anyway */
63
64 if (memory_corruption_check == -1) {
65 memory_corruption_check =
66#ifdef CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
67 1
68#else
69 0
70#endif
71 ;
72 }
73
74 if (corruption_check_size == 0)
75 memory_corruption_check = 0;
76
77 if (!memory_corruption_check)
78 return;
79
80 corruption_check_size = round_up(corruption_check_size, PAGE_SIZE);
81
82 while (addr < corruption_check_size && num_scan_areas < MAX_SCAN_AREAS) {
83 u64 size;
84 addr = find_e820_area_size(addr, &size, PAGE_SIZE);
85
86 if (addr == 0)
87 break;
88
89 if ((addr + size) > corruption_check_size)
90 size = corruption_check_size - addr;
91
92 if (size == 0)
93 break;
94
95 e820_update_range(addr, size, E820_RAM, E820_RESERVED);
96 scan_areas[num_scan_areas].addr = addr;
97 scan_areas[num_scan_areas].size = size;
98 num_scan_areas++;
99
100 /* Assume we've already mapped this early memory */
101 memset(__va(addr), 0, size);
102
103 addr += size;
104 }
105
106 printk(KERN_INFO "Scanning %d areas for low memory corruption\n",
107 num_scan_areas);
108 update_e820();
109}
110
111
112void check_for_bios_corruption(void)
113{
114 int i;
115 int corruption = 0;
116
117 if (!memory_corruption_check)
118 return;
119
120 for (i = 0; i < num_scan_areas; i++) {
121 unsigned long *addr = __va(scan_areas[i].addr);
122 unsigned long size = scan_areas[i].size;
123
124 for (; size; addr++, size -= sizeof(unsigned long)) {
125 if (!*addr)
126 continue;
127 printk(KERN_ERR "Corrupted low memory at %p (%lx phys) = %08lx\n",
128 addr, __pa(addr), *addr);
129 corruption = 1;
130 *addr = 0;
131 }
132 }
133
134 WARN_ONCE(corruption, KERN_ERR "Memory corruption detected in low memory\n");
135}
136
137static void check_corruption(struct work_struct *dummy);
138static DECLARE_DELAYED_WORK(bios_check_work, check_corruption);
139
140static void check_corruption(struct work_struct *dummy)
141{
142 check_for_bios_corruption();
143 schedule_delayed_work(&bios_check_work,
144 round_jiffies_relative(corruption_check_period*HZ));
145}
146
147static int start_periodic_check_for_corruption(void)
148{
149 if (!memory_corruption_check || corruption_check_period == 0)
150 return 0;
151
152 printk(KERN_INFO "Scanning for low memory corruption every %d seconds\n",
153 corruption_check_period);
154
155 /* First time we run the checks right away */
156 schedule_delayed_work(&bios_check_work, 0);
157 return 0;
158}
159
160module_init(start_periodic_check_for_corruption);
161
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index 82ec6075c057..82db7f45e2de 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -2,8 +2,14 @@
2# Makefile for x86-compatible CPU details and quirks 2# Makefile for x86-compatible CPU details and quirks
3# 3#
4 4
5# Don't trace early stages of a secondary CPU boot
6ifdef CONFIG_FUNCTION_TRACER
7CFLAGS_REMOVE_common.o = -pg
8endif
9
5obj-y := intel_cacheinfo.o addon_cpuid_features.o 10obj-y := intel_cacheinfo.o addon_cpuid_features.o
6obj-y += proc.o capflags.o powerflags.o common.o 11obj-y += proc.o capflags.o powerflags.o common.o
12obj-y += vmware.o hypervisor.o
7 13
8obj-$(CONFIG_X86_32) += bugs.o cmpxchg.o 14obj-$(CONFIG_X86_32) += bugs.o cmpxchg.o
9obj-$(CONFIG_X86_64) += bugs_64.o 15obj-$(CONFIG_X86_64) += bugs_64.o
diff --git a/arch/x86/kernel/cpu/addon_cpuid_features.c b/arch/x86/kernel/cpu/addon_cpuid_features.c
index ef8f831af823..2cf23634b6d9 100644
--- a/arch/x86/kernel/cpu/addon_cpuid_features.c
+++ b/arch/x86/kernel/cpu/addon_cpuid_features.c
@@ -120,9 +120,17 @@ void __cpuinit detect_extended_topology(struct cpuinfo_x86 *c)
120 c->cpu_core_id = phys_pkg_id(c->initial_apicid, ht_mask_width) 120 c->cpu_core_id = phys_pkg_id(c->initial_apicid, ht_mask_width)
121 & core_select_mask; 121 & core_select_mask;
122 c->phys_proc_id = phys_pkg_id(c->initial_apicid, core_plus_mask_width); 122 c->phys_proc_id = phys_pkg_id(c->initial_apicid, core_plus_mask_width);
123 /*
124 * Reinit the apicid, now that we have extended initial_apicid.
125 */
126 c->apicid = phys_pkg_id(c->initial_apicid, 0);
123#else 127#else
124 c->cpu_core_id = phys_pkg_id(ht_mask_width) & core_select_mask; 128 c->cpu_core_id = phys_pkg_id(ht_mask_width) & core_select_mask;
125 c->phys_proc_id = phys_pkg_id(core_plus_mask_width); 129 c->phys_proc_id = phys_pkg_id(core_plus_mask_width);
130 /*
131 * Reinit the apicid, now that we have extended initial_apicid.
132 */
133 c->apicid = phys_pkg_id(0);
126#endif 134#endif
127 c->x86_max_cores = (core_level_siblings / smp_num_siblings); 135 c->x86_max_cores = (core_level_siblings / smp_num_siblings);
128 136
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 8f1e31db2ad5..7c878f6aa919 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -283,9 +283,14 @@ static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
283{ 283{
284 early_init_amd_mc(c); 284 early_init_amd_mc(c);
285 285
286 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */ 286 /*
287 if (c->x86_power & (1<<8)) 287 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
288 * with P/T states and does not stop in deep C-states
289 */
290 if (c->x86_power & (1 << 8)) {
288 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 291 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
292 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
293 }
289 294
290#ifdef CONFIG_X86_64 295#ifdef CONFIG_X86_64
291 set_cpu_cap(c, X86_FEATURE_SYSCALL32); 296 set_cpu_cap(c, X86_FEATURE_SYSCALL32);
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index b9c9ea0217a9..42e0853030cb 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -36,6 +36,7 @@
36#include <asm/proto.h> 36#include <asm/proto.h>
37#include <asm/sections.h> 37#include <asm/sections.h>
38#include <asm/setup.h> 38#include <asm/setup.h>
39#include <asm/hypervisor.h>
39 40
40#include "cpu.h" 41#include "cpu.h"
41 42
@@ -703,6 +704,7 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
703 detect_ht(c); 704 detect_ht(c);
704#endif 705#endif
705 706
707 init_hypervisor(c);
706 /* 708 /*
707 * On SMP, boot_cpu_data holds the common feature set between 709 * On SMP, boot_cpu_data holds the common feature set between
708 * all CPUs; so make sure that we indicate which features are 710 * all CPUs; so make sure that we indicate which features are
@@ -862,7 +864,7 @@ EXPORT_SYMBOL(_cpu_pda);
862 864
863struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table }; 865struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
864 866
865char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss; 867static char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
866 868
867void __cpuinit pda_init(int cpu) 869void __cpuinit pda_init(int cpu)
868{ 870{
@@ -903,8 +905,8 @@ void __cpuinit pda_init(int cpu)
903 } 905 }
904} 906}
905 907
906char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + 908static char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
907 DEBUG_STKSZ] __page_aligned_bss; 909 DEBUG_STKSZ] __page_aligned_bss;
908 910
909extern asmlinkage void ignore_sysret(void); 911extern asmlinkage void ignore_sysret(void);
910 912
diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index 8e48c5d4467d..88ea02dcb622 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -33,6 +33,7 @@
33#include <linux/cpufreq.h> 33#include <linux/cpufreq.h>
34#include <linux/compiler.h> 34#include <linux/compiler.h>
35#include <linux/dmi.h> 35#include <linux/dmi.h>
36#include <linux/ftrace.h>
36 37
37#include <linux/acpi.h> 38#include <linux/acpi.h>
38#include <acpi/processor.h> 39#include <acpi/processor.h>
@@ -391,6 +392,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
391 unsigned int next_perf_state = 0; /* Index into perf table */ 392 unsigned int next_perf_state = 0; /* Index into perf table */
392 unsigned int i; 393 unsigned int i;
393 int result = 0; 394 int result = 0;
395 struct power_trace it;
394 396
395 dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu); 397 dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
396 398
@@ -427,6 +429,8 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
427 } 429 }
428 } 430 }
429 431
432 trace_power_mark(&it, POWER_PSTATE, next_perf_state);
433
430 switch (data->cpu_feature) { 434 switch (data->cpu_feature) {
431 case SYSTEM_INTEL_MSR_CAPABLE: 435 case SYSTEM_INTEL_MSR_CAPABLE:
432 cmd.type = SYSTEM_INTEL_MSR_CAPABLE; 436 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
diff --git a/arch/x86/kernel/cpu/hypervisor.c b/arch/x86/kernel/cpu/hypervisor.c
new file mode 100644
index 000000000000..fb5b86af0b01
--- /dev/null
+++ b/arch/x86/kernel/cpu/hypervisor.c
@@ -0,0 +1,58 @@
1/*
2 * Common hypervisor code
3 *
4 * Copyright (C) 2008, VMware, Inc.
5 * Author : Alok N Kataria <akataria@vmware.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
15 * NON INFRINGEMENT. See the GNU General Public License for more
16 * details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
21 *
22 */
23
24#include <asm/processor.h>
25#include <asm/vmware.h>
26#include <asm/hypervisor.h>
27
28static inline void __cpuinit
29detect_hypervisor_vendor(struct cpuinfo_x86 *c)
30{
31 if (vmware_platform()) {
32 c->x86_hyper_vendor = X86_HYPER_VENDOR_VMWARE;
33 } else {
34 c->x86_hyper_vendor = X86_HYPER_VENDOR_NONE;
35 }
36}
37
38unsigned long get_hypervisor_tsc_freq(void)
39{
40 if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE)
41 return vmware_get_tsc_khz();
42 return 0;
43}
44
45static inline void __cpuinit
46hypervisor_set_feature_bits(struct cpuinfo_x86 *c)
47{
48 if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE) {
49 vmware_set_feature_bits(c);
50 return;
51 }
52}
53
54void __cpuinit init_hypervisor(struct cpuinfo_x86 *c)
55{
56 detect_hypervisor_vendor(c);
57 hypervisor_set_feature_bits(c);
58}
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index cce0b6118d55..8ea6929e974c 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -11,7 +11,6 @@
11#include <asm/pgtable.h> 11#include <asm/pgtable.h>
12#include <asm/msr.h> 12#include <asm/msr.h>
13#include <asm/uaccess.h> 13#include <asm/uaccess.h>
14#include <asm/ptrace.h>
15#include <asm/ds.h> 14#include <asm/ds.h>
16#include <asm/bugs.h> 15#include <asm/bugs.h>
17 16
@@ -41,6 +40,16 @@ static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
41 if (c->x86 == 15 && c->x86_cache_alignment == 64) 40 if (c->x86 == 15 && c->x86_cache_alignment == 64)
42 c->x86_cache_alignment = 128; 41 c->x86_cache_alignment = 128;
43#endif 42#endif
43
44 /*
45 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
46 * with P/T states and does not stop in deep C-states
47 */
48 if (c->x86_power & (1 << 8)) {
49 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
50 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
51 }
52
44} 53}
45 54
46#ifdef CONFIG_X86_32 55#ifdef CONFIG_X86_32
@@ -242,6 +251,13 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
242 251
243 intel_workarounds(c); 252 intel_workarounds(c);
244 253
254 /*
255 * Detect the extended topology information if available. This
256 * will reinitialise the initial_apicid which will be used
257 * in init_intel_cacheinfo()
258 */
259 detect_extended_topology(c);
260
245 l2 = init_intel_cacheinfo(c); 261 l2 = init_intel_cacheinfo(c);
246 if (c->cpuid_level > 9) { 262 if (c->cpuid_level > 9) {
247 unsigned eax = cpuid_eax(10); 263 unsigned eax = cpuid_eax(10);
@@ -307,13 +323,8 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
307 set_cpu_cap(c, X86_FEATURE_P4); 323 set_cpu_cap(c, X86_FEATURE_P4);
308 if (c->x86 == 6) 324 if (c->x86 == 6)
309 set_cpu_cap(c, X86_FEATURE_P3); 325 set_cpu_cap(c, X86_FEATURE_P3);
310
311 if (cpu_has_bts)
312 ptrace_bts_init_intel(c);
313
314#endif 326#endif
315 327
316 detect_extended_topology(c);
317 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) { 328 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
318 /* 329 /*
319 * let's use the legacy cpuid vector 0x1 and 0x4 for topology 330 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index 3f46afbb1cf1..68b5d8681cbb 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -644,20 +644,17 @@ static inline ssize_t show_shared_cpu_list(struct _cpuid4_info *leaf, char *buf)
644 return show_shared_cpu_map_func(leaf, 1, buf); 644 return show_shared_cpu_map_func(leaf, 1, buf);
645} 645}
646 646
647static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) { 647static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf)
648 switch(this_leaf->eax.split.type) { 648{
649 case CACHE_TYPE_DATA: 649 switch (this_leaf->eax.split.type) {
650 case CACHE_TYPE_DATA:
650 return sprintf(buf, "Data\n"); 651 return sprintf(buf, "Data\n");
651 break; 652 case CACHE_TYPE_INST:
652 case CACHE_TYPE_INST:
653 return sprintf(buf, "Instruction\n"); 653 return sprintf(buf, "Instruction\n");
654 break; 654 case CACHE_TYPE_UNIFIED:
655 case CACHE_TYPE_UNIFIED:
656 return sprintf(buf, "Unified\n"); 655 return sprintf(buf, "Unified\n");
657 break; 656 default:
658 default:
659 return sprintf(buf, "Unknown\n"); 657 return sprintf(buf, "Unknown\n");
660 break;
661 } 658 }
662} 659}
663 660
diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
index 5eb390a4b2e9..748c8f9e7a05 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c
@@ -237,7 +237,7 @@ asmlinkage void mce_threshold_interrupt(void)
237 } 237 }
238 } 238 }
239out: 239out:
240 add_pda(irq_threshold_count, 1); 240 inc_irq_stat(irq_threshold_count);
241 irq_exit(); 241 irq_exit();
242} 242}
243 243
diff --git a/arch/x86/kernel/cpu/mcheck/mce_intel_64.c b/arch/x86/kernel/cpu/mcheck/mce_intel_64.c
index c17eaf5dd6dd..4b48f251fd39 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_intel_64.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_intel_64.c
@@ -26,7 +26,7 @@ asmlinkage void smp_thermal_interrupt(void)
26 if (therm_throt_process(msr_val & 1)) 26 if (therm_throt_process(msr_val & 1))
27 mce_log_therm_throt_event(smp_processor_id(), msr_val); 27 mce_log_therm_throt_event(smp_processor_id(), msr_val);
28 28
29 add_pda(irq_thermal_count, 1); 29 inc_irq_stat(irq_thermal_count);
30 irq_exit(); 30 irq_exit();
31} 31}
32 32
diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c
index c78c04821ea1..1159e269e596 100644
--- a/arch/x86/kernel/cpu/mtrr/main.c
+++ b/arch/x86/kernel/cpu/mtrr/main.c
@@ -803,6 +803,7 @@ x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
803} 803}
804 804
805static struct res_range __initdata range[RANGE_NUM]; 805static struct res_range __initdata range[RANGE_NUM];
806static int __initdata nr_range;
806 807
807#ifdef CONFIG_MTRR_SANITIZER 808#ifdef CONFIG_MTRR_SANITIZER
808 809
@@ -1206,39 +1207,43 @@ struct mtrr_cleanup_result {
1206#define PSHIFT (PAGE_SHIFT - 10) 1207#define PSHIFT (PAGE_SHIFT - 10)
1207 1208
1208static struct mtrr_cleanup_result __initdata result[NUM_RESULT]; 1209static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
1209static struct res_range __initdata range_new[RANGE_NUM];
1210static unsigned long __initdata min_loss_pfn[RANGE_NUM]; 1210static unsigned long __initdata min_loss_pfn[RANGE_NUM];
1211 1211
1212static int __init mtrr_cleanup(unsigned address_bits) 1212static void __init print_out_mtrr_range_state(void)
1213{ 1213{
1214 unsigned long extra_remove_base, extra_remove_size;
1215 unsigned long base, size, def, dummy;
1216 mtrr_type type;
1217 int nr_range, nr_range_new;
1218 u64 chunk_size, gran_size;
1219 unsigned long range_sums, range_sums_new;
1220 int index_good;
1221 int num_reg_good;
1222 int i; 1214 int i;
1215 char start_factor = 'K', size_factor = 'K';
1216 unsigned long start_base, size_base;
1217 mtrr_type type;
1223 1218
1224 /* extra one for all 0 */ 1219 for (i = 0; i < num_var_ranges; i++) {
1225 int num[MTRR_NUM_TYPES + 1];
1226 1220
1227 if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1) 1221 size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10);
1228 return 0; 1222 if (!size_base)
1229 rdmsr(MTRRdefType_MSR, def, dummy); 1223 continue;
1230 def &= 0xff;
1231 if (def != MTRR_TYPE_UNCACHABLE)
1232 return 0;
1233 1224
1234 /* get it and store it aside */ 1225 size_base = to_size_factor(size_base, &size_factor),
1235 memset(range_state, 0, sizeof(range_state)); 1226 start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
1236 for (i = 0; i < num_var_ranges; i++) { 1227 start_base = to_size_factor(start_base, &start_factor),
1237 mtrr_if->get(i, &base, &size, &type); 1228 type = range_state[i].type;
1238 range_state[i].base_pfn = base; 1229
1239 range_state[i].size_pfn = size; 1230 printk(KERN_DEBUG "reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
1240 range_state[i].type = type; 1231 i, start_base, start_factor,
1232 size_base, size_factor,
1233 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
1234 ((type == MTRR_TYPE_WRPROT) ? "WP" :
1235 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
1236 );
1241 } 1237 }
1238}
1239
1240static int __init mtrr_need_cleanup(void)
1241{
1242 int i;
1243 mtrr_type type;
1244 unsigned long size;
1245 /* extra one for all 0 */
1246 int num[MTRR_NUM_TYPES + 1];
1242 1247
1243 /* check entries number */ 1248 /* check entries number */
1244 memset(num, 0, sizeof(num)); 1249 memset(num, 0, sizeof(num));
@@ -1263,29 +1268,133 @@ static int __init mtrr_cleanup(unsigned address_bits)
1263 num_var_ranges - num[MTRR_NUM_TYPES]) 1268 num_var_ranges - num[MTRR_NUM_TYPES])
1264 return 0; 1269 return 0;
1265 1270
1266 /* print original var MTRRs at first, for debugging: */ 1271 return 1;
1267 printk(KERN_DEBUG "original variable MTRRs\n"); 1272}
1268 for (i = 0; i < num_var_ranges; i++) {
1269 char start_factor = 'K', size_factor = 'K';
1270 unsigned long start_base, size_base;
1271 1273
1272 size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10); 1274static unsigned long __initdata range_sums;
1273 if (!size_base) 1275static void __init mtrr_calc_range_state(u64 chunk_size, u64 gran_size,
1274 continue; 1276 unsigned long extra_remove_base,
1277 unsigned long extra_remove_size,
1278 int i)
1279{
1280 int num_reg;
1281 static struct res_range range_new[RANGE_NUM];
1282 static int nr_range_new;
1283 unsigned long range_sums_new;
1284
1285 /* convert ranges to var ranges state */
1286 num_reg = x86_setup_var_mtrrs(range, nr_range,
1287 chunk_size, gran_size);
1288
1289 /* we got new setting in range_state, check it */
1290 memset(range_new, 0, sizeof(range_new));
1291 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
1292 extra_remove_base, extra_remove_size);
1293 range_sums_new = sum_ranges(range_new, nr_range_new);
1294
1295 result[i].chunk_sizek = chunk_size >> 10;
1296 result[i].gran_sizek = gran_size >> 10;
1297 result[i].num_reg = num_reg;
1298 if (range_sums < range_sums_new) {
1299 result[i].lose_cover_sizek =
1300 (range_sums_new - range_sums) << PSHIFT;
1301 result[i].bad = 1;
1302 } else
1303 result[i].lose_cover_sizek =
1304 (range_sums - range_sums_new) << PSHIFT;
1275 1305
1276 size_base = to_size_factor(size_base, &size_factor), 1306 /* double check it */
1277 start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10); 1307 if (!result[i].bad && !result[i].lose_cover_sizek) {
1278 start_base = to_size_factor(start_base, &start_factor), 1308 if (nr_range_new != nr_range ||
1279 type = range_state[i].type; 1309 memcmp(range, range_new, sizeof(range)))
1310 result[i].bad = 1;
1311 }
1280 1312
1281 printk(KERN_DEBUG "reg %d, base: %ld%cB, range: %ld%cB, type %s\n", 1313 if (!result[i].bad && (range_sums - range_sums_new <
1282 i, start_base, start_factor, 1314 min_loss_pfn[num_reg])) {
1283 size_base, size_factor, 1315 min_loss_pfn[num_reg] =
1284 (type == MTRR_TYPE_UNCACHABLE) ? "UC" : 1316 range_sums - range_sums_new;
1285 ((type == MTRR_TYPE_WRPROT) ? "WP" :
1286 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
1287 );
1288 } 1317 }
1318}
1319
1320static void __init mtrr_print_out_one_result(int i)
1321{
1322 char gran_factor, chunk_factor, lose_factor;
1323 unsigned long gran_base, chunk_base, lose_base;
1324
1325 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1326 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1327 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1328 printk(KERN_INFO "%sgran_size: %ld%c \tchunk_size: %ld%c \t",
1329 result[i].bad ? "*BAD*" : " ",
1330 gran_base, gran_factor, chunk_base, chunk_factor);
1331 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ld%c\n",
1332 result[i].num_reg, result[i].bad ? "-" : "",
1333 lose_base, lose_factor);
1334}
1335
1336static int __init mtrr_search_optimal_index(void)
1337{
1338 int i;
1339 int num_reg_good;
1340 int index_good;
1341
1342 if (nr_mtrr_spare_reg >= num_var_ranges)
1343 nr_mtrr_spare_reg = num_var_ranges - 1;
1344 num_reg_good = -1;
1345 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
1346 if (!min_loss_pfn[i])
1347 num_reg_good = i;
1348 }
1349
1350 index_good = -1;
1351 if (num_reg_good != -1) {
1352 for (i = 0; i < NUM_RESULT; i++) {
1353 if (!result[i].bad &&
1354 result[i].num_reg == num_reg_good &&
1355 !result[i].lose_cover_sizek) {
1356 index_good = i;
1357 break;
1358 }
1359 }
1360 }
1361
1362 return index_good;
1363}
1364
1365
1366static int __init mtrr_cleanup(unsigned address_bits)
1367{
1368 unsigned long extra_remove_base, extra_remove_size;
1369 unsigned long base, size, def, dummy;
1370 mtrr_type type;
1371 u64 chunk_size, gran_size;
1372 int index_good;
1373 int i;
1374
1375 if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1)
1376 return 0;
1377 rdmsr(MTRRdefType_MSR, def, dummy);
1378 def &= 0xff;
1379 if (def != MTRR_TYPE_UNCACHABLE)
1380 return 0;
1381
1382 /* get it and store it aside */
1383 memset(range_state, 0, sizeof(range_state));
1384 for (i = 0; i < num_var_ranges; i++) {
1385 mtrr_if->get(i, &base, &size, &type);
1386 range_state[i].base_pfn = base;
1387 range_state[i].size_pfn = size;
1388 range_state[i].type = type;
1389 }
1390
1391 /* check if we need handle it and can handle it */
1392 if (!mtrr_need_cleanup())
1393 return 0;
1394
1395 /* print original var MTRRs at first, for debugging: */
1396 printk(KERN_DEBUG "original variable MTRRs\n");
1397 print_out_mtrr_range_state();
1289 1398
1290 memset(range, 0, sizeof(range)); 1399 memset(range, 0, sizeof(range));
1291 extra_remove_size = 0; 1400 extra_remove_size = 0;
@@ -1309,176 +1418,64 @@ static int __init mtrr_cleanup(unsigned address_bits)
1309 range_sums >> (20 - PAGE_SHIFT)); 1418 range_sums >> (20 - PAGE_SHIFT));
1310 1419
1311 if (mtrr_chunk_size && mtrr_gran_size) { 1420 if (mtrr_chunk_size && mtrr_gran_size) {
1312 int num_reg; 1421 i = 0;
1313 char gran_factor, chunk_factor, lose_factor; 1422 mtrr_calc_range_state(mtrr_chunk_size, mtrr_gran_size,
1314 unsigned long gran_base, chunk_base, lose_base; 1423 extra_remove_base, extra_remove_size, i);
1315
1316 debug_print++;
1317 /* convert ranges to var ranges state */
1318 num_reg = x86_setup_var_mtrrs(range, nr_range, mtrr_chunk_size,
1319 mtrr_gran_size);
1320 1424
1321 /* we got new setting in range_state, check it */ 1425 mtrr_print_out_one_result(i);
1322 memset(range_new, 0, sizeof(range_new));
1323 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
1324 extra_remove_base,
1325 extra_remove_size);
1326 range_sums_new = sum_ranges(range_new, nr_range_new);
1327 1426
1328 i = 0;
1329 result[i].chunk_sizek = mtrr_chunk_size >> 10;
1330 result[i].gran_sizek = mtrr_gran_size >> 10;
1331 result[i].num_reg = num_reg;
1332 if (range_sums < range_sums_new) {
1333 result[i].lose_cover_sizek =
1334 (range_sums_new - range_sums) << PSHIFT;
1335 result[i].bad = 1;
1336 } else
1337 result[i].lose_cover_sizek =
1338 (range_sums - range_sums_new) << PSHIFT;
1339
1340 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1341 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1342 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1343 printk(KERN_INFO "%sgran_size: %ld%c \tchunk_size: %ld%c \t",
1344 result[i].bad?"*BAD*":" ",
1345 gran_base, gran_factor, chunk_base, chunk_factor);
1346 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ld%c\n",
1347 result[i].num_reg, result[i].bad?"-":"",
1348 lose_base, lose_factor);
1349 if (!result[i].bad) { 1427 if (!result[i].bad) {
1350 set_var_mtrr_all(address_bits); 1428 set_var_mtrr_all(address_bits);
1351 return 1; 1429 return 1;
1352 } 1430 }
1353 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, " 1431 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, "
1354 "will find optimal one\n"); 1432 "will find optimal one\n");
1355 debug_print--;
1356 memset(result, 0, sizeof(result[0]));
1357 } 1433 }
1358 1434
1359 i = 0; 1435 i = 0;
1360 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn)); 1436 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
1361 memset(result, 0, sizeof(result)); 1437 memset(result, 0, sizeof(result));
1362 for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) { 1438 for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) {
1363 char gran_factor;
1364 unsigned long gran_base;
1365
1366 if (debug_print)
1367 gran_base = to_size_factor(gran_size >> 10, &gran_factor);
1368 1439
1369 for (chunk_size = gran_size; chunk_size < (1ULL<<32); 1440 for (chunk_size = gran_size; chunk_size < (1ULL<<32);
1370 chunk_size <<= 1) { 1441 chunk_size <<= 1) {
1371 int num_reg;
1372 1442
1373 if (debug_print) {
1374 char chunk_factor;
1375 unsigned long chunk_base;
1376
1377 chunk_base = to_size_factor(chunk_size>>10, &chunk_factor),
1378 printk(KERN_INFO "\n");
1379 printk(KERN_INFO "gran_size: %ld%c chunk_size: %ld%c \n",
1380 gran_base, gran_factor, chunk_base, chunk_factor);
1381 }
1382 if (i >= NUM_RESULT) 1443 if (i >= NUM_RESULT)
1383 continue; 1444 continue;
1384 1445
1385 /* convert ranges to var ranges state */ 1446 mtrr_calc_range_state(chunk_size, gran_size,
1386 num_reg = x86_setup_var_mtrrs(range, nr_range, 1447 extra_remove_base, extra_remove_size, i);
1387 chunk_size, gran_size); 1448 if (debug_print) {
1388 1449 mtrr_print_out_one_result(i);
1389 /* we got new setting in range_state, check it */ 1450 printk(KERN_INFO "\n");
1390 memset(range_new, 0, sizeof(range_new));
1391 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
1392 extra_remove_base, extra_remove_size);
1393 range_sums_new = sum_ranges(range_new, nr_range_new);
1394
1395 result[i].chunk_sizek = chunk_size >> 10;
1396 result[i].gran_sizek = gran_size >> 10;
1397 result[i].num_reg = num_reg;
1398 if (range_sums < range_sums_new) {
1399 result[i].lose_cover_sizek =
1400 (range_sums_new - range_sums) << PSHIFT;
1401 result[i].bad = 1;
1402 } else
1403 result[i].lose_cover_sizek =
1404 (range_sums - range_sums_new) << PSHIFT;
1405
1406 /* double check it */
1407 if (!result[i].bad && !result[i].lose_cover_sizek) {
1408 if (nr_range_new != nr_range ||
1409 memcmp(range, range_new, sizeof(range)))
1410 result[i].bad = 1;
1411 } 1451 }
1412 1452
1413 if (!result[i].bad && (range_sums - range_sums_new <
1414 min_loss_pfn[num_reg])) {
1415 min_loss_pfn[num_reg] =
1416 range_sums - range_sums_new;
1417 }
1418 i++; 1453 i++;
1419 } 1454 }
1420 } 1455 }
1421 1456
1422 /* print out all */
1423 for (i = 0; i < NUM_RESULT; i++) {
1424 char gran_factor, chunk_factor, lose_factor;
1425 unsigned long gran_base, chunk_base, lose_base;
1426
1427 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
1428 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
1429 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1430 printk(KERN_INFO "%sgran_size: %ld%c \tchunk_size: %ld%c \t",
1431 result[i].bad?"*BAD*":" ",
1432 gran_base, gran_factor, chunk_base, chunk_factor);
1433 printk(KERN_CONT "num_reg: %d \tlose cover RAM: %s%ld%c\n",
1434 result[i].num_reg, result[i].bad?"-":"",
1435 lose_base, lose_factor);
1436 }
1437
1438 /* try to find the optimal index */ 1457 /* try to find the optimal index */
1439 if (nr_mtrr_spare_reg >= num_var_ranges) 1458 index_good = mtrr_search_optimal_index();
1440 nr_mtrr_spare_reg = num_var_ranges - 1;
1441 num_reg_good = -1;
1442 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
1443 if (!min_loss_pfn[i])
1444 num_reg_good = i;
1445 }
1446
1447 index_good = -1;
1448 if (num_reg_good != -1) {
1449 for (i = 0; i < NUM_RESULT; i++) {
1450 if (!result[i].bad &&
1451 result[i].num_reg == num_reg_good &&
1452 !result[i].lose_cover_sizek) {
1453 index_good = i;
1454 break;
1455 }
1456 }
1457 }
1458 1459
1459 if (index_good != -1) { 1460 if (index_good != -1) {
1460 char gran_factor, chunk_factor, lose_factor;
1461 unsigned long gran_base, chunk_base, lose_base;
1462
1463 printk(KERN_INFO "Found optimal setting for mtrr clean up\n"); 1461 printk(KERN_INFO "Found optimal setting for mtrr clean up\n");
1464 i = index_good; 1462 i = index_good;
1465 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor), 1463 mtrr_print_out_one_result(i);
1466 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor), 1464
1467 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
1468 printk(KERN_INFO "gran_size: %ld%c \tchunk_size: %ld%c \t",
1469 gran_base, gran_factor, chunk_base, chunk_factor);
1470 printk(KERN_CONT "num_reg: %d \tlose RAM: %ld%c\n",
1471 result[i].num_reg, lose_base, lose_factor);
1472 /* convert ranges to var ranges state */ 1465 /* convert ranges to var ranges state */
1473 chunk_size = result[i].chunk_sizek; 1466 chunk_size = result[i].chunk_sizek;
1474 chunk_size <<= 10; 1467 chunk_size <<= 10;
1475 gran_size = result[i].gran_sizek; 1468 gran_size = result[i].gran_sizek;
1476 gran_size <<= 10; 1469 gran_size <<= 10;
1477 debug_print++;
1478 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size); 1470 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
1479 debug_print--;
1480 set_var_mtrr_all(address_bits); 1471 set_var_mtrr_all(address_bits);
1472 printk(KERN_DEBUG "New variable MTRRs\n");
1473 print_out_mtrr_range_state();
1481 return 1; 1474 return 1;
1475 } else {
1476 /* print out all */
1477 for (i = 0; i < NUM_RESULT; i++)
1478 mtrr_print_out_one_result(i);
1482 } 1479 }
1483 1480
1484 printk(KERN_INFO "mtrr_cleanup: can not find optimal value\n"); 1481 printk(KERN_INFO "mtrr_cleanup: can not find optimal value\n");
@@ -1562,7 +1559,6 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
1562{ 1559{
1563 unsigned long i, base, size, highest_pfn = 0, def, dummy; 1560 unsigned long i, base, size, highest_pfn = 0, def, dummy;
1564 mtrr_type type; 1561 mtrr_type type;
1565 int nr_range;
1566 u64 total_trim_size; 1562 u64 total_trim_size;
1567 1563
1568 /* extra one for all 0 */ 1564 /* extra one for all 0 */
diff --git a/arch/x86/kernel/cpu/vmware.c b/arch/x86/kernel/cpu/vmware.c
new file mode 100644
index 000000000000..284c399e3234
--- /dev/null
+++ b/arch/x86/kernel/cpu/vmware.c
@@ -0,0 +1,112 @@
1/*
2 * VMware Detection code.
3 *
4 * Copyright (C) 2008, VMware, Inc.
5 * Author : Alok N Kataria <akataria@vmware.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
15 * NON INFRINGEMENT. See the GNU General Public License for more
16 * details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
21 *
22 */
23
24#include <linux/dmi.h>
25#include <asm/div64.h>
26#include <asm/vmware.h>
27
28#define CPUID_VMWARE_INFO_LEAF 0x40000000
29#define VMWARE_HYPERVISOR_MAGIC 0x564D5868
30#define VMWARE_HYPERVISOR_PORT 0x5658
31
32#define VMWARE_PORT_CMD_GETVERSION 10
33#define VMWARE_PORT_CMD_GETHZ 45
34
35#define VMWARE_PORT(cmd, eax, ebx, ecx, edx) \
36 __asm__("inl (%%dx)" : \
37 "=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) : \
38 "0"(VMWARE_HYPERVISOR_MAGIC), \
39 "1"(VMWARE_PORT_CMD_##cmd), \
40 "2"(VMWARE_HYPERVISOR_PORT), "3"(UINT_MAX) : \
41 "memory");
42
43static inline int __vmware_platform(void)
44{
45 uint32_t eax, ebx, ecx, edx;
46 VMWARE_PORT(GETVERSION, eax, ebx, ecx, edx);
47 return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
48}
49
50static unsigned long __vmware_get_tsc_khz(void)
51{
52 uint64_t tsc_hz;
53 uint32_t eax, ebx, ecx, edx;
54
55 VMWARE_PORT(GETHZ, eax, ebx, ecx, edx);
56
57 if (ebx == UINT_MAX)
58 return 0;
59 tsc_hz = eax | (((uint64_t)ebx) << 32);
60 do_div(tsc_hz, 1000);
61 BUG_ON(tsc_hz >> 32);
62 return tsc_hz;
63}
64
65/*
66 * While checking the dmi string infomation, just checking the product
67 * serial key should be enough, as this will always have a VMware
68 * specific string when running under VMware hypervisor.
69 */
70int vmware_platform(void)
71{
72 if (cpu_has_hypervisor) {
73 unsigned int eax, ebx, ecx, edx;
74 char hyper_vendor_id[13];
75
76 cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &ebx, &ecx, &edx);
77 memcpy(hyper_vendor_id + 0, &ebx, 4);
78 memcpy(hyper_vendor_id + 4, &ecx, 4);
79 memcpy(hyper_vendor_id + 8, &edx, 4);
80 hyper_vendor_id[12] = '\0';
81 if (!strcmp(hyper_vendor_id, "VMwareVMware"))
82 return 1;
83 } else if (dmi_available && dmi_name_in_serial("VMware") &&
84 __vmware_platform())
85 return 1;
86
87 return 0;
88}
89
90unsigned long vmware_get_tsc_khz(void)
91{
92 BUG_ON(!vmware_platform());
93 return __vmware_get_tsc_khz();
94}
95
96/*
97 * VMware hypervisor takes care of exporting a reliable TSC to the guest.
98 * Still, due to timing difference when running on virtual cpus, the TSC can
99 * be marked as unstable in some cases. For example, the TSC sync check at
100 * bootup can fail due to a marginal offset between vcpus' TSCs (though the
101 * TSCs do not drift from each other). Also, the ACPI PM timer clocksource
102 * is not suitable as a watchdog when running on a hypervisor because the
103 * kernel may miss a wrap of the counter if the vcpu is descheduled for a
104 * long time. To skip these checks at runtime we set these capability bits,
105 * so that the kernel could just trust the hypervisor with providing a
106 * reliable virtual TSC that is suitable for timekeeping.
107 */
108void __cpuinit vmware_set_feature_bits(struct cpuinfo_x86 *c)
109{
110 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
111 set_cpu_cap(c, X86_FEATURE_TSC_RELIABLE);
112}
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index 268553817909..d84a852e4cd7 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -29,34 +29,17 @@
29 29
30#include <mach_ipi.h> 30#include <mach_ipi.h>
31 31
32/* This keeps a track of which one is crashing cpu. */
33static int crashing_cpu;
34 32
35#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC) 33#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
36static atomic_t waiting_for_crash_ipi;
37 34
38static int crash_nmi_callback(struct notifier_block *self, 35static void kdump_nmi_callback(int cpu, struct die_args *args)
39 unsigned long val, void *data)
40{ 36{
41 struct pt_regs *regs; 37 struct pt_regs *regs;
42#ifdef CONFIG_X86_32 38#ifdef CONFIG_X86_32
43 struct pt_regs fixed_regs; 39 struct pt_regs fixed_regs;
44#endif 40#endif
45 int cpu;
46 41
47 if (val != DIE_NMI_IPI) 42 regs = args->regs;
48 return NOTIFY_OK;
49
50 regs = ((struct die_args *)data)->regs;
51 cpu = raw_smp_processor_id();
52
53 /* Don't do anything if this handler is invoked on crashing cpu.
54 * Otherwise, system will completely hang. Crashing cpu can get
55 * an NMI if system was initially booted with nmi_watchdog parameter.
56 */
57 if (cpu == crashing_cpu)
58 return NOTIFY_STOP;
59 local_irq_disable();
60 43
61#ifdef CONFIG_X86_32 44#ifdef CONFIG_X86_32
62 if (!user_mode_vm(regs)) { 45 if (!user_mode_vm(regs)) {
@@ -65,54 +48,19 @@ static int crash_nmi_callback(struct notifier_block *self,
65 } 48 }
66#endif 49#endif
67 crash_save_cpu(regs, cpu); 50 crash_save_cpu(regs, cpu);
68 disable_local_APIC();
69 atomic_dec(&waiting_for_crash_ipi);
70 /* Assume hlt works */
71 halt();
72 for (;;)
73 cpu_relax();
74
75 return 1;
76}
77 51
78static void smp_send_nmi_allbutself(void) 52 disable_local_APIC();
79{
80 cpumask_t mask = cpu_online_map;
81 cpu_clear(safe_smp_processor_id(), mask);
82 if (!cpus_empty(mask))
83 send_IPI_mask(mask, NMI_VECTOR);
84} 53}
85 54
86static struct notifier_block crash_nmi_nb = { 55static void kdump_nmi_shootdown_cpus(void)
87 .notifier_call = crash_nmi_callback,
88};
89
90static void nmi_shootdown_cpus(void)
91{ 56{
92 unsigned long msecs; 57 nmi_shootdown_cpus(kdump_nmi_callback);
93
94 atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
95 /* Would it be better to replace the trap vector here? */
96 if (register_die_notifier(&crash_nmi_nb))
97 return; /* return what? */
98 /* Ensure the new callback function is set before sending
99 * out the NMI
100 */
101 wmb();
102 58
103 smp_send_nmi_allbutself();
104
105 msecs = 1000; /* Wait at most a second for the other cpus to stop */
106 while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
107 mdelay(1);
108 msecs--;
109 }
110
111 /* Leave the nmi callback set */
112 disable_local_APIC(); 59 disable_local_APIC();
113} 60}
61
114#else 62#else
115static void nmi_shootdown_cpus(void) 63static void kdump_nmi_shootdown_cpus(void)
116{ 64{
117 /* There are no cpus to shootdown */ 65 /* There are no cpus to shootdown */
118} 66}
@@ -131,9 +79,7 @@ void native_machine_crash_shutdown(struct pt_regs *regs)
131 /* The kernel is broken so disable interrupts */ 79 /* The kernel is broken so disable interrupts */
132 local_irq_disable(); 80 local_irq_disable();
133 81
134 /* Make a note of crashing cpu. Will be used in NMI callback.*/ 82 kdump_nmi_shootdown_cpus();
135 crashing_cpu = safe_smp_processor_id();
136 nmi_shootdown_cpus();
137 lapic_shutdown(); 83 lapic_shutdown();
138#if defined(CONFIG_X86_IO_APIC) 84#if defined(CONFIG_X86_IO_APIC)
139 disable_IO_APIC(); 85 disable_IO_APIC();
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index a2d1176c38ee..da91701a2348 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -6,14 +6,13 @@
6 * precise-event based sampling (PEBS). 6 * precise-event based sampling (PEBS).
7 * 7 *
8 * It manages: 8 * It manages:
9 * - per-thread and per-cpu allocation of BTS and PEBS 9 * - DS and BTS hardware configuration
10 * - buffer memory allocation (optional) 10 * - buffer overflow handling (to be done)
11 * - buffer overflow handling
12 * - buffer access 11 * - buffer access
13 * 12 *
14 * It assumes: 13 * It does not do:
15 * - get_task_struct on all parameter tasks 14 * - security checking (is the caller allowed to trace the task)
16 * - current is allowed to trace parameter tasks 15 * - buffer allocation (memory accounting)
17 * 16 *
18 * 17 *
19 * Copyright (C) 2007-2008 Intel Corporation. 18 * Copyright (C) 2007-2008 Intel Corporation.
@@ -28,22 +27,69 @@
28#include <linux/slab.h> 27#include <linux/slab.h>
29#include <linux/sched.h> 28#include <linux/sched.h>
30#include <linux/mm.h> 29#include <linux/mm.h>
30#include <linux/kernel.h>
31 31
32 32
33/* 33/*
34 * The configuration for a particular DS hardware implementation. 34 * The configuration for a particular DS hardware implementation.
35 */ 35 */
36struct ds_configuration { 36struct ds_configuration {
37 /* the size of the DS structure in bytes */ 37 /* the name of the configuration */
38 unsigned char sizeof_ds; 38 const char *name;
39 /* the size of one pointer-typed field in the DS structure in bytes; 39 /* the size of one pointer-typed field in the DS structure and
40 this covers the first 8 fields related to buffer management. */ 40 in the BTS and PEBS buffers in bytes;
41 this covers the first 8 DS fields related to buffer management. */
41 unsigned char sizeof_field; 42 unsigned char sizeof_field;
42 /* the size of a BTS/PEBS record in bytes */ 43 /* the size of a BTS/PEBS record in bytes */
43 unsigned char sizeof_rec[2]; 44 unsigned char sizeof_rec[2];
45 /* a series of bit-masks to control various features indexed
46 * by enum ds_feature */
47 unsigned long ctl[dsf_ctl_max];
44}; 48};
45static struct ds_configuration ds_cfg; 49static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
46 50
51#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id())
52
53#define MAX_SIZEOF_DS (12 * 8) /* maximal size of a DS configuration */
54#define MAX_SIZEOF_BTS (3 * 8) /* maximal size of a BTS record */
55#define DS_ALIGNMENT (1 << 3) /* BTS and PEBS buffer alignment */
56
57#define BTS_CONTROL \
58 (ds_cfg.ctl[dsf_bts] | ds_cfg.ctl[dsf_bts_kernel] | ds_cfg.ctl[dsf_bts_user] |\
59 ds_cfg.ctl[dsf_bts_overflow])
60
61
62/*
63 * A BTS or PEBS tracer.
64 *
65 * This holds the configuration of the tracer and serves as a handle
66 * to identify tracers.
67 */
68struct ds_tracer {
69 /* the DS context (partially) owned by this tracer */
70 struct ds_context *context;
71 /* the buffer provided on ds_request() and its size in bytes */
72 void *buffer;
73 size_t size;
74};
75
76struct bts_tracer {
77 /* the common DS part */
78 struct ds_tracer ds;
79 /* the trace including the DS configuration */
80 struct bts_trace trace;
81 /* buffer overflow notification function */
82 bts_ovfl_callback_t ovfl;
83};
84
85struct pebs_tracer {
86 /* the common DS part */
87 struct ds_tracer ds;
88 /* the trace including the DS configuration */
89 struct pebs_trace trace;
90 /* buffer overflow notification function */
91 pebs_ovfl_callback_t ovfl;
92};
47 93
48/* 94/*
49 * Debug Store (DS) save area configuration (see Intel64 and IA32 95 * Debug Store (DS) save area configuration (see Intel64 and IA32
@@ -109,32 +155,9 @@ static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
109 155
110 156
111/* 157/*
112 * Locking is done only for allocating BTS or PEBS resources and for 158 * Locking is done only for allocating BTS or PEBS resources.
113 * guarding context and buffer memory allocation.
114 *
115 * Most functions require the current task to own the ds context part
116 * they are going to access. All the locking is done when validating
117 * access to the context.
118 */ 159 */
119static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock); 160static DEFINE_SPINLOCK(ds_lock);
120
121/*
122 * Validate that the current task is allowed to access the BTS/PEBS
123 * buffer of the parameter task.
124 *
125 * Returns 0, if access is granted; -Eerrno, otherwise.
126 */
127static inline int ds_validate_access(struct ds_context *context,
128 enum ds_qualifier qual)
129{
130 if (!context)
131 return -EPERM;
132
133 if (context->owner[qual] == current)
134 return 0;
135
136 return -EPERM;
137}
138 161
139 162
140/* 163/*
@@ -150,27 +173,32 @@ static inline int ds_validate_access(struct ds_context *context,
150 * >0 number of per-thread tracers 173 * >0 number of per-thread tracers
151 * <0 number of per-cpu tracers 174 * <0 number of per-cpu tracers
152 * 175 *
153 * The below functions to get and put tracers and to check the
154 * allocation type require the ds_lock to be held by the caller.
155 *
156 * Tracers essentially gives the number of ds contexts for a certain 176 * Tracers essentially gives the number of ds contexts for a certain
157 * type of allocation. 177 * type of allocation.
158 */ 178 */
159static long tracers; 179static atomic_t tracers = ATOMIC_INIT(0);
160 180
161static inline void get_tracer(struct task_struct *task) 181static inline void get_tracer(struct task_struct *task)
162{ 182{
163 tracers += (task ? 1 : -1); 183 if (task)
184 atomic_inc(&tracers);
185 else
186 atomic_dec(&tracers);
164} 187}
165 188
166static inline void put_tracer(struct task_struct *task) 189static inline void put_tracer(struct task_struct *task)
167{ 190{
168 tracers -= (task ? 1 : -1); 191 if (task)
192 atomic_dec(&tracers);
193 else
194 atomic_inc(&tracers);
169} 195}
170 196
171static inline int check_tracer(struct task_struct *task) 197static inline int check_tracer(struct task_struct *task)
172{ 198{
173 return (task ? (tracers >= 0) : (tracers <= 0)); 199 return task ?
200 (atomic_read(&tracers) >= 0) :
201 (atomic_read(&tracers) <= 0);
174} 202}
175 203
176 204
@@ -183,99 +211,70 @@ static inline int check_tracer(struct task_struct *task)
183 * 211 *
184 * Contexts are use-counted. They are allocated on first access and 212 * Contexts are use-counted. They are allocated on first access and
185 * deallocated when the last user puts the context. 213 * deallocated when the last user puts the context.
186 *
187 * We distinguish between an allocating and a non-allocating get of a
188 * context:
189 * - the allocating get is used for requesting BTS/PEBS resources. It
190 * requires the caller to hold the global ds_lock.
191 * - the non-allocating get is used for all other cases. A
192 * non-existing context indicates an error. It acquires and releases
193 * the ds_lock itself for obtaining the context.
194 *
195 * A context and its DS configuration are allocated and deallocated
196 * together. A context always has a DS configuration of the
197 * appropriate size.
198 */
199static DEFINE_PER_CPU(struct ds_context *, system_context);
200
201#define this_system_context per_cpu(system_context, smp_processor_id())
202
203/*
204 * Returns the pointer to the parameter task's context or to the
205 * system-wide context, if task is NULL.
206 *
207 * Increases the use count of the returned context, if not NULL.
208 */ 214 */
209static inline struct ds_context *ds_get_context(struct task_struct *task) 215struct ds_context {
210{ 216 /* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */
211 struct ds_context *context; 217 unsigned char ds[MAX_SIZEOF_DS];
212 unsigned long irq; 218 /* the owner of the BTS and PEBS configuration, respectively */
219 struct bts_tracer *bts_master;
220 struct pebs_tracer *pebs_master;
221 /* use count */
222 unsigned long count;
223 /* a pointer to the context location inside the thread_struct
224 * or the per_cpu context array */
225 struct ds_context **this;
226 /* a pointer to the task owning this context, or NULL, if the
227 * context is owned by a cpu */
228 struct task_struct *task;
229};
213 230
214 spin_lock_irqsave(&ds_lock, irq); 231static DEFINE_PER_CPU(struct ds_context *, system_context_array);
215 232
216 context = (task ? task->thread.ds_ctx : this_system_context); 233#define system_context per_cpu(system_context_array, smp_processor_id())
217 if (context)
218 context->count++;
219 234
220 spin_unlock_irqrestore(&ds_lock, irq);
221
222 return context;
223}
224 235
225/* 236static inline struct ds_context *ds_get_context(struct task_struct *task)
226 * Same as ds_get_context, but allocates the context and it's DS
227 * structure, if necessary; returns NULL; if out of memory.
228 */
229static inline struct ds_context *ds_alloc_context(struct task_struct *task)
230{ 237{
231 struct ds_context **p_context = 238 struct ds_context **p_context =
232 (task ? &task->thread.ds_ctx : &this_system_context); 239 (task ? &task->thread.ds_ctx : &system_context);
233 struct ds_context *context = *p_context; 240 struct ds_context *context = NULL;
241 struct ds_context *new_context = NULL;
234 unsigned long irq; 242 unsigned long irq;
235 243
236 if (!context) { 244 /* Chances are small that we already have a context. */
237 context = kzalloc(sizeof(*context), GFP_KERNEL); 245 new_context = kzalloc(sizeof(*new_context), GFP_KERNEL);
238 if (!context) 246 if (!new_context)
239 return NULL; 247 return NULL;
240
241 context->ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL);
242 if (!context->ds) {
243 kfree(context);
244 return NULL;
245 }
246 248
247 spin_lock_irqsave(&ds_lock, irq); 249 spin_lock_irqsave(&ds_lock, irq);
248 250
249 if (*p_context) { 251 context = *p_context;
250 kfree(context->ds); 252 if (!context) {
251 kfree(context); 253 context = new_context;
252 254
253 context = *p_context; 255 context->this = p_context;
254 } else { 256 context->task = task;
255 *p_context = context; 257 context->count = 0;
256 258
257 context->this = p_context; 259 if (task)
258 context->task = task; 260 set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
259 261
260 if (task) 262 if (!task || (task == current))
261 set_tsk_thread_flag(task, TIF_DS_AREA_MSR); 263 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
262 264
263 if (!task || (task == current)) 265 *p_context = context;
264 wrmsrl(MSR_IA32_DS_AREA,
265 (unsigned long)context->ds);
266 }
267 spin_unlock_irqrestore(&ds_lock, irq);
268 } 266 }
269 267
270 context->count++; 268 context->count++;
271 269
270 spin_unlock_irqrestore(&ds_lock, irq);
271
272 if (context != new_context)
273 kfree(new_context);
274
272 return context; 275 return context;
273} 276}
274 277
275/*
276 * Decreases the use count of the parameter context, if not NULL.
277 * Deallocates the context, if the use count reaches zero.
278 */
279static inline void ds_put_context(struct ds_context *context) 278static inline void ds_put_context(struct ds_context *context)
280{ 279{
281 unsigned long irq; 280 unsigned long irq;
@@ -285,8 +284,10 @@ static inline void ds_put_context(struct ds_context *context)
285 284
286 spin_lock_irqsave(&ds_lock, irq); 285 spin_lock_irqsave(&ds_lock, irq);
287 286
288 if (--context->count) 287 if (--context->count) {
289 goto out; 288 spin_unlock_irqrestore(&ds_lock, irq);
289 return;
290 }
290 291
291 *(context->this) = NULL; 292 *(context->this) = NULL;
292 293
@@ -296,135 +297,263 @@ static inline void ds_put_context(struct ds_context *context)
296 if (!context->task || (context->task == current)) 297 if (!context->task || (context->task == current))
297 wrmsrl(MSR_IA32_DS_AREA, 0); 298 wrmsrl(MSR_IA32_DS_AREA, 0);
298 299
299 put_tracer(context->task); 300 spin_unlock_irqrestore(&ds_lock, irq);
300 301
301 /* free any leftover buffers from tracers that did not
302 * deallocate them properly. */
303 kfree(context->buffer[ds_bts]);
304 kfree(context->buffer[ds_pebs]);
305 kfree(context->ds);
306 kfree(context); 302 kfree(context);
307 out:
308 spin_unlock_irqrestore(&ds_lock, irq);
309} 303}
310 304
311 305
312/* 306/*
313 * Handle a buffer overflow 307 * Call the tracer's callback on a buffer overflow.
314 * 308 *
315 * task: the task whose buffers are overflowing;
316 * NULL for a buffer overflow on the current cpu
317 * context: the ds context 309 * context: the ds context
318 * qual: the buffer type 310 * qual: the buffer type
319 */ 311 */
320static void ds_overflow(struct task_struct *task, struct ds_context *context, 312static void ds_overflow(struct ds_context *context, enum ds_qualifier qual)
321 enum ds_qualifier qual)
322{ 313{
323 if (!context) 314 switch (qual) {
324 return; 315 case ds_bts:
325 316 if (context->bts_master &&
326 if (context->callback[qual]) 317 context->bts_master->ovfl)
327 (*context->callback[qual])(task); 318 context->bts_master->ovfl(context->bts_master);
328 319 break;
329 /* todo: do some more overflow handling */ 320 case ds_pebs:
321 if (context->pebs_master &&
322 context->pebs_master->ovfl)
323 context->pebs_master->ovfl(context->pebs_master);
324 break;
325 }
330} 326}
331 327
332 328
333/* 329/*
334 * Allocate a non-pageable buffer of the parameter size. 330 * Write raw data into the BTS or PEBS buffer.
335 * Checks the memory and the locked memory rlimit.
336 * 331 *
337 * Returns the buffer, if successful; 332 * The remainder of any partially written record is zeroed out.
338 * NULL, if out of memory or rlimit exceeded.
339 * 333 *
340 * size: the requested buffer size in bytes 334 * context: the DS context
341 * pages (out): if not NULL, contains the number of pages reserved 335 * qual: the buffer type
336 * record: the data to write
337 * size: the size of the data
342 */ 338 */
343static inline void *ds_allocate_buffer(size_t size, unsigned int *pages) 339static int ds_write(struct ds_context *context, enum ds_qualifier qual,
340 const void *record, size_t size)
344{ 341{
345 unsigned long rlim, vm, pgsz; 342 int bytes_written = 0;
346 void *buffer;
347 343
348 pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT; 344 if (!record)
345 return -EINVAL;
349 346
350 rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; 347 while (size) {
351 vm = current->mm->total_vm + pgsz; 348 unsigned long base, index, end, write_end, int_th;
352 if (rlim < vm) 349 unsigned long write_size, adj_write_size;
353 return NULL;
354 350
355 rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; 351 /*
356 vm = current->mm->locked_vm + pgsz; 352 * write as much as possible without producing an
357 if (rlim < vm) 353 * overflow interrupt.
358 return NULL; 354 *
355 * interrupt_threshold must either be
356 * - bigger than absolute_maximum or
357 * - point to a record between buffer_base and absolute_maximum
358 *
359 * index points to a valid record.
