aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-06-11 17:01:07 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-06-11 17:01:07 -0400
commit8a1ca8cedd108c8e76a6ab34079d0bbb4f244799 (patch)
tree636c715524f1718599209cc289908ea44b6cb859 /arch/x86
parentb640f042faa2a2fad6464f259a8afec06e2f6386 (diff)
parent940010c5a314a7bd9b498593bc6ba1718ac5aec5 (diff)
Merge branch 'perfcounters-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'perfcounters-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (574 commits) perf_counter: Turn off by default perf_counter: Add counter->id to the throttle event perf_counter: Better align code perf_counter: Rename L2 to LL cache perf_counter: Standardize event names perf_counter: Rename enums perf_counter tools: Clean up u64 usage perf_counter: Rename perf_counter_limit sysctl perf_counter: More paranoia settings perf_counter: powerpc: Implement generalized cache events for POWER processors perf_counters: powerpc: Add support for POWER7 processors perf_counter: Accurate period data perf_counter: Introduce struct for sample data perf_counter tools: Normalize data using per sample period data perf_counter: Annotate exit ctx recursion perf_counter tools: Propagate signals properly perf_counter tools: Small frequency related fixes perf_counter: More aggressive frequency adjustment perf_counter/x86: Fix the model number of Intel Core2 processors perf_counter, x86: Correct some event and umask values for Intel processors ...
Diffstat (limited to 'arch/x86')
-rw-r--r--arch/x86/Kconfig1
-rw-r--r--arch/x86/ia32/ia32entry.S3
-rw-r--r--arch/x86/include/asm/atomic_32.h236
-rw-r--r--arch/x86/include/asm/entry_arch.h2
-rw-r--r--arch/x86/include/asm/hardirq.h2
-rw-r--r--arch/x86/include/asm/hw_irq.h2
-rw-r--r--arch/x86/include/asm/intel_arch_perfmon.h31
-rw-r--r--arch/x86/include/asm/irq_vectors.h8
-rw-r--r--arch/x86/include/asm/perf_counter.h100
-rw-r--r--arch/x86/include/asm/unistd_32.h1
-rw-r--r--arch/x86/include/asm/unistd_64.h3
-rw-r--r--arch/x86/kernel/apic/apic.c3
-rw-r--r--arch/x86/kernel/cpu/Makefile12
-rw-r--r--arch/x86/kernel/cpu/common.c2
-rw-r--r--arch/x86/kernel/cpu/perf_counter.c1704
-rw-r--r--arch/x86/kernel/cpu/perfctr-watchdog.c4
-rw-r--r--arch/x86/kernel/entry_64.S5
-rw-r--r--arch/x86/kernel/irq.c10
-rw-r--r--arch/x86/kernel/irqinit.c15
-rw-r--r--arch/x86/kernel/signal.c1
-rw-r--r--arch/x86/kernel/syscall_table_32.S1
-rw-r--r--arch/x86/kernel/traps.c12
-rw-r--r--arch/x86/mm/fault.c12
-rw-r--r--arch/x86/oprofile/nmi_int.c7
-rw-r--r--arch/x86/oprofile/op_model_ppro.c10
-rw-r--r--arch/x86/vdso/vdso32-setup.c6
-rw-r--r--arch/x86/vdso/vma.c7
27 files changed, 2130 insertions, 70 deletions
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index aafae3b140de..68f5578fe38e 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -739,6 +739,7 @@ config X86_UP_IOAPIC
739config X86_LOCAL_APIC 739config X86_LOCAL_APIC
740 def_bool y 740 def_bool y
741 depends on X86_64 || SMP || X86_32_NON_STANDARD || X86_UP_APIC 741 depends on X86_64 || SMP || X86_32_NON_STANDARD || X86_UP_APIC
742 select HAVE_PERF_COUNTERS if (!M386 && !M486)
742 743
743config X86_IO_APIC 744config X86_IO_APIC
744 def_bool y 745 def_bool y
diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index dcef387ddc36..e590261ba059 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -825,10 +825,11 @@ ia32_sys_call_table:
825 .quad compat_sys_signalfd4 825 .quad compat_sys_signalfd4
826 .quad sys_eventfd2 826 .quad sys_eventfd2
827 .quad sys_epoll_create1 827 .quad sys_epoll_create1
828 .quad sys_dup3 /* 330 */ 828 .quad sys_dup3 /* 330 */
829 .quad sys_pipe2 829 .quad sys_pipe2
830 .quad sys_inotify_init1 830 .quad sys_inotify_init1
831 .quad compat_sys_preadv 831 .quad compat_sys_preadv
832 .quad compat_sys_pwritev 832 .quad compat_sys_pwritev
833 .quad compat_sys_rt_tgsigqueueinfo /* 335 */ 833 .quad compat_sys_rt_tgsigqueueinfo /* 335 */
834 .quad sys_perf_counter_open
834ia32_syscall_end: 835ia32_syscall_end:
diff --git a/arch/x86/include/asm/atomic_32.h b/arch/x86/include/asm/atomic_32.h
index 85b46fba4229..aff9f1fcdcd7 100644
--- a/arch/x86/include/asm/atomic_32.h
+++ b/arch/x86/include/asm/atomic_32.h
@@ -247,5 +247,241 @@ static inline int atomic_add_unless(atomic_t *v, int a, int u)
247#define smp_mb__before_atomic_inc() barrier() 247#define smp_mb__before_atomic_inc() barrier()
248#define smp_mb__after_atomic_inc() barrier() 248#define smp_mb__after_atomic_inc() barrier()
249 249
250/* An 64bit atomic type */
251
252typedef struct {
253 unsigned long long counter;
254} atomic64_t;
255
256#define ATOMIC64_INIT(val) { (val) }
257
258/**
259 * atomic64_read - read atomic64 variable
260 * @v: pointer of type atomic64_t
261 *
262 * Atomically reads the value of @v.
263 * Doesn't imply a read memory barrier.
264 */
265#define __atomic64_read(ptr) ((ptr)->counter)
266
267static inline unsigned long long
268cmpxchg8b(unsigned long long *ptr, unsigned long long old, unsigned long long new)
269{
270 asm volatile(
271
272 LOCK_PREFIX "cmpxchg8b (%[ptr])\n"
273
274 : "=A" (old)
275
276 : [ptr] "D" (ptr),
277 "A" (old),
278 "b" (ll_low(new)),
279 "c" (ll_high(new))
280
281 : "memory");
282
283 return old;
284}
285
286static inline unsigned long long
287atomic64_cmpxchg(atomic64_t *ptr, unsigned long long old_val,
288 unsigned long long new_val)
289{
290 return cmpxchg8b(&ptr->counter, old_val, new_val);
291}
292
293/**
294 * atomic64_xchg - xchg atomic64 variable
295 * @ptr: pointer to type atomic64_t
296 * @new_val: value to assign
297 * @old_val: old value that was there
298 *
299 * Atomically xchgs the value of @ptr to @new_val and returns
300 * the old value.
301 */
302
303static inline unsigned long long
304atomic64_xchg(atomic64_t *ptr, unsigned long long new_val)
305{
306 unsigned long long old_val;
307
308 do {
309 old_val = atomic_read(ptr);
310 } while (atomic64_cmpxchg(ptr, old_val, new_val) != old_val);
311
312 return old_val;
313}
314
315/**
316 * atomic64_set - set atomic64 variable
317 * @ptr: pointer to type atomic64_t
318 * @new_val: value to assign
319 *
320 * Atomically sets the value of @ptr to @new_val.
321 */
322static inline void atomic64_set(atomic64_t *ptr, unsigned long long new_val)
323{
324 atomic64_xchg(ptr, new_val);
325}
326
327/**
328 * atomic64_read - read atomic64 variable
329 * @ptr: pointer to type atomic64_t
330 *
331 * Atomically reads the value of @ptr and returns it.
332 */
333static inline unsigned long long atomic64_read(atomic64_t *ptr)
334{
335 unsigned long long curr_val;
336
337 do {
338 curr_val = __atomic64_read(ptr);
339 } while (atomic64_cmpxchg(ptr, curr_val, curr_val) != curr_val);
340
341 return curr_val;
342}
343
344/**
345 * atomic64_add_return - add and return
346 * @delta: integer value to add
347 * @ptr: pointer to type atomic64_t
348 *
349 * Atomically adds @delta to @ptr and returns @delta + *@ptr
350 */
351static inline unsigned long long
352atomic64_add_return(unsigned long long delta, atomic64_t *ptr)
353{
354 unsigned long long old_val, new_val;
355
356 do {
357 old_val = atomic_read(ptr);
358 new_val = old_val + delta;
359
360 } while (atomic64_cmpxchg(ptr, old_val, new_val) != old_val);
361
362 return new_val;
363}
364
365static inline long atomic64_sub_return(unsigned long long delta, atomic64_t *ptr)
366{
367 return atomic64_add_return(-delta, ptr);
368}
369
370static inline long atomic64_inc_return(atomic64_t *ptr)
371{
372 return atomic64_add_return(1, ptr);
373}
374
375static inline long atomic64_dec_return(atomic64_t *ptr)
376{
377 return atomic64_sub_return(1, ptr);
378}
379
380/**
381 * atomic64_add - add integer to atomic64 variable
382 * @delta: integer value to add
383 * @ptr: pointer to type atomic64_t
384 *
385 * Atomically adds @delta to @ptr.
386 */
387static inline void atomic64_add(unsigned long long delta, atomic64_t *ptr)
388{
389 atomic64_add_return(delta, ptr);
390}
391
392/**
393 * atomic64_sub - subtract the atomic64 variable
394 * @delta: integer value to subtract
395 * @ptr: pointer to type atomic64_t
396 *
397 * Atomically subtracts @delta from @ptr.
398 */
399static inline void atomic64_sub(unsigned long long delta, atomic64_t *ptr)
400{
401 atomic64_add(-delta, ptr);
402}
403
404/**
405 * atomic64_sub_and_test - subtract value from variable and test result
406 * @delta: integer value to subtract
407 * @ptr: pointer to type atomic64_t
408 *
409 * Atomically subtracts @delta from @ptr and returns
410 * true if the result is zero, or false for all
411 * other cases.
412 */
413static inline int
414atomic64_sub_and_test(unsigned long long delta, atomic64_t *ptr)
415{
416 unsigned long long old_val = atomic64_sub_return(delta, ptr);
417
418 return old_val == 0;
419}
420
421/**
422 * atomic64_inc - increment atomic64 variable
423 * @ptr: pointer to type atomic64_t
424 *
425 * Atomically increments @ptr by 1.
426 */
427static inline void atomic64_inc(atomic64_t *ptr)
428{
429 atomic64_add(1, ptr);
430}
431
432/**
433 * atomic64_dec - decrement atomic64 variable
434 * @ptr: pointer to type atomic64_t
435 *
436 * Atomically decrements @ptr by 1.
437 */
438static inline void atomic64_dec(atomic64_t *ptr)
439{
440 atomic64_sub(1, ptr);
441}
442
443/**
444 * atomic64_dec_and_test - decrement and test
445 * @ptr: pointer to type atomic64_t
446 *
447 * Atomically decrements @ptr by 1 and
448 * returns true if the result is 0, or false for all other
449 * cases.
450 */
451static inline int atomic64_dec_and_test(atomic64_t *ptr)
452{
453 return atomic64_sub_and_test(1, ptr);
454}
455
456/**
457 * atomic64_inc_and_test - increment and test
458 * @ptr: pointer to type atomic64_t
459 *
460 * Atomically increments @ptr by 1
461 * and returns true if the result is zero, or false for all
462 * other cases.
463 */
464static inline int atomic64_inc_and_test(atomic64_t *ptr)
465{
466 return atomic64_sub_and_test(-1, ptr);
467}
468
469/**
470 * atomic64_add_negative - add and test if negative
471 * @delta: integer value to add
472 * @ptr: pointer to type atomic64_t
473 *
474 * Atomically adds @delta to @ptr and returns true
475 * if the result is negative, or false when
476 * result is greater than or equal to zero.
