logo

oasis-root

Compiled tree of Oasis Linux based on own branch at <https://hacktivis.me/git/oasis/> git clone https://anongit.hacktivis.me/git/oasis-root.git

perf_event.h (43445B)


  1. /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
  2. /*
  3. * Performance events:
  4. *
  5. * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
  6. * Copyright (C) 2008-2011, Red Hat, Inc., Ingo Molnar
  7. * Copyright (C) 2008-2011, Red Hat, Inc., Peter Zijlstra
  8. *
  9. * Data type definitions, declarations, prototypes.
  10. *
  11. * Started by: Thomas Gleixner and Ingo Molnar
  12. *
  13. * For licencing details see kernel-base/COPYING
  14. */
  15. #ifndef _LINUX_PERF_EVENT_H
  16. #define _LINUX_PERF_EVENT_H
  17. #include <linux/types.h>
  18. #include <linux/ioctl.h>
  19. #include <asm/byteorder.h>
  20. /*
  21. * User-space ABI bits:
  22. */
  23. /*
  24. * attr.type
  25. */
  26. enum perf_type_id {
  27. PERF_TYPE_HARDWARE = 0,
  28. PERF_TYPE_SOFTWARE = 1,
  29. PERF_TYPE_TRACEPOINT = 2,
  30. PERF_TYPE_HW_CACHE = 3,
  31. PERF_TYPE_RAW = 4,
  32. PERF_TYPE_BREAKPOINT = 5,
  33. PERF_TYPE_MAX, /* non-ABI */
  34. };
  35. /*
  36. * attr.config layout for type PERF_TYPE_HARDWARE and PERF_TYPE_HW_CACHE
  37. * PERF_TYPE_HARDWARE: 0xEEEEEEEE000000AA
  38. * AA: hardware event ID
  39. * EEEEEEEE: PMU type ID
  40. * PERF_TYPE_HW_CACHE: 0xEEEEEEEE00DDCCBB
  41. * BB: hardware cache ID
  42. * CC: hardware cache op ID
  43. * DD: hardware cache op result ID
  44. * EEEEEEEE: PMU type ID
  45. * If the PMU type ID is 0, the PERF_TYPE_RAW will be applied.
  46. */
  47. #define PERF_PMU_TYPE_SHIFT 32
  48. #define PERF_HW_EVENT_MASK 0xffffffff
  49. /*
  50. * Generalized performance event event_id types, used by the
  51. * attr.event_id parameter of the sys_perf_event_open()
  52. * syscall:
  53. */
  54. enum perf_hw_id {
  55. /*
  56. * Common hardware events, generalized by the kernel:
  57. */
  58. PERF_COUNT_HW_CPU_CYCLES = 0,
  59. PERF_COUNT_HW_INSTRUCTIONS = 1,
  60. PERF_COUNT_HW_CACHE_REFERENCES = 2,
  61. PERF_COUNT_HW_CACHE_MISSES = 3,
  62. PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4,
  63. PERF_COUNT_HW_BRANCH_MISSES = 5,
  64. PERF_COUNT_HW_BUS_CYCLES = 6,
  65. PERF_COUNT_HW_STALLED_CYCLES_FRONTEND = 7,
  66. PERF_COUNT_HW_STALLED_CYCLES_BACKEND = 8,
  67. PERF_COUNT_HW_REF_CPU_CYCLES = 9,
  68. PERF_COUNT_HW_MAX, /* non-ABI */
  69. };
  70. /*
  71. * Generalized hardware cache events:
  72. *
  73. * { L1-D, L1-I, LLC, ITLB, DTLB, BPU, NODE } x
  74. * { read, write, prefetch } x
  75. * { accesses, misses }
  76. */
  77. enum perf_hw_cache_id {
  78. PERF_COUNT_HW_CACHE_L1D = 0,
  79. PERF_COUNT_HW_CACHE_L1I = 1,
  80. PERF_COUNT_HW_CACHE_LL = 2,
  81. PERF_COUNT_HW_CACHE_DTLB = 3,
  82. PERF_COUNT_HW_CACHE_ITLB = 4,
  83. PERF_COUNT_HW_CACHE_BPU = 5,
  84. PERF_COUNT_HW_CACHE_NODE = 6,
  85. PERF_COUNT_HW_CACHE_MAX, /* non-ABI */