360 */
361 base = ds_get(context->ds, qual, ds_buffer_base);
362 index = ds_get(context->ds, qual, ds_index);
363 end = ds_get(context->ds, qual, ds_absolute_maximum);
364 int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
359 365
360 buffer = kzalloc(size, GFP_KERNEL); 366 write_end = min(end, int_th);
361 if (!buffer)
362 return NULL;
363 367
364 current->mm->total_vm += pgsz; 368 /* if we are already beyond the interrupt threshold,
365 current->mm->locked_vm += pgsz; 369 * we fill the entire buffer */
370 if (write_end <= index)
371 write_end = end;
366 372
367 if (pages) 373 if (write_end <= index)
368 *pages = pgsz; 374 break;
375
376 write_size = min((unsigned long) size, write_end - index);
377 memcpy((void *)index, record, write_size);
369 378
370 return buffer; 379 record = (const char *)record + write_size;
380 size -= write_size;
381 bytes_written += write_size;
382
383 adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
384 adj_write_size *= ds_cfg.sizeof_rec[qual];
385
386 /* zero out trailing bytes */
387 memset((char *)index + write_size, 0,
388 adj_write_size - write_size);
389 index += adj_write_size;
390
391 if (index >= end)
392 index = base;
393 ds_set(context->ds, qual, ds_index, index);
394
395 if (index >= int_th)
396 ds_overflow(context, qual);
397 }
398
399 return bytes_written;
371} 400}
372 401
373static int ds_request(struct task_struct *task, void *base, size_t size, 402
374 ds_ovfl_callback_t ovfl, enum ds_qualifier qual) 403/*
404 * Branch Trace Store (BTS) uses the following format. Different
405 * architectures vary in the size of those fields.
406 * - source linear address
407 * - destination linear address
408 * - flags
409 *
410 * Later architectures use 64bit pointers throughout, whereas earlier
411 * architectures use 32bit pointers in 32bit mode.
412 *
413 * We compute the base address for the first 8 fields based on:
414 * - the field size stored in the DS configuration
415 * - the relative field position
416 *
417 * In order to store additional information in the BTS buffer, we use
418 * a special source address to indicate that the record requires
419 * special interpretation.
420 *
421 * Netburst indicated via a bit in the flags field whether the branch
422 * was predicted; this is ignored.
423 *
424 * We use two levels of abstraction:
425 * - the raw data level defined here
426 * - an arch-independent level defined in ds.h
427 */
428
429enum bts_field {
430 bts_from,
431 bts_to,
432 bts_flags,
433
434 bts_qual = bts_from,
435 bts_jiffies = bts_to,
436 bts_pid = bts_flags,
437
438 bts_qual_mask = (bts_qual_max - 1),
439 bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
440};
441
442static inline unsigned long bts_get(const char *base, enum bts_field field)
375{ 443{
376 struct ds_context *context; 444 base += (ds_cfg.sizeof_field * field);
377 unsigned long buffer, adj; 445 return *(unsigned long *)base;
378 const unsigned long alignment = (1 << 3); 446}
379 unsigned long irq; 447
380 int error = 0; 448static inline void bts_set(char *base, enum bts_field field, unsigned long val)
449{
450 base += (ds_cfg.sizeof_field * field);;
451 (*(unsigned long *)base) = val;
452}
381 453
382 if (!ds_cfg.sizeof_ds)
383 return -EOPNOTSUPP;
384 454
385 /* we require some space to do alignment adjustments below */ 455/*
386 if (size < (alignment + ds_cfg.sizeof_rec[qual])) 456 * The raw BTS data is architecture dependent.
457 *
458 * For higher-level users, we give an arch-independent view.
459 * - ds.h defines struct bts_struct
460 * - bts_read translates one raw bts record into a bts_struct
461 * - bts_write translates one bts_struct into the raw format and
462 * writes it into the top of the parameter tracer's buffer.
463 *
464 * return: bytes read/written on success; -Eerrno, otherwise
465 */
466static int bts_read(struct bts_tracer *tracer, const void *at,
467 struct bts_struct *out)
468{
469 if (!tracer)
387 return -EINVAL; 470 return -EINVAL;
388 471
389 /* buffer overflow notification is not yet implemented */ 472 if (at < tracer->trace.ds.begin)
390 if (ovfl) 473 return -EINVAL;
391 return -EOPNOTSUPP;
392 474
475 if (tracer->trace.ds.end < (at + tracer->trace.ds.size))
476 return -EINVAL;
393 477
394 context = ds_alloc_context(task); 478 memset(out, 0, sizeof(*out));
395 if (!context) 479 if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) {
396 return -ENOMEM; 480 out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask);
481 out->variant.timestamp.jiffies = bts_get(at, bts_jiffies);
482 out->variant.timestamp.pid = bts_get(at, bts_pid);
483 } else {
484 out->qualifier = bts_branch;
485 out->variant.lbr.from = bts_get(at, bts_from);
486 out->variant.lbr.to = bts_get(at, bts_to);
487
488 if (!out->variant.lbr.from && !out->variant.lbr.to)
489 out->qualifier = bts_invalid;
490 }
397 491
398 spin_lock_irqsave(&ds_lock, irq); 492 return ds_cfg.sizeof_rec[ds_bts];
493}
399 494
400 error = -EPERM; 495static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
401 if (!check_tracer(task)) 496{
402 goto out_unlock; 497 unsigned char raw[MAX_SIZEOF_BTS];
403 498
404 get_tracer(task); 499 if (!tracer)
500 return -EINVAL;
405 501
406 error = -EALREADY; 502 if (MAX_SIZEOF_BTS < ds_cfg.sizeof_rec[ds_bts])
407 if (context->owner[qual] == current) 503 return -EOVERFLOW;
408 goto out_put_tracer;
409 error = -EPERM;
410 if (context->owner[qual] != NULL)
411 goto out_put_tracer;
412 context->owner[qual] = current;
413 504
414 spin_unlock_irqrestore(&ds_lock, irq); 505 switch (in->qualifier) {
506 case bts_invalid:
507 bts_set(raw, bts_from, 0);
508 bts_set(raw, bts_to, 0);
509 bts_set(raw, bts_flags, 0);
510 break;
511 case bts_branch:
512 bts_set(raw, bts_from, in->variant.lbr.from);
513 bts_set(raw, bts_to, in->variant.lbr.to);
514 bts_set(raw, bts_flags, 0);
515 break;
516 case bts_task_arrives:
517 case bts_task_departs:
518 bts_set(raw, bts_qual, (bts_escape | in->qualifier));
519 bts_set(raw, bts_jiffies, in->variant.timestamp.jiffies);
520 bts_set(raw, bts_pid, in->variant.timestamp.pid);
521 break;
522 default:
523 return -EINVAL;
524 }
415 525
526 return ds_write(tracer->ds.context, ds_bts, raw,
527 ds_cfg.sizeof_rec[ds_bts]);
528}
416 529
417 error = -ENOMEM;
418 if (!base) {
419 base = ds_allocate_buffer(size, &context->pages[qual]);
420 if (!base)
421 goto out_release;
422 530
423 context->buffer[qual] = base; 531static void ds_write_config(struct ds_context *context,
424 } 532 struct ds_trace *cfg, enum ds_qualifier qual)
425 error = 0; 533{
534 unsigned char *ds = context->ds;
535
536 ds_set(ds, qual, ds_buffer_base, (unsigned long)cfg->begin);
537 ds_set(ds, qual, ds_index, (unsigned long)cfg->top);
538 ds_set(ds, qual, ds_absolute_maximum, (unsigned long)cfg->end);
539 ds_set(ds, qual, ds_interrupt_threshold, (unsigned long)cfg->ith);
540}
541
542static void ds_read_config(struct ds_context *context,
543 struct ds_trace *cfg, enum ds_qualifier qual)
544{
545 unsigned char *ds = context->ds;
426 546
427 context->callback[qual] = ovfl; 547 cfg->begin = (void *)ds_get(ds, qual, ds_buffer_base);
548 cfg->top = (void *)ds_get(ds, qual, ds_index);
549 cfg->end = (void *)ds_get(ds, qual, ds_absolute_maximum);
550 cfg->ith = (void *)ds_get(ds, qual, ds_interrupt_threshold);
551}
552
553static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
554 void *base, size_t size, size_t ith,
555 unsigned int flags) {
556 unsigned long buffer, adj;
428 557
429 /* adjust the buffer address and size to meet alignment 558 /* adjust the buffer address and size to meet alignment
430 * constraints: 559 * constraints:
@@ -436,410 +565,383 @@ static int ds_request(struct task_struct *task, void *base, size_t size,
436 */ 565 */
437 buffer = (unsigned long)base; 566 buffer = (unsigned long)base;
438 567
439 adj = ALIGN(buffer, alignment) - buffer; 568 adj = ALIGN(buffer, DS_ALIGNMENT) - buffer;
440 buffer += adj; 569 buffer += adj;
441 size -= adj; 570 size -= adj;
442 571
443 size /= ds_cfg.sizeof_rec[qual]; 572 trace->n = size / ds_cfg.sizeof_rec[qual];
444 size *= ds_cfg.sizeof_rec[qual]; 573 trace->size = ds_cfg.sizeof_rec[qual];
445
446 ds_set(context->ds, qual, ds_buffer_base, buffer);
447 ds_set(context->ds, qual, ds_index, buffer);
448 ds_set(context->ds, qual, ds_absolute_maximum, buffer + size);
449 574
450 if (ovfl) { 575 size = (trace->n * trace->size);
451 /* todo: select a suitable interrupt threshold */
452 } else
453 ds_set(context->ds, qual,
454 ds_interrupt_threshold, buffer + size + 1);
455 576
456 /* we keep the context until ds_release */ 577 trace->begin = (void *)buffer;
457 return error; 578 trace->top = trace->begin;
458 579 trace->end = (void *)(buffer + size);
459 out_release: 580 /* The value for 'no threshold' is -1, which will set the
460 context->owner[qual] = NULL; 581 * threshold outside of the buffer, just like we want it.
461 ds_put_context(context); 582 */
462 put_tracer(task); 583 trace->ith = (void *)(buffer + size - ith);
463 return error;
464
465 out_put_tracer:
466 spin_unlock_irqrestore(&ds_lock, irq);
467 ds_put_context(context);
468 put_tracer(task);
469 return error;
470 584
471 out_unlock: 585 trace->flags = flags;
472 spin_unlock_irqrestore(&ds_lock, irq);
473 ds_put_context(context);
474 return error;
475} 586}
476 587
477int ds_request_bts(struct task_struct *task, void *base, size_t size,
478 ds_ovfl_callback_t ovfl)
479{
480 return ds_request(task, base, size, ovfl, ds_bts);
481}
482 588
483int ds_request_pebs(struct task_struct *task, void *base, size_t size, 589static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
484 ds_ovfl_callback_t ovfl) 590 enum ds_qualifier qual, struct task_struct *task,
485{ 591 void *base, size_t size, size_t th, unsigned int flags)
486 return ds_request(task, base, size, ovfl, ds_pebs);
487}
488
489static int ds_release(struct task_struct *task, enum ds_qualifier qual)
490{ 592{
491 struct ds_context *context; 593 struct ds_context *context;
492 int error; 594 int error;
493 595
494 context = ds_get_context(task); 596 error = -EINVAL;
495 error = ds_validate_access(context, qual); 597 if (!base)
496 if (error < 0)
497 goto out; 598 goto out;
498 599
499 kfree(context->buffer[qual]); 600 /* we require some space to do alignment adjustments below */
500 context->buffer[qual] = NULL; 601 error = -EINVAL;
501 602 if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
502 current->mm->total_vm -= context->pages[qual]; 603 goto out;
503 current->mm->locked_vm -= context->pages[qual];
504 context->pages[qual] = 0;
505 context->owner[qual] = NULL;
506
507 /*
508 * we put the context twice:
509 * once for the ds_get_context
510 * once for the corresponding ds_request
511 */
512 ds_put_context(context);
513 out:
514 ds_put_context(context);
515 return error;
516}
517 604
518int ds_release_bts(struct task_struct *task) 605 if (th != (size_t)-1) {
519{ 606 th *= ds_cfg.sizeof_rec[qual];
520 return ds_release(task, ds_bts);
521}
522 607
523int ds_release_pebs(struct task_struct *task) 608 error = -EINVAL;
524{ 609 if (size <= th)
525 return ds_release(task, ds_pebs); 610 goto out;
526} 611 }
527 612
528static int ds_get_index(struct task_struct *task, size_t *pos, 613 tracer->buffer = base;
529 enum ds_qualifier qual) 614 tracer->size = size;
530{
531 struct ds_context *context;
532 unsigned long base, index;
533 int error;
534 615
616 error = -ENOMEM;
535 context = ds_get_context(task); 617 context = ds_get_context(task);
536 error = ds_validate_access(context, qual); 618 if (!context)
537 if (error < 0)
538 goto out; 619 goto out;
620 tracer->context = context;
539 621
540 base = ds_get(context->ds, qual, ds_buffer_base); 622 ds_init_ds_trace(trace, qual, base, size, th, flags);
541 index = ds_get(context->ds, qual, ds_index);
542 623
543 error = ((index - base) / ds_cfg.sizeof_rec[qual]); 624 error = 0;
544 if (pos)
545 *pos = error;
546 out: 625 out:
547 ds_put_context(context);
548 return error; 626 return error;
549} 627}
550 628
551int ds_get_bts_index(struct task_struct *task, size_t *pos) 629struct bts_tracer *ds_request_bts(struct task_struct *task,
552{ 630 void *base, size_t size,
553 return ds_get_index(task, pos, ds_bts); 631 bts_ovfl_callback_t ovfl, size_t th,
554} 632 unsigned int flags)
555
556int ds_get_pebs_index(struct task_struct *task, size_t *pos)
557{ 633{
558 return ds_get_index(task, pos, ds_pebs); 634 struct bts_tracer *tracer;
559} 635 unsigned long irq;
560
561static int ds_get_end(struct task_struct *task, size_t *pos,
562 enum ds_qualifier qual)
563{
564 struct ds_context *context;
565 unsigned long base, end;
566 int error; 636 int error;
567 637
568 context = ds_get_context(task); 638 error = -EOPNOTSUPP;
569 error = ds_validate_access(context, qual); 639 if (!ds_cfg.ctl[dsf_bts])
570 if (error < 0)
571 goto out; 640 goto out;
572 641
573 base = ds_get(context->ds, qual, ds_buffer_base); 642 /* buffer overflow notification is not yet implemented */
574 end = ds_get(context->ds, qual, ds_absolute_maximum); 643 error = -EOPNOTSUPP;
644 if (ovfl)
645 goto out;
575 646
576 error = ((end - base) / ds_cfg.sizeof_rec[qual]); 647 error = -ENOMEM;
577 if (pos) 648 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
578 *pos = error; 649 if (!tracer)
579 out: 650 goto out;
580 ds_put_context(context); 651 tracer->ovfl = ovfl;
581 return error;
582}
583 652
584int ds_get_bts_end(struct task_struct *task, size_t *pos) 653 error = ds_request(&tracer->ds, &tracer->trace.ds,
585{ 654 ds_bts, task, base, size, th, flags);
586 return ds_get_end(task, pos, ds_bts); 655 if (error < 0)
587} 656 goto out_tracer;
588 657
589int ds_get_pebs_end(struct task_struct *task, size_t *pos)
590{
591 return ds_get_end(task, pos, ds_pebs);
592}
593 658
594static int ds_access(struct task_struct *task, size_t index, 659 spin_lock_irqsave(&ds_lock, irq);
595 const void **record, enum ds_qualifier qual)
596{
597 struct ds_context *context;
598 unsigned long base, idx;
599 int error;
600 660
601 if (!record) 661 error = -EPERM;
602 return -EINVAL; 662 if (!check_tracer(task))
663 goto out_unlock;
664 get_tracer(task);
603 665
604 context = ds_get_context(task); 666 error = -EPERM;
605 error = ds_validate_access(context, qual); 667 if (tracer->ds.context->bts_master)
606 if (error < 0) 668 goto out_put_tracer;
607 goto out; 669 tracer->ds.context->bts_master = tracer;
608 670
609 base = ds_get(context->ds, qual, ds_buffer_base); 671 spin_unlock_irqrestore(&ds_lock, irq);
610 idx = base + (index * ds_cfg.sizeof_rec[qual]);
611 672
612 error = -EINVAL;
613 if (idx > ds_get(context->ds, qual, ds_absolute_maximum))
614 goto out;
615 673
616 *record = (const void *)idx; 674 tracer->trace.read = bts_read;
617 error = ds_cfg.sizeof_rec[qual]; 675 tracer->trace.write = bts_write;
618 out:
619 ds_put_context(context);
620 return error;
621}
622 676
623int ds_access_bts(struct task_struct *task, size_t index, const void **record) 677 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
624{ 678 ds_resume_bts(tracer);
625 return ds_access(task, index, record, ds_bts);
626}
627 679
628int ds_access_pebs(struct task_struct *task, size_t index, const void **record) 680 return tracer;
629{ 681
630 return ds_access(task, index, record, ds_pebs); 682 out_put_tracer:
683 put_tracer(task);
684 out_unlock:
685 spin_unlock_irqrestore(&ds_lock, irq);
686 ds_put_context(tracer->ds.context);
687 out_tracer:
688 kfree(tracer);
689 out:
690 return ERR_PTR(error);
631} 691}
632 692
633static int ds_write(struct task_struct *task, const void *record, size_t size, 693struct pebs_tracer *ds_request_pebs(struct task_struct *task,
634 enum ds_qualifier qual, int force) 694 void *base, size_t size,
695 pebs_ovfl_callback_t ovfl, size_t th,
696 unsigned int flags)
635{ 697{
636 struct ds_context *context; 698 struct pebs_tracer *tracer;
699 unsigned long irq;
637 int error; 700 int error;
638 701
639 if (!record) 702 /* buffer overflow notification is not yet implemented */
640 return -EINVAL; 703 error = -EOPNOTSUPP;
704 if (ovfl)
705 goto out;
641 706
642 error = -EPERM; 707 error = -ENOMEM;
643 context = ds_get_context(task); 708 tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
644 if (!context) 709 if (!tracer)
645 goto out; 710 goto out;
711 tracer->ovfl = ovfl;
646 712
647 if (!force) { 713 error = ds_request(&tracer->ds, &tracer->trace.ds,
648 error = ds_validate_access(context, qual); 714 ds_pebs, task, base, size, th, flags);
649 if (error < 0) 715 if (error < 0)
650 goto out; 716 goto out_tracer;
651 }
652 717
653 error = 0; 718 spin_lock_irqsave(&ds_lock, irq);
654 while (size) {
655 unsigned long base, index, end, write_end, int_th;
656 unsigned long write_size, adj_write_size;
657 719
658 /* 720 error = -EPERM;
659 * write as much as possible without producing an 721 if (!check_tracer(task))
660 * overflow interrupt. 722 goto out_unlock;
661 * 723 get_tracer(task);
662 * interrupt_threshold must either be
663 * - bigger than absolute_maximum or
664 * - point to a record between buffer_base and absolute_maximum
665 *
666 * index points to a valid record.
667 */
668 base = ds_get(context->ds, qual, ds_buffer_base);
669 index = ds_get(context->ds, qual, ds_index);
670 end = ds_get(context->ds, qual, ds_absolute_maximum);
671 int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
672 724
673 write_end = min(end, int_th); 725 error = -EPERM;
726 if (tracer->ds.context->pebs_master)
727 goto out_put_tracer;
728 tracer->ds.context->pebs_master = tracer;
674 729
675 /* if we are already beyond the interrupt threshold, 730 spin_unlock_irqrestore(&ds_lock, irq);
676 * we fill the entire buffer */
677 if (write_end <= index)
678 write_end = end;
679 731
680 if (write_end <= index) 732 ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
681 goto out; 733 ds_resume_pebs(tracer);
682 734
683 write_size = min((unsigned long) size, write_end - index); 735 return tracer;
684 memcpy((void *)index, record, write_size);
685 736
686 record = (const char *)record + write_size; 737 out_put_tracer:
687 size -= write_size; 738 put_tracer(task);
688 error += write_size; 739 out_unlock:
740 spin_unlock_irqrestore(&ds_lock, irq);
741 ds_put_context(tracer->ds.context);
742 out_tracer:
743 kfree(tracer);
744 out:
745 return ERR_PTR(error);
746}
689 747
690 adj_write_size = write_size / ds_cfg.sizeof_rec[qual]; 748void ds_release_bts(struct bts_tracer *tracer)
691 adj_write_size *= ds_cfg.sizeof_rec[qual]; 749{
750 if (!tracer)
751 return;
692 752
693 /* zero out trailing bytes */ 753 ds_suspend_bts(tracer);
694 memset((char *)index + write_size, 0,
695 adj_write_size - write_size);
696 index += adj_write_size;
697 754
698 if (index >= end) 755 WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
699 index = base; 756 tracer->ds.context->bts_master = NULL;
700 ds_set(context->ds, qual, ds_index, index);
701 757
702 if (index >= int_th) 758 put_tracer(tracer->ds.context->task);
703 ds_overflow(task, context, qual); 759 ds_put_context(tracer->ds.context);
704 }
705 760
706 out: 761 kfree(tracer);
707 ds_put_context(context);
708 return error;
709} 762}
710 763
711int ds_write_bts(struct task_struct *task, const void *record, size_t size) 764void ds_suspend_bts(struct bts_tracer *tracer)
712{ 765{
713 return ds_write(task, record, size, ds_bts, /* force = */ 0); 766 struct task_struct *task;
714}
715 767
716int ds_write_pebs(struct task_struct *task, const void *record, size_t size) 768 if (!tracer)
717{ 769 return;
718 return ds_write(task, record, size, ds_pebs, /* force = */ 0);
719}
720 770
721int ds_unchecked_write_bts(struct task_struct *task, 771 task = tracer->ds.context->task;
722 const void *record, size_t size)
723{
724 return ds_write(task, record, size, ds_bts, /* force = */ 1);
725}
726 772
727int ds_unchecked_write_pebs(struct task_struct *task, 773 if (!task || (task == current))
728 const void *record, size_t size) 774 update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL);
729{ 775
730 return ds_write(task, record, size, ds_pebs, /* force = */ 1); 776 if (task) {
777 task->thread.debugctlmsr &= ~BTS_CONTROL;
778
779 if (!task->thread.debugctlmsr)
780 clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
781 }
731} 782}
732 783
733static int ds_reset_or_clear(struct task_struct *task, 784void ds_resume_bts(struct bts_tracer *tracer)
734 enum ds_qualifier qual, int clear)
735{ 785{
736 struct ds_context *context; 786 struct task_struct *task;
737 unsigned long base, end; 787 unsigned long control;
738 int error;
739 788
740 context = ds_get_context(task); 789 if (!tracer)
741 error = ds_validate_access(context, qual); 790 return;
742 if (error < 0)
743 goto out;
744 791
745 base = ds_get(context->ds, qual, ds_buffer_base); 792 task = tracer->ds.context->task;
746 end = ds_get(context->ds, qual, ds_absolute_maximum);
747 793
748 if (clear) 794 control = ds_cfg.ctl[dsf_bts];
749 memset((void *)base, 0, end - base); 795 if (!(tracer->trace.ds.flags & BTS_KERNEL))
796 control |= ds_cfg.ctl[dsf_bts_kernel];
797 if (!(tracer->trace.ds.flags & BTS_USER))
798 control |= ds_cfg.ctl[dsf_bts_user];
750 799
751 ds_set(context->ds, qual, ds_index, base); 800 if (task) {
801 task->thread.debugctlmsr |= control;
802 set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
803 }
752 804
753 error = 0; 805 if (!task || (task == current))
754 out: 806 update_debugctlmsr(get_debugctlmsr() | control);
755 ds_put_context(context);
756 return error;
757} 807}
758 808
759int ds_reset_bts(struct task_struct *task) 809void ds_release_pebs(struct pebs_tracer *tracer)
760{ 810{
761 return ds_reset_or_clear(task, ds_bts, /* clear = */ 0); 811 if (!tracer)
812 return;
813
814 ds_suspend_pebs(tracer);
815
816 WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
817 tracer->ds.context->pebs_master = NULL;
818
819 put_tracer(tracer->ds.context->task);
820 ds_put_context(tracer->ds.context);
821
822 kfree(tracer);
762} 823}
763 824
764int ds_reset_pebs(struct task_struct *task) 825void ds_suspend_pebs(struct pebs_tracer *tracer)
765{ 826{
766 return ds_reset_or_clear(task, ds_pebs, /* clear = */ 0); 827
767} 828}
768 829
769int ds_clear_bts(struct task_struct *task) 830void ds_resume_pebs(struct pebs_tracer *tracer)
770{ 831{
771 return ds_reset_or_clear(task, ds_bts, /* clear = */ 1); 832
772} 833}
773 834
774int ds_clear_pebs(struct task_struct *task) 835const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
775{ 836{
776 return ds_reset_or_clear(task, ds_pebs, /* clear = */ 1); 837 if (!tracer)
838 return NULL;
839
840 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
841 return &tracer->trace;
777} 842}
778 843
779int ds_get_pebs_reset(struct task_struct *task, u64 *value) 844const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer)
780{ 845{
781 struct ds_context *context; 846 if (!tracer)
782 int error; 847 return NULL;
848
849 ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
850 tracer->trace.reset_value =
851 *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8));
783 852
784 if (!value) 853 return &tracer->trace;
854}
855
856int ds_reset_bts(struct bts_tracer *tracer)
857{
858 if (!tracer)
785 return -EINVAL; 859 return -EINVAL;
786 860
787 context = ds_get_context(task); 861 tracer->trace.ds.top = tracer->trace.ds.begin;
788 error = ds_validate_access(context, ds_pebs);
789 if (error < 0)
790 goto out;
791 862
792 *value = *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8)); 863 ds_set(tracer->ds.context->ds, ds_bts, ds_index,
864 (unsigned long)tracer->trace.ds.top);
793 865
794 error = 0; 866 return 0;
795 out:
796 ds_put_context(context);
797 return error;
798} 867}
799 868
800int ds_set_pebs_reset(struct task_struct *task, u64 value) 869int ds_reset_pebs(struct pebs_tracer *tracer)
801{ 870{
802 struct ds_context *context; 871 if (!tracer)
803 int error; 872 return -EINVAL;
804 873
805 context = ds_get_context(task); 874 tracer->trace.ds.top = tracer->trace.ds.begin;
806 error = ds_validate_access(context, ds_pebs);
807 if (error < 0)
808 goto out;
809 875
810 *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8)) = value; 876 ds_set(tracer->ds.context->ds, ds_bts, ds_index,
877 (unsigned long)tracer->trace.ds.top);
811 878
812 error = 0; 879 return 0;
813 out: 880}
814 ds_put_context(context); 881
815 return error; 882int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value)
883{
884 if (!tracer)
885 return -EINVAL;
886
887 *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)) = value;
888
889 return 0;
816} 890}
817 891
818static const struct ds_configuration ds_cfg_var = { 892static const struct ds_configuration ds_cfg_netburst = {
819 .sizeof_ds = sizeof(long) * 12, 893 .name = "netburst",
820 .sizeof_field = sizeof(long), 894 .ctl[dsf_bts] = (1 << 2) | (1 << 3),
821 .sizeof_rec[ds_bts] = sizeof(long) * 3, 895 .ctl[dsf_bts_kernel] = (1 << 5),
896 .ctl[dsf_bts_user] = (1 << 6),
897
898 .sizeof_field = sizeof(long),
899 .sizeof_rec[ds_bts] = sizeof(long) * 3,
822#ifdef __i386__ 900#ifdef __i386__
823 .sizeof_rec[ds_pebs] = sizeof(long) * 10 901 .sizeof_rec[ds_pebs] = sizeof(long) * 10,
824#else 902#else
825 .sizeof_rec[ds_pebs] = sizeof(long) * 18 903 .sizeof_rec[ds_pebs] = sizeof(long) * 18,
826#endif 904#endif
827}; 905};
828static const struct ds_configuration ds_cfg_64 = { 906static const struct ds_configuration ds_cfg_pentium_m = {
829 .sizeof_ds = 8 * 12, 907 .name = "pentium m",
830 .sizeof_field = 8, 908 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
831 .sizeof_rec[ds_bts] = 8 * 3, 909
910 .sizeof_field = sizeof(long),
911 .sizeof_rec[ds_bts] = sizeof(long) * 3,
832#ifdef __i386__ 912#ifdef __i386__
833 .sizeof_rec[ds_pebs] = 8 * 10 913 .sizeof_rec[ds_pebs] = sizeof(long) * 10,
834#else 914#else
835 .sizeof_rec[ds_pebs] = 8 * 18 915 .sizeof_rec[ds_pebs] = sizeof(long) * 18,
836#endif 916#endif
837}; 917};
918static const struct ds_configuration ds_cfg_core2 = {
919 .name = "core 2",
920 .ctl[dsf_bts] = (1 << 6) | (1 << 7),
921 .ctl[dsf_bts_kernel] = (1 << 9),
922 .ctl[dsf_bts_user] = (1 << 10),
923
924 .sizeof_field = 8,
925 .sizeof_rec[ds_bts] = 8 * 3,
926 .sizeof_rec[ds_pebs] = 8 * 18,
927};
838 928
839static inline void 929static void
840ds_configure(const struct ds_configuration *cfg) 930ds_configure(const struct ds_configuration *cfg)
841{ 931{
932 memset(&ds_cfg, 0, sizeof(ds_cfg));
842 ds_cfg = *cfg; 933 ds_cfg = *cfg;
934
935 printk(KERN_INFO "[ds] using %s configuration\n", ds_cfg.name);
936
937 if (!cpu_has_bts) {
938 ds_cfg.ctl[dsf_bts] = 0;
939 printk(KERN_INFO "[ds] bts not available\n");
940 }
941 if (!cpu_has_pebs)
942 printk(KERN_INFO "[ds] pebs not available\n");
943
944 WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_field));
843} 945}
844 946
845void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) 947void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
@@ -847,16 +949,15 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
847 switch (c->x86) { 949 switch (c->x86) {
848 case 0x6: 950 case 0x6:
849 switch (c->x86_model) { 951 switch (c->x86_model) {
952 case 0 ... 0xC:
953 /* sorry, don't know about them */
954 break;
850 case 0xD: 955 case 0xD:
851 case 0xE: /* Pentium M */ 956 case 0xE: /* Pentium M */
852 ds_configure(&ds_cfg_var); 957 ds_configure(&ds_cfg_pentium_m);
853 break; 958 break;
854 case 0xF: /* Core2 */ 959 default: /* Core2, Atom, ... */
855 case 0x1C: /* Atom */ 960 ds_configure(&ds_cfg_core2);
856 ds_configure(&ds_cfg_64);
857 break;
858 default:
859 /* sorry, don't know about them */
860 break; 961 break;
861 } 962 }
862 break; 963 break;
@@ -865,7 +966,7 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
865 case 0x0: 966 case 0x0:
866 case 0x1: 967 case 0x1:
867 case 0x2: /* Netburst */ 968 case 0x2: /* Netburst */
868 ds_configure(&ds_cfg_var); 969 ds_configure(&ds_cfg_netburst);
869 break; 970 break;
870 default: 971 default:
871 /* sorry, don't know about them */ 972 /* sorry, don't know about them */
@@ -878,12 +979,52 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
878 } 979 }
879} 980}
880 981
881void ds_free(struct ds_context *context) 982/*
983 * Change the DS configuration from tracing prev to tracing next.
984 */
985void ds_switch_to(struct task_struct *prev, struct task_struct *next)
986{
987 struct ds_context *prev_ctx = prev->thread.ds_ctx;
988 struct ds_context *next_ctx = next->thread.ds_ctx;
989
990 if (prev_ctx) {
991 update_debugctlmsr(0);
992
993 if (prev_ctx->bts_master &&
994 (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
995 struct bts_struct ts = {
996 .qualifier = bts_task_departs,
997 .variant.timestamp.jiffies = jiffies_64,
998 .variant.timestamp.pid = prev->pid
999 };
1000 bts_write(prev_ctx->bts_master, &ts);
1001 }
1002 }
1003
1004 if (next_ctx) {
1005 if (next_ctx->bts_master &&
1006 (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
1007 struct bts_struct ts = {
1008 .qualifier = bts_task_arrives,
1009 .variant.timestamp.jiffies = jiffies_64,
1010 .variant.timestamp.pid = next->pid
1011 };
1012 bts_write(next_ctx->bts_master, &ts);
1013 }
1014
1015 wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
1016 }
1017
1018 update_debugctlmsr(next->thread.debugctlmsr);
1019}
1020
1021void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)
1022{
1023 clear_tsk_thread_flag(tsk, TIF_DS_AREA_MSR);
1024 tsk->thread.ds_ctx = NULL;
1025}
1026
1027void ds_exit_thread(struct task_struct *tsk)
882{ 1028{
883 /* This is called when the task owning the parameter context 1029 WARN_ON(tsk->thread.ds_ctx);
884 * is dying. There should not be any user of that context left
885 * to disturb us, anymore. */
886 unsigned long leftovers = context->count;
887 while (leftovers--)
888 ds_put_context(context);
889} 1030}
diff --git a/arch/x86/kernel/dumpstack.c b/arch/x86/kernel/dumpstack.c
new file mode 100644
index 000000000000..6b1f6f6f8661
--- /dev/null
+++ b/arch/x86/kernel/dumpstack.c
@@ -0,0 +1,351 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 */
5#include <linux/kallsyms.h>
6#include <linux/kprobes.h>
7#include <linux/uaccess.h>
8#include <linux/utsname.h>
9#include <linux/hardirq.h>
10#include <linux/kdebug.h>
11#include <linux/module.h>
12#include <linux/ptrace.h>
13#include <linux/kexec.h>
14#include <linux/bug.h>
15#include <linux/nmi.h>
16#include <linux/sysfs.h>
17
18#include <asm/stacktrace.h>
19
20#include "dumpstack.h"
21
22int panic_on_unrecovered_nmi;
23unsigned int code_bytes = 64;
24int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
25static int die_counter;
26
27void printk_address(unsigned long address, int reliable)
28{
29 printk(" [<%p>] %s%pS\n", (void *) address,
30 reliable ? "" : "? ", (void *) address);
31}
32
33#ifdef CONFIG_FUNCTION_GRAPH_TRACER
34static void
35print_ftrace_graph_addr(unsigned long addr, void *data,
36 const struct stacktrace_ops *ops,
37 struct thread_info *tinfo, int *graph)
38{
39 struct task_struct *task = tinfo->task;
40 unsigned long ret_addr;
41 int index = task->curr_ret_stack;
42
43 if (addr != (unsigned long)return_to_handler)
44 return;
45
46 if (!task->ret_stack || index < *graph)
47 return;
48
49 index -= *graph;
50 ret_addr = task->ret_stack[index].ret;
51
52 ops->address(data, ret_addr, 1);
53
54 (*graph)++;
55}
56#else
57static inline void
58print_ftrace_graph_addr(unsigned long addr, void *data,
59 const struct stacktrace_ops *ops,
60 struct thread_info *tinfo, int *graph)
61{ }
62#endif
63
64/*
65 * x86-64 can have up to three kernel stacks:
66 * process stack
67 * interrupt stack
68 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
69 */
70
71static inline int valid_stack_ptr(struct thread_info *tinfo,
72 void *p, unsigned int size, void *end)
73{
74 void *t = tinfo;
75 if (end) {
76 if (p < end && p >= (end-THREAD_SIZE))
77 return 1;
78 else
79 return 0;
80 }
81 return p > t && p < t + THREAD_SIZE - size;
82}
83
84unsigned long
85print_context_stack(struct thread_info *tinfo,
86 unsigned long *stack, unsigned long bp,
87 const struct stacktrace_ops *ops, void *data,
88 unsigned long *end, int *graph)
89{
90 struct stack_frame *frame = (struct stack_frame *)bp;
91
92 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
93 unsigned long addr;
94
95 addr = *stack;
96 if (__kernel_text_address(addr)) {
97 if ((unsigned long) stack == bp + sizeof(long)) {
98 ops->address(data, addr, 1);
99 frame = frame->next_frame;
100 bp = (unsigned long) frame;
101 } else {
102 ops->address(data, addr, bp == 0);
103 }
104 print_ftrace_graph_addr(addr, data, ops, tinfo, graph);
105 }
106 stack++;
107 }
108 return bp;
109}
110
111
112static void
113print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
114{
115 printk(data);
116 print_symbol(msg, symbol);
117 printk("\n");
118}
119
120static void print_trace_warning(void *data, char *msg)
121{
122 printk("%s%s\n", (char *)data, msg);
123}
124
125static int print_trace_stack(void *data, char *name)
126{
127 printk("%s <%s> ", (char *)data, name);
128 return 0;
129}
130
131/*
132 * Print one address/symbol entries per line.
133 */
134static void print_trace_address(void *data, unsigned long addr, int reliable)
135{
136 touch_nmi_watchdog();
137 printk(data);
138 printk_address(addr, reliable);
139}
140
141static const struct stacktrace_ops print_trace_ops = {
142 .warning = print_trace_warning,
143 .warning_symbol = print_trace_warning_symbol,
144 .stack = print_trace_stack,
145 .address = print_trace_address,
146};
147
148void
149show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
150 unsigned long *stack, unsigned long bp, char *log_lvl)
151{
152 printk("%sCall Trace:\n", log_lvl);
153 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
154}
155
156void show_trace(struct task_struct *task, struct pt_regs *regs,
157 unsigned long *stack, unsigned long bp)
158{
159 show_trace_log_lvl(task, regs, stack, bp, "");
160}
161
162void show_stack(struct task_struct *task, unsigned long *sp)
163{
164 show_stack_log_lvl(task, NULL, sp, 0, "");
165}
166
167/*
168 * The architecture-independent dump_stack generator
169 */
170void dump_stack(void)
171{
172 unsigned long bp = 0;
173 unsigned long stack;
174
175#ifdef CONFIG_FRAME_POINTER
176 if (!bp)
177 get_bp(bp);
178#endif
179
180 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
181 current->pid, current->comm, print_tainted(),
182 init_utsname()->release,
183 (int)strcspn(init_utsname()->version, " "),
184 init_utsname()->version);
185 show_trace(NULL, NULL, &stack, bp);
186}
187EXPORT_SYMBOL(dump_stack);
188
189static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
190static int die_owner = -1;
191static unsigned int die_nest_count;
192
193unsigned __kprobes long oops_begin(void)
194{
195 int cpu;
196 unsigned long flags;
197
198 oops_enter();
199
200 /* racy, but better than risking deadlock. */
201 raw_local_irq_save(flags);
202 cpu = smp_processor_id();
203 if (!__raw_spin_trylock(&die_lock)) {
204 if (cpu == die_owner)
205 /* nested oops. should stop eventually */;
206 else
207 __raw_spin_lock(&die_lock);
208 }
209 die_nest_count++;
210 die_owner = cpu;
211 console_verbose();
212 bust_spinlocks(1);
213 return flags;
214}
215
216void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
217{
218 if (regs && kexec_should_crash(current))
219 crash_kexec(regs);
220
221 bust_spinlocks(0);
222 die_owner = -1;
223 add_taint(TAINT_DIE);
224 die_nest_count--;
225 if (!die_nest_count)
226 /* Nest count reaches zero, release the lock. */
227 __raw_spin_unlock(&die_lock);
228 raw_local_irq_restore(flags);
229 oops_exit();
230
231 if (!signr)
232 return;
233 if (in_interrupt())
234 panic("Fatal exception in interrupt");
235 if (panic_on_oops)
236 panic("Fatal exception");
237 do_exit(signr);
238}
239
240int __kprobes __die(const char *str, struct pt_regs *regs, long err)
241{
242#ifdef CONFIG_X86_32
243 unsigned short ss;
244 unsigned long sp;
245#endif
246 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
247#ifdef CONFIG_PREEMPT
248 printk("PREEMPT ");
249#endif
250#ifdef CONFIG_SMP
251 printk("SMP ");
252#endif
253#ifdef CONFIG_DEBUG_PAGEALLOC
254 printk("DEBUG_PAGEALLOC");
255#endif
256 printk("\n");
257 sysfs_printk_last_file();
258 if (notify_die(DIE_OOPS, str, regs, err,
259 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
260 return 1;
261
262 show_registers(regs);
263#ifdef CONFIG_X86_32
264 sp = (unsigned long) (&regs->sp);
265 savesegment(ss, ss);
266 if (user_mode(regs)) {
267 sp = regs->sp;
268 ss = regs->ss & 0xffff;
269 }
270 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
271 print_symbol("%s", regs->ip);
272 printk(" SS:ESP %04x:%08lx\n", ss, sp);
273#else
274 /* Executive summary in case the oops scrolled away */
275 printk(KERN_ALERT "RIP ");
276 printk_address(regs->ip, 1);
277 printk(" RSP <%016lx>\n", regs->sp);
278#endif
279 return 0;
280}
281
282/*
283 * This is gone through when something in the kernel has done something bad
284 * and is about to be terminated:
285 */
286void die(const char *str, struct pt_regs *regs, long err)
287{
288 unsigned long flags = oops_begin();
289 int sig = SIGSEGV;
290
291 if (!user_mode_vm(regs))
292 report_bug(regs->ip, regs);
293
294 if (__die(str, regs, err))
295 sig = 0;
296 oops_end(flags, regs, sig);
297}
298
299void notrace __kprobes
300die_nmi(char *str, struct pt_regs *regs, int do_panic)
301{
302 unsigned long flags;
303
304 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
305 return;
306
307 /*
308 * We are in trouble anyway, lets at least try
309 * to get a message out.
310 */
311 flags = oops_begin();
312 printk(KERN_EMERG "%s", str);
313 printk(" on CPU%d, ip %08lx, registers:\n",
314 smp_processor_id(), regs->ip);
315 show_registers(regs);
316 oops_end(flags, regs, 0);
317 if (do_panic || panic_on_oops)
318 panic("Non maskable interrupt");
319 nmi_exit();
320 local_irq_enable();
321 do_exit(SIGBUS);
322}
323
324static int __init oops_setup(char *s)
325{
326 if (!s)
327 return -EINVAL;
328 if (!strcmp(s, "panic"))
329 panic_on_oops = 1;
330 return 0;
331}
332early_param("oops", oops_setup);
333
334static int __init kstack_setup(char *s)
335{
336 if (!s)
337 return -EINVAL;
338 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
339 return 0;
340}
341early_param("kstack", kstack_setup);
342
343static int __init code_bytes_setup(char *s)
344{
345 code_bytes = simple_strtoul(s, NULL, 0);
346 if (code_bytes > 8192)
347 code_bytes = 8192;
348
349 return 1;
350}
351__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/dumpstack.h b/arch/x86/kernel/dumpstack.h
new file mode 100644
index 000000000000..da87590b8698
--- /dev/null
+++ b/arch/x86/kernel/dumpstack.h
@@ -0,0 +1,39 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 */
5
6#ifndef DUMPSTACK_H
7#define DUMPSTACK_H
8
9#ifdef CONFIG_X86_32
10#define STACKSLOTS_PER_LINE 8
11#define get_bp(bp) asm("movl %%ebp, %0" : "=r" (bp) :)
12#else
13#define STACKSLOTS_PER_LINE 4
14#define get_bp(bp) asm("movq %%rbp, %0" : "=r" (bp) :)
15#endif
16
17extern unsigned long
18print_context_stack(struct thread_info *tinfo,
19 unsigned long *stack, unsigned long bp,
20 const struct stacktrace_ops *ops, void *data,
21 unsigned long *end, int *graph);
22
23extern void
24show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
25 unsigned long *stack, unsigned long bp, char *log_lvl);
26
27extern void
28show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
29 unsigned long *sp, unsigned long bp, char *log_lvl);
30
31extern unsigned int code_bytes;
32extern int kstack_depth_to_print;
33
34/* The form of the top of the frame on the stack */
35struct stack_frame {
36 struct stack_frame *next_frame;
37 unsigned long return_address;
38};
39#endif
diff --git a/arch/x86/kernel/dumpstack_32.c b/arch/x86/kernel/dumpstack_32.c
index b3614752197b..d593cd1f58dc 100644
--- a/arch/x86/kernel/dumpstack_32.c
+++ b/arch/x86/kernel/dumpstack_32.c
@@ -17,69 +17,14 @@
17 17
18#include <asm/stacktrace.h> 18#include <asm/stacktrace.h>
19 19
20#define STACKSLOTS_PER_LINE 8 20#include "dumpstack.h"
21#define get_bp(bp) asm("movl %%ebp, %0" : "=r" (bp) :)
22
23int panic_on_unrecovered_nmi;
24int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
25static unsigned int code_bytes = 64;
26static int die_counter;
27
28void printk_address(unsigned long address, int reliable)
29{
30 printk(" [<%p>] %s%pS\n", (void *) address,
31 reliable ? "" : "? ", (void *) address);
32}
33
34static inline int valid_stack_ptr(struct thread_info *tinfo,
35 void *p, unsigned int size, void *end)
36{
37 void *t = tinfo;
38 if (end) {
39 if (p < end && p >= (end-THREAD_SIZE))
40 return 1;
41 else
42 return 0;
43 }
44 return p > t && p < t + THREAD_SIZE - size;
45}
46
47/* The form of the top of the frame on the stack */
48struct stack_frame {
49 struct stack_frame *next_frame;
50 unsigned long return_address;
51};
52
53static inline unsigned long
54print_context_stack(struct thread_info *tinfo,
55 unsigned long *stack, unsigned long bp,
56 const struct stacktrace_ops *ops, void *data,
57 unsigned long *end)
58{
59 struct stack_frame *frame = (struct stack_frame *)bp;
60
61 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
62 unsigned long addr;
63
64 addr = *stack;
65 if (__kernel_text_address(addr)) {
66 if ((unsigned long) stack == bp + sizeof(long)) {
67 ops->address(data, addr, 1);
68 frame = frame->next_frame;
69 bp = (unsigned long) frame;
70 } else {
71 ops->address(data, addr, bp == 0);
72 }
73 }
74 stack++;
75 }
76 return bp;
77}
78 21
79void dump_trace(struct task_struct *task, struct pt_regs *regs, 22void dump_trace(struct task_struct *task, struct pt_regs *regs,
80 unsigned long *stack, unsigned long bp, 23 unsigned long *stack, unsigned long bp,
81 const struct stacktrace_ops *ops, void *data) 24 const struct stacktrace_ops *ops, void *data)
82{ 25{
26 int graph = 0;
27
83 if (!task) 28 if (!task)
84 task = current; 29 task = current;
85 30
@@ -107,7 +52,8 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
107 52
108 context = (struct thread_info *) 53 context = (struct thread_info *)
109 ((unsigned long)stack & (~(THREAD_SIZE - 1))); 54 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
110 bp = print_context_stack(context, stack, bp, ops, data, NULL); 55 bp = print_context_stack(context, stack, bp, ops,
56 data, NULL, &graph);
111 57
112 stack = (unsigned long *)context->previous_esp; 58 stack = (unsigned long *)context->previous_esp;
113 if (!stack) 59 if (!stack)
@@ -119,57 +65,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
119} 65}
120EXPORT_SYMBOL(dump_trace); 66EXPORT_SYMBOL(dump_trace);
121 67
122static void 68void
123print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
124{
125 printk(data);
126 print_symbol(msg, symbol);
127 printk("\n");
128}
129
130static void print_trace_warning(void *data, char *msg)
131{
132 printk("%s%s\n", (char *)data, msg);
133}
134
135static int print_trace_stack(void *data, char *name)
136{
137 printk("%s <%s> ", (char *)data, name);
138 return 0;
139}
140
141/*
142 * Print one address/symbol entries per line.
143 */
144static void print_trace_address(void *data, unsigned long addr, int reliable)
145{
146 touch_nmi_watchdog();
147 printk(data);
148 printk_address(addr, reliable);
149}
150
151static const struct stacktrace_ops print_trace_ops = {
152 .warning = print_trace_warning,
153 .warning_symbol = print_trace_warning_symbol,
154 .stack = print_trace_stack,
155 .address = print_trace_address,
156};
157
158static void
159show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
160 unsigned long *stack, unsigned long bp, char *log_lvl)
161{
162 printk("%sCall Trace:\n", log_lvl);
163 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
164}
165
166void show_trace(struct task_struct *task, struct pt_regs *regs,
167 unsigned long *stack, unsigned long bp)
168{
169 show_trace_log_lvl(task, regs, stack, bp, "");
170}
171
172static void
173show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs, 69show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
174 unsigned long *sp, unsigned long bp, char *log_lvl) 70 unsigned long *sp, unsigned long bp, char *log_lvl)
175{ 71{
@@ -196,33 +92,6 @@ show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
196 show_trace_log_lvl(task, regs, sp, bp, log_lvl); 92 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
197} 93}
198 94
199void show_stack(struct task_struct *task, unsigned long *sp)
200{
201 show_stack_log_lvl(task, NULL, sp, 0, "");
202}
203
204/*
205 * The architecture-independent dump_stack generator
206 */
207void dump_stack(void)
208{
209 unsigned long bp = 0;
210 unsigned long stack;
211
212#ifdef CONFIG_FRAME_POINTER
213 if (!bp)
214 get_bp(bp);
215#endif
216
217 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
218 current->pid, current->comm, print_tainted(),
219 init_utsname()->release,
220 (int)strcspn(init_utsname()->version, " "),
221 init_utsname()->version);
222 show_trace(NULL, NULL, &stack, bp);
223}
224
225EXPORT_SYMBOL(dump_stack);
226 95
227void show_registers(struct pt_regs *regs) 96void show_registers(struct pt_regs *regs)
228{ 97{
@@ -283,167 +152,3 @@ int is_valid_bugaddr(unsigned long ip)
283 return ud2 == 0x0b0f; 152 return ud2 == 0x0b0f;
284} 153}
285 154
286static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
287static int die_owner = -1;
288static unsigned int die_nest_count;
289
290unsigned __kprobes long oops_begin(void)
291{
292 unsigned long flags;
293
294 oops_enter();
295
296 if (die_owner != raw_smp_processor_id()) {
297 console_verbose();
298 raw_local_irq_save(flags);
299 __raw_spin_lock(&die_lock);
300 die_owner = smp_processor_id();
301 die_nest_count = 0;
302 bust_spinlocks(1);
303 } else {
304 raw_local_irq_save(flags);
305 }
306 die_nest_count++;
307 return flags;
308}
309
310void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
311{
312 bust_spinlocks(0);
313 die_owner = -1;
314 add_taint(TAINT_DIE);
315 __raw_spin_unlock(&die_lock);
316 raw_local_irq_restore(flags);
317
318 if (!regs)
319 return;
320
321 if (kexec_should_crash(current))
322 crash_kexec(regs);
323 if (in_interrupt())
324 panic("Fatal exception in interrupt");
325 if (panic_on_oops)
326 panic("Fatal exception");
327 oops_exit();
328 do_exit(signr);
329}
330
331int __kprobes __die(const char *str, struct pt_regs *regs, long err)
332{
333 unsigned short ss;
334 unsigned long sp;
335
336 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
337#ifdef CONFIG_PREEMPT
338 printk("PREEMPT ");
339#endif
340#ifdef CONFIG_SMP
341 printk("SMP ");
342#endif
343#ifdef CONFIG_DEBUG_PAGEALLOC
344 printk("DEBUG_PAGEALLOC");
345#endif
346 printk("\n");
347 sysfs_printk_last_file();
348 if (notify_die(DIE_OOPS, str, regs, err,
349 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
350 return 1;
351
352 show_registers(regs);
353 /* Executive summary in case the oops scrolled away */
354 sp = (unsigned long) (&regs->sp);
355 savesegment(ss, ss);
356 if (user_mode(regs)) {
357 sp = regs->sp;
358 ss = regs->ss & 0xffff;
359 }
360 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
361 print_symbol("%s", regs->ip);
362 printk(" SS:ESP %04x:%08lx\n", ss, sp);
363 return 0;
364}
365
366/*
367 * This is gone through when something in the kernel has done something bad
368 * and is about to be terminated:
369 */
370void die(const char *str, struct pt_regs *regs, long err)
371{
372 unsigned long flags = oops_begin();
373
374 if (die_nest_count < 3) {
375 report_bug(regs->ip, regs);
376
377 if (__die(str, regs, err))
378 regs = NULL;
379 } else {
380 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
381 }
382
383 oops_end(flags, regs, SIGSEGV);
384}
385
386static DEFINE_SPINLOCK(nmi_print_lock);
387
388void notrace __kprobes
389die_nmi(char *str, struct pt_regs *regs, int do_panic)
390{
391 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
392 return;
393
394 spin_lock(&nmi_print_lock);
395 /*
396 * We are in trouble anyway, lets at least try
397 * to get a message out:
398 */
399 bust_spinlocks(1);
400 printk(KERN_EMERG "%s", str);
401 printk(" on CPU%d, ip %08lx, registers:\n",
402 smp_processor_id(), regs->ip);
403 show_registers(regs);
404 if (do_panic)
405 panic("Non maskable interrupt");
406 console_silent();
407 spin_unlock(&nmi_print_lock);
408
409 /*
410 * If we are in kernel we are probably nested up pretty bad
411 * and might aswell get out now while we still can:
412 */
413 if (!user_mode_vm(regs)) {
414 current->thread.trap_no = 2;
415 crash_kexec(regs);
416 }
417
418 bust_spinlocks(0);
419 do_exit(SIGSEGV);
420}
421
422static int __init oops_setup(char *s)
423{
424 if (!s)
425 return -EINVAL;
426 if (!strcmp(s, "panic"))
427 panic_on_oops = 1;
428 return 0;
429}
430early_param("oops", oops_setup);
431
432static int __init kstack_setup(char *s)
433{
434 if (!s)
435 return -EINVAL;
436 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
437 return 0;
438}
439early_param("kstack", kstack_setup);
440
441static int __init code_bytes_setup(char *s)
442{
443 code_bytes = simple_strtoul(s, NULL, 0);
444 if (code_bytes > 8192)
445 code_bytes = 8192;
446
447 return 1;
448}
449__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/dumpstack_64.c b/arch/x86/kernel/dumpstack_64.c
index 96a5db7da8a7..c302d0707048 100644
--- a/arch/x86/kernel/dumpstack_64.c
+++ b/arch/x86/kernel/dumpstack_64.c
@@ -17,19 +17,7 @@
17 17
18#include <asm/stacktrace.h> 18#include <asm/stacktrace.h>
19 19
20#define STACKSLOTS_PER_LINE 4 20#include "dumpstack.h"
21#define get_bp(bp) asm("movq %%rbp, %0" : "=r" (bp) :)
22
23int panic_on_unrecovered_nmi;
24int kstack_depth_to_print = 3 * STACKSLOTS_PER_LINE;
25static unsigned int code_bytes = 64;
26static int die_counter;
27
28void printk_address(unsigned long address, int reliable)
29{
30 printk(" [<%p>] %s%pS\n", (void *) address,
31 reliable ? "" : "? ", (void *) address);
32}
33 21
34static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack, 22static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
35 unsigned *usedp, char **idp) 23 unsigned *usedp, char **idp)
@@ -113,51 +101,6 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
113 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack 101 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
114 */ 102 */
115 103
116static inline int valid_stack_ptr(struct thread_info *tinfo,
117 void *p, unsigned int size, void *end)
118{
119 void *t = tinfo;
120 if (end) {
121 if (p < end && p >= (end-THREAD_SIZE))
122 return 1;
123 else
124 return 0;
125 }
126 return p > t && p < t + THREAD_SIZE - size;
127}
128
129/* The form of the top of the frame on the stack */
130struct stack_frame {
131 struct stack_frame *next_frame;
132 unsigned long return_address;
133};
134
135static inline unsigned long
136print_context_stack(struct thread_info *tinfo,
137 unsigned long *stack, unsigned long bp,
138 const struct stacktrace_ops *ops, void *data,
139 unsigned long *end)
140{
141 struct stack_frame *frame = (struct stack_frame *)bp;
142
143 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
144 unsigned long addr;
145
146 addr = *stack;
147 if (__kernel_text_address(addr)) {
148 if ((unsigned long) stack == bp + sizeof(long)) {
149 ops->address(data, addr, 1);
150 frame = frame->next_frame;
151 bp = (unsigned long) frame;
152 } else {
153 ops->address(data, addr, bp == 0);
154 }
155 }
156 stack++;
157 }
158 return bp;
159}
160
161void dump_trace(struct task_struct *task, struct pt_regs *regs, 104void dump_trace(struct task_struct *task, struct pt_regs *regs,
162 unsigned long *stack, unsigned long bp, 105 unsigned long *stack, unsigned long bp,
163 const struct stacktrace_ops *ops, void *data) 106 const struct stacktrace_ops *ops, void *data)
@@ -166,6 +109,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
166 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr; 109 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
167 unsigned used = 0; 110 unsigned used = 0;
168 struct thread_info *tinfo; 111 struct thread_info *tinfo;
112 int graph = 0;
169 113
170 if (!task) 114 if (!task)
171 task = current; 115 task = current;
@@ -206,7 +150,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
206 break; 150 break;
207 151
208 bp = print_context_stack(tinfo, stack, bp, ops, 152 bp = print_context_stack(tinfo, stack, bp, ops,
209 data, estack_end); 153 data, estack_end, &graph);
210 ops->stack(data, "<EOE>"); 154 ops->stack(data, "<EOE>");
211 /* 155 /*
212 * We link to the next stack via the 156 * We link to the next stack via the
@@ -225,7 +169,7 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
225 if (ops->stack(data, "IRQ") < 0) 169 if (ops->stack(data, "IRQ") < 0)
226 break; 170 break;
227 bp = print_context_stack(tinfo, stack, bp, 171 bp = print_context_stack(tinfo, stack, bp,
228 ops, data, irqstack_end); 172 ops, data, irqstack_end, &graph);
229 /* 173 /*
230 * We link to the next stack (which would be 174 * We link to the next stack (which would be
231 * the process stack normally) the last 175 * the process stack normally) the last
@@ -243,62 +187,12 @@ void dump_trace(struct task_struct *task, struct pt_regs *regs,
243 /* 187 /*
244 * This handles the process stack: 188 * This handles the process stack:
245 */ 189 */
246 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL); 190 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL, &graph);
247 put_cpu(); 191 put_cpu();
248} 192}
249EXPORT_SYMBOL(dump_trace); 193EXPORT_SYMBOL(dump_trace);
250 194
251static void 195void
252print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
253{
254 printk(data);
255 print_symbol(msg, symbol);
256 printk("\n");
257}
258
259static void print_trace_warning(void *data, char *msg)
260{
261 printk("%s%s\n", (char *)data, msg);
262}
263
264static int print_trace_stack(void *data, char *name)
265{
266 printk("%s <%s> ", (char *)data, name);
267 return 0;
268}
269
270/*
271 * Print one address/symbol entries per line.
