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 <pzijlstr@redhat.com>
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>
19 #define wrmsrl(msr, val) \
21 unsigned int _msr = (msr); \
23 trace_printk("wrmsrl(%x, %Lx)\n", (unsigned int)(_msr), \
24 (unsigned long long)(_val)); \
25 native_write_msr((_msr), (u32)(_val), (u32)(_val >> 32)); \
31 * register -------------------------------
32 * | HT | no HT | HT | no HT |
33 *-----------------------------------------
34 * offcore | core | core | cpu | core |
35 * lbr_sel | core | core | cpu | core |
36 * ld_lat | cpu | core | cpu | core |
37 *-----------------------------------------
39 * Given that there is a small number of shared regs,
40 * we can pre-allocate their slot in the per-cpu
41 * per-core reg tables.
44 EXTRA_REG_NONE = -1, /* not used */
46 EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */
47 EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */
48 EXTRA_REG_LBR = 2, /* lbr_select */
49 EXTRA_REG_LDLAT = 3, /* ld_lat_threshold */
51 EXTRA_REG_MAX /* number of entries needed */
54 struct event_constraint {
56 unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
66 * struct hw_perf_event.flags flags
68 #define PERF_X86_EVENT_PEBS_LDLAT 0x0001 /* ld+ldlat data address sampling */
69 #define PERF_X86_EVENT_PEBS_ST 0x0002 /* st data address sampling */
70 #define PERF_X86_EVENT_PEBS_ST_HSW 0x0004 /* haswell style datala, store */
71 #define PERF_X86_EVENT_COMMITTED 0x0008 /* event passed commit_txn */
72 #define PERF_X86_EVENT_PEBS_LD_HSW 0x0010 /* haswell style datala, load */
73 #define PERF_X86_EVENT_PEBS_NA_HSW 0x0020 /* haswell style datala, unknown */
74 #define PERF_X86_EVENT_EXCL 0x0040 /* HT exclusivity on counter */
75 #define PERF_X86_EVENT_DYNAMIC 0x0080 /* dynamic alloc'd constraint */
76 #define PERF_X86_EVENT_RDPMC_ALLOWED 0x0100 /* grant rdpmc permission */
77 #define PERF_X86_EVENT_EXCL_ACCT 0x0200 /* accounted EXCL event */
78 #define PERF_X86_EVENT_AUTO_RELOAD 0x0400 /* use PEBS auto-reload */
79 #define PERF_X86_EVENT_FREERUNNING 0x0800 /* use freerunning PEBS */
83 int nb_id; /* NorthBridge id */
84 int refcnt; /* reference count */
85 struct perf_event *owners[X86_PMC_IDX_MAX];
86 struct event_constraint event_constraints[X86_PMC_IDX_MAX];
89 /* The maximal number of PEBS events: */
90 #define MAX_PEBS_EVENTS 8
93 * Flags PEBS can handle without an PMI.
96 #define PEBS_FREERUNNING_FLAGS \
97 (PERF_SAMPLE_IP | PERF_SAMPLE_ADDR | \
98 PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
99 PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
100 PERF_SAMPLE_TRANSACTION)
103 * A debug store configuration.
105 * We only support architectures that use 64bit fields.
110 u64 bts_absolute_maximum;
111 u64 bts_interrupt_threshold;
112 u64 pebs_buffer_base;
114 u64 pebs_absolute_maximum;
115 u64 pebs_interrupt_threshold;
116 u64 pebs_event_reset[MAX_PEBS_EVENTS];
120 * Per register state.
123 raw_spinlock_t lock; /* per-core: protect structure */
124 u64 config; /* extra MSR config */
125 u64 reg; /* extra MSR number */
126 atomic_t ref; /* reference count */
132 * Used to coordinate shared registers between HT threads or
133 * among events on a single PMU.
