2 * Performance events x86 architecture header
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
9 * Copyright (C) 2009 Intel Corporation, <markus.t.metzger@intel.com>
10 * Copyright (C) 2009 Google, Inc., Stephane Eranian
12 * For licencing details see kernel-base/COPYING
15 #include <linux/perf_event.h>
17 /* To enable MSR tracing please use the generic trace points. */
21 * register -------------------------------
22 * | HT | no HT | HT | no HT |
23 *-----------------------------------------
24 * offcore | core | core | cpu | core |
25 * lbr_sel | core | core | cpu | core |
26 * ld_lat | cpu | core | cpu | core |
27 *-----------------------------------------
29 * Given that there is a small number of shared regs,
30 * we can pre-allocate their slot in the per-cpu
31 * per-core reg tables.
34 EXTRA_REG_NONE = -1, /* not used */
36 EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */
37 EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */
38 EXTRA_REG_LBR = 2, /* lbr_select */
39 EXTRA_REG_LDLAT = 3, /* ld_lat_threshold */
40 EXTRA_REG_FE = 4, /* fe_* */
42 EXTRA_REG_MAX /* number of entries needed */
45 struct event_constraint {
47 unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
57 * struct hw_perf_event.flags flags
59 #define PERF_X86_EVENT_PEBS_LDLAT 0x0001 /* ld+ldlat data address sampling */
60 #define PERF_X86_EVENT_PEBS_ST 0x0002 /* st data address sampling */
61 #define PERF_X86_EVENT_PEBS_ST_HSW 0x0004 /* haswell style datala, store */
62 #define PERF_X86_EVENT_COMMITTED 0x0008 /* event passed commit_txn */
63 #define PERF_X86_EVENT_PEBS_LD_HSW 0x0010 /* haswell style datala, load */
64 #define PERF_X86_EVENT_PEBS_NA_HSW 0x0020 /* haswell style datala, unknown */
65 #define PERF_X86_EVENT_EXCL 0x0040 /* HT exclusivity on counter */
66 #define PERF_X86_EVENT_DYNAMIC 0x0080 /* dynamic alloc'd constraint */
67 #define PERF_X86_EVENT_RDPMC_ALLOWED 0x0100 /* grant rdpmc permission */
68 #define PERF_X86_EVENT_EXCL_ACCT 0x0200 /* accounted EXCL event */
69 #define PERF_X86_EVENT_AUTO_RELOAD 0x0400 /* use PEBS auto-reload */
70 #define PERF_X86_EVENT_FREERUNNING 0x0800 /* use freerunning PEBS */
74 int nb_id; /* NorthBridge id */
75 int refcnt; /* reference count */
76 struct perf_event *owners[X86_PMC_IDX_MAX];
77 struct event_constraint event_constraints[X86_PMC_IDX_MAX];
80 /* The maximal number of PEBS events: */
81 #define MAX_PEBS_EVENTS 8
82 #define PEBS_COUNTER_MASK ((1ULL << MAX_PEBS_EVENTS) - 1)
85 * Flags PEBS can handle without an PMI.
87 * TID can only be handled by flushing at context switch.
88 * REGS_USER can be handled for events limited to ring 3.
91 #define PEBS_FREERUNNING_FLAGS \
92 (PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
93 PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
94 PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
95 PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
96 PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER)
99 * A debug store configuration.
101 * We only support architectures that use 64bit fields.
106 u64 bts_absolute_maximum;
107 u64 bts_interrupt_threshold;
108 u64 pebs_buffer_base;
110 u64 pebs_absolute_maximum;
111 u64 pebs_interrupt_threshold;
112 u64 pebs_event_reset[MAX_PEBS_EVENTS];
125 PERF_REG_X86_FLAGS | \
136 * Per register state.
139 raw_spinlock_t lock; /* per-core: protect structure */
140 u64 config; /* extra MSR config */
141 u64 reg; /* extra MSR number */
142 atomic_t ref; /* reference count */
148 * Used to coordinate shared registers between HT threads or
149 * among events on a single PMU.
