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 */
80 int nb_id; /* NorthBridge id */
81 int refcnt; /* reference count */
82 struct perf_event *owners[X86_PMC_IDX_MAX];
83 struct event_constraint event_constraints[X86_PMC_IDX_MAX];
86 /* The maximal number of PEBS events: */
87 #define MAX_PEBS_EVENTS 8
90 * A debug store configuration.
92 * We only support architectures that use 64bit fields.
97 u64 bts_absolute_maximum;
98 u64 bts_interrupt_threshold;
101 u64 pebs_absolute_maximum;
102 u64 pebs_interrupt_threshold;
103 u64 pebs_event_reset[MAX_PEBS_EVENTS];
107 * Per register state.
110 raw_spinlock_t lock; /* per-core: protect structure */
111 u64 config; /* extra MSR config */
112 u64 reg; /* extra MSR number */
113 atomic_t ref; /* reference count */
119 * Used to coordinate shared registers between HT threads or
120 * among events on a single PMU.
122 struct intel_shared_regs {
123 struct er_account regs[EXTRA_REG_MAX];
124 int refcnt; /* per-core: #HT threads */
125 unsigned core_id; /* per-core: core id */
128 enum intel_excl_state_type {
129 INTEL_EXCL_UNUSED = 0, /* counter is unused */
130 INTEL_EXCL_SHARED = 1, /* counter can be used by both threads */
131 INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
134 struct intel_excl_states {
135 enum intel_excl_state_type init_state[X86_PMC_IDX_MAX];
136 enum intel_excl_state_type state[X86_PMC_IDX_MAX];
137 int num_alloc_cntrs;/* #counters allocated */
138 int max_alloc_cntrs;/* max #counters allowed */
139 bool sched_started; /* true if scheduling has started */
142 struct intel_excl_cntrs {
145 struct intel_excl_states states[2];
147 int refcnt; /* per-core: #HT threads */
148 unsigned core_id; /* per-core: core id */
151 #define MAX_LBR_ENTRIES 16
154 X86_PERF_KFREE_SHARED = 0,
155 X86_PERF_KFREE_EXCL = 1,
159 struct cpu_hw_events {
161 * Generic x86 PMC bits
163 struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
164 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
165 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
168 int n_events; /* the # of events in the below arrays */
169 int n_added; /* the # last events in the below arrays;
170 they've never been enabled yet */
171 int n_txn; /* the # last events in the below arrays;
172 added in the current transaction */
173 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
174 u64 tags[X86_PMC_IDX_MAX];
175 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
177 unsigned int group_flag;
181 * Intel DebugStore bits
183 struct debug_store *ds;
191 struct perf_branch_stack lbr_stack;
192 struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
193 struct er_account *lbr_sel;
197 * Intel host/guest exclude bits
199 u64 intel_ctrl_guest_mask;
200 u64 intel_ctrl_host_mask;
201 struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
204 * Intel checkpoint mask
209 * manage shared (per-core, per-cpu) registers
210 * used on Intel NHM/WSM/SNB
212 struct intel_shared_regs *shared_regs;
214 * manage exclusive counter access between hyperthread
216 struct event_constraint *constraint_list; /* in enable order */
217 struct intel_excl_cntrs *excl_cntrs;
218 int excl_thread_id; /* 0 or 1 */
223 struct amd_nb *amd_nb;
224 /* Inverted mask of bits to clear in the perf_ctr ctrl registers */
225 u64 perf_ctr_virt_mask;
227 void *kfree_on_online[X86_PERF_KFREE_MAX];
230 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) {\
231 { .idxmsk64 = (n) }, \
239 #define EVENT_CONSTRAINT(c, n, m) \
240 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
242 #define INTEL_EXCLEVT_CONSTRAINT(c, n) \
243 __EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
244 0, PERF_X86_EVENT_EXCL)
247 * The overlap flag marks event constraints with overlapping counter
248 * masks. This is the case if the counter mask of such an event is not
249 * a subset of any other counter mask of a constraint with an equal or
250 * higher weight, e.g.:
252 * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
253 * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
254 * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
256 * The event scheduler may not select the correct counter in the first
257 * cycle because it needs to know which subsequent events will be
258 * scheduled. It may fail to schedule the events then. So we set the
259 * overlap flag for such constraints to give the scheduler a hint which
260 * events to select for counter rescheduling.