135 struct intel_shared_regs {
136 struct er_account regs[EXTRA_REG_MAX];
137 int refcnt; /* per-core: #HT threads */
138 unsigned core_id; /* per-core: core id */
141 enum intel_excl_state_type {
142 INTEL_EXCL_UNUSED = 0, /* counter is unused */
143 INTEL_EXCL_SHARED = 1, /* counter can be used by both threads */
144 INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
147 struct intel_excl_states {
148 enum intel_excl_state_type state[X86_PMC_IDX_MAX];
149 bool sched_started; /* true if scheduling has started */
152 struct intel_excl_cntrs {
155 struct intel_excl_states states[2];
158 u16 has_exclusive[2];
159 u32 exclusive_present;
162 int refcnt; /* per-core: #HT threads */
163 unsigned core_id; /* per-core: core id */
166 #define MAX_LBR_ENTRIES 16
169 X86_PERF_KFREE_SHARED = 0,
170 X86_PERF_KFREE_EXCL = 1,
174 struct cpu_hw_events {
176 * Generic x86 PMC bits
178 struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
179 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
180 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
183 int n_events; /* the # of events in the below arrays */
184 int n_added; /* the # last events in the below arrays;
185 they've never been enabled yet */
186 int n_txn; /* the # last events in the below arrays;
187 added in the current transaction */
188 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
189 u64 tags[X86_PMC_IDX_MAX];
191 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
192 struct event_constraint *event_constraint[X86_PMC_IDX_MAX];
194 int n_excl; /* the number of exclusive events */
196 unsigned int group_flag;
200 * Intel DebugStore bits
202 struct debug_store *ds;
210 struct perf_branch_stack lbr_stack;
211 struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
212 struct er_account *lbr_sel;
216 * Intel host/guest exclude bits
218 u64 intel_ctrl_guest_mask;
219 u64 intel_ctrl_host_mask;
220 struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
223 * Intel checkpoint mask
228 * manage shared (per-core, per-cpu) registers
229 * used on Intel NHM/WSM/SNB
231 struct intel_shared_regs *shared_regs;
233 * manage exclusive counter access between hyperthread
235 struct event_constraint *constraint_list; /* in enable order */
236 struct intel_excl_cntrs *excl_cntrs;
237 int excl_thread_id; /* 0 or 1 */
242 struct amd_nb *amd_nb;
243 /* Inverted mask of bits to clear in the perf_ctr ctrl registers */
244 u64 perf_ctr_virt_mask;
246 void *kfree_on_online[X86_PERF_KFREE_MAX];
249 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) {\
250 { .idxmsk64 = (n) }, \
258 #define EVENT_CONSTRAINT(c, n, m) \
259 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
261 #define INTEL_EXCLEVT_CONSTRAINT(c, n) \
262 __EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
263 0, PERF_X86_EVENT_EXCL)
266 * The overlap flag marks event constraints with overlapping counter
267 * masks. This is the case if the counter mask of such an event is not
268 * a subset of any other counter mask of a constraint with an equal or
269 * higher weight, e.g.:
271 * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
272 * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
273 * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
275 * The event scheduler may not select the correct counter in the first
276 * cycle because it needs to know which subsequent events will be
277 * scheduled. It may fail to schedule the events then. So we set the
278 * overlap flag for such constraints to give the scheduler a hint which
279 * events to select for counter rescheduling.
281 * Care must be taken as the rescheduling algorithm is O(n!) which
282 * will increase scheduling cycles for an over-commited system
283 * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros
284 * and its counter masks must be kept at a minimum.
286 #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \
287 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
290 * Constraint on the Event code.
292 #define INTEL_EVENT_CONSTRAINT(c, n) \
293 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
296 * Constraint on the Event code + UMask + fixed-mask
298 * filter mask to validate fixed counter events.
299 * the following filters disqualify for fixed counters:
304 * - in_tx_checkpointed
305 * The other filters are supported by fixed counters.
306 * The any-thread option is supported starting with v3.
308 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
309 #define FIXED_EVENT_CONSTRAINT(c, n) \
310 EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
313 * Constraint on the Event code + UMask
315 #define INTEL_UEVENT_CONSTRAINT(c, n) \
316 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
318 /* Like UEVENT_CONSTRAINT, but match flags too */
319 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \
320 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
322 #define INTEL_EXCLUEVT_CONSTRAINT(c, n) \
323 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
324 HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
326 #define INTEL_PLD_CONSTRAINT(c, n) \
327 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
328 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
330 #define INTEL_PST_CONSTRAINT(c, n) \
331 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
332 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
334 /* Event constraint, but match on all event flags too. */
335 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
336 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
338 /* Check only flags, but allow all event/umask */
339 #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
340 EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
342 /* Check flags and event code, and set the HSW store flag */
343 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
344 __EVENT_CONSTRAINT(code, n, \
345 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
346 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
348 /* Check flags and event code, and set the HSW load flag */
349 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
350 __EVENT_CONSTRAINT(code, n, \
351 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
352 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
354 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
355 __EVENT_CONSTRAINT(code, n, \
356 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
358 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
360 /* Check flags and event code/umask, and set the HSW store flag */
361 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
362 __EVENT_CONSTRAINT(code, n, \
363 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
364 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
366 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
367 __EVENT_CONSTRAINT(code, n, \
368 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
370 PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
372 /* Check flags and event code/umask, and set the HSW load flag */
373 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
374 __EVENT_CONSTRAINT(code, n, \
375 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
376 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
378 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
379 __EVENT_CONSTRAINT(code, n, \
380 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
382 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
384 /* Check flags and event code/umask, and set the HSW N/A flag */
385 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
386 __EVENT_CONSTRAINT(code, n, \
387 INTEL_ARCH_EVENT_MASK|INTEL_ARCH_EVENT_MASK, \
388 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
392 * We define the end marker as having a weight of -1
393 * to enable blacklisting of events using a counter bitmask
394 * of zero and thus a weight of zero.