272 */
273static void print_trace_address(void *data, unsigned long addr, int reliable)
274{
275 touch_nmi_watchdog();
276 printk(data);
277 printk_address(addr, reliable);
278}
279
280static const struct stacktrace_ops print_trace_ops = {
281 .warning = print_trace_warning,
282 .warning_symbol = print_trace_warning_symbol,
283 .stack = print_trace_stack,
284 .address = print_trace_address,
285};
286
287static void
288show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
289 unsigned long *stack, unsigned long bp, char *log_lvl)
290{
291 printk("%sCall Trace:\n", log_lvl);
292 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
293}
294
295void show_trace(struct task_struct *task, struct pt_regs *regs,
296 unsigned long *stack, unsigned long bp)
297{
298 show_trace_log_lvl(task, regs, stack, bp, "");
299}
300
301static void
302show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs, 196show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
303 unsigned long *sp, unsigned long bp, char *log_lvl) 197 unsigned long *sp, unsigned long bp, char *log_lvl)
304{ 198{
@@ -342,33 +236,6 @@ show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
342 show_trace_log_lvl(task, regs, sp, bp, log_lvl); 236 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
343} 237}
344 238
345void show_stack(struct task_struct *task, unsigned long *sp)
346{
347 show_stack_log_lvl(task, NULL, sp, 0, "");
348}
349
350/*
351 * The architecture-independent dump_stack generator
352 */
353void dump_stack(void)
354{
355 unsigned long bp = 0;
356 unsigned long stack;
357
358#ifdef CONFIG_FRAME_POINTER
359 if (!bp)
360 get_bp(bp);
361#endif
362
363 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
364 current->pid, current->comm, print_tainted(),
365 init_utsname()->release,
366 (int)strcspn(init_utsname()->version, " "),
367 init_utsname()->version);
368 show_trace(NULL, NULL, &stack, bp);
369}
370EXPORT_SYMBOL(dump_stack);
371
372void show_registers(struct pt_regs *regs) 239void show_registers(struct pt_regs *regs)
373{ 240{
374 int i; 241 int i;
@@ -429,147 +296,3 @@ int is_valid_bugaddr(unsigned long ip)
429 return ud2 == 0x0b0f; 296 return ud2 == 0x0b0f;
430} 297}
431 298
432static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
433static int die_owner = -1;
434static unsigned int die_nest_count;
435
436unsigned __kprobes long oops_begin(void)
437{
438 int cpu;
439 unsigned long flags;
440
441 oops_enter();
442
443 /* racy, but better than risking deadlock. */
444 raw_local_irq_save(flags);
445 cpu = smp_processor_id();
446 if (!__raw_spin_trylock(&die_lock)) {
447 if (cpu == die_owner)
448 /* nested oops. should stop eventually */;
449 else
450 __raw_spin_lock(&die_lock);
451 }
452 die_nest_count++;
453 die_owner = cpu;
454 console_verbose();
455 bust_spinlocks(1);
456 return flags;
457}
458
459void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
460{
461 die_owner = -1;
462 bust_spinlocks(0);
463 die_nest_count--;
464 if (!die_nest_count)
465 /* Nest count reaches zero, release the lock. */
466 __raw_spin_unlock(&die_lock);
467 raw_local_irq_restore(flags);
468 if (!regs) {
469 oops_exit();
470 return;
471 }
472 if (in_interrupt())
473 panic("Fatal exception in interrupt");
474 if (panic_on_oops)
475 panic("Fatal exception");
476 oops_exit();
477 do_exit(signr);
478}
479
480int __kprobes __die(const char *str, struct pt_regs *regs, long err)
481{
482 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
483#ifdef CONFIG_PREEMPT
484 printk("PREEMPT ");
485#endif
486#ifdef CONFIG_SMP
487 printk("SMP ");
488#endif
489#ifdef CONFIG_DEBUG_PAGEALLOC
490 printk("DEBUG_PAGEALLOC");
491#endif
492 printk("\n");
493 sysfs_printk_last_file();
494 if (notify_die(DIE_OOPS, str, regs, err,
495 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
496 return 1;
497
498 show_registers(regs);
499 add_taint(TAINT_DIE);
500 /* Executive summary in case the oops scrolled away */
501 printk(KERN_ALERT "RIP ");
502 printk_address(regs->ip, 1);
503 printk(" RSP <%016lx>\n", regs->sp);
504 if (kexec_should_crash(current))
505 crash_kexec(regs);
506 return 0;
507}
508
509void die(const char *str, struct pt_regs *regs, long err)
510{
511 unsigned long flags = oops_begin();
512
513 if (!user_mode(regs))
514 report_bug(regs->ip, regs);
515
516 if (__die(str, regs, err))
517 regs = NULL;
518 oops_end(flags, regs, SIGSEGV);
519}
520
521notrace __kprobes void
522die_nmi(char *str, struct pt_regs *regs, int do_panic)
523{
524 unsigned long flags;
525
526 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
527 return;
528
529 flags = oops_begin();
530 /*
531 * We are in trouble anyway, lets at least try
532 * to get a message out.
533 */
534 printk(KERN_EMERG "%s", str);
535 printk(" on CPU%d, ip %08lx, registers:\n",
536 smp_processor_id(), regs->ip);
537 show_registers(regs);
538 if (kexec_should_crash(current))
539 crash_kexec(regs);
540 if (do_panic || panic_on_oops)
541 panic("Non maskable interrupt");
542 oops_end(flags, NULL, SIGBUS);
543 nmi_exit();
544 local_irq_enable();
545 do_exit(SIGBUS);
546}
547
548static int __init oops_setup(char *s)
549{
550 if (!s)
551 return -EINVAL;
552 if (!strcmp(s, "panic"))
553 panic_on_oops = 1;
554 return 0;
555}
556early_param("oops", oops_setup);
557
558static int __init kstack_setup(char *s)
559{
560 if (!s)
561 return -EINVAL;
562 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
563 return 0;
564}
565early_param("kstack", kstack_setup);
566
567static int __init code_bytes_setup(char *s)
568{
569 code_bytes = simple_strtoul(s, NULL, 0);
570 if (code_bytes > 8192)
571 code_bytes = 8192;
572
573 return 1;
574}
575__setup("code_bytes=", code_bytes_setup);
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
index 7aafeb5263ef..65a13943e098 100644
--- a/arch/x86/kernel/e820.c
+++ b/arch/x86/kernel/e820.c
@@ -677,22 +677,6 @@ struct early_res {
677}; 677};
678static struct early_res early_res[MAX_EARLY_RES] __initdata = { 678static struct early_res early_res[MAX_EARLY_RES] __initdata = {
679 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */ 679 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
680#if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
681 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
682#endif
683#if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
684 /*
685 * But first pinch a few for the stack/trampoline stuff
686 * FIXME: Don't need the extra page at 4K, but need to fix
687 * trampoline before removing it. (see the GDT stuff)
688 */
689 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
690 /*
691 * Has to be in very low memory so we can execute
692 * real-mode AP code.
693 */
694 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
695#endif
696 {} 680 {}
697}; 681};
698 682
diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
index 1b894b72c0f5..744aa7fc49d5 100644
--- a/arch/x86/kernel/early-quirks.c
+++ b/arch/x86/kernel/early-quirks.c
@@ -17,6 +17,7 @@
17#include <asm/io_apic.h> 17#include <asm/io_apic.h>
18#include <asm/apic.h> 18#include <asm/apic.h>
19#include <asm/iommu.h> 19#include <asm/iommu.h>
20#include <asm/gart.h>
20 21
21static void __init fix_hypertransport_config(int num, int slot, int func) 22static void __init fix_hypertransport_config(int num, int slot, int func)
22{ 23{
diff --git a/arch/x86/kernel/early_printk.c b/arch/x86/kernel/early_printk.c
index 34ad997d3834..23b138e31e9c 100644
--- a/arch/x86/kernel/early_printk.c
+++ b/arch/x86/kernel/early_printk.c
@@ -875,49 +875,6 @@ static struct console early_dbgp_console = {
875}; 875};
876#endif 876#endif
877 877
878/* Console interface to a host file on AMD's SimNow! */
879
880static int simnow_fd;
881
882enum {
883 MAGIC1 = 0xBACCD00A,
884 MAGIC2 = 0xCA110000,
885 XOPEN = 5,
886 XWRITE = 4,
887};
888
889static noinline long simnow(long cmd, long a, long b, long c)
890{
891 long ret;
892
893 asm volatile("cpuid" :
894 "=a" (ret) :
895 "b" (a), "c" (b), "d" (c), "0" (MAGIC1), "D" (cmd + MAGIC2));
896 return ret;
897}
898
899static void __init simnow_init(char *str)
900{
901 char *fn = "klog";
902
903 if (*str == '=')
904 fn = ++str;
905 /* error ignored */
906 simnow_fd = simnow(XOPEN, (unsigned long)fn, O_WRONLY|O_APPEND|O_CREAT, 0644);
907}
908
909static void simnow_write(struct console *con, const char *s, unsigned n)
910{
911 simnow(XWRITE, simnow_fd, (unsigned long)s, n);
912}
913
914static struct console simnow_console = {
915 .name = "simnow",
916 .write = simnow_write,
917 .flags = CON_PRINTBUFFER,
918 .index = -1,
919};
920
921/* Direct interface for emergencies */ 878/* Direct interface for emergencies */
922static struct console *early_console = &early_vga_console; 879static struct console *early_console = &early_vga_console;
923static int __initdata early_console_initialized; 880static int __initdata early_console_initialized;
@@ -960,10 +917,6 @@ static int __init setup_early_printk(char *buf)
960 max_ypos = boot_params.screen_info.orig_video_lines; 917 max_ypos = boot_params.screen_info.orig_video_lines;
961 current_ypos = boot_params.screen_info.orig_y; 918 current_ypos = boot_params.screen_info.orig_y;
962 early_console = &early_vga_console; 919 early_console = &early_vga_console;
963 } else if (!strncmp(buf, "simnow", 6)) {
964 simnow_init(buf + 6);
965 early_console = &simnow_console;
966 keep_early = 1;
967#ifdef CONFIG_EARLY_PRINTK_DBGP 920#ifdef CONFIG_EARLY_PRINTK_DBGP
968 } else if (!strncmp(buf, "dbgp", 4)) { 921 } else if (!strncmp(buf, "dbgp", 4)) {
969 if (early_dbgp_init(buf+4) < 0) 922 if (early_dbgp_init(buf+4) < 0)
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index 28b597ef9ca1..d6f0490a7391 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -619,28 +619,37 @@ END(syscall_badsys)
61927:; 61927:;
620 620
621/* 621/*
622 * Build the entry stubs and pointer table with 622 * Build the entry stubs and pointer table with some assembler magic.
623 * some assembler magic. 623 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
624 * single cache line on all modern x86 implementations.
624 */ 625 */
625.section .rodata,"a" 626.section .init.rodata,"a"
626ENTRY(interrupt) 627ENTRY(interrupt)
627.text 628.text
628 629 .p2align 5
630 .p2align CONFIG_X86_L1_CACHE_SHIFT
629ENTRY(irq_entries_start) 631ENTRY(irq_entries_start)
630 RING0_INT_FRAME 632 RING0_INT_FRAME
631vector=0 633vector=FIRST_EXTERNAL_VECTOR
632.rept NR_VECTORS 634.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
633 ALIGN 635 .balign 32
634 .if vector 636 .rept 7
637 .if vector < NR_VECTORS
638 .if vector <> FIRST_EXTERNAL_VECTOR
635 CFI_ADJUST_CFA_OFFSET -4 639 CFI_ADJUST_CFA_OFFSET -4
636 .endif 640 .endif
6371: pushl $~(vector) 6411: pushl $(~vector+0x80) /* Note: always in signed byte range */
638 CFI_ADJUST_CFA_OFFSET 4 642 CFI_ADJUST_CFA_OFFSET 4
639 jmp common_interrupt 643 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
640 .previous 644 jmp 2f
645 .endif
646 .previous
641 .long 1b 647 .long 1b
642 .text 648 .text
643vector=vector+1 649vector=vector+1
650 .endif
651 .endr
6522: jmp common_interrupt
644.endr 653.endr
645END(irq_entries_start) 654END(irq_entries_start)
646 655
@@ -652,8 +661,9 @@ END(interrupt)
652 * the CPU automatically disables interrupts when executing an IRQ vector, 661 * the CPU automatically disables interrupts when executing an IRQ vector,
653 * so IRQ-flags tracing has to follow that: 662 * so IRQ-flags tracing has to follow that:
654 */ 663 */
655 ALIGN 664 .p2align CONFIG_X86_L1_CACHE_SHIFT
656common_interrupt: 665common_interrupt:
666 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
657 SAVE_ALL 667 SAVE_ALL
658 TRACE_IRQS_OFF 668 TRACE_IRQS_OFF
659 movl %esp,%eax 669 movl %esp,%eax
@@ -678,65 +688,6 @@ ENDPROC(name)
678/* The include is where all of the SMP etc. interrupts come from */ 688/* The include is where all of the SMP etc. interrupts come from */
679#include "entry_arch.h" 689#include "entry_arch.h"
680 690
681KPROBE_ENTRY(page_fault)
682 RING0_EC_FRAME
683 pushl $do_page_fault
684 CFI_ADJUST_CFA_OFFSET 4
685 ALIGN
686error_code:
687 /* the function address is in %fs's slot on the stack */
688 pushl %es
689 CFI_ADJUST_CFA_OFFSET 4
690 /*CFI_REL_OFFSET es, 0*/
691 pushl %ds
692 CFI_ADJUST_CFA_OFFSET 4
693 /*CFI_REL_OFFSET ds, 0*/
694 pushl %eax
695 CFI_ADJUST_CFA_OFFSET 4
696 CFI_REL_OFFSET eax, 0
697 pushl %ebp
698 CFI_ADJUST_CFA_OFFSET 4
699 CFI_REL_OFFSET ebp, 0
700 pushl %edi
701 CFI_ADJUST_CFA_OFFSET 4
702 CFI_REL_OFFSET edi, 0
703 pushl %esi
704 CFI_ADJUST_CFA_OFFSET 4
705 CFI_REL_OFFSET esi, 0
706 pushl %edx
707 CFI_ADJUST_CFA_OFFSET 4
708 CFI_REL_OFFSET edx, 0
709 pushl %ecx
710 CFI_ADJUST_CFA_OFFSET 4
711 CFI_REL_OFFSET ecx, 0
712 pushl %ebx
713 CFI_ADJUST_CFA_OFFSET 4
714 CFI_REL_OFFSET ebx, 0
715 cld
716 pushl %fs
717 CFI_ADJUST_CFA_OFFSET 4
718 /*CFI_REL_OFFSET fs, 0*/
719 movl $(__KERNEL_PERCPU), %ecx
720 movl %ecx, %fs
721 UNWIND_ESPFIX_STACK
722 popl %ecx
723 CFI_ADJUST_CFA_OFFSET -4
724 /*CFI_REGISTER es, ecx*/
725 movl PT_FS(%esp), %edi # get the function address
726 movl PT_ORIG_EAX(%esp), %edx # get the error code
727 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
728 mov %ecx, PT_FS(%esp)
729 /*CFI_REL_OFFSET fs, ES*/
730 movl $(__USER_DS), %ecx
731 movl %ecx, %ds
732 movl %ecx, %es
733 TRACE_IRQS_OFF
734 movl %esp,%eax # pt_regs pointer
735 call *%edi
736 jmp ret_from_exception
737 CFI_ENDPROC
738KPROBE_END(page_fault)
739
740ENTRY(coprocessor_error) 691ENTRY(coprocessor_error)
741 RING0_INT_FRAME 692 RING0_INT_FRAME
742 pushl $0 693 pushl $0
@@ -767,140 +718,6 @@ ENTRY(device_not_available)
767 CFI_ENDPROC 718 CFI_ENDPROC
768END(device_not_available) 719END(device_not_available)
769 720
770/*
771 * Debug traps and NMI can happen at the one SYSENTER instruction
772 * that sets up the real kernel stack. Check here, since we can't
773 * allow the wrong stack to be used.
774 *
775 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
776 * already pushed 3 words if it hits on the sysenter instruction:
777 * eflags, cs and eip.
778 *
779 * We just load the right stack, and push the three (known) values
780 * by hand onto the new stack - while updating the return eip past
781 * the instruction that would have done it for sysenter.
782 */
783#define FIX_STACK(offset, ok, label) \
784 cmpw $__KERNEL_CS,4(%esp); \
785 jne ok; \
786label: \
787 movl TSS_sysenter_sp0+offset(%esp),%esp; \
788 CFI_DEF_CFA esp, 0; \
789 CFI_UNDEFINED eip; \
790 pushfl; \
791 CFI_ADJUST_CFA_OFFSET 4; \
792 pushl $__KERNEL_CS; \
793 CFI_ADJUST_CFA_OFFSET 4; \
794 pushl $sysenter_past_esp; \
795 CFI_ADJUST_CFA_OFFSET 4; \
796 CFI_REL_OFFSET eip, 0
797
798KPROBE_ENTRY(debug)
799 RING0_INT_FRAME
800 cmpl $ia32_sysenter_target,(%esp)
801 jne debug_stack_correct
802 FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
803debug_stack_correct:
804 pushl $-1 # mark this as an int
805 CFI_ADJUST_CFA_OFFSET 4
806 SAVE_ALL
807 TRACE_IRQS_OFF
808 xorl %edx,%edx # error code 0
809 movl %esp,%eax # pt_regs pointer
810 call do_debug
811 jmp ret_from_exception
812 CFI_ENDPROC
813KPROBE_END(debug)
814
815/*
816 * NMI is doubly nasty. It can happen _while_ we're handling
817 * a debug fault, and the debug fault hasn't yet been able to
818 * clear up the stack. So we first check whether we got an
819 * NMI on the sysenter entry path, but after that we need to
820 * check whether we got an NMI on the debug path where the debug
821 * fault happened on the sysenter path.
822 */
823KPROBE_ENTRY(nmi)
824 RING0_INT_FRAME
825 pushl %eax
826 CFI_ADJUST_CFA_OFFSET 4
827 movl %ss, %eax
828 cmpw $__ESPFIX_SS, %ax
829 popl %eax
830 CFI_ADJUST_CFA_OFFSET -4
831 je nmi_espfix_stack
832 cmpl $ia32_sysenter_target,(%esp)
833 je nmi_stack_fixup
834 pushl %eax
835 CFI_ADJUST_CFA_OFFSET 4
836 movl %esp,%eax
837 /* Do not access memory above the end of our stack page,
838 * it might not exist.
839 */
840 andl $(THREAD_SIZE-1),%eax
841 cmpl $(THREAD_SIZE-20),%eax
842 popl %eax
843 CFI_ADJUST_CFA_OFFSET -4
844 jae nmi_stack_correct
845 cmpl $ia32_sysenter_target,12(%esp)
846 je nmi_debug_stack_check
847nmi_stack_correct:
848 /* We have a RING0_INT_FRAME here */
849 pushl %eax
850 CFI_ADJUST_CFA_OFFSET 4
851 SAVE_ALL
852 TRACE_IRQS_OFF
853 xorl %edx,%edx # zero error code
854 movl %esp,%eax # pt_regs pointer
855 call do_nmi
856 jmp restore_nocheck_notrace
857 CFI_ENDPROC
858
859nmi_stack_fixup:
860 RING0_INT_FRAME
861 FIX_STACK(12,nmi_stack_correct, 1)
862 jmp nmi_stack_correct
863
864nmi_debug_stack_check:
865 /* We have a RING0_INT_FRAME here */
866 cmpw $__KERNEL_CS,16(%esp)
867 jne nmi_stack_correct
868 cmpl $debug,(%esp)
869 jb nmi_stack_correct
870 cmpl $debug_esp_fix_insn,(%esp)
871 ja nmi_stack_correct
872 FIX_STACK(24,nmi_stack_correct, 1)
873 jmp nmi_stack_correct
874
875nmi_espfix_stack:
876 /* We have a RING0_INT_FRAME here.
877 *
878 * create the pointer to lss back
879 */
880 pushl %ss
881 CFI_ADJUST_CFA_OFFSET 4
882 pushl %esp
883 CFI_ADJUST_CFA_OFFSET 4
884 addw $4, (%esp)
885 /* copy the iret frame of 12 bytes */
886 .rept 3
887 pushl 16(%esp)
888 CFI_ADJUST_CFA_OFFSET 4
889 .endr
890 pushl %eax
891 CFI_ADJUST_CFA_OFFSET 4
892 SAVE_ALL
893 TRACE_IRQS_OFF
894 FIXUP_ESPFIX_STACK # %eax == %esp
895 xorl %edx,%edx # zero error code
896 call do_nmi
897 RESTORE_REGS
898 lss 12+4(%esp), %esp # back to espfix stack
899 CFI_ADJUST_CFA_OFFSET -24
900 jmp irq_return
901 CFI_ENDPROC
902KPROBE_END(nmi)
903
904#ifdef CONFIG_PARAVIRT 721#ifdef CONFIG_PARAVIRT
905ENTRY(native_iret) 722ENTRY(native_iret)
906 iret 723 iret
@@ -916,19 +733,6 @@ ENTRY(native_irq_enable_sysexit)
916END(native_irq_enable_sysexit) 733END(native_irq_enable_sysexit)
917#endif 734#endif
918 735
919KPROBE_ENTRY(int3)
920 RING0_INT_FRAME
921 pushl $-1 # mark this as an int
922 CFI_ADJUST_CFA_OFFSET 4
923 SAVE_ALL
924 TRACE_IRQS_OFF
925 xorl %edx,%edx # zero error code
926 movl %esp,%eax # pt_regs pointer
927 call do_int3
928 jmp ret_from_exception
929 CFI_ENDPROC
930KPROBE_END(int3)
931
932ENTRY(overflow) 736ENTRY(overflow)
933 RING0_INT_FRAME 737 RING0_INT_FRAME
934 pushl $0 738 pushl $0
@@ -993,14 +797,6 @@ ENTRY(stack_segment)
993 CFI_ENDPROC 797 CFI_ENDPROC
994END(stack_segment) 798END(stack_segment)
995 799
996KPROBE_ENTRY(general_protection)
997 RING0_EC_FRAME
998 pushl $do_general_protection
999 CFI_ADJUST_CFA_OFFSET 4
1000 jmp error_code
1001 CFI_ENDPROC
1002KPROBE_END(general_protection)
1003
1004ENTRY(alignment_check) 800ENTRY(alignment_check)
1005 RING0_EC_FRAME 801 RING0_EC_FRAME
1006 pushl $do_alignment_check 802 pushl $do_alignment_check
@@ -1051,6 +847,7 @@ ENTRY(kernel_thread_helper)
1051 push %eax 847 push %eax
1052 CFI_ADJUST_CFA_OFFSET 4 848 CFI_ADJUST_CFA_OFFSET 4
1053 call do_exit 849 call do_exit
850 ud2 # padding for call trace
1054 CFI_ENDPROC 851 CFI_ENDPROC
1055ENDPROC(kernel_thread_helper) 852ENDPROC(kernel_thread_helper)
1056 853
@@ -1157,6 +954,9 @@ ENTRY(mcount)
1157END(mcount) 954END(mcount)
1158 955
1159ENTRY(ftrace_caller) 956ENTRY(ftrace_caller)
957 cmpl $0, function_trace_stop
958 jne ftrace_stub
959
1160 pushl %eax 960 pushl %eax
1161 pushl %ecx 961 pushl %ecx
1162 pushl %edx 962 pushl %edx
@@ -1171,6 +971,11 @@ ftrace_call:
1171 popl %edx 971 popl %edx
1172 popl %ecx 972 popl %ecx
1173 popl %eax 973 popl %eax
974#ifdef CONFIG_FUNCTION_GRAPH_TRACER
975.globl ftrace_graph_call
976ftrace_graph_call:
977 jmp ftrace_stub
978#endif
1174 979
1175.globl ftrace_stub 980.globl ftrace_stub
1176ftrace_stub: 981ftrace_stub:
@@ -1180,8 +985,18 @@ END(ftrace_caller)
1180#else /* ! CONFIG_DYNAMIC_FTRACE */ 985#else /* ! CONFIG_DYNAMIC_FTRACE */
1181 986
1182ENTRY(mcount) 987ENTRY(mcount)
988 cmpl $0, function_trace_stop
989 jne ftrace_stub
990
1183 cmpl $ftrace_stub, ftrace_trace_function 991 cmpl $ftrace_stub, ftrace_trace_function
1184 jnz trace 992 jnz trace
993#ifdef CONFIG_FUNCTION_GRAPH_TRACER
994 cmpl $ftrace_stub, ftrace_graph_return
995 jnz ftrace_graph_caller
996
997 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
998 jnz ftrace_graph_caller
999#endif
1185.globl ftrace_stub 1000.globl ftrace_stub
1186ftrace_stub: 1001ftrace_stub:
1187 ret 1002 ret
@@ -1200,13 +1015,268 @@ trace:
1200 popl %edx 1015 popl %edx
1201 popl %ecx 1016 popl %ecx
1202 popl %eax 1017 popl %eax
1203
1204 jmp ftrace_stub 1018 jmp ftrace_stub
1205END(mcount) 1019END(mcount)
1206#endif /* CONFIG_DYNAMIC_FTRACE */ 1020#endif /* CONFIG_DYNAMIC_FTRACE */
1207#endif /* CONFIG_FUNCTION_TRACER */ 1021#endif /* CONFIG_FUNCTION_TRACER */
1208 1022
1023#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1024ENTRY(ftrace_graph_caller)
1025 cmpl $0, function_trace_stop
1026 jne ftrace_stub
1027
1028 pushl %eax
1029 pushl %ecx
1030 pushl %edx
1031 movl 0xc(%esp), %edx
1032 lea 0x4(%ebp), %eax
1033 subl $MCOUNT_INSN_SIZE, %edx
1034 call prepare_ftrace_return
1035 popl %edx
1036 popl %ecx
1037 popl %eax
1038 ret
1039END(ftrace_graph_caller)
1040
1041.globl return_to_handler
1042return_to_handler:
1043 pushl $0
1044 pushl %eax
1045 pushl %ecx
1046 pushl %edx
1047 call ftrace_return_to_handler
1048 movl %eax, 0xc(%esp)
1049 popl %edx
1050 popl %ecx
1051 popl %eax
1052 ret
1053#endif
1054
1209.section .rodata,"a" 1055.section .rodata,"a"
1210#include "syscall_table_32.S" 1056#include "syscall_table_32.S"
1211 1057
1212syscall_table_size=(.-sys_call_table) 1058syscall_table_size=(.-sys_call_table)
1059
1060/*
1061 * Some functions should be protected against kprobes
1062 */
1063 .pushsection .kprobes.text, "ax"
1064
1065ENTRY(page_fault)
1066 RING0_EC_FRAME
1067 pushl $do_page_fault
1068 CFI_ADJUST_CFA_OFFSET 4
1069 ALIGN
1070error_code:
1071 /* the function address is in %fs's slot on the stack */
1072 pushl %es
1073 CFI_ADJUST_CFA_OFFSET 4
1074 /*CFI_REL_OFFSET es, 0*/
1075 pushl %ds
1076 CFI_ADJUST_CFA_OFFSET 4
1077 /*CFI_REL_OFFSET ds, 0*/
1078 pushl %eax
1079 CFI_ADJUST_CFA_OFFSET 4
1080 CFI_REL_OFFSET eax, 0
1081 pushl %ebp
1082 CFI_ADJUST_CFA_OFFSET 4
1083 CFI_REL_OFFSET ebp, 0
1084 pushl %edi
1085 CFI_ADJUST_CFA_OFFSET 4
1086 CFI_REL_OFFSET edi, 0
1087 pushl %esi
1088 CFI_ADJUST_CFA_OFFSET 4
1089 CFI_REL_OFFSET esi, 0
1090 pushl %edx
1091 CFI_ADJUST_CFA_OFFSET 4
1092 CFI_REL_OFFSET edx, 0
1093 pushl %ecx
1094 CFI_ADJUST_CFA_OFFSET 4
1095 CFI_REL_OFFSET ecx, 0
1096 pushl %ebx
1097 CFI_ADJUST_CFA_OFFSET 4
1098 CFI_REL_OFFSET ebx, 0
1099 cld
1100 pushl %fs
1101 CFI_ADJUST_CFA_OFFSET 4
1102 /*CFI_REL_OFFSET fs, 0*/
1103 movl $(__KERNEL_PERCPU), %ecx
1104 movl %ecx, %fs
1105 UNWIND_ESPFIX_STACK
1106 popl %ecx
1107 CFI_ADJUST_CFA_OFFSET -4
1108 /*CFI_REGISTER es, ecx*/
1109 movl PT_FS(%esp), %edi # get the function address
1110 movl PT_ORIG_EAX(%esp), %edx # get the error code
1111 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1112 mov %ecx, PT_FS(%esp)
1113 /*CFI_REL_OFFSET fs, ES*/
1114 movl $(__USER_DS), %ecx
1115 movl %ecx, %ds
1116 movl %ecx, %es
1117 TRACE_IRQS_OFF
1118 movl %esp,%eax # pt_regs pointer
1119 call *%edi
1120 jmp ret_from_exception
1121 CFI_ENDPROC
1122END(page_fault)
1123
1124/*
1125 * Debug traps and NMI can happen at the one SYSENTER instruction
1126 * that sets up the real kernel stack. Check here, since we can't
1127 * allow the wrong stack to be used.
1128 *
1129 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1130 * already pushed 3 words if it hits on the sysenter instruction:
1131 * eflags, cs and eip.
1132 *
1133 * We just load the right stack, and push the three (known) values
1134 * by hand onto the new stack - while updating the return eip past
1135 * the instruction that would have done it for sysenter.
1136 */
1137#define FIX_STACK(offset, ok, label) \
1138 cmpw $__KERNEL_CS,4(%esp); \
1139 jne ok; \
1140label: \
1141 movl TSS_sysenter_sp0+offset(%esp),%esp; \
1142 CFI_DEF_CFA esp, 0; \
1143 CFI_UNDEFINED eip; \
1144 pushfl; \
1145 CFI_ADJUST_CFA_OFFSET 4; \
1146 pushl $__KERNEL_CS; \
1147 CFI_ADJUST_CFA_OFFSET 4; \
1148 pushl $sysenter_past_esp; \
1149 CFI_ADJUST_CFA_OFFSET 4; \
1150 CFI_REL_OFFSET eip, 0
1151
1152ENTRY(debug)
1153 RING0_INT_FRAME
1154 cmpl $ia32_sysenter_target,(%esp)
1155 jne debug_stack_correct
1156 FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
1157debug_stack_correct:
1158 pushl $-1 # mark this as an int
1159 CFI_ADJUST_CFA_OFFSET 4
1160 SAVE_ALL
1161 TRACE_IRQS_OFF
1162 xorl %edx,%edx # error code 0
1163 movl %esp,%eax # pt_regs pointer
1164 call do_debug
1165 jmp ret_from_exception
1166 CFI_ENDPROC
1167END(debug)
1168
1169/*
1170 * NMI is doubly nasty. It can happen _while_ we're handling
1171 * a debug fault, and the debug fault hasn't yet been able to
1172 * clear up the stack. So we first check whether we got an
1173 * NMI on the sysenter entry path, but after that we need to
1174 * check whether we got an NMI on the debug path where the debug
1175 * fault happened on the sysenter path.
1176 */
1177ENTRY(nmi)
1178 RING0_INT_FRAME
1179 pushl %eax
1180 CFI_ADJUST_CFA_OFFSET 4
1181 movl %ss, %eax
1182 cmpw $__ESPFIX_SS, %ax
1183 popl %eax
1184 CFI_ADJUST_CFA_OFFSET -4
1185 je nmi_espfix_stack
1186 cmpl $ia32_sysenter_target,(%esp)
1187 je nmi_stack_fixup
1188 pushl %eax
1189 CFI_ADJUST_CFA_OFFSET 4
1190 movl %esp,%eax
1191 /* Do not access memory above the end of our stack page,
1192 * it might not exist.
1193 */
1194 andl $(THREAD_SIZE-1),%eax
1195 cmpl $(THREAD_SIZE-20),%eax
1196 popl %eax
1197 CFI_ADJUST_CFA_OFFSET -4
1198 jae nmi_stack_correct
1199 cmpl $ia32_sysenter_target,12(%esp)
1200 je nmi_debug_stack_check
1201nmi_stack_correct:
1202 /* We have a RING0_INT_FRAME here */
1203 pushl %eax
1204 CFI_ADJUST_CFA_OFFSET 4
1205 SAVE_ALL
1206 TRACE_IRQS_OFF
1207 xorl %edx,%edx # zero error code
1208 movl %esp,%eax # pt_regs pointer
1209 call do_nmi
1210 jmp restore_nocheck_notrace
1211 CFI_ENDPROC
1212
1213nmi_stack_fixup:
1214 RING0_INT_FRAME
1215 FIX_STACK(12,nmi_stack_correct, 1)
1216 jmp nmi_stack_correct
1217
1218nmi_debug_stack_check:
1219 /* We have a RING0_INT_FRAME here */
1220 cmpw $__KERNEL_CS,16(%esp)
1221 jne nmi_stack_correct
1222 cmpl $debug,(%esp)
1223 jb nmi_stack_correct
1224 cmpl $debug_esp_fix_insn,(%esp)
1225 ja nmi_stack_correct
1226 FIX_STACK(24,nmi_stack_correct, 1)
1227 jmp nmi_stack_correct
1228
1229nmi_espfix_stack:
1230 /* We have a RING0_INT_FRAME here.
1231 *
1232 * create the pointer to lss back
1233 */
1234 pushl %ss
1235 CFI_ADJUST_CFA_OFFSET 4
1236 pushl %esp
1237 CFI_ADJUST_CFA_OFFSET 4
1238 addw $4, (%esp)
1239 /* copy the iret frame of 12 bytes */
1240 .rept 3
1241 pushl 16(%esp)
1242 CFI_ADJUST_CFA_OFFSET 4
1243 .endr
1244 pushl %eax
1245 CFI_ADJUST_CFA_OFFSET 4
1246 SAVE_ALL
1247 TRACE_IRQS_OFF
1248 FIXUP_ESPFIX_STACK # %eax == %esp
1249 xorl %edx,%edx # zero error code
1250 call do_nmi
1251 RESTORE_REGS
1252 lss 12+4(%esp), %esp # back to espfix stack
1253 CFI_ADJUST_CFA_OFFSET -24
1254 jmp irq_return
1255 CFI_ENDPROC
1256END(nmi)
1257
1258ENTRY(int3)
1259 RING0_INT_FRAME
1260 pushl $-1 # mark this as an int
1261 CFI_ADJUST_CFA_OFFSET 4
1262 SAVE_ALL
1263 TRACE_IRQS_OFF
1264 xorl %edx,%edx # zero error code
1265 movl %esp,%eax # pt_regs pointer
1266 call do_int3
1267 jmp ret_from_exception
1268 CFI_ENDPROC
1269END(int3)
1270
1271ENTRY(general_protection)
1272 RING0_EC_FRAME
1273 pushl $do_general_protection
1274 CFI_ADJUST_CFA_OFFSET 4
1275 jmp error_code
1276 CFI_ENDPROC
1277END(general_protection)
1278
1279/*
1280 * End of kprobes section
1281 */
1282 .popsection
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index b86f332c96a6..e28c7a987793 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -11,15 +11,15 @@
11 * 11 *
12 * NOTE: This code handles signal-recognition, which happens every time 12 * NOTE: This code handles signal-recognition, which happens every time
13 * after an interrupt and after each system call. 13 * after an interrupt and after each system call.
14 * 14 *
15 * Normal syscalls and interrupts don't save a full stack frame, this is 15 * Normal syscalls and interrupts don't save a full stack frame, this is
16 * only done for syscall tracing, signals or fork/exec et.al. 16 * only done for syscall tracing, signals or fork/exec et.al.
17 * 17 *
18 * A note on terminology: 18 * A note on terminology:
19 * - top of stack: Architecture defined interrupt frame from SS to RIP 19 * - top of stack: Architecture defined interrupt frame from SS to RIP
20 * at the top of the kernel process stack. 20 * at the top of the kernel process stack.
21 * - partial stack frame: partially saved registers upto R11. 21 * - partial stack frame: partially saved registers upto R11.
22 * - full stack frame: Like partial stack frame, but all register saved. 22 * - full stack frame: Like partial stack frame, but all register saved.
23 * 23 *
24 * Some macro usage: 24 * Some macro usage:
25 * - CFI macros are used to generate dwarf2 unwind information for better 25 * - CFI macros are used to generate dwarf2 unwind information for better
@@ -60,7 +60,6 @@
60#define __AUDIT_ARCH_LE 0x40000000 60#define __AUDIT_ARCH_LE 0x40000000
61 61
62 .code64 62 .code64
63
64#ifdef CONFIG_FUNCTION_TRACER 63#ifdef CONFIG_FUNCTION_TRACER
65#ifdef CONFIG_DYNAMIC_FTRACE 64#ifdef CONFIG_DYNAMIC_FTRACE
66ENTRY(mcount) 65ENTRY(mcount)
@@ -68,16 +67,10 @@ ENTRY(mcount)
68END(mcount) 67END(mcount)
69 68
70ENTRY(ftrace_caller) 69ENTRY(ftrace_caller)
70 cmpl $0, function_trace_stop
71 jne ftrace_stub
71 72
72 /* taken from glibc */ 73 MCOUNT_SAVE_FRAME
73 subq $0x38, %rsp
74 movq %rax, (%rsp)
75 movq %rcx, 8(%rsp)
76 movq %rdx, 16(%rsp)
77 movq %rsi, 24(%rsp)
78 movq %rdi, 32(%rsp)
79 movq %r8, 40(%rsp)
80 movq %r9, 48(%rsp)
81 74
82 movq 0x38(%rsp), %rdi 75 movq 0x38(%rsp), %rdi
83 movq 8(%rbp), %rsi 76 movq 8(%rbp), %rsi
@@ -87,14 +80,13 @@ ENTRY(ftrace_caller)
87ftrace_call: 80ftrace_call:
88 call ftrace_stub 81 call ftrace_stub
89 82
90 movq 48(%rsp), %r9 83 MCOUNT_RESTORE_FRAME
91 movq 40(%rsp), %r8 84
92 movq 32(%rsp), %rdi 85#ifdef CONFIG_FUNCTION_GRAPH_TRACER
93 movq 24(%rsp), %rsi 86.globl ftrace_graph_call
94 movq 16(%rsp), %rdx 87ftrace_graph_call:
95 movq 8(%rsp), %rcx 88 jmp ftrace_stub
96 movq (%rsp), %rax 89#endif
97 addq $0x38, %rsp
98 90
99.globl ftrace_stub 91.globl ftrace_stub
100ftrace_stub: 92ftrace_stub:
@@ -103,15 +95,63 @@ END(ftrace_caller)
103 95
104#else /* ! CONFIG_DYNAMIC_FTRACE */ 96#else /* ! CONFIG_DYNAMIC_FTRACE */
105ENTRY(mcount) 97ENTRY(mcount)
98 cmpl $0, function_trace_stop
99 jne ftrace_stub
100
106 cmpq $ftrace_stub, ftrace_trace_function 101 cmpq $ftrace_stub, ftrace_trace_function
107 jnz trace 102 jnz trace
103
104#ifdef CONFIG_FUNCTION_GRAPH_TRACER
105 cmpq $ftrace_stub, ftrace_graph_return
106 jnz ftrace_graph_caller
107
108 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
109 jnz ftrace_graph_caller
110#endif
111
108.globl ftrace_stub 112.globl ftrace_stub
109ftrace_stub: 113ftrace_stub:
110 retq 114 retq
111 115
112trace: 116trace:
113 /* taken from glibc */ 117 MCOUNT_SAVE_FRAME
114 subq $0x38, %rsp 118
119 movq 0x38(%rsp), %rdi
120 movq 8(%rbp), %rsi
121 subq $MCOUNT_INSN_SIZE, %rdi
122
123 call *ftrace_trace_function
124
125 MCOUNT_RESTORE_FRAME
126
127 jmp ftrace_stub
128END(mcount)
129#endif /* CONFIG_DYNAMIC_FTRACE */
130#endif /* CONFIG_FUNCTION_TRACER */
131
132#ifdef CONFIG_FUNCTION_GRAPH_TRACER
133ENTRY(ftrace_graph_caller)
134 cmpl $0, function_trace_stop
135 jne ftrace_stub
136
137 MCOUNT_SAVE_FRAME
138
139 leaq 8(%rbp), %rdi
140 movq 0x38(%rsp), %rsi
141 subq $MCOUNT_INSN_SIZE, %rsi
142
143 call prepare_ftrace_return
144
145 MCOUNT_RESTORE_FRAME
146
147 retq
148END(ftrace_graph_caller)
149
150
151.globl return_to_handler
152return_to_handler:
153 subq $80, %rsp
154
115 movq %rax, (%rsp) 155 movq %rax, (%rsp)
116 movq %rcx, 8(%rsp) 156 movq %rcx, 8(%rsp)
117 movq %rdx, 16(%rsp) 157 movq %rdx, 16(%rsp)
@@ -119,13 +159,14 @@ trace:
119 movq %rdi, 32(%rsp) 159 movq %rdi, 32(%rsp)
120 movq %r8, 40(%rsp) 160 movq %r8, 40(%rsp)
121 movq %r9, 48(%rsp) 161 movq %r9, 48(%rsp)
162 movq %r10, 56(%rsp)
163 movq %r11, 64(%rsp)
122 164
123 movq 0x38(%rsp), %rdi 165 call ftrace_return_to_handler
124 movq 8(%rbp), %rsi
125 subq $MCOUNT_INSN_SIZE, %rdi
126
127 call *ftrace_trace_function
128 166
167 movq %rax, 72(%rsp)
168 movq 64(%rsp), %r11
169 movq 56(%rsp), %r10
129 movq 48(%rsp), %r9 170 movq 48(%rsp), %r9
130 movq 40(%rsp), %r8 171 movq 40(%rsp), %r8
131 movq 32(%rsp), %rdi 172 movq 32(%rsp), %rdi
@@ -133,16 +174,14 @@ trace:
133 movq 16(%rsp), %rdx 174 movq 16(%rsp), %rdx
134 movq 8(%rsp), %rcx 175 movq 8(%rsp), %rcx
135 movq (%rsp), %rax 176 movq (%rsp), %rax
136 addq $0x38, %rsp 177 addq $72, %rsp
178 retq
179#endif
137 180
138 jmp ftrace_stub
139END(mcount)
140#endif /* CONFIG_DYNAMIC_FTRACE */
141#endif /* CONFIG_FUNCTION_TRACER */
142 181
143#ifndef CONFIG_PREEMPT 182#ifndef CONFIG_PREEMPT
144#define retint_kernel retint_restore_args 183#define retint_kernel retint_restore_args
145#endif 184#endif
146 185
147#ifdef CONFIG_PARAVIRT 186#ifdef CONFIG_PARAVIRT
148ENTRY(native_usergs_sysret64) 187ENTRY(native_usergs_sysret64)
@@ -161,29 +200,29 @@ ENTRY(native_usergs_sysret64)
161.endm 200.endm
162 201
163/* 202/*
164 * C code is not supposed to know about undefined top of stack. Every time 203 * C code is not supposed to know about undefined top of stack. Every time
165 * a C function with an pt_regs argument is called from the SYSCALL based 204 * a C function with an pt_regs argument is called from the SYSCALL based
166 * fast path FIXUP_TOP_OF_STACK is needed. 205 * fast path FIXUP_TOP_OF_STACK is needed.
167 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs 206 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
168 * manipulation. 207 * manipulation.
169 */ 208 */
170 209
171 /* %rsp:at FRAMEEND */ 210 /* %rsp:at FRAMEEND */
172 .macro FIXUP_TOP_OF_STACK tmp 211 .macro FIXUP_TOP_OF_STACK tmp offset=0
173 movq %gs:pda_oldrsp,\tmp 212 movq %gs:pda_oldrsp,\tmp
174 movq \tmp,RSP(%rsp) 213 movq \tmp,RSP+\offset(%rsp)
175 movq $__USER_DS,SS(%rsp) 214 movq $__USER_DS,SS+\offset(%rsp)
176 movq $__USER_CS,CS(%rsp) 215 movq $__USER_CS,CS+\offset(%rsp)
177 movq $-1,RCX(%rsp) 216 movq $-1,RCX+\offset(%rsp)
178 movq R11(%rsp),\tmp /* get eflags */ 217 movq R11+\offset(%rsp),\tmp /* get eflags */
179 movq \tmp,EFLAGS(%rsp) 218 movq \tmp,EFLAGS+\offset(%rsp)
180 .endm 219 .endm
181 220
182 .macro RESTORE_TOP_OF_STACK tmp,offset=0 221 .macro RESTORE_TOP_OF_STACK tmp offset=0
183 movq RSP-\offset(%rsp),\tmp 222 movq RSP+\offset(%rsp),\tmp
184 movq \tmp,%gs:pda_oldrsp 223 movq \tmp,%gs:pda_oldrsp
185 movq EFLAGS-\offset(%rsp),\tmp 224 movq EFLAGS+\offset(%rsp),\tmp
186 movq \tmp,R11-\offset(%rsp) 225 movq \tmp,R11+\offset(%rsp)
187 .endm 226 .endm
188 227
189 .macro FAKE_STACK_FRAME child_rip 228 .macro FAKE_STACK_FRAME child_rip
@@ -195,7 +234,7 @@ ENTRY(native_usergs_sysret64)
195 pushq %rax /* rsp */ 234 pushq %rax /* rsp */
196 CFI_ADJUST_CFA_OFFSET 8 235 CFI_ADJUST_CFA_OFFSET 8
197 CFI_REL_OFFSET rsp,0 236 CFI_REL_OFFSET rsp,0
198 pushq $(1<<9) /* eflags - interrupts on */ 237 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
199 CFI_ADJUST_CFA_OFFSET 8 238 CFI_ADJUST_CFA_OFFSET 8
200 /*CFI_REL_OFFSET rflags,0*/ 239 /*CFI_REL_OFFSET rflags,0*/
201 pushq $__KERNEL_CS /* cs */ 240 pushq $__KERNEL_CS /* cs */
@@ -213,62 +252,184 @@ ENTRY(native_usergs_sysret64)
213 CFI_ADJUST_CFA_OFFSET -(6*8) 252 CFI_ADJUST_CFA_OFFSET -(6*8)
214 .endm 253 .endm
215 254
216 .macro CFI_DEFAULT_STACK start=1 255/*
256 * initial frame state for interrupts (and exceptions without error code)
257 */
258 .macro EMPTY_FRAME start=1 offset=0
217 .if \start 259 .if \start
218 CFI_STARTPROC simple 260 CFI_STARTPROC simple
219 CFI_SIGNAL_FRAME 261 CFI_SIGNAL_FRAME
220 CFI_DEF_CFA rsp,SS+8 262 CFI_DEF_CFA rsp,8+\offset
221 .else 263 .else
222 CFI_DEF_CFA_OFFSET SS+8 264 CFI_DEF_CFA_OFFSET 8+\offset
223 .endif 265 .endif
224 CFI_REL_OFFSET r15,R15
225 CFI_REL_OFFSET r14,R14
226 CFI_REL_OFFSET r13,R13
227 CFI_REL_OFFSET r12,R12
228 CFI_REL_OFFSET rbp,RBP
229 CFI_REL_OFFSET rbx,RBX
230 CFI_REL_OFFSET r11,R11
231 CFI_REL_OFFSET r10,R10
232 CFI_REL_OFFSET r9,R9
233 CFI_REL_OFFSET r8,R8
234 CFI_REL_OFFSET rax,RAX
235 CFI_REL_OFFSET rcx,RCX
236 CFI_REL_OFFSET rdx,RDX
237 CFI_REL_OFFSET rsi,RSI
238 CFI_REL_OFFSET rdi,RDI
239 CFI_REL_OFFSET rip,RIP
240 /*CFI_REL_OFFSET cs,CS*/
241 /*CFI_REL_OFFSET rflags,EFLAGS*/
242 CFI_REL_OFFSET rsp,RSP
243 /*CFI_REL_OFFSET ss,SS*/
244 .endm 266 .endm
267
268/*
269 * initial frame state for interrupts (and exceptions without error code)
270 */
271 .macro INTR_FRAME start=1 offset=0
272 EMPTY_FRAME \start, SS+8+\offset-RIP
273 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
274 CFI_REL_OFFSET rsp, RSP+\offset-RIP
275 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
276 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
277 CFI_REL_OFFSET rip, RIP+\offset-RIP
278 .endm
279
280/*
281 * initial frame state for exceptions with error code (and interrupts
282 * with vector already pushed)
283 */
284 .macro XCPT_FRAME start=1 offset=0
285 INTR_FRAME \start, RIP+\offset-ORIG_RAX
286 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
287 .endm
288
245/* 289/*
246 * A newly forked process directly context switches into this. 290 * frame that enables calling into C.
247 */ 291 */
248/* rdi: prev */ 292 .macro PARTIAL_FRAME start=1 offset=0
293 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
294 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
295 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
296 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
297 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
298 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
299 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
300 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
301 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
302 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
303 .endm
304
305/*
306 * frame that enables passing a complete pt_regs to a C function.
307 */
308 .macro DEFAULT_FRAME start=1 offset=0
309 PARTIAL_FRAME \start, R11+\offset-R15
310 CFI_REL_OFFSET rbx, RBX+\offset
311 CFI_REL_OFFSET rbp, RBP+\offset
312 CFI_REL_OFFSET r12, R12+\offset
313 CFI_REL_OFFSET r13, R13+\offset
314 CFI_REL_OFFSET r14, R14+\offset
315 CFI_REL_OFFSET r15, R15+\offset
316 .endm
317
318/* save partial stack frame */
319ENTRY(save_args)
320 XCPT_FRAME
321 cld
322 movq_cfi rdi, RDI+16-ARGOFFSET
323 movq_cfi rsi, RSI+16-ARGOFFSET
324 movq_cfi rdx, RDX+16-ARGOFFSET
325 movq_cfi rcx, RCX+16-ARGOFFSET
326 movq_cfi rax, RAX+16-ARGOFFSET
327 movq_cfi r8, R8+16-ARGOFFSET
328 movq_cfi r9, R9+16-ARGOFFSET
329 movq_cfi r10, R10+16-ARGOFFSET
330 movq_cfi r11, R11+16-ARGOFFSET
331
332 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
333 movq_cfi rbp, 8 /* push %rbp */
334 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
335 testl $3, CS(%rdi)
336 je 1f
337 SWAPGS
338 /*
339 * irqcount is used to check if a CPU is already on an interrupt stack
340 * or not. While this is essentially redundant with preempt_count it is
341 * a little cheaper to use a separate counter in the PDA (short of
342 * moving irq_enter into assembly, which would be too much work)
343 */
3441: incl %gs:pda_irqcount
345 jne 2f
346 popq_cfi %rax /* move return address... */
347 mov %gs:pda_irqstackptr,%rsp
348 EMPTY_FRAME 0
349 pushq_cfi %rax /* ... to the new stack */
350 /*
351 * We entered an interrupt context - irqs are off:
352 */
3532: TRACE_IRQS_OFF
354 ret
355 CFI_ENDPROC
356END(save_args)
357
358ENTRY(save_rest)
359 PARTIAL_FRAME 1 REST_SKIP+8
360 movq 5*8+16(%rsp), %r11 /* save return address */
361 movq_cfi rbx, RBX+16
362 movq_cfi rbp, RBP+16
363 movq_cfi r12, R12+16
364 movq_cfi r13, R13+16
365 movq_cfi r14, R14+16
366 movq_cfi r15, R15+16
367 movq %r11, 8(%rsp) /* return address */
368 FIXUP_TOP_OF_STACK %r11, 16
369 ret
370 CFI_ENDPROC
371END(save_rest)
372
373/* save complete stack frame */
374ENTRY(save_paranoid)
375 XCPT_FRAME 1 RDI+8
376 cld
377 movq_cfi rdi, RDI+8
378 movq_cfi rsi, RSI+8
379 movq_cfi rdx, RDX+8
380 movq_cfi rcx, RCX+8
381 movq_cfi rax, RAX+8
382 movq_cfi r8, R8+8
383 movq_cfi r9, R9+8
384 movq_cfi r10, R10+8
385 movq_cfi r11, R11+8
386 movq_cfi rbx, RBX+8
387 movq_cfi rbp, RBP+8
388 movq_cfi r12, R12+8
389 movq_cfi r13, R13+8
390 movq_cfi r14, R14+8
391 movq_cfi r15, R15+8
392 movl $1,%ebx
393 movl $MSR_GS_BASE,%ecx
394 rdmsr
395 testl %edx,%edx
396 js 1f /* negative -> in kernel */
397 SWAPGS
398 xorl %ebx,%ebx
3991: ret
400 CFI_ENDPROC
401END(save_paranoid)
402
403/*
404 * A newly forked process directly context switches into this address.
405 *
406 * rdi: prev task we switched from
407 */
249ENTRY(ret_from_fork) 408ENTRY(ret_from_fork)
250 CFI_DEFAULT_STACK 409 DEFAULT_FRAME
410
251 push kernel_eflags(%rip) 411 push kernel_eflags(%rip)
252 CFI_ADJUST_CFA_OFFSET 8 412 CFI_ADJUST_CFA_OFFSET 8
253 popf # reset kernel eflags 413 popf # reset kernel eflags
254 CFI_ADJUST_CFA_OFFSET -8 414 CFI_ADJUST_CFA_OFFSET -8
255 call schedule_tail 415
416 call schedule_tail # rdi: 'prev' task parameter
417
256 GET_THREAD_INFO(%rcx) 418 GET_THREAD_INFO(%rcx)
257 testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),TI_flags(%rcx) 419
258 jnz rff_trace 420 CFI_REMEMBER_STATE
259rff_action:
260 RESTORE_REST 421 RESTORE_REST
261 testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread? 422
423 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
262 je int_ret_from_sys_call 424 je int_ret_from_sys_call
263 testl $_TIF_IA32,TI_flags(%rcx) 425
426 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
264 jnz int_ret_from_sys_call 427 jnz int_ret_from_sys_call
265 RESTORE_TOP_OF_STACK %rdi,ARGOFFSET 428
266 jmp ret_from_sys_call 429 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
267rff_trace: 430 jmp ret_from_sys_call # go to the SYSRET fastpath
268 movq %rsp,%rdi 431
269 call syscall_trace_leave 432 CFI_RESTORE_STATE
270 GET_THREAD_INFO(%rcx)
271 jmp rff_action
272 CFI_ENDPROC 433 CFI_ENDPROC
273END(ret_from_fork) 434END(ret_from_fork)
274 435
@@ -278,20 +439,20 @@ END(ret_from_fork)
278 * SYSCALL does not save anything on the stack and does not change the 439 * SYSCALL does not save anything on the stack and does not change the
279 * stack pointer. 440 * stack pointer.