151 struct intel_shared_regs {
152 struct er_account regs[EXTRA_REG_MAX];
153 int refcnt; /* per-core: #HT threads */
154 unsigned core_id; /* per-core: core id */
157 enum intel_excl_state_type {
158 INTEL_EXCL_UNUSED = 0, /* counter is unused */
159 INTEL_EXCL_SHARED = 1, /* counter can be used by both threads */
160 INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
163 struct intel_excl_states {
164 enum intel_excl_state_type state[X86_PMC_IDX_MAX];
165 bool sched_started; /* true if scheduling has started */
168 struct intel_excl_cntrs {
171 struct intel_excl_states states[2];
174 u16 has_exclusive[2];
175 u32 exclusive_present;
178 int refcnt; /* per-core: #HT threads */
179 unsigned core_id; /* per-core: core id */
182 #define MAX_LBR_ENTRIES 32
185 X86_PERF_KFREE_SHARED = 0,
186 X86_PERF_KFREE_EXCL = 1,
190 struct cpu_hw_events {
192 * Generic x86 PMC bits
194 struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
195 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
196 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
199 int n_events; /* the # of events in the below arrays */
200 int n_added; /* the # last events in the below arrays;
201 they've never been enabled yet */
202 int n_txn; /* the # last events in the below arrays;
203 added in the current transaction */
204 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
205 u64 tags[X86_PMC_IDX_MAX];
207 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
208 struct event_constraint *event_constraint[X86_PMC_IDX_MAX];
210 int n_excl; /* the number of exclusive events */
212 unsigned int txn_flags;
216 * Intel DebugStore bits
218 struct debug_store *ds;
227 struct perf_branch_stack lbr_stack;
228 struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
229 struct er_account *lbr_sel;
233 * Intel host/guest exclude bits
235 u64 intel_ctrl_guest_mask;
236 u64 intel_ctrl_host_mask;
237 struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
240 * Intel checkpoint mask
245 * manage shared (per-core, per-cpu) registers
246 * used on Intel NHM/WSM/SNB
248 struct intel_shared_regs *shared_regs;
250 * manage exclusive counter access between hyperthread
252 struct event_constraint *constraint_list; /* in enable order */
253 struct intel_excl_cntrs *excl_cntrs;
254 int excl_thread_id; /* 0 or 1 */
259 struct amd_nb *amd_nb;
260 /* Inverted mask of bits to clear in the perf_ctr ctrl registers */
261 u64 perf_ctr_virt_mask;
263 void *kfree_on_online[X86_PERF_KFREE_MAX];
266 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) {\
267 { .idxmsk64 = (n) }, \
275 #define EVENT_CONSTRAINT(c, n, m) \
276 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
278 #define INTEL_EXCLEVT_CONSTRAINT(c, n) \
279 __EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
280 0, PERF_X86_EVENT_EXCL)
283 * The overlap flag marks event constraints with overlapping counter
284 * masks. This is the case if the counter mask of such an event is not
285 * a subset of any other counter mask of a constraint with an equal or
286 * higher weight, e.g.:
288 * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
289 * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
290 * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
292 * The event scheduler may not select the correct counter in the first
293 * cycle because it needs to know which subsequent events will be
294 * scheduled. It may fail to schedule the events then. So we set the
295 * overlap flag for such constraints to give the scheduler a hint which
296 * events to select for counter rescheduling.
298 * Care must be taken as the rescheduling algorithm is O(n!) which
299 * will increase scheduling cycles for an over-committed system
300 * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros
301 * and its counter masks must be kept at a minimum.
303 #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \
304 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
307 * Constraint on the Event code.
309 #define INTEL_EVENT_CONSTRAINT(c, n) \
310 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
313 * Constraint on the Event code + UMask + fixed-mask
315 * filter mask to validate fixed counter events.
316 * the following filters disqualify for fixed counters:
321 * - in_tx_checkpointed
322 * The other filters are supported by fixed counters.
323 * The any-thread option is supported starting with v3.
