perf_event.h 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516
  1. /*
  2. * Performance events:
  3. *
  4. * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
  5. * Copyright (C) 2008-2011, Red Hat, Inc., Ingo Molnar
  6. * Copyright (C) 2008-2011, Red Hat, Inc., Peter Zijlstra
  7. *
  8. * Data type definitions, declarations, prototypes.
  9. *
  10. * Started by: Thomas Gleixner and Ingo Molnar
  11. *
  12. * For licencing details see kernel-base/COPYING
  13. */
  14. #ifndef _LINUX_PERF_EVENT_H
  15. #define _LINUX_PERF_EVENT_H
  16. #include <linux/types.h>
  17. #include <linux/ioctl.h>
  18. #include <asm/byteorder.h>
  19. /*
  20. * User-space ABI bits:
  21. */
  22. /*
  23. * attr.type
  24. */
  25. enum perf_type_id {
  26. PERF_TYPE_HARDWARE = 0,
  27. PERF_TYPE_SOFTWARE = 1,
  28. PERF_TYPE_TRACEPOINT = 2,
  29. PERF_TYPE_HW_CACHE = 3,
  30. PERF_TYPE_RAW = 4,
  31. PERF_TYPE_BREAKPOINT = 5,
  32. PERF_TYPE_MAX, /* non-ABI */
  33. };
  34. /*
  35. * Generalized performance event event_id types, used by the
  36. * attr.event_id parameter of the sys_perf_event_open()
  37. * syscall:
  38. */
  39. enum perf_hw_id {
  40. /*
  41. * Common hardware events, generalized by the kernel:
  42. */
  43. PERF_COUNT_HW_CPU_CYCLES = 0,
  44. PERF_COUNT_HW_INSTRUCTIONS = 1,
  45. PERF_COUNT_HW_CACHE_REFERENCES = 2,
  46. PERF_COUNT_HW_CACHE_MISSES = 3,
  47. PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4,
  48. PERF_COUNT_HW_BRANCH_MISSES = 5,
  49. PERF_COUNT_HW_BUS_CYCLES = 6,
  50. PERF_COUNT_HW_STALLED_CYCLES_FRONTEND = 7,
  51. PERF_COUNT_HW_STALLED_CYCLES_BACKEND = 8,
  52. PERF_COUNT_HW_MAX, /* non-ABI */
  53. };
  54. /*
  55. * Generalized hardware cache events:
  56. *
  57. * { L1-D, L1-I, LLC, ITLB, DTLB, BPU, NODE } x
  58. * { read, write, prefetch } x
  59. * { accesses, misses }
  60. */
  61. enum perf_hw_cache_id {
  62. PERF_COUNT_HW_CACHE_L1D = 0,
  63. PERF_COUNT_HW_CACHE_L1I = 1,
  64. PERF_COUNT_HW_CACHE_LL = 2,
  65. PERF_COUNT_HW_CACHE_DTLB = 3,
  66. PERF_COUNT_HW_CACHE_ITLB = 4,
  67. PERF_COUNT_HW_CACHE_BPU = 5,
  68. PERF_COUNT_HW_CACHE_NODE = 6,
  69. PERF_COUNT_HW_CACHE_MAX, /* non-ABI */
  70. };
  71. enum perf_hw_cache_op_id {
  72. PERF_COUNT_HW_CACHE_OP_READ = 0,
  73. PERF_COUNT_HW_CACHE_OP_WRITE = 1,
  74. PERF_COUNT_HW_CACHE_OP_PREFETCH = 2,
  75. PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */
  76. };
  77. enum perf_hw_cache_op_result_id {
  78. PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0,
  79. PERF_COUNT_HW_CACHE_RESULT_MISS = 1,
  80. PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */
  81. };
  82. /*
  83. * Special "software" events provided by the kernel, even if the hardware
  84. * does not support performance events. These events measure various
  85. * physical and sw events of the kernel (and allow the profiling of them as
  86. * well):
  87. */
  88. enum perf_sw_ids {
  89. PERF_COUNT_SW_CPU_CLOCK = 0,
  90. PERF_COUNT_SW_TASK_CLOCK = 1,
  91. PERF_COUNT_SW_PAGE_FAULTS = 2,
  92. PERF_COUNT_SW_CONTEXT_SWITCHES = 3,
  93. PERF_COUNT_SW_CPU_MIGRATIONS = 4,
  94. PERF_COUNT_SW_PAGE_FAULTS_MIN = 5,
  95. PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6,
  96. PERF_COUNT_SW_ALIGNMENT_FAULTS = 7,
  97. PERF_COUNT_SW_EMULATION_FAULTS = 8,
  98. PERF_COUNT_SW_MAX, /* non-ABI */
  99. };
  100. /*
  101. * Bits that can be set in attr.sample_type to request information
  102. * in the overflow packets.