280 */ 441 */
281 442
282/* 443/*
283 * Register setup: 444 * Register setup:
284 * rax system call number 445 * rax system call number
285 * rdi arg0 446 * rdi arg0
286 * rcx return address for syscall/sysret, C arg3 447 * rcx return address for syscall/sysret, C arg3
287 * rsi arg1 448 * rsi arg1
288 * rdx arg2 449 * rdx arg2
289 * r10 arg3 (--> moved to rcx for C) 450 * r10 arg3 (--> moved to rcx for C)
290 * r8 arg4 451 * r8 arg4
291 * r9 arg5 452 * r9 arg5
292 * r11 eflags for syscall/sysret, temporary for C 453 * r11 eflags for syscall/sysret, temporary for C
293 * r12-r15,rbp,rbx saved by C code, not touched. 454 * r12-r15,rbp,rbx saved by C code, not touched.
294 * 455 *
295 * Interrupts are off on entry. 456 * Interrupts are off on entry.
296 * Only called from user space. 457 * Only called from user space.
297 * 458 *
@@ -301,7 +462,7 @@ END(ret_from_fork)
301 * When user can change the frames always force IRET. That is because 462 * When user can change the frames always force IRET. That is because
302 * it deals with uncanonical addresses better. SYSRET has trouble 463 * it deals with uncanonical addresses better. SYSRET has trouble
303 * with them due to bugs in both AMD and Intel CPUs. 464 * with them due to bugs in both AMD and Intel CPUs.
304 */ 465 */
305 466
306ENTRY(system_call) 467ENTRY(system_call)
307 CFI_STARTPROC simple 468 CFI_STARTPROC simple
@@ -317,7 +478,7 @@ ENTRY(system_call)
317 */ 478 */
318ENTRY(system_call_after_swapgs) 479ENTRY(system_call_after_swapgs)
319 480
320 movq %rsp,%gs:pda_oldrsp 481 movq %rsp,%gs:pda_oldrsp
321 movq %gs:pda_kernelstack,%rsp 482 movq %gs:pda_kernelstack,%rsp
322 /* 483 /*
323 * No need to follow this irqs off/on section - it's straight 484 * No need to follow this irqs off/on section - it's straight
@@ -325,7 +486,7 @@ ENTRY(system_call_after_swapgs)
325 */ 486 */
326 ENABLE_INTERRUPTS(CLBR_NONE) 487 ENABLE_INTERRUPTS(CLBR_NONE)
327 SAVE_ARGS 8,1 488 SAVE_ARGS 8,1
328 movq %rax,ORIG_RAX-ARGOFFSET(%rsp) 489 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
329 movq %rcx,RIP-ARGOFFSET(%rsp) 490 movq %rcx,RIP-ARGOFFSET(%rsp)
330 CFI_REL_OFFSET rip,RIP-ARGOFFSET 491 CFI_REL_OFFSET rip,RIP-ARGOFFSET
331 GET_THREAD_INFO(%rcx) 492 GET_THREAD_INFO(%rcx)
@@ -339,19 +500,19 @@ system_call_fastpath:
339 movq %rax,RAX-ARGOFFSET(%rsp) 500 movq %rax,RAX-ARGOFFSET(%rsp)
340/* 501/*
341 * Syscall return path ending with SYSRET (fast path) 502 * Syscall return path ending with SYSRET (fast path)
342 * Has incomplete stack frame and undefined top of stack. 503 * Has incomplete stack frame and undefined top of stack.
343 */ 504 */
344ret_from_sys_call: 505ret_from_sys_call:
345 movl $_TIF_ALLWORK_MASK,%edi 506 movl $_TIF_ALLWORK_MASK,%edi
346 /* edi: flagmask */ 507 /* edi: flagmask */
347sysret_check: 508sysret_check:
348 LOCKDEP_SYS_EXIT 509 LOCKDEP_SYS_EXIT
349 GET_THREAD_INFO(%rcx) 510 GET_THREAD_INFO(%rcx)
350 DISABLE_INTERRUPTS(CLBR_NONE) 511 DISABLE_INTERRUPTS(CLBR_NONE)
351 TRACE_IRQS_OFF 512 TRACE_IRQS_OFF
352 movl TI_flags(%rcx),%edx 513 movl TI_flags(%rcx),%edx
353 andl %edi,%edx 514 andl %edi,%edx
354 jnz sysret_careful 515 jnz sysret_careful
355 CFI_REMEMBER_STATE 516 CFI_REMEMBER_STATE
356 /* 517 /*
357 * sysretq will re-enable interrupts: 518 * sysretq will re-enable interrupts:
@@ -366,7 +527,7 @@ sysret_check:
366 527
367 CFI_RESTORE_STATE 528 CFI_RESTORE_STATE
368 /* Handle reschedules */ 529 /* Handle reschedules */
369 /* edx: work, edi: workmask */ 530 /* edx: work, edi: workmask */
370sysret_careful: 531sysret_careful:
371 bt $TIF_NEED_RESCHED,%edx 532 bt $TIF_NEED_RESCHED,%edx
372 jnc sysret_signal 533 jnc sysret_signal
@@ -379,7 +540,7 @@ sysret_careful:
379 CFI_ADJUST_CFA_OFFSET -8 540 CFI_ADJUST_CFA_OFFSET -8
380 jmp sysret_check 541 jmp sysret_check
381 542
382 /* Handle a signal */ 543 /* Handle a signal */
383sysret_signal: 544sysret_signal:
384 TRACE_IRQS_ON 545 TRACE_IRQS_ON
385 ENABLE_INTERRUPTS(CLBR_NONE) 546 ENABLE_INTERRUPTS(CLBR_NONE)
@@ -388,17 +549,20 @@ sysret_signal:
388 jc sysret_audit 549 jc sysret_audit
389#endif 550#endif
390 /* edx: work flags (arg3) */ 551 /* edx: work flags (arg3) */
391 leaq do_notify_resume(%rip),%rax
392 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1 552 leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
393 xorl %esi,%esi # oldset -> arg2 553 xorl %esi,%esi # oldset -> arg2
394 call ptregscall_common 554 SAVE_REST
555 FIXUP_TOP_OF_STACK %r11
556 call do_notify_resume
557 RESTORE_TOP_OF_STACK %r11
558 RESTORE_REST
395 movl $_TIF_WORK_MASK,%edi 559 movl $_TIF_WORK_MASK,%edi
396 /* Use IRET because user could have changed frame. This 560 /* Use IRET because user could have changed frame. This
397 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */ 561 works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
398 DISABLE_INTERRUPTS(CLBR_NONE) 562 DISABLE_INTERRUPTS(CLBR_NONE)
399 TRACE_IRQS_OFF 563 TRACE_IRQS_OFF
400 jmp int_with_check 564 jmp int_with_check
401 565
402badsys: 566badsys:
403 movq $-ENOSYS,RAX-ARGOFFSET(%rsp) 567 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
404 jmp ret_from_sys_call 568 jmp ret_from_sys_call
@@ -437,7 +601,7 @@ sysret_audit:
437#endif /* CONFIG_AUDITSYSCALL */ 601#endif /* CONFIG_AUDITSYSCALL */
438 602
439 /* Do syscall tracing */ 603 /* Do syscall tracing */
440tracesys: 604tracesys:
441#ifdef CONFIG_AUDITSYSCALL 605#ifdef CONFIG_AUDITSYSCALL
442 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx) 606 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
443 jz auditsys 607 jz auditsys
@@ -460,8 +624,8 @@ tracesys:
460 call *sys_call_table(,%rax,8) 624 call *sys_call_table(,%rax,8)
461 movq %rax,RAX-ARGOFFSET(%rsp) 625 movq %rax,RAX-ARGOFFSET(%rsp)
462 /* Use IRET because user could have changed frame */ 626 /* Use IRET because user could have changed frame */
463 627
464/* 628/*
465 * Syscall return path ending with IRET. 629 * Syscall return path ending with IRET.
466 * Has correct top of stack, but partial stack frame. 630 * Has correct top of stack, but partial stack frame.
467 */ 631 */
@@ -505,18 +669,18 @@ int_very_careful:
505 TRACE_IRQS_ON 669 TRACE_IRQS_ON
506 ENABLE_INTERRUPTS(CLBR_NONE) 670 ENABLE_INTERRUPTS(CLBR_NONE)
507 SAVE_REST 671 SAVE_REST
508 /* Check for syscall exit trace */ 672 /* Check for syscall exit trace */
509 testl $_TIF_WORK_SYSCALL_EXIT,%edx 673 testl $_TIF_WORK_SYSCALL_EXIT,%edx
510 jz int_signal 674 jz int_signal
511 pushq %rdi 675 pushq %rdi
512 CFI_ADJUST_CFA_OFFSET 8 676 CFI_ADJUST_CFA_OFFSET 8
513 leaq 8(%rsp),%rdi # &ptregs -> arg1 677 leaq 8(%rsp),%rdi # &ptregs -> arg1
514 call syscall_trace_leave 678 call syscall_trace_leave
515 popq %rdi 679 popq %rdi
516 CFI_ADJUST_CFA_OFFSET -8 680 CFI_ADJUST_CFA_OFFSET -8
517 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi 681 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
518 jmp int_restore_rest 682 jmp int_restore_rest
519 683
520int_signal: 684int_signal:
521 testl $_TIF_DO_NOTIFY_MASK,%edx 685 testl $_TIF_DO_NOTIFY_MASK,%edx
522 jz 1f 686 jz 1f
@@ -531,22 +695,24 @@ int_restore_rest:
531 jmp int_with_check 695 jmp int_with_check
532 CFI_ENDPROC 696 CFI_ENDPROC
533END(system_call) 697END(system_call)
534 698
535/* 699/*
536 * Certain special system calls that need to save a complete full stack frame. 700 * Certain special system calls that need to save a complete full stack frame.
537 */ 701 */
538
539 .macro PTREGSCALL label,func,arg 702 .macro PTREGSCALL label,func,arg
540 .globl \label 703ENTRY(\label)
541\label: 704 PARTIAL_FRAME 1 8 /* offset 8: return address */
542 leaq \func(%rip),%rax 705 subq $REST_SKIP, %rsp
543 leaq -ARGOFFSET+8(%rsp),\arg /* 8 for return address */ 706 CFI_ADJUST_CFA_OFFSET REST_SKIP
544 jmp ptregscall_common 707 call save_rest
708 DEFAULT_FRAME 0 8 /* offset 8: return address */
709 leaq 8(%rsp), \arg /* pt_regs pointer */
710 call \func
711 jmp ptregscall_common
712 CFI_ENDPROC
545END(\label) 713END(\label)
546 .endm 714 .endm
547 715
548 CFI_STARTPROC
549
550 PTREGSCALL stub_clone, sys_clone, %r8 716 PTREGSCALL stub_clone, sys_clone, %r8
551 PTREGSCALL stub_fork, sys_fork, %rdi 717 PTREGSCALL stub_fork, sys_fork, %rdi
552 PTREGSCALL stub_vfork, sys_vfork, %rdi 718 PTREGSCALL stub_vfork, sys_vfork, %rdi
@@ -554,25 +720,18 @@ END(\label)
554 PTREGSCALL stub_iopl, sys_iopl, %rsi 720 PTREGSCALL stub_iopl, sys_iopl, %rsi
555 721
556ENTRY(ptregscall_common) 722ENTRY(ptregscall_common)
557 popq %r11 723 DEFAULT_FRAME 1 8 /* offset 8: return address */
558 CFI_ADJUST_CFA_OFFSET -8 724 RESTORE_TOP_OF_STACK %r11, 8
559 CFI_REGISTER rip, r11 725 movq_cfi_restore R15+8, r15
560 SAVE_REST 726 movq_cfi_restore R14+8, r14
561 movq %r11, %r15 727 movq_cfi_restore R13+8, r13
562 CFI_REGISTER rip, r15 728 movq_cfi_restore R12+8, r12
563 FIXUP_TOP_OF_STACK %r11 729 movq_cfi_restore RBP+8, rbp
564 call *%rax 730 movq_cfi_restore RBX+8, rbx
565 RESTORE_TOP_OF_STACK %r11 731 ret $REST_SKIP /* pop extended registers */
566 movq %r15, %r11
567 CFI_REGISTER rip, r11
568 RESTORE_REST
569 pushq %r11
570 CFI_ADJUST_CFA_OFFSET 8
571 CFI_REL_OFFSET rip, 0
572 ret
573 CFI_ENDPROC 732 CFI_ENDPROC
574END(ptregscall_common) 733END(ptregscall_common)
575 734
576ENTRY(stub_execve) 735ENTRY(stub_execve)
577 CFI_STARTPROC 736 CFI_STARTPROC
578 popq %r11 737 popq %r11
@@ -588,11 +747,11 @@ ENTRY(stub_execve)
588 jmp int_ret_from_sys_call 747 jmp int_ret_from_sys_call
589 CFI_ENDPROC 748 CFI_ENDPROC
590END(stub_execve) 749END(stub_execve)
591 750
592/* 751/*
593 * sigreturn is special because it needs to restore all registers on return. 752 * sigreturn is special because it needs to restore all registers on return.
594 * This cannot be done with SYSRET, so use the IRET return path instead. 753 * This cannot be done with SYSRET, so use the IRET return path instead.
595 */ 754 */
596ENTRY(stub_rt_sigreturn) 755ENTRY(stub_rt_sigreturn)
597 CFI_STARTPROC 756 CFI_STARTPROC
598 addq $8, %rsp 757 addq $8, %rsp
@@ -608,70 +767,70 @@ ENTRY(stub_rt_sigreturn)
608END(stub_rt_sigreturn) 767END(stub_rt_sigreturn)
609 768
610/* 769/*
611 * initial frame state for interrupts and exceptions 770 * Build the entry stubs and pointer table with some assembler magic.
771 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
772 * single cache line on all modern x86 implementations.
612 */ 773 */
613 .macro _frame ref 774 .section .init.rodata,"a"
614 CFI_STARTPROC simple 775ENTRY(interrupt)
615 CFI_SIGNAL_FRAME 776 .text
616 CFI_DEF_CFA rsp,SS+8-\ref 777 .p2align 5
617 /*CFI_REL_OFFSET ss,SS-\ref*/ 778 .p2align CONFIG_X86_L1_CACHE_SHIFT
618 CFI_REL_OFFSET rsp,RSP-\ref 779ENTRY(irq_entries_start)
619 /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/ 780 INTR_FRAME
620 /*CFI_REL_OFFSET cs,CS-\ref*/ 781vector=FIRST_EXTERNAL_VECTOR
621 CFI_REL_OFFSET rip,RIP-\ref 782.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
622 .endm 783 .balign 32
784 .rept 7
785 .if vector < NR_VECTORS
786 .if vector <> FIRST_EXTERNAL_VECTOR
787 CFI_ADJUST_CFA_OFFSET -8
788 .endif
7891: pushq $(~vector+0x80) /* Note: always in signed byte range */
790 CFI_ADJUST_CFA_OFFSET 8
791 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
792 jmp 2f
793 .endif
794 .previous
795 .quad 1b
796 .text
797vector=vector+1
798 .endif
799 .endr
8002: jmp common_interrupt
801.endr
802 CFI_ENDPROC
803END(irq_entries_start)
623 804
624/* initial frame state for interrupts (and exceptions without error code) */ 805.previous
625#define INTR_FRAME _frame RIP 806END(interrupt)
626/* initial frame state for exceptions with error code (and interrupts with 807.previous
627 vector already pushed) */
628#define XCPT_FRAME _frame ORIG_RAX
629 808
630/* 809/*
631 * Interrupt entry/exit. 810 * Interrupt entry/exit.
632 * 811 *
633 * Interrupt entry points save only callee clobbered registers in fast path. 812 * Interrupt entry points save only callee clobbered registers in fast path.
634 * 813 *
635 * Entry runs with interrupts off. 814 * Entry runs with interrupts off.
636 */ 815 */
637 816
638/* 0(%rsp): interrupt number */ 817/* 0(%rsp): ~(interrupt number) */
639 .macro interrupt func 818 .macro interrupt func
640 cld 819 subq $10*8, %rsp
641 SAVE_ARGS 820 CFI_ADJUST_CFA_OFFSET 10*8
642 leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler 821 call save_args
643 pushq %rbp 822 PARTIAL_FRAME 0
644 /*
645 * Save rbp twice: One is for marking the stack frame, as usual, and the
646 * other, to fill pt_regs properly. This is because bx comes right
647 * before the last saved register in that structure, and not bp. If the
648 * base pointer were in the place bx is today, this would not be needed.
649 */
650 movq %rbp, -8(%rsp)
651 CFI_ADJUST_CFA_OFFSET 8
652 CFI_REL_OFFSET rbp, 0
653 movq %rsp,%rbp
654 CFI_DEF_CFA_REGISTER rbp
655 testl $3,CS(%rdi)
656 je 1f
657 SWAPGS
658 /* irqcount is used to check if a CPU is already on an interrupt
659 stack or not. While this is essentially redundant with preempt_count
660 it is a little cheaper to use a separate counter in the PDA
661 (short of moving irq_enter into assembly, which would be too
662 much work) */
6631: incl %gs:pda_irqcount
664 cmoveq %gs:pda_irqstackptr,%rsp
665 push %rbp # backlink for old unwinder
666 /*
667 * We entered an interrupt context - irqs are off:
668 */
669 TRACE_IRQS_OFF
670 call \func 823 call \func
671 .endm 824 .endm
672 825
673ENTRY(common_interrupt) 826 /*
827 * The interrupt stubs push (~vector+0x80) onto the stack and
828 * then jump to common_interrupt.
829 */
830 .p2align CONFIG_X86_L1_CACHE_SHIFT
831common_interrupt:
674 XCPT_FRAME 832 XCPT_FRAME
833 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
675 interrupt do_IRQ 834 interrupt do_IRQ
676 /* 0(%rsp): oldrsp-ARGOFFSET */ 835 /* 0(%rsp): oldrsp-ARGOFFSET */
677ret_from_intr: 836ret_from_intr:
@@ -685,12 +844,12 @@ exit_intr:
685 GET_THREAD_INFO(%rcx) 844 GET_THREAD_INFO(%rcx)
686 testl $3,CS-ARGOFFSET(%rsp) 845 testl $3,CS-ARGOFFSET(%rsp)
687 je retint_kernel 846 je retint_kernel
688 847
689 /* Interrupt came from user space */ 848 /* Interrupt came from user space */
690 /* 849 /*
691 * Has a correct top of stack, but a partial stack frame 850 * Has a correct top of stack, but a partial stack frame
692 * %rcx: thread info. Interrupts off. 851 * %rcx: thread info. Interrupts off.
693 */ 852 */
694retint_with_reschedule: 853retint_with_reschedule:
695 movl $_TIF_WORK_MASK,%edi 854 movl $_TIF_WORK_MASK,%edi
696retint_check: 855retint_check:
@@ -763,20 +922,20 @@ retint_careful:
763 pushq %rdi 922 pushq %rdi
764 CFI_ADJUST_CFA_OFFSET 8 923 CFI_ADJUST_CFA_OFFSET 8
765 call schedule 924 call schedule
766 popq %rdi 925 popq %rdi
767 CFI_ADJUST_CFA_OFFSET -8 926 CFI_ADJUST_CFA_OFFSET -8
768 GET_THREAD_INFO(%rcx) 927 GET_THREAD_INFO(%rcx)
769 DISABLE_INTERRUPTS(CLBR_NONE) 928 DISABLE_INTERRUPTS(CLBR_NONE)
770 TRACE_IRQS_OFF 929 TRACE_IRQS_OFF
771 jmp retint_check 930 jmp retint_check
772 931
773retint_signal: 932retint_signal:
774 testl $_TIF_DO_NOTIFY_MASK,%edx 933 testl $_TIF_DO_NOTIFY_MASK,%edx
775 jz retint_swapgs 934 jz retint_swapgs
776 TRACE_IRQS_ON 935 TRACE_IRQS_ON
777 ENABLE_INTERRUPTS(CLBR_NONE) 936 ENABLE_INTERRUPTS(CLBR_NONE)
778 SAVE_REST 937 SAVE_REST
779 movq $-1,ORIG_RAX(%rsp) 938 movq $-1,ORIG_RAX(%rsp)
780 xorl %esi,%esi # oldset 939 xorl %esi,%esi # oldset
781 movq %rsp,%rdi # &pt_regs 940 movq %rsp,%rdi # &pt_regs
782 call do_notify_resume 941 call do_notify_resume
@@ -798,324 +957,211 @@ ENTRY(retint_kernel)
798 jnc retint_restore_args 957 jnc retint_restore_args
799 call preempt_schedule_irq 958 call preempt_schedule_irq
800 jmp exit_intr 959 jmp exit_intr
801#endif 960#endif
802 961
803 CFI_ENDPROC 962 CFI_ENDPROC
804END(common_interrupt) 963END(common_interrupt)
805 964
806/* 965/*
807 * APIC interrupts. 966 * APIC interrupts.
808 */ 967 */
809 .macro apicinterrupt num,func 968.macro apicinterrupt num sym do_sym
969ENTRY(\sym)
810 INTR_FRAME 970 INTR_FRAME
811 pushq $~(\num) 971 pushq $~(\num)
812 CFI_ADJUST_CFA_OFFSET 8 972 CFI_ADJUST_CFA_OFFSET 8
813 interrupt \func 973 interrupt \do_sym
814 jmp ret_from_intr 974 jmp ret_from_intr
815 CFI_ENDPROC 975 CFI_ENDPROC
816 .endm 976END(\sym)
817 977.endm
818ENTRY(thermal_interrupt)
819 apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
820END(thermal_interrupt)
821
822ENTRY(threshold_interrupt)
823 apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
824END(threshold_interrupt)
825
826#ifdef CONFIG_SMP
827ENTRY(reschedule_interrupt)
828 apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
829END(reschedule_interrupt)
830
831 .macro INVALIDATE_ENTRY num
832ENTRY(invalidate_interrupt\num)
833 apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
834END(invalidate_interrupt\num)
835 .endm
836 978
837 INVALIDATE_ENTRY 0 979#ifdef CONFIG_SMP
838 INVALIDATE_ENTRY 1 980apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
839 INVALIDATE_ENTRY 2 981 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
840 INVALIDATE_ENTRY 3
841 INVALIDATE_ENTRY 4
842 INVALIDATE_ENTRY 5
843 INVALIDATE_ENTRY 6
844 INVALIDATE_ENTRY 7
845
846ENTRY(call_function_interrupt)
847 apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
848END(call_function_interrupt)
849ENTRY(call_function_single_interrupt)
850 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR,smp_call_function_single_interrupt
851END(call_function_single_interrupt)
852ENTRY(irq_move_cleanup_interrupt)
853 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
854END(irq_move_cleanup_interrupt)
855#endif 982#endif
856 983
857ENTRY(apic_timer_interrupt) 984apicinterrupt UV_BAU_MESSAGE \
858 apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt 985 uv_bau_message_intr1 uv_bau_message_interrupt
859END(apic_timer_interrupt) 986apicinterrupt LOCAL_TIMER_VECTOR \
987 apic_timer_interrupt smp_apic_timer_interrupt
988
989#ifdef CONFIG_SMP
990apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
991 invalidate_interrupt0 smp_invalidate_interrupt
992apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
993 invalidate_interrupt1 smp_invalidate_interrupt
994apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
995 invalidate_interrupt2 smp_invalidate_interrupt
996apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
997 invalidate_interrupt3 smp_invalidate_interrupt
998apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
999 invalidate_interrupt4 smp_invalidate_interrupt
1000apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
1001 invalidate_interrupt5 smp_invalidate_interrupt
1002apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
1003 invalidate_interrupt6 smp_invalidate_interrupt
1004apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
1005 invalidate_interrupt7 smp_invalidate_interrupt
1006#endif
860 1007
861ENTRY(uv_bau_message_intr1) 1008apicinterrupt THRESHOLD_APIC_VECTOR \
862 apicinterrupt 220,uv_bau_message_interrupt 1009 threshold_interrupt mce_threshold_interrupt
863END(uv_bau_message_intr1) 1010apicinterrupt THERMAL_APIC_VECTOR \
1011 thermal_interrupt smp_thermal_interrupt
1012
1013#ifdef CONFIG_SMP
1014apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1015 call_function_single_interrupt smp_call_function_single_interrupt
1016apicinterrupt CALL_FUNCTION_VECTOR \
1017 call_function_interrupt smp_call_function_interrupt
1018apicinterrupt RESCHEDULE_VECTOR \
1019 reschedule_interrupt smp_reschedule_interrupt
1020#endif
864 1021
865ENTRY(error_interrupt) 1022apicinterrupt ERROR_APIC_VECTOR \
866 apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt 1023 error_interrupt smp_error_interrupt
867END(error_interrupt) 1024apicinterrupt SPURIOUS_APIC_VECTOR \
1025 spurious_interrupt smp_spurious_interrupt
868 1026
869ENTRY(spurious_interrupt)
870 apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
871END(spurious_interrupt)
872
873/* 1027/*
874 * Exception entry points. 1028 * Exception entry points.
875 */ 1029 */
876 .macro zeroentry sym 1030.macro zeroentry sym do_sym
1031ENTRY(\sym)
877 INTR_FRAME 1032 INTR_FRAME
878 PARAVIRT_ADJUST_EXCEPTION_FRAME 1033 PARAVIRT_ADJUST_EXCEPTION_FRAME
879 pushq $0 /* push error code/oldrax */ 1034 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
880 CFI_ADJUST_CFA_OFFSET 8 1035 subq $15*8,%rsp
881 pushq %rax /* push real oldrax to the rdi slot */ 1036 CFI_ADJUST_CFA_OFFSET 15*8
882 CFI_ADJUST_CFA_OFFSET 8 1037 call error_entry
883 CFI_REL_OFFSET rax,0 1038 DEFAULT_FRAME 0
884 leaq \sym(%rip),%rax 1039 movq %rsp,%rdi /* pt_regs pointer */
885 jmp error_entry 1040 xorl %esi,%esi /* no error code */
1041 call \do_sym
1042 jmp error_exit /* %ebx: no swapgs flag */
886 CFI_ENDPROC 1043 CFI_ENDPROC
887 .endm 1044END(\sym)
1045.endm
888 1046
889 .macro errorentry sym 1047.macro paranoidzeroentry sym do_sym
890 XCPT_FRAME 1048ENTRY(\sym)
1049 INTR_FRAME
891 PARAVIRT_ADJUST_EXCEPTION_FRAME 1050 PARAVIRT_ADJUST_EXCEPTION_FRAME
892 pushq %rax 1051 pushq $-1 /* ORIG_RAX: no syscall to restart */
893 CFI_ADJUST_CFA_OFFSET 8 1052 CFI_ADJUST_CFA_OFFSET 8
894 CFI_REL_OFFSET rax,0 1053 subq $15*8, %rsp
895 leaq \sym(%rip),%rax 1054 call save_paranoid
896 jmp error_entry 1055 TRACE_IRQS_OFF
1056 movq %rsp,%rdi /* pt_regs pointer */
1057 xorl %esi,%esi /* no error code */
1058 call \do_sym
1059 jmp paranoid_exit /* %ebx: no swapgs flag */
897 CFI_ENDPROC 1060 CFI_ENDPROC
898 .endm 1061END(\sym)
1062.endm
899 1063
900 /* error code is on the stack already */ 1064.macro paranoidzeroentry_ist sym do_sym ist
901 /* handle NMI like exceptions that can happen everywhere */ 1065ENTRY(\sym)
902 .macro paranoidentry sym, ist=0, irqtrace=1 1066 INTR_FRAME
903 SAVE_ALL 1067 PARAVIRT_ADJUST_EXCEPTION_FRAME
904 cld 1068 pushq $-1 /* ORIG_RAX: no syscall to restart */
905 movl $1,%ebx 1069 CFI_ADJUST_CFA_OFFSET 8
906 movl $MSR_GS_BASE,%ecx 1070 subq $15*8, %rsp
907 rdmsr 1071 call save_paranoid
908 testl %edx,%edx
909 js 1f
910 SWAPGS
911 xorl %ebx,%ebx
9121:
913 .if \ist
914 movq %gs:pda_data_offset, %rbp
915 .endif
916 .if \irqtrace
917 TRACE_IRQS_OFF
918 .endif
919 movq %rsp,%rdi
920 movq ORIG_RAX(%rsp),%rsi
921 movq $-1,ORIG_RAX(%rsp)
922 .if \ist
923 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
924 .endif
925 call \sym
926 .if \ist
927 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
928 .endif
929 DISABLE_INTERRUPTS(CLBR_NONE)
930 .if \irqtrace
931 TRACE_IRQS_OFF 1072 TRACE_IRQS_OFF
932 .endif 1073 movq %rsp,%rdi /* pt_regs pointer */
933 .endm 1074 xorl %esi,%esi /* no error code */
1075 movq %gs:pda_data_offset, %rbp
1076 subq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1077 call \do_sym
1078 addq $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
1079 jmp paranoid_exit /* %ebx: no swapgs flag */
1080 CFI_ENDPROC
1081END(\sym)
1082.endm
934 1083
935 /* 1084.macro errorentry sym do_sym
936 * "Paranoid" exit path from exception stack. 1085ENTRY(\sym)
937 * Paranoid because this is used by NMIs and cannot take 1086 XCPT_FRAME
938 * any kernel state for granted. 1087 PARAVIRT_ADJUST_EXCEPTION_FRAME
939 * We don't do kernel preemption checks here, because only 1088 subq $15*8,%rsp
940 * NMI should be common and it does not enable IRQs and 1089 CFI_ADJUST_CFA_OFFSET 15*8
941 * cannot get reschedule ticks. 1090 call error_entry
942 * 1091 DEFAULT_FRAME 0
943 * "trace" is 0 for the NMI handler only, because irq-tracing 1092 movq %rsp,%rdi /* pt_regs pointer */
944 * is fundamentally NMI-unsafe. (we cannot change the soft and 1093 movq ORIG_RAX(%rsp),%rsi /* get error code */
945 * hard flags at once, atomically) 1094 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
946 */ 1095 call \do_sym
947 .macro paranoidexit trace=1 1096 jmp error_exit /* %ebx: no swapgs flag */
948 /* ebx: no swapgs flag */
949paranoid_exit\trace:
950 testl %ebx,%ebx /* swapgs needed? */
951 jnz paranoid_restore\trace
952 testl $3,CS(%rsp)
953 jnz paranoid_userspace\trace
954paranoid_swapgs\trace:
955 .if \trace
956 TRACE_IRQS_IRETQ 0
957 .endif
958 SWAPGS_UNSAFE_STACK
959paranoid_restore\trace:
960 RESTORE_ALL 8
961 jmp irq_return
962paranoid_userspace\trace:
963 GET_THREAD_INFO(%rcx)
964 movl TI_flags(%rcx),%ebx
965 andl $_TIF_WORK_MASK,%ebx
966 jz paranoid_swapgs\trace
967 movq %rsp,%rdi /* &pt_regs */
968 call sync_regs
969 movq %rax,%rsp /* switch stack for scheduling */
970 testl $_TIF_NEED_RESCHED,%ebx
971 jnz paranoid_schedule\trace
972 movl %ebx,%edx /* arg3: thread flags */
973 .if \trace
974 TRACE_IRQS_ON
975 .endif
976 ENABLE_INTERRUPTS(CLBR_NONE)
977 xorl %esi,%esi /* arg2: oldset */
978 movq %rsp,%rdi /* arg1: &pt_regs */
979 call do_notify_resume
980 DISABLE_INTERRUPTS(CLBR_NONE)
981 .if \trace
982 TRACE_IRQS_OFF
983 .endif
984 jmp paranoid_userspace\trace
985paranoid_schedule\trace:
986 .if \trace
987 TRACE_IRQS_ON
988 .endif
989 ENABLE_INTERRUPTS(CLBR_ANY)
990 call schedule
991 DISABLE_INTERRUPTS(CLBR_ANY)
992 .if \trace
993 TRACE_IRQS_OFF
994 .endif
995 jmp paranoid_userspace\trace
996 CFI_ENDPROC 1097 CFI_ENDPROC
997 .endm 1098END(\sym)
1099.endm
998 1100
999/* 1101 /* error code is on the stack already */
1000 * Exception entry point. This expects an error code/orig_rax on the stack 1102.macro paranoiderrorentry sym do_sym
1001 * and the exception handler in %rax. 1103ENTRY(\sym)
1002 */ 1104 XCPT_FRAME
1003KPROBE_ENTRY(error_entry) 1105 PARAVIRT_ADJUST_EXCEPTION_FRAME
1004 _frame RDI 1106 subq $15*8,%rsp
1005 CFI_REL_OFFSET rax,0 1107 CFI_ADJUST_CFA_OFFSET 15*8
1006 /* rdi slot contains rax, oldrax contains error code */ 1108 call save_paranoid
1007 cld 1109 DEFAULT_FRAME 0
1008 subq $14*8,%rsp
1009 CFI_ADJUST_CFA_OFFSET (14*8)
1010 movq %rsi,13*8(%rsp)
1011 CFI_REL_OFFSET rsi,RSI
1012 movq 14*8(%rsp),%rsi /* load rax from rdi slot */
1013 CFI_REGISTER rax,rsi
1014 movq %rdx,12*8(%rsp)
1015 CFI_REL_OFFSET rdx,RDX
1016 movq %rcx,11*8(%rsp)
1017 CFI_REL_OFFSET rcx,RCX
1018 movq %rsi,10*8(%rsp) /* store rax */
1019 CFI_REL_OFFSET rax,RAX
1020 movq %r8, 9*8(%rsp)
1021 CFI_REL_OFFSET r8,R8
1022 movq %r9, 8*8(%rsp)
1023 CFI_REL_OFFSET r9,R9
1024 movq %r10,7*8(%rsp)
1025 CFI_REL_OFFSET r10,R10
1026 movq %r11,6*8(%rsp)
1027 CFI_REL_OFFSET r11,R11
1028 movq %rbx,5*8(%rsp)
1029 CFI_REL_OFFSET rbx,RBX
1030 movq %rbp,4*8(%rsp)
1031 CFI_REL_OFFSET rbp,RBP
1032 movq %r12,3*8(%rsp)
1033 CFI_REL_OFFSET r12,R12
1034 movq %r13,2*8(%rsp)
1035 CFI_REL_OFFSET r13,R13
1036 movq %r14,1*8(%rsp)
1037 CFI_REL_OFFSET r14,R14
1038 movq %r15,(%rsp)
1039 CFI_REL_OFFSET r15,R15
1040 xorl %ebx,%ebx
1041 testl $3,CS(%rsp)
1042 je error_kernelspace
1043error_swapgs:
1044 SWAPGS
1045error_sti:
1046 TRACE_IRQS_OFF
1047 movq %rdi,RDI(%rsp)
1048 CFI_REL_OFFSET rdi,RDI
1049 movq %rsp,%rdi
1050 movq ORIG_RAX(%rsp),%rsi /* get error code */
1051 movq $-1,ORIG_RAX(%rsp)
1052 call *%rax
1053 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1054error_exit:
1055 movl %ebx,%eax
1056 RESTORE_REST
1057 DISABLE_INTERRUPTS(CLBR_NONE)
1058 TRACE_IRQS_OFF 1110 TRACE_IRQS_OFF
1059 GET_THREAD_INFO(%rcx) 1111 movq %rsp,%rdi /* pt_regs pointer */
1060 testl %eax,%eax 1112 movq ORIG_RAX(%rsp),%rsi /* get error code */
1061 jne retint_kernel 1113 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1062 LOCKDEP_SYS_EXIT_IRQ 1114 call \do_sym
1063 movl TI_flags(%rcx),%edx 1115 jmp paranoid_exit /* %ebx: no swapgs flag */
1064 movl $_TIF_WORK_MASK,%edi
1065 andl %edi,%edx
1066 jnz retint_careful
1067 jmp retint_swapgs
1068 CFI_ENDPROC 1116 CFI_ENDPROC
1117END(\sym)
1118.endm
1069 1119
1070error_kernelspace: 1120zeroentry divide_error do_divide_error
1071 incl %ebx 1121zeroentry overflow do_overflow
1072 /* There are two places in the kernel that can potentially fault with 1122zeroentry bounds do_bounds
1073 usergs. Handle them here. The exception handlers after 1123zeroentry invalid_op do_invalid_op
1074 iret run with kernel gs again, so don't set the user space flag. 1124zeroentry device_not_available do_device_not_available
1075 B stepping K8s sometimes report an truncated RIP for IRET 1125paranoiderrorentry double_fault do_double_fault
1076 exceptions returning to compat mode. Check for these here too. */ 1126zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1077 leaq irq_return(%rip),%rcx 1127errorentry invalid_TSS do_invalid_TSS
1078 cmpq %rcx,RIP(%rsp) 1128errorentry segment_not_present do_segment_not_present
1079 je error_swapgs 1129zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1080 movl %ecx,%ecx /* zero extend */ 1130zeroentry coprocessor_error do_coprocessor_error
1081 cmpq %rcx,RIP(%rsp) 1131errorentry alignment_check do_alignment_check
1082 je error_swapgs 1132zeroentry simd_coprocessor_error do_simd_coprocessor_error
1083 cmpq $gs_change,RIP(%rsp) 1133
1084 je error_swapgs 1134 /* Reload gs selector with exception handling */
1085 jmp error_sti 1135 /* edi: new selector */
1086KPROBE_END(error_entry)
1087
1088 /* Reload gs selector with exception handling */
1089 /* edi: new selector */
1090ENTRY(native_load_gs_index) 1136ENTRY(native_load_gs_index)
1091 CFI_STARTPROC 1137 CFI_STARTPROC
1092 pushf 1138 pushf
1093 CFI_ADJUST_CFA_OFFSET 8 1139 CFI_ADJUST_CFA_OFFSET 8
1094 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI)) 1140 DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
1095 SWAPGS 1141 SWAPGS
1096gs_change: 1142gs_change:
1097 movl %edi,%gs 1143 movl %edi,%gs
10982: mfence /* workaround */ 11442: mfence /* workaround */
1099 SWAPGS 1145 SWAPGS
1100 popf 1146 popf
1101 CFI_ADJUST_CFA_OFFSET -8 1147 CFI_ADJUST_CFA_OFFSET -8
1102 ret 1148 ret
1103 CFI_ENDPROC 1149 CFI_ENDPROC
1104ENDPROC(native_load_gs_index) 1150END(native_load_gs_index)
1105 1151
1106 .section __ex_table,"a" 1152 .section __ex_table,"a"
1107 .align 8 1153 .align 8
1108 .quad gs_change,bad_gs 1154 .quad gs_change,bad_gs
1109 .previous 1155 .previous
1110 .section .fixup,"ax" 1156 .section .fixup,"ax"
1111 /* running with kernelgs */ 1157 /* running with kernelgs */
1112bad_gs: 1158bad_gs:
1113 SWAPGS /* switch back to user gs */ 1159 SWAPGS /* switch back to user gs */
1114 xorl %eax,%eax 1160 xorl %eax,%eax
1115 movl %eax,%gs 1161 movl %eax,%gs
1116 jmp 2b 1162 jmp 2b
1117 .previous 1163 .previous
1118 1164
1119/* 1165/*
1120 * Create a kernel thread. 1166 * Create a kernel thread.
1121 * 1167 *
@@ -1138,7 +1184,7 @@ ENTRY(kernel_thread)
1138 1184
1139 xorl %r8d,%r8d 1185 xorl %r8d,%r8d
1140 xorl %r9d,%r9d 1186 xorl %r9d,%r9d
1141 1187
1142 # clone now 1188 # clone now
1143 call do_fork 1189 call do_fork
1144 movq %rax,RAX(%rsp) 1190 movq %rax,RAX(%rsp)
@@ -1149,15 +1195,15 @@ ENTRY(kernel_thread)
1149 * so internally to the x86_64 port you can rely on kernel_thread() 1195 * so internally to the x86_64 port you can rely on kernel_thread()
1150 * not to reschedule the child before returning, this avoids the need 1196 * not to reschedule the child before returning, this avoids the need
1151 * of hacks for example to fork off the per-CPU idle tasks. 1197 * of hacks for example to fork off the per-CPU idle tasks.
1152 * [Hopefully no generic code relies on the reschedule -AK] 1198 * [Hopefully no generic code relies on the reschedule -AK]
1153 */ 1199 */
1154 RESTORE_ALL 1200 RESTORE_ALL
1155 UNFAKE_STACK_FRAME 1201 UNFAKE_STACK_FRAME
1156 ret 1202 ret
1157 CFI_ENDPROC 1203 CFI_ENDPROC
1158ENDPROC(kernel_thread) 1204END(kernel_thread)
1159 1205
1160child_rip: 1206ENTRY(child_rip)
1161 pushq $0 # fake return address 1207 pushq $0 # fake return address
1162 CFI_STARTPROC 1208 CFI_STARTPROC
1163 /* 1209 /*
@@ -1170,8 +1216,9 @@ child_rip:
1170 # exit 1216 # exit
1171 mov %eax, %edi 1217 mov %eax, %edi
1172 call do_exit 1218 call do_exit
1219 ud2 # padding for call trace
1173 CFI_ENDPROC 1220 CFI_ENDPROC
1174ENDPROC(child_rip) 1221END(child_rip)
1175 1222
1176/* 1223/*
1177 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly. 1224 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
@@ -1191,10 +1238,10 @@ ENDPROC(child_rip)
1191ENTRY(kernel_execve) 1238ENTRY(kernel_execve)
1192 CFI_STARTPROC 1239 CFI_STARTPROC
1193 FAKE_STACK_FRAME $0 1240 FAKE_STACK_FRAME $0
1194 SAVE_ALL 1241 SAVE_ALL
1195 movq %rsp,%rcx 1242 movq %rsp,%rcx
1196 call sys_execve 1243 call sys_execve
1197 movq %rax, RAX(%rsp) 1244 movq %rax, RAX(%rsp)
1198 RESTORE_REST 1245 RESTORE_REST
1199 testq %rax,%rax 1246 testq %rax,%rax
1200 je int_ret_from_sys_call 1247 je int_ret_from_sys_call
@@ -1202,129 +1249,7 @@ ENTRY(kernel_execve)
1202 UNFAKE_STACK_FRAME 1249 UNFAKE_STACK_FRAME
1203 ret 1250 ret
1204 CFI_ENDPROC 1251 CFI_ENDPROC
1205ENDPROC(kernel_execve) 1252END(kernel_execve)
1206
1207KPROBE_ENTRY(page_fault)
1208 errorentry do_page_fault
1209KPROBE_END(page_fault)
1210
1211ENTRY(coprocessor_error)
1212 zeroentry do_coprocessor_error
1213END(coprocessor_error)
1214
1215ENTRY(simd_coprocessor_error)
1216 zeroentry do_simd_coprocessor_error
1217END(simd_coprocessor_error)
1218
1219ENTRY(device_not_available)
1220 zeroentry do_device_not_available
1221END(device_not_available)
1222
1223 /* runs on exception stack */
1224KPROBE_ENTRY(debug)
1225 INTR_FRAME
1226 PARAVIRT_ADJUST_EXCEPTION_FRAME
1227 pushq $0
1228 CFI_ADJUST_CFA_OFFSET 8
1229 paranoidentry do_debug, DEBUG_STACK
1230 paranoidexit
1231KPROBE_END(debug)
1232
1233 /* runs on exception stack */
1234KPROBE_ENTRY(nmi)
1235 INTR_FRAME
1236 PARAVIRT_ADJUST_EXCEPTION_FRAME
1237 pushq $-1
1238 CFI_ADJUST_CFA_OFFSET 8
1239 paranoidentry do_nmi, 0, 0
1240#ifdef CONFIG_TRACE_IRQFLAGS
1241 paranoidexit 0
1242#else
1243 jmp paranoid_exit1
1244 CFI_ENDPROC
1245#endif
1246KPROBE_END(nmi)
1247
1248KPROBE_ENTRY(int3)
1249 INTR_FRAME
1250 PARAVIRT_ADJUST_EXCEPTION_FRAME
1251 pushq $0
1252 CFI_ADJUST_CFA_OFFSET 8
1253 paranoidentry do_int3, DEBUG_STACK
1254 jmp paranoid_exit1
1255 CFI_ENDPROC
1256KPROBE_END(int3)
1257
1258ENTRY(overflow)
1259 zeroentry do_overflow
1260END(overflow)
1261
1262ENTRY(bounds)
1263 zeroentry do_bounds
1264END(bounds)
1265
1266ENTRY(invalid_op)
1267 zeroentry do_invalid_op
1268END(invalid_op)
1269
1270ENTRY(coprocessor_segment_overrun)
1271 zeroentry do_coprocessor_segment_overrun
1272END(coprocessor_segment_overrun)
1273
1274 /* runs on exception stack */
1275ENTRY(double_fault)
1276 XCPT_FRAME
1277 PARAVIRT_ADJUST_EXCEPTION_FRAME
1278 paranoidentry do_double_fault
1279 jmp paranoid_exit1
1280 CFI_ENDPROC
1281END(double_fault)
1282
1283ENTRY(invalid_TSS)
1284 errorentry do_invalid_TSS
1285END(invalid_TSS)
1286
1287ENTRY(segment_not_present)
1288 errorentry do_segment_not_present
1289END(segment_not_present)
1290
1291 /* runs on exception stack */
1292ENTRY(stack_segment)
1293 XCPT_FRAME
1294 PARAVIRT_ADJUST_EXCEPTION_FRAME
1295 paranoidentry do_stack_segment
1296 jmp paranoid_exit1
1297 CFI_ENDPROC
1298END(stack_segment)
1299
1300KPROBE_ENTRY(general_protection)
1301 errorentry do_general_protection
1302KPROBE_END(general_protection)
1303
1304ENTRY(alignment_check)
1305 errorentry do_alignment_check
1306END(alignment_check)
1307
1308ENTRY(divide_error)
1309 zeroentry do_divide_error
1310END(divide_error)
1311
1312ENTRY(spurious_interrupt_bug)
1313 zeroentry do_spurious_interrupt_bug
1314END(spurious_interrupt_bug)
1315
1316#ifdef CONFIG_X86_MCE
1317 /* runs on exception stack */
1318ENTRY(machine_check)
1319 INTR_FRAME
1320 PARAVIRT_ADJUST_EXCEPTION_FRAME
1321 pushq $0
1322 CFI_ADJUST_CFA_OFFSET 8
1323 paranoidentry do_machine_check
1324 jmp paranoid_exit1
1325 CFI_ENDPROC
1326END(machine_check)
1327#endif
1328 1253
1329/* Call softirq on interrupt stack. Interrupts are off. */ 1254/* Call softirq on interrupt stack. Interrupts are off. */
1330ENTRY(call_softirq) 1255ENTRY(call_softirq)
@@ -1344,40 +1269,33 @@ ENTRY(call_softirq)
1344 decl %gs:pda_irqcount 1269 decl %gs:pda_irqcount
1345 ret 1270 ret
1346 CFI_ENDPROC 1271 CFI_ENDPROC
1347ENDPROC(call_softirq) 1272END(call_softirq)
1348
1349KPROBE_ENTRY(ignore_sysret)
1350 CFI_STARTPROC
1351 mov $-ENOSYS,%eax
1352 sysret
1353 CFI_ENDPROC
1354ENDPROC(ignore_sysret)
1355 1273
1356#ifdef CONFIG_XEN 1274#ifdef CONFIG_XEN
1357ENTRY(xen_hypervisor_callback) 1275zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1358 zeroentry xen_do_hypervisor_callback
1359END(xen_hypervisor_callback)
1360 1276
1361/* 1277/*
1362# A note on the "critical region" in our callback handler. 1278 * A note on the "critical region" in our callback handler.
1363# We want to avoid stacking callback handlers due to events occurring 1279 * We want to avoid stacking callback handlers due to events occurring
1364# during handling of the last event. To do this, we keep events disabled 1280 * during handling of the last event. To do this, we keep events disabled
1365# until we've done all processing. HOWEVER, we must enable events before 1281 * until we've done all processing. HOWEVER, we must enable events before
1366# popping the stack frame (can't be done atomically) and so it would still 1282 * popping the stack frame (can't be done atomically) and so it would still
1367# be possible to get enough handler activations to overflow the stack. 1283 * be possible to get enough handler activations to overflow the stack.
1368# Although unlikely, bugs of that kind are hard to track down, so we'd 1284 * Although unlikely, bugs of that kind are hard to track down, so we'd
1369# like to avoid the possibility. 1285 * like to avoid the possibility.
1370# So, on entry to the handler we detect whether we interrupted an 1286 * So, on entry to the handler we detect whether we interrupted an
1371# existing activation in its critical region -- if so, we pop the current 1287 * existing activation in its critical region -- if so, we pop the current
1372# activation and restart the handler using the previous one. 1288 * activation and restart the handler using the previous one.
1373*/ 1289 */
1374ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs) 1290ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1375 CFI_STARTPROC 1291 CFI_STARTPROC
1376/* Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will 1292/*
1377 see the correct pointer to the pt_regs */ 1293 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1294 * see the correct pointer to the pt_regs
1295 */
1378 movq %rdi, %rsp # we don't return, adjust the stack frame 1296 movq %rdi, %rsp # we don't return, adjust the stack frame
1379 CFI_ENDPROC 1297 CFI_ENDPROC
1380 CFI_DEFAULT_STACK 1298 DEFAULT_FRAME
138111: incl %gs:pda_irqcount 129911: incl %gs:pda_irqcount
1382 movq %rsp,%rbp 1300 movq %rsp,%rbp
1383 CFI_DEF_CFA_REGISTER rbp 1301 CFI_DEF_CFA_REGISTER rbp
@@ -1392,23 +1310,26 @@ ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1392END(do_hypervisor_callback) 1310END(do_hypervisor_callback)
1393 1311
1394/* 1312/*
1395# Hypervisor uses this for application faults while it executes. 1313 * Hypervisor uses this for application faults while it executes.
1396# We get here for two reasons: 1314 * We get here for two reasons:
1397# 1. Fault while reloading DS, ES, FS or GS 1315 * 1. Fault while reloading DS, ES, FS or GS
1398# 2. Fault while executing IRET 1316 * 2. Fault while executing IRET
1399# Category 1 we do not need to fix up as Xen has already reloaded all segment 1317 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1400# registers that could be reloaded and zeroed the others. 1318 * registers that could be reloaded and zeroed the others.
1401# Category 2 we fix up by killing the current process. We cannot use the 1319 * Category 2 we fix up by killing the current process. We cannot use the
1402# normal Linux return path in this case because if we use the IRET hypercall 1320 * normal Linux return path in this case because if we use the IRET hypercall
1403# to pop the stack frame we end up in an infinite loop of failsafe callbacks. 1321 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1404# We distinguish between categories by comparing each saved segment register 1322 * We distinguish between categories by comparing each saved segment register
1405# with its current contents: any discrepancy means we in category 1. 1323 * with its current contents: any discrepancy means we in category 1.
1406*/ 1324 */
1407ENTRY(xen_failsafe_callback) 1325ENTRY(xen_failsafe_callback)
1408 framesz = (RIP-0x30) /* workaround buggy gas */ 1326 INTR_FRAME 1 (6*8)
1409 _frame framesz 1327 /*CFI_REL_OFFSET gs,GS*/
1410 CFI_REL_OFFSET rcx, 0 1328 /*CFI_REL_OFFSET fs,FS*/
1411 CFI_REL_OFFSET r11, 8 1329 /*CFI_REL_OFFSET es,ES*/
1330 /*CFI_REL_OFFSET ds,DS*/
1331 CFI_REL_OFFSET r11,8
1332 CFI_REL_OFFSET rcx,0
1412 movw %ds,%cx 1333 movw %ds,%cx
1413 cmpw %cx,0x10(%rsp) 1334 cmpw %cx,0x10(%rsp)
1414 CFI_REMEMBER_STATE 1335 CFI_REMEMBER_STATE
@@ -1429,12 +1350,9 @@ ENTRY(xen_failsafe_callback)
1429 CFI_RESTORE r11 1350 CFI_RESTORE r11
1430 addq $0x30,%rsp 1351 addq $0x30,%rsp
1431 CFI_ADJUST_CFA_OFFSET -0x30 1352 CFI_ADJUST_CFA_OFFSET -0x30
1432 pushq $0 1353 pushq_cfi $0 /* RIP */
1433 CFI_ADJUST_CFA_OFFSET 8 1354 pushq_cfi %r11
1434 pushq %r11 1355 pushq_cfi %rcx
1435 CFI_ADJUST_CFA_OFFSET 8
1436 pushq %rcx
1437 CFI_ADJUST_CFA_OFFSET 8
1438 jmp general_protection 1356 jmp general_protection
1439 CFI_RESTORE_STATE 1357 CFI_RESTORE_STATE
14401: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */ 13581: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
@@ -1444,11 +1362,223 @@ ENTRY(xen_failsafe_callback)
1444 CFI_RESTORE r11 1362 CFI_RESTORE r11
1445 addq $0x30,%rsp 1363 addq $0x30,%rsp
1446 CFI_ADJUST_CFA_OFFSET -0x30 1364 CFI_ADJUST_CFA_OFFSET -0x30
1447 pushq $0 1365 pushq_cfi $0
1448 CFI_ADJUST_CFA_OFFSET 8
1449 SAVE_ALL 1366 SAVE_ALL
1450 jmp error_exit 1367 jmp error_exit
1451 CFI_ENDPROC 1368 CFI_ENDPROC
1452END(xen_failsafe_callback) 1369END(xen_failsafe_callback)
1453 1370
1454#endif /* CONFIG_XEN */ 1371#endif /* CONFIG_XEN */
1372
1373/*
1374 * Some functions should be protected against kprobes
1375 */
1376 .pushsection .kprobes.text, "ax"
1377
1378paranoidzeroentry_ist debug do_debug DEBUG_STACK
1379paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1380paranoiderrorentry stack_segment do_stack_segment
1381errorentry general_protection do_general_protection
1382errorentry page_fault do_page_fault
1383#ifdef CONFIG_X86_MCE
1384paranoidzeroentry machine_check do_machine_check
1385#endif
1386
1387 /*
1388 * "Paranoid" exit path from exception stack.
1389 * Paranoid because this is used by NMIs and cannot take
1390 * any kernel state for granted.
1391 * We don't do kernel preemption checks here, because only
1392 * NMI should be common and it does not enable IRQs and
1393 * cannot get reschedule ticks.
1394 *
1395 * "trace" is 0 for the NMI handler only, because irq-tracing
1396 * is fundamentally NMI-unsafe. (we cannot change the soft and
1397 * hard flags at once, atomically)
1398 */
1399
1400 /* ebx: no swapgs flag */
1401ENTRY(paranoid_exit)
1402 INTR_FRAME
1403 DISABLE_INTERRUPTS(CLBR_NONE)
1404 TRACE_IRQS_OFF
1405 testl %ebx,%ebx /* swapgs needed? */
1406 jnz paranoid_restore
1407 testl $3,CS(%rsp)
1408 jnz paranoid_userspace
1409paranoid_swapgs:
1410 TRACE_IRQS_IRETQ 0
1411 SWAPGS_UNSAFE_STACK
1412paranoid_restore:
1413 RESTORE_ALL 8
1414 jmp irq_return
1415paranoid_userspace:
1416 GET_THREAD_INFO(%rcx)
1417 movl TI_flags(%rcx),%ebx
1418 andl $_TIF_WORK_MASK,%ebx
1419 jz paranoid_swapgs
1420 movq %rsp,%rdi /* &pt_regs */
1421 call sync_regs
1422 movq %rax,%rsp /* switch stack for scheduling */
1423 testl $_TIF_NEED_RESCHED,%ebx
1424 jnz paranoid_schedule
1425 movl %ebx,%edx /* arg3: thread flags */
1426 TRACE_IRQS_ON
1427 ENABLE_INTERRUPTS(CLBR_NONE)
1428 xorl %esi,%esi /* arg2: oldset */
1429 movq %rsp,%rdi /* arg1: &pt_regs */
1430 call do_notify_resume
1431 DISABLE_INTERRUPTS(CLBR_NONE)
1432 TRACE_IRQS_OFF
1433 jmp paranoid_userspace
1434paranoid_schedule:
1435 TRACE_IRQS_ON
1436 ENABLE_INTERRUPTS(CLBR_ANY)
1437 call schedule
1438 DISABLE_INTERRUPTS(CLBR_ANY)
1439 TRACE_IRQS_OFF
1440 jmp paranoid_userspace
1441 CFI_ENDPROC
1442END(paranoid_exit)
1443
1444/*
1445 * Exception entry point. This expects an error code/orig_rax on the stack.
1446 * returns in "no swapgs flag" in %ebx.