325 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
326 #define FIXED_EVENT_CONSTRAINT(c, n) \
327 EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
330 * Constraint on the Event code + UMask
332 #define INTEL_UEVENT_CONSTRAINT(c, n) \
333 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
335 /* Constraint on specific umask bit only + event */
336 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n) \
337 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
339 /* Like UEVENT_CONSTRAINT, but match flags too */
340 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \
341 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
343 #define INTEL_EXCLUEVT_CONSTRAINT(c, n) \
344 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
345 HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
347 #define INTEL_PLD_CONSTRAINT(c, n) \
348 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
349 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
351 #define INTEL_PST_CONSTRAINT(c, n) \
352 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
353 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
355 /* Event constraint, but match on all event flags too. */
356 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
357 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
359 /* Check only flags, but allow all event/umask */
360 #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
361 EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
363 /* Check flags and event code, and set the HSW store flag */
364 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
365 __EVENT_CONSTRAINT(code, n, \
366 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
367 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
369 /* Check flags and event code, and set the HSW load flag */
370 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
371 __EVENT_CONSTRAINT(code, n, \
372 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
373 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
375 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
376 __EVENT_CONSTRAINT(code, n, \
377 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
379 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
381 /* Check flags and event code/umask, and set the HSW store flag */
382 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
383 __EVENT_CONSTRAINT(code, n, \
384 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
385 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
387 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
388 __EVENT_CONSTRAINT(code, n, \
389 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
391 PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
393 /* Check flags and event code/umask, and set the HSW load flag */
394 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
395 __EVENT_CONSTRAINT(code, n, \
396 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
397 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
399 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
400 __EVENT_CONSTRAINT(code, n, \
401 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
403 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
405 /* Check flags and event code/umask, and set the HSW N/A flag */
406 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
407 __EVENT_CONSTRAINT(code, n, \
408 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
409 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
413 * We define the end marker as having a weight of -1
414 * to enable blacklisting of events using a counter bitmask
415 * of zero and thus a weight of zero.
416 * The end marker has a weight that cannot possibly be
417 * obtained from counting the bits in the bitmask.
419 #define EVENT_CONSTRAINT_END { .weight = -1 }
422 * Check for end marker with weight == -1
424 #define for_each_event_constraint(e, c) \
425 for ((e) = (c); (e)->weight != -1; (e)++)
428 * Extra registers for specific events.
430 * Some events need large masks and require external MSRs.
431 * Those extra MSRs end up being shared for all events on
432 * a PMU and sometimes between PMU of sibling HT threads.
433 * In either case, the kernel needs to handle conflicting