  103. */
  104. enum perf_event_sample_format {
  105. PERF_SAMPLE_IP = 1U << 0,
  106. PERF_SAMPLE_TID = 1U << 1,
  107. PERF_SAMPLE_TIME = 1U << 2,
  108. PERF_SAMPLE_ADDR = 1U << 3,
  109. PERF_SAMPLE_READ = 1U << 4,
  110. PERF_SAMPLE_CALLCHAIN = 1U << 5,
  111. PERF_SAMPLE_ID = 1U << 6,
  112. PERF_SAMPLE_CPU = 1U << 7,
  113. PERF_SAMPLE_PERIOD = 1U << 8,
  114. PERF_SAMPLE_STREAM_ID = 1U << 9,
  115. PERF_SAMPLE_RAW = 1U << 10,
  116. PERF_SAMPLE_BRANCH_STACK = 1U << 11,
  117. PERF_SAMPLE_MAX = 1U << 12, /* non-ABI */
  118. };
  119. /*
  120. * values to program into branch_sample_type when PERF_SAMPLE_BRANCH is set
  121. *
  122. * If the user does not pass priv level information via branch_sample_type,
  123. * the kernel uses the event's priv level. Branch and event priv levels do
  124. * not have to match. Branch priv level is checked for permissions.
  125. *
  126. * The branch types can be combined, however BRANCH_ANY covers all types
  127. * of branches and therefore it supersedes all the other types.
  128. */
  129. enum perf_branch_sample_type {
  130. PERF_SAMPLE_BRANCH_USER = 1U << 0, /* user branches */
  131. PERF_SAMPLE_BRANCH_KERNEL = 1U << 1, /* kernel branches */
  132. PERF_SAMPLE_BRANCH_HV = 1U << 2, /* hypervisor branches */
  133. PERF_SAMPLE_BRANCH_ANY = 1U << 3, /* any branch types */
  134. PERF_SAMPLE_BRANCH_ANY_CALL = 1U << 4, /* any call branch */
  135. PERF_SAMPLE_BRANCH_ANY_RETURN = 1U << 5, /* any return branch */
  136. PERF_SAMPLE_BRANCH_IND_CALL = 1U << 6, /* indirect calls */
  137. PERF_SAMPLE_BRANCH_MAX = 1U << 7, /* non-ABI */
  138. };
  139. #define PERF_SAMPLE_BRANCH_PLM_ALL \
  140. (PERF_SAMPLE_BRANCH_USER|\
  141. PERF_SAMPLE_BRANCH_KERNEL|\
  142. PERF_SAMPLE_BRANCH_HV)
  143. /*
  144. * The format of the data returned by read() on a perf event fd,
  145. * as specified by attr.read_format:
  146. *
  147. * struct read_format {
  148. * { u64 value;
  149. * { u64 time_enabled; } && PERF_FORMAT_TOTAL_TIME_ENABLED
  150. * { u64 time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  151. * { u64 id; } && PERF_FORMAT_ID
  152. * } && !PERF_FORMAT_GROUP
  153. *
  154. * { u64 nr;
  155. * { u64 time_enabled; } && PERF_FORMAT_TOTAL_TIME_ENABLED
  156. * { u64 time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  157. * { u64 value;
  158. * { u64 id; } && PERF_FORMAT_ID
  159. * } cntr[nr];
  160. * } && PERF_FORMAT_GROUP
  161. * };
  162. */
  163. enum perf_event_read_format {
  164. PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,
  165. PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,
  166. PERF_FORMAT_ID = 1U << 2,
  167. PERF_FORMAT_GROUP = 1U << 3,
  168. PERF_FORMAT_MAX = 1U << 4, /* non-ABI */
  169. };
  170. #define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */
  171. /*
  172. * Hardware event_id to monitor via a performance monitoring event:
  173. */
  174. struct perf_event_attr {
  175. /*
  176. * Major type: hardware/software/tracepoint/etc.