262 * Care must be taken as the rescheduling algorithm is O(n!) which
263 * will increase scheduling cycles for an over-commited system
264 * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros
265 * and its counter masks must be kept at a minimum.
267 #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \
268 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
271 * Constraint on the Event code.
273 #define INTEL_EVENT_CONSTRAINT(c, n) \
274 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
277 * Constraint on the Event code + UMask + fixed-mask
279 * filter mask to validate fixed counter events.
280 * the following filters disqualify for fixed counters:
285 * - in_tx_checkpointed
286 * The other filters are supported by fixed counters.
287 * The any-thread option is supported starting with v3.
289 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
290 #define FIXED_EVENT_CONSTRAINT(c, n) \
291 EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
294 * Constraint on the Event code + UMask
296 #define INTEL_UEVENT_CONSTRAINT(c, n) \
297 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
299 /* Like UEVENT_CONSTRAINT, but match flags too */
300 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \
301 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
303 #define INTEL_EXCLUEVT_CONSTRAINT(c, n) \
304 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
305 HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
307 #define INTEL_PLD_CONSTRAINT(c, n) \
308 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
309 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
311 #define INTEL_PST_CONSTRAINT(c, n) \
312 __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
313 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
315 /* Event constraint, but match on all event flags too. */
316 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
317 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
319 /* Check only flags, but allow all event/umask */
320 #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
321 EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
323 /* Check flags and event code, and set the HSW store flag */
324 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
325 __EVENT_CONSTRAINT(code, n, \
326 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
327 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
329 /* Check flags and event code, and set the HSW load flag */
330 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
331 __EVENT_CONSTRAINT(code, n, \
332 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
333 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
335 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
336 __EVENT_CONSTRAINT(code, n, \
337 ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
339 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
341 /* Check flags and event code/umask, and set the HSW store flag */
342 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
343 __EVENT_CONSTRAINT(code, n, \
344 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
345 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
347 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
348 __EVENT_CONSTRAINT(code, n, \
349 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
351 PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
353 /* Check flags and event code/umask, and set the HSW load flag */
354 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
355 __EVENT_CONSTRAINT(code, n, \
356 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
357 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
359 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
360 __EVENT_CONSTRAINT(code, n, \
361 INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
363 PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
365 /* Check flags and event code/umask, and set the HSW N/A flag */
366 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
367 __EVENT_CONSTRAINT(code, n, \
368 INTEL_ARCH_EVENT_MASK|INTEL_ARCH_EVENT_MASK, \
369 HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
373 * We define the end marker as having a weight of -1
374 * to enable blacklisting of events using a counter bitmask
375 * of zero and thus a weight of zero.
376 * The end marker has a weight that cannot possibly be
377 * obtained from counting the bits in the bitmask.
379 #define EVENT_CONSTRAINT_END { .weight = -1 }
382 * Check for end marker with weight == -1
384 #define for_each_event_constraint(e, c) \
385 for ((e) = (c); (e)->weight != -1; (e)++)
388 * Extra registers for specific events.
390 * Some events need large masks and require external MSRs.
391 * Those extra MSRs end up being shared for all events on
392 * a PMU and sometimes between PMU of sibling HT threads.
393 * In either case, the kernel needs to handle conflicting
394 * accesses to those extra, shared, regs. The data structure
395 * to manage those registers is stored in cpu_hw_event.
