perf_event.h 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779
  1. /*
  2. * Performance events x86 architecture header
  3. *
  4. * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
  5. * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
  6. * Copyright (C) 2009 Jaswinder Singh Rajput
  7. * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
  8. * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
  9. * Copyright (C) 2009 Intel Corporation, <markus.t.metzger@intel.com>
  10. * Copyright (C) 2009 Google, Inc., Stephane Eranian
  11. *
  12. * For licencing details see kernel-base/COPYING
  13. */
  14. #include <linux/perf_event.h>
  15. #if 0
  16. #undef wrmsrl
  17. #define wrmsrl(msr, val) \
  18. do { \
  19. unsigned int _msr = (msr); \
  20. u64 _val = (val); \
  21. trace_printk("wrmsrl(%x, %Lx)\n", (unsigned int)(_msr), \
  22. (unsigned long long)(_val)); \
  23. native_write_msr((_msr), (u32)(_val), (u32)(_val >> 32)); \
  24. } while (0)
  25. #endif
  26. /*
  27. * | NHM/WSM | SNB |
  28. * register -------------------------------
  29. * | HT | no HT | HT | no HT |
  30. *-----------------------------------------
  31. * offcore | core | core | cpu | core |
  32. * lbr_sel | core | core | cpu | core |
  33. * ld_lat | cpu | core | cpu | core |
  34. *-----------------------------------------
  35. *
  36. * Given that there is a small number of shared regs,
  37. * we can pre-allocate their slot in the per-cpu
  38. * per-core reg tables.
  39. */
  40. enum extra_reg_type {
  41. EXTRA_REG_NONE = -1, /* not used */
  42. EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */
  43. EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */
  44. EXTRA_REG_LBR = 2, /* lbr_select */
  45. EXTRA_REG_LDLAT = 3, /* ld_lat_threshold */
  46. EXTRA_REG_MAX /* number of entries needed */
  47. };
  48. struct event_constraint {
  49. union {
  50. unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
  51. u64 idxmsk64;
  52. };
  53. u64 code;
  54. u64 cmask;
  55. int weight;
  56. int overlap;
  57. int flags;
  58. };
  59. /*
  60. * struct hw_perf_event.flags flags
  61. */
  62. #define PERF_X86_EVENT_PEBS_LDLAT 0x1 /* ld+ldlat data address sampling */
  63. #define PERF_X86_EVENT_PEBS_ST 0x2 /* st data address sampling */
  64. #define PERF_X86_EVENT_PEBS_ST_HSW 0x4 /* haswell style datala, store */
  65. #define PERF_X86_EVENT_COMMITTED 0x8 /* event passed commit_txn */
  66. #define PERF_X86_EVENT_PEBS_LD_HSW 0x10 /* haswell style datala, load */
  67. #define PERF_X86_EVENT_PEBS_NA_HSW 0x20 /* haswell style datala, unknown */
  68. struct amd_nb {
  69. int nb_id; /* NorthBridge id */
  70. int refcnt; /* reference count */
  71. struct perf_event *owners[X86_PMC_IDX_MAX];
  72. struct event_constraint event_constraints[X86_PMC_IDX_MAX];
  73. };
  74. /* The maximal number of PEBS events: */
  75. #define MAX_PEBS_EVENTS 8
  76. /*
  77. * A debug store configuration.
  78. *
  79. * We only support architectures that use 64bit fields.
  80. */
  81. struct debug_store {
  82. u64 bts_buffer_base;
  83. u64 bts_index;
  84. u64 bts_absolute_maximum;
  85. u64 bts_interrupt_threshold;
  86. u64 pebs_buffer_base;
  87. u64 pebs_index;
  88. u64 pebs_absolute_maximum;
  89. u64 pebs_interrupt_threshold;
  90. u64 pebs_event_reset[MAX_PEBS_EVENTS];
  91. };
  92. /*
  93. * Per register state.