  177. */
  178. __u32 type;
  179. /*
  180. * Size of the attr structure, for fwd/bwd compat.
  181. */
  182. __u32 size;
  183. /*
  184. * Type specific configuration information.
  185. */
  186. __u64 config;
  187. union {
  188. __u64 sample_period;
  189. __u64 sample_freq;
  190. };
  191. __u64 sample_type;
  192. __u64 read_format;
  193. __u64 disabled : 1, /* off by default */
  194. inherit : 1, /* children inherit it */
  195. pinned : 1, /* must always be on PMU */
  196. exclusive : 1, /* only group on PMU */
  197. exclude_user : 1, /* don't count user */
  198. exclude_kernel : 1, /* ditto kernel */
  199. exclude_hv : 1, /* ditto hypervisor */
  200. exclude_idle : 1, /* don't count when idle */
  201. mmap : 1, /* include mmap data */
  202. comm : 1, /* include comm data */
  203. freq : 1, /* use freq, not period */
  204. inherit_stat : 1, /* per task counts */
  205. enable_on_exec : 1, /* next exec enables */
  206. task : 1, /* trace fork/exit */
  207. watermark : 1, /* wakeup_watermark */
  208. /*
  209. * precise_ip:
  210. *
  211. * 0 - SAMPLE_IP can have arbitrary skid
  212. * 1 - SAMPLE_IP must have constant skid
  213. * 2 - SAMPLE_IP requested to have 0 skid
  214. * 3 - SAMPLE_IP must have 0 skid
  215. *
  216. * See also PERF_RECORD_MISC_EXACT_IP
  217. */
  218. precise_ip : 2, /* skid constraint */
  219. mmap_data : 1, /* non-exec mmap data */
  220. sample_id_all : 1, /* sample_type all events */
  221. exclude_host : 1, /* don't count in host */
  222. exclude_guest : 1, /* don't count in guest */
  223. __reserved_1 : 43;
  224. union {
  225. __u32 wakeup_events; /* wakeup every n events */
  226. __u32 wakeup_watermark; /* bytes before wakeup */
  227. };
  228. __u32 bp_type;
  229. union {
  230. __u64 bp_addr;
  231. __u64 config1; /* extension of config */
  232. };
  233. union {
  234. __u64 bp_len;
  235. __u64 config2; /* extension of config1 */
  236. };
  237. __u64 branch_sample_type; /* enum branch_sample_type */
  238. };
  239. /*
  240. * Ioctls that can be done on a perf event fd:
  241. */
  242. #define PERF_EVENT_IOC_ENABLE _IO ('$', 0)
  243. #define PERF_EVENT_IOC_DISABLE _IO ('$', 1)
  244. #define PERF_EVENT_IOC_REFRESH _IO ('$', 2)
  245. #define PERF_EVENT_IOC_RESET _IO ('$', 3)
  246. #define PERF_EVENT_IOC_PERIOD _IOW('$', 4, __u64)
  247. #define PERF_EVENT_IOC_SET_OUTPUT _IO ('$', 5)
  248. #define PERF_EVENT_IOC_SET_FILTER _IOW('$', 6, char *)
  249. enum perf_event_ioc_flags {
  250. PERF_IOC_FLAG_GROUP = 1U << 0,
  251. };
  252. /*
  253. * Structure of the page that can be mapped via mmap
  254. */
  255. struct perf_event_mmap_page {
  256. __u32 version; /* version number of this structure */
  257. __u32 compat_version; /* lowest version this is compat with */
  258. /*
  259. * Bits needed to read the hw events in user-space.
