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 #include <asm/fpu/xstate.h>
18 #include <asm/intel_ds.h>
21 /* To enable MSR tracing please use the generic trace points. */
25 * register -------------------------------
26 * | HT | no HT | HT | no HT |
27 *-----------------------------------------
28 * offcore | core | core | cpu | core |
29 * lbr_sel | core | core | cpu | core |
30 * ld_lat | cpu | core | cpu | core |
31 *-----------------------------------------
33 * Given that there is a small number of shared regs,
34 * we can pre-allocate their slot in the per-cpu
35 * per-core reg tables.
38 EXTRA_REG_NONE = -1, /* not used */
40 EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */
41 EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */
42 EXTRA_REG_LBR = 2, /* lbr_select */
43 EXTRA_REG_LDLAT = 3, /* ld_lat_threshold */
44 EXTRA_REG_FE = 4, /* fe_* */
45 EXTRA_REG_SNOOP_0 = 5, /* snoop response 0 */
46 EXTRA_REG_SNOOP_1 = 6, /* snoop response 1 */
48 EXTRA_REG_MAX /* number of entries needed */
51 struct event_constraint {
53 unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
64 static inline bool constraint_match(struct event_constraint *c, u64 ecode)
66 return ((ecode & c->cmask) - c->code) <= (u64)c->size;
69 #define PERF_ARCH(name, val) \
70 PERF_X86_EVENT_##name = val,
73 * struct hw_perf_event.flags flags
76 #include "perf_event_flags.h"
81 #define PERF_ARCH(name, val) \
82 static_assert((PERF_X86_EVENT_##name & PERF_EVENT_FLAG_ARCH) == \
83 PERF_X86_EVENT_##name);
85 #include "perf_event_flags.h"
89 static inline bool is_topdown_count(struct perf_event *event)
91 return event->hw.flags & PERF_X86_EVENT_TOPDOWN;
94 static inline bool is_metric_event(struct perf_event *event)
96 u64 config = event->attr.config;
98 return ((config & ARCH_PERFMON_EVENTSEL_EVENT) == 0) &&
99 ((config & INTEL_ARCH_EVENT_MASK) >= INTEL_TD_METRIC_RETIRING) &&
100 ((config & INTEL_ARCH_EVENT_MASK) <= INTEL_TD_METRIC_MAX);
103 static inline bool is_slots_event(struct perf_event *event)
105 return (event->attr.config & INTEL_ARCH_EVENT_MASK) == INTEL_TD_SLOTS;
108 static inline bool is_topdown_event(struct perf_event *event)
110 return is_metric_event(event) || is_slots_event(event);
114 int nb_id; /* NorthBridge id */
115 int refcnt; /* reference count */
116 struct perf_event *owners[X86_PMC_IDX_MAX];
117 struct event_constraint event_constraints[X86_PMC_IDX_MAX];
120 #define PEBS_COUNTER_MASK ((1ULL << MAX_PEBS_EVENTS) - 1)
121 #define PEBS_PMI_AFTER_EACH_RECORD BIT_ULL(60)
122 #define PEBS_OUTPUT_OFFSET 61
123 #define PEBS_OUTPUT_MASK (3ull << PEBS_OUTPUT_OFFSET)
124 #define PEBS_OUTPUT_PT (1ull << PEBS_OUTPUT_OFFSET)
125 #define PEBS_VIA_PT_MASK (PEBS_OUTPUT_PT | PEBS_PMI_AFTER_EACH_RECORD)
128 * Flags PEBS can handle without an PMI.
130 * TID can only be handled by flushing at context switch.
131 * REGS_USER can be handled for events limited to ring 3.
134 #define LARGE_PEBS_FLAGS \
135 (PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
136 PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
137 PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
138 PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
139 PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
140 PERF_SAMPLE_PERIOD | PERF_SAMPLE_CODE_PAGE_SIZE | \
141 PERF_SAMPLE_WEIGHT_TYPE)
143 #define PEBS_GP_REGS \
144 ((1ULL << PERF_REG_X86_AX) | \
145 (1ULL << PERF_REG_X86_BX) | \
146 (1ULL << PERF_REG_X86_CX) | \
147 (1ULL << PERF_REG_X86_DX) | \
148 (1ULL << PERF_REG_X86_DI) | \
149 (1ULL << PERF_REG_X86_SI) | \
150 (1ULL << PERF_REG_X86_SP) | \
151 (1ULL << PERF_REG_X86_BP) | \
152 (1ULL << PERF_REG_X86_IP) | \
153 (1ULL << PERF_REG_X86_FLAGS) | \
154 (1ULL << PERF_REG_X86_R8) | \
155 (1ULL << PERF_REG_X86_R9) | \
156 (1ULL << PERF_REG_X86_R10) | \
157 (1ULL << PERF_REG_X86_R11) | \
158 (1ULL << PERF_REG_X86_R12) | \
159 (1ULL << PERF_REG_X86_R13) | \
160 (1ULL << PERF_REG_X86_R14) | \
161 (1ULL << PERF_REG_X86_R15))
164 * Per register state.
167 raw_spinlock_t lock; /* per-core: protect structure */
168 u64 config; /* extra MSR config */
169 u64 reg; /* extra MSR number */
170 atomic_t ref; /* reference count */
176 * Used to coordinate shared registers between HT threads or
177 * among events on a single PMU.