1447 */
1448ENTRY(error_entry)
1449 XCPT_FRAME
1450 CFI_ADJUST_CFA_OFFSET 15*8
1451 /* oldrax contains error code */
1452 cld
1453 movq_cfi rdi, RDI+8
1454 movq_cfi rsi, RSI+8
1455 movq_cfi rdx, RDX+8
1456 movq_cfi rcx, RCX+8
1457 movq_cfi rax, RAX+8
1458 movq_cfi r8, R8+8
1459 movq_cfi r9, R9+8
1460 movq_cfi r10, R10+8
1461 movq_cfi r11, R11+8
1462 movq_cfi rbx, RBX+8
1463 movq_cfi rbp, RBP+8
1464 movq_cfi r12, R12+8
1465 movq_cfi r13, R13+8
1466 movq_cfi r14, R14+8
1467 movq_cfi r15, R15+8
1468 xorl %ebx,%ebx
1469 testl $3,CS+8(%rsp)
1470 je error_kernelspace
1471error_swapgs:
1472 SWAPGS
1473error_sti:
1474 TRACE_IRQS_OFF
1475 ret
1476 CFI_ENDPROC
1477
1478/*
1479 * There are two places in the kernel that can potentially fault with
1480 * usergs. Handle them here. The exception handlers after iret run with
1481 * kernel gs again, so don't set the user space flag. B stepping K8s
1482 * sometimes report an truncated RIP for IRET exceptions returning to
1483 * compat mode. Check for these here too.
1484 */
1485error_kernelspace:
1486 incl %ebx
1487 leaq irq_return(%rip),%rcx
1488 cmpq %rcx,RIP+8(%rsp)
1489 je error_swapgs
1490 movl %ecx,%ecx /* zero extend */
1491 cmpq %rcx,RIP+8(%rsp)
1492 je error_swapgs
1493 cmpq $gs_change,RIP+8(%rsp)
1494 je error_swapgs
1495 jmp error_sti
1496END(error_entry)
1497
1498
1499/* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1500ENTRY(error_exit)
1501 DEFAULT_FRAME
1502 movl %ebx,%eax
1503 RESTORE_REST
1504 DISABLE_INTERRUPTS(CLBR_NONE)
1505 TRACE_IRQS_OFF
1506 GET_THREAD_INFO(%rcx)
1507 testl %eax,%eax
1508 jne retint_kernel
1509 LOCKDEP_SYS_EXIT_IRQ
1510 movl TI_flags(%rcx),%edx
1511 movl $_TIF_WORK_MASK,%edi
1512 andl %edi,%edx
1513 jnz retint_careful
1514 jmp retint_swapgs
1515 CFI_ENDPROC
1516END(error_exit)
1517
1518
1519 /* runs on exception stack */
1520ENTRY(nmi)
1521 INTR_FRAME
1522 PARAVIRT_ADJUST_EXCEPTION_FRAME
1523 pushq_cfi $-1
1524 subq $15*8, %rsp
1525 CFI_ADJUST_CFA_OFFSET 15*8
1526 call save_paranoid
1527 DEFAULT_FRAME 0
1528 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1529 movq %rsp,%rdi
1530 movq $-1,%rsi
1531 call do_nmi
1532#ifdef CONFIG_TRACE_IRQFLAGS
1533 /* paranoidexit; without TRACE_IRQS_OFF */
1534 /* ebx: no swapgs flag */
1535 DISABLE_INTERRUPTS(CLBR_NONE)
1536 testl %ebx,%ebx /* swapgs needed? */
1537 jnz nmi_restore
1538 testl $3,CS(%rsp)
1539 jnz nmi_userspace
1540nmi_swapgs:
1541 SWAPGS_UNSAFE_STACK
1542nmi_restore:
1543 RESTORE_ALL 8
1544 jmp irq_return
1545nmi_userspace:
1546 GET_THREAD_INFO(%rcx)
1547 movl TI_flags(%rcx),%ebx
1548 andl $_TIF_WORK_MASK,%ebx
1549 jz nmi_swapgs
1550 movq %rsp,%rdi /* &pt_regs */
1551 call sync_regs
1552 movq %rax,%rsp /* switch stack for scheduling */
1553 testl $_TIF_NEED_RESCHED,%ebx
1554 jnz nmi_schedule
1555 movl %ebx,%edx /* arg3: thread flags */
1556 ENABLE_INTERRUPTS(CLBR_NONE)
1557 xorl %esi,%esi /* arg2: oldset */
1558 movq %rsp,%rdi /* arg1: &pt_regs */
1559 call do_notify_resume
1560 DISABLE_INTERRUPTS(CLBR_NONE)
1561 jmp nmi_userspace
1562nmi_schedule:
1563 ENABLE_INTERRUPTS(CLBR_ANY)
1564 call schedule
1565 DISABLE_INTERRUPTS(CLBR_ANY)
1566 jmp nmi_userspace
1567 CFI_ENDPROC
1568#else
1569 jmp paranoid_exit
1570 CFI_ENDPROC
1571#endif
1572END(nmi)
1573
1574ENTRY(ignore_sysret)
1575 CFI_STARTPROC
1576 mov $-ENOSYS,%eax
1577 sysret
1578 CFI_ENDPROC
1579END(ignore_sysret)
1580
1581/*
1582 * End of kprobes section
1583 */
1584 .popsection
diff --git a/arch/x86/kernel/es7000_32.c b/arch/x86/kernel/es7000_32.c
index 0aa2c443d600..53699c931ad4 100644
--- a/arch/x86/kernel/es7000_32.c
+++ b/arch/x86/kernel/es7000_32.c
@@ -38,8 +38,11 @@
38#include <asm/io.h> 38#include <asm/io.h>
39#include <asm/nmi.h> 39#include <asm/nmi.h>
40#include <asm/smp.h> 40#include <asm/smp.h>
41#include <asm/atomic.h>
41#include <asm/apicdef.h> 42#include <asm/apicdef.h>
42#include <mach_mpparse.h> 43#include <mach_mpparse.h>
44#include <asm/genapic.h>
45#include <asm/setup.h>
43 46
44/* 47/*
45 * ES7000 chipsets 48 * ES7000 chipsets
@@ -161,6 +164,43 @@ es7000_rename_gsi(int ioapic, int gsi)
161 return gsi; 164 return gsi;
162} 165}
163 166
167static int wakeup_secondary_cpu_via_mip(int cpu, unsigned long eip)
168{
169 unsigned long vect = 0, psaival = 0;
170
171 if (psai == NULL)
172 return -1;
173
174 vect = ((unsigned long)__pa(eip)/0x1000) << 16;
175 psaival = (0x1000000 | vect | cpu);
176
177 while (*psai & 0x1000000)
178 ;
179
180 *psai = psaival;
181
182 return 0;
183}
184
185static void noop_wait_for_deassert(atomic_t *deassert_not_used)
186{
187}
188
189static int __init es7000_update_genapic(void)
190{
191 genapic->wakeup_cpu = wakeup_secondary_cpu_via_mip;
192
193 /* MPENTIUMIII */
194 if (boot_cpu_data.x86 == 6 &&
195 (boot_cpu_data.x86_model >= 7 || boot_cpu_data.x86_model <= 11)) {
196 es7000_update_genapic_to_cluster();
197 genapic->wait_for_init_deassert = noop_wait_for_deassert;
198 genapic->wakeup_cpu = wakeup_secondary_cpu_via_mip;
199 }
200
201 return 0;
202}
203
164void __init 204void __init
165setup_unisys(void) 205setup_unisys(void)
166{ 206{
@@ -176,6 +216,8 @@ setup_unisys(void)
176 else 216 else
177 es7000_plat = ES7000_CLASSIC; 217 es7000_plat = ES7000_CLASSIC;
178 ioapic_renumber_irq = es7000_rename_gsi; 218 ioapic_renumber_irq = es7000_rename_gsi;
219
220 x86_quirks->update_genapic = es7000_update_genapic;
179} 221}
180 222
181/* 223/*
@@ -317,26 +359,6 @@ es7000_mip_write(struct mip_reg *mip_reg)
317 return status; 359 return status;
318} 360}
319 361
320int
321es7000_start_cpu(int cpu, unsigned long eip)
322{
323 unsigned long vect = 0, psaival = 0;
324
325 if (psai == NULL)
326 return -1;
327
328 vect = ((unsigned long)__pa(eip)/0x1000) << 16;
329 psaival = (0x1000000 | vect | cpu);
330
331 while (*psai & 0x1000000)
332 ;
333
334 *psai = psaival;
335
336 return 0;
337
338}
339
340void __init 362void __init
341es7000_sw_apic(void) 363es7000_sw_apic(void)
342{ 364{
diff --git a/arch/x86/kernel/ftrace.c b/arch/x86/kernel/ftrace.c
index 50ea0ac8c9bf..1b43086b097a 100644
--- a/arch/x86/kernel/ftrace.c
+++ b/arch/x86/kernel/ftrace.c
@@ -14,14 +14,17 @@
14#include <linux/uaccess.h> 14#include <linux/uaccess.h>
15#include <linux/ftrace.h> 15#include <linux/ftrace.h>
16#include <linux/percpu.h> 16#include <linux/percpu.h>
17#include <linux/sched.h>
17#include <linux/init.h> 18#include <linux/init.h>
18#include <linux/list.h> 19#include <linux/list.h>
19 20
20#include <asm/ftrace.h> 21#include <asm/ftrace.h>
22#include <linux/ftrace.h>
21#include <asm/nops.h> 23#include <asm/nops.h>
24#include <asm/nmi.h>
22 25
23 26
24static unsigned char ftrace_nop[MCOUNT_INSN_SIZE]; 27#ifdef CONFIG_DYNAMIC_FTRACE
25 28
26union ftrace_code_union { 29union ftrace_code_union {
27 char code[MCOUNT_INSN_SIZE]; 30 char code[MCOUNT_INSN_SIZE];
@@ -31,18 +34,12 @@ union ftrace_code_union {
31 } __attribute__((packed)); 34 } __attribute__((packed));
32}; 35};
33 36
34
35static int ftrace_calc_offset(long ip, long addr) 37static int ftrace_calc_offset(long ip, long addr)
36{ 38{
37 return (int)(addr - ip); 39 return (int)(addr - ip);
38} 40}
39 41
40unsigned char *ftrace_nop_replace(void) 42static unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr)
41{
42 return ftrace_nop;
43}
44
45unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr)
46{ 43{
47 static union ftrace_code_union calc; 44 static union ftrace_code_union calc;
48 45
@@ -56,7 +53,142 @@ unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr)
56 return calc.code; 53 return calc.code;
57} 54}
58 55
59int 56/*
57 * Modifying code must take extra care. On an SMP machine, if
58 * the code being modified is also being executed on another CPU
59 * that CPU will have undefined results and possibly take a GPF.
60 * We use kstop_machine to stop other CPUS from exectuing code.
61 * But this does not stop NMIs from happening. We still need
62 * to protect against that. We separate out the modification of
63 * the code to take care of this.
64 *
65 * Two buffers are added: An IP buffer and a "code" buffer.
66 *
67 * 1) Put the instruction pointer into the IP buffer
68 * and the new code into the "code" buffer.
69 * 2) Set a flag that says we are modifying code
70 * 3) Wait for any running NMIs to finish.
71 * 4) Write the code
72 * 5) clear the flag.
73 * 6) Wait for any running NMIs to finish.
74 *
75 * If an NMI is executed, the first thing it does is to call
76 * "ftrace_nmi_enter". This will check if the flag is set to write
77 * and if it is, it will write what is in the IP and "code" buffers.
78 *
79 * The trick is, it does not matter if everyone is writing the same
80 * content to the code location. Also, if a CPU is executing code
81 * it is OK to write to that code location if the contents being written
82 * are the same as what exists.
83 */
84
85static atomic_t in_nmi = ATOMIC_INIT(0);
86static int mod_code_status; /* holds return value of text write */
87static int mod_code_write; /* set when NMI should do the write */
88static void *mod_code_ip; /* holds the IP to write to */
89static void *mod_code_newcode; /* holds the text to write to the IP */
90
91static unsigned nmi_wait_count;
92static atomic_t nmi_update_count = ATOMIC_INIT(0);
93
94int ftrace_arch_read_dyn_info(char *buf, int size)
95{
96 int r;
97
98 r = snprintf(buf, size, "%u %u",
99 nmi_wait_count,
100 atomic_read(&nmi_update_count));
101 return r;
102}
103
104static void ftrace_mod_code(void)
105{
106 /*
107 * Yes, more than one CPU process can be writing to mod_code_status.
108 * (and the code itself)
109 * But if one were to fail, then they all should, and if one were
110 * to succeed, then they all should.
111 */
112 mod_code_status = probe_kernel_write(mod_code_ip, mod_code_newcode,
113 MCOUNT_INSN_SIZE);
114}
115
116void ftrace_nmi_enter(void)
117{
118 atomic_inc(&in_nmi);
119 /* Must have in_nmi seen before reading write flag */
120 smp_mb();
121 if (mod_code_write) {
122 ftrace_mod_code();
123 atomic_inc(&nmi_update_count);
124 }
125}
126
127void ftrace_nmi_exit(void)
128{
129 /* Finish all executions before clearing in_nmi */
130 smp_wmb();
131 atomic_dec(&in_nmi);
132}
133
134static void wait_for_nmi(void)
135{
136 int waited = 0;
137
138 while (atomic_read(&in_nmi)) {
139 waited = 1;
140 cpu_relax();
141 }
142
143 if (waited)
144 nmi_wait_count++;
145}
146
147static int
148do_ftrace_mod_code(unsigned long ip, void *new_code)
149{
150 mod_code_ip = (void *)ip;
151 mod_code_newcode = new_code;
152
153 /* The buffers need to be visible before we let NMIs write them */
154 smp_wmb();
155
156 mod_code_write = 1;
157
158 /* Make sure write bit is visible before we wait on NMIs */
159 smp_mb();
160
161 wait_for_nmi();
162
163 /* Make sure all running NMIs have finished before we write the code */
164 smp_mb();
165
166 ftrace_mod_code();
167
168 /* Make sure the write happens before clearing the bit */
169 smp_wmb();
170
171 mod_code_write = 0;
172
173 /* make sure NMIs see the cleared bit */
174 smp_mb();
175
176 wait_for_nmi();
177
178 return mod_code_status;
179}
180
181
182
183
184static unsigned char ftrace_nop[MCOUNT_INSN_SIZE];
185
186static unsigned char *ftrace_nop_replace(void)
187{
188 return ftrace_nop;
189}
190
191static int
60ftrace_modify_code(unsigned long ip, unsigned char *old_code, 192ftrace_modify_code(unsigned long ip, unsigned char *old_code,
61 unsigned char *new_code) 193 unsigned char *new_code)
62{ 194{
@@ -81,7 +213,7 @@ ftrace_modify_code(unsigned long ip, unsigned char *old_code,
81 return -EINVAL; 213 return -EINVAL;
82 214
83 /* replace the text with the new text */ 215 /* replace the text with the new text */
84 if (probe_kernel_write((void *)ip, new_code, MCOUNT_INSN_SIZE)) 216 if (do_ftrace_mod_code(ip, new_code))
85 return -EPERM; 217 return -EPERM;
86 218
87 sync_core(); 219 sync_core();
@@ -89,6 +221,29 @@ ftrace_modify_code(unsigned long ip, unsigned char *old_code,
89 return 0; 221 return 0;
90} 222}
91 223
224int ftrace_make_nop(struct module *mod,
225 struct dyn_ftrace *rec, unsigned long addr)
226{
227 unsigned char *new, *old;
228 unsigned long ip = rec->ip;
229
230 old = ftrace_call_replace(ip, addr);
231 new = ftrace_nop_replace();
232
233 return ftrace_modify_code(rec->ip, old, new);
234}
235
236int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
237{
238 unsigned char *new, *old;
239 unsigned long ip = rec->ip;
240
241 old = ftrace_nop_replace();
242 new = ftrace_call_replace(ip, addr);
243
244 return ftrace_modify_code(rec->ip, old, new);
245}
246
92int ftrace_update_ftrace_func(ftrace_func_t func) 247int ftrace_update_ftrace_func(ftrace_func_t func)
93{ 248{
94 unsigned long ip = (unsigned long)(&ftrace_call); 249 unsigned long ip = (unsigned long)(&ftrace_call);
@@ -165,3 +320,218 @@ int __init ftrace_dyn_arch_init(void *data)
165 320
166 return 0; 321 return 0;
167} 322}
323#endif
324
325#ifdef CONFIG_FUNCTION_GRAPH_TRACER
326
327#ifdef CONFIG_DYNAMIC_FTRACE
328extern void ftrace_graph_call(void);
329
330static int ftrace_mod_jmp(unsigned long ip,
331 int old_offset, int new_offset)
332{
333 unsigned char code[MCOUNT_INSN_SIZE];
334
335 if (probe_kernel_read(code, (void *)ip, MCOUNT_INSN_SIZE))
336 return -EFAULT;
337
338 if (code[0] != 0xe9 || old_offset != *(int *)(&code[1]))
339 return -EINVAL;
340
341 *(int *)(&code[1]) = new_offset;
342
343 if (do_ftrace_mod_code(ip, &code))
344 return -EPERM;
345
346 return 0;
347}
348
349int ftrace_enable_ftrace_graph_caller(void)
350{
351 unsigned long ip = (unsigned long)(&ftrace_graph_call);
352 int old_offset, new_offset;
353
354 old_offset = (unsigned long)(&ftrace_stub) - (ip + MCOUNT_INSN_SIZE);
355 new_offset = (unsigned long)(&ftrace_graph_caller) - (ip + MCOUNT_INSN_SIZE);
356
357 return ftrace_mod_jmp(ip, old_offset, new_offset);
358}
359
360int ftrace_disable_ftrace_graph_caller(void)
361{
362 unsigned long ip = (unsigned long)(&ftrace_graph_call);
363 int old_offset, new_offset;
364
365 old_offset = (unsigned long)(&ftrace_graph_caller) - (ip + MCOUNT_INSN_SIZE);
366 new_offset = (unsigned long)(&ftrace_stub) - (ip + MCOUNT_INSN_SIZE);
367
368 return ftrace_mod_jmp(ip, old_offset, new_offset);
369}
370
371#else /* CONFIG_DYNAMIC_FTRACE */
372
373/*
374 * These functions are picked from those used on
375 * this page for dynamic ftrace. They have been
376 * simplified to ignore all traces in NMI context.
377 */
378static atomic_t in_nmi;
379
380void ftrace_nmi_enter(void)
381{
382 atomic_inc(&in_nmi);
383}
384
385void ftrace_nmi_exit(void)
386{
387 atomic_dec(&in_nmi);
388}
389
390#endif /* !CONFIG_DYNAMIC_FTRACE */
391
392/* Add a function return address to the trace stack on thread info.*/
393static int push_return_trace(unsigned long ret, unsigned long long time,
394 unsigned long func, int *depth)
395{
396 int index;
397
398 if (!current->ret_stack)
399 return -EBUSY;
400
401 /* The return trace stack is full */
402 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
403 atomic_inc(&current->trace_overrun);
404 return -EBUSY;
405 }
406
407 index = ++current->curr_ret_stack;
408 barrier();
409 current->ret_stack[index].ret = ret;
410 current->ret_stack[index].func = func;
411 current->ret_stack[index].calltime = time;
412 *depth = index;
413
414 return 0;
415}
416
417/* Retrieve a function return address to the trace stack on thread info.*/
418static void pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret)
419{
420 int index;
421
422 index = current->curr_ret_stack;
423
424 if (unlikely(index < 0)) {
425 ftrace_graph_stop();
426 WARN_ON(1);
427 /* Might as well panic, otherwise we have no where to go */
428 *ret = (unsigned long)panic;
429 return;
430 }
431
432 *ret = current->ret_stack[index].ret;
433 trace->func = current->ret_stack[index].func;
434 trace->calltime = current->ret_stack[index].calltime;
435 trace->overrun = atomic_read(&current->trace_overrun);
436 trace->depth = index;
437 barrier();
438 current->curr_ret_stack--;
439
440}
441
442/*
443 * Send the trace to the ring-buffer.
444 * @return the original return address.
445 */
446unsigned long ftrace_return_to_handler(void)
447{
448 struct ftrace_graph_ret trace;
449 unsigned long ret;
450
451 pop_return_trace(&trace, &ret);
452 trace.rettime = cpu_clock(raw_smp_processor_id());
453 ftrace_graph_return(&trace);
454
455 if (unlikely(!ret)) {
456 ftrace_graph_stop();
457 WARN_ON(1);
458 /* Might as well panic. What else to do? */
459 ret = (unsigned long)panic;
460 }
461
462 return ret;
463}
464
465/*
466 * Hook the return address and push it in the stack of return addrs
467 * in current thread info.
468 */
469void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
470{
471 unsigned long old;
472 unsigned long long calltime;
473 int faulted;
474 struct ftrace_graph_ent trace;
475 unsigned long return_hooker = (unsigned long)
476 &return_to_handler;
477
478 /* Nmi's are currently unsupported */
479 if (unlikely(atomic_read(&in_nmi)))
480 return;
481
482 if (unlikely(atomic_read(&current->tracing_graph_pause)))
483 return;
484
485 /*
486 * Protect against fault, even if it shouldn't
487 * happen. This tool is too much intrusive to
488 * ignore such a protection.
489 */
490 asm volatile(
491 "1: " _ASM_MOV " (%[parent_old]), %[old]\n"
492 "2: " _ASM_MOV " %[return_hooker], (%[parent_replaced])\n"
493 " movl $0, %[faulted]\n"
494
495 ".section .fixup, \"ax\"\n"
496 "3: movl $1, %[faulted]\n"
497 ".previous\n"
498
499 _ASM_EXTABLE(1b, 3b)
500 _ASM_EXTABLE(2b, 3b)
501
502 : [parent_replaced] "=r" (parent), [old] "=r" (old),
503 [faulted] "=r" (faulted)
504 : [parent_old] "0" (parent), [return_hooker] "r" (return_hooker)
505 : "memory"
506 );
507
508 if (unlikely(faulted)) {
509 ftrace_graph_stop();
510 WARN_ON(1);
511 return;
512 }
513
514 if (unlikely(!__kernel_text_address(old))) {
515 ftrace_graph_stop();
516 *parent = old;
517 WARN_ON(1);
518 return;
519 }
520
521 calltime = cpu_clock(raw_smp_processor_id());
522
523 if (push_return_trace(old, calltime,
524 self_addr, &trace.depth) == -EBUSY) {
525 *parent = old;
526 return;
527 }
528
529 trace.func = self_addr;
530
531 /* Only trace if the calling function expects to */
532 if (!ftrace_graph_entry(&trace)) {
533 current->curr_ret_stack--;
534 *parent = old;
535 }
536}
537#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
diff --git a/arch/x86/kernel/genapic_64.c b/arch/x86/kernel/genapic_64.c
index 6c9bfc9e1e95..2bced78b0b8e 100644
--- a/arch/x86/kernel/genapic_64.c
+++ b/arch/x86/kernel/genapic_64.c
@@ -21,6 +21,7 @@
21#include <asm/smp.h> 21#include <asm/smp.h>
22#include <asm/ipi.h> 22#include <asm/ipi.h>
23#include <asm/genapic.h> 23#include <asm/genapic.h>
24#include <asm/setup.h>
24 25
25extern struct genapic apic_flat; 26extern struct genapic apic_flat;
26extern struct genapic apic_physflat; 27extern struct genapic apic_physflat;
@@ -53,6 +54,9 @@ void __init setup_apic_routing(void)
53 genapic = &apic_physflat; 54 genapic = &apic_physflat;
54 printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name); 55 printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name);
55 } 56 }
57
58 if (x86_quirks->update_genapic)
59 x86_quirks->update_genapic();
56} 60}
57 61
58/* Same for both flat and physical. */ 62/* Same for both flat and physical. */
diff --git a/arch/x86/kernel/genx2apic_uv_x.c b/arch/x86/kernel/genx2apic_uv_x.c
index 2c7dbdb98278..dece17289731 100644
--- a/arch/x86/kernel/genx2apic_uv_x.c
+++ b/arch/x86/kernel/genx2apic_uv_x.c
@@ -10,6 +10,7 @@
10 10
11#include <linux/kernel.h> 11#include <linux/kernel.h>
12#include <linux/threads.h> 12#include <linux/threads.h>
13#include <linux/cpu.h>
13#include <linux/cpumask.h> 14#include <linux/cpumask.h>
14#include <linux/string.h> 15#include <linux/string.h>
15#include <linux/ctype.h> 16#include <linux/ctype.h>
@@ -17,6 +18,9 @@
17#include <linux/sched.h> 18#include <linux/sched.h>
18#include <linux/module.h> 19#include <linux/module.h>
19#include <linux/hardirq.h> 20#include <linux/hardirq.h>
21#include <linux/timer.h>
22#include <linux/proc_fs.h>
23#include <asm/current.h>
20#include <asm/smp.h> 24#include <asm/smp.h>
21#include <asm/ipi.h> 25#include <asm/ipi.h>
22#include <asm/genapic.h> 26#include <asm/genapic.h>
@@ -356,6 +360,103 @@ static __init void uv_rtc_init(void)
356} 360}
357 361
358/* 362/*
363 * percpu heartbeat timer
364 */
365static void uv_heartbeat(unsigned long ignored)
366{
367 struct timer_list *timer = &uv_hub_info->scir.timer;
368 unsigned char bits = uv_hub_info->scir.state;
369
370 /* flip heartbeat bit */
371 bits ^= SCIR_CPU_HEARTBEAT;
372
373 /* is this cpu idle? */
374 if (idle_cpu(raw_smp_processor_id()))
375 bits &= ~SCIR_CPU_ACTIVITY;
376 else
377 bits |= SCIR_CPU_ACTIVITY;
378
379 /* update system controller interface reg */
380 uv_set_scir_bits(bits);
381
382 /* enable next timer period */
383 mod_timer(timer, jiffies + SCIR_CPU_HB_INTERVAL);
384}
385
386static void __cpuinit uv_heartbeat_enable(int cpu)
387{
388 if (!uv_cpu_hub_info(cpu)->scir.enabled) {
389 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
390
391 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
392 setup_timer(timer, uv_heartbeat, cpu);
393 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
394 add_timer_on(timer, cpu);
395 uv_cpu_hub_info(cpu)->scir.enabled = 1;
396 }
397
398 /* check boot cpu */
399 if (!uv_cpu_hub_info(0)->scir.enabled)
400 uv_heartbeat_enable(0);
401}
402
403#ifdef CONFIG_HOTPLUG_CPU
404static void __cpuinit uv_heartbeat_disable(int cpu)
405{
406 if (uv_cpu_hub_info(cpu)->scir.enabled) {
407 uv_cpu_hub_info(cpu)->scir.enabled = 0;
408 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
409 }
410 uv_set_cpu_scir_bits(cpu, 0xff);
411}
412
413/*
414 * cpu hotplug notifier
415 */
416static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
417 unsigned long action, void *hcpu)
418{
419 long cpu = (long)hcpu;
420
421 switch (action) {
422 case CPU_ONLINE:
423 uv_heartbeat_enable(cpu);
424 break;
425 case CPU_DOWN_PREPARE:
426 uv_heartbeat_disable(cpu);
427 break;
428 default:
429 break;
430 }
431 return NOTIFY_OK;
432}
433
434static __init void uv_scir_register_cpu_notifier(void)
435{
436 hotcpu_notifier(uv_scir_cpu_notify, 0);
437}
438
439#else /* !CONFIG_HOTPLUG_CPU */
440
441static __init void uv_scir_register_cpu_notifier(void)
442{
443}
444
445static __init int uv_init_heartbeat(void)
446{
447 int cpu;
448
449 if (is_uv_system())
450 for_each_online_cpu(cpu)
451 uv_heartbeat_enable(cpu);
452 return 0;
453}
454
455late_initcall(uv_init_heartbeat);
456
457#endif /* !CONFIG_HOTPLUG_CPU */
458
459/*
359 * Called on each cpu to initialize the per_cpu UV data area. 460 * Called on each cpu to initialize the per_cpu UV data area.
360 * ZZZ hotplug not supported yet 461 * ZZZ hotplug not supported yet
361 */ 462 */
@@ -428,7 +529,7 @@ void __init uv_system_init(void)
428 529
429 uv_bios_init(); 530 uv_bios_init();
430 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, 531 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id,
431 &uv_coherency_id, &uv_region_size); 532 &sn_coherency_id, &sn_region_size);
432 uv_rtc_init(); 533 uv_rtc_init();
433 534
434 for_each_present_cpu(cpu) { 535 for_each_present_cpu(cpu) {
@@ -439,8 +540,7 @@ void __init uv_system_init(void)
439 uv_blade_info[blade].nr_possible_cpus++; 540 uv_blade_info[blade].nr_possible_cpus++;
440 541
441 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base; 542 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
442 uv_cpu_hub_info(cpu)->lowmem_remap_top = 543 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
443 lowmem_redir_base + lowmem_redir_size;
444 uv_cpu_hub_info(cpu)->m_val = m_val; 544 uv_cpu_hub_info(cpu)->m_val = m_val;
445 uv_cpu_hub_info(cpu)->n_val = m_val; 545 uv_cpu_hub_info(cpu)->n_val = m_val;
446 uv_cpu_hub_info(cpu)->numa_blade_id = blade; 546 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
@@ -450,7 +550,8 @@ void __init uv_system_init(void)
450 uv_cpu_hub_info(cpu)->gpa_mask = (1 << (m_val + n_val)) - 1; 550 uv_cpu_hub_info(cpu)->gpa_mask = (1 << (m_val + n_val)) - 1;
451 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper; 551 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
452 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base; 552 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
453 uv_cpu_hub_info(cpu)->coherency_domain_number = uv_coherency_id; 553 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
554 uv_cpu_hub_info(cpu)->scir.offset = SCIR_LOCAL_MMR_BASE + lcpu;
454 uv_node_to_blade[nid] = blade; 555 uv_node_to_blade[nid] = blade;
455 uv_cpu_to_blade[cpu] = blade; 556 uv_cpu_to_blade[cpu] = blade;
456 max_pnode = max(pnode, max_pnode); 557 max_pnode = max(pnode, max_pnode);
@@ -467,4 +568,6 @@ void __init uv_system_init(void)
467 map_mmioh_high(max_pnode); 568 map_mmioh_high(max_pnode);
468 569
469 uv_cpu_init(); 570 uv_cpu_init();
571 uv_scir_register_cpu_notifier();
572 proc_mkdir("sgi_uv", NULL);
470} 573}
diff --git a/arch/x86/kernel/head.c b/arch/x86/kernel/head.c
index 1dcb0f13897e..3e66bd364a9d 100644
--- a/arch/x86/kernel/head.c
+++ b/arch/x86/kernel/head.c
@@ -35,7 +35,6 @@ void __init reserve_ebda_region(void)
35 35
36 /* start of EBDA area */ 36 /* start of EBDA area */
37 ebda_addr = get_bios_ebda(); 37 ebda_addr = get_bios_ebda();
38 printk(KERN_INFO "BIOS EBDA/lowmem at: %08x/%08x\n", ebda_addr, lowmem);
39 38
40 /* Fixup: bios puts an EBDA in the top 64K segment */ 39 /* Fixup: bios puts an EBDA in the top 64K segment */
41 /* of conventional memory, but does not adjust lowmem. */ 40 /* of conventional memory, but does not adjust lowmem. */
diff --git a/arch/x86/kernel/head32.c b/arch/x86/kernel/head32.c
index fa1d25dd83e3..ac108d1fe182 100644
--- a/arch/x86/kernel/head32.c
+++ b/arch/x86/kernel/head32.c
@@ -12,9 +12,12 @@
12#include <asm/sections.h> 12#include <asm/sections.h>
13#include <asm/e820.h> 13#include <asm/e820.h>
14#include <asm/bios_ebda.h> 14#include <asm/bios_ebda.h>
15#include <asm/trampoline.h>
15 16
16void __init i386_start_kernel(void) 17void __init i386_start_kernel(void)
17{ 18{
19 reserve_trampoline_memory();
20
18 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS"); 21 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS");
19 22
20#ifdef CONFIG_BLK_DEV_INITRD 23#ifdef CONFIG_BLK_DEV_INITRD
diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index d16084f90649..388e05a5fc17 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -24,6 +24,7 @@
24#include <asm/kdebug.h> 24#include <asm/kdebug.h>
25#include <asm/e820.h> 25#include <asm/e820.h>
26#include <asm/bios_ebda.h> 26#include <asm/bios_ebda.h>
27#include <asm/trampoline.h>
27 28
28/* boot cpu pda */ 29/* boot cpu pda */
29static struct x8664_pda _boot_cpu_pda __read_mostly; 30static struct x8664_pda _boot_cpu_pda __read_mostly;
@@ -120,6 +121,8 @@ void __init x86_64_start_reservations(char *real_mode_data)
120{ 121{
121 copy_bootdata(__va(real_mode_data)); 122 copy_bootdata(__va(real_mode_data));
122 123
124 reserve_trampoline_memory();
125
123 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS"); 126 reserve_early(__pa_symbol(&_text), __pa_symbol(&_end), "TEXT DATA BSS");
124 127
125#ifdef CONFIG_BLK_DEV_INITRD 128#ifdef CONFIG_BLK_DEV_INITRD
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index 067d8de913f6..3f0a3edf0a57 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -33,7 +33,9 @@
33 * HPET address is set in acpi/boot.c, when an ACPI entry exists 33 * HPET address is set in acpi/boot.c, when an ACPI entry exists
34 */ 34 */
35unsigned long hpet_address; 35unsigned long hpet_address;
36unsigned long hpet_num_timers; 36#ifdef CONFIG_PCI_MSI
37static unsigned long hpet_num_timers;
38#endif
37static void __iomem *hpet_virt_address; 39static void __iomem *hpet_virt_address;
38 40
39struct hpet_dev { 41struct hpet_dev {
diff --git a/arch/x86/kernel/init_task.c b/arch/x86/kernel/init_task.c
index a4f93b4120c1..d39918076bb4 100644
--- a/arch/x86/kernel/init_task.c
+++ b/arch/x86/kernel/init_task.c
@@ -14,7 +14,6 @@ static struct fs_struct init_fs = INIT_FS;
14static struct signal_struct init_signals = INIT_SIGNALS(init_signals); 14static struct signal_struct init_signals = INIT_SIGNALS(init_signals);
15static struct sighand_struct init_sighand = INIT_SIGHAND(init_sighand); 15static struct sighand_struct init_sighand = INIT_SIGHAND(init_sighand);
16struct mm_struct init_mm = INIT_MM(init_mm); 16struct mm_struct init_mm = INIT_MM(init_mm);
17EXPORT_UNUSED_SYMBOL(init_mm); /* will be removed in 2.6.26 */
18 17
19/* 18/*
20 * Initial thread structure. 19 * Initial thread structure.
diff --git a/arch/x86/kernel/io_apic.c b/arch/x86/kernel/io_apic.c
index 9043251210fb..679e7bbbbcd6 100644
--- a/arch/x86/kernel/io_apic.c
+++ b/arch/x86/kernel/io_apic.c
@@ -2216,10 +2216,9 @@ static void set_ir_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
2216asmlinkage void smp_irq_move_cleanup_interrupt(void) 2216asmlinkage void smp_irq_move_cleanup_interrupt(void)
2217{ 2217{
2218 unsigned vector, me; 2218 unsigned vector, me;
2219
2219 ack_APIC_irq(); 2220 ack_APIC_irq();
2220#ifdef CONFIG_X86_64
2221 exit_idle(); 2221 exit_idle();
2222#endif
2223 irq_enter(); 2222 irq_enter();
2224 2223
2225 me = smp_processor_id(); 2224 me = smp_processor_id();
diff --git a/arch/x86/kernel/irq_64.c b/arch/x86/kernel/irq_64.c
index 60eb84eb77a0..1df869e5bd0b 100644
--- a/arch/x86/kernel/irq_64.c
+++ b/arch/x86/kernel/irq_64.c
@@ -13,12 +13,12 @@
13#include <linux/seq_file.h> 13#include <linux/seq_file.h>
14#include <linux/module.h> 14#include <linux/module.h>
15#include <linux/delay.h> 15#include <linux/delay.h>
16#include <linux/ftrace.h>
16#include <asm/uaccess.h> 17#include <asm/uaccess.h>
17#include <asm/io_apic.h> 18#include <asm/io_apic.h>
18#include <asm/idle.h> 19#include <asm/idle.h>
19#include <asm/smp.h> 20#include <asm/smp.h>
20 21
21#ifdef CONFIG_DEBUG_STACKOVERFLOW
22/* 22/*
23 * Probabilistic stack overflow check: 23 * Probabilistic stack overflow check:
24 * 24 *
@@ -28,26 +28,25 @@
28 */ 28 */
29static inline void stack_overflow_check(struct pt_regs *regs) 29static inline void stack_overflow_check(struct pt_regs *regs)
30{ 30{
31#ifdef CONFIG_DEBUG_STACKOVERFLOW
31 u64 curbase = (u64)task_stack_page(current); 32 u64 curbase = (u64)task_stack_page(current);
32 static unsigned long warned = -60*HZ; 33
33 34 WARN_ONCE(regs->sp >= curbase &&
34 if (regs->sp >= curbase && regs->sp <= curbase + THREAD_SIZE && 35 regs->sp <= curbase + THREAD_SIZE &&
35 regs->sp < curbase + sizeof(struct thread_info) + 128 && 36 regs->sp < curbase + sizeof(struct thread_info) +
36 time_after(jiffies, warned + 60*HZ)) { 37 sizeof(struct pt_regs) + 128,
37 printk("do_IRQ: %s near stack overflow (cur:%Lx,sp:%lx)\n", 38
38 current->comm, curbase, regs->sp); 39 "do_IRQ: %s near stack overflow (cur:%Lx,sp:%lx)\n",
39 show_stack(NULL,NULL); 40 current->comm, curbase, regs->sp);
40 warned = jiffies;
41 }
42}
43#endif 41#endif
42}
44 43
45/* 44/*
46 * do_IRQ handles all normal device IRQ's (the special 45 * do_IRQ handles all normal device IRQ's (the special
47 * SMP cross-CPU interrupts have their own specific 46 * SMP cross-CPU interrupts have their own specific
48 * handlers). 47 * handlers).
49 */ 48 */
50asmlinkage unsigned int do_IRQ(struct pt_regs *regs) 49asmlinkage unsigned int __irq_entry do_IRQ(struct pt_regs *regs)
51{ 50{
52 struct pt_regs *old_regs = set_irq_regs(regs); 51 struct pt_regs *old_regs = set_irq_regs(regs);
53 struct irq_desc *desc; 52 struct irq_desc *desc;
@@ -60,9 +59,7 @@ asmlinkage unsigned int do_IRQ(struct pt_regs *regs)
60 irq_enter(); 59 irq_enter();
61 irq = __get_cpu_var(vector_irq)[vector]; 60 irq = __get_cpu_var(vector_irq)[vector];
62 61
63#ifdef CONFIG_DEBUG_STACKOVERFLOW
64 stack_overflow_check(regs); 62 stack_overflow_check(regs);
65#endif
66 63
67 desc = irq_to_desc(irq); 64 desc = irq_to_desc(irq);
68 if (likely(desc)) 65 if (likely(desc))
diff --git a/arch/x86/kernel/irqinit_32.c b/arch/x86/kernel/irqinit_32.c
index 845aa9803e80..607db63044a5 100644
--- a/arch/x86/kernel/irqinit_32.c
+++ b/arch/x86/kernel/irqinit_32.c
@@ -129,7 +129,7 @@ void __init native_init_IRQ(void)
129 for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) { 129 for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
130 /* SYSCALL_VECTOR was reserved in trap_init. */ 130 /* SYSCALL_VECTOR was reserved in trap_init. */
131 if (i != SYSCALL_VECTOR) 131 if (i != SYSCALL_VECTOR)
132 set_intr_gate(i, interrupt[i]); 132 set_intr_gate(i, interrupt[i-FIRST_EXTERNAL_VECTOR]);
133 } 133 }
134 134
135 135
diff --git a/arch/x86/kernel/irqinit_64.c b/arch/x86/kernel/irqinit_64.c
index ff0235391285..8670b3ce626e 100644
--- a/arch/x86/kernel/irqinit_64.c
+++ b/arch/x86/kernel/irqinit_64.c
@@ -24,41 +24,6 @@
24#include <asm/i8259.h> 24#include <asm/i8259.h>
25 25
26/* 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: 27 * ISA PIC or low IO-APIC triggered (INTA-cycle or APIC) interrupts:
63 * (these are usually mapped to vectors 0x30-0x3f) 28 * (these are usually mapped to vectors 0x30-0x3f)
64 */ 29 */
@@ -73,37 +38,6 @@
73 * 38 *
74 * (these are usually mapped into the 0x30-0xff vector range) 39 * (these are usually mapped into the 0x30-0xff vector range)
75 */ 40 */
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 41
108/* 42/*
109 * IRQ2 is cascade interrupt to second interrupt controller 43 * IRQ2 is cascade interrupt to second interrupt controller
diff --git a/arch/x86/kernel/machine_kexec_32.c b/arch/x86/kernel/machine_kexec_32.c
index 7a385746509a..37f420018a41 100644
--- a/arch/x86/kernel/machine_kexec_32.c
+++ b/arch/x86/kernel/machine_kexec_32.c
@@ -13,6 +13,7 @@
13#include <linux/numa.h> 13#include <linux/numa.h>
14#include <linux/ftrace.h> 14#include <linux/ftrace.h>
15#include <linux/suspend.h> 15#include <linux/suspend.h>
16#include <linux/gfp.h>
16 17
17#include <asm/pgtable.h> 18#include <asm/pgtable.h>
18#include <asm/pgalloc.h> 19#include <asm/pgalloc.h>
@@ -25,15 +26,6 @@
25#include <asm/system.h> 26#include <asm/system.h>
26#include <asm/cacheflush.h> 27#include <asm/cacheflush.h>
27 28
28#define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE)))
29static u32 kexec_pgd[1024] PAGE_ALIGNED;
30#ifdef CONFIG_X86_PAE
31static u32 kexec_pmd0[1024] PAGE_ALIGNED;
32static u32 kexec_pmd1[1024] PAGE_ALIGNED;
33#endif
34static u32 kexec_pte0[1024] PAGE_ALIGNED;
35static u32 kexec_pte1[1024] PAGE_ALIGNED;
36
37static void set_idt(void *newidt, __u16 limit) 29static void set_idt(void *newidt, __u16 limit)
38{ 30{
39 struct desc_ptr curidt; 31 struct desc_ptr curidt;
@@ -76,6 +68,76 @@ static void load_segments(void)
76#undef __STR 68#undef __STR
77} 69}
78 70
71static void machine_kexec_free_page_tables(struct kimage *image)
72{
73 free_page((unsigned long)image->arch.pgd);
74#ifdef CONFIG_X86_PAE
75 free_page((unsigned long)image->arch.pmd0);
76 free_page((unsigned long)image->arch.pmd1);
77#endif
78 free_page((unsigned long)image->arch.pte0);
79 free_page((unsigned long)image->arch.pte1);
80}
81
82static int machine_kexec_alloc_page_tables(struct kimage *image)
83{
84 image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
85#ifdef CONFIG_X86_PAE
86 image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
87 image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
88#endif
89 image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
90 image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
91 if (!image->arch.pgd ||
92#ifdef CONFIG_X86_PAE
93 !image->arch.pmd0 || !image->arch.pmd1 ||
94#endif
95 !image->arch.pte0 || !image->arch.pte1) {
96 machine_kexec_free_page_tables(image);
97 return -ENOMEM;
98 }
99 return 0;
100}
101
102static void machine_kexec_page_table_set_one(
103 pgd_t *pgd, pmd_t *pmd, pte_t *pte,
104 unsigned long vaddr, unsigned long paddr)
105{
106 pud_t *pud;
107
108 pgd += pgd_index(vaddr);
109#ifdef CONFIG_X86_PAE
110 if (!(pgd_val(*pgd) & _PAGE_PRESENT))
111 set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
112#endif
113 pud = pud_offset(pgd, vaddr);
114 pmd = pmd_offset(pud, vaddr);
115 if (!(pmd_val(*pmd) & _PAGE_PRESENT))
116 set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
117 pte = pte_offset_kernel(pmd, vaddr);
118 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
119}
120
121static void machine_kexec_prepare_page_tables(struct kimage *image)
122{
123 void *control_page;
124 pmd_t *pmd = 0;
125
126 control_page = page_address(image->control_code_page);
127#ifdef CONFIG_X86_PAE
128 pmd = image->arch.pmd0;
129#endif
130 machine_kexec_page_table_set_one(
131 image->arch.pgd, pmd, image->arch.pte0,
132 (unsigned long)control_page, __pa(control_page));
133#ifdef CONFIG_X86_PAE
134 pmd = image->arch.pmd1;
135#endif
136 machine_kexec_page_table_set_one(
137 image->arch.pgd, pmd, image->arch.pte1,
138 __pa(control_page), __pa(control_page));
139}
140
79/* 141/*
80 * A architecture hook called to validate the 142 * A architecture hook called to validate the
81 * proposed image and prepare the control pages 143 * proposed image and prepare the control pages
@@ -87,12 +149,20 @@ static void load_segments(void)
87 * reboot code buffer to allow us to avoid allocations 149 * reboot code buffer to allow us to avoid allocations
88 * later. 150 * later.
89 * 151 *
90 * Make control page executable. 152 * - Make control page executable.
153 * - Allocate page tables
154 * - Setup page tables
91 */ 155 */
92int machine_kexec_prepare(struct kimage *image) 156int machine_kexec_prepare(struct kimage *image)
93{ 157{
158 int error;
159
94 if (nx_enabled) 160 if (nx_enabled)
95 set_pages_x(image->control_code_page, 1); 161 set_pages_x(image->control_code_page, 1);
162 error = machine_kexec_alloc_page_tables(image);
163 if (error)
164 return error;
165 machine_kexec_prepare_page_tables(image);
96 return 0; 166 return 0;
97} 167}
98 168
@@ -104,6 +174,7 @@ void machine_kexec_cleanup(struct kimage *image)
104{ 174{
105 if (nx_enabled) 175 if (nx_enabled)
106 set_pages_nx(image->control_code_page, 1); 176 set_pages_nx(image->control_code_page, 1);
177 machine_kexec_free_page_tables(image);
107} 178}
108 179
109/* 180/*
@@ -150,18 +221,7 @@ void machine_kexec(struct kimage *image)
150 relocate_kernel_ptr = control_page; 221 relocate_kernel_ptr = control_page;
151 page_list[PA_CONTROL_PAGE] = __pa(control_page); 222 page_list[PA_CONTROL_PAGE] = __pa(control_page);
152 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page; 223 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
153 page_list[PA_PGD] = __pa(kexec_pgd); 224 page_list[PA_PGD] = __pa(image->arch.pgd);
154 page_list[VA_PGD] = (unsigned long)kexec_pgd;
155#ifdef CONFIG_X86_PAE
156 page_list[PA_PMD_0] = __pa(kexec_pmd0);
157 page_list[VA_PMD_0] = (unsigned long)kexec_pmd0;
158 page_list[PA_PMD_1] = __pa(kexec_pmd1);
159 page_list[VA_PMD_1] = (unsigned long)kexec_pmd1;
160#endif
161 page_list[PA_PTE_0] = __pa(kexec_pte0);
162 page_list[VA_PTE_0] = (unsigned long)kexec_pte0;
163 page_list[PA_PTE_1] = __pa(kexec_pte1);
164 page_list[VA_PTE_1] = (unsigned long)kexec_pte1;
165 225
166 if (image->type == KEXEC_TYPE_DEFAULT) 226 if (image->type == KEXEC_TYPE_DEFAULT)
167 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page) 227 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
diff --git a/arch/x86/kernel/microcode_amd.c b/arch/x86/kernel/microcode_amd.c
index 5f8e5d75a254..c25fdb382292 100644
--- a/arch/x86/kernel/microcode_amd.c
+++ b/arch/x86/kernel/microcode_amd.c
@@ -10,7 +10,7 @@
10 * This driver allows to upgrade microcode on AMD 10 * This driver allows to upgrade microcode on AMD
11 * family 0x10 and 0x11 processors. 11 * family 0x10 and 0x11 processors.
12 * 12 *
13 * Licensed unter the terms of the GNU General Public 13 * Licensed under the terms of the GNU General Public
14 * License version 2. See file COPYING for details. 14 * License version 2. See file COPYING for details.
