perf_event.h 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776
  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. #define INTEL_PLD_CONSTRAINT(c, n) \
  228. __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  229. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
  230. #define INTEL_PST_CONSTRAINT(c, n) \
  231. __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  232. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
  233. /* Event constraint, but match on all event flags too. */
  234. #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
  235. EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
  236. /* Check only flags, but allow all event/umask */
  237. #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \
  238. EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
  239. /* Check flags and event code, and set the HSW store flag */
  240. #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
  241. __EVENT_CONSTRAINT(code, n, \
  242. ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
  243. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
  244. /* Check flags and event code, and set the HSW load flag */
  245. #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
  246. __EVENT_CONSTRAINT(code, n, \
  247. ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
  248. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
  249. /* Check flags and event code/umask, and set the HSW store flag */
  250. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
  251. __EVENT_CONSTRAINT(code, n, \
  252. INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  253. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
  254. /* Check flags and event code/umask, and set the HSW load flag */
  255. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
  256. __EVENT_CONSTRAINT(code, n, \
  257. INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
  258. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
  259. /* Check flags and event code/umask, and set the HSW N/A flag */
  260. #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
  261. __EVENT_CONSTRAINT(code, n, \
  262. INTEL_ARCH_EVENT_MASK|INTEL_ARCH_EVENT_MASK, \
  263. HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
  264. /*
  265. * We define the end marker as having a weight of -1
  266. * to enable blacklisting of events using a counter bitmask
  267. * of zero and thus a weight of zero.
  268. * The end marker has a weight that cannot possibly be
  269. * obtained from counting the bits in the bitmask.
  270. */
  271. #define EVENT_CONSTRAINT_END { .weight = -1 }
  272. /*
  273. * Check for end marker with weight == -1
  274. */
  275. #define for_each_event_constraint(e, c) \
  276. for ((e) = (c); (e)->weight != -1; (e)++)
  277. /*
  278. * Extra registers for specific events.
  279. *
  280. * Some events need large masks and require external MSRs.
  281. * Those extra MSRs end up being shared for all events on
  282. * a PMU and sometimes between PMU of sibling HT threads.
  283. * In either case, the kernel needs to handle conflicting
  284. * accesses to those extra, shared, regs. The data structure
  285. * to manage those registers is stored in cpu_hw_event.
  286. */
  287. struct extra_reg {
  288. unsigned int event;
  289. unsigned int msr;
  290. u64 config_mask;
  291. u64 valid_mask;
  292. int idx; /* per_xxx->regs[] reg index */
  293. bool extra_msr_access;
  294. };
  295. #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
  296. .event = (e), \
  297. .msr = (ms), \
  298. .config_mask = (m), \
  299. .valid_mask = (vm), \
  300. .idx = EXTRA_REG_##i, \
  301. .extra_msr_access = true, \
  302. }
  303. #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
  304. EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
  305. #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
  306. EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
  307. ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
  308. #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
  309. INTEL_UEVENT_EXTRA_REG(c, \
  310. MSR_PEBS_LD_LAT_THRESHOLD, \
  311. 0xffff, \
  312. LDLAT)
  313. #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
  314. union perf_capabilities {
  315. struct {
  316. u64 lbr_format:6;
  317. u64 pebs_trap:1;
  318. u64 pebs_arch_reg:1;
  319. u64 pebs_format:4;
  320. u64 smm_freeze:1;
  321. /*
  322. * PMU supports separate counter range for writing
  323. * values > 32bit.