477 */
478static inline int
479atomic64_add_negative(unsigned long long delta, atomic64_t *ptr)
480{
481 long long old_val = atomic64_add_return(delta, ptr);
482
483 return old_val < 0;
484}
485
250#include <asm-generic/atomic.h> 486#include <asm-generic/atomic.h>
251#endif /* _ASM_X86_ATOMIC_32_H */ 487#endif /* _ASM_X86_ATOMIC_32_H */
diff --git a/arch/x86/include/asm/entry_arch.h b/arch/x86/include/asm/entry_arch.h
index c2e6bedaf258..d750a10ccad6 100644
--- a/arch/x86/include/asm/entry_arch.h
+++ b/arch/x86/include/asm/entry_arch.h
@@ -49,7 +49,7 @@ BUILD_INTERRUPT(error_interrupt,ERROR_APIC_VECTOR)
49BUILD_INTERRUPT(spurious_interrupt,SPURIOUS_APIC_VECTOR) 49BUILD_INTERRUPT(spurious_interrupt,SPURIOUS_APIC_VECTOR)
50 50
51#ifdef CONFIG_PERF_COUNTERS 51#ifdef CONFIG_PERF_COUNTERS
52BUILD_INTERRUPT(perf_counter_interrupt, LOCAL_PERF_VECTOR) 52BUILD_INTERRUPT(perf_pending_interrupt, LOCAL_PENDING_VECTOR)
53#endif 53#endif
54 54
55#ifdef CONFIG_X86_MCE_P4THERMAL 55#ifdef CONFIG_X86_MCE_P4THERMAL
diff --git a/arch/x86/include/asm/hardirq.h b/arch/x86/include/asm/hardirq.h
index 37555e52f980..9ebc5c255032 100644
--- a/arch/x86/include/asm/hardirq.h
+++ b/arch/x86/include/asm/hardirq.h
@@ -13,6 +13,8 @@ typedef struct {
13 unsigned int irq_spurious_count; 13 unsigned int irq_spurious_count;
14#endif 14#endif
15 unsigned int generic_irqs; /* arch dependent */ 15 unsigned int generic_irqs; /* arch dependent */
16 unsigned int apic_perf_irqs;
17 unsigned int apic_pending_irqs;
16#ifdef CONFIG_SMP 18#ifdef CONFIG_SMP
17 unsigned int irq_resched_count; 19 unsigned int irq_resched_count;
18 unsigned int irq_call_count; 20 unsigned int irq_call_count;
diff --git a/arch/x86/include/asm/hw_irq.h b/arch/x86/include/asm/hw_irq.h
index 3bd1777a4c8b..6df45f639666 100644
--- a/arch/x86/include/asm/hw_irq.h
+++ b/arch/x86/include/asm/hw_irq.h
@@ -29,6 +29,8 @@
29extern void apic_timer_interrupt(void); 29extern void apic_timer_interrupt(void);
30extern void generic_interrupt(void); 30extern void generic_interrupt(void);
31extern void error_interrupt(void); 31extern void error_interrupt(void);
32extern void perf_pending_interrupt(void);
33
32extern void spurious_interrupt(void); 34extern void spurious_interrupt(void);
33extern void thermal_interrupt(void); 35extern void thermal_interrupt(void);
34extern void reschedule_interrupt(void); 36extern void reschedule_interrupt(void);
diff --git a/arch/x86/include/asm/intel_arch_perfmon.h b/arch/x86/include/asm/intel_arch_perfmon.h
deleted file mode 100644
index fa0fd068bc2e..000000000000
--- a/arch/x86/include/asm/intel_arch_perfmon.h
+++ /dev/null
@@ -1,31 +0,0 @@
1#ifndef _ASM_X86_INTEL_ARCH_PERFMON_H
2#define _ASM_X86_INTEL_ARCH_PERFMON_H
3
4#define MSR_ARCH_PERFMON_PERFCTR0 0xc1
5#define MSR_ARCH_PERFMON_PERFCTR1 0xc2
6
7#define MSR_ARCH_PERFMON_EVENTSEL0 0x186
8#define MSR_ARCH_PERFMON_EVENTSEL1 0x187
9
10#define ARCH_PERFMON_EVENTSEL0_ENABLE (1 << 22)
11#define ARCH_PERFMON_EVENTSEL_INT (1 << 20)
12#define ARCH_PERFMON_EVENTSEL_OS (1 << 17)
13#define ARCH_PERFMON_EVENTSEL_USR (1 << 16)
14
15#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL (0x3c)
16#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8)
17#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX (0)
18#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \
19 (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX))
20
21union cpuid10_eax {
22 struct {
23 unsigned int version_id:8;
24 unsigned int num_counters:8;
25 unsigned int bit_width:8;
26 unsigned int mask_length:8;
27 } split;
28 unsigned int full;
29};
30
31#endif /* _ASM_X86_INTEL_ARCH_PERFMON_H */
diff --git a/arch/x86/include/asm/irq_vectors.h b/arch/x86/include/asm/irq_vectors.h
index 910b5a3d6751..e997be98c9b9 100644
--- a/arch/x86/include/asm/irq_vectors.h
+++ b/arch/x86/include/asm/irq_vectors.h
@@ -108,14 +108,14 @@
108#define LOCAL_TIMER_VECTOR 0xef 108#define LOCAL_TIMER_VECTOR 0xef
109 109
110/* 110/*
111 * Performance monitoring interrupt vector: 111 * Generic system vector for platform specific use
112 */ 112 */
113#define LOCAL_PERF_VECTOR 0xee 113#define GENERIC_INTERRUPT_VECTOR 0xed
114 114
115/* 115/*
116 * Generic system vector for platform specific use 116 * Performance monitoring pending work vector:
117 */ 117 */
118#define GENERIC_INTERRUPT_VECTOR 0xed 118#define LOCAL_PENDING_VECTOR 0xec
119 119
120/* 120/*
121 * First APIC vector available to drivers: (vectors 0x30-0xee) we 121 * First APIC vector available to drivers: (vectors 0x30-0xee) we
diff --git a/arch/x86/include/asm/perf_counter.h b/arch/x86/include/asm/perf_counter.h
new file mode 100644
index 000000000000..876ed97147b3
--- /dev/null
+++ b/arch/x86/include/asm/perf_counter.h
@@ -0,0 +1,100 @@
1#ifndef _ASM_X86_PERF_COUNTER_H
2#define _ASM_X86_PERF_COUNTER_H
3
4/*
5 * Performance counter hw details:
6 */
7
8#define X86_PMC_MAX_GENERIC 8
9#define X86_PMC_MAX_FIXED 3
10
11#define X86_PMC_IDX_GENERIC 0
12#define X86_PMC_IDX_FIXED 32
13#define X86_PMC_IDX_MAX 64
14
15#define MSR_ARCH_PERFMON_PERFCTR0 0xc1
16#define MSR_ARCH_PERFMON_PERFCTR1 0xc2
17
18#define MSR_ARCH_PERFMON_EVENTSEL0 0x186
19#define MSR_ARCH_PERFMON_EVENTSEL1 0x187
20
21#define ARCH_PERFMON_EVENTSEL0_ENABLE (1 << 22)
22#define ARCH_PERFMON_EVENTSEL_INT (1 << 20)
23#define ARCH_PERFMON_EVENTSEL_OS (1 << 17)
24#define ARCH_PERFMON_EVENTSEL_USR (1 << 16)
25
26/*
27 * Includes eventsel and unit mask as well:
28 */
29#define ARCH_PERFMON_EVENT_MASK 0xffff
30
31#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c
32#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8)
33#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0
34#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \
35 (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX))
36
37#define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6
38
39/*
40 * Intel "Architectural Performance Monitoring" CPUID
41 * detection/enumeration details:
42 */
43union cpuid10_eax {
44 struct {
45 unsigned int version_id:8;
46 unsigned int num_counters:8;
47 unsigned int bit_width:8;
48 unsigned int mask_length:8;
49 } split;
50 unsigned int full;
51};
52
53union cpuid10_edx {
54 struct {
55 unsigned int num_counters_fixed:4;
56 unsigned int reserved:28;
57 } split;
58 unsigned int full;
59};
60
61
62/*
63 * Fixed-purpose performance counters:
64 */
65
66/*
67 * All 3 fixed-mode PMCs are configured via this single MSR:
68 */
69#define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d
70
71/*
72 * The counts are available in three separate MSRs:
73 */
74
75/* Instr_Retired.Any: */
76#define MSR_ARCH_PERFMON_FIXED_CTR0 0x309
77#define X86_PMC_IDX_FIXED_INSTRUCTIONS (X86_PMC_IDX_FIXED + 0)
78
79/* CPU_CLK_Unhalted.Core: */
80#define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a
81#define X86_PMC_IDX_FIXED_CPU_CYCLES (X86_PMC_IDX_FIXED + 1)
82
83/* CPU_CLK_Unhalted.Ref: */
84#define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b
85#define X86_PMC_IDX_FIXED_BUS_CYCLES (X86_PMC_IDX_FIXED + 2)
86
87extern void set_perf_counter_pending(void);
88
89#define clear_perf_counter_pending() do { } while (0)
90#define test_perf_counter_pending() (0)
91
92#ifdef CONFIG_PERF_COUNTERS
93extern void init_hw_perf_counters(void);
94extern void perf_counters_lapic_init(void);
95#else
96static inline void init_hw_perf_counters(void) { }
97static inline void perf_counters_lapic_init(void) { }
98#endif
99
100#endif /* _ASM_X86_PERF_COUNTER_H */
diff --git a/arch/x86/include/asm/unistd_32.h b/arch/x86/include/asm/unistd_32.h
index 708dae61262d..732a30706153 100644
--- a/arch/x86/include/asm/unistd_32.h
+++ b/arch/x86/include/asm/unistd_32.h
@@ -341,6 +341,7 @@
341#define __NR_preadv 333 341#define __NR_preadv 333
342#define __NR_pwritev 334 342#define __NR_pwritev 334
343#define __NR_rt_tgsigqueueinfo 335 343#define __NR_rt_tgsigqueueinfo 335
344#define __NR_perf_counter_open 336
344 345
345#ifdef __KERNEL__ 346#ifdef __KERNEL__
346 347
diff --git a/arch/x86/include/asm/unistd_64.h b/arch/x86/include/asm/unistd_64.h
index 4e2b05404400..900e1617e672 100644
--- a/arch/x86/include/asm/unistd_64.h
+++ b/arch/x86/include/asm/unistd_64.h
@@ -659,7 +659,8 @@ __SYSCALL(__NR_preadv, sys_preadv)
659__SYSCALL(__NR_pwritev, sys_pwritev) 659__SYSCALL(__NR_pwritev, sys_pwritev)
660#define __NR_rt_tgsigqueueinfo 297 660#define __NR_rt_tgsigqueueinfo 297
661__SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo) 661__SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo)
662 662#define __NR_perf_counter_open 298
663__SYSCALL(__NR_perf_counter_open, sys_perf_counter_open)
663 664
664#ifndef __NO_STUBS 665#ifndef __NO_STUBS
665#define __ARCH_WANT_OLD_READDIR 666#define __ARCH_WANT_OLD_READDIR
diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
index a4c9cf0bf70b..076d3881f3da 100644
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -14,6 +14,7 @@
14 * Mikael Pettersson : PM converted to driver model. 14 * Mikael Pettersson : PM converted to driver model.