434 * accesses to those extra, shared, regs. The data structure
435 * to manage those registers is stored in cpu_hw_event.
442 int idx; /* per_xxx->regs[] reg index */
443 bool extra_msr_access;
446 #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
449 .config_mask = (m), \
450 .valid_mask = (vm), \
451 .idx = EXTRA_REG_##i, \
452 .extra_msr_access = true, \
455 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
456 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
458 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
459 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
460 ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
462 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
463 INTEL_UEVENT_EXTRA_REG(c, \
464 MSR_PEBS_LD_LAT_THRESHOLD, \
468 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
470 union perf_capabilities {
478 * PMU supports separate counter range for writing
481 u64 full_width_write:1;
486 struct x86_pmu_quirk {
487 struct x86_pmu_quirk *next;
491 union x86_pmu_config {
512 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
515 x86_lbr_exclusive_lbr,
516 x86_lbr_exclusive_bts,
517 x86_lbr_exclusive_pt,
518 x86_lbr_exclusive_max,
522 * struct x86_pmu - generic x86 pmu
526 * Generic x86 PMC bits
530 int (*handle_irq)(struct pt_regs *);
531 void (*disable_all)(void);
532 void (*enable_all)(int added);
533 void (*enable)(struct perf_event *);
534 void (*disable)(struct perf_event *);
535 void (*add)(struct perf_event *);
536 void (*del)(struct perf_event *);
537 int (*hw_config)(struct perf_event *event);
538 int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
541 int (*addr_offset)(int index, bool eventsel);
542 int (*rdpmc_index)(int index);
543 u64 (*event_map)(int);
546 int num_counters_fixed;
550 unsigned long events_maskl;
551 unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
556 struct event_constraint *
557 (*get_event_constraints)(struct cpu_hw_events *cpuc,
559 struct perf_event *event);
561 void (*put_event_constraints)(struct cpu_hw_events *cpuc,
562 struct perf_event *event);
564 void (*start_scheduling)(struct cpu_hw_events *cpuc);
566 void (*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
568 void (*stop_scheduling)(struct cpu_hw_events *cpuc);
570 struct event_constraint *event_constraints;
571 struct x86_pmu_quirk *quirks;
572 int perfctr_second_write;
574 unsigned (*limit_period)(struct perf_event *event, unsigned l);
579 int attr_rdpmc_broken;
581 struct attribute **format_attrs;
582 struct attribute **event_attrs;
583 struct attribute **caps_attrs;
585 ssize_t (*events_sysfs_show)(char *page, u64 config);
586 struct attribute **cpu_events;
588 unsigned long attr_freeze_on_smi;
589 struct attribute **attrs;
594 int (*cpu_prepare)(int cpu);
595 void (*cpu_starting)(int cpu);
596 void (*cpu_dying)(int cpu);
597 void (*cpu_dead)(int cpu);
599 void (*check_microcode)(void);
600 void (*sched_task)(struct perf_event_context *ctx,
604 * Intel Arch Perfmon v2+
607 union perf_capabilities intel_cap;
610 * Intel DebugStore bits
619 int pebs_record_size;
620 int pebs_buffer_size;
621 void (*drain_pebs)(struct pt_regs *regs);
622 struct event_constraint *pebs_constraints;
623 void (*pebs_aliases)(struct perf_event *event);
625 unsigned long free_running_flags;
630 unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
631 int lbr_nr; /* hardware stack size */
632 u64 lbr_sel_mask; /* LBR_SELECT valid bits */
633 const int *lbr_sel_map; /* lbr_select mappings */
634 bool lbr_double_abort; /* duplicated lbr aborts */
635 bool lbr_pt_coexist; /* (LBR|BTS) may coexist with PT */
638 * Intel PT/LBR/BTS are exclusive
640 atomic_t lbr_exclusive[x86_lbr_exclusive_max];
645 unsigned int amd_nb_constraints : 1;
648 * Extra registers for events
650 struct extra_reg *extra_regs;
654 * Intel host/guest support (KVM)
656 struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
659 struct x86_perf_task_context {
660 u64 lbr_from[MAX_LBR_ENTRIES];
661 u64 lbr_to[MAX_LBR_ENTRIES];
662 u64 lbr_info[MAX_LBR_ENTRIES];
664 int lbr_callstack_users;
668 #define x86_add_quirk(func_) \
670 static struct x86_pmu_quirk __quirk __initdata = { \
673 __quirk.next = x86_pmu.quirks; \
674 x86_pmu.quirks = &__quirk; \