  324. */
  325. u64 full_width_write:1;
  326. };
  327. u64 capabilities;
  328. };
  329. struct x86_pmu_quirk {
  330. struct x86_pmu_quirk *next;
  331. void (*func)(void);
  332. };
  333. union x86_pmu_config {
  334. struct {
  335. u64 event:8,
  336. umask:8,
  337. usr:1,
  338. os:1,
  339. edge:1,
  340. pc:1,
  341. interrupt:1,
  342. __reserved1:1,
  343. en:1,
  344. inv:1,
  345. cmask:8,
  346. event2:4,
  347. __reserved2:4,
  348. go:1,
  349. ho:1;
  350. } bits;
  351. u64 value;
  352. };
  353. #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
  354. /*
  355. * struct x86_pmu - generic x86 pmu
  356. */
  357. struct x86_pmu {
  358. /*
  359. * Generic x86 PMC bits
  360. */
  361. const char *name;
  362. int version;
  363. int (*handle_irq)(struct pt_regs *);
  364. void (*disable_all)(void);
  365. void (*enable_all)(int added);
  366. void (*enable)(struct perf_event *);
  367. void (*disable)(struct perf_event *);
  368. int (*hw_config)(struct perf_event *event);
  369. int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
  370. unsigned eventsel;
  371. unsigned perfctr;
  372. int (*addr_offset)(int index, bool eventsel);
  373. int (*rdpmc_index)(int index);
  374. u64 (*event_map)(int);
  375. int max_events;
  376. int num_counters;
  377. int num_counters_fixed;
  378. int cntval_bits;
  379. u64 cntval_mask;
  380. union {
  381. unsigned long events_maskl;
  382. unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
  383. };
  384. int events_mask_len;
  385. int apic;
  386. u64 max_period;
  387. struct event_constraint *
  388. (*get_event_constraints)(struct cpu_hw_events *cpuc,
  389. struct perf_event *event);
  390. void (*put_event_constraints)(struct cpu_hw_events *cpuc,
  391. struct perf_event *event);
  392. struct event_constraint *event_constraints;
  393. struct x86_pmu_quirk *quirks;
  394. int perfctr_second_write;
  395. bool late_ack;
  396. unsigned (*limit_period)(struct perf_event *event, unsigned l);
  397. /*
  398. * sysfs attrs
  399. */
  400. int attr_rdpmc_broken;
  401. int attr_rdpmc;
  402. struct attribute **format_attrs;
  403. struct attribute **event_attrs;
  404. ssize_t (*events_sysfs_show)(char *page, u64 config);
  405. struct attribute **cpu_events;
  406. /*
  407. * CPU Hotplug hooks
  408. */
  409. int (*cpu_prepare)(int cpu);
  410. void (*cpu_starting)(int cpu);
  411. void (*cpu_dying)(int cpu);
  412. void (*cpu_dead)(int cpu);
  413. void (*check_microcode)(void);
  414. void (*flush_branch_stack)(void);
  415. /*
  416. * Intel Arch Perfmon v2+
  417. */
  418. u64 intel_ctrl;
  419. union perf_capabilities intel_cap;
  420. /*
  421. * Intel DebugStore bits
  422. */
  423. unsigned int bts :1,
  424. bts_active :1,
  425. pebs :1,
  426. pebs_active :1,
  427. pebs_broken :1;
  428. int pebs_record_size;
  429. void (*drain_pebs)(struct pt_regs *regs);
  430. struct event_constraint *pebs_constraints;
  431. void (*pebs_aliases)(struct perf_event *event);
  432. int max_pebs_events;
  433. /*
  434. * Intel LBR
  435. */
  436. unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
  437. int lbr_nr; /* hardware stack size */
  438. u64 lbr_sel_mask; /* LBR_SELECT valid bits */
  439. const int *lbr_sel_map; /* lbr_select mappings */
  440. bool lbr_double_abort; /* duplicated lbr aborts */
  441. /*
  442. * Extra registers for events
  443. */
  444. struct extra_reg *extra_regs;
  445. unsigned int er_flags;
  446. /*
  447. * Intel host/guest support (KVM)
  448. */
  449. struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
  450. };
  451. #define x86_add_quirk(func_) \
  452. do { \
  453. static struct x86_pmu_quirk __quirk __initdata = { \