  94. */
  95. struct er_account {
  96. raw_spinlock_t lock; /* per-core: protect structure */
  97. u64 config; /* extra MSR config */
  98. u64 reg; /* extra MSR number */
  99. atomic_t ref; /* reference count */
  100. };
  101. /*
  102. * Per core/cpu state
  103. *
  104. * Used to coordinate shared registers between HT threads or
  105. * among events on a single PMU.
  106. */
  107. struct intel_shared_regs {
  108. struct er_account regs[EXTRA_REG_MAX];
  109. int refcnt; /* per-core: #HT threads */
  110. unsigned core_id; /* per-core: core id */
  111. };
  112. #define MAX_LBR_ENTRIES 16
  113. struct cpu_hw_events {
  114. /*
  115. * Generic x86 PMC bits
  116. */
  117. struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
  118. unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
  119. unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
  120. int enabled;
  121. int n_events; /* the # of events in the below arrays */
  122. int n_added; /* the # last events in the below arrays;
  123. they've never been enabled yet */
  124. int n_txn; /* the # last events in the below arrays;
  125. added in the current transaction */
  126. int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
  127. u64 tags[X86_PMC_IDX_MAX];
  128. struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
  129. unsigned int group_flag;
  130. int is_fake;
  131. /*
  132. * Intel DebugStore bits
  133. */
  134. struct debug_store *ds;
  135. u64 pebs_enabled;
  136. /*
  137. * Intel LBR bits
  138. */
  139. int lbr_users;
  140. void *lbr_context;
  141. struct perf_branch_stack lbr_stack;
  142. struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
  143. struct er_account *lbr_sel;
  144. u64 br_sel;
  145. /*
  146. * Intel host/guest exclude bits
  147. */
  148. u64 intel_ctrl_guest_mask;
  149. u64 intel_ctrl_host_mask;
  150. struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
  151. /*
  152. * Intel checkpoint mask
  153. */
  154. u64 intel_cp_status;
  155. /*
  156. * manage shared (per-core, per-cpu) registers
  157. * used on Intel NHM/WSM/SNB
  158. */
  159. struct intel_shared_regs *shared_regs;
  160. /*
  161. * AMD specific bits
  162. */
  163. struct amd_nb *amd_nb;
  164. /* Inverted mask of bits to clear in the perf_ctr ctrl registers */
  165. u64 perf_ctr_virt_mask;
  166. void *kfree_on_online;
  167. };
  168. #define __EVENT_CONSTRAINT(c, n, m, w, o, f) {\
  169. { .idxmsk64 = (n) }, \
  170. .code = (c), \
  171. .cmask = (m), \
  172. .weight = (w), \
  173. .overlap = (o), \
  174. .flags = f, \
  175. }
  176. #define EVENT_CONSTRAINT(c, n, m) \
  177. __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
  178. /*
  179. * The overlap flag marks event constraints with overlapping counter
  180. * masks. This is the case if the counter mask of such an event is not
  181. * a subset of any other counter mask of a constraint with an equal or
  182. * higher weight, e.g.:
  183. *
  184. * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
  185. * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
  186. * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
  187. *
  188. * The event scheduler may not select the correct counter in the first
  189. * cycle because it needs to know which subsequent events will be
  190. * scheduled. It may fail to schedule the events then. So we set the
  191. * overlap flag for such constraints to give the scheduler a hint which
  192. * events to select for counter rescheduling.
  193. *
  194. * Care must be taken as the rescheduling algorithm is O(n!) which
  195. * will increase scheduling cycles for an over-commited system
  196. * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros
  197. * and its counter masks must be kept at a minimum.
  198. */
  199. #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \
  200. __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
  201. /*
  202. * Constraint on the Event code.
  203. */
  204. #define INTEL_EVENT_CONSTRAINT(c, n) \
  205. EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
  206. /*
  207. * Constraint on the Event code + UMask + fixed-mask
  208. *
  209. * filter mask to validate fixed counter events.
  210. * the following filters disqualify for fixed counters:
  211. * - inv
  212. * - edge
  213. * - cnt-mask
  214. * - in_tx
  215. * - in_tx_checkpointed
  216. * The other filters are supported by fixed counters.