15 */ 15 */
16 16
17#include <linux/perf_counter.h>
17#include <linux/kernel_stat.h> 18#include <linux/kernel_stat.h>
18#include <linux/mc146818rtc.h> 19#include <linux/mc146818rtc.h>
19#include <linux/acpi_pmtmr.h> 20#include <linux/acpi_pmtmr.h>
@@ -34,6 +35,7 @@
34#include <linux/smp.h> 35#include <linux/smp.h>
35#include <linux/mm.h> 36#include <linux/mm.h>
36 37
38#include <asm/perf_counter.h>
37#include <asm/pgalloc.h> 39#include <asm/pgalloc.h>
38#include <asm/atomic.h> 40#include <asm/atomic.h>
39#include <asm/mpspec.h> 41#include <asm/mpspec.h>
@@ -1187,6 +1189,7 @@ void __cpuinit setup_local_APIC(void)
1187 apic_write(APIC_ESR, 0); 1189 apic_write(APIC_ESR, 0);
1188 } 1190 }
1189#endif 1191#endif
1192 perf_counters_lapic_init();
1190 1193
1191 preempt_disable(); 1194 preempt_disable();
1192 1195
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index 4e242f9a06e4..3efcb2b96a15 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -1,5 +1,5 @@
1# 1#
2# Makefile for x86-compatible CPU details and quirks 2# Makefile for x86-compatible CPU details, features and quirks
3# 3#
4 4
5# Don't trace early stages of a secondary CPU boot 5# Don't trace early stages of a secondary CPU boot
@@ -23,11 +23,13 @@ obj-$(CONFIG_CPU_SUP_CENTAUR) += centaur.o
23obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o 23obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o
24obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o 24obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o
25 25
26obj-$(CONFIG_X86_MCE) += mcheck/ 26obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o
27obj-$(CONFIG_MTRR) += mtrr/
28obj-$(CONFIG_CPU_FREQ) += cpufreq/
29 27
30obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o 28obj-$(CONFIG_X86_MCE) += mcheck/
29obj-$(CONFIG_MTRR) += mtrr/
30obj-$(CONFIG_CPU_FREQ) += cpufreq/
31
32obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o
31 33
32quiet_cmd_mkcapflags = MKCAP $@ 34quiet_cmd_mkcapflags = MKCAP $@
33 cmd_mkcapflags = $(PERL) $(srctree)/$(src)/mkcapflags.pl $< $@ 35 cmd_mkcapflags = $(PERL) $(srctree)/$(src)/mkcapflags.pl $< $@
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index b0517aa2bd3b..3ffdcfa9abdf 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -13,6 +13,7 @@
13#include <linux/io.h> 13#include <linux/io.h>
14 14
15#include <asm/stackprotector.h> 15#include <asm/stackprotector.h>
16#include <asm/perf_counter.h>
16#include <asm/mmu_context.h> 17#include <asm/mmu_context.h>
17#include <asm/hypervisor.h> 18#include <asm/hypervisor.h>
18#include <asm/processor.h> 19#include <asm/processor.h>
@@ -874,6 +875,7 @@ void __init identify_boot_cpu(void)
874#else 875#else
875 vgetcpu_set_mode(); 876 vgetcpu_set_mode();
876#endif 877#endif
878 init_hw_perf_counters();
877} 879}
878 880
879void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) 881void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
new file mode 100644
index 000000000000..895c82e78455
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -0,0 +1,1704 @@
1/*
2 * Performance counter x86 architecture code
3 *
4 * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
5 * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
6 * Copyright (C) 2009 Jaswinder Singh Rajput
7 * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
8 * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
9 *
10 * For licencing details see kernel-base/COPYING
11 */
12
13#include <linux/perf_counter.h>
14#include <linux/capability.h>
15#include <linux/notifier.h>
16#include <linux/hardirq.h>
17#include <linux/kprobes.h>
18#include <linux/module.h>
19#include <linux/kdebug.h>
20#include <linux/sched.h>
21#include <linux/uaccess.h>
22
23#include <asm/apic.h>
24#include <asm/stacktrace.h>
25#include <asm/nmi.h>
26
27static u64 perf_counter_mask __read_mostly;
28
29struct cpu_hw_counters {
30 struct perf_counter *counters[X86_PMC_IDX_MAX];
31 unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
32 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
33 unsigned long interrupts;
34 int enabled;
35};
36
37/*
38 * struct x86_pmu - generic x86 pmu
39 */
40struct x86_pmu {
41 const char *name;
42 int version;
43 int (*handle_irq)(struct pt_regs *);
44 void (*disable_all)(void);
45 void (*enable_all)(void);
46 void (*enable)(struct hw_perf_counter *, int);
47 void (*disable)(struct hw_perf_counter *, int);
48 unsigned eventsel;
49 unsigned perfctr;
50 u64 (*event_map)(int);
51 u64 (*raw_event)(u64);
52 int max_events;
53 int num_counters;
54 int num_counters_fixed;
55 int counter_bits;
56 u64 counter_mask;
57 u64 max_period;
58 u64 intel_ctrl;
59};
60
61static struct x86_pmu x86_pmu __read_mostly;
62
63static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = {
64 .enabled = 1,
65};
66
67/*
68 * Intel PerfMon v3. Used on Core2 and later.
69 */
70static const u64 intel_perfmon_event_map[] =
71{
72 [PERF_COUNT_HW_CPU_CYCLES] = 0x003c,
73 [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
74 [PERF_COUNT_HW_CACHE_REFERENCES] = 0x4f2e,
75 [PERF_COUNT_HW_CACHE_MISSES] = 0x412e,
76 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
77 [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
78 [PERF_COUNT_HW_BUS_CYCLES] = 0x013c,
79};
80
81static u64 intel_pmu_event_map(int event)
82{
83 return intel_perfmon_event_map[event];
84}
85
86/*
87 * Generalized hw caching related event table, filled
88 * in on a per model basis. A value of 0 means
89 * 'not supported', -1 means 'event makes no sense on
90 * this CPU', any other value means the raw event
91 * ID.
92 */
93
94#define C(x) PERF_COUNT_HW_CACHE_##x
95
96static u64 __read_mostly hw_cache_event_ids
97 [PERF_COUNT_HW_CACHE_MAX]
98 [PERF_COUNT_HW_CACHE_OP_MAX]
99 [PERF_COUNT_HW_CACHE_RESULT_MAX];
100
101static const u64 nehalem_hw_cache_event_ids
102 [PERF_COUNT_HW_CACHE_MAX]
103 [PERF_COUNT_HW_CACHE_OP_MAX]
104 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
105{
106 [ C(L1D) ] = {
107 [ C(OP_READ) ] = {
108 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
109 [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
110 },
111 [ C(OP_WRITE) ] = {
112 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
113 [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
114 },
115 [ C(OP_PREFETCH) ] = {
116 [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */
117 [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */
118 },
119 },
120 [ C(L1I ) ] = {
121 [ C(OP_READ) ] = {
122 [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
123 [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
124 },
125 [ C(OP_WRITE) ] = {
126 [ C(RESULT_ACCESS) ] = -1,
127 [ C(RESULT_MISS) ] = -1,
128 },
129 [ C(OP_PREFETCH) ] = {
130 [ C(RESULT_ACCESS) ] = 0x0,
131 [ C(RESULT_MISS) ] = 0x0,
132 },
133 },
134 [ C(LL ) ] = {
135 [ C(OP_READ) ] = {
136 [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */
137 [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */
138 },
139 [ C(OP_WRITE) ] = {
140 [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */
141 [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */
142 },
143 [ C(OP_PREFETCH) ] = {
144 [ C(RESULT_ACCESS) ] = 0x4f2e, /* LLC Reference */
145 [ C(RESULT_MISS) ] = 0x412e, /* LLC Misses */
146 },
147 },
148 [ C(DTLB) ] = {
149 [ C(OP_READ) ] = {
150 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
151 [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */
152 },
153 [ C(OP_WRITE) ] = {
154 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
155 [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */
156 },
157 [ C(OP_PREFETCH) ] = {
158 [ C(RESULT_ACCESS) ] = 0x0,
159 [ C(RESULT_MISS) ] = 0x0,
160 },
161 },
162 [ C(ITLB) ] = {
163 [ C(OP_READ) ] = {
164 [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */
165 [ C(RESULT_MISS) ] = 0x20c8, /* ITLB_MISS_RETIRED */
166 },
167 [ C(OP_WRITE) ] = {
168 [ C(RESULT_ACCESS) ] = -1,
169 [ C(RESULT_MISS) ] = -1,
170 },
171 [ C(OP_PREFETCH) ] = {
172 [ C(RESULT_ACCESS) ] = -1,
173 [ C(RESULT_MISS) ] = -1,
174 },
175 },
176 [ C(BPU ) ] = {
177 [ C(OP_READ) ] = {
178 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */
179 [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */
180 },
181 [ C(OP_WRITE) ] = {
182 [ C(RESULT_ACCESS) ] = -1,
183 [ C(RESULT_MISS) ] = -1,
184 },
185 [ C(OP_PREFETCH) ] = {
186 [ C(RESULT_ACCESS) ] = -1,
187 [ C(RESULT_MISS) ] = -1,
188 },
189 },
190};
191
192static const u64 core2_hw_cache_event_ids
193 [PERF_COUNT_HW_CACHE_MAX]
194 [PERF_COUNT_HW_CACHE_OP_MAX]
195 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
196{
197 [ C(L1D) ] = {
198 [ C(OP_READ) ] = {
199 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
200 [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
201 },
202 [ C(OP_WRITE) ] = {
203 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
204 [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
205 },
206 [ C(OP_PREFETCH) ] = {
207 [ C(RESULT_ACCESS) ] = 0x104e, /* L1D_PREFETCH.REQUESTS */
208 [ C(RESULT_MISS) ] = 0,
209 },
210 },
211 [ C(L1I ) ] = {
212 [ C(OP_READ) ] = {
213 [ C(RESULT_ACCESS) ] = 0x0080, /* L1I.READS */
214 [ C(RESULT_MISS) ] = 0x0081, /* L1I.MISSES */
215 },
216 [ C(OP_WRITE) ] = {
217 [ C(RESULT_ACCESS) ] = -1,
218 [ C(RESULT_MISS) ] = -1,
219 },
220 [ C(OP_PREFETCH) ] = {
221 [ C(RESULT_ACCESS) ] = 0,
222 [ C(RESULT_MISS) ] = 0,
223 },
224 },
225 [ C(LL ) ] = {
226 [ C(OP_READ) ] = {
227 [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
228 [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
229 },
230 [ C(OP_WRITE) ] = {
231 [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
232 [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
233 },
234 [ C(OP_PREFETCH) ] = {
235 [ C(RESULT_ACCESS) ] = 0,
236 [ C(RESULT_MISS) ] = 0,
237 },
238 },
239 [ C(DTLB) ] = {
240 [ C(OP_READ) ] = {
241 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
242 [ C(RESULT_MISS) ] = 0x0208, /* DTLB_MISSES.MISS_LD */
243 },
244 [ C(OP_WRITE) ] = {
245 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
246 [ C(RESULT_MISS) ] = 0x0808, /* DTLB_MISSES.MISS_ST */
247 },
248 [ C(OP_PREFETCH) ] = {
249 [ C(RESULT_ACCESS) ] = 0,
250 [ C(RESULT_MISS) ] = 0,
251 },
252 },
253 [ C(ITLB) ] = {
254 [ C(OP_READ) ] = {
255 [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
256 [ C(RESULT_MISS) ] = 0x1282, /* ITLBMISSES */
257 },
258 [ C(OP_WRITE) ] = {
259 [ C(RESULT_ACCESS) ] = -1,
260 [ C(RESULT_MISS) ] = -1,
261 },
262 [ C(OP_PREFETCH) ] = {
263 [ C(RESULT_ACCESS) ] = -1,
264 [ C(RESULT_MISS) ] = -1,
265 },
266 },
267 [ C(BPU ) ] = {
268 [ C(OP_READ) ] = {
269 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
270 [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
271 },
272 [ C(OP_WRITE) ] = {
273 [ C(RESULT_ACCESS) ] = -1,
274 [ C(RESULT_MISS) ] = -1,
275 },
276 [ C(OP_PREFETCH) ] = {
277 [ C(RESULT_ACCESS) ] = -1,