179 struct intel_shared_regs {
180 struct er_account regs[EXTRA_REG_MAX];
181 int refcnt; /* per-core: #HT threads */
182 unsigned core_id; /* per-core: core id */
185 enum intel_excl_state_type {
186 INTEL_EXCL_UNUSED = 0, /* counter is unused */
187 INTEL_EXCL_SHARED = 1, /* counter can be used by both threads */
188 INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
191 struct intel_excl_states {
192 enum intel_excl_state_type state[X86_PMC_IDX_MAX];
193 bool sched_started; /* true if scheduling has started */
196 struct intel_excl_cntrs {
199 struct intel_excl_states states[2];
202 u16 has_exclusive[2];
203 u32 exclusive_present;
206 int refcnt; /* per-core: #HT threads */
207 unsigned core_id; /* per-core: core id */
210 struct x86_perf_task_context;
211 #define MAX_LBR_ENTRIES 32
214 LBR_FORMAT_32 = 0x00,
215 LBR_FORMAT_LIP = 0x01,
216 LBR_FORMAT_EIP = 0x02,
217 LBR_FORMAT_EIP_FLAGS = 0x03,
218 LBR_FORMAT_EIP_FLAGS2 = 0x04,
219 LBR_FORMAT_INFO = 0x05,
220 LBR_FORMAT_TIME = 0x06,
221 LBR_FORMAT_INFO2 = 0x07,
222 LBR_FORMAT_MAX_KNOWN = LBR_FORMAT_INFO2,
226 X86_PERF_KFREE_SHARED = 0,
227 X86_PERF_KFREE_EXCL = 1,
231 struct cpu_hw_events {
233 * Generic x86 PMC bits
235 struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
236 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
237 unsigned long dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
240 int n_events; /* the # of events in the below arrays */
241 int n_added; /* the # last events in the below arrays;
242 they've never been enabled yet */
243 int n_txn; /* the # last events in the below arrays;
244 added in the current transaction */
247 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
248 u64 tags[X86_PMC_IDX_MAX];
250 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
251 struct event_constraint *event_constraint[X86_PMC_IDX_MAX];
253 int n_excl; /* the number of exclusive events */
255 unsigned int txn_flags;
259 * Intel DebugStore bits
261 struct debug_store *ds;
270 /* Current super set of events hardware configuration */
272 u64 active_pebs_data_cfg;
273 int pebs_record_size;
275 /* Intel Fixed counter configuration */
277 u64 active_fixed_ctrl_val;
284 struct perf_branch_stack lbr_stack;
285 struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
287 struct er_account *lbr_sel;
288 struct er_account *lbr_ctl;
297 * Intel host/guest exclude bits
299 u64 intel_ctrl_guest_mask;
300 u64 intel_ctrl_host_mask;
301 struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
304 * Intel checkpoint mask
309 * manage shared (per-core, per-cpu) registers
310 * used on Intel NHM/WSM/SNB
312 struct intel_shared_regs *shared_regs;
314 * manage exclusive counter access between hyperthread
316 struct event_constraint *constraint_list; /* in enable order */
317 struct intel_excl_cntrs *excl_cntrs;
318 int excl_thread_id; /* 0 or 1 */
321 * SKL TSX_FORCE_ABORT shadow
328 /* number of accepted metrics events */
334 struct amd_nb *amd_nb;
335 int brs_active; /* BRS is enabled */
337 /* Inverted mask of bits to clear in the perf_ctr ctrl registers */
338 u64 perf_ctr_virt_mask;
339 int n_pair; /* Large increment events */
341 void *kfree_on_online[X86_PERF_KFREE_MAX];
346 #define __EVENT_CONSTRAINT_RANGE(c, e, n, m, w, o, f) { \
347 { .idxmsk64 = (n) }, \
356 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) \
357 __EVENT_CONSTRAINT_RANGE(c, c, n, m, w, o, f)
359 #define EVENT_CONSTRAINT(c, n, m) \
360 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
363 * The constraint_match() function only works for 'simple' event codes
364 * and not for extended (AMD64_EVENTSEL_EVENT) events codes.
366 #define EVENT_CONSTRAINT_RANGE(c, e, n, m) \
367 __EVENT_CONSTRAINT_RANGE(c, e, n, m, HWEIGHT(n), 0, 0)
369 #define INTEL_EXCLEVT_CONSTRAINT(c, n) \
370 __EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
371 0, PERF_X86_EVENT_EXCL)
374 * The overlap flag marks event constraints with overlapping counter
375 * masks. This is the case if the counter mask of such an event is not
376 * a subset of any other counter mask of a constraint with an equal or
377 * higher weight, e.g.:
379 * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
380 * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
381 * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
383 * The event scheduler may not select the correct counter in the first
384 * cycle because it needs to know which subsequent events will be
385 * scheduled. It may fail to schedule the events then. So we set the
386 * overlap flag for such constraints to give the scheduler a hint which
387 * events to select for counter rescheduling.
389 * Care must be taken as the rescheduling algorithm is O(n!) which
390 * will increase scheduling cycles for an over-committed system
391 * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros
392 * and its counter masks must be kept at a minimum.
394 #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \
395 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
398 * Constraint on the Event code.
400 #define INTEL_EVENT_CONSTRAINT(c, n) \
401 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
404 * Constraint on a range of Event codes
406 #define INTEL_EVENT_CONSTRAINT_RANGE(c, e, n) \
407 EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT)
410 * Constraint on the Event code + UMask + fixed-mask
412 * filter mask to validate fixed counter events.
413 * the following filters disqualify for fixed counters:
418 * - in_tx_checkpointed
419 * The other filters are supported by fixed counters.
420 * The any-thread option is supported starting with v3.
422 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
423 #define FIXED_EVENT_CONSTRAINT(c, n) \
424 EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
427 * The special metric counters do not actually exist. They are calculated from
428 * the combination of the FxCtr3 + MSR_PERF_METRICS.