680 #define PMU_FL_NO_HT_SHARING 0x1 /* no hyper-threading resource sharing */
681 #define PMU_FL_HAS_RSP_1 0x2 /* has 2 equivalent offcore_rsp regs */
682 #define PMU_FL_EXCL_CNTRS 0x4 /* has exclusive counter requirements */
683 #define PMU_FL_EXCL_ENABLED 0x8 /* exclusive counter active */
685 #define EVENT_VAR(_id) event_attr_##_id
686 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
688 #define EVENT_ATTR(_name, _id) \
689 static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
690 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
691 .id = PERF_COUNT_HW_##_id, \
695 #define EVENT_ATTR_STR(_name, v, str) \
696 static struct perf_pmu_events_attr event_attr_##v = { \
697 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
702 #define EVENT_ATTR_STR_HT(_name, v, noht, ht) \
703 static struct perf_pmu_events_ht_attr event_attr_##v = { \
704 .attr = __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
706 .event_str_noht = noht, \
707 .event_str_ht = ht, \
710 extern struct x86_pmu x86_pmu __read_mostly;
712 static inline bool x86_pmu_has_lbr_callstack(void)
714 return x86_pmu.lbr_sel_map &&
715 x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
718 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
720 int x86_perf_event_set_period(struct perf_event *event);
723 * Generalized hw caching related hw_event table, filled
724 * in on a per model basis. A value of 0 means
725 * 'not supported', -1 means 'hw_event makes no sense on
726 * this CPU', any other value means the raw hw_event
730 #define C(x) PERF_COUNT_HW_CACHE_##x
732 extern u64 __read_mostly hw_cache_event_ids
733 [PERF_COUNT_HW_CACHE_MAX]
734 [PERF_COUNT_HW_CACHE_OP_MAX]
735 [PERF_COUNT_HW_CACHE_RESULT_MAX];
736 extern u64 __read_mostly hw_cache_extra_regs
737 [PERF_COUNT_HW_CACHE_MAX]
738 [PERF_COUNT_HW_CACHE_OP_MAX]
739 [PERF_COUNT_HW_CACHE_RESULT_MAX];
741 u64 x86_perf_event_update(struct perf_event *event);
743 static inline unsigned int x86_pmu_config_addr(int index)
745 return x86_pmu.eventsel + (x86_pmu.addr_offset ?
746 x86_pmu.addr_offset(index, true) : index);
749 static inline unsigned int x86_pmu_event_addr(int index)
751 return x86_pmu.perfctr + (x86_pmu.addr_offset ?
752 x86_pmu.addr_offset(index, false) : index);
755 static inline int x86_pmu_rdpmc_index(int index)
757 return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
760 int x86_add_exclusive(unsigned int what);
762 void x86_del_exclusive(unsigned int what);
764 int x86_reserve_hardware(void);
766 void x86_release_hardware(void);
768 int x86_pmu_max_precise(void);
770 void hw_perf_lbr_event_destroy(struct perf_event *event);
772 int x86_setup_perfctr(struct perf_event *event);
774 int x86_pmu_hw_config(struct perf_event *event);
776 void x86_pmu_disable_all(void);
778 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
781 u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
783 if (hwc->extra_reg.reg)
784 wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
785 wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
788 void x86_pmu_enable_all(int added);
790 int perf_assign_events(struct event_constraint **constraints, int n,
791 int wmin, int wmax, int gpmax, int *assign);
792 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
794 void x86_pmu_stop(struct perf_event *event, int flags);
796 static inline void x86_pmu_disable_event(struct perf_event *event)
798 struct hw_perf_event *hwc = &event->hw;
800 wrmsrl(hwc->config_base, hwc->config);
803 void x86_pmu_enable_event(struct perf_event *event);
805 int x86_pmu_handle_irq(struct pt_regs *regs);
807 extern struct event_constraint emptyconstraint;
809 extern struct event_constraint unconstrained;
811 static inline bool kernel_ip(unsigned long ip)
814 return ip > PAGE_OFFSET;
821 * Not all PMUs provide the right context information to place the reported IP
822 * into full context. Specifically segment registers are typically not
825 * Assuming the address is a linear address (it is for IBS), we fake the CS and
826 * vm86 mode using the known zero-based code segment and 'fix up' the registers
829 * Intel PEBS/LBR appear to typically provide the effective address, nothing
830 * much we can do about that but pray and treat it like a linear address.