395 * The end marker has a weight that cannot possibly be
396 * obtained from counting the bits in the bitmask.
398 #define EVENT_CONSTRAINT_END { .weight = -1 }
401 * Check for end marker with weight == -1
403 #define for_each_event_constraint(e, c) \
404 for ((e) = (c); (e)->weight != -1; (e)++)
407 * Extra registers for specific events.
409 * Some events need large masks and require external MSRs.
410 * Those extra MSRs end up being shared for all events on
411 * a PMU and sometimes between PMU of sibling HT threads.
412 * In either case, the kernel needs to handle conflicting
413 * accesses to those extra, shared, regs. The data structure
414 * to manage those registers is stored in cpu_hw_event.
421 int idx; /* per_xxx->regs[] reg index */
422 bool extra_msr_access;
425 #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
428 .config_mask = (m), \
429 .valid_mask = (vm), \
430 .idx = EXTRA_REG_##i, \
431 .extra_msr_access = true, \
434 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
435 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
437 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
438 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
439 ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
441 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
442 INTEL_UEVENT_EXTRA_REG(c, \
443 MSR_PEBS_LD_LAT_THRESHOLD, \
447 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
449 union perf_capabilities {
457 * PMU supports separate counter range for writing
460 u64 full_width_write:1;
465 struct x86_pmu_quirk {
466 struct x86_pmu_quirk *next;
470 union x86_pmu_config {
491 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
494 x86_lbr_exclusive_lbr,
495 x86_lbr_exclusive_bts,
496 x86_lbr_exclusive_pt,
497 x86_lbr_exclusive_max,
501 * struct x86_pmu - generic x86 pmu
505 * Generic x86 PMC bits
509 int (*handle_irq)(struct pt_regs *);
510 void (*disable_all)(void);
511 void (*enable_all)(int added);
512 void (*enable)(struct perf_event *);
513 void (*disable)(struct perf_event *);
514 int (*hw_config)(struct perf_event *event);
515 int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
518 int (*addr_offset)(int index, bool eventsel);
519 int (*rdpmc_index)(int index);
520 u64 (*event_map)(int);
523 int num_counters_fixed;
527 unsigned long events_maskl;
528 unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
533 struct event_constraint *
534 (*get_event_constraints)(struct cpu_hw_events *cpuc,
536 struct perf_event *event);
538 void (*put_event_constraints)(struct cpu_hw_events *cpuc,
539 struct perf_event *event);
541 void (*start_scheduling)(struct cpu_hw_events *cpuc);
543 void (*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
545 void (*stop_scheduling)(struct cpu_hw_events *cpuc);
547 struct event_constraint *event_constraints;
548 struct x86_pmu_quirk *quirks;
549 int perfctr_second_write;
551 unsigned (*limit_period)(struct perf_event *event, unsigned l);
556 int attr_rdpmc_broken;
558 struct attribute **format_attrs;
559 struct attribute **event_attrs;
561 ssize_t (*events_sysfs_show)(char *page, u64 config);
562 struct attribute **cpu_events;
567 int (*cpu_prepare)(int cpu);
568 void (*cpu_starting)(int cpu);
569 void (*cpu_dying)(int cpu);
570 void (*cpu_dead)(int cpu);
572 void (*check_microcode)(void);
573 void (*sched_task)(struct perf_event_context *ctx,
577 * Intel Arch Perfmon v2+
580 union perf_capabilities intel_cap;
583 * Intel DebugStore bits
590 int pebs_record_size;
591 void (*drain_pebs)(struct pt_regs *regs);
592 struct event_constraint *pebs_constraints;
593 void (*pebs_aliases)(struct perf_event *event);
599 unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
600 int lbr_nr; /* hardware stack size */
601 u64 lbr_sel_mask; /* LBR_SELECT valid bits */
602 const int *lbr_sel_map; /* lbr_select mappings */
603 bool lbr_double_abort; /* duplicated lbr aborts */
606 * Intel PT/LBR/BTS are exclusive
608 atomic_t lbr_exclusive[x86_lbr_exclusive_max];
611 * Extra registers for events
613 struct extra_reg *extra_regs;
617 * Intel host/guest support (KVM)
619 struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
622 struct x86_perf_task_context {