278 [ C(RESULT_MISS) ] = -1,
279 },
280 },
281};
282
283static const u64 atom_hw_cache_event_ids
284 [PERF_COUNT_HW_CACHE_MAX]
285 [PERF_COUNT_HW_CACHE_OP_MAX]
286 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
287{
288 [ C(L1D) ] = {
289 [ C(OP_READ) ] = {
290 [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE.LD */
291 [ C(RESULT_MISS) ] = 0,
292 },
293 [ C(OP_WRITE) ] = {
294 [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE.ST */
295 [ C(RESULT_MISS) ] = 0,
296 },
297 [ C(OP_PREFETCH) ] = {
298 [ C(RESULT_ACCESS) ] = 0x0,
299 [ C(RESULT_MISS) ] = 0,
300 },
301 },
302 [ C(L1I ) ] = {
303 [ C(OP_READ) ] = {
304 [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
305 [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
306 },
307 [ C(OP_WRITE) ] = {
308 [ C(RESULT_ACCESS) ] = -1,
309 [ C(RESULT_MISS) ] = -1,
310 },
311 [ C(OP_PREFETCH) ] = {
312 [ C(RESULT_ACCESS) ] = 0,
313 [ C(RESULT_MISS) ] = 0,
314 },
315 },
316 [ C(LL ) ] = {
317 [ C(OP_READ) ] = {
318 [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
319 [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
320 },
321 [ C(OP_WRITE) ] = {
322 [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
323 [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
324 },
325 [ C(OP_PREFETCH) ] = {
326 [ C(RESULT_ACCESS) ] = 0,
327 [ C(RESULT_MISS) ] = 0,
328 },
329 },
330 [ C(DTLB) ] = {
331 [ C(OP_READ) ] = {
332 [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE_LD.MESI (alias) */
333 [ C(RESULT_MISS) ] = 0x0508, /* DTLB_MISSES.MISS_LD */
334 },
335 [ C(OP_WRITE) ] = {
336 [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE_ST.MESI (alias) */
337 [ C(RESULT_MISS) ] = 0x0608, /* DTLB_MISSES.MISS_ST */
338 },
339 [ C(OP_PREFETCH) ] = {
340 [ C(RESULT_ACCESS) ] = 0,
341 [ C(RESULT_MISS) ] = 0,
342 },
343 },
344 [ C(ITLB) ] = {
345 [ C(OP_READ) ] = {
346 [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
347 [ C(RESULT_MISS) ] = 0x0282, /* ITLB.MISSES */
348 },
349 [ C(OP_WRITE) ] = {
350 [ C(RESULT_ACCESS) ] = -1,
351 [ C(RESULT_MISS) ] = -1,
352 },
353 [ C(OP_PREFETCH) ] = {
354 [ C(RESULT_ACCESS) ] = -1,
355 [ C(RESULT_MISS) ] = -1,
356 },
357 },
358 [ C(BPU ) ] = {
359 [ C(OP_READ) ] = {
360 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
361 [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
362 },
363 [ C(OP_WRITE) ] = {
364 [ C(RESULT_ACCESS) ] = -1,
365 [ C(RESULT_MISS) ] = -1,
366 },
367 [ C(OP_PREFETCH) ] = {
368 [ C(RESULT_ACCESS) ] = -1,
369 [ C(RESULT_MISS) ] = -1,
370 },
371 },
372};
373
374static u64 intel_pmu_raw_event(u64 event)
375{
376#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL
377#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL
378#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL
379#define CORE_EVNTSEL_INV_MASK 0x00800000ULL
380#define CORE_EVNTSEL_COUNTER_MASK 0xFF000000ULL
381
382#define CORE_EVNTSEL_MASK \
383 (CORE_EVNTSEL_EVENT_MASK | \
384 CORE_EVNTSEL_UNIT_MASK | \
385 CORE_EVNTSEL_EDGE_MASK | \
386 CORE_EVNTSEL_INV_MASK | \
387 CORE_EVNTSEL_COUNTER_MASK)
388
389 return event & CORE_EVNTSEL_MASK;
390}
391
392static const u64 amd_0f_hw_cache_event_ids
393 [PERF_COUNT_HW_CACHE_MAX]
394 [PERF_COUNT_HW_CACHE_OP_MAX]
395 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
396{
397 [ C(L1D) ] = {
398 [ C(OP_READ) ] = {
399 [ C(RESULT_ACCESS) ] = 0,
400 [ C(RESULT_MISS) ] = 0,
401 },
402 [ C(OP_WRITE) ] = {
403 [ C(RESULT_ACCESS) ] = 0,
404 [ C(RESULT_MISS) ] = 0,
405 },
406 [ C(OP_PREFETCH) ] = {
407 [ C(RESULT_ACCESS) ] = 0,
408 [ C(RESULT_MISS) ] = 0,
409 },
410 },
411 [ C(L1I ) ] = {
412 [ C(OP_READ) ] = {
413 [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction cache fetches */
414 [ C(RESULT_MISS) ] = 0x0081, /* Instruction cache misses */
415 },
416 [ C(OP_WRITE) ] = {
417 [ C(RESULT_ACCESS) ] = -1,
418 [ C(RESULT_MISS) ] = -1,
419 },
420 [ C(OP_PREFETCH) ] = {
421 [ C(RESULT_ACCESS) ] = 0,
422 [ C(RESULT_MISS) ] = 0,
423 },
424 },
425 [ C(LL ) ] = {
426 [ C(OP_READ) ] = {
427 [ C(RESULT_ACCESS) ] = 0,
428 [ C(RESULT_MISS) ] = 0,
429 },
430 [ C(OP_WRITE) ] = {
431 [ C(RESULT_ACCESS) ] = 0,
432 [ C(RESULT_MISS) ] = 0,
433 },
434 [ C(OP_PREFETCH) ] = {
435 [ C(RESULT_ACCESS) ] = 0,
436 [ C(RESULT_MISS) ] = 0,
437 },
438 },
439 [ C(DTLB) ] = {
440 [ C(OP_READ) ] = {
441 [ C(RESULT_ACCESS) ] = 0,
442 [ C(RESULT_MISS) ] = 0,
443 },
444 [ C(OP_WRITE) ] = {
445 [ C(RESULT_ACCESS) ] = 0,
446 [ C(RESULT_MISS) ] = 0,
447 },
448 [ C(OP_PREFETCH) ] = {
449 [ C(RESULT_ACCESS) ] = 0,
450 [ C(RESULT_MISS) ] = 0,
451 },
452 },
453 [ C(ITLB) ] = {
454 [ C(OP_READ) ] = {
455 [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction fecthes */
456 [ C(RESULT_MISS) ] = 0x0085, /* Instr. fetch ITLB misses */
457 },
458 [ C(OP_WRITE) ] = {
459 [ C(RESULT_ACCESS) ] = -1,
460 [ C(RESULT_MISS) ] = -1,
461 },
462 [ C(OP_PREFETCH) ] = {
463 [ C(RESULT_ACCESS) ] = -1,
464 [ C(RESULT_MISS) ] = -1,
465 },
466 },
467 [ C(BPU ) ] = {
468 [ C(OP_READ) ] = {
469 [ C(RESULT_ACCESS) ] = 0x00c2, /* Retired Branch Instr. */
470 [ C(RESULT_MISS) ] = 0x00c3, /* Retired Mispredicted BI */
471 },
472 [ C(OP_WRITE) ] = {
473 [ C(RESULT_ACCESS) ] = -1,
474 [ C(RESULT_MISS) ] = -1,
475 },
476 [ C(OP_PREFETCH) ] = {
477 [ C(RESULT_ACCESS) ] = -1,
478 [ C(RESULT_MISS) ] = -1,
479 },
480 },
481};
482
483/*
484 * AMD Performance Monitor K7 and later.
485 */
486static const u64 amd_perfmon_event_map[] =
487{
488 [PERF_COUNT_HW_CPU_CYCLES] = 0x0076,
489 [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
490 [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0080,
491 [PERF_COUNT_HW_CACHE_MISSES] = 0x0081,
492 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
493 [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
494};
495
496static u64 amd_pmu_event_map(int event)
497{
498 return amd_perfmon_event_map[event];
499}
500
501static u64 amd_pmu_raw_event(u64 event)
502{
503#define K7_EVNTSEL_EVENT_MASK 0x7000000FFULL
504#define K7_EVNTSEL_UNIT_MASK 0x00000FF00ULL
505#define K7_EVNTSEL_EDGE_MASK 0x000040000ULL
506#define K7_EVNTSEL_INV_MASK 0x000800000ULL
507#define K7_EVNTSEL_COUNTER_MASK 0x0FF000000ULL
508
509#define K7_EVNTSEL_MASK \
510 (K7_EVNTSEL_EVENT_MASK | \
511 K7_EVNTSEL_UNIT_MASK | \
512 K7_EVNTSEL_EDGE_MASK | \
513 K7_EVNTSEL_INV_MASK | \
514 K7_EVNTSEL_COUNTER_MASK)
515
516 return event & K7_EVNTSEL_MASK;
517}
518
519/*
520 * Propagate counter elapsed time into the generic counter.
521 * Can only be executed on the CPU where the counter is active.
522 * Returns the delta events processed.
523 */
524static u64
525x86_perf_counter_update(struct perf_counter *counter,
526 struct hw_perf_counter *hwc, int idx)
527{
528 int shift = 64 - x86_pmu.counter_bits;
529 u64 prev_raw_count, new_raw_count;
530 s64 delta;
531
532 /*
533 * Careful: an NMI might modify the previous counter value.
534 *
535 * Our tactic to handle this is to first atomically read and
536 * exchange a new raw count - then add that new-prev delta
537 * count to the generic counter atomically:
538 */
539again:
540 prev_raw_count = atomic64_read(&hwc->prev_count);
541 rdmsrl(hwc->counter_base + idx, new_raw_count);
542
543 if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count,
544 new_raw_count) != prev_raw_count)
545 goto again;
546
547 /*
548 * Now we have the new raw value and have updated the prev
549 * timestamp already. We can now calculate the elapsed delta
550 * (counter-)time and add that to the generic counter.
551 *
552 * Careful, not all hw sign-extends above the physical width
553 * of the count.
554 */
555 delta = (new_raw_count << shift) - (prev_raw_count << shift);
556 delta >>= shift;
557
558 atomic64_add(delta, &counter->count);
559 atomic64_sub(delta, &hwc->period_left);
560
561 return new_raw_count;
562}
563
564static atomic_t active_counters;
565static DEFINE_MUTEX(pmc_reserve_mutex);
566
567static bool reserve_pmc_hardware(void)
568{
569 int i;
570
571 if (nmi_watchdog == NMI_LOCAL_APIC)
572 disable_lapic_nmi_watchdog();
573
574 for (i = 0; i < x86_pmu.num_counters; i++) {
575 if (!reserve_perfctr_nmi(x86_pmu.perfctr + i))
576 goto perfctr_fail;
577 }
578
579 for (i = 0; i < x86_pmu.num_counters; i++) {
580 if (!reserve_evntsel_nmi(x86_pmu.eventsel + i))
581 goto eventsel_fail;
582 }
583
584 return true;
585
586eventsel_fail:
587 for (i--; i >= 0; i--)
588 release_evntsel_nmi(x86_pmu.eventsel + i);
589
590 i = x86_pmu.num_counters;
591
592perfctr_fail:
593 for (i--; i >= 0; i--)
594 release_perfctr_nmi(x86_pmu.perfctr + i);
595
596 if (nmi_watchdog == NMI_LOCAL_APIC)
597 enable_lapic_nmi_watchdog();
598
599 return false;
600}
601
602static void release_pmc_hardware(void)
603{
604 int i;
605
606 for (i = 0; i < x86_pmu.num_counters; i++) {
607 release_perfctr_nmi(x86_pmu.perfctr + i);
608 release_evntsel_nmi(x86_pmu.eventsel + i);
609 }
610
611 if (nmi_watchdog == NMI_LOCAL_APIC)
612 enable_lapic_nmi_watchdog();
613}
614
615static void hw_perf_counter_destroy(struct perf_counter *counter)
616{
617 if (atomic_dec_and_mutex_lock(&active_counters, &pmc_reserve_mutex)) {
618 release_pmc_hardware();
619 mutex_unlock(&pmc_reserve_mutex);
620 }
621}
622
623static inline int x86_pmu_initialized(void)
624{
625 return x86_pmu.handle_irq != NULL;
626}
627
628static inline int
629set_ext_hw_attr(struct hw_perf_counter *hwc, struct perf_counter_attr *attr)
630{
631 unsigned int cache_type, cache_op, cache_result;
632 u64 config, val;
633
634 config = attr->config;
635
636 cache_type = (config >> 0) & 0xff;
637 if (cache_type >= PERF_COUNT_HW_CACHE_MAX)
638 return -EINVAL;
639
640 cache_op = (config >> 8) & 0xff;
641 if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX)
642 return -EINVAL;
643
644 cache_result = (config >> 16) & 0xff;
645 if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
646 return -EINVAL;
647
648 val = hw_cache_event_ids[cache_type][cache_op][cache_result];
649
650 if (val == 0)
651 return -ENOENT;
652
653 if (val == -1)
654 return -EINVAL;
655
656 hwc->config |= val;
657
658 return 0;
659}
660
661/*
662 * Setup the hardware configuration for a given attr_type
663 */
664static int __hw_perf_counter_init(struct perf_counter *counter)
665{
666 struct perf_counter_attr *attr = &counter->attr;
667 struct hw_perf_counter *hwc = &counter->hw;
668 int err;
669
670 if (!x86_pmu_initialized())
671 return -ENODEV;
672
673 err = 0;
674 if (!atomic_inc_not_zero(&active_counters)) {
675 mutex_lock(&pmc_reserve_mutex);
676 if (atomic_read(&active_counters) == 0 && !reserve_pmc_hardware())
677 err = -EBUSY;
678 else
679 atomic_inc(&active_counters);
680 mutex_unlock(&pmc_reserve_mutex);
681 }
682 if (err)
683 return err;
684
685 /*
686 * Generate PMC IRQs:
687 * (keep 'enabled' bit clear for now)
688 */
689 hwc->config = ARCH_PERFMON_EVENTSEL_INT;
690
691 /*
692 * Count user and OS events unless requested not to.