430 * The special metric counters are mapped to a dummy offset for the scheduler.
431 * The sharing between multiple users of the same metric without multiplexing
432 * is not allowed, even though the hardware supports that in principle.
435 #define METRIC_EVENT_CONSTRAINT(c, n) \
436 EVENT_CONSTRAINT(c, (1ULL << (INTEL_PMC_IDX_METRIC_BASE + n)), \
437 INTEL_ARCH_EVENT_MASK)
440 * Constraint on the Event code + UMask
442 #define INTEL_UEVENT_CONSTRAINT(c, n) \
443 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
445 /* Constraint on specific umask bit only + event */
446 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n) \
447 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
449 /* Like UEVENT_CONSTRAINT, but match flags too */
450 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \
451 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
453 #define INTEL_EXCLUEVT_CONSTRAINT(c, n) \
454 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
455 HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
457 #define INTEL_PLD_CONSTRAINT(c, n) \
458 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
459 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
461 #define INTEL_PSD_CONSTRAINT(c, n) \
462 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
463 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_STLAT)
465 #define INTEL_PST_CONSTRAINT(c, n) \
466 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
467 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
469 #define INTEL_HYBRID_LAT_CONSTRAINT(c, n) \
470 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
471 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID)
473 /* Event constraint, but match on all event flags too. */
474 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
475 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
477 #define INTEL_FLAGS_EVENT_CONSTRAINT_RANGE(c, e, n) \
478 EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
480 /* Check only flags, but allow all event/umask */
481 #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
482 EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
484 /* Check flags and event code, and set the HSW store flag */
485 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
486 __EVENT_CONSTRAINT(code, n, \
487 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
488 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
490 /* Check flags and event code, and set the HSW load flag */
491 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
492 __EVENT_CONSTRAINT(code, n, \
493 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
494 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
496 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD_RANGE(code, end, n) \
497 __EVENT_CONSTRAINT_RANGE(code, end, n, \
498 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
499 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
501 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
502 __EVENT_CONSTRAINT(code, n, \
503 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
505 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
507 /* Check flags and event code/umask, and set the HSW store flag */
508 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
509 __EVENT_CONSTRAINT(code, n, \
510 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
511 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
513 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
514 __EVENT_CONSTRAINT(code, n, \
515 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
517 PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
519 /* Check flags and event code/umask, and set the HSW load flag */
520 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
521 __EVENT_CONSTRAINT(code, n, \
522 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
523 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
525 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
526 __EVENT_CONSTRAINT(code, n, \
527 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
529 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
531 /* Check flags and event code/umask, and set the HSW N/A flag */
532 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
533 __EVENT_CONSTRAINT(code, n, \
534 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
535 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
539 * We define the end marker as having a weight of -1
540 * to enable blacklisting of events using a counter bitmask
541 * of zero and thus a weight of zero.
542 * The end marker has a weight that cannot possibly be
543 * obtained from counting the bits in the bitmask.
545 #define EVENT_CONSTRAINT_END { .weight = -1 }
548 * Check for end marker with weight == -1
550 #define for_each_event_constraint(e, c) \
551 for ((e) = (c); (e)->weight != -1; (e)++)
554 * Extra registers for specific events.
556 * Some events need large masks and require external MSRs.
557 * Those extra MSRs end up being shared for all events on
558 * a PMU and sometimes between PMU of sibling HT threads.
559 * In either case, the kernel needs to handle conflicting
560 * accesses to those extra, shared, regs. The data structure
561 * to manage those registers is stored in cpu_hw_event.
568 int idx; /* per_xxx->regs[] reg index */
569 bool extra_msr_access;
572 #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
575 .config_mask = (m), \
576 .valid_mask = (vm), \
577 .idx = EXTRA_REG_##i, \
578 .extra_msr_access = true, \
581 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
582 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
584 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
585 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
586 ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
588 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