15*/ 15*/
16 16
@@ -32,9 +32,9 @@
32#include <linux/platform_device.h> 32#include <linux/platform_device.h>
33#include <linux/pci.h> 33#include <linux/pci.h>
34#include <linux/pci_ids.h> 34#include <linux/pci_ids.h>
35#include <linux/uaccess.h>
35 36
36#include <asm/msr.h> 37#include <asm/msr.h>
37#include <asm/uaccess.h>
38#include <asm/processor.h> 38#include <asm/processor.h>
39#include <asm/microcode.h> 39#include <asm/microcode.h>
40 40
@@ -47,43 +47,38 @@ MODULE_LICENSE("GPL v2");
47#define UCODE_UCODE_TYPE 0x00000001 47#define UCODE_UCODE_TYPE 0x00000001
48 48
49struct equiv_cpu_entry { 49struct equiv_cpu_entry {
50 unsigned int installed_cpu; 50 u32 installed_cpu;
51 unsigned int fixed_errata_mask; 51 u32 fixed_errata_mask;
52 unsigned int fixed_errata_compare; 52 u32 fixed_errata_compare;
53 unsigned int equiv_cpu; 53 u16 equiv_cpu;
54}; 54 u16 res;
55} __attribute__((packed));
55 56
56struct microcode_header_amd { 57struct microcode_header_amd {
57 unsigned int data_code; 58 u32 data_code;
58 unsigned int patch_id; 59 u32 patch_id;
59 unsigned char mc_patch_data_id[2]; 60 u16 mc_patch_data_id;
60 unsigned char mc_patch_data_len; 61 u8 mc_patch_data_len;
61 unsigned char init_flag; 62 u8 init_flag;
62 unsigned int mc_patch_data_checksum; 63 u32 mc_patch_data_checksum;
63 unsigned int nb_dev_id; 64 u32 nb_dev_id;
64 unsigned int sb_dev_id; 65 u32 sb_dev_id;
65 unsigned char processor_rev_id[2]; 66 u16 processor_rev_id;
66 unsigned char nb_rev_id; 67 u8 nb_rev_id;
67 unsigned char sb_rev_id; 68 u8 sb_rev_id;
68 unsigned char bios_api_rev; 69 u8 bios_api_rev;
69 unsigned char reserved1[3]; 70 u8 reserved1[3];
70 unsigned int match_reg[8]; 71 u32 match_reg[8];
71}; 72} __attribute__((packed));
72 73
73struct microcode_amd { 74struct microcode_amd {
74 struct microcode_header_amd hdr; 75 struct microcode_header_amd hdr;
75 unsigned int mpb[0]; 76 unsigned int mpb[0];
76}; 77};
77 78
78#define UCODE_MAX_SIZE (2048) 79#define UCODE_MAX_SIZE 2048
79#define DEFAULT_UCODE_DATASIZE (896) 80#define UCODE_CONTAINER_SECTION_HDR 8
80#define MC_HEADER_SIZE (sizeof(struct microcode_header_amd)) 81#define UCODE_CONTAINER_HEADER_SIZE 12
81#define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE)
82#define DWSIZE (sizeof(u32))
83/* For now we support a fixed ucode total size only */
84#define get_totalsize(mc) \
85 ((((struct microcode_amd *)mc)->hdr.mc_patch_data_len * 28) \
86 + MC_HEADER_SIZE)
87 82
88/* serialize access to the physical write */ 83/* serialize access to the physical write */
89static DEFINE_SPINLOCK(microcode_update_lock); 84static DEFINE_SPINLOCK(microcode_update_lock);
@@ -93,31 +88,24 @@ static struct equiv_cpu_entry *equiv_cpu_table;
93static int collect_cpu_info_amd(int cpu, struct cpu_signature *csig) 88static int collect_cpu_info_amd(int cpu, struct cpu_signature *csig)
94{ 89{
95 struct cpuinfo_x86 *c = &cpu_data(cpu); 90 struct cpuinfo_x86 *c = &cpu_data(cpu);
91 u32 dummy;
96 92
97 memset(csig, 0, sizeof(*csig)); 93 memset(csig, 0, sizeof(*csig));
98
99 if (c->x86_vendor != X86_VENDOR_AMD || c->x86 < 0x10) { 94 if (c->x86_vendor != X86_VENDOR_AMD || c->x86 < 0x10) {
100 printk(KERN_ERR "microcode: CPU%d not a capable AMD processor\n", 95 printk(KERN_WARNING "microcode: CPU%d: AMD CPU family 0x%x not "
101 cpu); 96 "supported\n", cpu, c->x86);
102 return -1; 97 return -1;
103 } 98 }
104 99 rdmsr(MSR_AMD64_PATCH_LEVEL, csig->rev, dummy);
105 asm volatile("movl %1, %%ecx; rdmsr" 100 printk(KERN_INFO "microcode: CPU%d: patch_level=0x%x\n", cpu, csig->rev);
106 : "=a" (csig->rev)
107 : "i" (0x0000008B) : "ecx");
108
109 printk(KERN_INFO "microcode: collect_cpu_info_amd : patch_id=0x%x\n",
110 csig->rev);
111
112 return 0; 101 return 0;
113} 102}
114 103
115static int get_matching_microcode(int cpu, void *mc, int rev) 104static int get_matching_microcode(int cpu, void *mc, int rev)
116{ 105{
117 struct microcode_header_amd *mc_header = mc; 106 struct microcode_header_amd *mc_header = mc;
118 struct pci_dev *nb_pci_dev, *sb_pci_dev;
119 unsigned int current_cpu_id; 107 unsigned int current_cpu_id;
120 unsigned int equiv_cpu_id = 0x00; 108 u16 equiv_cpu_id = 0;
121 unsigned int i = 0; 109 unsigned int i = 0;
122 110
123 BUG_ON(equiv_cpu_table == NULL); 111 BUG_ON(equiv_cpu_table == NULL);
@@ -132,57 +120,25 @@ static int get_matching_microcode(int cpu, void *mc, int rev)
132 } 120 }
133 121
134 if (!equiv_cpu_id) { 122 if (!equiv_cpu_id) {
135 printk(KERN_ERR "microcode: CPU%d cpu_id " 123 printk(KERN_WARNING "microcode: CPU%d: cpu revision "
136 "not found in equivalent cpu table \n", cpu); 124 "not listed in equivalent cpu table\n", cpu);
137 return 0; 125 return 0;
138 } 126 }
139 127
140 if ((mc_header->processor_rev_id[0]) != (equiv_cpu_id & 0xff)) { 128 if (mc_header->processor_rev_id != equiv_cpu_id) {
141 printk(KERN_ERR 129 printk(KERN_ERR "microcode: CPU%d: patch mismatch "
142 "microcode: CPU%d patch does not match " 130 "(processor_rev_id: %x, equiv_cpu_id: %x)\n",
143 "(patch is %x, cpu extended is %x) \n", 131 cpu, mc_header->processor_rev_id, equiv_cpu_id);
144 cpu, mc_header->processor_rev_id[0],
145 (equiv_cpu_id & 0xff));
146 return 0; 132 return 0;
147 } 133 }
148 134
149 if ((mc_header->processor_rev_id[1]) != ((equiv_cpu_id >> 16) & 0xff)) { 135 /* ucode might be chipset specific -- currently we don't support this */
150 printk(KERN_ERR "microcode: CPU%d patch does not match " 136 if (mc_header->nb_dev_id || mc_header->sb_dev_id) {
151 "(patch is %x, cpu base id is %x) \n", 137 printk(KERN_ERR "microcode: CPU%d: loading of chipset "
152 cpu, mc_header->processor_rev_id[1], 138 "specific code not yet supported\n", cpu);
153 ((equiv_cpu_id >> 16) & 0xff));
154
155 return 0; 139 return 0;
156 } 140 }
157 141
158 /* ucode may be northbridge specific */
159 if (mc_header->nb_dev_id) {
160 nb_pci_dev = pci_get_device(PCI_VENDOR_ID_AMD,
161 (mc_header->nb_dev_id & 0xff),
162 NULL);
163 if ((!nb_pci_dev) ||
164 (mc_header->nb_rev_id != nb_pci_dev->revision)) {
165 printk(KERN_ERR "microcode: CPU%d NB mismatch \n", cpu);
166 pci_dev_put(nb_pci_dev);
167 return 0;
168 }
169 pci_dev_put(nb_pci_dev);
170 }
171
172 /* ucode may be southbridge specific */
173 if (mc_header->sb_dev_id) {
174 sb_pci_dev = pci_get_device(PCI_VENDOR_ID_AMD,
175 (mc_header->sb_dev_id & 0xff),
176 NULL);
177 if ((!sb_pci_dev) ||
178 (mc_header->sb_rev_id != sb_pci_dev->revision)) {
179 printk(KERN_ERR "microcode: CPU%d SB mismatch \n", cpu);
180 pci_dev_put(sb_pci_dev);
181 return 0;
182 }
183 pci_dev_put(sb_pci_dev);
184 }
185
186 if (mc_header->patch_id <= rev) 142 if (mc_header->patch_id <= rev)
187 return 0; 143 return 0;
188 144
@@ -192,12 +148,10 @@ static int get_matching_microcode(int cpu, void *mc, int rev)
192static void apply_microcode_amd(int cpu) 148static void apply_microcode_amd(int cpu)
193{ 149{
194 unsigned long flags; 150 unsigned long flags;
195 unsigned int eax, edx; 151 u32 rev, dummy;
196 unsigned int rev;
197 int cpu_num = raw_smp_processor_id(); 152 int cpu_num = raw_smp_processor_id();
198 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num; 153 struct ucode_cpu_info *uci = ucode_cpu_info + cpu_num;
199 struct microcode_amd *mc_amd = uci->mc; 154 struct microcode_amd *mc_amd = uci->mc;
200 unsigned long addr;
201 155
202 /* We should bind the task to the CPU */ 156 /* We should bind the task to the CPU */
203 BUG_ON(cpu_num != cpu); 157 BUG_ON(cpu_num != cpu);
@@ -206,42 +160,34 @@ static void apply_microcode_amd(int cpu)
206 return; 160 return;
207 161
208 spin_lock_irqsave(&microcode_update_lock, flags); 162 spin_lock_irqsave(&microcode_update_lock, flags);
209 163 wrmsrl(MSR_AMD64_PATCH_LOADER, (u64)(long)&mc_amd->hdr.data_code);
210 addr = (unsigned long)&mc_amd->hdr.data_code;
211 edx = (unsigned int)(((unsigned long)upper_32_bits(addr)));
212 eax = (unsigned int)(((unsigned long)lower_32_bits(addr)));
213
214 asm volatile("movl %0, %%ecx; wrmsr" :
215 : "i" (0xc0010020), "a" (eax), "d" (edx) : "ecx");
216
217 /* get patch id after patching */ 164 /* get patch id after patching */
218 asm volatile("movl %1, %%ecx; rdmsr" 165 rdmsr(MSR_AMD64_PATCH_LEVEL, rev, dummy);
219 : "=a" (rev)
220 : "i" (0x0000008B) : "ecx");
221
222 spin_unlock_irqrestore(&microcode_update_lock, flags); 166 spin_unlock_irqrestore(&microcode_update_lock, flags);
223 167
224 /* check current patch id and patch's id for match */ 168 /* check current patch id and patch's id for match */
225 if (rev != mc_amd->hdr.patch_id) { 169 if (rev != mc_amd->hdr.patch_id) {
226 printk(KERN_ERR "microcode: CPU%d update from revision " 170 printk(KERN_ERR "microcode: CPU%d: update failed "
227 "0x%x to 0x%x failed\n", cpu_num, 171 "(for patch_level=0x%x)\n", cpu, mc_amd->hdr.patch_id);
228 mc_amd->hdr.patch_id, rev);
229 return; 172 return;
230 } 173 }
231 174
232 printk(KERN_INFO "microcode: CPU%d updated from revision " 175 printk(KERN_INFO "microcode: CPU%d: updated (new patch_level=0x%x)\n",
233 "0x%x to 0x%x \n", 176 cpu, rev);
234 cpu_num, uci->cpu_sig.rev, mc_amd->hdr.patch_id);
235 177
236 uci->cpu_sig.rev = rev; 178 uci->cpu_sig.rev = rev;
237} 179}
238 180
239static void * get_next_ucode(u8 *buf, unsigned int size, 181static int get_ucode_data(void *to, const u8 *from, size_t n)
240 int (*get_ucode_data)(void *, const void *, size_t), 182{
241 unsigned int *mc_size) 183 memcpy(to, from, n);
184 return 0;
185}
186
187static void *get_next_ucode(const u8 *buf, unsigned int size,
188 unsigned int *mc_size)
242{ 189{
243 unsigned int total_size; 190 unsigned int total_size;
244#define UCODE_CONTAINER_SECTION_HDR 8
245 u8 section_hdr[UCODE_CONTAINER_SECTION_HDR]; 191 u8 section_hdr[UCODE_CONTAINER_SECTION_HDR];
246 void *mc; 192 void *mc;
247 193
@@ -249,39 +195,37 @@ static void * get_next_ucode(u8 *buf, unsigned int size,
249 return NULL; 195 return NULL;
250 196
251 if (section_hdr[0] != UCODE_UCODE_TYPE) { 197 if (section_hdr[0] != UCODE_UCODE_TYPE) {
252 printk(KERN_ERR "microcode: error! " 198 printk(KERN_ERR "microcode: error: invalid type field in "
253 "Wrong microcode payload type field\n"); 199 "container file section header\n");
254 return NULL; 200 return NULL;
255 } 201 }
256 202
257 total_size = (unsigned long) (section_hdr[4] + (section_hdr[5] << 8)); 203 total_size = (unsigned long) (section_hdr[4] + (section_hdr[5] << 8));
258 204
259 printk(KERN_INFO "microcode: size %u, total_size %u\n", 205 printk(KERN_DEBUG "microcode: size %u, total_size %u\n",
260 size, total_size); 206 size, total_size);
261 207
262 if (total_size > size || total_size > UCODE_MAX_SIZE) { 208 if (total_size > size || total_size > UCODE_MAX_SIZE) {
263 printk(KERN_ERR "microcode: error! Bad data in microcode data file\n"); 209 printk(KERN_ERR "microcode: error: size mismatch\n");
264 return NULL; 210 return NULL;
265 } 211 }
266 212
267 mc = vmalloc(UCODE_MAX_SIZE); 213 mc = vmalloc(UCODE_MAX_SIZE);
268 if (mc) { 214 if (mc) {
269 memset(mc, 0, UCODE_MAX_SIZE); 215 memset(mc, 0, UCODE_MAX_SIZE);
270 if (get_ucode_data(mc, buf + UCODE_CONTAINER_SECTION_HDR, total_size)) { 216 if (get_ucode_data(mc, buf + UCODE_CONTAINER_SECTION_HDR,
217 total_size)) {
271 vfree(mc); 218 vfree(mc);
272 mc = NULL; 219 mc = NULL;
273 } else 220 } else
274 *mc_size = total_size + UCODE_CONTAINER_SECTION_HDR; 221 *mc_size = total_size + UCODE_CONTAINER_SECTION_HDR;
275 } 222 }
276#undef UCODE_CONTAINER_SECTION_HDR
277 return mc; 223 return mc;
278} 224}
279 225
280 226
281static int install_equiv_cpu_table(u8 *buf, 227static int install_equiv_cpu_table(const u8 *buf)
282 int (*get_ucode_data)(void *, const void *, size_t))
283{ 228{
284#define UCODE_CONTAINER_HEADER_SIZE 12
285 u8 *container_hdr[UCODE_CONTAINER_HEADER_SIZE]; 229 u8 *container_hdr[UCODE_CONTAINER_HEADER_SIZE];
286 unsigned int *buf_pos = (unsigned int *)container_hdr; 230 unsigned int *buf_pos = (unsigned int *)container_hdr;
287 unsigned long size; 231 unsigned long size;
@@ -292,14 +236,15 @@ static int install_equiv_cpu_table(u8 *buf,
292 size = buf_pos[2]; 236 size = buf_pos[2];
293 237
294 if (buf_pos[1] != UCODE_EQUIV_CPU_TABLE_TYPE || !size) { 238 if (buf_pos[1] != UCODE_EQUIV_CPU_TABLE_TYPE || !size) {
295 printk(KERN_ERR "microcode: error! " 239 printk(KERN_ERR "microcode: error: invalid type field in "
296 "Wrong microcode equivalnet cpu table\n"); 240 "container file section header\n");
297 return 0; 241 return 0;
298 } 242 }
299 243
300 equiv_cpu_table = (struct equiv_cpu_entry *) vmalloc(size); 244 equiv_cpu_table = (struct equiv_cpu_entry *) vmalloc(size);
301 if (!equiv_cpu_table) { 245 if (!equiv_cpu_table) {
302 printk(KERN_ERR "microcode: error, can't allocate memory for equiv CPU table\n"); 246 printk(KERN_ERR "microcode: failed to allocate "
247 "equivalent CPU table\n");
303 return 0; 248 return 0;
304 } 249 }
305 250
@@ -310,7 +255,6 @@ static int install_equiv_cpu_table(u8 *buf,
310 } 255 }
311 256
312 return size + UCODE_CONTAINER_HEADER_SIZE; /* add header length */ 257 return size + UCODE_CONTAINER_HEADER_SIZE; /* add header length */
313#undef UCODE_CONTAINER_HEADER_SIZE
314} 258}
315 259
316static void free_equiv_cpu_table(void) 260static void free_equiv_cpu_table(void)
@@ -321,18 +265,20 @@ static void free_equiv_cpu_table(void)
321 } 265 }
322} 266}
323 267
324static int generic_load_microcode(int cpu, void *data, size_t size, 268static int generic_load_microcode(int cpu, const u8 *data, size_t size)
325 int (*get_ucode_data)(void *, const void *, size_t))
326{ 269{
327 struct ucode_cpu_info *uci = ucode_cpu_info + cpu; 270 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
328 u8 *ucode_ptr = data, *new_mc = NULL, *mc; 271 const u8 *ucode_ptr = data;
272 void *new_mc = NULL;
273 void *mc;
329 int new_rev = uci->cpu_sig.rev; 274 int new_rev = uci->cpu_sig.rev;
330 unsigned int leftover; 275 unsigned int leftover;
331 unsigned long offset; 276 unsigned long offset;
332 277
333 offset = install_equiv_cpu_table(ucode_ptr, get_ucode_data); 278 offset = install_equiv_cpu_table(ucode_ptr);
334 if (!offset) { 279 if (!offset) {
335 printk(KERN_ERR "microcode: installing equivalent cpu table failed\n"); 280 printk(KERN_ERR "microcode: failed to create "
281 "equivalent cpu table\n");
336 return -EINVAL; 282 return -EINVAL;
337 } 283 }
338 284
@@ -343,7 +289,7 @@ static int generic_load_microcode(int cpu, void *data, size_t size,
343 unsigned int uninitialized_var(mc_size); 289 unsigned int uninitialized_var(mc_size);
344 struct microcode_header_amd *mc_header; 290 struct microcode_header_amd *mc_header;
345 291
346 mc = get_next_ucode(ucode_ptr, leftover, get_ucode_data, &mc_size); 292 mc = get_next_ucode(ucode_ptr, leftover, &mc_size);
347 if (!mc) 293 if (!mc)
348 break; 294 break;
349 295
@@ -353,7 +299,7 @@ static int generic_load_microcode(int cpu, void *data, size_t size,
353 vfree(new_mc); 299 vfree(new_mc);
354 new_rev = mc_header->patch_id; 300 new_rev = mc_header->patch_id;
355 new_mc = mc; 301 new_mc = mc;
356 } else 302 } else
357 vfree(mc); 303 vfree(mc);
358 304
359 ucode_ptr += mc_size; 305 ucode_ptr += mc_size;
@@ -365,9 +311,9 @@ static int generic_load_microcode(int cpu, void *data, size_t size,
365 if (uci->mc) 311 if (uci->mc)
366 vfree(uci->mc); 312 vfree(uci->mc);
367 uci->mc = new_mc; 313 uci->mc = new_mc;
368 pr_debug("microcode: CPU%d found a matching microcode update with" 314 pr_debug("microcode: CPU%d found a matching microcode "
369 " version 0x%x (current=0x%x)\n", 315 "update with version 0x%x (current=0x%x)\n",
370 cpu, new_rev, uci->cpu_sig.rev); 316 cpu, new_rev, uci->cpu_sig.rev);
371 } else 317 } else
372 vfree(new_mc); 318 vfree(new_mc);
373 } 319 }
@@ -377,12 +323,6 @@ static int generic_load_microcode(int cpu, void *data, size_t size,
377 return (int)leftover; 323 return (int)leftover;
378} 324}
379 325
380static int get_ucode_fw(void *to, const void *from, size_t n)
381{
382 memcpy(to, from, n);
383 return 0;
384}
385
386static int request_microcode_fw(int cpu, struct device *device) 326static int request_microcode_fw(int cpu, struct device *device)
387{ 327{
388 const char *fw_name = "amd-ucode/microcode_amd.bin"; 328 const char *fw_name = "amd-ucode/microcode_amd.bin";
@@ -394,12 +334,11 @@ static int request_microcode_fw(int cpu, struct device *device)
394 334
395 ret = request_firmware(&firmware, fw_name, device); 335 ret = request_firmware(&firmware, fw_name, device);
396 if (ret) { 336 if (ret) {
397 printk(KERN_ERR "microcode: ucode data file %s load failed\n", fw_name); 337 printk(KERN_ERR "microcode: failed to load file %s\n", fw_name);
398 return ret; 338 return ret;
399 } 339 }
400 340
401 ret = generic_load_microcode(cpu, (void*)firmware->data, firmware->size, 341 ret = generic_load_microcode(cpu, firmware->data, firmware->size);
402 &get_ucode_fw);
403 342
404 release_firmware(firmware); 343 release_firmware(firmware);
405 344
@@ -408,8 +347,8 @@ static int request_microcode_fw(int cpu, struct device *device)
408 347
409static int request_microcode_user(int cpu, const void __user *buf, size_t size) 348static int request_microcode_user(int cpu, const void __user *buf, size_t size)
410{ 349{
411 printk(KERN_WARNING "microcode: AMD microcode update via /dev/cpu/microcode" 350 printk(KERN_INFO "microcode: AMD microcode update via "
412 "is not supported\n"); 351 "/dev/cpu/microcode not supported\n");
413 return -1; 352 return -1;
414} 353}
415 354
@@ -433,3 +372,4 @@ struct microcode_ops * __init init_amd_microcode(void)
433{ 372{
434 return &microcode_amd_ops; 373 return &microcode_amd_ops;
435} 374}
375
diff --git a/arch/x86/kernel/microcode_core.c b/arch/x86/kernel/microcode_core.c
index c4b5b24e0217..c9b721ba968c 100644
--- a/arch/x86/kernel/microcode_core.c
+++ b/arch/x86/kernel/microcode_core.c
@@ -99,7 +99,7 @@ MODULE_LICENSE("GPL");
99 99
100#define MICROCODE_VERSION "2.00" 100#define MICROCODE_VERSION "2.00"
101 101
102struct microcode_ops *microcode_ops; 102static struct microcode_ops *microcode_ops;
103 103
104/* no concurrent ->write()s are allowed on /dev/cpu/microcode */ 104/* no concurrent ->write()s are allowed on /dev/cpu/microcode */
105static DEFINE_MUTEX(microcode_mutex); 105static DEFINE_MUTEX(microcode_mutex);
@@ -203,7 +203,7 @@ MODULE_ALIAS_MISCDEV(MICROCODE_MINOR);
203#endif 203#endif
204 204
205/* fake device for request_firmware */ 205/* fake device for request_firmware */
206struct platform_device *microcode_pdev; 206static struct platform_device *microcode_pdev;
207 207
208static ssize_t reload_store(struct sys_device *dev, 208static ssize_t reload_store(struct sys_device *dev,
209 struct sysdev_attribute *attr, 209 struct sysdev_attribute *attr,
@@ -328,7 +328,7 @@ static int microcode_resume_cpu(int cpu)
328 return 0; 328 return 0;
329} 329}
330 330
331void microcode_update_cpu(int cpu) 331static void microcode_update_cpu(int cpu)
332{ 332{
333 struct ucode_cpu_info *uci = ucode_cpu_info + cpu; 333 struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
334 int err = 0; 334 int err = 0;
diff --git a/arch/x86/kernel/microcode_intel.c b/arch/x86/kernel/microcode_intel.c
index a8e62792d171..b7f4c929e615 100644
--- a/arch/x86/kernel/microcode_intel.c
+++ b/arch/x86/kernel/microcode_intel.c
@@ -471,7 +471,7 @@ static void microcode_fini_cpu(int cpu)
471 uci->mc = NULL; 471 uci->mc = NULL;
472} 472}
473 473
474struct microcode_ops microcode_intel_ops = { 474static struct microcode_ops microcode_intel_ops = {
475 .request_microcode_user = request_microcode_user, 475 .request_microcode_user = request_microcode_user,
476 .request_microcode_fw = request_microcode_fw, 476 .request_microcode_fw = request_microcode_fw,
477 .collect_cpu_info = collect_cpu_info, 477 .collect_cpu_info = collect_cpu_info,
diff --git a/arch/x86/kernel/mpparse.c b/arch/x86/kernel/mpparse.c
index 0f4c1fd5a1f4..45e3b69808ba 100644
--- a/arch/x86/kernel/mpparse.c
+++ b/arch/x86/kernel/mpparse.c
@@ -586,26 +586,23 @@ static void __init __get_smp_config(unsigned int early)
586{ 586{
587 struct intel_mp_floating *mpf = mpf_found; 587 struct intel_mp_floating *mpf = mpf_found;
588 588
589 if (x86_quirks->mach_get_smp_config) { 589 if (!mpf)
590 if (x86_quirks->mach_get_smp_config(early)) 590 return;
591 return; 591
592 }
593 if (acpi_lapic && early) 592 if (acpi_lapic && early)
594 return; 593 return;
594
595 /* 595 /*
596 * ACPI supports both logical (e.g. Hyper-Threading) and physical 596 * MPS doesn't support hyperthreading, aka only have
597 * processors, where MPS only supports physical. 597 * thread 0 apic id in MPS table
598 */ 598 */
599 if (acpi_lapic && acpi_ioapic) { 599 if (acpi_lapic && acpi_ioapic)
600 printk(KERN_INFO "Using ACPI (MADT) for SMP configuration "
601 "information\n");
602 return; 600 return;
603 } else if (acpi_lapic)
604 printk(KERN_INFO "Using ACPI for processor (LAPIC) "
605 "configuration information\n");
606 601
607 if (!mpf) 602 if (x86_quirks->mach_get_smp_config) {
608 return; 603 if (x86_quirks->mach_get_smp_config(early))
604 return;
605 }
609 606
610 printk(KERN_INFO "Intel MultiProcessor Specification v1.%d\n", 607 printk(KERN_INFO "Intel MultiProcessor Specification v1.%d\n",
611 mpf->mpf_specification); 608 mpf->mpf_specification);
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index 2c97f07f1c2c..8bd1bf9622a7 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -131,6 +131,11 @@ static void report_broken_nmi(int cpu, int *prev_nmi_count)
131 atomic_dec(&nmi_active); 131 atomic_dec(&nmi_active);
132} 132}
133 133
134static void __acpi_nmi_disable(void *__unused)
135{
136 apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED);
137}
138
134int __init check_nmi_watchdog(void) 139int __init check_nmi_watchdog(void)
135{ 140{
136 unsigned int *prev_nmi_count; 141 unsigned int *prev_nmi_count;
@@ -179,8 +184,12 @@ int __init check_nmi_watchdog(void)
179 kfree(prev_nmi_count); 184 kfree(prev_nmi_count);
180 return 0; 185 return 0;
181error: 186error:
182 if (nmi_watchdog == NMI_IO_APIC && !timer_through_8259) 187 if (nmi_watchdog == NMI_IO_APIC) {
183 disable_8259A_irq(0); 188 if (!timer_through_8259)
189 disable_8259A_irq(0);
190 on_each_cpu(__acpi_nmi_disable, NULL, 1);
191 }
192
184#ifdef CONFIG_X86_32 193#ifdef CONFIG_X86_32
185 timer_ack = 0; 194 timer_ack = 0;
186#endif 195#endif
@@ -199,12 +208,17 @@ static int __init setup_nmi_watchdog(char *str)
199 ++str; 208 ++str;
200 } 209 }
201 210
202 get_option(&str, &nmi); 211 if (!strncmp(str, "lapic", 5))
203 212 nmi_watchdog = NMI_LOCAL_APIC;
204 if (nmi >= NMI_INVALID) 213 else if (!strncmp(str, "ioapic", 6))
205 return 0; 214 nmi_watchdog = NMI_IO_APIC;
215 else {
216 get_option(&str, &nmi);
217 if (nmi >= NMI_INVALID)
218 return 0;
219 nmi_watchdog = nmi;
220 }
206 221
207 nmi_watchdog = nmi;
208 return 1; 222 return 1;
209} 223}
210__setup("nmi_watchdog=", setup_nmi_watchdog); 224__setup("nmi_watchdog=", setup_nmi_watchdog);
@@ -285,11 +299,6 @@ void acpi_nmi_enable(void)
285 on_each_cpu(__acpi_nmi_enable, NULL, 1); 299 on_each_cpu(__acpi_nmi_enable, NULL, 1);
286} 300}
287 301
288static void __acpi_nmi_disable(void *__unused)
289{
290 apic_write(APIC_LVT0, APIC_DM_NMI | APIC_LVT_MASKED);
291}
292
293/* 302/*
294 * Disable timer based NMIs on all CPUs: 303 * Disable timer based NMIs on all CPUs:
295 */ 304 */
@@ -340,6 +349,8 @@ void stop_apic_nmi_watchdog(void *unused)
340 return; 349 return;
341 if (nmi_watchdog == NMI_LOCAL_APIC) 350 if (nmi_watchdog == NMI_LOCAL_APIC)
342 lapic_watchdog_stop(); 351 lapic_watchdog_stop();
352 else
353 __acpi_nmi_disable(NULL);
343 __get_cpu_var(wd_enabled) = 0; 354 __get_cpu_var(wd_enabled) = 0;
344 atomic_dec(&nmi_active); 355 atomic_dec(&nmi_active);
345} 356}
@@ -465,6 +476,24 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason)
465 476
466#ifdef CONFIG_SYSCTL 477#ifdef CONFIG_SYSCTL
467 478
479static void enable_ioapic_nmi_watchdog_single(void *unused)
480{
481 __get_cpu_var(wd_enabled) = 1;
482 atomic_inc(&nmi_active);
483 __acpi_nmi_enable(NULL);
484}
485
486static void enable_ioapic_nmi_watchdog(void)
487{
488 on_each_cpu(enable_ioapic_nmi_watchdog_single, NULL, 1);
489 touch_nmi_watchdog();
490}
491
492static void disable_ioapic_nmi_watchdog(void)
493{
494 on_each_cpu(stop_apic_nmi_watchdog, NULL, 1);
495}
496
468static int __init setup_unknown_nmi_panic(char *str) 497static int __init setup_unknown_nmi_panic(char *str)
469{ 498{
470 unknown_nmi_panic = 1; 499 unknown_nmi_panic = 1;
@@ -507,6 +536,11 @@ int proc_nmi_enabled(struct ctl_table *table, int write, struct file *file,
507 enable_lapic_nmi_watchdog(); 536 enable_lapic_nmi_watchdog();
508 else 537 else
509 disable_lapic_nmi_watchdog(); 538 disable_lapic_nmi_watchdog();
539 } else if (nmi_watchdog == NMI_IO_APIC) {
540 if (nmi_watchdog_enabled)
541 enable_ioapic_nmi_watchdog();
542 else
543 disable_ioapic_nmi_watchdog();
510 } else { 544 } else {
511 printk(KERN_WARNING 545 printk(KERN_WARNING
512 "NMI watchdog doesn't know what hardware to touch\n"); 546 "NMI watchdog doesn't know what hardware to touch\n");
diff --git a/arch/x86/kernel/numaq_32.c b/arch/x86/kernel/numaq_32.c
index 4caff39078e0..0deea37a53cf 100644
--- a/arch/x86/kernel/numaq_32.c
+++ b/arch/x86/kernel/numaq_32.c
@@ -31,7 +31,7 @@
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> 34#include <asm/genapic.h>
35#include <asm/e820.h> 35#include <asm/e820.h>
36#include <asm/setup.h> 36#include <asm/setup.h>
37 37
@@ -235,6 +235,13 @@ static int __init numaq_setup_ioapic_ids(void)
235 return 1; 235 return 1;
236} 236}
237 237
238static int __init numaq_update_genapic(void)
239{
240 genapic->wakeup_cpu = wakeup_secondary_cpu_via_nmi;
241
242 return 0;
243}
244
238static struct x86_quirks numaq_x86_quirks __initdata = { 245static struct x86_quirks numaq_x86_quirks __initdata = {
239 .arch_pre_time_init = numaq_pre_time_init, 246 .arch_pre_time_init = numaq_pre_time_init,
240 .arch_time_init = NULL, 247 .arch_time_init = NULL,
@@ -250,6 +257,7 @@ static struct x86_quirks numaq_x86_quirks __initdata = {
250 .mpc_oem_pci_bus = mpc_oem_pci_bus, 257 .mpc_oem_pci_bus = mpc_oem_pci_bus,
251 .smp_read_mpc_oem = smp_read_mpc_oem, 258 .smp_read_mpc_oem = smp_read_mpc_oem,
252 .setup_ioapic_ids = numaq_setup_ioapic_ids, 259 .setup_ioapic_ids = numaq_setup_ioapic_ids,
260 .update_genapic = numaq_update_genapic,
253}; 261};
254 262
255void numaq_mps_oem_check(struct mp_config_table *mpc, char *oem, 263void numaq_mps_oem_check(struct mp_config_table *mpc, char *oem,
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index 192624820217..7a3dfceb90e4 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -6,6 +6,7 @@
6#include <asm/proto.h> 6#include <asm/proto.h>
7#include <asm/dma.h> 7#include <asm/dma.h>
8#include <asm/iommu.h> 8#include <asm/iommu.h>
9#include <asm/gart.h>
9#include <asm/calgary.h> 10#include <asm/calgary.h>
10#include <asm/amd_iommu.h> 11#include <asm/amd_iommu.h>
11 12
@@ -30,11 +31,6 @@ int no_iommu __read_mostly;
30/* Set this to 1 if there is a HW IOMMU in the system */ 31/* Set this to 1 if there is a HW IOMMU in the system */
31int iommu_detected __read_mostly = 0; 32int iommu_detected __read_mostly = 0;
32 33
33/* This tells the BIO block layer to assume merging. Default to off
34 because we cannot guarantee merging later. */
35int iommu_bio_merge __read_mostly = 0;
36EXPORT_SYMBOL(iommu_bio_merge);
37
38dma_addr_t bad_dma_address __read_mostly = 0; 34dma_addr_t bad_dma_address __read_mostly = 0;
39EXPORT_SYMBOL(bad_dma_address); 35EXPORT_SYMBOL(bad_dma_address);
40 36
@@ -188,7 +184,6 @@ static __init int iommu_setup(char *p)
188 } 184 }
189 185
190 if (!strncmp(p, "biomerge", 8)) { 186 if (!strncmp(p, "biomerge", 8)) {
191 iommu_bio_merge = 4096;
192 iommu_merge = 1; 187 iommu_merge = 1;
193 force_iommu = 1; 188 force_iommu = 1;
194 } 189 }
@@ -300,8 +295,8 @@ fs_initcall(pci_iommu_init);
300static __devinit void via_no_dac(struct pci_dev *dev) 295static __devinit void via_no_dac(struct pci_dev *dev)
301{ 296{
302 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI && forbid_dac == 0) { 297 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI && forbid_dac == 0) {
303 printk(KERN_INFO "PCI: VIA PCI bridge detected." 298 printk(KERN_INFO
304 "Disabling DAC.\n"); 299 "PCI: VIA PCI bridge detected. Disabling DAC.\n");
305 forbid_dac = 1; 300 forbid_dac = 1;
306 } 301 }
307} 302}
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index c622772744d8..e68bb9e30864 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -1,13 +1,16 @@
1#include <linux/errno.h> 1#include <linux/errno.h>
2#include <linux/kernel.h> 2#include <linux/kernel.h>
3#include <linux/mm.h> 3#include <linux/mm.h>
4#include <asm/idle.h>
4#include <linux/smp.h> 5#include <linux/smp.h>
5#include <linux/slab.h> 6#include <linux/slab.h>
6#include <linux/sched.h> 7#include <linux/sched.h>
7#include <linux/module.h> 8#include <linux/module.h>
8#include <linux/pm.h> 9#include <linux/pm.h>
9#include <linux/clockchips.h> 10#include <linux/clockchips.h>
11#include <linux/ftrace.h>
10#include <asm/system.h> 12#include <asm/system.h>
13#include <asm/apic.h>
11 14
12unsigned long idle_halt; 15unsigned long idle_halt;
13EXPORT_SYMBOL(idle_halt); 16EXPORT_SYMBOL(idle_halt);
@@ -100,6 +103,9 @@ static inline int hlt_use_halt(void)
100void default_idle(void) 103void default_idle(void)
101{ 104{
102 if (hlt_use_halt()) { 105 if (hlt_use_halt()) {
106 struct power_trace it;
107
108 trace_power_start(&it, POWER_CSTATE, 1);
103 current_thread_info()->status &= ~TS_POLLING; 109 current_thread_info()->status &= ~TS_POLLING;
104 /* 110 /*
105 * TS_POLLING-cleared state must be visible before we 111 * TS_POLLING-cleared state must be visible before we
@@ -112,6 +118,7 @@ void default_idle(void)
112 else 118 else
113 local_irq_enable(); 119 local_irq_enable();
114 current_thread_info()->status |= TS_POLLING; 120 current_thread_info()->status |= TS_POLLING;
121 trace_power_end(&it);
115 } else { 122 } else {
116 local_irq_enable(); 123 local_irq_enable();
117 /* loop is done by the caller */ 124 /* loop is done by the caller */
@@ -122,6 +129,21 @@ void default_idle(void)
122EXPORT_SYMBOL(default_idle); 129EXPORT_SYMBOL(default_idle);
123#endif 130#endif
124 131
132void stop_this_cpu(void *dummy)
133{
134 local_irq_disable();
135 /*
136 * Remove this CPU:
137 */
138 cpu_clear(smp_processor_id(), cpu_online_map);
139 disable_local_APIC();
140
141 for (;;) {
142 if (hlt_works(smp_processor_id()))
143 halt();
144 }
145}
146
125static void do_nothing(void *unused) 147static void do_nothing(void *unused)
126{ 148{
127} 149}
@@ -154,24 +176,31 @@ EXPORT_SYMBOL_GPL(cpu_idle_wait);
154 */ 176 */
155void mwait_idle_with_hints(unsigned long ax, unsigned long cx) 177void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
156{ 178{
179 struct power_trace it;
180
181 trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
157 if (!need_resched()) { 182 if (!need_resched()) {
158 __monitor((void *)&current_thread_info()->flags, 0, 0); 183 __monitor((void *)&current_thread_info()->flags, 0, 0);
159 smp_mb(); 184 smp_mb();
160 if (!need_resched()) 185 if (!need_resched())
161 __mwait(ax, cx); 186 __mwait(ax, cx);
162 } 187 }
188 trace_power_end(&it);
163} 189}
164 190
165/* Default MONITOR/MWAIT with no hints, used for default C1 state */ 191/* Default MONITOR/MWAIT with no hints, used for default C1 state */
166static void mwait_idle(void) 192static void mwait_idle(void)
167{ 193{
194 struct power_trace it;
168 if (!need_resched()) { 195 if (!need_resched()) {
196 trace_power_start(&it, POWER_CSTATE, 1);
169 __monitor((void *)&current_thread_info()->flags, 0, 0); 197 __monitor((void *)&current_thread_info()->flags, 0, 0);
170 smp_mb(); 198 smp_mb();
171 if (!need_resched()) 199 if (!need_resched())
172 __sti_mwait(0, 0); 200 __sti_mwait(0, 0);
173 else 201 else
174 local_irq_enable(); 202 local_irq_enable();
203 trace_power_end(&it);
175 } else 204 } else
176 local_irq_enable(); 205 local_irq_enable();
177} 206}
@@ -183,9 +212,13 @@ static void mwait_idle(void)
183 */ 212 */
184static void poll_idle(void) 213static void poll_idle(void)
185{ 214{
215 struct power_trace it;
216
217 trace_power_start(&it, POWER_CSTATE, 0);
186 local_irq_enable(); 218 local_irq_enable();
187 while (!need_resched()) 219 while (!need_resched())
188 cpu_relax(); 220 cpu_relax();
221 trace_power_end(&it);
189} 222}
190 223
191/* 224/*
@@ -270,7 +303,7 @@ static void c1e_idle(void)
270 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); 303 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
271 if (lo & K8_INTP_C1E_ACTIVE_MASK) { 304 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
272 c1e_detected = 1; 305 c1e_detected = 1;
273 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) 306 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
274 mark_tsc_unstable("TSC halt in AMD C1E"); 307 mark_tsc_unstable("TSC halt in AMD C1E");
275 printk(KERN_INFO "System has AMD C1E enabled\n"); 308 printk(KERN_INFO "System has AMD C1E enabled\n");
276 set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E); 309 set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 0a1302fe6d45..3ba155d24884 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -38,6 +38,7 @@
38#include <linux/percpu.h> 38#include <linux/percpu.h>
39#include <linux/prctl.h> 39#include <linux/prctl.h>
40#include <linux/dmi.h> 40#include <linux/dmi.h>
41#include <linux/ftrace.h>
41 42
42#include <asm/uaccess.h> 43#include <asm/uaccess.h>
43#include <asm/pgtable.h> 44#include <asm/pgtable.h>
@@ -59,6 +60,7 @@
59#include <asm/idle.h> 60#include <asm/idle.h>
60#include <asm/syscalls.h> 61#include <asm/syscalls.h>
61#include <asm/smp.h> 62#include <asm/smp.h>
63#include <asm/ds.h>
62 64
63asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 65asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
64 66
@@ -250,14 +252,8 @@ void exit_thread(void)
250 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET; 252 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
251 put_cpu(); 253 put_cpu();
252 } 254 }
253#ifdef CONFIG_X86_DS 255
254 /* Free any DS contexts that have not been properly released. */ 256 ds_exit_thread(current);
255 if (unlikely(current->thread.ds_ctx)) {
256 /* we clear debugctl to make sure DS is not used. */
257 update_debugctlmsr(0);
258 ds_free(current->thread.ds_ctx);
259 }
260#endif /* CONFIG_X86_DS */
261} 257}
262 258
263void flush_thread(void) 259void flush_thread(void)
@@ -339,6 +335,12 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
339 kfree(p->thread.io_bitmap_ptr); 335 kfree(p->thread.io_bitmap_ptr);
340 p->thread.io_bitmap_max = 0; 336 p->thread.io_bitmap_max = 0;
341 } 337 }
338
339 ds_copy_thread(p, current);
340
341 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
342 p->thread.debugctlmsr = 0;
343
342 return err; 344 return err;
343} 345}
344 346
@@ -419,48 +421,19 @@ int set_tsc_mode(unsigned int val)
419 return 0; 421 return 0;
420} 422}
421 423
422#ifdef CONFIG_X86_DS
423static int update_debugctl(struct thread_struct *prev,
424 struct thread_struct *next, unsigned long debugctl)
425{
426 unsigned long ds_prev = 0;
427 unsigned long ds_next = 0;
428
429 if (prev->ds_ctx)
430 ds_prev = (unsigned long)prev->ds_ctx->ds;
431 if (next->ds_ctx)
432 ds_next = (unsigned long)next->ds_ctx->ds;
433
434 if (ds_next != ds_prev) {
435 /* we clear debugctl to make sure DS
436 * is not in use when we change it */
437 debugctl = 0;
438 update_debugctlmsr(0);
439 wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
440 }
441 return debugctl;
442}
443#else
444static int update_debugctl(struct thread_struct *prev,
445 struct thread_struct *next, unsigned long debugctl)
446{
447 return debugctl;
448}
449#endif /* CONFIG_X86_DS */
450
451static noinline void 424static noinline void
452__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, 425__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
453 struct tss_struct *tss) 426 struct tss_struct *tss)
454{ 427{
455 struct thread_struct *prev, *next; 428 struct thread_struct *prev, *next;
456 unsigned long debugctl;
457 429
458 prev = &prev_p->thread; 430 prev = &prev_p->thread;
459 next = &next_p->thread; 431 next = &next_p->thread;
460 432
461 debugctl = update_debugctl(prev, next, prev->debugctlmsr); 433 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
462 434 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
463 if (next->debugctlmsr != debugctl) 435 ds_switch_to(prev_p, next_p);
436 else if (next->debugctlmsr != prev->debugctlmsr)
464 update_debugctlmsr(next->debugctlmsr); 437 update_debugctlmsr(next->debugctlmsr);
465 438
466 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { 439 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
@@ -482,15 +455,6 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
482 hard_enable_TSC(); 455 hard_enable_TSC();
483 } 456 }
484 457
485#ifdef CONFIG_X86_PTRACE_BTS
486 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
487 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
488
489 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
490 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
491#endif /* CONFIG_X86_PTRACE_BTS */
492
493
494 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { 458 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
495 /* 459 /*
496 * Disable the bitmap via an invalid offset. We still cache 460 * Disable the bitmap via an invalid offset. We still cache
@@ -548,7 +512,8 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
548 * the task-switch, and shows up in ret_from_fork in entry.S, 512 * the task-switch, and shows up in ret_from_fork in entry.S,
549 * for example. 513 * for example.
550 */ 514 */
551struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 515__notrace_funcgraph struct task_struct *
516__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
552{ 517{
553 struct thread_struct *prev = &prev_p->thread, 518 struct thread_struct *prev = &prev_p->thread,
554 *next = &next_p->thread; 519 *next = &next_p->thread;
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index c958120fb1b6..416fb9282f4f 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -39,6 +39,7 @@
39#include <linux/prctl.h> 39#include <linux/prctl.h>
40#include <linux/uaccess.h> 40#include <linux/uaccess.h>
41#include <linux/io.h> 41#include <linux/io.h>
42#include <linux/ftrace.h>
42 43
43#include <asm/pgtable.h> 44#include <asm/pgtable.h>
44#include <asm/system.h> 45#include <asm/system.h>
@@ -52,6 +53,7 @@
52#include <asm/ia32.h> 53#include <asm/ia32.h>
53#include <asm/idle.h> 54#include <asm/idle.h>
54#include <asm/syscalls.h> 55#include <asm/syscalls.h>
56#include <asm/ds.h>
55 57
56asmlinkage extern void ret_from_fork(void); 58asmlinkage extern void ret_from_fork(void);
57 59
@@ -235,14 +237,8 @@ void exit_thread(void)
235 t->io_bitmap_max = 0; 237 t->io_bitmap_max = 0;
236 put_cpu(); 238 put_cpu();
237 } 239 }
238#ifdef CONFIG_X86_DS 240
239 /* Free any DS contexts that have not been properly released. */ 241 ds_exit_thread(current);
240 if (unlikely(t->ds_ctx)) {
241 /* we clear debugctl to make sure DS is not used. */
242 update_debugctlmsr(0);
243 ds_free(t->ds_ctx);
244 }
245#endif /* CONFIG_X86_DS */
246} 242}
247 243
248void flush_thread(void) 244void flush_thread(void)
@@ -372,6 +368,12 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
372 if (err) 368 if (err)
373 goto out; 369 goto out;
374 } 370 }
371
372 ds_copy_thread(p, me);
373
374 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
375 p->thread.debugctlmsr = 0;
376
375 err = 0; 377 err = 0;
376out: 378out:
377 if (err && p->thread.io_bitmap_ptr) { 379 if (err && p->thread.io_bitmap_ptr) {
@@ -470,35 +472,14 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
470 struct tss_struct *tss) 472 struct tss_struct *tss)
471{ 473{
472 struct thread_struct *prev, *next; 474 struct thread_struct *prev, *next;
473 unsigned long debugctl;
474 475
475 prev = &prev_p->thread, 476 prev = &prev_p->thread,
476 next = &next_p->thread; 477 next = &next_p->thread;
477 478
478 debugctl = prev->debugctlmsr; 479 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
479 480 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
480#ifdef CONFIG_X86_DS 481 ds_switch_to(prev_p, next_p);
481 { 482 else if (next->debugctlmsr != prev->debugctlmsr)
482 unsigned long ds_prev = 0, ds_next = 0;
483
484 if (prev->ds_ctx)
485 ds_prev = (unsigned long)prev->ds_ctx->ds;
486 if (next->ds_ctx)
487 ds_next = (unsigned long)next->ds_ctx->ds;
488
489 if (ds_next != ds_prev) {
490 /*
491 * We clear debugctl to make sure DS
492 * is not in use when we change it:
493 */
494 debugctl = 0;
495 update_debugctlmsr(0);
496 wrmsrl(MSR_IA32_DS_AREA, ds_next);
497 }
498 }
499#endif /* CONFIG_X86_DS */
500
501 if (next->debugctlmsr != debugctl)
502 update_debugctlmsr(next->debugctlmsr); 483 update_debugctlmsr(next->debugctlmsr);
503 484
504 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { 485 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
@@ -533,14 +514,6 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
533 */ 514 */
534 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); 515 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
535 } 516 }
536
537#ifdef CONFIG_X86_PTRACE_BTS
538 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
539 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
540
541 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
542 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
543#endif /* CONFIG_X86_PTRACE_BTS */
544} 517}
545 518
546/* 519/*
@@ -551,8 +524,9 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
551 * - could test fs/gs bitsliced 524 * - could test fs/gs bitsliced
552 * 525 *
553 * Kprobes not supported here. Set the probe on schedule instead. 526 * Kprobes not supported here. Set the probe on schedule instead.
527 * Function graph tracer not supported too.
554 */ 528 */
555struct task_struct * 529__notrace_funcgraph struct task_struct *
556__switch_to(struct task_struct *prev_p, struct task_struct *next_p) 530__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
557{ 531{
558 struct thread_struct *prev = &prev_p->thread; 532 struct thread_struct *prev = &prev_p->thread;
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index 0a6d8c12e10d..0a5df5f82fb9 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -581,158 +581,91 @@ static int ioperm_get(struct task_struct *target,
581} 581}
582 582
583#ifdef CONFIG_X86_PTRACE_BTS 583#ifdef CONFIG_X86_PTRACE_BTS
584/*
585 * The configuration for a particular BTS hardware implementation.
586 */
587struct bts_configuration {
588 /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */
589 unsigned char sizeof_bts;
590 /* the size of a field in the BTS record in bytes */
591 unsigned char sizeof_field;
592 /* a bitmask to enable/disable BTS in DEBUGCTL MSR */
593 unsigned long debugctl_mask;
594};
595static struct bts_configuration bts_cfg;
596
597#define BTS_MAX_RECORD_SIZE (8 * 3)
598
599
600/*
601 * Branch Trace Store (BTS) uses the following format. Different
602 * architectures vary in the size of those fields.
603 * - source linear address
604 * - destination linear address
605 * - flags
606 *
607 * Later architectures use 64bit pointers throughout, whereas earlier
608 * architectures use 32bit pointers in 32bit mode.
609 *
610 * We compute the base address for the first 8 fields based on:
611 * - the field size stored in the DS configuration
612 * - the relative field position
613 *
614 * In order to store additional information in the BTS buffer, we use
615 * a special source address to indicate that the record requires
616 * special interpretation.
617 *
618 * Netburst indicated via a bit in the flags field whether the branch
619 * was predicted; this is ignored.
620 */
621
622enum bts_field {
623 bts_from = 0,
624 bts_to,
625 bts_flags,
626
627 bts_escape = (unsigned long)-1,
628 bts_qual = bts_to,
629 bts_jiffies = bts_flags
630};
631
632static inline unsigned long bts_get(const char *base, enum bts_field field)
633{
634 base += (bts_cfg.sizeof_field * field);
635 return *(unsigned long *)base;
636}
637
638static inline void bts_set(char *base, enum bts_field field, unsigned long val)
639{
640 base += (bts_cfg.sizeof_field * field);;
641 (*(unsigned long *)base) = val;
642}
643
644/*
645 * Translate a BTS record from the raw format into the bts_struct format
646 *
647 * out (out): bts_struct interpretation
648 * raw: raw BTS record
649 */
650static void ptrace_bts_translate_record(struct bts_struct *out, const void *raw)
651{
652 memset(out, 0, sizeof(*out));
653 if (bts_get(raw, bts_from) == bts_escape) {
654 out->qualifier = bts_get(raw, bts_qual);
655 out->variant.jiffies = bts_get(raw, bts_jiffies);
656 } else {
657 out->qualifier = BTS_BRANCH;
658 out->variant.lbr.from_ip = bts_get(raw, bts_from);
659 out->variant.lbr.to_ip = bts_get(raw, bts_to);
660 }
661}
662
663static int ptrace_bts_read_record(struct task_struct *child, size_t index, 584static int ptrace_bts_read_record(struct task_struct *child, size_t index,
664 struct bts_struct __user *out) 585 struct bts_struct __user *out)
665{ 586{
666 struct bts_struct ret; 587 const struct bts_trace *trace;
667 const void *bts_record; 588 struct bts_struct bts;
668 size_t bts_index, bts_end; 589 const unsigned char *at;
669 int error; 590 int error;
670 591
671 error = ds_get_bts_end(child, &bts_end); 592 trace = ds_read_bts(child->bts);
672 if (error < 0) 593 if (!trace)
673 return error; 594 return -EPERM;
674
675 if (bts_end <= index)
676 return -EINVAL;
677 595
678 error = ds_get_bts_index(child, &bts_index); 596 at = trace->ds.top - ((index + 1) * trace->ds.size);
679 if (error < 0) 597 if ((void *)at < trace->ds.begin)
680 return error; 598 at += (trace->ds.n * trace->ds.size);
681 599
682 /* translate the ptrace bts index into the ds bts index */ 600 if (!trace->read)
683 bts_index += bts_end - (index + 1); 601 return -EOPNOTSUPP;
684 if (bts_end <= bts_index)
685 bts_index -= bts_end;
686 602
687 error = ds_access_bts(child, bts_index, &bts_record); 603 error = trace->read(child->bts, at, &bts);
688 if (error < 0) 604 if (error < 0)
689 return error; 605 return error;
690 606
691 ptrace_bts_translate_record(&ret, bts_record); 607 if (copy_to_user(out, &bts, sizeof(bts)))
692
693 if (copy_to_user(out, &ret, sizeof(ret)))
694 return -EFAULT; 608 return -EFAULT;
695 609
696 return sizeof(ret); 610 return sizeof(bts);
697} 611}
698 612
699static int ptrace_bts_drain(struct task_struct *child, 613static int ptrace_bts_drain(struct task_struct *child,
700 long size, 614 long size,
701 struct bts_struct __user *out) 615 struct bts_struct __user *out)
702{ 616{
703 struct bts_struct ret; 617 const struct bts_trace *trace;
704 const unsigned char *raw; 618 const unsigned char *at;
705 size_t end, i; 619 int error, drained = 0;
706 int error;
707 620
708 error = ds_get_bts_index(child, &end); 621 trace = ds_read_bts(child->bts);
709 if (error < 0) 622 if (!trace)
710 return error; 623 return -EPERM;
711 624
712 if (size < (end * sizeof(struct bts_struct))) 625 if (!trace->read)
626 return -EOPNOTSUPP;
627
628 if (size < (trace->ds.top - trace->ds.begin))
713 return -EIO; 629 return -EIO;
714 630
715 error = ds_access_bts(child, 0, (const void **)&raw); 631 for (at = trace->ds.begin; (void *)at < trace->ds.top;
716 if (error < 0) 632 out++, drained++, at += trace->ds.size) {
717 return error; 633 struct bts_struct bts;
634 int error;
718 635
719 for (i = 0; i < end; i++, out++, raw += bts_cfg.sizeof_bts) { 636 error = trace->read(child->bts, at, &bts);
720 ptrace_bts_translate_record(&ret, raw); 637 if (error < 0)
638 return error;
721 639
722 if (copy_to_user(out, &ret, sizeof(ret))) 640 if (copy_to_user(out, &bts, sizeof(bts)))
723 return -EFAULT; 641 return -EFAULT;
724 } 642 }
725 643
726 error = ds_clear_bts(child); 644 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
645
646 error = ds_reset_bts(child->bts);
727 if (error < 0) 647 if (error < 0)
728 return error; 648 return error;
729 649
730 return end; 650 return drained;
731} 651}
732 652
733static void ptrace_bts_ovfl(struct task_struct *child) 653static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
734{ 654{
735 send_sig(child->thread.bts_ovfl_signal, child, 0); 655 child->bts_buffer = alloc_locked_buffer(size);
656 if (!child->bts_buffer)
657 return -ENOMEM;
658
659 child->bts_size = size;
660
661 return 0;
662}
663
664static void ptrace_bts_free_buffer(struct task_struct *child)
665{
666 free_locked_buffer(child->bts_buffer, child->bts_size);
667 child->bts_buffer = NULL;
668 child->bts_size = 0;
736} 669}
737 670
738static int ptrace_bts_config(struct task_struct *child, 671static int ptrace_bts_config(struct task_struct *child,
@@ -740,114 +673,86 @@ static int ptrace_bts_config(struct task_struct *child,
740 const struct ptrace_bts_config __user *ucfg) 673 const struct ptrace_bts_config __user *ucfg)
741{ 674{
742 struct ptrace_bts_config cfg; 675 struct ptrace_bts_config cfg;
743 int error = 0; 676 unsigned int flags = 0;
744
745 error = -EOPNOTSUPP;
746 if (!bts_cfg.sizeof_bts)
747 goto errout;
748 677
749 error = -EIO;
750 if (cfg_size < sizeof(cfg)) 678 if (cfg_size < sizeof(cfg))
751 goto errout; 679 return -EIO;
752 680
753 error = -EFAULT;
754 if (copy_from_user(&cfg, ucfg, sizeof(cfg))) 681 if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
755 goto errout; 682 return -EFAULT;
756 683
757 error = -EINVAL; 684 if (child->bts) {
758 if ((cfg.flags & PTRACE_BTS_O_SIGNAL) && 685 ds_release_bts(child->bts);
759 !(cfg.flags & PTRACE_BTS_O_ALLOC)) 686 child->bts = NULL;
760 goto errout; 687 }
761 688
762 if (cfg.flags & PTRACE_BTS_O_ALLOC) { 689 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
763 ds_ovfl_callback_t ovfl = NULL; 690 if (!cfg.signal)
764 unsigned int sig = 0; 691 return -EINVAL;
765 692
766 /* we ignore the error in case we were not tracing child */ 693 return -EOPNOTSUPP;
767 (void)ds_release_bts(child);
768 694
769 if (cfg.flags & PTRACE_BTS_O_SIGNAL) { 695 child->thread.bts_ovfl_signal = cfg.signal;
770 if (!cfg.signal) 696 }
771 goto errout;
772 697
773 sig = cfg.signal; 698 if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
774 ovfl = ptrace_bts_ovfl; 699 (cfg.size != child->bts_size)) {
775 } 700 int error;
776 701
777 error = ds_request_bts(child, /* base = */ NULL, cfg.size, ovfl); 702 ptrace_bts_free_buffer(child);
778 if (error < 0)
779 goto errout;
780 703
781 child->thread.bts_ovfl_signal = sig; 704 error = ptrace_bts_allocate_buffer(child, cfg.size);
705 if (error < 0)
706 return error;
782 } 707 }
783 708
784 error = -EINVAL;
785 if (!child->thread.ds_ctx && cfg.flags)
786 goto errout;
787
788 if (cfg.flags & PTRACE_BTS_O_TRACE) 709 if (cfg.flags & PTRACE_BTS_O_TRACE)
789 child->thread.debugctlmsr |= bts_cfg.debugctl_mask; 710 flags |= BTS_USER;
790 else
791 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
792 711
793 if (cfg.flags & PTRACE_BTS_O_SCHED) 712 if (cfg.flags & PTRACE_BTS_O_SCHED)
794 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS); 713 flags |= BTS_TIMESTAMPS;
795 else
796 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
797 714
798 error = sizeof(cfg); 715 child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
716 /* ovfl = */ NULL, /* th = */ (size_t)-1,
717 flags);
718 if (IS_ERR(child->bts)) {
719 int error = PTR_ERR(child->bts);
799 720
800out: 721 ptrace_bts_free_buffer(child);
801 if (child->thread.debugctlmsr) 722 child->bts = NULL;
802 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
803 else
804 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
805 723
806 return error; 724 return error;
725 }
807 726
808errout: 727 return sizeof(cfg);
809 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
810 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
811 goto out;
812} 728}
813 729
814static int ptrace_bts_status(struct task_struct *child, 730static int ptrace_bts_status(struct task_struct *child,
815 long cfg_size, 731 long cfg_size,
816 struct ptrace_bts_config __user *ucfg) 732 struct ptrace_bts_config __user *ucfg)
817{ 733{
734 const struct bts_trace *trace;
818 struct ptrace_bts_config cfg; 735 struct ptrace_bts_config cfg;
819 size_t end;
820 const void *base, *max;
821 int error;
822 736
823 if (cfg_size < sizeof(cfg)) 737 if (cfg_size < sizeof(cfg))
824 return -EIO; 738 return -EIO;
825 739
826 error = ds_get_bts_end(child, &end); 740 trace = ds_read_bts(child->bts);
827 if (error < 0) 741 if (!trace)
828 return error; 742 return -EPERM;
829
830 error = ds_access_bts(child, /* index = */ 0, &base);
831 if (error < 0)
832 return error;
833
834 error = ds_access_bts(child, /* index = */ end, &max);
835 if (error < 0)
836 return error;
837 743
838 memset(&cfg, 0, sizeof(cfg)); 744 memset(&cfg, 0, sizeof(cfg));
839 cfg.size = (max - base); 745 cfg.size = trace->ds.end - trace->ds.begin;
840 cfg.signal = child->thread.bts_ovfl_signal; 746 cfg.signal = child->thread.bts_ovfl_signal;
841 cfg.bts_size = sizeof(struct bts_struct); 747 cfg.bts_size = sizeof(struct bts_struct);
842 748
843 if (cfg.signal) 749 if (cfg.signal)
844 cfg.flags |= PTRACE_BTS_O_SIGNAL; 750 cfg.flags |= PTRACE_BTS_O_SIGNAL;
845 751
846 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) && 752 if (trace->ds.flags & BTS_USER)
847 child->thread.debugctlmsr & bts_cfg.debugctl_mask)
848 cfg.flags |= PTRACE_BTS_O_TRACE; 753 cfg.flags |= PTRACE_BTS_O_TRACE;
849 754
850 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS)) 755 if (trace->ds.flags & BTS_TIMESTAMPS)
851 cfg.flags |= PTRACE_BTS_O_SCHED; 756 cfg.flags |= PTRACE_BTS_O_SCHED;
852 757
853 if (copy_to_user(ucfg, &cfg, sizeof(cfg))) 758 if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
@@ -856,110 +761,77 @@ static int ptrace_bts_status(struct task_struct *child,
856 return sizeof(cfg); 761 return sizeof(cfg);
857} 762}
858 763
859static int ptrace_bts_write_record(struct task_struct *child, 764static int ptrace_bts_clear(struct task_struct *child)
860 const struct bts_struct *in)
861{ 765{
862 unsigned char bts_record[BTS_MAX_RECORD_SIZE]; 766 const struct bts_trace *trace;
863 767
864 BUG_ON(BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts); 768 trace = ds_read_bts(child->bts);
769 if (!trace)
770 return -EPERM;
865 771
866 memset(bts_record, 0, bts_cfg.sizeof_bts); 772 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
867 switch (in->qualifier) {
868 case BTS_INVALID:
869 break;
870 773
871 case BTS_BRANCH: 774 return ds_reset_bts(child->bts);
872 bts_set(bts_record, bts_from, in->variant.lbr.from_ip); 775}
873 bts_set(bts_record, bts_to, in->variant.lbr.to_ip);
874 break;
875 776
876 case BTS_TASK_ARRIVES: 777static int ptrace_bts_size(struct task_struct *child)
877 case BTS_TASK_DEPARTS: 778{
878 bts_set(bts_record, bts_from, bts_escape); 779 const struct bts_trace *trace;
879 bts_set(bts_record, bts_qual, in->qualifier);
880 bts_set(bts_record, bts_jiffies, in->variant.jiffies);
881 break;
882 780
883 default: 781 trace = ds_read_bts(child->bts);
884 return -EINVAL; 782 if (!trace)
885 } 783 return -EPERM;
886 784
887 /* The writing task will be the switched-to task on a context 785 return (trace->ds.top - trace->ds.begin) / trace->ds.size;
888 * switch. It needs to write into the switched-from task's BTS
889 * buffer. */
890 return ds_unchecked_write_bts(child, bts_record, bts_cfg.sizeof_bts);
891} 786}
892 787
893void ptrace_bts_take_timestamp(struct task_struct *tsk, 788static void ptrace_bts_fork(struct task_struct *tsk)
894 enum bts_qualifier qualifier)
895{ 789{
896 struct bts_struct rec = { 790 tsk->bts = NULL;
897 .qualifier = qualifier, 791 tsk->bts_buffer = NULL;
898 .variant.jiffies = jiffies_64 792 tsk->bts_size = 0;
899 }; 793 tsk->thread.bts_ovfl_signal = 0;
900
901 ptrace_bts_write_record(tsk, &rec);
902} 794}
903 795
904static const struct bts_configuration bts_cfg_netburst = { 796static void ptrace_bts_untrace(struct task_struct *child)
905 .sizeof_bts = sizeof(long) * 3, 797{
906 .sizeof_field = sizeof(long), 798 if (unlikely(child->bts)) {
907 .debugctl_mask = (1<<2)|(1<<3)|(1<<5) 799 ds_release_bts(child->bts);
908}; 800 child->bts = NULL;
801
802 /* We cannot update total_vm and locked_vm since
803 child's mm is already gone. But we can reclaim the
804 memory. */
805 kfree(child->bts_buffer);
806 child->bts_buffer = NULL;
807 child->bts_size = 0;
808 }
809}
909 810
910static const struct bts_configuration bts_cfg_pentium_m = { 811static void ptrace_bts_detach(struct task_struct *child)
911 .sizeof_bts = sizeof(long) * 3, 812{
912 .sizeof_field = sizeof(long), 813 if (unlikely(child->bts)) {
913 .debugctl_mask = (1<<6)|(1<<7) 814 ds_release_bts(child->bts);
914}; 815 child->bts = NULL;
915 816
916static const struct bts_configuration bts_cfg_core2 = { 817 ptrace_bts_free_buffer(child);
917 .sizeof_bts = 8 * 3, 818 }
918 .sizeof_field = 8, 819}
919 .debugctl_mask = (1<<6)|(1<<7)|(1<<9) 820#else
920}; 821static inline void ptrace_bts_fork(struct task_struct *tsk) {}
822static inline void ptrace_bts_detach(struct task_struct *child) {}
823static inline void ptrace_bts_untrace(struct task_struct *child) {}
824#endif /* CONFIG_X86_PTRACE_BTS */
921 825
922static inline void bts_configure(const struct bts_configuration *cfg) 826void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
923{ 827{
924 bts_cfg = *cfg; 828 ptrace_bts_fork(child);
925} 829}
926 830
927void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c) 831void x86_ptrace_untrace(struct task_struct *child)
928{ 832{
929 switch (c->x86) { 833 ptrace_bts_untrace(child);
930 case 0x6:
931 switch (c->x86_model) {
932 case 0xD:
933 case 0xE: /* Pentium M */
934 bts_configure(&bts_cfg_pentium_m);
935 break;
936 case 0xF: /* Core2 */
937 case 0x1C: /* Atom */
938 bts_configure(&bts_cfg_core2);
939 break;
940 default:
941 /* sorry, don't know about them */
942 break;
943 }
944 break;
945 case 0xF:
946 switch (c->x86_model) {
947 case 0x0:
948 case 0x1:
949 case 0x2: /* Netburst */
950 bts_configure(&bts_cfg_netburst);
951 break;
952 default:
953 /* sorry, don't know about them */
954 break;
955 }
956 break;
957 default:
958 /* sorry, don't know about them */
959 break;
960 }
961} 834}
962#endif /* CONFIG_X86_PTRACE_BTS */
963 835
964/* 836/*
965 * Called by kernel/ptrace.c when detaching.. 837 * Called by kernel/ptrace.c when detaching..