  217. * The any-thread option is supported starting with v3.
  218. */
  219. #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
  220. #define FIXED_EVENT_CONSTRAINT(c, n) \
  221. EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
  222. /*
  223. * Constraint on the Event code + UMask
  224. */
  225. #define INTEL_UEVENT_CONSTRAINT(c, n) \
  226. EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
  227. /* Like UEVENT_CONSTRAINT, but match flags too */
  228. #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \
  229. EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
  230. #define INTEL_PLD_CONSTRAINT(c, n) \
  231. __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  232. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
  233. #define INTEL_PST_CONSTRAINT(c, n) \
  234. __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  235. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
  236. /* Event constraint, but match on all event flags too. */
  237. #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
  238. EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
  239. /* Check only flags, but allow all event/umask */
  240. #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
  241. EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
  242. /* Check flags and event code, and set the HSW store flag */
  243. #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
  244. __EVENT_CONSTRAINT(code, n, \
  245. ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
  246. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
  247. /* Check flags and event code, and set the HSW load flag */
  248. #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
  249. __EVENT_CONSTRAINT(code, n, \
  250. ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
  251. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
  252. /* Check flags and event code/umask, and set the HSW store flag */
  253. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
  254. __EVENT_CONSTRAINT(code, n, \
  255. INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  256. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
  257. /* Check flags and event code/umask, and set the HSW load flag */
  258. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
  259. __EVENT_CONSTRAINT(code, n, \
  260. INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  261. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
  262. /* Check flags and event code/umask, and set the HSW N/A flag */
  263. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
  264. __EVENT_CONSTRAINT(code, n, \
  265. INTEL_ARCH_EVENT_MASK|INTEL_ARCH_EVENT_MASK, \
  266. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
  267. /*
  268. * We define the end marker as having a weight of -1
  269. * to enable blacklisting of events using a counter bitmask
  270. * of zero and thus a weight of zero.
  271. * The end marker has a weight that cannot possibly be
  272. * obtained from counting the bits in the bitmask.
  273. */
  274. #define EVENT_CONSTRAINT_END { .weight = -1 }
  275. /*
  276. * Check for end marker with weight == -1
  277. */
  278. #define for_each_event_constraint(e, c) \
  279. for ((e) = (c); (e)->weight != -1; (e)++)
  280. /*
  281. * Extra registers for specific events.
  282. *
  283. * Some events need large masks and require external MSRs.
  284. * Those extra MSRs end up being shared for all events on
  285. * a PMU and sometimes between PMU of sibling HT threads.
  286. * In either case, the kernel needs to handle conflicting
  287. * accesses to those extra, shared, regs. The data structure
  288. * to manage those registers is stored in cpu_hw_event.
  289. */
  290. struct extra_reg {
  291. unsigned int event;
  292. unsigned int msr;
  293. u64 config_mask;
  294. u64 valid_mask;
  295. int idx; /* per_xxx->regs[] reg index */
  296. bool extra_msr_access;
  297. };
  298. #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
  299. .event = (e), \
  300. .msr = (ms), \
  301. .config_mask = (m), \
  302. .valid_mask = (vm), \
  303. .idx = EXTRA_REG_##i, \
  304. .extra_msr_access = true, \
  305. }
  306. #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
  307. EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
  308. #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
  309. EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
  310. ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
  311. #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
  312. INTEL_UEVENT_EXTRA_REG(c, \
  313. MSR_PEBS_LD_LAT_THRESHOLD, \
  314. 0xffff, \
  315. LDLAT)
  316. #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
  317. union perf_capabilities {
  318. struct {
  319. u64 lbr_format:6;
  320. u64 pebs_trap:1;
  321. u64 pebs_arch_reg:1;
  322. u64 pebs_format:4;
  323. u64 smm_freeze:1;
  324. /*
  325. * PMU supports separate counter range for writing
  326. * values > 32bit.