623 u64 lbr_from[MAX_LBR_ENTRIES];
624 u64 lbr_to[MAX_LBR_ENTRIES];
625 int lbr_callstack_users;
629 #define x86_add_quirk(func_) \
631 static struct x86_pmu_quirk __quirk __initdata = { \
634 __quirk.next = x86_pmu.quirks; \
635 x86_pmu.quirks = &__quirk; \
641 #define PMU_FL_NO_HT_SHARING 0x1 /* no hyper-threading resource sharing */
642 #define PMU_FL_HAS_RSP_1 0x2 /* has 2 equivalent offcore_rsp regs */
643 #define PMU_FL_EXCL_CNTRS 0x4 /* has exclusive counter requirements */
644 #define PMU_FL_EXCL_ENABLED 0x8 /* exclusive counter active */
646 #define EVENT_VAR(_id) event_attr_##_id
647 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
649 #define EVENT_ATTR(_name, _id) \
650 static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
651 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
652 .id = PERF_COUNT_HW_##_id, \
656 #define EVENT_ATTR_STR(_name, v, str) \
657 static struct perf_pmu_events_attr event_attr_##v = { \
658 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
663 extern struct x86_pmu x86_pmu __read_mostly;
665 static inline bool x86_pmu_has_lbr_callstack(void)
667 return x86_pmu.lbr_sel_map &&
668 x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
671 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
673 int x86_perf_event_set_period(struct perf_event *event);
676 * Generalized hw caching related hw_event table, filled
677 * in on a per model basis. A value of 0 means
678 * 'not supported', -1 means 'hw_event makes no sense on
679 * this CPU', any other value means the raw hw_event
683 #define C(x) PERF_COUNT_HW_CACHE_##x
685 extern u64 __read_mostly hw_cache_event_ids
686 [PERF_COUNT_HW_CACHE_MAX]
687 [PERF_COUNT_HW_CACHE_OP_MAX]
688 [PERF_COUNT_HW_CACHE_RESULT_MAX];
689 extern u64 __read_mostly hw_cache_extra_regs
690 [PERF_COUNT_HW_CACHE_MAX]
691 [PERF_COUNT_HW_CACHE_OP_MAX]
692 [PERF_COUNT_HW_CACHE_RESULT_MAX];
694 u64 x86_perf_event_update(struct perf_event *event);
696 static inline unsigned int x86_pmu_config_addr(int index)
698 return x86_pmu.eventsel + (x86_pmu.addr_offset ?
699 x86_pmu.addr_offset(index, true) : index);
702 static inline unsigned int x86_pmu_event_addr(int index)
704 return x86_pmu.perfctr + (x86_pmu.addr_offset ?
705 x86_pmu.addr_offset(index, false) : index);
708 static inline int x86_pmu_rdpmc_index(int index)
710 return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
713 int x86_add_exclusive(unsigned int what);
715 void x86_del_exclusive(unsigned int what);
717 void hw_perf_lbr_event_destroy(struct perf_event *event);
719 int x86_setup_perfctr(struct perf_event *event);
721 int x86_pmu_hw_config(struct perf_event *event);
723 void x86_pmu_disable_all(void);
725 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
728 u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
730 if (hwc->extra_reg.reg)
731 wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
732 wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
735 void x86_pmu_enable_all(int added);
737 int perf_assign_events(struct event_constraint **constraints, int n,
738 int wmin, int wmax, int gpmax, int *assign);
739 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
741 void x86_pmu_stop(struct perf_event *event, int flags);
743 static inline void x86_pmu_disable_event(struct perf_event *event)
745 struct hw_perf_event *hwc = &event->hw;
747 wrmsrl(hwc->config_base, hwc->config);
750 void x86_pmu_enable_event(struct perf_event *event);
752 int x86_pmu_handle_irq(struct pt_regs *regs);
754 extern struct event_constraint emptyconstraint;
756 extern struct event_constraint unconstrained;
758 static inline bool kernel_ip(unsigned long ip)
761 return ip > PAGE_OFFSET;
768 * Not all PMUs provide the right context information to place the reported IP
769 * into full context. Specifically segment registers are typically not
772 * Assuming the address is a linear address (it is for IBS), we fake the CS and
773 * vm86 mode using the known zero-based code segment and 'fix up' the registers
776 * Intel PEBS/LBR appear to typically provide the effective address, nothing
777 * much we can do about that but pray and treat it like a linear address.