@@ -972,15 +844,7 @@ void ptrace_disable(struct task_struct *child)
972#ifdef TIF_SYSCALL_EMU 844#ifdef TIF_SYSCALL_EMU
973 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU); 845 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
974#endif 846#endif
975#ifdef CONFIG_X86_PTRACE_BTS 847 ptrace_bts_detach(child);
976 (void)ds_release_bts(child);
977
978 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
979 if (!child->thread.debugctlmsr)
980 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
981
982 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
983#endif /* CONFIG_X86_PTRACE_BTS */
984} 848}
985 849
986#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION 850#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
@@ -1112,7 +976,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1112 break; 976 break;
1113 977
1114 case PTRACE_BTS_SIZE: 978 case PTRACE_BTS_SIZE:
1115 ret = ds_get_bts_index(child, /* pos = */ NULL); 979 ret = ptrace_bts_size(child);
1116 break; 980 break;
1117 981
1118 case PTRACE_BTS_GET: 982 case PTRACE_BTS_GET:
@@ -1121,7 +985,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1121 break; 985 break;
1122 986
1123 case PTRACE_BTS_CLEAR: 987 case PTRACE_BTS_CLEAR:
1124 ret = ds_clear_bts(child); 988 ret = ptrace_bts_clear(child);
1125 break; 989 break;
1126 990
1127 case PTRACE_BTS_DRAIN: 991 case PTRACE_BTS_DRAIN:
@@ -1384,6 +1248,14 @@ long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1384 1248
1385 case PTRACE_GET_THREAD_AREA: 1249 case PTRACE_GET_THREAD_AREA:
1386 case PTRACE_SET_THREAD_AREA: 1250 case PTRACE_SET_THREAD_AREA:
1251#ifdef CONFIG_X86_PTRACE_BTS
1252 case PTRACE_BTS_CONFIG:
1253 case PTRACE_BTS_STATUS:
1254 case PTRACE_BTS_SIZE:
1255 case PTRACE_BTS_GET:
1256 case PTRACE_BTS_CLEAR:
1257 case PTRACE_BTS_DRAIN:
1258#endif /* CONFIG_X86_PTRACE_BTS */
1387 return arch_ptrace(child, request, addr, data); 1259 return arch_ptrace(child, request, addr, data);
1388 1260
1389 default: 1261 default:
diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c
index cc5a2545dd41..61f718df6eec 100644
--- a/arch/x86/kernel/reboot.c
+++ b/arch/x86/kernel/reboot.c
@@ -21,6 +21,9 @@
21# include <asm/iommu.h> 21# include <asm/iommu.h>
22#endif 22#endif
23 23
24#include <mach_ipi.h>
25
26
24/* 27/*
25 * Power off function, if any 28 * Power off function, if any
26 */ 29 */
@@ -36,7 +39,10 @@ int reboot_force;
36static int reboot_cpu = -1; 39static int reboot_cpu = -1;
37#endif 40#endif
38 41
39/* reboot=b[ios] | s[mp] | t[riple] | k[bd] | e[fi] [, [w]arm | [c]old] 42/* This is set by the PCI code if either type 1 or type 2 PCI is detected */
43bool port_cf9_safe = false;
44
45/* reboot=b[ios] | s[mp] | t[riple] | k[bd] | e[fi] [, [w]arm | [c]old] | p[ci]
40 warm Don't set the cold reboot flag 46 warm Don't set the cold reboot flag
41 cold Set the cold reboot flag 47 cold Set the cold reboot flag
42 bios Reboot by jumping through the BIOS (only for X86_32) 48 bios Reboot by jumping through the BIOS (only for X86_32)
@@ -45,6 +51,7 @@ static int reboot_cpu = -1;
45 kbd Use the keyboard controller. cold reset (default) 51 kbd Use the keyboard controller. cold reset (default)
46 acpi Use the RESET_REG in the FADT 52 acpi Use the RESET_REG in the FADT
47 efi Use efi reset_system runtime service 53 efi Use efi reset_system runtime service
54 pci Use the so-called "PCI reset register", CF9
48 force Avoid anything that could hang. 55 force Avoid anything that could hang.
49 */ 56 */
50static int __init reboot_setup(char *str) 57static int __init reboot_setup(char *str)
@@ -79,6 +86,7 @@ static int __init reboot_setup(char *str)
79 case 'k': 86 case 'k':
80 case 't': 87 case 't':
81 case 'e': 88 case 'e':
89 case 'p':
82 reboot_type = *str; 90 reboot_type = *str;
83 break; 91 break;
84 92
@@ -404,12 +412,27 @@ static void native_machine_emergency_restart(void)
404 reboot_type = BOOT_KBD; 412 reboot_type = BOOT_KBD;
405 break; 413 break;
406 414
407
408 case BOOT_EFI: 415 case BOOT_EFI:
409 if (efi_enabled) 416 if (efi_enabled)
410 efi.reset_system(reboot_mode ? EFI_RESET_WARM : EFI_RESET_COLD, 417 efi.reset_system(reboot_mode ?
418 EFI_RESET_WARM :
419 EFI_RESET_COLD,
411 EFI_SUCCESS, 0, NULL); 420 EFI_SUCCESS, 0, NULL);
421 reboot_type = BOOT_KBD;
422 break;
412 423
424 case BOOT_CF9:
425 port_cf9_safe = true;
426 /* fall through */
427
428 case BOOT_CF9_COND:
429 if (port_cf9_safe) {
430 u8 cf9 = inb(0xcf9) & ~6;
431 outb(cf9|2, 0xcf9); /* Request hard reset */
432 udelay(50);
433 outb(cf9|6, 0xcf9); /* Actually do the reset */
434 udelay(50);
435 }
413 reboot_type = BOOT_KBD; 436 reboot_type = BOOT_KBD;
414 break; 437 break;
415 } 438 }
@@ -470,6 +493,11 @@ static void native_machine_restart(char *__unused)
470 493
471static void native_machine_halt(void) 494static void native_machine_halt(void)
472{ 495{
496 /* stop other cpus and apics */
497 machine_shutdown();
498
499 /* stop this cpu */
500 stop_this_cpu(NULL);
473} 501}
474 502
475static void native_machine_power_off(void) 503static void native_machine_power_off(void)
@@ -523,3 +551,95 @@ void machine_crash_shutdown(struct pt_regs *regs)
523 machine_ops.crash_shutdown(regs); 551 machine_ops.crash_shutdown(regs);
524} 552}
525#endif 553#endif
554
555
556#if defined(CONFIG_SMP)
557
558/* This keeps a track of which one is crashing cpu. */
559static int crashing_cpu;
560static nmi_shootdown_cb shootdown_callback;
561
562static atomic_t waiting_for_crash_ipi;
563
564static int crash_nmi_callback(struct notifier_block *self,
565 unsigned long val, void *data)
566{
567 int cpu;
568
569 if (val != DIE_NMI_IPI)
570 return NOTIFY_OK;
571
572 cpu = raw_smp_processor_id();
573
574 /* Don't do anything if this handler is invoked on crashing cpu.
575 * Otherwise, system will completely hang. Crashing cpu can get
576 * an NMI if system was initially booted with nmi_watchdog parameter.
577 */
578 if (cpu == crashing_cpu)
579 return NOTIFY_STOP;
580 local_irq_disable();
581
582 shootdown_callback(cpu, (struct die_args *)data);
583
584 atomic_dec(&waiting_for_crash_ipi);
585 /* Assume hlt works */
586 halt();
587 for (;;)
588 cpu_relax();
589
590 return 1;
591}
592
593static void smp_send_nmi_allbutself(void)
594{
595 cpumask_t mask = cpu_online_map;
596 cpu_clear(safe_smp_processor_id(), mask);
597 if (!cpus_empty(mask))
598 send_IPI_mask(mask, NMI_VECTOR);
599}
600
601static struct notifier_block crash_nmi_nb = {
602 .notifier_call = crash_nmi_callback,
603};
604
605/* Halt all other CPUs, calling the specified function on each of them
606 *
607 * This function can be used to halt all other CPUs on crash
608 * or emergency reboot time. The function passed as parameter
609 * will be called inside a NMI handler on all CPUs.
610 */
611void nmi_shootdown_cpus(nmi_shootdown_cb callback)
612{
613 unsigned long msecs;
614 local_irq_disable();
615
616 /* Make a note of crashing cpu. Will be used in NMI callback.*/
617 crashing_cpu = safe_smp_processor_id();
618
619 shootdown_callback = callback;
620
621 atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
622 /* Would it be better to replace the trap vector here? */
623 if (register_die_notifier(&crash_nmi_nb))
624 return; /* return what? */
625 /* Ensure the new callback function is set before sending
626 * out the NMI
627 */
628 wmb();
629
630 smp_send_nmi_allbutself();
631
632 msecs = 1000; /* Wait at most a second for the other cpus to stop */
633 while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
634 mdelay(1);
635 msecs--;
636 }
637
638 /* Leave the nmi callback set */
639}
640#else /* !CONFIG_SMP */
641void nmi_shootdown_cpus(nmi_shootdown_cb callback)
642{
643 /* No other CPUs to shoot down */
644}
645#endif
diff --git a/arch/x86/kernel/relocate_kernel_32.S b/arch/x86/kernel/relocate_kernel_32.S
index 6f50664b2ba5..a160f3119725 100644
--- a/arch/x86/kernel/relocate_kernel_32.S
+++ b/arch/x86/kernel/relocate_kernel_32.S
@@ -10,15 +10,12 @@
10#include <asm/page.h> 10#include <asm/page.h>
11#include <asm/kexec.h> 11#include <asm/kexec.h>
12#include <asm/processor-flags.h> 12#include <asm/processor-flags.h>
13#include <asm/pgtable.h>
14 13
15/* 14/*
16 * Must be relocatable PIC code callable as a C function 15 * Must be relocatable PIC code callable as a C function
17 */ 16 */
18 17
19#define PTR(x) (x << 2) 18#define PTR(x) (x << 2)
20#define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
21#define PAE_PGD_ATTR (_PAGE_PRESENT)
22 19
23/* control_page + KEXEC_CONTROL_CODE_MAX_SIZE 20/* control_page + KEXEC_CONTROL_CODE_MAX_SIZE
24 * ~ control_page + PAGE_SIZE are used as data storage and stack for 21 * ~ control_page + PAGE_SIZE are used as data storage and stack for
@@ -39,7 +36,6 @@
39#define CP_PA_BACKUP_PAGES_MAP DATA(0x1c) 36#define CP_PA_BACKUP_PAGES_MAP DATA(0x1c)
40 37
41 .text 38 .text
42 .align PAGE_SIZE
43 .globl relocate_kernel 39 .globl relocate_kernel
44relocate_kernel: 40relocate_kernel:
45 /* Save the CPU context, used for jumping back */ 41 /* Save the CPU context, used for jumping back */
@@ -60,117 +56,6 @@ relocate_kernel:
60 movl %cr4, %eax 56 movl %cr4, %eax
61 movl %eax, CR4(%edi) 57 movl %eax, CR4(%edi)
62 58
63#ifdef CONFIG_X86_PAE
64 /* map the control page at its virtual address */
65
66 movl PTR(VA_PGD)(%ebp), %edi
67 movl PTR(VA_CONTROL_PAGE)(%ebp), %eax
68 andl $0xc0000000, %eax
69 shrl $27, %eax
70 addl %edi, %eax
71
72 movl PTR(PA_PMD_0)(%ebp), %edx
73 orl $PAE_PGD_ATTR, %edx
74 movl %edx, (%eax)
75
76 movl PTR(VA_PMD_0)(%ebp), %edi
77 movl PTR(VA_CONTROL_PAGE)(%ebp), %eax
78 andl $0x3fe00000, %eax
79 shrl $18, %eax
80 addl %edi, %eax
81
82 movl PTR(PA_PTE_0)(%ebp), %edx
83 orl $PAGE_ATTR, %edx
84 movl %edx, (%eax)
85
86 movl PTR(VA_PTE_0)(%ebp), %edi
87 movl PTR(VA_CONTROL_PAGE)(%ebp), %eax
88 andl $0x001ff000, %eax
89 shrl $9, %eax
90 addl %edi, %eax
91
92 movl PTR(PA_CONTROL_PAGE)(%ebp), %edx
93 orl $PAGE_ATTR, %edx
94 movl %edx, (%eax)
95
96 /* identity map the control page at its physical address */
97
98 movl PTR(VA_PGD)(%ebp), %edi
99 movl PTR(PA_CONTROL_PAGE)(%ebp), %eax
100 andl $0xc0000000, %eax
101 shrl $27, %eax
102 addl %edi, %eax
103
104 movl PTR(PA_PMD_1)(%ebp), %edx
105 orl $PAE_PGD_ATTR, %edx
106 movl %edx, (%eax)
107
108 movl PTR(VA_PMD_1)(%ebp), %edi
109 movl PTR(PA_CONTROL_PAGE)(%ebp), %eax
110 andl $0x3fe00000, %eax
111 shrl $18, %eax
112 addl %edi, %eax
113
114 movl PTR(PA_PTE_1)(%ebp), %edx
115 orl $PAGE_ATTR, %edx
116 movl %edx, (%eax)
117
118 movl PTR(VA_PTE_1)(%ebp), %edi
119 movl PTR(PA_CONTROL_PAGE)(%ebp), %eax
120 andl $0x001ff000, %eax
121 shrl $9, %eax
122 addl %edi, %eax
123
124 movl PTR(PA_CONTROL_PAGE)(%ebp), %edx
125 orl $PAGE_ATTR, %edx
126 movl %edx, (%eax)
127#else
128 /* map the control page at its virtual address */
129
130 movl PTR(VA_PGD)(%ebp), %edi
131 movl PTR(VA_CONTROL_PAGE)(%ebp), %eax
132 andl $0xffc00000, %eax
133 shrl $20, %eax
134 addl %edi, %eax
135
136 movl PTR(PA_PTE_0)(%ebp), %edx
137 orl $PAGE_ATTR, %edx
138 movl %edx, (%eax)
139
140 movl PTR(VA_PTE_0)(%ebp), %edi
141 movl PTR(VA_CONTROL_PAGE)(%ebp), %eax
142 andl $0x003ff000, %eax
143 shrl $10, %eax
144 addl %edi, %eax
145
146 movl PTR(PA_CONTROL_PAGE)(%ebp), %edx
147 orl $PAGE_ATTR, %edx
148 movl %edx, (%eax)
149
150 /* identity map the control page at its physical address */
151
152 movl PTR(VA_PGD)(%ebp), %edi
153 movl PTR(PA_CONTROL_PAGE)(%ebp), %eax
154 andl $0xffc00000, %eax
155 shrl $20, %eax
156 addl %edi, %eax
157
158 movl PTR(PA_PTE_1)(%ebp), %edx
159 orl $PAGE_ATTR, %edx
160 movl %edx, (%eax)
161
162 movl PTR(VA_PTE_1)(%ebp), %edi
163 movl PTR(PA_CONTROL_PAGE)(%ebp), %eax
164 andl $0x003ff000, %eax
165 shrl $10, %eax
166 addl %edi, %eax
167
168 movl PTR(PA_CONTROL_PAGE)(%ebp), %edx
169 orl $PAGE_ATTR, %edx
170 movl %edx, (%eax)
171#endif
172
173relocate_new_kernel:
174 /* read the arguments and say goodbye to the stack */ 59 /* read the arguments and say goodbye to the stack */
175 movl 20+4(%esp), %ebx /* page_list */ 60 movl 20+4(%esp), %ebx /* page_list */
176 movl 20+8(%esp), %ebp /* list of pages */ 61 movl 20+8(%esp), %ebp /* list of pages */
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index bdec76e55594..08e02e8453c9 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -93,11 +93,13 @@
93#include <asm/desc.h> 93#include <asm/desc.h>
94#include <asm/dma.h> 94#include <asm/dma.h>
95#include <asm/iommu.h> 95#include <asm/iommu.h>
96#include <asm/gart.h>
96#include <asm/mmu_context.h> 97#include <asm/mmu_context.h>
97#include <asm/proto.h> 98#include <asm/proto.h>
98 99
99#include <mach_apic.h> 100#include <mach_apic.h>
100#include <asm/paravirt.h> 101#include <asm/paravirt.h>
102#include <asm/hypervisor.h>
101 103
102#include <asm/percpu.h> 104#include <asm/percpu.h>
103#include <asm/topology.h> 105#include <asm/topology.h>
@@ -448,6 +450,7 @@ static void __init reserve_early_setup_data(void)
448 * @size: Size of the crashkernel memory to reserve. 450 * @size: Size of the crashkernel memory to reserve.
449 * Returns the base address on success, and -1ULL on failure. 451 * Returns the base address on success, and -1ULL on failure.
450 */ 452 */
453static
451unsigned long long __init find_and_reserve_crashkernel(unsigned long long size) 454unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
452{ 455{
453 const unsigned long long alignment = 16<<20; /* 16M */ 456 const unsigned long long alignment = 16<<20; /* 16M */
@@ -583,161 +586,24 @@ static int __init setup_elfcorehdr(char *arg)
583early_param("elfcorehdr", setup_elfcorehdr); 586early_param("elfcorehdr", setup_elfcorehdr);
584#endif 587#endif
585 588
586static struct x86_quirks default_x86_quirks __initdata; 589static int __init default_update_genapic(void)
587
588struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
589
590/*
591 * Some BIOSes seem to corrupt the low 64k of memory during events
592 * like suspend/resume and unplugging an HDMI cable. Reserve all
593 * remaining free memory in that area and fill it with a distinct
594 * pattern.
595 */
596#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
597#define MAX_SCAN_AREAS 8
598
599static int __read_mostly memory_corruption_check = -1;
600
601static unsigned __read_mostly corruption_check_size = 64*1024;
602static unsigned __read_mostly corruption_check_period = 60; /* seconds */
603
604static struct e820entry scan_areas[MAX_SCAN_AREAS];
605static int num_scan_areas;
606
607
608static int set_corruption_check(char *arg)
609{
610 char *end;
611
612 memory_corruption_check = simple_strtol(arg, &end, 10);
613
614 return (*end == 0) ? 0 : -EINVAL;
615}
616early_param("memory_corruption_check", set_corruption_check);
617
618static int set_corruption_check_period(char *arg)
619{
620 char *end;
621
622 corruption_check_period = simple_strtoul(arg, &end, 10);
623
624 return (*end == 0) ? 0 : -EINVAL;
625}
626early_param("memory_corruption_check_period", set_corruption_check_period);
627
628static int set_corruption_check_size(char *arg)
629{ 590{
630 char *end; 591#ifdef CONFIG_X86_SMP
631 unsigned size; 592# if defined(CONFIG_X86_GENERICARCH) || defined(CONFIG_X86_64)
632 593 genapic->wakeup_cpu = wakeup_secondary_cpu_via_init;
633 size = memparse(arg, &end); 594# endif
634
635 if (*end == '\0')
636 corruption_check_size = size;
637
638 return (size == corruption_check_size) ? 0 : -EINVAL;
639}
640early_param("memory_corruption_check_size", set_corruption_check_size);
641
642
643static void __init setup_bios_corruption_check(void)
644{
645 u64 addr = PAGE_SIZE; /* assume first page is reserved anyway */
646
647 if (memory_corruption_check == -1) {
648 memory_corruption_check =
649#ifdef CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
650 1
651#else
652 0
653#endif 595#endif
654 ;
655 }
656
657 if (corruption_check_size == 0)
658 memory_corruption_check = 0;
659
660 if (!memory_corruption_check)
661 return;
662
663 corruption_check_size = round_up(corruption_check_size, PAGE_SIZE);
664 596
665 while(addr < corruption_check_size && num_scan_areas < MAX_SCAN_AREAS) { 597 return 0;
666 u64 size;
667 addr = find_e820_area_size(addr, &size, PAGE_SIZE);
668
669 if (addr == 0)
670 break;
671
672 if ((addr + size) > corruption_check_size)
673 size = corruption_check_size - addr;
674
675 if (size == 0)
676 break;
677
678 e820_update_range(addr, size, E820_RAM, E820_RESERVED);
679 scan_areas[num_scan_areas].addr = addr;
680 scan_areas[num_scan_areas].size = size;
681 num_scan_areas++;
682
683 /* Assume we've already mapped this early memory */
684 memset(__va(addr), 0, size);
685
686 addr += size;
687 }
688
689 printk(KERN_INFO "Scanning %d areas for low memory corruption\n",
690 num_scan_areas);
691 update_e820();
692}
693
694static struct timer_list periodic_check_timer;
695
696void check_for_bios_corruption(void)
697{
698 int i;
699 int corruption = 0;
700
701 if (!memory_corruption_check)
702 return;
703
704 for(i = 0; i < num_scan_areas; i++) {
705 unsigned long *addr = __va(scan_areas[i].addr);
706 unsigned long size = scan_areas[i].size;
707
708 for(; size; addr++, size -= sizeof(unsigned long)) {
709 if (!*addr)
710 continue;
711 printk(KERN_ERR "Corrupted low memory at %p (%lx phys) = %08lx\n",
712 addr, __pa(addr), *addr);
713 corruption = 1;
714 *addr = 0;
715 }
716 }
717
718 WARN(corruption, KERN_ERR "Memory corruption detected in low memory\n");
719}
720
721static void periodic_check_for_corruption(unsigned long data)
722{
723 check_for_bios_corruption();
724 mod_timer(&periodic_check_timer, round_jiffies(jiffies + corruption_check_period*HZ));
725} 598}
726 599
727void start_periodic_check_for_corruption(void) 600static struct x86_quirks default_x86_quirks __initdata = {
728{ 601 .update_genapic = default_update_genapic,
729 if (!memory_corruption_check || corruption_check_period == 0) 602};
730 return;
731
732 printk(KERN_INFO "Scanning for low memory corruption every %d seconds\n",
733 corruption_check_period);
734 603
735 init_timer(&periodic_check_timer); 604struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
736 periodic_check_timer.function = &periodic_check_for_corruption;
737 periodic_check_for_corruption(0);
738}
739#endif
740 605
606#ifdef CONFIG_X86_RESERVE_LOW_64K
741static int __init dmi_low_memory_corruption(const struct dmi_system_id *d) 607static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
742{ 608{
743 printk(KERN_NOTICE 609 printk(KERN_NOTICE
@@ -749,6 +615,7 @@ static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
749 615
750 return 0; 616 return 0;
751} 617}
618#endif
752 619
753/* List of systems that have known low memory corruption BIOS problems */ 620/* List of systems that have known low memory corruption BIOS problems */
754static struct dmi_system_id __initdata bad_bios_dmi_table[] = { 621static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
@@ -907,6 +774,12 @@ void __init setup_arch(char **cmdline_p)
907 774
908 dmi_check_system(bad_bios_dmi_table); 775 dmi_check_system(bad_bios_dmi_table);
909 776
777 /*
778 * VMware detection requires dmi to be available, so this
779 * needs to be done after dmi_scan_machine, for the BP.
780 */
781 init_hypervisor(&boot_cpu_data);
782
910#ifdef CONFIG_X86_32 783#ifdef CONFIG_X86_32
911 probe_roms(); 784 probe_roms();
912#endif 785#endif
diff --git a/arch/x86/kernel/sigframe.h b/arch/x86/kernel/sigframe.h
deleted file mode 100644
index cc673aa55ce4..000000000000
--- a/arch/x86/kernel/sigframe.h
+++ /dev/null
@@ -1,42 +0,0 @@
1#ifdef CONFIG_X86_32
2struct sigframe {
3 char __user *pretcode;
4 int sig;
5 struct sigcontext sc;
6 /*
7 * fpstate is unused. fpstate is moved/allocated after
8 * retcode[] below. This movement allows to have the FP state and the
9 * future state extensions (xsave) stay together.
10 * And at the same time retaining the unused fpstate, prevents changing
11 * the offset of extramask[] in the sigframe and thus prevent any
12 * legacy application accessing/modifying it.
13 */
14 struct _fpstate fpstate_unused;
15 unsigned long extramask[_NSIG_WORDS-1];
16 char retcode[8];
17 /* fp state follows here */
18};
19
20struct rt_sigframe {
21 char __user *pretcode;
22 int sig;
23 struct siginfo __user *pinfo;
24 void __user *puc;
25 struct siginfo info;
26 struct ucontext uc;
27 char retcode[8];
28 /* fp state follows here */
29};
30#else
31struct rt_sigframe {
32 char __user *pretcode;
33 struct ucontext uc;
34 struct siginfo info;
35 /* fp state follows here */
36};
37
38int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
39 sigset_t *set, struct pt_regs *regs);
40int ia32_setup_frame(int sig, struct k_sigaction *ka,
41 sigset_t *set, struct pt_regs *regs);
42#endif
diff --git a/arch/x86/kernel/signal_32.c b/arch/x86/kernel/signal.c
index d6dd057d0f22..89bb7668041d 100644
--- a/arch/x86/kernel/signal_32.c
+++ b/arch/x86/kernel/signal.c
@@ -1,36 +1,41 @@
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 * 1997-11-28 Modified for POSIX.1b signals by Richard Henderson 5 * 1997-11-28 Modified for POSIX.1b signals by Richard Henderson
5 * 2000-06-20 Pentium III FXSR, SSE support by Gareth Hughes 6 * 2000-06-20 Pentium III FXSR, SSE support by Gareth Hughes
7 * 2000-2002 x86-64 support by Andi Kleen
6 */ 8 */
7#include <linux/list.h>
8 9
9#include <linux/personality.h> 10#include <linux/sched.h>
10#include <linux/binfmts.h> 11#include <linux/mm.h>
11#include <linux/suspend.h> 12#include <linux/smp.h>
12#include <linux/kernel.h> 13#include <linux/kernel.h>
13#include <linux/ptrace.h>
14#include <linux/signal.h> 14#include <linux/signal.h>
15#include <linux/stddef.h>
16#include <linux/unistd.h>
17#include <linux/errno.h> 15#include <linux/errno.h>
18#include <linux/sched.h>
19#include <linux/wait.h> 16#include <linux/wait.h>
17#include <linux/ptrace.h>
20#include <linux/tracehook.h> 18#include <linux/tracehook.h>
21#include <linux/elf.h> 19#include <linux/unistd.h>
22#include <linux/smp.h> 20#include <linux/stddef.h>
23#include <linux/mm.h> 21#include <linux/personality.h>
22#include <linux/uaccess.h>
24 23
25#include <asm/processor.h> 24#include <asm/processor.h>
26#include <asm/ucontext.h> 25#include <asm/ucontext.h>
27#include <asm/uaccess.h>
28#include <asm/i387.h> 26#include <asm/i387.h>
29#include <asm/vdso.h> 27#include <asm/vdso.h>
28
29#ifdef CONFIG_X86_64
30#include <asm/proto.h>
31#include <asm/ia32_unistd.h>
32#include <asm/mce.h>
33#endif /* CONFIG_X86_64 */
34
30#include <asm/syscall.h> 35#include <asm/syscall.h>
31#include <asm/syscalls.h> 36#include <asm/syscalls.h>
32 37
33#include "sigframe.h" 38#include <asm/sigframe.h>
34 39
35#define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP))) 40#define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
36 41
@@ -45,74 +50,6 @@
45# define FIX_EFLAGS __FIX_EFLAGS 50# define FIX_EFLAGS __FIX_EFLAGS
46#endif 51#endif
47 52
48/*
49 * Atomically swap in the new signal mask, and wait for a signal.
50 */
51asmlinkage int
52sys_sigsuspend(int history0, int history1, old_sigset_t mask)
53{
54 mask &= _BLOCKABLE;
55 spin_lock_irq(&current->sighand->siglock);
56 current->saved_sigmask = current->blocked;
57 siginitset(&current->blocked, mask);
58 recalc_sigpending();
59 spin_unlock_irq(&current->sighand->siglock);
60
61 current->state = TASK_INTERRUPTIBLE;
62 schedule();
63 set_restore_sigmask();
64
65 return -ERESTARTNOHAND;
66}
67
68asmlinkage int
69sys_sigaction(int sig, const struct old_sigaction __user *act,
70 struct old_sigaction __user *oact)
71{
72 struct k_sigaction new_ka, old_ka;
73 int ret;
74
75 if (act) {
76 old_sigset_t mask;
77
78 if (!access_ok(VERIFY_READ, act, sizeof(*act)) ||
79 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
80 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer))
81 return -EFAULT;
82
83 __get_user(new_ka.sa.sa_flags, &act->sa_flags);
84 __get_user(mask, &act->sa_mask);
85 siginitset(&new_ka.sa.sa_mask, mask);
86 }
87
88 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
89
90 if (!ret && oact) {
91 if (!access_ok(VERIFY_WRITE, oact, sizeof(*oact)) ||
92 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
93 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer))
94 return -EFAULT;
95
96 __put_user(old_ka.sa.sa_flags, &oact->sa_flags);
97 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask);
98 }
99
100 return ret;
101}
102
103asmlinkage int sys_sigaltstack(unsigned long bx)
104{
105 /*
106 * This is needed to make gcc realize it doesn't own the
107 * "struct pt_regs"
108 */
109 struct pt_regs *regs = (struct pt_regs *)&bx;
110 const stack_t __user *uss = (const stack_t __user *)bx;
111 stack_t __user *uoss = (stack_t __user *)regs->cx;
112
113 return do_sigaltstack(uss, uoss, regs->sp);
114}
115
116#define COPY(x) { \ 53#define COPY(x) { \
117 err |= __get_user(regs->x, &sc->x); \ 54 err |= __get_user(regs->x, &sc->x); \
118} 55}
@@ -123,7 +60,7 @@ asmlinkage int sys_sigaltstack(unsigned long bx)
123 regs->seg = tmp; \ 60 regs->seg = tmp; \
124} 61}
125 62
126#define COPY_SEG_STRICT(seg) { \ 63#define COPY_SEG_CPL3(seg) { \
127 unsigned short tmp; \ 64 unsigned short tmp; \
128 err |= __get_user(tmp, &sc->seg); \ 65 err |= __get_user(tmp, &sc->seg); \
129 regs->seg = tmp | 3; \ 66 regs->seg = tmp | 3; \
@@ -135,9 +72,6 @@ asmlinkage int sys_sigaltstack(unsigned long bx)
135 loadsegment(seg, tmp); \ 72 loadsegment(seg, tmp); \
136} 73}
137 74
138/*
139 * Do a signal return; undo the signal stack.
140 */
141static int 75static int
142restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc, 76restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
143 unsigned long *pax) 77 unsigned long *pax)
@@ -149,14 +83,36 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
149 /* Always make any pending restarted system calls return -EINTR */ 83 /* Always make any pending restarted system calls return -EINTR */
150 current_thread_info()->restart_block.fn = do_no_restart_syscall; 84 current_thread_info()->restart_block.fn = do_no_restart_syscall;
151 85
86#ifdef CONFIG_X86_32
152 GET_SEG(gs); 87 GET_SEG(gs);
153 COPY_SEG(fs); 88 COPY_SEG(fs);
154 COPY_SEG(es); 89 COPY_SEG(es);
155 COPY_SEG(ds); 90 COPY_SEG(ds);
91#endif /* CONFIG_X86_32 */
92
156 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx); 93 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx);
157 COPY(dx); COPY(cx); COPY(ip); 94 COPY(dx); COPY(cx); COPY(ip);
158 COPY_SEG_STRICT(cs); 95
159 COPY_SEG_STRICT(ss); 96#ifdef CONFIG_X86_64
97 COPY(r8);
98 COPY(r9);
99 COPY(r10);
100 COPY(r11);
101 COPY(r12);
102 COPY(r13);
103 COPY(r14);
104 COPY(r15);
105#endif /* CONFIG_X86_64 */
106
107#ifdef CONFIG_X86_32
108 COPY_SEG_CPL3(cs);
109 COPY_SEG_CPL3(ss);
110#else /* !CONFIG_X86_32 */
111 /* Kernel saves and restores only the CS segment register on signals,
112 * which is the bare minimum needed to allow mixed 32/64-bit code.
113 * App's signal handler can save/restore other segments if needed. */
114 COPY_SEG_CPL3(cs);
115#endif /* CONFIG_X86_32 */
160 116
161 err |= __get_user(tmpflags, &sc->flags); 117 err |= __get_user(tmpflags, &sc->flags);
162 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS); 118 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
@@ -169,102 +125,24 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
169 return err; 125 return err;
170} 126}
171 127
172asmlinkage unsigned long sys_sigreturn(unsigned long __unused)
173{
174 struct sigframe __user *frame;
175 struct pt_regs *regs;
176 unsigned long ax;
177 sigset_t set;
178
179 regs = (struct pt_regs *) &__unused;
180 frame = (struct sigframe __user *)(regs->sp - 8);
181
182 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
183 goto badframe;
184 if (__get_user(set.sig[0], &frame->sc.oldmask) || (_NSIG_WORDS > 1
185 && __copy_from_user(&set.sig[1], &frame->extramask,
186 sizeof(frame->extramask))))
187 goto badframe;
188
189 sigdelsetmask(&set, ~_BLOCKABLE);
190 spin_lock_irq(&current->sighand->siglock);
191 current->blocked = set;
192 recalc_sigpending();
193 spin_unlock_irq(&current->sighand->siglock);
194
195 if (restore_sigcontext(regs, &frame->sc, &ax))
196 goto badframe;
197 return ax;
198
199badframe:
200 if (show_unhandled_signals && printk_ratelimit()) {
201 printk("%s%s[%d] bad frame in sigreturn frame:"
202 "%p ip:%lx sp:%lx oeax:%lx",
203 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
204 current->comm, task_pid_nr(current), frame, regs->ip,
205 regs->sp, regs->orig_ax);
206 print_vma_addr(" in ", regs->ip);
207 printk(KERN_CONT "\n");
208 }
209
210 force_sig(SIGSEGV, current);
211
212 return 0;
213}
214
215static long do_rt_sigreturn(struct pt_regs *regs)
216{
217 struct rt_sigframe __user *frame;
218 unsigned long ax;
219 sigset_t set;
220
221 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long));
222 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
223 goto badframe;
224 if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
225 goto badframe;
226
227 sigdelsetmask(&set, ~_BLOCKABLE);
228 spin_lock_irq(&current->sighand->siglock);
229 current->blocked = set;
230 recalc_sigpending();
231 spin_unlock_irq(&current->sighand->siglock);
232
233 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &ax))
234 goto badframe;
235
236 if (do_sigaltstack(&frame->uc.uc_stack, NULL, regs->sp) == -EFAULT)
237 goto badframe;
238
239 return ax;
240
241badframe:
242 signal_fault(regs, frame, "rt_sigreturn");
243 return 0;
244}
245
246asmlinkage int sys_rt_sigreturn(unsigned long __unused)
247{
248 struct pt_regs *regs = (struct pt_regs *)&__unused;
249
250 return do_rt_sigreturn(regs);
251}
252
253/*
254 * Set up a signal frame.
255 */
256static int 128static int
257setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate, 129setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
258 struct pt_regs *regs, unsigned long mask) 130 struct pt_regs *regs, unsigned long mask)
259{ 131{
260 int tmp, err = 0; 132 int err = 0;
261 133
262 err |= __put_user(regs->fs, (unsigned int __user *)&sc->fs); 134#ifdef CONFIG_X86_32
263 savesegment(gs, tmp); 135 {
264 err |= __put_user(tmp, (unsigned int __user *)&sc->gs); 136 unsigned int tmp;
265 137
138 savesegment(gs, tmp);
139 err |= __put_user(tmp, (unsigned int __user *)&sc->gs);
140 }
141 err |= __put_user(regs->fs, (unsigned int __user *)&sc->fs);
266 err |= __put_user(regs->es, (unsigned int __user *)&sc->es); 142 err |= __put_user(regs->es, (unsigned int __user *)&sc->es);
267 err |= __put_user(regs->ds, (unsigned int __user *)&sc->ds); 143 err |= __put_user(regs->ds, (unsigned int __user *)&sc->ds);
144#endif /* CONFIG_X86_32 */
145
268 err |= __put_user(regs->di, &sc->di); 146 err |= __put_user(regs->di, &sc->di);
269 err |= __put_user(regs->si, &sc->si); 147 err |= __put_user(regs->si, &sc->si);
270 err |= __put_user(regs->bp, &sc->bp); 148 err |= __put_user(regs->bp, &sc->bp);
@@ -273,19 +151,33 @@ setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
273 err |= __put_user(regs->dx, &sc->dx); 151 err |= __put_user(regs->dx, &sc->dx);
274 err |= __put_user(regs->cx, &sc->cx); 152 err |= __put_user(regs->cx, &sc->cx);
275 err |= __put_user(regs->ax, &sc->ax); 153 err |= __put_user(regs->ax, &sc->ax);
154#ifdef CONFIG_X86_64
155 err |= __put_user(regs->r8, &sc->r8);
156 err |= __put_user(regs->r9, &sc->r9);
157 err |= __put_user(regs->r10, &sc->r10);
158 err |= __put_user(regs->r11, &sc->r11);
159 err |= __put_user(regs->r12, &sc->r12);
160 err |= __put_user(regs->r13, &sc->r13);
161 err |= __put_user(regs->r14, &sc->r14);
162 err |= __put_user(regs->r15, &sc->r15);
163#endif /* CONFIG_X86_64 */
164
276 err |= __put_user(current->thread.trap_no, &sc->trapno); 165 err |= __put_user(current->thread.trap_no, &sc->trapno);
277 err |= __put_user(current->thread.error_code, &sc->err); 166 err |= __put_user(current->thread.error_code, &sc->err);
278 err |= __put_user(regs->ip, &sc->ip); 167 err |= __put_user(regs->ip, &sc->ip);
168#ifdef CONFIG_X86_32
279 err |= __put_user(regs->cs, (unsigned int __user *)&sc->cs); 169 err |= __put_user(regs->cs, (unsigned int __user *)&sc->cs);
280 err |= __put_user(regs->flags, &sc->flags); 170 err |= __put_user(regs->flags, &sc->flags);
281 err |= __put_user(regs->sp, &sc->sp_at_signal); 171 err |= __put_user(regs->sp, &sc->sp_at_signal);
282 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss); 172 err |= __put_user(regs->ss, (unsigned int __user *)&sc->ss);
173#else /* !CONFIG_X86_32 */
174 err |= __put_user(regs->flags, &sc->flags);
175 err |= __put_user(regs->cs, &sc->cs);
176 err |= __put_user(0, &sc->gs);
177 err |= __put_user(0, &sc->fs);
178#endif /* CONFIG_X86_32 */
283 179
284 tmp = save_i387_xstate(fpstate); 180 err |= __put_user(fpstate, &sc->fpstate);
285 if (tmp < 0)
286 err = 1;
287 else
288 err |= __put_user(tmp ? fpstate : NULL, &sc->fpstate);
289 181
290 /* non-iBCS2 extensions.. */ 182 /* non-iBCS2 extensions.. */
291 err |= __put_user(mask, &sc->oldmask); 183 err |= __put_user(mask, &sc->oldmask);
@@ -295,6 +187,32 @@ setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
295} 187}
296 188
297/* 189/*
190 * Set up a signal frame.
191 */
192#ifdef CONFIG_X86_32
193static const struct {
194 u16 poplmovl;
195 u32 val;
196 u16 int80;
197} __attribute__((packed)) retcode = {
198 0xb858, /* popl %eax; movl $..., %eax */
199 __NR_sigreturn,
200 0x80cd, /* int $0x80 */
201};
202
203static const struct {
204 u8 movl;
205 u32 val;
206 u16 int80;
207 u8 pad;
208} __attribute__((packed)) rt_retcode = {
209 0xb8, /* movl $..., %eax */
210 __NR_rt_sigreturn,
211 0x80cd, /* int $0x80 */
212 0
213};
214
215/*
298 * Determine which stack to use.. 216 * Determine which stack to use..
299 */ 217 */
300static inline void __user * 218static inline void __user *
@@ -328,6 +246,8 @@ get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size,
328 if (used_math()) { 246 if (used_math()) {
329 sp = sp - sig_xstate_size; 247 sp = sp - sig_xstate_size;
330 *fpstate = (struct _fpstate *) sp; 248 *fpstate = (struct _fpstate *) sp;
249 if (save_i387_xstate(*fpstate) < 0)
250 return (void __user *)-1L;
331 } 251 }
332 252
333 sp -= frame_size; 253 sp -= frame_size;
@@ -383,9 +303,7 @@ __setup_frame(int sig, struct k_sigaction *ka, sigset_t *set,
383 * reasons and because gdb uses it as a signature to notice 303 * reasons and because gdb uses it as a signature to notice
384 * signal handler stack frames. 304 * signal handler stack frames.
385 */ 305 */
386 err |= __put_user(0xb858, (short __user *)(frame->retcode+0)); 306 err |= __put_user(*((u64 *)&retcode), (u64 *)frame->retcode);
387 err |= __put_user(__NR_sigreturn, (int __user *)(frame->retcode+2));
388 err |= __put_user(0x80cd, (short __user *)(frame->retcode+6));
389 307
390 if (err) 308 if (err)
391 return -EFAULT; 309 return -EFAULT;
@@ -454,9 +372,7 @@ static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
454 * reasons and because gdb uses it as a signature to notice 372 * reasons and because gdb uses it as a signature to notice
455 * signal handler stack frames. 373 * signal handler stack frames.
456 */ 374 */
457 err |= __put_user(0xb8, (char __user *)(frame->retcode+0)); 375 err |= __put_user(*((u64 *)&rt_retcode), (u64 *)frame->retcode);
458 err |= __put_user(__NR_rt_sigreturn, (int __user *)(frame->retcode+1));
459 err |= __put_user(0x80cd, (short __user *)(frame->retcode+5));
460 376
461 if (err) 377 if (err)
462 return -EFAULT; 378 return -EFAULT;
@@ -475,23 +391,293 @@ static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
475 391
476 return 0; 392 return 0;
477} 393}
394#else /* !CONFIG_X86_32 */
395/*
396 * Determine which stack to use..
397 */
398static void __user *
399get_stack(struct k_sigaction *ka, unsigned long sp, unsigned long size)
400{
401 /* Default to using normal stack - redzone*/
402 sp -= 128;
403
404 /* This is the X/Open sanctioned signal stack switching. */
405 if (ka->sa.sa_flags & SA_ONSTACK) {
406 if (sas_ss_flags(sp) == 0)
407 sp = current->sas_ss_sp + current->sas_ss_size;
408 }
409
410 return (void __user *)round_down(sp - size, 64);
411}
412
413static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
414 sigset_t *set, struct pt_regs *regs)
415{
416 struct rt_sigframe __user *frame;
417 void __user *fp = NULL;
418 int err = 0;
419 struct task_struct *me = current;
420
421 if (used_math()) {
422 fp = get_stack(ka, regs->sp, sig_xstate_size);
423 frame = (void __user *)round_down(
424 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8;
425
426 if (save_i387_xstate(fp) < 0)
427 return -EFAULT;
428 } else
429 frame = get_stack(ka, regs->sp, sizeof(struct rt_sigframe)) - 8;
430
431 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
432 return -EFAULT;
433
434 if (ka->sa.sa_flags & SA_SIGINFO) {
435 if (copy_siginfo_to_user(&frame->info, info))
436 return -EFAULT;
437 }
438
439 /* Create the ucontext. */
440 if (cpu_has_xsave)
441 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
442 else
443 err |= __put_user(0, &frame->uc.uc_flags);
444 err |= __put_user(0, &frame->uc.uc_link);
445 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
446 err |= __put_user(sas_ss_flags(regs->sp),
447 &frame->uc.uc_stack.ss_flags);
448 err |= __put_user(me->sas_ss_size, &frame->uc.uc_stack.ss_size);
449 err |= setup_sigcontext(&frame->uc.uc_mcontext, fp, regs, set->sig[0]);
450 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
451
452 /* Set up to return from userspace. If provided, use a stub
453 already in userspace. */
454 /* x86-64 should always use SA_RESTORER. */
455 if (ka->sa.sa_flags & SA_RESTORER) {
456 err |= __put_user(ka->sa.sa_restorer, &frame->pretcode);
457 } else {
458 /* could use a vstub here */
459 return -EFAULT;
460 }
461
462 if (err)
463 return -EFAULT;
464
465 /* Set up registers for signal handler */
466 regs->di = sig;
467 /* In case the signal handler was declared without prototypes */
468 regs->ax = 0;
469
470 /* This also works for non SA_SIGINFO handlers because they expect the
471 next argument after the signal number on the stack. */
472 regs->si = (unsigned long)&frame->info;
473 regs->dx = (unsigned long)&frame->uc;
474 regs->ip = (unsigned long) ka->sa.sa_handler;
475
476 regs->sp = (unsigned long)frame;
477
478 /* Set up the CS register to run signal handlers in 64-bit mode,
479 even if the handler happens to be interrupting 32-bit code. */
480 regs->cs = __USER_CS;
481
482 return 0;
483}
484#endif /* CONFIG_X86_32 */
485
486#ifdef CONFIG_X86_32
487/*
488 * Atomically swap in the new signal mask, and wait for a signal.
489 */
490asmlinkage int
491sys_sigsuspend(int history0, int history1, old_sigset_t mask)
492{
493 mask &= _BLOCKABLE;
494 spin_lock_irq(&current->sighand->siglock);
495 current->saved_sigmask = current->blocked;
496 siginitset(&current->blocked, mask);
497 recalc_sigpending();
498 spin_unlock_irq(&current->sighand->siglock);
499
500 current->state = TASK_INTERRUPTIBLE;
501 schedule();
502 set_restore_sigmask();
503
504 return -ERESTARTNOHAND;
505}
506
507asmlinkage int
508sys_sigaction(int sig, const struct old_sigaction __user *act,
509 struct old_sigaction __user *oact)
510{
511 struct k_sigaction new_ka, old_ka;
512 int ret;
513
514 if (act) {
515 old_sigset_t mask;
516
517 if (!access_ok(VERIFY_READ, act, sizeof(*act)) ||
518 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
519 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer))
520 return -EFAULT;
521
522 __get_user(new_ka.sa.sa_flags, &act->sa_flags);
523 __get_user(mask, &act->sa_mask);
524 siginitset(&new_ka.sa.sa_mask, mask);
525 }
526
527 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
528
529 if (!ret && oact) {
530 if (!access_ok(VERIFY_WRITE, oact, sizeof(*oact)) ||
531 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
532 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer))
533 return -EFAULT;
534
535 __put_user(old_ka.sa.sa_flags, &oact->sa_flags);
536 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask);
537 }
538
539 return ret;
540}
541#endif /* CONFIG_X86_32 */
542
543#ifdef CONFIG_X86_32
544asmlinkage int sys_sigaltstack(unsigned long bx)
545{
546 /*
547 * This is needed to make gcc realize it doesn't own the
548 * "struct pt_regs"
549 */
550 struct pt_regs *regs = (struct pt_regs *)&bx;
551 const stack_t __user *uss = (const stack_t __user *)bx;
552 stack_t __user *uoss = (stack_t __user *)regs->cx;
553
554 return do_sigaltstack(uss, uoss, regs->sp);
555}
556#else /* !CONFIG_X86_32 */
557asmlinkage long
558sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
559 struct pt_regs *regs)
560{
561 return do_sigaltstack(uss, uoss, regs->sp);
562}
563#endif /* CONFIG_X86_32 */
564
565/*
566 * Do a signal return; undo the signal stack.
567 */
568#ifdef CONFIG_X86_32
569asmlinkage unsigned long sys_sigreturn(unsigned long __unused)
570{
571 struct sigframe __user *frame;
572 struct pt_regs *regs;
573 unsigned long ax;
574 sigset_t set;
575
576 regs = (struct pt_regs *) &__unused;
577 frame = (struct sigframe __user *)(regs->sp - 8);
578
579 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
580 goto badframe;
581 if (__get_user(set.sig[0], &frame->sc.oldmask) || (_NSIG_WORDS > 1
582 && __copy_from_user(&set.sig[1], &frame->extramask,
583 sizeof(frame->extramask))))
584 goto badframe;
585
586 sigdelsetmask(&set, ~_BLOCKABLE);
587 spin_lock_irq(&current->sighand->siglock);
588 current->blocked = set;
589 recalc_sigpending();
590 spin_unlock_irq(&current->sighand->siglock);
591
592 if (restore_sigcontext(regs, &frame->sc, &ax))
593 goto badframe;
594 return ax;
595
596badframe:
597 signal_fault(regs, frame, "sigreturn");
598
599 return 0;
600}
601#endif /* CONFIG_X86_32 */
602
603static long do_rt_sigreturn(struct pt_regs *regs)
604{
605 struct rt_sigframe __user *frame;
606 unsigned long ax;
607 sigset_t set;
608
609 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long));
610 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
611 goto badframe;
612 if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
613 goto badframe;
614
615 sigdelsetmask(&set, ~_BLOCKABLE);
616 spin_lock_irq(&current->sighand->siglock);
617 current->blocked = set;
618 recalc_sigpending();
619 spin_unlock_irq(&current->sighand->siglock);
620
621 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &ax))
622 goto badframe;
623
624 if (do_sigaltstack(&frame->uc.uc_stack, NULL, regs->sp) == -EFAULT)
625 goto badframe;
626
627 return ax;
628
629badframe:
630 signal_fault(regs, frame, "rt_sigreturn");
631 return 0;
632}
633
634#ifdef CONFIG_X86_32
635asmlinkage int sys_rt_sigreturn(struct pt_regs regs)
636{
637 return do_rt_sigreturn(&regs);
638}
639#else /* !CONFIG_X86_32 */
640asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
641{
642 return do_rt_sigreturn(regs);
643}
644#endif /* CONFIG_X86_32 */
478 645
479/* 646/*
480 * OK, we're invoking a handler: 647 * OK, we're invoking a handler:
481 */ 648 */
482static int signr_convert(int sig) 649static int signr_convert(int sig)
483{ 650{
651#ifdef CONFIG_X86_32
484 struct thread_info *info = current_thread_info(); 652 struct thread_info *info = current_thread_info();
485 653
486 if (info->exec_domain && info->exec_domain->signal_invmap && sig < 32) 654 if (info->exec_domain && info->exec_domain->signal_invmap && sig < 32)
487 return info->exec_domain->signal_invmap[sig]; 655 return info->exec_domain->signal_invmap[sig];
656#endif /* CONFIG_X86_32 */
488 return sig; 657 return sig;
489} 658}
490 659
660#ifdef CONFIG_X86_32
661
491#define is_ia32 1 662#define is_ia32 1
492#define ia32_setup_frame __setup_frame 663#define ia32_setup_frame __setup_frame
493#define ia32_setup_rt_frame __setup_rt_frame 664#define ia32_setup_rt_frame __setup_rt_frame
494 665
666#else /* !CONFIG_X86_32 */
667
668#ifdef CONFIG_IA32_EMULATION
669#define is_ia32 test_thread_flag(TIF_IA32)
670#else /* !CONFIG_IA32_EMULATION */
671#define is_ia32 0
672#endif /* CONFIG_IA32_EMULATION */
673
674int ia32_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
675 sigset_t *set, struct pt_regs *regs);
676int ia32_setup_frame(int sig, struct k_sigaction *ka,
677 sigset_t *set, struct pt_regs *regs);
678
679#endif /* CONFIG_X86_32 */
680
495static int 681static int
496setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info, 682setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
497 sigset_t *set, struct pt_regs *regs) 683 sigset_t *set, struct pt_regs *regs)
@@ -592,7 +778,13 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
592 return 0; 778 return 0;
593} 779}
594 780
781#ifdef CONFIG_X86_32
595#define NR_restart_syscall __NR_restart_syscall 782#define NR_restart_syscall __NR_restart_syscall
783#else /* !CONFIG_X86_32 */
784#define NR_restart_syscall \
785 test_thread_flag(TIF_IA32) ? __NR_ia32_restart_syscall : __NR_restart_syscall
786#endif /* CONFIG_X86_32 */
787
596/* 788/*
597 * Note that 'init' is a special process: it doesn't get signals it doesn't 789 * Note that 'init' is a special process: it doesn't get signals it doesn't
598 * want to handle. Thus you cannot kill init even with a SIGKILL even by 790 * want to handle. Thus you cannot kill init even with a SIGKILL even by
@@ -704,8 +896,9 @@ void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
704 struct task_struct *me = current; 896 struct task_struct *me = current;
705 897
706 if (show_unhandled_signals && printk_ratelimit()) { 898 if (show_unhandled_signals && printk_ratelimit()) {
707 printk(KERN_INFO 899 printk("%s"
708 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx", 900 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
901 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
709 me->comm, me->pid, where, frame, 902 me->comm, me->pid, where, frame,
710 regs->ip, regs->sp, regs->orig_ax); 903 regs->ip, regs->sp, regs->orig_ax);
711 print_vma_addr(" in ", regs->ip); 904 print_vma_addr(" in ", regs->ip);
diff --git a/arch/x86/kernel/signal_64.c b/arch/x86/kernel/signal_64.c
deleted file mode 100644
index a5c9627f4db9..000000000000
--- a/arch/x86/kernel/signal_64.c
+++ /dev/null
@@ -1,516 +0,0 @@
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
4 *
5 * 1997-11-28 Modified for POSIX.1b signals by Richard Henderson
6 * 2000-06-20 Pentium III FXSR, SSE support by Gareth Hughes
7 * 2000-2002 x86-64 support by Andi Kleen
8 */
9
10#include <linux/sched.h>
11#include <linux/mm.h>
12#include <linux/smp.h>
13#include <linux/kernel.h>
14#include <linux/signal.h>
15#include <linux/errno.h>
16#include <linux/wait.h>
17#include <linux/ptrace.h>
18#include <linux/tracehook.h>
19#include <linux/unistd.h>
20#include <linux/stddef.h>
21#include <linux/personality.h>
22#include <linux/compiler.h>
23#include <linux/uaccess.h>
24
25#include <asm/processor.h>
26#include <asm/ucontext.h>
27#include <asm/i387.h>
28#include <asm/proto.h>
29#include <asm/ia32_unistd.h>
30#include <asm/mce.h>
31#include <asm/syscall.h>
32#include <asm/syscalls.h>
33#include "sigframe.h"
34
35#define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
36
37#define __FIX_EFLAGS (X86_EFLAGS_AC | X86_EFLAGS_OF | \
38 X86_EFLAGS_DF | X86_EFLAGS_TF | X86_EFLAGS_SF | \
39 X86_EFLAGS_ZF | X86_EFLAGS_AF | X86_EFLAGS_PF | \
40 X86_EFLAGS_CF)
41
42#ifdef CONFIG_X86_32
43# define FIX_EFLAGS (__FIX_EFLAGS | X86_EFLAGS_RF)
44#else
45# define FIX_EFLAGS __FIX_EFLAGS
46#endif
47
48asmlinkage long
49sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss,
50 struct pt_regs *regs)
51{
52 return do_sigaltstack(uss, uoss, regs->sp);
53}
54
55#define COPY(x) { \
56 err |= __get_user(regs->x, &sc->x); \
57}
58
59#define COPY_SEG_STRICT(seg) { \
60 unsigned short tmp; \
61 err |= __get_user(tmp, &sc->seg); \
62 regs->seg = tmp | 3; \
63}
64
65/*
66 * Do a signal return; undo the signal stack.