402 int idx; /* per_xxx->regs[] reg index */
403 bool extra_msr_access;
406 #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
409 .config_mask = (m), \
410 .valid_mask = (vm), \
411 .idx = EXTRA_REG_##i, \
412 .extra_msr_access = true, \
415 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
416 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
418 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
419 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
420 ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
422 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
423 INTEL_UEVENT_EXTRA_REG(c, \
424 MSR_PEBS_LD_LAT_THRESHOLD, \
428 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
430 union perf_capabilities {
438 * PMU supports separate counter range for writing
441 u64 full_width_write:1;
446 struct x86_pmu_quirk {
447 struct x86_pmu_quirk *next;
451 union x86_pmu_config {
472 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
475 x86_lbr_exclusive_lbr,
476 x86_lbr_exclusive_bts,
477 x86_lbr_exclusive_pt,
478 x86_lbr_exclusive_max,
482 * struct x86_pmu - generic x86 pmu
486 * Generic x86 PMC bits
490 int (*handle_irq)(struct pt_regs *);
491 void (*disable_all)(void);
492 void (*enable_all)(int added);
493 void (*enable)(struct perf_event *);
494 void (*disable)(struct perf_event *);
495 int (*hw_config)(struct perf_event *event);
496 int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
499 int (*addr_offset)(int index, bool eventsel);
500 int (*rdpmc_index)(int index);
501 u64 (*event_map)(int);
504 int num_counters_fixed;
508 unsigned long events_maskl;
509 unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
514 struct event_constraint *
515 (*get_event_constraints)(struct cpu_hw_events *cpuc,
517 struct perf_event *event);
519 void (*put_event_constraints)(struct cpu_hw_events *cpuc,
520 struct perf_event *event);
522 void (*commit_scheduling)(struct cpu_hw_events *cpuc,
523 struct perf_event *event,
526 void (*start_scheduling)(struct cpu_hw_events *cpuc);
528 void (*stop_scheduling)(struct cpu_hw_events *cpuc);
530 struct event_constraint *event_constraints;
531 struct x86_pmu_quirk *quirks;
532 int perfctr_second_write;
534 unsigned (*limit_period)(struct perf_event *event, unsigned l);
539 int attr_rdpmc_broken;
541 struct attribute **format_attrs;
542 struct attribute **event_attrs;
544 ssize_t (*events_sysfs_show)(char *page, u64 config);
545 struct attribute **cpu_events;
550 int (*cpu_prepare)(int cpu);
551 void (*cpu_starting)(int cpu);
552 void (*cpu_dying)(int cpu);
553 void (*cpu_dead)(int cpu);
555 void (*check_microcode)(void);
556 void (*sched_task)(struct perf_event_context *ctx,
560 * Intel Arch Perfmon v2+
563 union perf_capabilities intel_cap;
566 * Intel DebugStore bits
573 int pebs_record_size;
574 void (*drain_pebs)(struct pt_regs *regs);
575 struct event_constraint *pebs_constraints;
576 void (*pebs_aliases)(struct perf_event *event);
582 unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
583 int lbr_nr; /* hardware stack size */
584 u64 lbr_sel_mask; /* LBR_SELECT valid bits */
585 const int *lbr_sel_map; /* lbr_select mappings */
586 bool lbr_double_abort; /* duplicated lbr aborts */
589 * Intel PT/LBR/BTS are exclusive
591 atomic_t lbr_exclusive[x86_lbr_exclusive_max];
594 * Extra registers for events
596 struct extra_reg *extra_regs;
600 * Intel host/guest support (KVM)
602 struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
605 struct x86_perf_task_context {
606 u64 lbr_from[MAX_LBR_ENTRIES];
607 u64 lbr_to[MAX_LBR_ENTRIES];
608 int lbr_callstack_users;
612 #define x86_add_quirk(func_) \
614 static struct x86_pmu_quirk __quirk __initdata = { \
617 __quirk.next = x86_pmu.quirks; \
618 x86_pmu.quirks = &__quirk; \
624 #define PMU_FL_NO_HT_SHARING 0x1 /* no hyper-threading resource sharing */