693 */
694 if (!attr->exclude_user)
695 hwc->config |= ARCH_PERFMON_EVENTSEL_USR;
696 if (!attr->exclude_kernel)
697 hwc->config |= ARCH_PERFMON_EVENTSEL_OS;
698
699 if (!hwc->sample_period) {
700 hwc->sample_period = x86_pmu.max_period;
701 hwc->last_period = hwc->sample_period;
702 atomic64_set(&hwc->period_left, hwc->sample_period);
703 }
704
705 counter->destroy = hw_perf_counter_destroy;
706
707 /*
708 * Raw event type provide the config in the event structure
709 */
710 if (attr->type == PERF_TYPE_RAW) {
711 hwc->config |= x86_pmu.raw_event(attr->config);
712 return 0;
713 }
714
715 if (attr->type == PERF_TYPE_HW_CACHE)
716 return set_ext_hw_attr(hwc, attr);
717
718 if (attr->config >= x86_pmu.max_events)
719 return -EINVAL;
720 /*
721 * The generic map:
722 */
723 hwc->config |= x86_pmu.event_map(attr->config);
724
725 return 0;
726}
727
728static void intel_pmu_disable_all(void)
729{
730 wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0);
731}
732
733static void amd_pmu_disable_all(void)
734{
735 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
736 int idx;
737
738 if (!cpuc->enabled)
739 return;
740
741 cpuc->enabled = 0;
742 /*
743 * ensure we write the disable before we start disabling the
744 * counters proper, so that amd_pmu_enable_counter() does the
745 * right thing.
746 */
747 barrier();
748
749 for (idx = 0; idx < x86_pmu.num_counters; idx++) {
750 u64 val;
751
752 if (!test_bit(idx, cpuc->active_mask))
753 continue;
754 rdmsrl(MSR_K7_EVNTSEL0 + idx, val);
755 if (!(val & ARCH_PERFMON_EVENTSEL0_ENABLE))
756 continue;
757 val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE;
758 wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
759 }
760}
761
762void hw_perf_disable(void)
763{
764 if (!x86_pmu_initialized())
765 return;
766 return x86_pmu.disable_all();
767}
768
769static void intel_pmu_enable_all(void)
770{
771 wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
772}
773
774static void amd_pmu_enable_all(void)
775{
776 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
777 int idx;
778
779 if (cpuc->enabled)
780 return;
781
782 cpuc->enabled = 1;
783 barrier();
784
785 for (idx = 0; idx < x86_pmu.num_counters; idx++) {
786 u64 val;
787
788 if (!test_bit(idx, cpuc->active_mask))
789 continue;
790 rdmsrl(MSR_K7_EVNTSEL0 + idx, val);
791 if (val & ARCH_PERFMON_EVENTSEL0_ENABLE)
792 continue;
793 val |= ARCH_PERFMON_EVENTSEL0_ENABLE;
794 wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
795 }
796}
797
798void hw_perf_enable(void)
799{
800 if (!x86_pmu_initialized())
801 return;
802 x86_pmu.enable_all();
803}
804
805static inline u64 intel_pmu_get_status(void)
806{
807 u64 status;
808
809 rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
810
811 return status;
812}
813
814static inline void intel_pmu_ack_status(u64 ack)
815{
816 wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack);
817}
818
819static inline void x86_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
820{
821 int err;
822 err = checking_wrmsrl(hwc->config_base + idx,
823 hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE);
824}
825
826static inline void x86_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
827{
828 int err;
829 err = checking_wrmsrl(hwc->config_base + idx,
830 hwc->config);
831}
832
833static inline void
834intel_pmu_disable_fixed(struct hw_perf_counter *hwc, int __idx)
835{
836 int idx = __idx - X86_PMC_IDX_FIXED;
837 u64 ctrl_val, mask;
838 int err;
839
840 mask = 0xfULL << (idx * 4);
841
842 rdmsrl(hwc->config_base, ctrl_val);
843 ctrl_val &= ~mask;
844 err = checking_wrmsrl(hwc->config_base, ctrl_val);
845}
846
847static inline void
848intel_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
849{
850 if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
851 intel_pmu_disable_fixed(hwc, idx);
852 return;
853 }
854
855 x86_pmu_disable_counter(hwc, idx);
856}
857
858static inline void
859amd_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
860{
861 x86_pmu_disable_counter(hwc, idx);
862}
863
864static DEFINE_PER_CPU(u64, prev_left[X86_PMC_IDX_MAX]);
865
866/*
867 * Set the next IRQ period, based on the hwc->period_left value.
868 * To be called with the counter disabled in hw:
869 */
870static int
871x86_perf_counter_set_period(struct perf_counter *counter,
872 struct hw_perf_counter *hwc, int idx)
873{
874 s64 left = atomic64_read(&hwc->period_left);
875 s64 period = hwc->sample_period;
876 int err, ret = 0;
877
878 /*
879 * If we are way outside a reasoable range then just skip forward:
880 */
881 if (unlikely(left <= -period)) {
882 left = period;
883 atomic64_set(&hwc->period_left, left);
884 hwc->last_period = period;
885 ret = 1;
886 }
887
888 if (unlikely(left <= 0)) {
889 left += period;
890 atomic64_set(&hwc->period_left, left);
891 hwc->last_period = period;
892 ret = 1;
893 }
894 /*
895 * Quirk: certain CPUs dont like it if just 1 event is left:
896 */
897 if (unlikely(left < 2))
898 left = 2;
899
900 if (left > x86_pmu.max_period)
901 left = x86_pmu.max_period;
902
903 per_cpu(prev_left[idx], smp_processor_id()) = left;
904
905 /*
906 * The hw counter starts counting from this counter offset,
907 * mark it to be able to extra future deltas:
908 */
909 atomic64_set(&hwc->prev_count, (u64)-left);
910
911 err = checking_wrmsrl(hwc->counter_base + idx,
912 (u64)(-left) & x86_pmu.counter_mask);
913
914 return ret;
915}
916
917static inline void
918intel_pmu_enable_fixed(struct hw_perf_counter *hwc, int __idx)
919{
920 int idx = __idx - X86_PMC_IDX_FIXED;
921 u64 ctrl_val, bits, mask;
922 int err;
923
924 /*
925 * Enable IRQ generation (0x8),
926 * and enable ring-3 counting (0x2) and ring-0 counting (0x1)
927 * if requested:
928 */
929 bits = 0x8ULL;
930 if (hwc->config & ARCH_PERFMON_EVENTSEL_USR)
931 bits |= 0x2;
932 if (hwc->config & ARCH_PERFMON_EVENTSEL_OS)
933 bits |= 0x1;
934 bits <<= (idx * 4);
935 mask = 0xfULL << (idx * 4);
936
937 rdmsrl(hwc->config_base, ctrl_val);
938 ctrl_val &= ~mask;
939 ctrl_val |= bits;
940 err = checking_wrmsrl(hwc->config_base, ctrl_val);
941}
942
943static void intel_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
944{
945 if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
946 intel_pmu_enable_fixed(hwc, idx);
947 return;
948 }
949
950 x86_pmu_enable_counter(hwc, idx);
951}
952
953static void amd_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
954{
955 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
956
957 if (cpuc->enabled)
958 x86_pmu_enable_counter(hwc, idx);
959 else
960 x86_pmu_disable_counter(hwc, idx);
961}
962
963static int
964fixed_mode_idx(struct perf_counter *counter, struct hw_perf_counter *hwc)
965{
966 unsigned int event;
967
968 if (!x86_pmu.num_counters_fixed)
969 return -1;
970
971 event = hwc->config & ARCH_PERFMON_EVENT_MASK;
972
973 if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS)))
974 return X86_PMC_IDX_FIXED_INSTRUCTIONS;
975 if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_CPU_CYCLES)))
976 return X86_PMC_IDX_FIXED_CPU_CYCLES;
977 if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_BUS_CYCLES)))
978 return X86_PMC_IDX_FIXED_BUS_CYCLES;
979
980 return -1;
981}
982
983/*
984 * Find a PMC slot for the freshly enabled / scheduled in counter:
985 */
986static int x86_pmu_enable(struct perf_counter *counter)
987{
988 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
989 struct hw_perf_counter *hwc = &counter->hw;
990 int idx;
991
992 idx = fixed_mode_idx(counter, hwc);
993 if (idx >= 0) {
994 /*
995 * Try to get the fixed counter, if that is already taken
996 * then try to get a generic counter:
997 */
998 if (test_and_set_bit(idx, cpuc->used_mask))
999 goto try_generic;
1000
1001 hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL;
1002 /*
1003 * We set it so that counter_base + idx in wrmsr/rdmsr maps to
1004 * MSR_ARCH_PERFMON_FIXED_CTR0 ... CTR2:
1005 */
1006 hwc->counter_base =
1007 MSR_ARCH_PERFMON_FIXED_CTR0 - X86_PMC_IDX_FIXED;
1008 hwc->idx = idx;
1009 } else {
1010 idx = hwc->idx;
1011 /* Try to get the previous generic counter again */
1012 if (test_and_set_bit(idx, cpuc->used_mask)) {
1013try_generic:
1014 idx = find_first_zero_bit(cpuc->used_mask,
1015 x86_pmu.num_counters);
1016 if (idx == x86_pmu.num_counters)
1017 return -EAGAIN;
1018
1019 set_bit(idx, cpuc->used_mask);
1020 hwc->idx = idx;
1021 }
1022 hwc->config_base = x86_pmu.eventsel;
1023 hwc->counter_base = x86_pmu.perfctr;
1024 }
1025
1026 perf_counters_lapic_init();
1027
1028 x86_pmu.disable(hwc, idx);
1029
1030 cpuc->counters[idx] = counter;
1031 set_bit(idx, cpuc->active_mask);
1032
1033 x86_perf_counter_set_period(counter, hwc, idx);
1034 x86_pmu.enable(hwc, idx);
1035
1036 return 0;
1037}
1038
1039static void x86_pmu_unthrottle(struct perf_counter *counter)
1040{
1041 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
1042 struct hw_perf_counter *hwc = &counter->hw;
1043
1044 if (WARN_ON_ONCE(hwc->idx >= X86_PMC_IDX_MAX ||
1045 cpuc->counters[hwc->idx] != counter))
1046 return;
1047
1048 x86_pmu.enable(hwc, hwc->idx);
1049}
1050
1051void perf_counter_print_debug(void)
1052{
1053 u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed;
1054 struct cpu_hw_counters *cpuc;
1055 unsigned long flags;