779 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
781 regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
782 if (regs->flags & X86_VM_MASK)
783 regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
787 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
788 ssize_t intel_event_sysfs_show(char *page, u64 config);
790 #ifdef CONFIG_CPU_SUP_AMD
792 int amd_pmu_init(void);
794 #else /* CONFIG_CPU_SUP_AMD */
796 static inline int amd_pmu_init(void)
801 #endif /* CONFIG_CPU_SUP_AMD */
803 #ifdef CONFIG_CPU_SUP_INTEL
805 static inline bool intel_pmu_needs_lbr_smpl(struct perf_event *event)
807 /* user explicitly requested branch sampling */
808 if (has_branch_stack(event))
811 /* implicit branch sampling to correct PEBS skid */
812 if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1 &&
813 x86_pmu.intel_cap.pebs_format < 2)
819 static inline bool intel_pmu_has_bts(struct perf_event *event)
821 if (event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS &&
822 !event->attr.freq && event->hw.sample_period == 1)
828 int intel_pmu_save_and_restart(struct perf_event *event);
830 struct event_constraint *
831 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
832 struct perf_event *event);
834 struct intel_shared_regs *allocate_shared_regs(int cpu);
836 int intel_pmu_init(void);
838 void init_debug_store_on_cpu(int cpu);
840 void fini_debug_store_on_cpu(int cpu);
842 void release_ds_buffers(void);
844 void reserve_ds_buffers(void);
846 extern struct event_constraint bts_constraint;
848 void intel_pmu_enable_bts(u64 config);
850 void intel_pmu_disable_bts(void);
852 int intel_pmu_drain_bts_buffer(void);
854 extern struct event_constraint intel_core2_pebs_event_constraints[];
856 extern struct event_constraint intel_atom_pebs_event_constraints[];
858 extern struct event_constraint intel_slm_pebs_event_constraints[];
860 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
862 extern struct event_constraint intel_westmere_pebs_event_constraints[];
864 extern struct event_constraint intel_snb_pebs_event_constraints[];
866 extern struct event_constraint intel_ivb_pebs_event_constraints[];
868 extern struct event_constraint intel_hsw_pebs_event_constraints[];
870 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
872 void intel_pmu_pebs_enable(struct perf_event *event);
874 void intel_pmu_pebs_disable(struct perf_event *event);
876 void intel_pmu_pebs_enable_all(void);
878 void intel_pmu_pebs_disable_all(void);
880 void intel_ds_init(void);
882 void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);
884 void intel_pmu_lbr_reset(void);
886 void intel_pmu_lbr_enable(struct perf_event *event);
888 void intel_pmu_lbr_disable(struct perf_event *event);
890 void intel_pmu_lbr_enable_all(bool pmi);
892 void intel_pmu_lbr_disable_all(void);
894 void intel_pmu_lbr_read(void);
896 void intel_pmu_lbr_init_core(void);
898 void intel_pmu_lbr_init_nhm(void);
900 void intel_pmu_lbr_init_atom(void);
902 void intel_pmu_lbr_init_snb(void);
904 void intel_pmu_lbr_init_hsw(void);
906 int intel_pmu_setup_lbr_filter(struct perf_event *event);
908 void intel_pt_interrupt(void);
910 int intel_bts_interrupt(void);
912 void intel_bts_enable_local(void);
914 void intel_bts_disable_local(void);
916 int p4_pmu_init(void);
918 int p6_pmu_init(void);
920 int knc_pmu_init(void);
922 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
925 static inline int is_ht_workaround_enabled(void)
927 return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
929 #else /* CONFIG_CPU_SUP_INTEL */
931 static inline void reserve_ds_buffers(void)
935 static inline void release_ds_buffers(void)
939 static inline int intel_pmu_init(void)
944 static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
949 static inline int is_ht_workaround_enabled(void)
953 #endif /* CONFIG_CPU_SUP_INTEL */