  454. .func = func_, \
  455. }; \
  456. __quirk.next = x86_pmu.quirks; \
  457. x86_pmu.quirks = &__quirk; \
  458. } while (0)
  459. #define ERF_NO_HT_SHARING 1
  460. #define ERF_HAS_RSP_1 2
  461. #define EVENT_VAR(_id) event_attr_##_id
  462. #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
  463. #define EVENT_ATTR(_name, _id) \
  464. static struct perf_pmu_events_attr EVENT_VAR(_id) = { \
  465. .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
  466. .id = PERF_COUNT_HW_##_id, \
  467. .event_str = NULL, \
  468. };
  469. #define EVENT_ATTR_STR(_name, v, str) \
  470. static struct perf_pmu_events_attr event_attr_##v = { \
  471. .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \
  472. .id = 0, \
  473. .event_str = str, \
  474. };
  475. extern struct x86_pmu x86_pmu __read_mostly;
  476. DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
  477. int x86_perf_event_set_period(struct perf_event *event);
  478. /*
  479. * Generalized hw caching related hw_event table, filled
  480. * in on a per model basis. A value of 0 means
  481. * 'not supported', -1 means 'hw_event makes no sense on
  482. * this CPU', any other value means the raw hw_event
  483. * ID.
  484. */
  485. #define C(x) PERF_COUNT_HW_CACHE_##x
  486. extern u64 __read_mostly hw_cache_event_ids
  487. [PERF_COUNT_HW_CACHE_MAX]
  488. [PERF_COUNT_HW_CACHE_OP_MAX]
  489. [PERF_COUNT_HW_CACHE_RESULT_MAX];
  490. extern u64 __read_mostly hw_cache_extra_regs
  491. [PERF_COUNT_HW_CACHE_MAX]
  492. [PERF_COUNT_HW_CACHE_OP_MAX]
  493. [PERF_COUNT_HW_CACHE_RESULT_MAX];
  494. u64 x86_perf_event_update(struct perf_event *event);
  495. static inline unsigned int x86_pmu_config_addr(int index)
  496. {
  497. return x86_pmu.eventsel + (x86_pmu.addr_offset ?
  498. x86_pmu.addr_offset(index, true) : index);
  499. }
  500. static inline unsigned int x86_pmu_event_addr(int index)
  501. {
  502. return x86_pmu.perfctr + (x86_pmu.addr_offset ?
  503. x86_pmu.addr_offset(index, false) : index);
  504. }
  505. static inline int x86_pmu_rdpmc_index(int index)
  506. {
  507. return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
  508. }
  509. int x86_setup_perfctr(struct perf_event *event);
  510. int x86_pmu_hw_config(struct perf_event *event);
  511. void x86_pmu_disable_all(void);
  512. static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
  513. u64 enable_mask)
  514. {
  515. u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
  516. if (hwc->extra_reg.reg)
  517. wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
  518. wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
  519. }
  520. void x86_pmu_enable_all(int added);
  521. int perf_assign_events(struct perf_event **events, int n,
  522. int wmin, int wmax, int *assign);
  523. int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
  524. void x86_pmu_stop(struct perf_event *event, int flags);
  525. static inline void x86_pmu_disable_event(struct perf_event *event)
  526. {
  527. struct hw_perf_event *hwc = &event->hw;
  528. wrmsrl(hwc->config_base, hwc->config);
  529. }
  530. void x86_pmu_enable_event(struct perf_event *event);
  531. int x86_pmu_handle_irq(struct pt_regs *regs);
  532. extern struct event_constraint emptyconstraint;
  533. extern struct event_constraint unconstrained;
  534. static inline bool kernel_ip(unsigned long ip)
  535. {
  536. #ifdef CONFIG_X86_32
  537. return ip > PAGE_OFFSET;
  538. #else
  539. return (long)ip < 0;
  540. #endif
  541. }
  542. /*
  543. * Not all PMUs provide the right context information to place the reported IP
  544. * into full context. Specifically segment registers are typically not
  545. * supplied.