  86. };
  87. enum perf_hw_cache_op_id {
  88. PERF_COUNT_HW_CACHE_OP_READ = 0,
  89. PERF_COUNT_HW_CACHE_OP_WRITE = 1,
  90. PERF_COUNT_HW_CACHE_OP_PREFETCH = 2,
  91. PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */
  92. };
  93. enum perf_hw_cache_op_result_id {
  94. PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0,
  95. PERF_COUNT_HW_CACHE_RESULT_MISS = 1,
  96. PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */
  97. };
  98. /*
  99. * Special "software" events provided by the kernel, even if the hardware
  100. * does not support performance events. These events measure various
  101. * physical and sw events of the kernel (and allow the profiling of them as
  102. * well):
  103. */
  104. enum perf_sw_ids {
  105. PERF_COUNT_SW_CPU_CLOCK = 0,
  106. PERF_COUNT_SW_TASK_CLOCK = 1,
  107. PERF_COUNT_SW_PAGE_FAULTS = 2,
  108. PERF_COUNT_SW_CONTEXT_SWITCHES = 3,
  109. PERF_COUNT_SW_CPU_MIGRATIONS = 4,
  110. PERF_COUNT_SW_PAGE_FAULTS_MIN = 5,
  111. PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6,
  112. PERF_COUNT_SW_ALIGNMENT_FAULTS = 7,
  113. PERF_COUNT_SW_EMULATION_FAULTS = 8,
  114. PERF_COUNT_SW_DUMMY = 9,
  115. PERF_COUNT_SW_BPF_OUTPUT = 10,
  116. PERF_COUNT_SW_CGROUP_SWITCHES = 11,
  117. PERF_COUNT_SW_MAX, /* non-ABI */
  118. };
  119. /*
  120. * Bits that can be set in attr.sample_type to request information
  121. * in the overflow packets.
  122. */
  123. enum perf_event_sample_format {
  124. PERF_SAMPLE_IP = 1U << 0,
  125. PERF_SAMPLE_TID = 1U << 1,
  126. PERF_SAMPLE_TIME = 1U << 2,
  127. PERF_SAMPLE_ADDR = 1U << 3,
  128. PERF_SAMPLE_READ = 1U << 4,
  129. PERF_SAMPLE_CALLCHAIN = 1U << 5,
  130. PERF_SAMPLE_ID = 1U << 6,
  131. PERF_SAMPLE_CPU = 1U << 7,
  132. PERF_SAMPLE_PERIOD = 1U << 8,
  133. PERF_SAMPLE_STREAM_ID = 1U << 9,
  134. PERF_SAMPLE_RAW = 1U << 10,
  135. PERF_SAMPLE_BRANCH_STACK = 1U << 11,
  136. PERF_SAMPLE_REGS_USER = 1U << 12,
  137. PERF_SAMPLE_STACK_USER = 1U << 13,
  138. PERF_SAMPLE_WEIGHT = 1U << 14,
  139. PERF_SAMPLE_DATA_SRC = 1U << 15,
  140. PERF_SAMPLE_IDENTIFIER = 1U << 16,
  141. PERF_SAMPLE_TRANSACTION = 1U << 17,
  142. PERF_SAMPLE_REGS_INTR = 1U << 18,
  143. PERF_SAMPLE_PHYS_ADDR = 1U << 19,
  144. PERF_SAMPLE_AUX = 1U << 20,
  145. PERF_SAMPLE_CGROUP = 1U << 21,
  146. PERF_SAMPLE_DATA_PAGE_SIZE = 1U << 22,
  147. PERF_SAMPLE_CODE_PAGE_SIZE = 1U << 23,
  148. PERF_SAMPLE_WEIGHT_STRUCT = 1U << 24,
  149. PERF_SAMPLE_MAX = 1U << 25, /* non-ABI */
  150. };
  151. #define PERF_SAMPLE_WEIGHT_TYPE (PERF_SAMPLE_WEIGHT | PERF_SAMPLE_WEIGHT_STRUCT)
  152. /*
  153. * values to program into branch_sample_type when PERF_SAMPLE_BRANCH is set
  154. *
  155. * If the user does not pass priv level information via branch_sample_type,
  156. * the kernel uses the event's priv level. Branch and event priv levels do
  157. * not have to match. Branch priv level is checked for permissions.