1056 int cpu, idx;
1057
1058 if (!x86_pmu.num_counters)
1059 return;
1060
1061 local_irq_save(flags);
1062
1063 cpu = smp_processor_id();
1064 cpuc = &per_cpu(cpu_hw_counters, cpu);
1065
1066 if (x86_pmu.version >= 2) {
1067 rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
1068 rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
1069 rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
1070 rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR_CTRL, fixed);
1071
1072 pr_info("\n");
1073 pr_info("CPU#%d: ctrl: %016llx\n", cpu, ctrl);
1074 pr_info("CPU#%d: status: %016llx\n", cpu, status);
1075 pr_info("CPU#%d: overflow: %016llx\n", cpu, overflow);
1076 pr_info("CPU#%d: fixed: %016llx\n", cpu, fixed);
1077 }
1078 pr_info("CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used_mask);
1079
1080 for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1081 rdmsrl(x86_pmu.eventsel + idx, pmc_ctrl);
1082 rdmsrl(x86_pmu.perfctr + idx, pmc_count);
1083
1084 prev_left = per_cpu(prev_left[idx], cpu);
1085
1086 pr_info("CPU#%d: gen-PMC%d ctrl: %016llx\n",
1087 cpu, idx, pmc_ctrl);
1088 pr_info("CPU#%d: gen-PMC%d count: %016llx\n",
1089 cpu, idx, pmc_count);
1090 pr_info("CPU#%d: gen-PMC%d left: %016llx\n",
1091 cpu, idx, prev_left);
1092 }
1093 for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
1094 rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count);
1095
1096 pr_info("CPU#%d: fixed-PMC%d count: %016llx\n",
1097 cpu, idx, pmc_count);
1098 }
1099 local_irq_restore(flags);
1100}
1101
1102static void x86_pmu_disable(struct perf_counter *counter)
1103{
1104 struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
1105 struct hw_perf_counter *hwc = &counter->hw;
1106 int idx = hwc->idx;
1107
1108 /*
1109 * Must be done before we disable, otherwise the nmi handler
1110 * could reenable again:
1111 */
1112 clear_bit(idx, cpuc->active_mask);
1113 x86_pmu.disable(hwc, idx);
1114
1115 /*
1116 * Make sure the cleared pointer becomes visible before we
1117 * (potentially) free the counter:
1118 */
1119 barrier();
1120
1121 /*
1122 * Drain the remaining delta count out of a counter
1123 * that we are disabling:
1124 */
1125 x86_perf_counter_update(counter, hwc, idx);
1126 cpuc->counters[idx] = NULL;
1127 clear_bit(idx, cpuc->used_mask);
1128}
1129
1130/*
1131 * Save and restart an expired counter. Called by NMI contexts,
1132 * so it has to be careful about preempting normal counter ops:
1133 */
1134static int intel_pmu_save_and_restart(struct perf_counter *counter)
1135{
1136 struct hw_perf_counter *hwc = &counter->hw;
1137 int idx = hwc->idx;
1138 int ret;
1139
1140 x86_perf_counter_update(counter, hwc, idx);
1141 ret = x86_perf_counter_set_period(counter, hwc, idx);
1142
1143 if (counter->state == PERF_COUNTER_STATE_ACTIVE)
1144 intel_pmu_enable_counter(hwc, idx);
1145
1146 return ret;
1147}
1148
1149static void intel_pmu_reset(void)
1150{
1151 unsigned long flags;
1152 int idx;
1153
1154 if (!x86_pmu.num_counters)
1155 return;
1156
1157 local_irq_save(flags);
1158
1159 printk("clearing PMU state on CPU#%d\n", smp_processor_id());
1160
1161 for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1162 checking_wrmsrl(x86_pmu.eventsel + idx, 0ull);
1163 checking_wrmsrl(x86_pmu.perfctr + idx, 0ull);
1164 }
1165 for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
1166 checking_wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, 0ull);
1167 }
1168
1169 local_irq_restore(flags);
1170}
1171
1172
1173/*
1174 * This handler is triggered by the local APIC, so the APIC IRQ handling
1175 * rules apply:
1176 */
1177static int intel_pmu_handle_irq(struct pt_regs *regs)
1178{
1179 struct perf_sample_data data;
1180 struct cpu_hw_counters *cpuc;
1181 int bit, cpu, loops;
1182 u64 ack, status;
1183
1184 data.regs = regs;
1185 data.addr = 0;
1186
1187 cpu = smp_processor_id();
1188 cpuc = &per_cpu(cpu_hw_counters, cpu);
1189
1190 perf_disable();
1191 status = intel_pmu_get_status();
1192 if (!status) {
1193 perf_enable();
1194 return 0;
1195 }
1196
1197 loops = 0;
1198again:
1199 if (++loops > 100) {
1200 WARN_ONCE(1, "perfcounters: irq loop stuck!\n");
1201 perf_counter_print_debug();
1202 intel_pmu_reset();
1203 perf_enable();
1204 return 1;
1205 }
1206
1207 inc_irq_stat(apic_perf_irqs);
1208 ack = status;
1209 for_each_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) {
1210 struct perf_counter *counter = cpuc->counters[bit];
1211
1212 clear_bit(bit, (unsigned long *) &status);
1213 if (!test_bit(bit, cpuc->active_mask))
1214 continue;
1215
1216 if (!intel_pmu_save_and_restart(counter))
1217 continue;
1218
1219 if (perf_counter_overflow(counter, 1, &data))
1220 intel_pmu_disable_counter(&counter->hw, bit);
1221 }
1222
1223 intel_pmu_ack_status(ack);
1224
1225 /*
1226 * Repeat if there is more work to be done:
1227 */
1228 status = intel_pmu_get_status();
1229 if (status)
1230 goto again;
1231
1232 perf_enable();
1233
1234 return 1;
1235}
1236
1237static int amd_pmu_handle_irq(struct pt_regs *regs)
1238{
1239 struct perf_sample_data data;
1240 struct cpu_hw_counters *cpuc;
1241 struct perf_counter *counter;
1242 struct hw_perf_counter *hwc;
1243 int cpu, idx, handled = 0;
1244 u64 val;
1245
1246 data.regs = regs;
1247 data.addr = 0;
1248
1249 cpu = smp_processor_id();
1250 cpuc = &per_cpu(cpu_hw_counters, cpu);
1251
1252 for (idx = 0; idx < x86_pmu.num_counters; idx++) {
1253 if (!test_bit(idx, cpuc->active_mask))
1254 continue;
1255
1256 counter = cpuc->counters[idx];
1257 hwc = &counter->hw;
1258
1259 val = x86_perf_counter_update(counter, hwc, idx);
1260 if (val & (1ULL << (x86_pmu.counter_bits - 1)))
1261 continue;
1262
1263 /*
1264 * counter overflow
1265 */
1266 handled = 1;
1267 data.period = counter->hw.last_period;
1268
1269 if (!x86_perf_counter_set_period(counter, hwc, idx))
1270 continue;
1271
1272 if (perf_counter_overflow(counter, 1, &data))
1273 amd_pmu_disable_counter(hwc, idx);
1274 }
1275
1276 if (handled)
1277 inc_irq_stat(apic_perf_irqs);
1278
1279 return handled;
1280}
1281
1282void smp_perf_pending_interrupt(struct pt_regs *regs)
1283{
1284 irq_enter();
1285 ack_APIC_irq();
1286 inc_irq_stat(apic_pending_irqs);
1287 perf_counter_do_pending();
1288 irq_exit();
1289}
1290
1291void set_perf_counter_pending(void)
1292{
1293 apic->send_IPI_self(LOCAL_PENDING_VECTOR);
1294}
1295
1296void perf_counters_lapic_init(void)
1297{
1298 if (!x86_pmu_initialized())
1299 return;
1300
1301 /*
1302 * Always use NMI for PMU
1303 */
1304 apic_write(APIC_LVTPC, APIC_DM_NMI);
1305}
1306
1307static int __kprobes
1308perf_counter_nmi_handler(struct notifier_block *self,
1309 unsigned long cmd, void *__args)
1310{
1311 struct die_args *args = __args;
1312 struct pt_regs *regs;
1313
1314 if (!atomic_read(&active_counters))
1315 return NOTIFY_DONE;
1316
1317 switch (cmd) {
1318 case DIE_NMI:
1319 case DIE_NMI_IPI:
1320 break;
1321
1322 default:
1323 return NOTIFY_DONE;
1324 }
1325
1326 regs = args->regs;
1327
1328 apic_write(APIC_LVTPC, APIC_DM_NMI);
1329 /*
1330 * Can't rely on the handled return value to say it was our NMI, two
1331 * counters could trigger 'simultaneously' raising two back-to-back NMIs.
1332 *
1333 * If the first NMI handles both, the latter will be empty and daze
1334 * the CPU.
1335 */
1336 x86_pmu.handle_irq(regs);
1337
1338 return NOTIFY_STOP;
1339}
1340
1341static __read_mostly struct notifier_block perf_counter_nmi_notifier = {
1342 .notifier_call = perf_counter_nmi_handler,
1343 .next = NULL,
1344 .priority = 1
1345};
1346
1347static struct x86_pmu intel_pmu = {
1348 .name = "Intel",
1349 .handle_irq = intel_pmu_handle_irq,
1350 .disable_all = intel_pmu_disable_all,
1351 .enable_all = intel_pmu_enable_all,
1352 .enable = intel_pmu_enable_counter,
1353 .disable = intel_pmu_disable_counter,
1354 .eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
1355 .perfctr = MSR_ARCH_PERFMON_PERFCTR0,
1356 .event_map = intel_pmu_event_map,
1357 .raw_event = intel_pmu_raw_event,
1358 .max_events = ARRAY_SIZE(intel_perfmon_event_map),
1359 /*
1360 * Intel PMCs cannot be accessed sanely above 32 bit width,
1361 * so we install an artificial 1<<31 period regardless of
1362 * the generic counter period:
1363 */
1364 .max_period = (1ULL << 31) - 1,
1365};
1366
1367static struct x86_pmu amd_pmu = {
1368 .name = "AMD",
1369 .handle_irq = amd_pmu_handle_irq,
1370 .disable_all = amd_pmu_disable_all,
1371 .enable_all = amd_pmu_enable_all,
1372 .enable = amd_pmu_enable_counter,
1373 .disable = amd_pmu_disable_counter,
1374 .eventsel = MSR_K7_EVNTSEL0,
1375 .perfctr = MSR_K7_PERFCTR0,
1376 .event_map = amd_pmu_event_map,
1377 .raw_event = amd_pmu_raw_event,
1378 .max_events = ARRAY_SIZE(amd_perfmon_event_map),
1379 .num_counters = 4,
1380 .counter_bits = 48,
1381 .counter_mask = (1ULL << 48) - 1,
1382 /* use highest bit to detect overflow */
1383 .max_period = (1ULL << 47) - 1,
1384};
1385
1386static int intel_pmu_init(void)
1387{
1388 union cpuid10_edx edx;
1389 union cpuid10_eax eax;
1390 unsigned int unused;
1391 unsigned int ebx;
1392 int version;
1393
1394 if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
1395 return -ENODEV;
1396
1397 /*
1398 * Check whether the Architectural PerfMon supports
1399 * Branch Misses Retired Event or not.