589 INTEL_UEVENT_EXTRA_REG(c, \
590 MSR_PEBS_LD_LAT_THRESHOLD, \
594 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
596 union perf_capabilities {
604 * PMU supports separate counter range for writing
607 u64 full_width_write:1;
610 u64 pebs_output_pt_available:1;
611 u64 pebs_timing_info:1;
612 u64 anythread_deprecated:1;
617 struct x86_pmu_quirk {
618 struct x86_pmu_quirk *next;
622 union x86_pmu_config {
643 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
646 x86_lbr_exclusive_lbr,
647 x86_lbr_exclusive_bts,
648 x86_lbr_exclusive_pt,
649 x86_lbr_exclusive_max,
652 #define PERF_PEBS_DATA_SOURCE_MAX 0x10
653 #define PERF_PEBS_DATA_SOURCE_MASK (PERF_PEBS_DATA_SOURCE_MAX - 1)
655 struct x86_hybrid_pmu {
659 cpumask_t supported_cpus;
660 union perf_capabilities intel_cap;
664 int num_counters_fixed;
665 struct event_constraint unconstrained;
667 u64 hw_cache_event_ids
668 [PERF_COUNT_HW_CACHE_MAX]
669 [PERF_COUNT_HW_CACHE_OP_MAX]
670 [PERF_COUNT_HW_CACHE_RESULT_MAX];
671 u64 hw_cache_extra_regs
672 [PERF_COUNT_HW_CACHE_MAX]
673 [PERF_COUNT_HW_CACHE_OP_MAX]
674 [PERF_COUNT_HW_CACHE_RESULT_MAX];
675 struct event_constraint *event_constraints;
676 struct event_constraint *pebs_constraints;
677 struct extra_reg *extra_regs;
679 unsigned int late_ack :1,
683 u64 pebs_data_source[PERF_PEBS_DATA_SOURCE_MAX];
686 static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)
688 return container_of(pmu, struct x86_hybrid_pmu, pmu);
691 extern struct static_key_false perf_is_hybrid;
692 #define is_hybrid() static_branch_unlikely(&perf_is_hybrid)
694 #define hybrid(_pmu, _field) \
696 typeof(&x86_pmu._field) __Fp = &x86_pmu._field; \
698 if (is_hybrid() && (_pmu)) \
699 __Fp = &hybrid_pmu(_pmu)->_field; \
704 #define hybrid_var(_pmu, _var) \
706 typeof(&_var) __Fp = &_var; \
708 if (is_hybrid() && (_pmu)) \
709 __Fp = &hybrid_pmu(_pmu)->_var; \
714 #define hybrid_bit(_pmu, _field) \
716 bool __Fp = x86_pmu._field; \
718 if (is_hybrid() && (_pmu)) \
719 __Fp = hybrid_pmu(_pmu)->_field; \
724 enum hybrid_pmu_type {
728 hybrid_big_small = hybrid_big | hybrid_small,
731 #define X86_HYBRID_PMU_ATOM_IDX 0
732 #define X86_HYBRID_PMU_CORE_IDX 1
734 #define X86_HYBRID_NUM_PMUS 2
737 * struct x86_pmu - generic x86 pmu
741 * Generic x86 PMC bits
745 int (*handle_irq)(struct pt_regs *);
746 void (*disable_all)(void);
747 void (*enable_all)(int added);
748 void (*enable)(struct perf_event *);
749 void (*disable)(struct perf_event *);
750 void (*assign)(struct perf_event *event, int idx);
751 void (*add)(struct perf_event *);
752 void (*del)(struct perf_event *);
753 void (*read)(struct perf_event *event);
754 int (*set_period)(struct perf_event *event);
755 u64 (*update)(struct perf_event *event);
756 int (*hw_config)(struct perf_event *event);
757 int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
760 int (*addr_offset)(int index, bool eventsel);
761 int (*rdpmc_index)(int index);
762 u64 (*event_map)(int);
765 int num_counters_fixed;
769 unsigned long events_maskl;
770 unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
775 struct event_constraint *
776 (*get_event_constraints)(struct cpu_hw_events *cpuc,
778 struct perf_event *event);
780 void (*put_event_constraints)(struct cpu_hw_events *cpuc,
781 struct perf_event *event);
783 void (*start_scheduling)(struct cpu_hw_events *cpuc);
785 void (*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
787 void (*stop_scheduling)(struct cpu_hw_events *cpuc);
789 struct event_constraint *event_constraints;
790 struct x86_pmu_quirk *quirks;
791 void (*limit_period)(struct perf_event *event, s64 *l);
793 /* PMI handler bits */
794 unsigned int late_ack :1,
800 int attr_rdpmc_broken;
802 struct attribute **format_attrs;
804 ssize_t (*events_sysfs_show)(char *page, u64 config);
805 const struct attribute_group **attr_update;
807 unsigned long attr_freeze_on_smi;
812 int (*cpu_prepare)(int cpu);
813 void (*cpu_starting)(int cpu);
814 void (*cpu_dying)(int cpu);
815 void (*cpu_dead)(int cpu);
817 void (*check_microcode)(void);
818 void (*sched_task)(struct perf_event_pmu_context *pmu_ctx,
822 * Intel Arch Perfmon v2+
825 union perf_capabilities intel_cap;
828 * Intel DebugStore bits
837 pebs_no_isolation :1,
840 int pebs_record_size;
841 int pebs_buffer_size;
843 void (*drain_pebs)(struct pt_regs *regs, struct perf_sample_data *data);
844 struct event_constraint *pebs_constraints;
845 void (*pebs_aliases)(struct perf_event *event);
846 u64 (*pebs_latency_data)(struct perf_event *event, u64 status);
847 unsigned long large_pebs_flags;
854 unsigned int lbr_tos, lbr_from, lbr_to,
855 lbr_info, lbr_nr; /* LBR base regs and size */
857 u64 lbr_sel_mask; /* LBR_SELECT valid bits */
858 u64 lbr_ctl_mask; /* LBR_CTL valid bits */
861 const int *lbr_sel_map; /* lbr_select mappings */
862 int *lbr_ctl_map; /* LBR_CTL mappings */
864 bool lbr_double_abort; /* duplicated lbr aborts */
865 bool lbr_pt_coexist; /* (LBR|BTS) may coexist with PT */
867 unsigned int lbr_has_info:1;
868 unsigned int lbr_has_tsx:1;
869 unsigned int lbr_from_flags:1;
870 unsigned int lbr_to_cycles:1;
873 * Intel Architectural LBR CPUID Enumeration
875 unsigned int lbr_depth_mask:8;
876 unsigned int lbr_deep_c_reset:1;
877 unsigned int lbr_lip:1;
878 unsigned int lbr_cpl:1;
879 unsigned int lbr_filter:1;
880 unsigned int lbr_call_stack:1;
881 unsigned int lbr_mispred:1;
882 unsigned int lbr_timed_lbr:1;
883 unsigned int lbr_br_type:1;
885 void (*lbr_reset)(void);
886 void (*lbr_read)(struct cpu_hw_events *cpuc);
887 void (*lbr_save)(void *ctx);
888 void (*lbr_restore)(void *ctx);
891 * Intel PT/LBR/BTS are exclusive
893 atomic_t lbr_exclusive[x86_lbr_exclusive_max];
898 int num_topdown_events;
901 * perf task context (i.e. struct perf_event_pmu_context::task_ctx_data)
902 * switch helper to bridge calls from perf/core to perf/x86.