832 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
834 regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
835 if (regs->flags & X86_VM_MASK)
836 regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
840 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
841 ssize_t intel_event_sysfs_show(char *page, u64 config);
843 struct attribute **merge_attr(struct attribute **a, struct attribute **b);
845 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
847 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
850 #ifdef CONFIG_CPU_SUP_AMD
852 int amd_pmu_init(void);
854 #else /* CONFIG_CPU_SUP_AMD */
856 static inline int amd_pmu_init(void)
861 #endif /* CONFIG_CPU_SUP_AMD */
863 #ifdef CONFIG_CPU_SUP_INTEL
865 static inline bool intel_pmu_has_bts(struct perf_event *event)
867 if (event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS &&
868 !event->attr.freq && event->hw.sample_period == 1)
874 int intel_pmu_save_and_restart(struct perf_event *event);
876 struct event_constraint *
877 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
878 struct perf_event *event);
880 struct intel_shared_regs *allocate_shared_regs(int cpu);
882 int intel_pmu_init(void);
884 void init_debug_store_on_cpu(int cpu);
886 void fini_debug_store_on_cpu(int cpu);
888 void release_ds_buffers(void);
890 void reserve_ds_buffers(void);
892 extern struct event_constraint bts_constraint;
894 void intel_pmu_enable_bts(u64 config);
896 void intel_pmu_disable_bts(void);
898 int intel_pmu_drain_bts_buffer(void);
900 extern struct event_constraint intel_core2_pebs_event_constraints[];
902 extern struct event_constraint intel_atom_pebs_event_constraints[];
904 extern struct event_constraint intel_slm_pebs_event_constraints[];
906 extern struct event_constraint intel_glm_pebs_event_constraints[];
908 extern struct event_constraint intel_glp_pebs_event_constraints[];
910 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
912 extern struct event_constraint intel_westmere_pebs_event_constraints[];
914 extern struct event_constraint intel_snb_pebs_event_constraints[];
916 extern struct event_constraint intel_ivb_pebs_event_constraints[];
918 extern struct event_constraint intel_hsw_pebs_event_constraints[];
920 extern struct event_constraint intel_bdw_pebs_event_constraints[];
922 extern struct event_constraint intel_skl_pebs_event_constraints[];
924 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
926 void intel_pmu_pebs_add(struct perf_event *event);
928 void intel_pmu_pebs_del(struct perf_event *event);
930 void intel_pmu_pebs_enable(struct perf_event *event);
932 void intel_pmu_pebs_disable(struct perf_event *event);
934 void intel_pmu_pebs_enable_all(void);
936 void intel_pmu_pebs_disable_all(void);
938 void intel_pmu_pebs_sched_task(struct perf_event_context *ctx, bool sched_in);
940 void intel_ds_init(void);
942 void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);
944 u64 lbr_from_signext_quirk_wr(u64 val);
946 void intel_pmu_lbr_reset(void);
948 void intel_pmu_lbr_add(struct perf_event *event);
950 void intel_pmu_lbr_del(struct perf_event *event);
952 void intel_pmu_lbr_enable_all(bool pmi);
954 void intel_pmu_lbr_disable_all(void);
956 void intel_pmu_lbr_read(void);
958 void intel_pmu_lbr_init_core(void);
960 void intel_pmu_lbr_init_nhm(void);
962 void intel_pmu_lbr_init_atom(void);
964 void intel_pmu_lbr_init_slm(void);
966 void intel_pmu_lbr_init_snb(void);
968 void intel_pmu_lbr_init_hsw(void);
970 void intel_pmu_lbr_init_skl(void);
972 void intel_pmu_lbr_init_knl(void);
974 void intel_pmu_pebs_data_source_nhm(void);
976 void intel_pmu_pebs_data_source_skl(bool pmem);
978 int intel_pmu_setup_lbr_filter(struct perf_event *event);
980 void intel_pt_interrupt(void);
982 int intel_bts_interrupt(void);
984 void intel_bts_enable_local(void);
986 void intel_bts_disable_local(void);
988 int p4_pmu_init(void);
990 int p6_pmu_init(void);
992 int knc_pmu_init(void);
994 static inline int is_ht_workaround_enabled(void)
996 return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
999 #else /* CONFIG_CPU_SUP_INTEL */
1001 static inline void reserve_ds_buffers(void)
1005 static inline void release_ds_buffers(void)
1009 static inline int intel_pmu_init(void)
1014 static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
1019 static inline int is_ht_workaround_enabled(void)
1023 #endif /* CONFIG_CPU_SUP_INTEL */