1400 */
1401 cpuid(10, &eax.full, &ebx, &unused, &edx.full);
1402 if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
1403 return -ENODEV;
1404
1405 version = eax.split.version_id;
1406 if (version < 2)
1407 return -ENODEV;
1408
1409 x86_pmu = intel_pmu;
1410 x86_pmu.version = version;
1411 x86_pmu.num_counters = eax.split.num_counters;
1412 x86_pmu.counter_bits = eax.split.bit_width;
1413 x86_pmu.counter_mask = (1ULL << eax.split.bit_width) - 1;
1414
1415 /*
1416 * Quirk: v2 perfmon does not report fixed-purpose counters, so
1417 * assume at least 3 counters:
1418 */
1419 x86_pmu.num_counters_fixed = max((int)edx.split.num_counters_fixed, 3);
1420
1421 rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
1422
1423 /*
1424 * Install the hw-cache-events table:
1425 */
1426 switch (boot_cpu_data.x86_model) {
1427 case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
1428 case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
1429 case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
1430 case 29: /* six-core 45 nm xeon "Dunnington" */
1431 memcpy(hw_cache_event_ids, core2_hw_cache_event_ids,
1432 sizeof(hw_cache_event_ids));
1433
1434 pr_cont("Core2 events, ");
1435 break;
1436 default:
1437 case 26:
1438 memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids,
1439 sizeof(hw_cache_event_ids));
1440
1441 pr_cont("Nehalem/Corei7 events, ");
1442 break;
1443 case 28:
1444 memcpy(hw_cache_event_ids, atom_hw_cache_event_ids,
1445 sizeof(hw_cache_event_ids));
1446
1447 pr_cont("Atom events, ");
1448 break;
1449 }
1450 return 0;
1451}
1452
1453static int amd_pmu_init(void)
1454{
1455 x86_pmu = amd_pmu;
1456
1457 switch (boot_cpu_data.x86) {
1458 case 0x0f:
1459 case 0x10:
1460 case 0x11:
1461 memcpy(hw_cache_event_ids, amd_0f_hw_cache_event_ids,
1462 sizeof(hw_cache_event_ids));
1463
1464 pr_cont("AMD Family 0f/10/11 events, ");
1465 break;
1466 }
1467 return 0;
1468}
1469
1470void __init init_hw_perf_counters(void)
1471{
1472 int err;
1473
1474 pr_info("Performance Counters: ");
1475
1476 switch (boot_cpu_data.x86_vendor) {
1477 case X86_VENDOR_INTEL:
1478 err = intel_pmu_init();
1479 break;
1480 case X86_VENDOR_AMD:
1481 err = amd_pmu_init();
1482 break;
1483 default:
1484 return;
1485 }
1486 if (err != 0) {
1487 pr_cont("no PMU driver, software counters only.\n");
1488 return;
1489 }
1490
1491 pr_cont("%s PMU driver.\n", x86_pmu.name);
1492
1493 if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) {
1494 x86_pmu.num_counters = X86_PMC_MAX_GENERIC;
1495 WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
1496 x86_pmu.num_counters, X86_PMC_MAX_GENERIC);
1497 }
1498 perf_counter_mask = (1 << x86_pmu.num_counters) - 1;
1499 perf_max_counters = x86_pmu.num_counters;
1500
1501 if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) {
1502 x86_pmu.num_counters_fixed = X86_PMC_MAX_FIXED;
1503 WARN(1, KERN_ERR "hw perf counters fixed %d > max(%d), clipping!",
1504 x86_pmu.num_counters_fixed, X86_PMC_MAX_FIXED);
1505 }
1506
1507 perf_counter_mask |=
1508 ((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED;
1509
1510 perf_counters_lapic_init();
1511 register_die_notifier(&perf_counter_nmi_notifier);
1512
1513 pr_info("... version: %d\n", x86_pmu.version);
1514 pr_info("... bit width: %d\n", x86_pmu.counter_bits);
1515 pr_info("... generic counters: %d\n", x86_pmu.num_counters);
1516 pr_info("... value mask: %016Lx\n", x86_pmu.counter_mask);
1517 pr_info("... max period: %016Lx\n", x86_pmu.max_period);
1518 pr_info("... fixed-purpose counters: %d\n", x86_pmu.num_counters_fixed);
1519 pr_info("... counter mask: %016Lx\n", perf_counter_mask);
1520}
1521
1522static inline void x86_pmu_read(struct perf_counter *counter)
1523{
1524 x86_perf_counter_update(counter, &counter->hw, counter->hw.idx);
1525}
1526
1527static const struct pmu pmu = {
1528 .enable = x86_pmu_enable,
1529 .disable = x86_pmu_disable,
1530 .read = x86_pmu_read,
1531 .unthrottle = x86_pmu_unthrottle,
1532};
1533
1534const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
1535{
1536 int err;
1537
1538 err = __hw_perf_counter_init(counter);
1539 if (err)
1540 return ERR_PTR(err);
1541
1542 return &pmu;
1543}
1544
1545/*
1546 * callchain support
1547 */
1548
1549static inline
1550void callchain_store(struct perf_callchain_entry *entry, unsigned long ip)
1551{
1552 if (entry->nr < MAX_STACK_DEPTH)
1553 entry->ip[entry->nr++] = ip;
1554}
1555
1556static DEFINE_PER_CPU(struct perf_callchain_entry, irq_entry);
1557static DEFINE_PER_CPU(struct perf_callchain_entry, nmi_entry);
1558
1559
1560static void
1561backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
1562{
1563 /* Ignore warnings */
1564}
1565
1566static void backtrace_warning(void *data, char *msg)
1567{
1568 /* Ignore warnings */
1569}
1570
1571static int backtrace_stack(void *data, char *name)
1572{
1573 /* Don't bother with IRQ stacks for now */
1574 return -1;
1575}
1576
1577static void backtrace_address(void *data, unsigned long addr, int reliable)
1578{
1579 struct perf_callchain_entry *entry = data;
1580
1581 if (reliable)
1582 callchain_store(entry, addr);
1583}
1584
1585static const struct stacktrace_ops backtrace_ops = {
1586 .warning = backtrace_warning,
1587 .warning_symbol = backtrace_warning_symbol,
1588 .stack = backtrace_stack,
1589 .address = backtrace_address,
1590};
1591
1592static void
1593perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry)
1594{
1595 unsigned long bp;
1596 char *stack;
1597 int nr = entry->nr;
1598
1599 callchain_store(entry, instruction_pointer(regs));
1600
1601 stack = ((char *)regs + sizeof(struct pt_regs));
1602#ifdef CONFIG_FRAME_POINTER
1603 bp = frame_pointer(regs);
1604#else
1605 bp = 0;
1606#endif
1607
1608 dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, entry);
1609
1610 entry->kernel = entry->nr - nr;
1611}
1612
1613
1614struct stack_frame {
1615 const void __user *next_fp;
1616 unsigned long return_address;
1617};
1618
1619static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
1620{
1621 int ret;
1622
1623 if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
1624 return 0;
1625
1626 ret = 1;
1627 pagefault_disable();
1628 if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
1629 ret = 0;
1630 pagefault_enable();
1631
1632 return ret;
1633}
1634
1635static void
1636perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
1637{
1638 struct stack_frame frame;
1639 const void __user *fp;
1640 int nr = entry->nr;
1641
1642 regs = (struct pt_regs *)current->thread.sp0 - 1;
1643 fp = (void __user *)regs->bp;
1644
1645 callchain_store(entry, regs->ip);
1646
1647 while (entry->nr < MAX_STACK_DEPTH) {
1648 frame.next_fp = NULL;
1649 frame.return_address = 0;
1650
1651 if (!copy_stack_frame(fp, &frame))
1652 break;
1653
1654 if ((unsigned long)fp < user_stack_pointer(regs))
1655 break;
1656
1657 callchain_store(entry, frame.return_address);
1658 fp = frame.next_fp;
1659 }
1660
1661 entry->user = entry->nr - nr;
1662}
1663
1664static void
1665perf_do_callchain(struct pt_regs *regs, struct perf_callchain_entry *entry)
1666{
1667 int is_user;
1668
1669 if (!regs)
1670 return;
1671
1672 is_user = user_mode(regs);
1673
1674 if (!current || current->pid == 0)
1675 return;
1676
1677 if (is_user && current->state != TASK_RUNNING)
1678 return;
1679
1680 if (!is_user)
1681 perf_callchain_kernel(regs, entry);
1682
1683 if (current->mm)
1684 perf_callchain_user(regs, entry);
1685}
1686
1687struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
1688{
1689 struct perf_callchain_entry *entry;
1690
1691 if (in_nmi())
1692 entry = &__get_cpu_var(nmi_entry);
1693 else
1694 entry = &__get_cpu_var(irq_entry);
1695
1696 entry->nr = 0;
1697 entry->hv = 0;
1698 entry->kernel = 0;
1699 entry->user = 0;
1700
1701 perf_do_callchain(regs, entry);
1702
1703 return entry;
1704}
diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c
index f6c70a164e32..d6f5b9fbde32 100644
--- a/arch/x86/kernel/cpu/perfctr-watchdog.c
+++ b/arch/x86/kernel/cpu/perfctr-watchdog.c
@@ -19,8 +19,8 @@
19#include <linux/nmi.h> 19#include <linux/nmi.h>
20#include <linux/kprobes.h> 20#include <linux/kprobes.h>
21 21
22#include <asm/genapic.h> 22#include <asm/apic.h>
23#include <asm/intel_arch_perfmon.h> 23#include <asm/perf_counter.h>
24 24
25struct nmi_watchdog_ctlblk { 25struct nmi_watchdog_ctlblk {
26 unsigned int cccr_msr; 26 unsigned int cccr_msr;
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 1c17d7c751a4..a4742a340d8d 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -1012,6 +1012,11 @@ apicinterrupt ERROR_APIC_VECTOR \
1012apicinterrupt SPURIOUS_APIC_VECTOR \ 1012apicinterrupt SPURIOUS_APIC_VECTOR \
1013 spurious_interrupt smp_spurious_interrupt 1013 spurious_interrupt smp_spurious_interrupt
1014 1014
1015#ifdef CONFIG_PERF_COUNTERS
1016apicinterrupt LOCAL_PENDING_VECTOR \
1017 perf_pending_interrupt smp_perf_pending_interrupt
1018#endif
1019
1015/* 1020/*
1016 * Exception entry points. 1021 * Exception entry points.