  327. */
  328. u64 full_width_write:1;
  329. };
  330. u64 capabilities;
  331. };
  332. struct x86_pmu_quirk {
  333. struct x86_pmu_quirk *next;
  334. void (*func)(void);
  335. };
  336. union x86_pmu_config {
  337. struct {
  338. u64 event:8,
  339. umask:8,
  340. usr:1,
  341. os:1,
  342. edge:1,
  343. pc:1,
  344. interrupt:1,
  345. __reserved1:1,
  346. en:1,
  347. inv:1,
  348. cmask:8,
  349. event2:4,
  350. __reserved2:4,
  351. go:1,
  352. ho:1;
  353. } bits;
  354. u64 value;
  355. };
  356. #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
  357. /*
  358. * struct x86_pmu - generic x86 pmu
  359. */
  360. struct x86_pmu {
  361. /*
  362. * Generic x86 PMC bits
  363. */
  364. const char *name;
  365. int version;
  366. int (*handle_irq)(struct pt_regs *);
  367. void (*disable_all)(void);
  368. void (*enable_all)(int added);
  369. void (*enable)(struct perf_event *);
  370. void (*disable)(struct perf_event *);
  371. int (*hw_config)(struct perf_event *event);
  372. int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
  373. unsigned eventsel;
  374. unsigned perfctr;
  375. int (*addr_offset)(int index, bool eventsel);
  376. int (*rdpmc_index)(int index);
  377. u64 (*event_map)(int);
  378. int max_events;
  379. int num_counters;
  380. int num_counters_fixed;
  381. int cntval_bits;
  382. u64 cntval_mask;
  383. union {
  384. unsigned long events_maskl;
  385. unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
  386. };
  387. int events_mask_len;
  388. int apic;
  389. u64 max_period;
  390. struct event_constraint *
  391. (*get_event_constraints)(struct cpu_hw_events *cpuc,
  392. struct perf_event *event);
  393. void (*put_event_constraints)(struct cpu_hw_events *cpuc,
  394. struct perf_event *event);
  395. struct event_constraint *event_constraints;
  396. struct x86_pmu_quirk *quirks;
  397. int perfctr_second_write;
  398. bool late_ack;
  399. /*
  400. * sysfs attrs
  401. */
  402. int attr_rdpmc_broken;
  403. int attr_rdpmc;
  404. struct attribute **format_attrs;
  405. struct attribute **event_attrs;
  406. ssize_t (*events_sysfs_show)(char *page, u64 config);
  407. struct attribute **cpu_events;
  408. /*
  409. * CPU Hotplug hooks
  410. */
  411. int (*cpu_prepare)(int cpu);
  412. void (*cpu_starting)(int cpu);
  413. void (*cpu_dying)(int cpu);
  414. void (*cpu_dead)(int cpu);
  415. void (*check_microcode)(void);
  416. void (*flush_branch_stack)(void);
  417. /*
  418. * Intel Arch Perfmon v2+
  419. */
  420. u64 intel_ctrl;
  421. union perf_capabilities intel_cap;
  422. /*
  423. * Intel DebugStore bits
  424. */
  425. unsigned int bts :1,
  426. bts_active :1,
  427. pebs :1,
  428. pebs_active :1,
  429. pebs_broken :1;
  430. int pebs_record_size;
  431. void (*drain_pebs)(struct pt_regs *regs);
  432. struct event_constraint *pebs_constraints;
  433. void (*pebs_aliases)(struct perf_event *event);
  434. int max_pebs_events;
  435. /*
  436. * Intel LBR
  437. */
  438. unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
  439. int lbr_nr; /* hardware stack size */
  440. u64 lbr_sel_mask; /* LBR_SELECT valid bits */
  441. const int *lbr_sel_map; /* lbr_select mappings */
  442. bool lbr_double_abort; /* duplicated lbr aborts */
  443. /*
  444. * Extra registers for events
  445. */
  446. struct extra_reg *extra_regs;
  447. unsigned int er_flags;
  448. /*
  449. * Intel host/guest support (KVM)
  450. */
  451. struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
  452. };
  453. #define x86_add_quirk(func_) \
  454. do { \
  455. static struct x86_pmu_quirk __quirk __initdata = { \
  456. .func = func_, \