625 #define PMU_FL_HAS_RSP_1 0x2 /* has 2 equivalent offcore_rsp regs */
626 #define PMU_FL_EXCL_CNTRS 0x4 /* has exclusive counter requirements */
627 #define PMU_FL_EXCL_ENABLED 0x8 /* exclusive counter active */
629 #define EVENT_VAR(_id) event_attr_##_id
630 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
632 #define EVENT_ATTR(_name, _id) \
633 static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
634 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
635 .id = PERF_COUNT_HW_##_id, \
639 #define EVENT_ATTR_STR(_name, v, str) \
640 static struct perf_pmu_events_attr event_attr_##v = { \
641 .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
646 extern struct x86_pmu x86_pmu __read_mostly;
648 static inline bool x86_pmu_has_lbr_callstack(void)
650 return x86_pmu.lbr_sel_map &&
651 x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
654 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
656 int x86_perf_event_set_period(struct perf_event *event);
659 * Generalized hw caching related hw_event table, filled
660 * in on a per model basis. A value of 0 means
661 * 'not supported', -1 means 'hw_event makes no sense on
662 * this CPU', any other value means the raw hw_event
666 #define C(x) PERF_COUNT_HW_CACHE_##x
668 extern u64 __read_mostly hw_cache_event_ids
669 [PERF_COUNT_HW_CACHE_MAX]
670 [PERF_COUNT_HW_CACHE_OP_MAX]
671 [PERF_COUNT_HW_CACHE_RESULT_MAX];
672 extern u64 __read_mostly hw_cache_extra_regs
673 [PERF_COUNT_HW_CACHE_MAX]
674 [PERF_COUNT_HW_CACHE_OP_MAX]
675 [PERF_COUNT_HW_CACHE_RESULT_MAX];
677 u64 x86_perf_event_update(struct perf_event *event);
679 static inline unsigned int x86_pmu_config_addr(int index)
681 return x86_pmu.eventsel + (x86_pmu.addr_offset ?
682 x86_pmu.addr_offset(index, true) : index);
685 static inline unsigned int x86_pmu_event_addr(int index)
687 return x86_pmu.perfctr + (x86_pmu.addr_offset ?
688 x86_pmu.addr_offset(index, false) : index);
691 static inline int x86_pmu_rdpmc_index(int index)
693 return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
696 int x86_add_exclusive(unsigned int what);
698 void x86_del_exclusive(unsigned int what);
700 void hw_perf_lbr_event_destroy(struct perf_event *event);
702 int x86_setup_perfctr(struct perf_event *event);
704 int x86_pmu_hw_config(struct perf_event *event);
706 void x86_pmu_disable_all(void);
708 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
711 u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
713 if (hwc->extra_reg.reg)
714 wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
715 wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
718 void x86_pmu_enable_all(int added);
720 int perf_assign_events(struct perf_event **events, int n,
721 int wmin, int wmax, int *assign);
722 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
724 void x86_pmu_stop(struct perf_event *event, int flags);
726 static inline void x86_pmu_disable_event(struct perf_event *event)
728 struct hw_perf_event *hwc = &event->hw;
730 wrmsrl(hwc->config_base, hwc->config);
733 void x86_pmu_enable_event(struct perf_event *event);
735 int x86_pmu_handle_irq(struct pt_regs *regs);
737 extern struct event_constraint emptyconstraint;
739 extern struct event_constraint unconstrained;
741 static inline bool kernel_ip(unsigned long ip)
744 return ip > PAGE_OFFSET;
751 * Not all PMUs provide the right context information to place the reported IP
752 * into full context. Specifically segment registers are typically not
755 * Assuming the address is a linear address (it is for IBS), we fake the CS and
756 * vm86 mode using the known zero-based code segment and 'fix up' the registers
759 * Intel PEBS/LBR appear to typically provide the effective address, nothing
760 * much we can do about that but pray and treat it like a linear address.