1017 */ 1022 */
diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c
index 9a391bbb8ba8..38287b5f116e 100644
--- a/arch/x86/kernel/irq.c
+++ b/arch/x86/kernel/irq.c
@@ -62,6 +62,14 @@ static int show_other_interrupts(struct seq_file *p, int prec)
62 for_each_online_cpu(j) 62 for_each_online_cpu(j)
63 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count); 63 seq_printf(p, "%10u ", irq_stats(j)->irq_spurious_count);
64 seq_printf(p, " Spurious interrupts\n"); 64 seq_printf(p, " Spurious interrupts\n");
65 seq_printf(p, "%*s: ", prec, "CNT");
66 for_each_online_cpu(j)
67 seq_printf(p, "%10u ", irq_stats(j)->apic_perf_irqs);
68 seq_printf(p, " Performance counter interrupts\n");
69 seq_printf(p, "%*s: ", prec, "PND");
70 for_each_online_cpu(j)
71 seq_printf(p, "%10u ", irq_stats(j)->apic_pending_irqs);
72 seq_printf(p, " Performance pending work\n");
65#endif 73#endif
66 if (generic_interrupt_extension) { 74 if (generic_interrupt_extension) {
67 seq_printf(p, "%*s: ", prec, "PLT"); 75 seq_printf(p, "%*s: ", prec, "PLT");
@@ -165,6 +173,8 @@ u64 arch_irq_stat_cpu(unsigned int cpu)
165#ifdef CONFIG_X86_LOCAL_APIC 173#ifdef CONFIG_X86_LOCAL_APIC
166 sum += irq_stats(cpu)->apic_timer_irqs; 174 sum += irq_stats(cpu)->apic_timer_irqs;
167 sum += irq_stats(cpu)->irq_spurious_count; 175 sum += irq_stats(cpu)->irq_spurious_count;
176 sum += irq_stats(cpu)->apic_perf_irqs;
177 sum += irq_stats(cpu)->apic_pending_irqs;
168#endif 178#endif
169 if (generic_interrupt_extension) 179 if (generic_interrupt_extension)
170 sum += irq_stats(cpu)->generic_irqs; 180 sum += irq_stats(cpu)->generic_irqs;
diff --git a/arch/x86/kernel/irqinit.c b/arch/x86/kernel/irqinit.c
index 2e08b10ad51a..267c6624c77f 100644
--- a/arch/x86/kernel/irqinit.c
+++ b/arch/x86/kernel/irqinit.c
@@ -181,10 +181,15 @@ static void __init apic_intr_init(void)
181{ 181{
182 smp_intr_init(); 182 smp_intr_init();
183 183
184#ifdef CONFIG_X86_64 184#ifdef CONFIG_X86_THERMAL_VECTOR
185 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt); 185 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
186#endif
187#ifdef CONFIG_X86_THRESHOLD
186 alloc_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt); 188 alloc_intr_gate(THRESHOLD_APIC_VECTOR, threshold_interrupt);
187#endif 189#endif
190#if defined(CONFIG_X86_NEW_MCE) && defined(CONFIG_X86_LOCAL_APIC)
191 alloc_intr_gate(MCE_SELF_VECTOR, mce_self_interrupt);
192#endif
188 193
189#if defined(CONFIG_X86_64) || defined(CONFIG_X86_LOCAL_APIC) 194#if defined(CONFIG_X86_64) || defined(CONFIG_X86_LOCAL_APIC)
190 /* self generated IPI for local APIC timer */ 195 /* self generated IPI for local APIC timer */
@@ -199,18 +204,10 @@ static void __init apic_intr_init(void)
199 204
200 /* Performance monitoring interrupts: */ 205 /* Performance monitoring interrupts: */
201# ifdef CONFIG_PERF_COUNTERS 206# ifdef CONFIG_PERF_COUNTERS
202 alloc_intr_gate(LOCAL_PERF_VECTOR, perf_counter_interrupt);
203 alloc_intr_gate(LOCAL_PENDING_VECTOR, perf_pending_interrupt); 207 alloc_intr_gate(LOCAL_PENDING_VECTOR, perf_pending_interrupt);
204# endif 208# endif
205 209
206#endif 210#endif
207
208#ifdef CONFIG_X86_32
209#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86_MCE_P4THERMAL)
210 /* thermal monitor LVT interrupt */
211 alloc_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
212#endif
213#endif
214} 211}
215 212
216/** 213/**
diff --git a/arch/x86/kernel/signal.c b/arch/x86/kernel/signal.c
index 14425166b8e3..0a813b17b172 100644
--- a/arch/x86/kernel/signal.c
+++ b/arch/x86/kernel/signal.c
@@ -6,7 +6,6 @@
6 * 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 7 * 2000-2002 x86-64 support by Andi Kleen
8 */ 8 */
9
10#include <linux/sched.h> 9#include <linux/sched.h>
11#include <linux/mm.h> 10#include <linux/mm.h>
12#include <linux/smp.h> 11#include <linux/smp.h>
diff --git a/arch/x86/kernel/syscall_table_32.S b/arch/x86/kernel/syscall_table_32.S
index 734f92c02dde..d51321ddafda 100644
--- a/arch/x86/kernel/syscall_table_32.S
+++ b/arch/x86/kernel/syscall_table_32.S
@@ -335,3 +335,4 @@ ENTRY(sys_call_table)
335 .long sys_preadv 335 .long sys_preadv
336 .long sys_pwritev 336 .long sys_pwritev
337 .long sys_rt_tgsigqueueinfo /* 335 */ 337 .long sys_rt_tgsigqueueinfo /* 335 */
338 .long sys_perf_counter_open
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index ede024531f8f..07d60c870ce2 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -942,8 +942,13 @@ void __init trap_init(void)
942#endif 942#endif
943 set_intr_gate(19, &simd_coprocessor_error); 943 set_intr_gate(19, &simd_coprocessor_error);
944 944
945 /* Reserve all the builtin and the syscall vector: */
946 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
947 set_bit(i, used_vectors);
948
945#ifdef CONFIG_IA32_EMULATION 949#ifdef CONFIG_IA32_EMULATION
946 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall); 950 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
951 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
947#endif 952#endif
948 953
949#ifdef CONFIG_X86_32 954#ifdef CONFIG_X86_32
@@ -960,14 +965,9 @@ void __init trap_init(void)
960 } 965 }
961 966
962 set_system_trap_gate(SYSCALL_VECTOR, &system_call); 967 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
968 set_bit(SYSCALL_VECTOR, used_vectors);
963#endif 969#endif
964 970
965 /* Reserve all the builtin and the syscall vector: */
966 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
967 set_bit(i, used_vectors);
968
969 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
970
971 /* 971 /*
972 * Should be a barrier for any external CPU state: 972 * Should be a barrier for any external CPU state:
973 */ 973 */
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index 5ec7ae366615..c6acc6326374 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -10,6 +10,7 @@
10#include <linux/bootmem.h> /* max_low_pfn */ 10#include <linux/bootmem.h> /* max_low_pfn */
11#include <linux/kprobes.h> /* __kprobes, ... */ 11#include <linux/kprobes.h> /* __kprobes, ... */
12#include <linux/mmiotrace.h> /* kmmio_handler, ... */ 12#include <linux/mmiotrace.h> /* kmmio_handler, ... */
13#include <linux/perf_counter.h> /* perf_swcounter_event */
13 14
14#include <asm/traps.h> /* dotraplinkage, ... */ 15#include <asm/traps.h> /* dotraplinkage, ... */
15#include <asm/pgalloc.h> /* pgd_*(), ... */ 16#include <asm/pgalloc.h> /* pgd_*(), ... */
@@ -1013,6 +1014,8 @@ do_page_fault(struct pt_regs *regs, unsigned long error_code)
1013 if (unlikely(error_code & PF_RSVD)) 1014 if (unlikely(error_code & PF_RSVD))
1014 pgtable_bad(regs, error_code, address); 1015 pgtable_bad(regs, error_code, address);
1015 1016
1017 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
1018
1016 /* 1019 /*
1017 * If we're in an interrupt, have no user context or are running 1020 * If we're in an interrupt, have no user context or are running
1018 * in an atomic region then we must not take the fault: 1021 * in an atomic region then we must not take the fault:
@@ -1106,10 +1109,15 @@ good_area:
1106 return; 1109 return;
1107 } 1110 }
1108 1111
1109 if (fault & VM_FAULT_MAJOR) 1112 if (fault & VM_FAULT_MAJOR) {
1110 tsk->maj_flt++; 1113 tsk->maj_flt++;
1111 else 1114 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
1115 regs, address);
1116 } else {
1112 tsk->min_flt++; 1117 tsk->min_flt++;
1118 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
1119 regs, address);
1120 }
1113 1121
1114 check_v8086_mode(regs, address, tsk); 1122 check_v8086_mode(regs, address, tsk);
1115 1123
diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c
index 3b285e656e27..b07dd8d0b321 100644
--- a/arch/x86/oprofile/nmi_int.c
+++ b/arch/x86/oprofile/nmi_int.c
@@ -40,8 +40,9 @@ static int profile_exceptions_notify(struct notifier_block *self,
40 40
41 switch (val) { 41 switch (val) {
42 case DIE_NMI: 42 case DIE_NMI:
43 if (model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu))) 43 case DIE_NMI_IPI:
44 ret = NOTIFY_STOP; 44 model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu));
45 ret = NOTIFY_STOP;
45 break; 46 break;
46 default: 47 default:
47 break; 48 break;
@@ -134,7 +135,7 @@ static void nmi_cpu_setup(void *dummy)
134static struct notifier_block profile_exceptions_nb = { 135static struct notifier_block profile_exceptions_nb = {
135 .notifier_call = profile_exceptions_notify, 136 .notifier_call = profile_exceptions_notify,
136 .next = NULL, 137 .next = NULL,
137 .priority = 0 138 .priority = 2
138}; 139};
139 140
140static int nmi_setup(void) 141static int nmi_setup(void)
diff --git a/arch/x86/oprofile/op_model_ppro.c b/arch/x86/oprofile/op_model_ppro.c
index 10131fbdaada..4da7230b3d17 100644
--- a/arch/x86/oprofile/op_model_ppro.c
+++ b/arch/x86/oprofile/op_model_ppro.c
@@ -18,7 +18,7 @@
18#include <asm/msr.h> 18#include <asm/msr.h>
19#include <asm/apic.h> 19#include <asm/apic.h>
20#include <asm/nmi.h> 20#include <asm/nmi.h>
21#include <asm/intel_arch_perfmon.h> 21#include <asm/perf_counter.h>
22 22
23#include "op_x86_model.h" 23#include "op_x86_model.h"
24#include "op_counter.h" 24#include "op_counter.h"
@@ -136,6 +136,13 @@ static int ppro_check_ctrs(struct pt_regs * const regs,
136 u64 val; 136 u64 val;
137 int i; 137 int i;
138 138
139 /*
140 * This can happen if perf counters are in use when
141 * we steal the die notifier NMI.
142 */
143 if (unlikely(!reset_value))
144 goto out;
145
139 for (i = 0 ; i < num_counters; ++i) { 146 for (i = 0 ; i < num_counters; ++i) {
140 if (!reset_value[i]) 147 if (!reset_value[i])
141 continue; 148 continue;
@@ -146,6 +153,7 @@ static int ppro_check_ctrs(struct pt_regs * const regs,
146 } 153 }
147 } 154 }
148 155
156out:
149 /* Only P6 based Pentium M need to re-unmask the apic vector but it 157 /* Only P6 based Pentium M need to re-unmask the apic vector but it
150 * doesn't hurt other P6 variant */ 158 * doesn't hurt other P6 variant */
151 apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED); 159 apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
diff --git a/arch/x86/vdso/vdso32-setup.c b/arch/x86/vdso/vdso32-setup.c
index 1241f118ab56..58bc00f68b12 100644
--- a/arch/x86/vdso/vdso32-setup.c
+++ b/arch/x86/vdso/vdso32-setup.c
@@ -338,6 +338,8 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
338 } 338 }
339 } 339 }
340 340
341 current->mm->context.vdso = (void *)addr;
342
341 if (compat_uses_vma || !compat) { 343 if (compat_uses_vma || !compat) {
342 /* 344 /*
343 * MAYWRITE to allow gdb to COW and set breakpoints 345 * MAYWRITE to allow gdb to COW and set breakpoints
@@ -358,11 +360,13 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
358 goto up_fail; 360 goto up_fail;
359 } 361 }
360 362
361 current->mm->context.vdso = (void *)addr;
362 current_thread_info()->sysenter_return = 363 current_thread_info()->sysenter_return =
363 VDSO32_SYMBOL(addr, SYSENTER_RETURN); 364 VDSO32_SYMBOL(addr, SYSENTER_RETURN);
364 365
365 up_fail: 366 up_fail:
367 if (ret)
368 current->mm->context.vdso = NULL;
369
366 up_write(&mm->mmap_sem); 370 up_write(&mm->mmap_sem);
367 371
368 return ret; 372 return ret;
diff --git a/arch/x86/vdso/vma.c b/arch/x86/vdso/vma.c
index cac083386e03..21e1aeb9f3ea 100644
--- a/arch/x86/vdso/vma.c
+++ b/arch/x86/vdso/vma.c
@@ -116,15 +116,18 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
116 goto up_fail; 116 goto up_fail;
117 } 117 }
118 118
119 current->mm->context.vdso = (void *)addr;
120
119 ret = install_special_mapping(mm, addr, vdso_size, 121 ret = install_special_mapping(mm, addr, vdso_size,
120 VM_READ|VM_EXEC| 122 VM_READ|VM_EXEC|
121 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC| 123 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
122 VM_ALWAYSDUMP, 124 VM_ALWAYSDUMP,
123 vdso_pages); 125 vdso_pages);
124 if (ret) 126 if (ret) {
127 current->mm->context.vdso = NULL;
125 goto up_fail; 128 goto up_fail;
129 }
126 130
127 current->mm->context.vdso = (void *)addr;
128up_fail: 131up_fail:
129 up_write(&mm->mmap_sem); 132 up_write(&mm->mmap_sem);
130 return ret; 133 return ret;
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-04 14:25:57 -0400