  546. *
  547. * Assuming the address is a linear address (it is for IBS), we fake the CS and
  548. * vm86 mode using the known zero-based code segment and 'fix up' the registers
  549. * to reflect this.
  550. *
  551. * Intel PEBS/LBR appear to typically provide the effective address, nothing
  552. * much we can do about that but pray and treat it like a linear address.
  553. */
  554. static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
  555. {
  556. regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
  557. if (regs->flags & X86_VM_MASK)
  558. regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
  559. regs->ip = ip;
  560. }
  561. ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
  562. ssize_t intel_event_sysfs_show(char *page, u64 config);
  563. #ifdef CONFIG_CPU_SUP_AMD
  564. int amd_pmu_init(void);
  565. #else /* CONFIG_CPU_SUP_AMD */
  566. static inline int amd_pmu_init(void)
  567. {
  568. return 0;
  569. }
  570. #endif /* CONFIG_CPU_SUP_AMD */
  571. #ifdef CONFIG_CPU_SUP_INTEL
  572. int intel_pmu_save_and_restart(struct perf_event *event);
  573. struct event_constraint *
  574. x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event);
  575. struct intel_shared_regs *allocate_shared_regs(int cpu);
  576. int intel_pmu_init(void);
  577. void init_debug_store_on_cpu(int cpu);
  578. void fini_debug_store_on_cpu(int cpu);
  579. void release_ds_buffers(void);
  580. void reserve_ds_buffers(void);
  581. extern struct event_constraint bts_constraint;
  582. void intel_pmu_enable_bts(u64 config);
  583. void intel_pmu_disable_bts(void);
  584. int intel_pmu_drain_bts_buffer(void);
  585. extern struct event_constraint intel_core2_pebs_event_constraints[];
  586. extern struct event_constraint intel_atom_pebs_event_constraints[];
  587. extern struct event_constraint intel_slm_pebs_event_constraints[];
  588. extern struct event_constraint intel_nehalem_pebs_event_constraints[];
  589. extern struct event_constraint intel_westmere_pebs_event_constraints[];
  590. extern struct event_constraint intel_snb_pebs_event_constraints[];
  591. extern struct event_constraint intel_ivb_pebs_event_constraints[];
  592. extern struct event_constraint intel_hsw_pebs_event_constraints[];
  593. struct event_constraint *intel_pebs_constraints(struct perf_event *event);
  594. void intel_pmu_pebs_enable(struct perf_event *event);
  595. void intel_pmu_pebs_disable(struct perf_event *event);
  596. void intel_pmu_pebs_enable_all(void);
  597. void intel_pmu_pebs_disable_all(void);
  598. void intel_ds_init(void);
  599. void intel_pmu_lbr_reset(void);
  600. void intel_pmu_lbr_enable(struct perf_event *event);
  601. void intel_pmu_lbr_disable(struct perf_event *event);
  602. void intel_pmu_lbr_enable_all(void);
  603. void intel_pmu_lbr_disable_all(void);
  604. void intel_pmu_lbr_read(void);
  605. void intel_pmu_lbr_init_core(void);
  606. void intel_pmu_lbr_init_nhm(void);
  607. void intel_pmu_lbr_init_atom(void);
  608. void intel_pmu_lbr_init_snb(void);
  609. int intel_pmu_setup_lbr_filter(struct perf_event *event);
  610. int p4_pmu_init(void);
  611. int p6_pmu_init(void);
  612. int knc_pmu_init(void);
  613. ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
  614. char *page);
  615. #else /* CONFIG_CPU_SUP_INTEL */
  616. static inline void reserve_ds_buffers(void)
  617. {
  618. }
  619. static inline void release_ds_buffers(void)
  620. {
  621. }
  622. static inline int intel_pmu_init(void)
  623. {
  624. return 0;
  625. }
  626. static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
  627. {
  628. return NULL;
  629. }
  630. #endif /* CONFIG_CPU_SUP_INTEL */