  260. *
  261. * u32 seq;
  262. * s64 count;
  263. *
  264. * do {
  265. * seq = pc->lock;
  266. *
  267. * barrier()
  268. * if (pc->index) {
  269. * count = pmc_read(pc->index - 1);
  270. * count += pc->offset;
  271. * } else
  272. * goto regular_read;
  273. *
  274. * barrier();
  275. * } while (pc->lock != seq);
  276. *
  277. * NOTE: for obvious reason this only works on self-monitoring
  278. * processes.
  279. */
  280. __u32 lock; /* seqlock for synchronization */
  281. __u32 index; /* hardware event identifier */
  282. __s64 offset; /* add to hardware event value */
  283. __u64 time_enabled; /* time event active */
  284. __u64 time_running; /* time event on cpu */
  285. /*
  286. * Hole for extension of the self monitor capabilities
  287. */
  288. __u64 __reserved[123]; /* align to 1k */
  289. /*
  290. * Control data for the mmap() data buffer.
  291. *
  292. * User-space reading the @data_head value should issue an rmb(), on
  293. * SMP capable platforms, after reading this value -- see
  294. * perf_event_wakeup().
  295. *
  296. * When the mapping is PROT_WRITE the @data_tail value should be
  297. * written by userspace to reflect the last read data. In this case
  298. * the kernel will not over-write unread data.
  299. */
  300. __u64 data_head; /* head in the data section */
  301. __u64 data_tail; /* user-space written tail */
  302. };
  303. #define PERF_RECORD_MISC_CPUMODE_MASK (7 << 0)
  304. #define PERF_RECORD_MISC_CPUMODE_UNKNOWN (0 << 0)
  305. #define PERF_RECORD_MISC_KERNEL (1 << 0)
  306. #define PERF_RECORD_MISC_USER (2 << 0)
  307. #define PERF_RECORD_MISC_HYPERVISOR (3 << 0)
  308. #define PERF_RECORD_MISC_GUEST_KERNEL (4 << 0)
  309. #define PERF_RECORD_MISC_GUEST_USER (5 << 0)
  310. /*
  311. * Indicates that the content of PERF_SAMPLE_IP points to
  312. * the actual instruction that triggered the event. See also
  313. * perf_event_attr::precise_ip.
  314. */
  315. #define PERF_RECORD_MISC_EXACT_IP (1 << 14)
  316. /*
  317. * Reserve the last bit to indicate some extended misc field
  318. */
  319. #define PERF_RECORD_MISC_EXT_RESERVED (1 << 15)
  320. struct perf_event_header {
  321. __u32 type;
  322. __u16 misc;
  323. __u16 size;
  324. };
  325. enum perf_event_type {
  326. /*
  327. * If perf_event_attr.sample_id_all is set then all event types will
  328. * have the sample_type selected fields related to where/when
  329. * (identity) an event took place (TID, TIME, ID, CPU, STREAM_ID)
  330. * described in PERF_RECORD_SAMPLE below, it will be stashed just after
  331. * the perf_event_header and the fields already present for the existing
  332. * fields, i.e. at the end of the payload. That way a newer perf.data
  333. * file will be supported by older perf tools, with these new optional
  334. * fields being ignored.