  457. }; \
  458. __quirk.next = x86_pmu.quirks; \
  459. x86_pmu.quirks = &__quirk; \
  460. } while (0)
  461. #define ERF_NO_HT_SHARING 1
  462. #define ERF_HAS_RSP_1 2
  463. #define EVENT_VAR(_id) event_attr_##_id
  464. #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
  465. #define EVENT_ATTR(_name, _id) \
  466. static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
  467. .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
  468. .id = PERF_COUNT_HW_##_id, \
  469. .event_str = NULL, \
  470. };
  471. #define EVENT_ATTR_STR(_name, v, str) \
  472. static struct perf_pmu_events_attr event_attr_##v = { \
  473. .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
  474. .id = 0, \
  475. .event_str = str, \
  476. };
  477. extern struct x86_pmu x86_pmu __read_mostly;
  478. DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
  479. int x86_perf_event_set_period(struct perf_event *event);
  480. /*
  481. * Generalized hw caching related hw_event table, filled
  482. * in on a per model basis. A value of 0 means
  483. * 'not supported', -1 means 'hw_event makes no sense on
  484. * this CPU', any other value means the raw hw_event
  485. * ID.
  486. */
  487. #define C(x) PERF_COUNT_HW_CACHE_##x
  488. extern u64 __read_mostly hw_cache_event_ids
  489. [PERF_COUNT_HW_CACHE_MAX]
  490. [PERF_COUNT_HW_CACHE_OP_MAX]
  491. [PERF_COUNT_HW_CACHE_RESULT_MAX];
  492. extern u64 __read_mostly hw_cache_extra_regs
  493. [PERF_COUNT_HW_CACHE_MAX]
  494. [PERF_COUNT_HW_CACHE_OP_MAX]
  495. [PERF_COUNT_HW_CACHE_RESULT_MAX];
  496. u64 x86_perf_event_update(struct perf_event *event);
  497. static inline unsigned int x86_pmu_config_addr(int index)
  498. {
  499. return x86_pmu.eventsel + (x86_pmu.addr_offset ?
  500. x86_pmu.addr_offset(index, true) : index);
  501. }
  502. static inline unsigned int x86_pmu_event_addr(int index)
  503. {
  504. return x86_pmu.perfctr + (x86_pmu.addr_offset ?
  505. x86_pmu.addr_offset(index, false) : index);
  506. }
  507. static inline int x86_pmu_rdpmc_index(int index)
  508. {
  509. return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
  510. }
  511. int x86_setup_perfctr(struct perf_event *event);
  512. int x86_pmu_hw_config(struct perf_event *event);
  513. void x86_pmu_disable_all(void);
  514. static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
  515. u64 enable_mask)
  516. {
  517. u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
  518. if (hwc->extra_reg.reg)
  519. wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
  520. wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
  521. }
  522. void x86_pmu_enable_all(int added);
  523. int perf_assign_events(struct perf_event **events, int n,
  524. int wmin, int wmax, int *assign);
  525. int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
  526. void x86_pmu_stop(struct perf_event *event, int flags);
  527. static inline void x86_pmu_disable_event(struct perf_event *event)
  528. {
  529. struct hw_perf_event *hwc = &event->hw;
  530. wrmsrl(hwc->config_base, hwc->config);
  531. }
  532. void x86_pmu_enable_event(struct perf_event *event);
  533. int x86_pmu_handle_irq(struct pt_regs *regs);
  534. extern struct event_constraint emptyconstraint;
  535. extern struct event_constraint unconstrained;
  536. static inline bool kernel_ip(unsigned long ip)
  537. {
  538. #ifdef CONFIG_X86_32
  539. return ip > PAGE_OFFSET;
  540. #else
  541. return (long)ip < 0;
  542. #endif
  543. }
  544. /*
  545. * Not all PMUs provide the right context information to place the reported IP
  546. * into full context. Specifically segment registers are typically not
  547. * supplied.