762 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
764 regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
765 if (regs->flags & X86_VM_MASK)
766 regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
770 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
771 ssize_t intel_event_sysfs_show(char *page, u64 config);
773 #ifdef CONFIG_CPU_SUP_AMD
775 int amd_pmu_init(void);
777 #else /* CONFIG_CPU_SUP_AMD */
779 static inline int amd_pmu_init(void)
784 #endif /* CONFIG_CPU_SUP_AMD */
786 #ifdef CONFIG_CPU_SUP_INTEL
788 static inline bool intel_pmu_needs_lbr_smpl(struct perf_event *event)
790 /* user explicitly requested branch sampling */
791 if (has_branch_stack(event))
794 /* implicit branch sampling to correct PEBS skid */
795 if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1 &&
796 x86_pmu.intel_cap.pebs_format < 2)
802 static inline bool intel_pmu_has_bts(struct perf_event *event)
804 if (event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS &&
805 !event->attr.freq && event->hw.sample_period == 1)
811 int intel_pmu_save_and_restart(struct perf_event *event);
813 struct event_constraint *
814 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
815 struct perf_event *event);
817 struct intel_shared_regs *allocate_shared_regs(int cpu);
819 int intel_pmu_init(void);
821 void init_debug_store_on_cpu(int cpu);
823 void fini_debug_store_on_cpu(int cpu);
825 void release_ds_buffers(void);
827 void reserve_ds_buffers(void);
829 extern struct event_constraint bts_constraint;
831 void intel_pmu_enable_bts(u64 config);
833 void intel_pmu_disable_bts(void);
835 int intel_pmu_drain_bts_buffer(void);
837 extern struct event_constraint intel_core2_pebs_event_constraints[];
839 extern struct event_constraint intel_atom_pebs_event_constraints[];
841 extern struct event_constraint intel_slm_pebs_event_constraints[];
843 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
845 extern struct event_constraint intel_westmere_pebs_event_constraints[];
847 extern struct event_constraint intel_snb_pebs_event_constraints[];
849 extern struct event_constraint intel_ivb_pebs_event_constraints[];
851 extern struct event_constraint intel_hsw_pebs_event_constraints[];
853 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
855 void intel_pmu_pebs_enable(struct perf_event *event);
857 void intel_pmu_pebs_disable(struct perf_event *event);
859 void intel_pmu_pebs_enable_all(void);
861 void intel_pmu_pebs_disable_all(void);
863 void intel_ds_init(void);
865 void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);
867 void intel_pmu_lbr_reset(void);
869 void intel_pmu_lbr_enable(struct perf_event *event);
871 void intel_pmu_lbr_disable(struct perf_event *event);
873 void intel_pmu_lbr_enable_all(bool pmi);
875 void intel_pmu_lbr_disable_all(void);
877 void intel_pmu_lbr_read(void);
879 void intel_pmu_lbr_init_core(void);
881 void intel_pmu_lbr_init_nhm(void);
883 void intel_pmu_lbr_init_atom(void);
885 void intel_pmu_lbr_init_snb(void);
887 void intel_pmu_lbr_init_hsw(void);
889 int intel_pmu_setup_lbr_filter(struct perf_event *event);
891 void intel_pt_interrupt(void);
893 int intel_bts_interrupt(void);
895 void intel_bts_enable_local(void);
897 void intel_bts_disable_local(void);
899 int p4_pmu_init(void);
901 int p6_pmu_init(void);
903 int knc_pmu_init(void);
905 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
908 static inline int is_ht_workaround_enabled(void)
910 return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
912 #else /* CONFIG_CPU_SUP_INTEL */
914 static inline void reserve_ds_buffers(void)
918 static inline void release_ds_buffers(void)
922 static inline int intel_pmu_init(void)
927 static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
932 #endif /* CONFIG_CPU_SUP_INTEL */