  335. *
  336. * The MMAP events record the PROT_EXEC mappings so that we can
  337. * correlate userspace IPs to code. They have the following structure:
  338. *
  339. * struct {
  340. * struct perf_event_header header;
  341. *
  342. * u32 pid, tid;
  343. * u64 addr;
  344. * u64 len;
  345. * u64 pgoff;
  346. * char filename[];
  347. * };
  348. */
  349. PERF_RECORD_MMAP = 1,
  350. /*
  351. * struct {
  352. * struct perf_event_header header;
  353. * u64 id;
  354. * u64 lost;
  355. * };
  356. */
  357. PERF_RECORD_LOST = 2,
  358. /*
  359. * struct {
  360. * struct perf_event_header header;
  361. *
  362. * u32 pid, tid;
  363. * char comm[];
  364. * };
  365. */
  366. PERF_RECORD_COMM = 3,
  367. /*
  368. * struct {
  369. * struct perf_event_header header;
  370. * u32 pid, ppid;
  371. * u32 tid, ptid;
  372. * u64 time;
  373. * };
  374. */
  375. PERF_RECORD_EXIT = 4,
  376. /*
  377. * struct {
  378. * struct perf_event_header header;
  379. * u64 time;
  380. * u64 id;
  381. * u64 stream_id;
  382. * };
  383. */
  384. PERF_RECORD_THROTTLE = 5,
  385. PERF_RECORD_UNTHROTTLE = 6,
  386. /*
  387. * struct {
  388. * struct perf_event_header header;
  389. * u32 pid, ppid;
  390. * u32 tid, ptid;
  391. * u64 time;
  392. * };
  393. */
  394. PERF_RECORD_FORK = 7,
  395. /*
  396. * struct {
  397. * struct perf_event_header header;
  398. * u32 pid, tid;
  399. *
  400. * struct read_format values;
  401. * };
  402. */
  403. PERF_RECORD_READ = 8,
  404. /*
  405. * struct {
  406. * struct perf_event_header header;
  407. *
  408. * { u64 ip; } && PERF_SAMPLE_IP
  409. * { u32 pid, tid; } && PERF_SAMPLE_TID
  410. * { u64 time; } && PERF_SAMPLE_TIME
  411. * { u64 addr; } && PERF_SAMPLE_ADDR
  412. * { u64 id; } && PERF_SAMPLE_ID
  413. * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID
  414. * { u32 cpu, res; } && PERF_SAMPLE_CPU
  415. * { u64 period; } && PERF_SAMPLE_PERIOD
  416. *
  417. * { struct read_format values; } && PERF_SAMPLE_READ
  418. *
  419. * { u64 nr,
  420. * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
  421. *
  422. * #
  423. * # The RAW record below is opaque data wrt the ABI
  424. * #
  425. * # That is, the ABI doesn't make any promises wrt to
  426. * # the stability of its content, it may vary depending
  427. * # on event, hardware, kernel version and phase of
  428. * # the moon.
  429. * #
  430. * # In other words, PERF_SAMPLE_RAW contents are not an ABI.
  431. * #
  432. *
  433. * { u32 size;
  434. * char data[size];}&& PERF_SAMPLE_RAW
  435. *
  436. * { u64 from, to, flags } lbr[nr];} && PERF_SAMPLE_BRANCH_STACK
  437. * };
  438. */
  439. PERF_RECORD_SAMPLE = 9,
  440. PERF_RECORD_MAX, /* non-ABI */
  441. };
  442. #define PERF_MAX_STACK_DEPTH 255
  443. enum perf_callchain_context {
  444. PERF_CONTEXT_HV = (__u64)-32,
  445. PERF_CONTEXT_KERNEL = (__u64)-128,
  446. PERF_CONTEXT_USER = (__u64)-512,
  447. PERF_CONTEXT_GUEST = (__u64)-2048,
  448. PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176,
  449. PERF_CONTEXT_GUEST_USER = (__u64)-2560,
  450. PERF_CONTEXT_MAX = (__u64)-4095,
  451. };
  452. #define PERF_FLAG_FD_NO_GROUP (1U << 0)
  453. #define PERF_FLAG_FD_OUTPUT (1U << 1)
  454. #define PERF_FLAG_PID_CGROUP (1U << 2) /* pid=cgroup id, per-cpu mode only */
  455. #endif /* _LINUX_PERF_EVENT_H */