903 * See struct pmu::swap_task_ctx() usage for examples;
905 void (*swap_task_ctx)(struct perf_event_pmu_context *prev_epc,
906 struct perf_event_pmu_context *next_epc);
911 unsigned int amd_nb_constraints : 1;
912 u64 perf_ctr_pair_en;
915 * Extra registers for events
917 struct extra_reg *extra_regs;
921 * Intel host/guest support (KVM)
923 struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr, void *data);
926 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
928 int (*check_period) (struct perf_event *event, u64 period);
930 int (*aux_output_match) (struct perf_event *event);
932 void (*filter)(struct pmu *pmu, int cpu, bool *ret);
936 * Most PMU capabilities are the same among different hybrid PMUs.
937 * The global x86_pmu saves the architecture capabilities, which
938 * are available for all PMUs. The hybrid_pmu only includes the
939 * unique capabilities.
942 struct x86_hybrid_pmu *hybrid_pmu;
943 u8 (*get_hybrid_cpu_type) (void);
946 struct x86_perf_task_context_opt {
947 int lbr_callstack_users;
952 struct x86_perf_task_context {
956 struct x86_perf_task_context_opt opt;
957 struct lbr_entry lbr[MAX_LBR_ENTRIES];
960 struct x86_perf_task_context_arch_lbr {
961 struct x86_perf_task_context_opt opt;
962 struct lbr_entry entries[];
966 * Add padding to guarantee the 64-byte alignment of the state buffer.
968 * The structure is dynamically allocated. The size of the LBR state may vary
969 * based on the number of LBR registers.
971 * Do not put anything after the LBR state.
973 struct x86_perf_task_context_arch_lbr_xsave {
974 struct x86_perf_task_context_opt opt;
977 struct xregs_state xsave;
979 struct fxregs_state i387;
980 struct xstate_header header;
981 struct arch_lbr_state lbr;
982 } __attribute__ ((packed, aligned (XSAVE_ALIGNMENT)));
986 #define x86_add_quirk(func_) \
988 static struct x86_pmu_quirk __quirk __initdata = { \
991 __quirk.next = x86_pmu.quirks; \
992 x86_pmu.quirks = &__quirk; \
998 #define PMU_FL_NO_HT_SHARING 0x1 /* no hyper-threading resource sharing */
999 #define PMU_FL_HAS_RSP_1 0x2 /* has 2 equivalent offcore_rsp regs */
1000 #define PMU_FL_EXCL_CNTRS 0x4 /* has exclusive counter requirements */
1001 #define PMU_FL_EXCL_ENABLED 0x8 /* exclusive counter active */
1002 #define PMU_FL_PEBS_ALL 0x10 /* all events are valid PEBS events */
1003 #define PMU_FL_TFA 0x20 /* deal with TSX force abort */
1004 #define PMU_FL_PAIR 0x40 /* merge counters for large incr. events */
1005 #define PMU_FL_INSTR_LATENCY 0x80 /* Support Instruction Latency in PEBS Memory Info Record */
1006 #define PMU_FL_MEM_LOADS_AUX 0x100 /* Require an auxiliary event for the complete memory info */
1007 #define PMU_FL_RETIRE_LATENCY 0x200 /* Support Retire Latency in PEBS */
1009 #define EVENT_VAR(_id) event_attr_##_id
1010 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
1012 #define EVENT_ATTR(_name, _id) \
1013 static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
1014 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
1015 .id = PERF_COUNT_HW_##_id, \
1016 .event_str = NULL, \
1019 #define EVENT_ATTR_STR(_name, v, str) \
1020 static struct perf_pmu_events_attr event_attr_##v = { \
1021 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
1026 #define EVENT_ATTR_STR_HT(_name, v, noht, ht) \
1027 static struct perf_pmu_events_ht_attr event_attr_##v = { \
1028 .attr = __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
1030 .event_str_noht = noht, \
1031 .event_str_ht = ht, \
1034 #define EVENT_ATTR_STR_HYBRID(_name, v, str, _pmu) \
1035 static struct perf_pmu_events_hybrid_attr event_attr_##v = { \
1036 .attr = __ATTR(_name, 0444, events_hybrid_sysfs_show, NULL),\
1042 #define FORMAT_HYBRID_PTR(_id) (&format_attr_hybrid_##_id.attr.attr)
1044 #define FORMAT_ATTR_HYBRID(_name, _pmu) \
1045 static struct perf_pmu_format_hybrid_attr format_attr_hybrid_##_name = {\
1046 .attr = __ATTR_RO(_name), \
1050 struct pmu *x86_get_pmu(unsigned int cpu);
1051 extern struct x86_pmu x86_pmu __read_mostly;
1053 DECLARE_STATIC_CALL(x86_pmu_set_period, *x86_pmu.set_period);
1054 DECLARE_STATIC_CALL(x86_pmu_update, *x86_pmu.update);
1056 static __always_inline struct x86_perf_task_context_opt *task_context_opt(void *ctx)
1058 if (static_cpu_has(X86_FEATURE_ARCH_LBR))
1059 return &((struct x86_perf_task_context_arch_lbr *)ctx)->opt;
1061 return &((struct x86_perf_task_context *)ctx)->opt;
1064 static inline bool x86_pmu_has_lbr_callstack(void)
1066 return x86_pmu.lbr_sel_map &&
1067 x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
1070 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
1071 DECLARE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
1073 int x86_perf_event_set_period(struct perf_event *event);
1076 * Generalized hw caching related hw_event table, filled
1077 * in on a per model basis. A value of 0 means
1078 * 'not supported', -1 means 'hw_event makes no sense on
1079 * this CPU', any other value means the raw hw_event
1083 #define C(x) PERF_COUNT_HW_CACHE_##x
1085 extern u64 __read_mostly hw_cache_event_ids
1086 [PERF_COUNT_HW_CACHE_MAX]
1087 [PERF_COUNT_HW_CACHE_OP_MAX]
1088 [PERF_COUNT_HW_CACHE_RESULT_MAX];
1089 extern u64 __read_mostly hw_cache_extra_regs
1090 [PERF_COUNT_HW_CACHE_MAX]
1091 [PERF_COUNT_HW_CACHE_OP_MAX]
1092 [PERF_COUNT_HW_CACHE_RESULT_MAX];
1094 u64 x86_perf_event_update(struct perf_event *event);
1096 static inline unsigned int x86_pmu_config_addr(int index)
1098 return x86_pmu.eventsel + (x86_pmu.addr_offset ?