67 */
68static int
69restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc,
70 unsigned long *pax)
71{
72 void __user *buf;
73 unsigned int tmpflags;
74 unsigned int err = 0;
75
76 /* Always make any pending restarted system calls return -EINTR */
77 current_thread_info()->restart_block.fn = do_no_restart_syscall;
78
79 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx);
80 COPY(dx); COPY(cx); COPY(ip);
81 COPY(r8);
82 COPY(r9);
83 COPY(r10);
84 COPY(r11);
85 COPY(r12);
86 COPY(r13);
87 COPY(r14);
88 COPY(r15);
89
90 /* Kernel saves and restores only the CS segment register on signals,
91 * which is the bare minimum needed to allow mixed 32/64-bit code.
92 * App's signal handler can save/restore other segments if needed. */
93 COPY_SEG_STRICT(cs);
94
95 err |= __get_user(tmpflags, &sc->flags);
96 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
97 regs->orig_ax = -1; /* disable syscall checks */
98
99 err |= __get_user(buf, &sc->fpstate);
100 err |= restore_i387_xstate(buf);
101
102 err |= __get_user(*pax, &sc->ax);
103 return err;
104}
105
106static long do_rt_sigreturn(struct pt_regs *regs)
107{
108 struct rt_sigframe __user *frame;
109 unsigned long ax;
110 sigset_t set;
111
112 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long));
113 if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
114 goto badframe;
115 if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
116 goto badframe;
117
118 sigdelsetmask(&set, ~_BLOCKABLE);
119 spin_lock_irq(&current->sighand->siglock);
120 current->blocked = set;
121 recalc_sigpending();
122 spin_unlock_irq(&current->sighand->siglock);
123
124 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &ax))
125 goto badframe;
126
127 if (do_sigaltstack(&frame->uc.uc_stack, NULL, regs->sp) == -EFAULT)
128 goto badframe;
129
130 return ax;
131
132badframe:
133 signal_fault(regs, frame, "rt_sigreturn");
134 return 0;
135}
136
137asmlinkage long sys_rt_sigreturn(struct pt_regs *regs)
138{
139 return do_rt_sigreturn(regs);
140}
141
142/*
143 * Set up a signal frame.
144 */
145
146static inline int
147setup_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs,
148 unsigned long mask, struct task_struct *me)
149{
150 int err = 0;
151
152 err |= __put_user(regs->cs, &sc->cs);
153 err |= __put_user(0, &sc->gs);
154 err |= __put_user(0, &sc->fs);
155
156 err |= __put_user(regs->di, &sc->di);
157 err |= __put_user(regs->si, &sc->si);
158 err |= __put_user(regs->bp, &sc->bp);
159 err |= __put_user(regs->sp, &sc->sp);
160 err |= __put_user(regs->bx, &sc->bx);
161 err |= __put_user(regs->dx, &sc->dx);
162 err |= __put_user(regs->cx, &sc->cx);
163 err |= __put_user(regs->ax, &sc->ax);
164 err |= __put_user(regs->r8, &sc->r8);
165 err |= __put_user(regs->r9, &sc->r9);
166 err |= __put_user(regs->r10, &sc->r10);
167 err |= __put_user(regs->r11, &sc->r11);
168 err |= __put_user(regs->r12, &sc->r12);
169 err |= __put_user(regs->r13, &sc->r13);
170 err |= __put_user(regs->r14, &sc->r14);
171 err |= __put_user(regs->r15, &sc->r15);
172 err |= __put_user(me->thread.trap_no, &sc->trapno);
173 err |= __put_user(me->thread.error_code, &sc->err);
174 err |= __put_user(regs->ip, &sc->ip);
175 err |= __put_user(regs->flags, &sc->flags);
176 err |= __put_user(mask, &sc->oldmask);
177 err |= __put_user(me->thread.cr2, &sc->cr2);
178
179 return err;
180}
181
182/*
183 * Determine which stack to use..
184 */
185
186static void __user *
187get_stack(struct k_sigaction *ka, struct pt_regs *regs, unsigned long size)
188{
189 unsigned long sp;
190
191 /* Default to using normal stack - redzone*/
192 sp = regs->sp - 128;
193
194 /* This is the X/Open sanctioned signal stack switching. */
195 if (ka->sa.sa_flags & SA_ONSTACK) {
196 if (sas_ss_flags(sp) == 0)
197 sp = current->sas_ss_sp + current->sas_ss_size;
198 }
199
200 return (void __user *)round_down(sp - size, 64);
201}
202
203static int __setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
204 sigset_t *set, struct pt_regs *regs)
205{
206 struct rt_sigframe __user *frame;
207 void __user *fp = NULL;
208 int err = 0;
209 struct task_struct *me = current;
210
211 if (used_math()) {
212 fp = get_stack(ka, regs, sig_xstate_size);
213 frame = (void __user *)round_down(
214 (unsigned long)fp - sizeof(struct rt_sigframe), 16) - 8;
215
216 if (save_i387_xstate(fp) < 0)
217 return -EFAULT;
218 } else
219 frame = get_stack(ka, regs, sizeof(struct rt_sigframe)) - 8;
220
221 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
222 return -EFAULT;
223
224 if (ka->sa.sa_flags & SA_SIGINFO) {
225 if (copy_siginfo_to_user(&frame->info, info))
226 return -EFAULT;
227 }
228
229 /* Create the ucontext. */
230 if (cpu_has_xsave)
231 err |= __put_user(UC_FP_XSTATE, &frame->uc.uc_flags);
232 else
233 err |= __put_user(0, &frame->uc.uc_flags);
234 err |= __put_user(0, &frame->uc.uc_link);
235 err |= __put_user(me->sas_ss_sp, &frame->uc.uc_stack.ss_sp);
236 err |= __put_user(sas_ss_flags(regs->sp),
237 &frame->uc.uc_stack.ss_flags);
238 err |= __put_user(me->sas_ss_size, &frame->uc.uc_stack.ss_size);
239 err |= setup_sigcontext(&frame->uc.uc_mcontext, regs, set->sig[0], me);
240 err |= __put_user(fp, &frame->uc.uc_mcontext.fpstate);
241 if (sizeof(*set) == 16) {
242 __put_user(set->sig[0], &frame->uc.uc_sigmask.sig[0]);
243 __put_user(set->sig[1], &frame->uc.uc_sigmask.sig[1]);
244 } else
245 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
246
247 /* Set up to return from userspace. If provided, use a stub
248 already in userspace. */
249 /* x86-64 should always use SA_RESTORER. */
250 if (ka->sa.sa_flags & SA_RESTORER) {
251 err |= __put_user(ka->sa.sa_restorer, &frame->pretcode);
252 } else {
253 /* could use a vstub here */
254 return -EFAULT;
255 }
256
257 if (err)
258 return -EFAULT;
259
260 /* Set up registers for signal handler */
261 regs->di = sig;
262 /* In case the signal handler was declared without prototypes */
263 regs->ax = 0;
264
265 /* This also works for non SA_SIGINFO handlers because they expect the
266 next argument after the signal number on the stack. */
267 regs->si = (unsigned long)&frame->info;
268 regs->dx = (unsigned long)&frame->uc;
269 regs->ip = (unsigned long) ka->sa.sa_handler;
270
271 regs->sp = (unsigned long)frame;
272
273 /* Set up the CS register to run signal handlers in 64-bit mode,
274 even if the handler happens to be interrupting 32-bit code. */
275 regs->cs = __USER_CS;
276
277 return 0;
278}
279
280/*
281 * OK, we're invoking a handler
282 */
283static int signr_convert(int sig)
284{
285 return sig;
286}
287
288#ifdef CONFIG_IA32_EMULATION
289#define is_ia32 test_thread_flag(TIF_IA32)
290#else
291#define is_ia32 0
292#endif
293
294static int
295setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
296 sigset_t *set, struct pt_regs *regs)
297{
298 int usig = signr_convert(sig);
299 int ret;
300
301 /* Set up the stack frame */
302 if (is_ia32) {
303 if (ka->sa.sa_flags & SA_SIGINFO)
304 ret = ia32_setup_rt_frame(usig, ka, info, set, regs);
305 else
306 ret = ia32_setup_frame(usig, ka, set, regs);
307 } else
308 ret = __setup_rt_frame(sig, ka, info, set, regs);
309
310 if (ret) {
311 force_sigsegv(sig, current);
312 return -EFAULT;
313 }
314
315 return ret;
316}
317
318static int
319handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
320 sigset_t *oldset, struct pt_regs *regs)
321{
322 int ret;
323
324 /* Are we from a system call? */
325 if (syscall_get_nr(current, regs) >= 0) {
326 /* If so, check system call restarting.. */
327 switch (syscall_get_error(current, regs)) {
328 case -ERESTART_RESTARTBLOCK:
329 case -ERESTARTNOHAND:
330 regs->ax = -EINTR;
331 break;
332
333 case -ERESTARTSYS:
334 if (!(ka->sa.sa_flags & SA_RESTART)) {
335 regs->ax = -EINTR;
336 break;
337 }
338 /* fallthrough */
339 case -ERESTARTNOINTR:
340 regs->ax = regs->orig_ax;
341 regs->ip -= 2;
342 break;
343 }
344 }
345
346 /*
347 * If TF is set due to a debugger (TIF_FORCED_TF), clear the TF
348 * flag so that register information in the sigcontext is correct.
349 */
350 if (unlikely(regs->flags & X86_EFLAGS_TF) &&
351 likely(test_and_clear_thread_flag(TIF_FORCED_TF)))
352 regs->flags &= ~X86_EFLAGS_TF;
353
354 ret = setup_rt_frame(sig, ka, info, oldset, regs);
355
356 if (ret)
357 return ret;
358
359#ifdef CONFIG_X86_64
360 /*
361 * This has nothing to do with segment registers,
362 * despite the name. This magic affects uaccess.h
363 * macros' behavior. Reset it to the normal setting.
364 */
365 set_fs(USER_DS);
366#endif
367
368 /*
369 * Clear the direction flag as per the ABI for function entry.
370 */
371 regs->flags &= ~X86_EFLAGS_DF;
372
373 /*
374 * Clear TF when entering the signal handler, but
375 * notify any tracer that was single-stepping it.
376 * The tracer may want to single-step inside the
377 * handler too.
378 */
379 regs->flags &= ~X86_EFLAGS_TF;
380
381 spin_lock_irq(&current->sighand->siglock);
382 sigorsets(&current->blocked, &current->blocked, &ka->sa.sa_mask);
383 if (!(ka->sa.sa_flags & SA_NODEFER))
384 sigaddset(&current->blocked, sig);
385 recalc_sigpending();
386 spin_unlock_irq(&current->sighand->siglock);
387
388 tracehook_signal_handler(sig, info, ka, regs,
389 test_thread_flag(TIF_SINGLESTEP));
390
391 return 0;
392}
393
394#define NR_restart_syscall \
395 test_thread_flag(TIF_IA32) ? __NR_ia32_restart_syscall : __NR_restart_syscall
396/*
397 * Note that 'init' is a special process: it doesn't get signals it doesn't
398 * want to handle. Thus you cannot kill init even with a SIGKILL even by
399 * mistake.
400 */
401static void do_signal(struct pt_regs *regs)
402{
403 struct k_sigaction ka;
404 siginfo_t info;
405 int signr;
406 sigset_t *oldset;
407
408 /*
409 * We want the common case to go fast, which is why we may in certain
410 * cases get here from kernel mode. Just return without doing anything
411 * if so.
412 * X86_32: vm86 regs switched out by assembly code before reaching
413 * here, so testing against kernel CS suffices.
414 */
415 if (!user_mode(regs))
416 return;
417
418 if (current_thread_info()->status & TS_RESTORE_SIGMASK)
419 oldset = &current->saved_sigmask;
420 else
421 oldset = &current->blocked;
422
423 signr = get_signal_to_deliver(&info, &ka, regs, NULL);
424 if (signr > 0) {
425 /*
426 * Re-enable any watchpoints before delivering the
427 * signal to user space. The processor register will
428 * have been cleared if the watchpoint triggered
429 * inside the kernel.
430 */
431 if (current->thread.debugreg7)
432 set_debugreg(current->thread.debugreg7, 7);
433
434 /* Whee! Actually deliver the signal. */
435 if (handle_signal(signr, &info, &ka, oldset, regs) == 0) {
436 /*
437 * A signal was successfully delivered; the saved
438 * sigmask will have been stored in the signal frame,
439 * and will be restored by sigreturn, so we can simply
440 * clear the TS_RESTORE_SIGMASK flag.
441 */
442 current_thread_info()->status &= ~TS_RESTORE_SIGMASK;
443 }
444 return;
445 }
446
447 /* Did we come from a system call? */
448 if (syscall_get_nr(current, regs) >= 0) {
449 /* Restart the system call - no handlers present */
450 switch (syscall_get_error(current, regs)) {
451 case -ERESTARTNOHAND:
452 case -ERESTARTSYS:
453 case -ERESTARTNOINTR:
454 regs->ax = regs->orig_ax;
455 regs->ip -= 2;
456 break;
457
458 case -ERESTART_RESTARTBLOCK:
459 regs->ax = NR_restart_syscall;
460 regs->ip -= 2;
461 break;
462 }
463 }
464
465 /*
466 * If there's no signal to deliver, we just put the saved sigmask
467 * back.
468 */
469 if (current_thread_info()->status & TS_RESTORE_SIGMASK) {
470 current_thread_info()->status &= ~TS_RESTORE_SIGMASK;
471 sigprocmask(SIG_SETMASK, &current->saved_sigmask, NULL);
472 }
473}
474
475/*
476 * notification of userspace execution resumption
477 * - triggered by the TIF_WORK_MASK flags
478 */
479void
480do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags)
481{
482#if defined(CONFIG_X86_64) && defined(CONFIG_X86_MCE)
483 /* notify userspace of pending MCEs */
484 if (thread_info_flags & _TIF_MCE_NOTIFY)
485 mce_notify_user();
486#endif /* CONFIG_X86_64 && CONFIG_X86_MCE */
487
488 /* deal with pending signal delivery */
489 if (thread_info_flags & _TIF_SIGPENDING)
490 do_signal(regs);
491
492 if (thread_info_flags & _TIF_NOTIFY_RESUME) {
493 clear_thread_flag(TIF_NOTIFY_RESUME);
494 tracehook_notify_resume(regs);
495 }
496
497#ifdef CONFIG_X86_32
498 clear_thread_flag(TIF_IRET);
499#endif /* CONFIG_X86_32 */
500}
501
502void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
503{
504 struct task_struct *me = current;
505
506 if (show_unhandled_signals && printk_ratelimit()) {
507 printk(KERN_INFO
508 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
509 me->comm, me->pid, where, frame,
510 regs->ip, regs->sp, regs->orig_ax);
511 print_vma_addr(" in ", regs->ip);
512 printk(KERN_CONT "\n");
513 }
514
515 force_sig(SIGSEGV, me);
516}
diff --git a/arch/x86/kernel/smp.c b/arch/x86/kernel/smp.c
index 18f9b19f5f8f..7e558db362c1 100644
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -140,19 +140,6 @@ void native_send_call_func_ipi(cpumask_t mask)
140 send_IPI_mask(mask, CALL_FUNCTION_VECTOR); 140 send_IPI_mask(mask, CALL_FUNCTION_VECTOR);
141} 141}
142 142
143static void stop_this_cpu(void *dummy)
144{
145 local_irq_disable();
146 /*
147 * Remove this CPU:
148 */
149 cpu_clear(smp_processor_id(), cpu_online_map);
150 disable_local_APIC();
151 if (hlt_works(smp_processor_id()))
152 for (;;) halt();
153 for (;;);
154}
155
156/* 143/*
157 * this function calls the 'stop' function on all other CPUs in the system. 144 * this function calls the 'stop' function on all other CPUs in the system.
158 */ 145 */
@@ -178,11 +165,7 @@ static void native_smp_send_stop(void)
178void smp_reschedule_interrupt(struct pt_regs *regs) 165void smp_reschedule_interrupt(struct pt_regs *regs)
179{ 166{
180 ack_APIC_irq(); 167 ack_APIC_irq();
181#ifdef CONFIG_X86_32 168 inc_irq_stat(irq_resched_count);
182 __get_cpu_var(irq_stat).irq_resched_count++;
183#else
184 add_pda(irq_resched_count, 1);
185#endif
186} 169}
187 170
188void smp_call_function_interrupt(struct pt_regs *regs) 171void smp_call_function_interrupt(struct pt_regs *regs)
@@ -190,11 +173,7 @@ void smp_call_function_interrupt(struct pt_regs *regs)
190 ack_APIC_irq(); 173 ack_APIC_irq();
191 irq_enter(); 174 irq_enter();
192 generic_smp_call_function_interrupt(); 175 generic_smp_call_function_interrupt();
193#ifdef CONFIG_X86_32 176 inc_irq_stat(irq_call_count);
194 __get_cpu_var(irq_stat).irq_call_count++;
195#else
196 add_pda(irq_call_count, 1);
197#endif
198 irq_exit(); 177 irq_exit();
199} 178}
200 179
@@ -203,11 +182,7 @@ void smp_call_function_single_interrupt(struct pt_regs *regs)
203 ack_APIC_irq(); 182 ack_APIC_irq();
204 irq_enter(); 183 irq_enter();
205 generic_smp_call_function_single_interrupt(); 184 generic_smp_call_function_single_interrupt();
206#ifdef CONFIG_X86_32 185 inc_irq_stat(irq_call_count);
207 __get_cpu_var(irq_stat).irq_call_count++;
208#else
209 add_pda(irq_call_count, 1);
210#endif
211 irq_exit(); 186 irq_exit();
212} 187}
213 188
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index f71f96fc9e62..f8500c969442 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -62,6 +62,7 @@
62#include <asm/mtrr.h> 62#include <asm/mtrr.h>
63#include <asm/vmi.h> 63#include <asm/vmi.h>
64#include <asm/genapic.h> 64#include <asm/genapic.h>
65#include <asm/setup.h>
65#include <linux/mc146818rtc.h> 66#include <linux/mc146818rtc.h>
66 67
67#include <mach_apic.h> 68#include <mach_apic.h>
@@ -287,7 +288,7 @@ static int __cpuinitdata unsafe_smp;
287/* 288/*
288 * Activate a secondary processor. 289 * Activate a secondary processor.
289 */ 290 */
290static void __cpuinit start_secondary(void *unused) 291notrace static void __cpuinit start_secondary(void *unused)
291{ 292{
292 /* 293 /*
293 * Don't put *anything* before cpu_init(), SMP booting is too 294 * Don't put *anything* before cpu_init(), SMP booting is too
@@ -534,7 +535,7 @@ static void impress_friends(void)
534 pr_debug("Before bogocount - setting activated=1.\n"); 535 pr_debug("Before bogocount - setting activated=1.\n");
535} 536}
536 537
537static inline void __inquire_remote_apic(int apicid) 538void __inquire_remote_apic(int apicid)
538{ 539{
539 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 }; 540 unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
540 char *names[] = { "ID", "VERSION", "SPIV" }; 541 char *names[] = { "ID", "VERSION", "SPIV" };
@@ -573,14 +574,13 @@ static inline void __inquire_remote_apic(int apicid)
573 } 574 }
574} 575}
575 576
576#ifdef WAKE_SECONDARY_VIA_NMI
577/* 577/*
578 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal 578 * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
579 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this 579 * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
580 * won't ... remember to clear down the APIC, etc later. 580 * won't ... remember to clear down the APIC, etc later.
581 */ 581 */
582static int __devinit 582int __devinit
583wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip) 583wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
584{ 584{
585 unsigned long send_status, accept_status = 0; 585 unsigned long send_status, accept_status = 0;
586 int maxlvt; 586 int maxlvt;
@@ -597,7 +597,7 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
597 * Give the other CPU some time to accept the IPI. 597 * Give the other CPU some time to accept the IPI.
598 */ 598 */
599 udelay(200); 599 udelay(200);
600 if (APIC_INTEGRATED(apic_version[phys_apicid])) { 600 if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
601 maxlvt = lapic_get_maxlvt(); 601 maxlvt = lapic_get_maxlvt();
602 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ 602 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
603 apic_write(APIC_ESR, 0); 603 apic_write(APIC_ESR, 0);
@@ -612,11 +612,9 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
612 612
613 return (send_status | accept_status); 613 return (send_status | accept_status);
614} 614}
615#endif /* WAKE_SECONDARY_VIA_NMI */
616 615
617#ifdef WAKE_SECONDARY_VIA_INIT 616int __devinit
618static int __devinit 617wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
619wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
620{ 618{
621 unsigned long send_status, accept_status = 0; 619 unsigned long send_status, accept_status = 0;
622 int maxlvt, num_starts, j; 620 int maxlvt, num_starts, j;
@@ -735,7 +733,6 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
735 733
736 return (send_status | accept_status); 734 return (send_status | accept_status);
737} 735}
738#endif /* WAKE_SECONDARY_VIA_INIT */
739 736
740struct create_idle { 737struct create_idle {
741 struct work_struct work; 738 struct work_struct work;
@@ -1084,8 +1081,10 @@ static int __init smp_sanity_check(unsigned max_cpus)
1084#endif 1081#endif
1085 1082
1086 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) { 1083 if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
1087 printk(KERN_WARNING "weird, boot CPU (#%d) not listed" 1084 printk(KERN_WARNING
1088 "by the BIOS.\n", hard_smp_processor_id()); 1085 "weird, boot CPU (#%d) not listed by the BIOS.\n",
1086 hard_smp_processor_id());
1087
1089 physid_set(hard_smp_processor_id(), phys_cpu_present_map); 1088 physid_set(hard_smp_processor_id(), phys_cpu_present_map);
1090 } 1089 }
1091 1090
diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
index a03e7f6d90c3..10786af95545 100644
--- a/arch/x86/kernel/stacktrace.c
+++ b/arch/x86/kernel/stacktrace.c
@@ -6,6 +6,7 @@
6#include <linux/sched.h> 6#include <linux/sched.h>
7#include <linux/stacktrace.h> 7#include <linux/stacktrace.h>
8#include <linux/module.h> 8#include <linux/module.h>
9#include <linux/uaccess.h>
9#include <asm/stacktrace.h> 10#include <asm/stacktrace.h>
10 11
11static void save_stack_warning(void *data, char *msg) 12static void save_stack_warning(void *data, char *msg)
@@ -83,3 +84,66 @@ void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
83 trace->entries[trace->nr_entries++] = ULONG_MAX; 84 trace->entries[trace->nr_entries++] = ULONG_MAX;
84} 85}
85EXPORT_SYMBOL_GPL(save_stack_trace_tsk); 86EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
87
88/* Userspace stacktrace - based on kernel/trace/trace_sysprof.c */
89
90struct stack_frame {
91 const void __user *next_fp;
92 unsigned long ret_addr;
93};
94
95static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
96{
97 int ret;
98
99 if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
100 return 0;
101
102 ret = 1;
103 pagefault_disable();
104 if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
105 ret = 0;
106 pagefault_enable();
107
108 return ret;
109}
110
111static inline void __save_stack_trace_user(struct stack_trace *trace)
112{
113 const struct pt_regs *regs = task_pt_regs(current);
114 const void __user *fp = (const void __user *)regs->bp;
115
116 if (trace->nr_entries < trace->max_entries)
117 trace->entries[trace->nr_entries++] = regs->ip;
118
119 while (trace->nr_entries < trace->max_entries) {
120 struct stack_frame frame;
121
122 frame.next_fp = NULL;
123 frame.ret_addr = 0;
124 if (!copy_stack_frame(fp, &frame))
125 break;
126 if ((unsigned long)fp < regs->sp)
127 break;
128 if (frame.ret_addr) {
129 trace->entries[trace->nr_entries++] =
130 frame.ret_addr;
131 }
132 if (fp == frame.next_fp)
133 break;
134 fp = frame.next_fp;
135 }
136}
137
138void save_stack_trace_user(struct stack_trace *trace)
139{
140 /*
141 * Trace user stack if we are not a kernel thread
142 */
143 if (current->mm) {
144 __save_stack_trace_user(trace);
145 }
146 if (trace->nr_entries < trace->max_entries)
147 trace->entries[trace->nr_entries++] = ULONG_MAX;
148}
149
diff --git a/arch/x86/kernel/time_32.c b/arch/x86/kernel/time_32.c
index 77b400f06ea2..65309e4cb1c0 100644
--- a/arch/x86/kernel/time_32.c
+++ b/arch/x86/kernel/time_32.c
@@ -75,7 +75,7 @@ EXPORT_SYMBOL(profile_pc);
75irqreturn_t timer_interrupt(int irq, void *dev_id) 75irqreturn_t timer_interrupt(int irq, void *dev_id)
76{ 76{
77 /* Keep nmi watchdog up to date */ 77 /* Keep nmi watchdog up to date */
78 per_cpu(irq_stat, smp_processor_id()).irq0_irqs++; 78 inc_irq_stat(irq0_irqs);
79 79
80#ifdef CONFIG_X86_IO_APIC 80#ifdef CONFIG_X86_IO_APIC
81 if (timer_ack) { 81 if (timer_ack) {
diff --git a/arch/x86/kernel/time_64.c b/arch/x86/kernel/time_64.c
index cb19d650c216..891e7a7c4334 100644
--- a/arch/x86/kernel/time_64.c
+++ b/arch/x86/kernel/time_64.c
@@ -49,9 +49,9 @@ unsigned long profile_pc(struct pt_regs *regs)
49} 49}
50EXPORT_SYMBOL(profile_pc); 50EXPORT_SYMBOL(profile_pc);
51 51
52irqreturn_t timer_interrupt(int irq, void *dev_id) 52static irqreturn_t timer_interrupt(int irq, void *dev_id)
53{ 53{
54 add_pda(irq0_irqs, 1); 54 inc_irq_stat(irq0_irqs);
55 55
56 global_clock_event->event_handler(global_clock_event); 56 global_clock_event->event_handler(global_clock_event);
57 57
@@ -80,6 +80,8 @@ unsigned long __init calibrate_cpu(void)
80 break; 80 break;
81 no_ctr_free = (i == 4); 81 no_ctr_free = (i == 4);
82 if (no_ctr_free) { 82 if (no_ctr_free) {
83 WARN(1, KERN_WARNING "Warning: AMD perfctrs busy ... "
84 "cpu_khz value may be incorrect.\n");
83 i = 3; 85 i = 3;
84 rdmsrl(MSR_K7_EVNTSEL3, evntsel3); 86 rdmsrl(MSR_K7_EVNTSEL3, evntsel3);
85 wrmsrl(MSR_K7_EVNTSEL3, 0); 87 wrmsrl(MSR_K7_EVNTSEL3, 0);
diff --git a/arch/x86/kernel/tlb_32.c b/arch/x86/kernel/tlb_32.c
index f4049f3513b6..8da059f949be 100644
--- a/arch/x86/kernel/tlb_32.c
+++ b/arch/x86/kernel/tlb_32.c
@@ -34,9 +34,8 @@ static DEFINE_SPINLOCK(tlbstate_lock);
34 */ 34 */
35void leave_mm(int cpu) 35void leave_mm(int cpu)
36{ 36{
37 if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK) 37 BUG_ON(x86_read_percpu(cpu_tlbstate.state) == TLBSTATE_OK);
38 BUG(); 38 cpu_clear(cpu, x86_read_percpu(cpu_tlbstate.active_mm)->cpu_vm_mask);
39 cpu_clear(cpu, per_cpu(cpu_tlbstate, cpu).active_mm->cpu_vm_mask);
40 load_cr3(swapper_pg_dir); 39 load_cr3(swapper_pg_dir);
41} 40}
42EXPORT_SYMBOL_GPL(leave_mm); 41EXPORT_SYMBOL_GPL(leave_mm);
@@ -104,8 +103,8 @@ void smp_invalidate_interrupt(struct pt_regs *regs)
104 * BUG(); 103 * BUG();
105 */ 104 */
106 105
107 if (flush_mm == per_cpu(cpu_tlbstate, cpu).active_mm) { 106 if (flush_mm == x86_read_percpu(cpu_tlbstate.active_mm)) {
108 if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_OK) { 107 if (x86_read_percpu(cpu_tlbstate.state) == TLBSTATE_OK) {
109 if (flush_va == TLB_FLUSH_ALL) 108 if (flush_va == TLB_FLUSH_ALL)
110 local_flush_tlb(); 109 local_flush_tlb();
111 else 110 else
@@ -119,7 +118,7 @@ void smp_invalidate_interrupt(struct pt_regs *regs)
119 smp_mb__after_clear_bit(); 118 smp_mb__after_clear_bit();
120out: 119out:
121 put_cpu_no_resched(); 120 put_cpu_no_resched();
122 __get_cpu_var(irq_stat).irq_tlb_count++; 121 inc_irq_stat(irq_tlb_count);
123} 122}
124 123
125void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm, 124void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm,
@@ -238,7 +237,7 @@ static void do_flush_tlb_all(void *info)
238 unsigned long cpu = smp_processor_id(); 237 unsigned long cpu = smp_processor_id();
239 238
240 __flush_tlb_all(); 239 __flush_tlb_all();
241 if (per_cpu(cpu_tlbstate, cpu).state == TLBSTATE_LAZY) 240 if (x86_read_percpu(cpu_tlbstate.state) == TLBSTATE_LAZY)
242 leave_mm(cpu); 241 leave_mm(cpu);
243} 242}
244 243
diff --git a/arch/x86/kernel/tlb_64.c b/arch/x86/kernel/tlb_64.c
index 8f919ca69494..29887d7081a9 100644
--- a/arch/x86/kernel/tlb_64.c
+++ b/arch/x86/kernel/tlb_64.c
@@ -154,7 +154,7 @@ asmlinkage void smp_invalidate_interrupt(struct pt_regs *regs)
154out: 154out:
155 ack_APIC_irq(); 155 ack_APIC_irq();
156 cpu_clear(cpu, f->flush_cpumask); 156 cpu_clear(cpu, f->flush_cpumask);
157 add_pda(irq_tlb_count, 1); 157 inc_irq_stat(irq_tlb_count);
158} 158}
159 159
160void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm, 160void native_flush_tlb_others(const cpumask_t *cpumaskp, struct mm_struct *mm,
diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
index 04431f34fd16..6a00e5faaa74 100644
--- a/arch/x86/kernel/tlb_uv.c
+++ b/arch/x86/kernel/tlb_uv.c
@@ -566,14 +566,10 @@ static int __init uv_ptc_init(void)
566 if (!is_uv_system()) 566 if (!is_uv_system())
567 return 0; 567 return 0;
568 568
569 if (!proc_mkdir("sgi_uv", NULL))
570 return -EINVAL;
571
572 proc_uv_ptc = create_proc_entry(UV_PTC_BASENAME, 0444, NULL); 569 proc_uv_ptc = create_proc_entry(UV_PTC_BASENAME, 0444, NULL);
573 if (!proc_uv_ptc) { 570 if (!proc_uv_ptc) {
574 printk(KERN_ERR "unable to create %s proc entry\n", 571 printk(KERN_ERR "unable to create %s proc entry\n",
575 UV_PTC_BASENAME); 572 UV_PTC_BASENAME);
576 remove_proc_entry("sgi_uv", NULL);
577 return -EINVAL; 573 return -EINVAL;
578 } 574 }
579 proc_uv_ptc->proc_fops = &proc_uv_ptc_operations; 575 proc_uv_ptc->proc_fops = &proc_uv_ptc_operations;
diff --git a/arch/x86/kernel/trampoline.c b/arch/x86/kernel/trampoline.c
index 1106fac6024d..808031a5ba19 100644
--- a/arch/x86/kernel/trampoline.c
+++ b/arch/x86/kernel/trampoline.c
@@ -1,10 +1,26 @@
1#include <linux/io.h> 1#include <linux/io.h>
2 2
3#include <asm/trampoline.h> 3#include <asm/trampoline.h>
4#include <asm/e820.h>
4 5
5/* ready for x86_64 and x86 */ 6/* ready for x86_64 and x86 */
6unsigned char *trampoline_base = __va(TRAMPOLINE_BASE); 7unsigned char *trampoline_base = __va(TRAMPOLINE_BASE);
7 8
9void __init reserve_trampoline_memory(void)
10{
11#ifdef CONFIG_X86_32
12 /*
13 * But first pinch a few for the stack/trampoline stuff
14 * FIXME: Don't need the extra page at 4K, but need to fix
15 * trampoline before removing it. (see the GDT stuff)
16 */
17 reserve_early(PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE");
18#endif
19 /* Has to be in very low memory so we can execute real-mode AP code. */
20 reserve_early(TRAMPOLINE_BASE, TRAMPOLINE_BASE + TRAMPOLINE_SIZE,
21 "TRAMPOLINE");
22}
23
8/* 24/*
9 * Currently trivial. Write the real->protected mode 25 * Currently trivial. Write the real->protected mode
10 * bootstrap into the page concerned. The caller 26 * bootstrap into the page concerned. The caller
@@ -12,7 +28,6 @@ unsigned char *trampoline_base = __va(TRAMPOLINE_BASE);
12 */ 28 */
13unsigned long setup_trampoline(void) 29unsigned long setup_trampoline(void)
14{ 30{
15 memcpy(trampoline_base, trampoline_data, 31 memcpy(trampoline_base, trampoline_data, TRAMPOLINE_SIZE);
16 trampoline_end - trampoline_data);
17 return virt_to_phys(trampoline_base); 32 return virt_to_phys(trampoline_base);
18} 33}
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 04d242ab0161..141907ab6e22 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -481,11 +481,7 @@ do_nmi(struct pt_regs *regs, long error_code)
481{ 481{
482 nmi_enter(); 482 nmi_enter();
483 483
484#ifdef CONFIG_X86_32 484 inc_irq_stat(__nmi_count);
485 { int cpu; cpu = smp_processor_id(); ++nmi_count(cpu); }
486#else
487 add_pda(__nmi_count, 1);
488#endif
489 485
490 if (!ignore_nmis) 486 if (!ignore_nmis)
491 default_do_nmi(regs); 487 default_do_nmi(regs);
@@ -664,7 +660,7 @@ void math_error(void __user *ip)
664{ 660{
665 struct task_struct *task; 661 struct task_struct *task;
666 siginfo_t info; 662 siginfo_t info;
667 unsigned short cwd, swd; 663 unsigned short cwd, swd, err;
668 664
669 /* 665 /*
670 * Save the info for the exception handler and clear the error. 666 * Save the info for the exception handler and clear the error.
@@ -675,7 +671,6 @@ void math_error(void __user *ip)
675 task->thread.error_code = 0; 671 task->thread.error_code = 0;
676 info.si_signo = SIGFPE; 672 info.si_signo = SIGFPE;
677 info.si_errno = 0; 673 info.si_errno = 0;
678 info.si_code = __SI_FAULT;
679 info.si_addr = ip; 674 info.si_addr = ip;
680 /* 675 /*
681 * (~cwd & swd) will mask out exceptions that are not set to unmasked 676 * (~cwd & swd) will mask out exceptions that are not set to unmasked
@@ -689,34 +684,31 @@ void math_error(void __user *ip)
689 */ 684 */
690 cwd = get_fpu_cwd(task); 685 cwd = get_fpu_cwd(task);
691 swd = get_fpu_swd(task); 686 swd = get_fpu_swd(task);
692 switch (swd & ~cwd & 0x3f) { 687
693 case 0x000: /* No unmasked exception */ 688 err = swd & ~cwd & 0x3f;
689
694#ifdef CONFIG_X86_32 690#ifdef CONFIG_X86_32
691 if (!err)
695 return; 692 return;
696#endif 693#endif
697 default: /* Multiple exceptions */ 694
698 break; 695 if (err & 0x001) { /* Invalid op */
699 case 0x001: /* Invalid Op */
700 /* 696 /*
701 * swd & 0x240 == 0x040: Stack Underflow 697 * swd & 0x240 == 0x040: Stack Underflow
702 * swd & 0x240 == 0x240: Stack Overflow 698 * swd & 0x240 == 0x240: Stack Overflow
703 * User must clear the SF bit (0x40) if set 699 * User must clear the SF bit (0x40) if set
704 */ 700 */
705 info.si_code = FPE_FLTINV; 701 info.si_code = FPE_FLTINV;
706 break; 702 } else if (err & 0x004) { /* Divide by Zero */
707 case 0x002: /* Denormalize */
708 case 0x010: /* Underflow */
709 info.si_code = FPE_FLTUND;
710 break;
711 case 0x004: /* Zero Divide */
712 info.si_code = FPE_FLTDIV; 703 info.si_code = FPE_FLTDIV;
713 break; 704 } else if (err & 0x008) { /* Overflow */
714 case 0x008: /* Overflow */
715 info.si_code = FPE_FLTOVF; 705 info.si_code = FPE_FLTOVF;
716 break; 706 } else if (err & 0x012) { /* Denormal, Underflow */
717 case 0x020: /* Precision */ 707 info.si_code = FPE_FLTUND;
708 } else if (err & 0x020) { /* Precision */
718 info.si_code = FPE_FLTRES; 709 info.si_code = FPE_FLTRES;
719 break; 710 } else {
711 info.si_code = __SI_FAULT|SI_KERNEL; /* WTF? */
720 } 712 }
721 force_sig_info(SIGFPE, &info, task); 713 force_sig_info(SIGFPE, &info, task);
722} 714}
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 424093b157d3..599e58168631 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -15,6 +15,7 @@
15#include <asm/vgtod.h> 15#include <asm/vgtod.h>
16#include <asm/time.h> 16#include <asm/time.h>
17#include <asm/delay.h> 17#include <asm/delay.h>
18#include <asm/hypervisor.h>
18 19
19unsigned int cpu_khz; /* TSC clocks / usec, not used here */ 20unsigned int cpu_khz; /* TSC clocks / usec, not used here */
20EXPORT_SYMBOL(cpu_khz); 21EXPORT_SYMBOL(cpu_khz);
@@ -31,6 +32,7 @@ static int tsc_unstable;
31 erroneous rdtsc usage on !cpu_has_tsc processors */ 32 erroneous rdtsc usage on !cpu_has_tsc processors */
32static int tsc_disabled = -1; 33static int tsc_disabled = -1;
33 34
35static int tsc_clocksource_reliable;
34/* 36/*
35 * Scheduler clock - returns current time in nanosec units. 37 * Scheduler clock - returns current time in nanosec units.
36 */ 38 */
@@ -98,6 +100,15 @@ int __init notsc_setup(char *str)
98 100
99__setup("notsc", notsc_setup); 101__setup("notsc", notsc_setup);
100 102
103static int __init tsc_setup(char *str)
104{
105 if (!strcmp(str, "reliable"))
106 tsc_clocksource_reliable = 1;
107 return 1;
108}
109
110__setup("tsc=", tsc_setup);
111
101#define MAX_RETRIES 5 112#define MAX_RETRIES 5
102#define SMI_TRESHOLD 50000 113#define SMI_TRESHOLD 50000
103 114
@@ -352,9 +363,15 @@ unsigned long native_calibrate_tsc(void)
352{ 363{
353 u64 tsc1, tsc2, delta, ref1, ref2; 364 u64 tsc1, tsc2, delta, ref1, ref2;
354 unsigned long tsc_pit_min = ULONG_MAX, tsc_ref_min = ULONG_MAX; 365 unsigned long tsc_pit_min = ULONG_MAX, tsc_ref_min = ULONG_MAX;
355 unsigned long flags, latch, ms, fast_calibrate; 366 unsigned long flags, latch, ms, fast_calibrate, tsc_khz;
356 int hpet = is_hpet_enabled(), i, loopmin; 367 int hpet = is_hpet_enabled(), i, loopmin;
357 368
369 tsc_khz = get_hypervisor_tsc_freq();
370 if (tsc_khz) {
371 printk(KERN_INFO "TSC: Frequency read from the hypervisor\n");
372 return tsc_khz;
373 }
374
358 local_irq_save(flags); 375 local_irq_save(flags);
359 fast_calibrate = quick_pit_calibrate(); 376 fast_calibrate = quick_pit_calibrate();
360 local_irq_restore(flags); 377 local_irq_restore(flags);
@@ -731,24 +748,21 @@ static struct dmi_system_id __initdata bad_tsc_dmi_table[] = {
731 {} 748 {}
732}; 749};
733 750
734/* 751static void __init check_system_tsc_reliable(void)
735 * Geode_LX - the OLPC CPU has a possibly a very reliable TSC 752{
736 */
737#ifdef CONFIG_MGEODE_LX 753#ifdef CONFIG_MGEODE_LX
738/* RTSC counts during suspend */ 754 /* RTSC counts during suspend */
739#define RTSC_SUSP 0x100 755#define RTSC_SUSP 0x100
740
741static void __init check_geode_tsc_reliable(void)
742{
743 unsigned long res_low, res_high; 756 unsigned long res_low, res_high;
744 757
745 rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high); 758 rdmsr_safe(MSR_GEODE_BUSCONT_CONF0, &res_low, &res_high);
759 /* Geode_LX - the OLPC CPU has a possibly a very reliable TSC */
746 if (res_low & RTSC_SUSP) 760 if (res_low & RTSC_SUSP)
747 clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY; 761 tsc_clocksource_reliable = 1;
748}
749#else
750static inline void check_geode_tsc_reliable(void) { }
751#endif 762#endif
763 if (boot_cpu_has(X86_FEATURE_TSC_RELIABLE))
764 tsc_clocksource_reliable = 1;
765}
752 766
753/* 767/*
754 * Make an educated guess if the TSC is trustworthy and synchronized 768 * Make an educated guess if the TSC is trustworthy and synchronized
@@ -783,6 +797,8 @@ static void __init init_tsc_clocksource(void)
783{ 797{
784 clocksource_tsc.mult = clocksource_khz2mult(tsc_khz, 798 clocksource_tsc.mult = clocksource_khz2mult(tsc_khz,
785 clocksource_tsc.shift); 799 clocksource_tsc.shift);
800 if (tsc_clocksource_reliable)
801 clocksource_tsc.flags &= ~CLOCK_SOURCE_MUST_VERIFY;
786 /* lower the rating if we already know its unstable: */ 802 /* lower the rating if we already know its unstable: */
787 if (check_tsc_unstable()) { 803 if (check_tsc_unstable()) {
788 clocksource_tsc.rating = 0; 804 clocksource_tsc.rating = 0;
@@ -843,7 +859,7 @@ void __init tsc_init(void)
843 if (unsynchronized_tsc()) 859 if (unsynchronized_tsc())
844 mark_tsc_unstable("TSCs unsynchronized"); 860 mark_tsc_unstable("TSCs unsynchronized");
845 861
846 check_geode_tsc_reliable(); 862 check_system_tsc_reliable();
847 init_tsc_clocksource(); 863 init_tsc_clocksource();
848} 864}
849 865
diff --git a/arch/x86/kernel/tsc_sync.c b/arch/x86/kernel/tsc_sync.c
index 1c0dfbca87c1..bf36328f6ef9 100644
--- a/arch/x86/kernel/tsc_sync.c
+++ b/arch/x86/kernel/tsc_sync.c
@@ -112,6 +112,12 @@ void __cpuinit check_tsc_sync_source(int cpu)
112 if (unsynchronized_tsc()) 112 if (unsynchronized_tsc())
113 return; 113 return;
114 114
115 if (boot_cpu_has(X86_FEATURE_TSC_RELIABLE)) {
116 printk(KERN_INFO
117 "Skipping synchronization checks as TSC is reliable.\n");
118 return;
119 }
120
115 printk(KERN_INFO "checking TSC synchronization [CPU#%d -> CPU#%d]:", 121 printk(KERN_INFO "checking TSC synchronization [CPU#%d -> CPU#%d]:",
116 smp_processor_id(), cpu); 122 smp_processor_id(), cpu);
117 123
@@ -165,7 +171,7 @@ void __cpuinit check_tsc_sync_target(void)
165{ 171{
166 int cpus = 2; 172 int cpus = 2;
167 173
168 if (unsynchronized_tsc()) 174 if (unsynchronized_tsc() || boot_cpu_has(X86_FEATURE_TSC_RELIABLE))
169 return; 175 return;
170 176
171 /* 177 /*
diff --git a/arch/x86/kernel/vmi_32.c b/arch/x86/kernel/vmi_32.c
index 22fd6577156a..23206ba16874 100644
--- a/arch/x86/kernel/vmi_32.c
+++ b/arch/x86/kernel/vmi_32.c
@@ -266,109 +266,6 @@ static void vmi_nop(void)
266{ 266{
267} 267}
268 268
269#ifdef CONFIG_DEBUG_PAGE_TYPE
270
271#ifdef CONFIG_X86_PAE
272#define MAX_BOOT_PTS (2048+4+1)
273#else
274#define MAX_BOOT_PTS (1024+1)
275#endif
276
277/*
278 * During boot, mem_map is not yet available in paging_init, so stash
279 * all the boot page allocations here.
280 */
281static struct {
282 u32 pfn;
283 int type;
284} boot_page_allocations[MAX_BOOT_PTS];
285static int num_boot_page_allocations;
286static int boot_allocations_applied;
287
288void vmi_apply_boot_page_allocations(void)
289{
290 int i;
291 BUG_ON(!mem_map);
292 for (i = 0; i < num_boot_page_allocations; i++) {
293 struct page *page = pfn_to_page(boot_page_allocations[i].pfn);
294 page->type = boot_page_allocations[i].type;
295 page->type = boot_page_allocations[i].type &
296 ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
297 }
298 boot_allocations_applied = 1;
299}
300
301static void record_page_type(u32 pfn, int type)
302{
303 BUG_ON(num_boot_page_allocations >= MAX_BOOT_PTS);
304 boot_page_allocations[num_boot_page_allocations].pfn = pfn;
305 boot_page_allocations[num_boot_page_allocations].type = type;
306 num_boot_page_allocations++;
307}
308
309static void check_zeroed_page(u32 pfn, int type, struct page *page)
310{
311 u32 *ptr;
312 int i;
313 int limit = PAGE_SIZE / sizeof(int);
314
315 if (page_address(page))
316 ptr = (u32 *)page_address(page);
317 else
318 ptr = (u32 *)__va(pfn << PAGE_SHIFT);
319 /*
320 * When cloning the root in non-PAE mode, only the userspace
321 * pdes need to be zeroed.
322 */
323 if (type & VMI_PAGE_CLONE)
324 limit = KERNEL_PGD_BOUNDARY;
325 for (i = 0; i < limit; i++)
326 BUG_ON(ptr[i]);
327}
328
329/*
330 * We stash the page type into struct page so we can verify the page
331 * types are used properly.
332 */
333static void vmi_set_page_type(u32 pfn, int type)
334{
335 /* PAE can have multiple roots per page - don't track */
336 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
337 return;
338
339 if (boot_allocations_applied) {
340 struct page *page = pfn_to_page(pfn);
341 if (type != VMI_PAGE_NORMAL)
342 BUG_ON(page->type);
343 else
344 BUG_ON(page->type == VMI_PAGE_NORMAL);
345 page->type = type & ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
346 if (type & VMI_PAGE_ZEROED)
347 check_zeroed_page(pfn, type, page);
348 } else {
349 record_page_type(pfn, type);
350 }
351}
352
353static void vmi_check_page_type(u32 pfn, int type)
354{
355 /* PAE can have multiple roots per page - skip checks */
356 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
357 return;
358
359 type &= ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
360 if (boot_allocations_applied) {
361 struct page *page = pfn_to_page(pfn);
362 BUG_ON((page->type ^ type) & VMI_PAGE_PAE);
363 BUG_ON(type == VMI_PAGE_NORMAL && page->type);
364 BUG_ON((type & page->type) == 0);
365 }
366}
367#else
368#define vmi_set_page_type(p,t) do { } while (0)
369#define vmi_check_page_type(p,t) do { } while (0)
370#endif
371
372#ifdef CONFIG_HIGHPTE 269#ifdef CONFIG_HIGHPTE
373static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type) 270static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
374{ 271{
@@ -395,7 +292,6 @@ static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
395 292
396static void vmi_allocate_pte(struct mm_struct *mm, unsigned long pfn) 293static void vmi_allocate_pte(struct mm_struct *mm, unsigned long pfn)
397{ 294{
398 vmi_set_page_type(pfn, VMI_PAGE_L1);
399 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0); 295 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
400} 296}
401 297
@@ -406,27 +302,22 @@ static void vmi_allocate_pmd(struct mm_struct *mm, unsigned long pfn)
406 * It is called only for swapper_pg_dir, which already has 302 * It is called only for swapper_pg_dir, which already has
407 * data on it. 303 * data on it.
408 */ 304 */
409 vmi_set_page_type(pfn, VMI_PAGE_L2);
410 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0); 305 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
411} 306}
412 307
413static void vmi_allocate_pmd_clone(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count) 308static void vmi_allocate_pmd_clone(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count)
414{ 309{
415 vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE);
416 vmi_check_page_type(clonepfn, VMI_PAGE_L2);
417 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count); 310 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
418} 311}
419 312
420static void vmi_release_pte(unsigned long pfn) 313static void vmi_release_pte(unsigned long pfn)
421{ 314{
422 vmi_ops.release_page(pfn, VMI_PAGE_L1); 315 vmi_ops.release_page(pfn, VMI_PAGE_L1);
423 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
424} 316}
425 317
426static void vmi_release_pmd(unsigned long pfn) 318static void vmi_release_pmd(unsigned long pfn)
427{ 319{
428 vmi_ops.release_page(pfn, VMI_PAGE_L2); 320 vmi_ops.release_page(pfn, VMI_PAGE_L2);
429 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
430} 321}
431 322
432/* 323/*
@@ -450,26 +341,22 @@ static void vmi_release_pmd(unsigned long pfn)
450 341
451static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 342static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
452{ 343{
453 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
454 vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); 344 vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
455} 345}
456 346
457static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 347static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
458{ 348{
459 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
460 vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0)); 349 vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0));
461} 350}
462 351
463static void vmi_set_pte(pte_t *ptep, pte_t pte) 352static void vmi_set_pte(pte_t *ptep, pte_t pte)
464{ 353{
465 /* XXX because of set_pmd_pte, this can be called on PT or PD layers */ 354 /* XXX because of set_pmd_pte, this can be called on PT or PD layers */
466 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE | VMI_PAGE_PD);
467 vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT); 355 vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT);
468} 356}
469 357
470static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte) 358static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
471{ 359{
472 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
473 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); 360 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
474} 361}
475 362
@@ -477,10 +364,8 @@ static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
477{ 364{
478#ifdef CONFIG_X86_PAE 365#ifdef CONFIG_X86_PAE
479 const pte_t pte = { .pte = pmdval.pmd }; 366 const pte_t pte = { .pte = pmdval.pmd };
480 vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PMD);
481#else 367#else
482 const pte_t pte = { pmdval.pud.pgd.pgd }; 368 const pte_t pte = { pmdval.pud.pgd.pgd };
483 vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PGD);
484#endif 369#endif
485 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD); 370 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
486} 371}
@@ -502,7 +387,6 @@ static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
502 387
503static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte) 388static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
504{ 389{
505 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
506 vmi_ops.set_pte(pte, ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 1)); 390 vmi_ops.set_pte(pte, ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 1));
507} 391}
508 392
@@ -510,21 +394,18 @@ static void vmi_set_pud(pud_t *pudp, pud_t pudval)
510{ 394{
511 /* Um, eww */ 395 /* Um, eww */
512 const pte_t pte = { .pte = pudval.pgd.pgd }; 396 const pte_t pte = { .pte = pudval.pgd.pgd };
513 vmi_check_page_type(__pa(pudp) >> PAGE_SHIFT, VMI_PAGE_PGD);
514 vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP); 397 vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP);
515} 398}
516 399
517static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) 400static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
518{ 401{
519 const pte_t pte = { .pte = 0 }; 402 const pte_t pte = { .pte = 0 };
520 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
521 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); 403 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
522} 404}
523 405
524static void vmi_pmd_clear(pmd_t *pmd) 406static void vmi_pmd_clear(pmd_t *pmd)
525{ 407{
526 const pte_t pte = { .pte = 0 }; 408 const pte_t pte = { .pte = 0 };
527 vmi_check_page_type(__pa(pmd) >> PAGE_SHIFT, VMI_PAGE_PMD);
528 vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD); 409 vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD);
529} 410}
530#endif 411#endif
diff --git a/arch/x86/kernel/vmlinux_32.lds.S b/arch/x86/kernel/vmlinux_32.lds.S
index a9b8560adbc2..82c67559dde7 100644
--- a/arch/x86/kernel/vmlinux_32.lds.S
+++ b/arch/x86/kernel/vmlinux_32.lds.S
@@ -44,6 +44,7 @@ SECTIONS
44 SCHED_TEXT 44 SCHED_TEXT
45 LOCK_TEXT 45 LOCK_TEXT
46 KPROBES_TEXT 46 KPROBES_TEXT
47 IRQENTRY_TEXT
47 *(.fixup) 48 *(.fixup)
48 *(.gnu.warning) 49 *(.gnu.warning)
49 _etext = .; /* End of text section */ 50 _etext = .; /* End of text section */
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index 46e05447405b..1a614c0e6bef 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -35,6 +35,7 @@ SECTIONS
35 SCHED_TEXT 35 SCHED_TEXT
36 LOCK_TEXT 36 LOCK_TEXT
37 KPROBES_TEXT 37 KPROBES_TEXT
38 IRQENTRY_TEXT
38 *(.fixup) 39 *(.fixup)
39 *(.gnu.warning) 40 *(.gnu.warning)
40 _etext = .; /* End of text section */ 41 _etext = .; /* End of text section */
diff --git a/arch/x86/kernel/vsyscall_64.c b/arch/x86/kernel/vsyscall_64.c
index 0b8b6690a86d..44153afc9067 100644
--- a/arch/x86/kernel/vsyscall_64.c
+++ b/arch/x86/kernel/vsyscall_64.c
@@ -17,6 +17,9 @@
17 * want per guest time just set the kernel.vsyscall64 sysctl to 0. 17 * want per guest time just set the kernel.vsyscall64 sysctl to 0.
18 */ 18 */
19 19
20/* Disable profiling for userspace code: */
21#define DISABLE_BRANCH_PROFILING
22
20#include <linux/time.h> 23#include <linux/time.h>
21#include <linux/init.h> 24#include <linux/init.h>
22#include <linux/kernel.h> 25#include <linux/kernel.h>
@@ -128,7 +131,16 @@ static __always_inline void do_vgettimeofday(struct timeval * tv)
128 gettimeofday(tv,NULL); 131 gettimeofday(tv,NULL);
129 return; 132 return;
130 } 133 }
134
135 /*
136 * Surround the RDTSC by barriers, to make sure it's not
137 * speculated to outside the seqlock critical section and
138 * does not cause time warps:
139 */
140 rdtsc_barrier();
131 now = vread(); 141 now = vread();
142 rdtsc_barrier();
143
132 base = __vsyscall_gtod_data.clock.cycle_last; 144 base = __vsyscall_gtod_data.clock.cycle_last;
133 mask = __vsyscall_gtod_data.clock.mask; 145 mask = __vsyscall_gtod_data.clock.mask;
134 mult = __vsyscall_gtod_data.clock.mult; 146 mult = __vsyscall_gtod_data.clock.mult;
generated by cgit v1.2.2 (git 2.25.0) at 2026-01-06 16:08:36 -0500