  548. *
  549. * Assuming the address is a linear address (it is for IBS), we fake the CS and
  550. * vm86 mode using the known zero-based code segment and 'fix up' the registers
  551. * to reflect this.
  552. *
  553. * Intel PEBS/LBR appear to typically provide the effective address, nothing
  554. * much we can do about that but pray and treat it like a linear address.
  555. */
  556. static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
  557. {
  558. regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
  559. if (regs->flags & X86_VM_MASK)
  560. regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
  561. regs->ip = ip;
  562. }
  563. ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
  564. ssize_t intel_event_sysfs_show(char *page, u64 config);
  565. #ifdef CONFIG_CPU_SUP_AMD
  566. int amd_pmu_init(void);
  567. #else /* CONFIG_CPU_SUP_AMD */
  568. static inline int amd_pmu_init(void)
  569. {
  570. return 0;
  571. }
  572. #endif /* CONFIG_CPU_SUP_AMD */
  573. #ifdef CONFIG_CPU_SUP_INTEL
  574. int intel_pmu_save_and_restart(struct perf_event *event);
  575. struct event_constraint *
  576. x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event);
  577. struct intel_shared_regs *allocate_shared_regs(int cpu);
  578. int intel_pmu_init(void);
  579. void init_debug_store_on_cpu(int cpu);
  580. void fini_debug_store_on_cpu(int cpu);
  581. void release_ds_buffers(void);
  582. void reserve_ds_buffers(void);
  583. extern struct event_constraint bts_constraint;
  584. void intel_pmu_enable_bts(u64 config);
  585. void intel_pmu_disable_bts(void);
  586. int intel_pmu_drain_bts_buffer(void);
  587. extern struct event_constraint intel_core2_pebs_event_constraints[];
  588. extern struct event_constraint intel_atom_pebs_event_constraints[];
  589. extern struct event_constraint intel_slm_pebs_event_constraints[];
  590. extern struct event_constraint intel_nehalem_pebs_event_constraints[];
  591. extern struct event_constraint intel_westmere_pebs_event_constraints[];
  592. extern struct event_constraint intel_snb_pebs_event_constraints[];
  593. extern struct event_constraint intel_ivb_pebs_event_constraints[];
  594. extern struct event_constraint intel_hsw_pebs_event_constraints[];
  595. struct event_constraint *intel_pebs_constraints(struct perf_event *event);
  596. void intel_pmu_pebs_enable(struct perf_event *event);
  597. void intel_pmu_pebs_disable(struct perf_event *event);
  598. void intel_pmu_pebs_enable_all(void);
  599. void intel_pmu_pebs_disable_all(void);
  600. void intel_ds_init(void);
  601. void intel_pmu_lbr_reset(void);
  602. void intel_pmu_lbr_enable(struct perf_event *event);
  603. void intel_pmu_lbr_disable(struct perf_event *event);
  604. void intel_pmu_lbr_enable_all(void);
  605. void intel_pmu_lbr_disable_all(void);
  606. void intel_pmu_lbr_read(void);
  607. void intel_pmu_lbr_init_core(void);
  608. void intel_pmu_lbr_init_nhm(void);
  609. void intel_pmu_lbr_init_atom(void);
  610. void intel_pmu_lbr_init_snb(void);
  611. int intel_pmu_setup_lbr_filter(struct perf_event *event);
  612. int p4_pmu_init(void);
  613. int p6_pmu_init(void);
  614. int knc_pmu_init(void);
  615. ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
  616. char *page);
  617. #else /* CONFIG_CPU_SUP_INTEL */
  618. static inline void reserve_ds_buffers(void)
  619. {
  620. }
  621. static inline void release_ds_buffers(void)
  622. {
  623. }
  624. static inline int intel_pmu_init(void)
  625. {
  626. return 0;
  627. }
  628. static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
  629. {
  630. return NULL;
  631. }
  632. #endif /* CONFIG_CPU_SUP_INTEL */