1099 x86_pmu.addr_offset(index, true) : index);
1102 static inline unsigned int x86_pmu_event_addr(int index)
1104 return x86_pmu.perfctr + (x86_pmu.addr_offset ?
1105 x86_pmu.addr_offset(index, false) : index);
1108 static inline int x86_pmu_rdpmc_index(int index)
1110 return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
1113 bool check_hw_exists(struct pmu *pmu, int num_counters,
1114 int num_counters_fixed);
1116 int x86_add_exclusive(unsigned int what);
1118 void x86_del_exclusive(unsigned int what);
1120 int x86_reserve_hardware(void);
1122 void x86_release_hardware(void);
1124 int x86_pmu_max_precise(void);
1126 void hw_perf_lbr_event_destroy(struct perf_event *event);
1128 int x86_setup_perfctr(struct perf_event *event);
1130 int x86_pmu_hw_config(struct perf_event *event);
1132 void x86_pmu_disable_all(void);
1134 static inline bool has_amd_brs(struct hw_perf_event *hwc)
1136 return hwc->flags & PERF_X86_EVENT_AMD_BRS;
1139 static inline bool is_counter_pair(struct hw_perf_event *hwc)
1141 return hwc->flags & PERF_X86_EVENT_PAIR;
1144 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
1147 u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1149 if (hwc->extra_reg.reg)
1150 wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
1153 * Add enabled Merge event on next counter
1154 * if large increment event being enabled on this counter
1156 if (is_counter_pair(hwc))
1157 wrmsrl(x86_pmu_config_addr(hwc->idx + 1), x86_pmu.perf_ctr_pair_en);
1159 wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
1162 void x86_pmu_enable_all(int added);
1164 int perf_assign_events(struct event_constraint **constraints, int n,
1165 int wmin, int wmax, int gpmax, int *assign);
1166 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
1168 void x86_pmu_stop(struct perf_event *event, int flags);
1170 static inline void x86_pmu_disable_event(struct perf_event *event)
1172 u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1173 struct hw_perf_event *hwc = &event->hw;
1175 wrmsrl(hwc->config_base, hwc->config & ~disable_mask);
1177 if (is_counter_pair(hwc))
1178 wrmsrl(x86_pmu_config_addr(hwc->idx + 1), 0);
1181 void x86_pmu_enable_event(struct perf_event *event);
1183 int x86_pmu_handle_irq(struct pt_regs *regs);
1185 void x86_pmu_show_pmu_cap(int num_counters, int num_counters_fixed,
1188 extern struct event_constraint emptyconstraint;
1190 extern struct event_constraint unconstrained;
1192 static inline bool kernel_ip(unsigned long ip)
1194 #ifdef CONFIG_X86_32
1195 return ip > PAGE_OFFSET;
1197 return (long)ip < 0;
1202 * Not all PMUs provide the right context information to place the reported IP
1203 * into full context. Specifically segment registers are typically not
1206 * Assuming the address is a linear address (it is for IBS), we fake the CS and
1207 * vm86 mode using the known zero-based code segment and 'fix up' the registers
1210 * Intel PEBS/LBR appear to typically provide the effective address, nothing
1211 * much we can do about that but pray and treat it like a linear address.
1213 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
1215 regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
1216 if (regs->flags & X86_VM_MASK)
1217 regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
1222 * x86control flow change classification
1223 * x86control flow changes include branches, interrupts, traps, faults
1226 X86_BR_NONE = 0, /* unknown */
1228 X86_BR_USER = 1 << 0, /* branch target is user */
1229 X86_BR_KERNEL = 1 << 1, /* branch target is kernel */
1231 X86_BR_CALL = 1 << 2, /* call */
1232 X86_BR_RET = 1 << 3, /* return */
1233 X86_BR_SYSCALL = 1 << 4, /* syscall */
1234 X86_BR_SYSRET = 1 << 5, /* syscall return */
1235 X86_BR_INT = 1 << 6, /* sw interrupt */
1236 X86_BR_IRET = 1 << 7, /* return from interrupt */
1237 X86_BR_JCC = 1 << 8, /* conditional */
1238 X86_BR_JMP = 1 << 9, /* jump */
1239 X86_BR_IRQ = 1 << 10,/* hw interrupt or trap or fault */
1240 X86_BR_IND_CALL = 1 << 11,/* indirect calls */
1241 X86_BR_ABORT = 1 << 12,/* transaction abort */
1242 X86_BR_IN_TX = 1 << 13,/* in transaction */
1243 X86_BR_NO_TX = 1 << 14,/* not in transaction */
1244 X86_BR_ZERO_CALL = 1 << 15,/* zero length call */
1245 X86_BR_CALL_STACK = 1 << 16,/* call stack */
1246 X86_BR_IND_JMP = 1 << 17,/* indirect jump */
1248 X86_BR_TYPE_SAVE = 1 << 18,/* indicate to save branch type */
1252 #define X86_BR_PLM (X86_BR_USER | X86_BR_KERNEL)
1253 #define X86_BR_ANYTX (X86_BR_NO_TX | X86_BR_IN_TX)
1255 #define X86_BR_ANY \
1270 #define X86_BR_ALL (X86_BR_PLM | X86_BR_ANY)
1272 #define X86_BR_ANY_CALL \
1280 int common_branch_type(int type);
1281 int branch_type(unsigned long from, unsigned long to, int abort);
1282 int branch_type_fused(unsigned long from, unsigned long to, int abort,
1285 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
1286 ssize_t intel_event_sysfs_show(char *page, u64 config);
1288 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
1290 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
1292 ssize_t events_hybrid_sysfs_show(struct device *dev,
1293 struct device_attribute *attr,
1296 static inline bool fixed_counter_disabled(int i, struct pmu *pmu)
1298 u64 intel_ctrl = hybrid(pmu, intel_ctrl);
1300 return !(intel_ctrl >> (i + INTEL_PMC_IDX_FIXED));
1303 #ifdef CONFIG_CPU_SUP_AMD
1305 int amd_pmu_init(void);
1307 int amd_pmu_lbr_init(void);
1308 void amd_pmu_lbr_reset(void);
1309 void amd_pmu_lbr_read(void);
1310 void amd_pmu_lbr_add(struct perf_event *event);
1311 void amd_pmu_lbr_del(struct perf_event *event);
1312 void amd_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1313 void amd_pmu_lbr_enable_all(void);
1314 void amd_pmu_lbr_disable_all(void);
1315 int amd_pmu_lbr_hw_config(struct perf_event *event);
1317 #ifdef CONFIG_PERF_EVENTS_AMD_BRS
1319 #define AMD_FAM19H_BRS_EVENT 0xc4 /* RETIRED_TAKEN_BRANCH_INSTRUCTIONS */
1321 int amd_brs_init(void);
1322 void amd_brs_disable(void);
1323 void amd_brs_enable(void);
1324 void amd_brs_enable_all(void);
1325 void amd_brs_disable_all(void);
1326 void amd_brs_drain(void);
1327 void amd_brs_lopwr_init(void);
1328 int amd_brs_hw_config(struct perf_event *event);
1329 void amd_brs_reset(void);
1331 static inline void amd_pmu_brs_add(struct perf_event *event)
1333 struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1335 perf_sched_cb_inc(event->pmu);
1338 * No need to reset BRS because it is reset
1339 * on brs_enable() and it is saturating
1343 static inline void amd_pmu_brs_del(struct perf_event *event)
1345 struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1348 WARN_ON_ONCE(cpuc->lbr_users < 0);
1350 perf_sched_cb_dec(event->pmu);
1353 void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1355 static inline int amd_brs_init(void)
1359 static inline void amd_brs_disable(void) {}
1360 static inline void amd_brs_enable(void) {}
1361 static inline void amd_brs_drain(void) {}
1362 static inline void amd_brs_lopwr_init(void) {}
1363 static inline void amd_brs_disable_all(void) {}
1364 static inline int amd_brs_hw_config(struct perf_event *event)
1368 static inline void amd_brs_reset(void) {}
1370 static inline void amd_pmu_brs_add(struct perf_event *event)
1374 static inline void amd_pmu_brs_del(struct perf_event *event)
1378 static inline void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in)
1382 static inline void amd_brs_enable_all(void)
1388 #else /* CONFIG_CPU_SUP_AMD */
1390 static inline int amd_pmu_init(void)
1395 static inline int amd_brs_init(void)
1400 static inline void amd_brs_drain(void)
1404 static inline void amd_brs_enable_all(void)
1408 static inline void amd_brs_disable_all(void)
1411 #endif /* CONFIG_CPU_SUP_AMD */
1413 static inline int is_pebs_pt(struct perf_event *event)
1415 return !!(event->hw.flags & PERF_X86_EVENT_PEBS_VIA_PT);
1418 #ifdef CONFIG_CPU_SUP_INTEL
1420 static inline bool intel_pmu_has_bts_period(struct perf_event *event, u64 period)
1422 struct hw_perf_event *hwc = &event->hw;
1423 unsigned int hw_event, bts_event;
1425 if (event->attr.freq)
1428 hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
1429 bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
1431 return hw_event == bts_event && period == 1;
1434 static inline bool intel_pmu_has_bts(struct perf_event *event)
1436 struct hw_perf_event *hwc = &event->hw;
1438 return intel_pmu_has_bts_period(event, hwc->sample_period);
1441 static __always_inline void __intel_pmu_pebs_disable_all(void)
1443 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1446 static __always_inline void __intel_pmu_arch_lbr_disable(void)
1448 wrmsrl(MSR_ARCH_LBR_CTL, 0);
1451 static __always_inline void __intel_pmu_lbr_disable(void)
1455 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1456 debugctl &= ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1457 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1460 int intel_pmu_save_and_restart(struct perf_event *event);
1462 struct event_constraint *
1463 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
1464 struct perf_event *event);
1466 extern int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu);
1467 extern void intel_cpuc_finish(struct cpu_hw_events *cpuc);
1469 int intel_pmu_init(void);
1471 void init_debug_store_on_cpu(int cpu);
1473 void fini_debug_store_on_cpu(int cpu);
1475 void release_ds_buffers(void);
1477 void reserve_ds_buffers(void);
1479 void release_lbr_buffers(void);
1481 void reserve_lbr_buffers(void);
1483 extern struct event_constraint bts_constraint;
1484 extern struct event_constraint vlbr_constraint;
1486 void intel_pmu_enable_bts(u64 config);
1488 void intel_pmu_disable_bts(void);
1490 int intel_pmu_drain_bts_buffer(void);
1492 u64 adl_latency_data_small(struct perf_event *event, u64 status);
1494 u64 mtl_latency_data_small(struct perf_event *event, u64 status);
1496 extern struct event_constraint intel_core2_pebs_event_constraints[];
1498 extern struct event_constraint intel_atom_pebs_event_constraints[];
1500 extern struct event_constraint intel_slm_pebs_event_constraints[];
1502 extern struct event_constraint intel_glm_pebs_event_constraints[];
1504 extern struct event_constraint intel_glp_pebs_event_constraints[];
1506 extern struct event_constraint intel_grt_pebs_event_constraints[];
1508 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
1510 extern struct event_constraint intel_westmere_pebs_event_constraints[];
1512 extern struct event_constraint intel_snb_pebs_event_constraints[];
1514 extern struct event_constraint intel_ivb_pebs_event_constraints[];
1516 extern struct event_constraint intel_hsw_pebs_event_constraints[];
1518 extern struct event_constraint intel_bdw_pebs_event_constraints[];
1520 extern struct event_constraint intel_skl_pebs_event_constraints[];
1522 extern struct event_constraint intel_icl_pebs_event_constraints[];
1524 extern struct event_constraint intel_spr_pebs_event_constraints[];
1526 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
1528 void intel_pmu_pebs_add(struct perf_event *event);
1530 void intel_pmu_pebs_del(struct perf_event *event);
1532 void intel_pmu_pebs_enable(struct perf_event *event);
1534 void intel_pmu_pebs_disable(struct perf_event *event);
1536 void intel_pmu_pebs_enable_all(void);
1538 void intel_pmu_pebs_disable_all(void);
1540 void intel_pmu_pebs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1542 void intel_pmu_auto_reload_read(struct perf_event *event);
1544 void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
1546 void intel_ds_init(void);
1548 void intel_pmu_lbr_swap_task_ctx(struct perf_event_pmu_context *prev_epc,
1549 struct perf_event_pmu_context *next_epc);
1551 void intel_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1553 u64 lbr_from_signext_quirk_wr(u64 val);
1555 void intel_pmu_lbr_reset(void);
1557 void intel_pmu_lbr_reset_32(void);
1559 void intel_pmu_lbr_reset_64(void);
1561 void intel_pmu_lbr_add(struct perf_event *event);
1563 void intel_pmu_lbr_del(struct perf_event *event);
1565 void intel_pmu_lbr_enable_all(bool pmi);
1567 void intel_pmu_lbr_disable_all(void);
1569 void intel_pmu_lbr_read(void);
1571 void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc);
1573 void intel_pmu_lbr_read_64(struct cpu_hw_events *cpuc);
1575 void intel_pmu_lbr_save(void *ctx);
1577 void intel_pmu_lbr_restore(void *ctx);
1579 void intel_pmu_lbr_init_core(void);
1581 void intel_pmu_lbr_init_nhm(void);
1583 void intel_pmu_lbr_init_atom(void);
1585 void intel_pmu_lbr_init_slm(void);
1587 void intel_pmu_lbr_init_snb(void);
1589 void intel_pmu_lbr_init_hsw(void);
1591 void intel_pmu_lbr_init_skl(void);
1593 void intel_pmu_lbr_init_knl(void);
1595 void intel_pmu_lbr_init(void);
1597 void intel_pmu_arch_lbr_init(void);
1599 void intel_pmu_pebs_data_source_nhm(void);
1601 void intel_pmu_pebs_data_source_skl(bool pmem);
1603 void intel_pmu_pebs_data_source_adl(void);
1605 void intel_pmu_pebs_data_source_grt(void);
1607 void intel_pmu_pebs_data_source_mtl(void);
1609 int intel_pmu_setup_lbr_filter(struct perf_event *event);
1611 void intel_pt_interrupt(void);
1613 int intel_bts_interrupt(void);
1615 void intel_bts_enable_local(void);
1617 void intel_bts_disable_local(void);
1619 int p4_pmu_init(void);
1621 int p6_pmu_init(void);
1623 int knc_pmu_init(void);
1625 static inline int is_ht_workaround_enabled(void)
1627 return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
1630 #else /* CONFIG_CPU_SUP_INTEL */
1632 static inline void reserve_ds_buffers(void)
1636 static inline void release_ds_buffers(void)
1640 static inline void release_lbr_buffers(void)
1644 static inline void reserve_lbr_buffers(void)
1648 static inline int intel_pmu_init(void)
1653 static inline int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
1658 static inline void intel_cpuc_finish(struct cpu_hw_events *cpuc)
1662 static inline int is_ht_workaround_enabled(void)
1666 #endif /* CONFIG_CPU_SUP_INTEL */
1668 #if ((defined CONFIG_CPU_SUP_CENTAUR) || (defined CONFIG_CPU_SUP_ZHAOXIN))
1669 int zhaoxin_pmu_init(void);
1671 static inline int zhaoxin_pmu_init(void)
1675 #endif /*CONFIG_CPU_SUP_CENTAUR or CONFIG_CPU_SUP_ZHAOXIN*/