  158. *
  159. * The branch types can be combined, however BRANCH_ANY covers all types
  160. * of branches and therefore it supersedes all the other types.
  161. */
  162. enum perf_branch_sample_type_shift {
  163. PERF_SAMPLE_BRANCH_USER_SHIFT = 0, /* user branches */
  164. PERF_SAMPLE_BRANCH_KERNEL_SHIFT = 1, /* kernel branches */
  165. PERF_SAMPLE_BRANCH_HV_SHIFT = 2, /* hypervisor branches */
  166. PERF_SAMPLE_BRANCH_ANY_SHIFT = 3, /* any branch types */
  167. PERF_SAMPLE_BRANCH_ANY_CALL_SHIFT = 4, /* any call branch */
  168. PERF_SAMPLE_BRANCH_ANY_RETURN_SHIFT = 5, /* any return branch */
  169. PERF_SAMPLE_BRANCH_IND_CALL_SHIFT = 6, /* indirect calls */
  170. PERF_SAMPLE_BRANCH_ABORT_TX_SHIFT = 7, /* transaction aborts */
  171. PERF_SAMPLE_BRANCH_IN_TX_SHIFT = 8, /* in transaction */
  172. PERF_SAMPLE_BRANCH_NO_TX_SHIFT = 9, /* not in transaction */
  173. PERF_SAMPLE_BRANCH_COND_SHIFT = 10, /* conditional branches */
  174. PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT = 11, /* call/ret stack */
  175. PERF_SAMPLE_BRANCH_IND_JUMP_SHIFT = 12, /* indirect jumps */
  176. PERF_SAMPLE_BRANCH_CALL_SHIFT = 13, /* direct call */
  177. PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT = 14, /* no flags */
  178. PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT = 15, /* no cycles */
  179. PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT = 16, /* save branch type */
  180. PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT = 17, /* save low level index of raw branch records */
  181. PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT = 18, /* save privilege mode */
  182. PERF_SAMPLE_BRANCH_COUNTERS_SHIFT = 19, /* save occurrences of events on a branch */
  183. PERF_SAMPLE_BRANCH_MAX_SHIFT /* non-ABI */
  184. };
  185. enum perf_branch_sample_type {
  186. PERF_SAMPLE_BRANCH_USER = 1U << PERF_SAMPLE_BRANCH_USER_SHIFT,
  187. PERF_SAMPLE_BRANCH_KERNEL = 1U << PERF_SAMPLE_BRANCH_KERNEL_SHIFT,
  188. PERF_SAMPLE_BRANCH_HV = 1U << PERF_SAMPLE_BRANCH_HV_SHIFT,
  189. PERF_SAMPLE_BRANCH_ANY = 1U << PERF_SAMPLE_BRANCH_ANY_SHIFT,
  190. PERF_SAMPLE_BRANCH_ANY_CALL = 1U << PERF_SAMPLE_BRANCH_ANY_CALL_SHIFT,
  191. PERF_SAMPLE_BRANCH_ANY_RETURN = 1U << PERF_SAMPLE_BRANCH_ANY_RETURN_SHIFT,
  192. PERF_SAMPLE_BRANCH_IND_CALL = 1U << PERF_SAMPLE_BRANCH_IND_CALL_SHIFT,
  193. PERF_SAMPLE_BRANCH_ABORT_TX = 1U << PERF_SAMPLE_BRANCH_ABORT_TX_SHIFT,
  194. PERF_SAMPLE_BRANCH_IN_TX = 1U << PERF_SAMPLE_BRANCH_IN_TX_SHIFT,
  195. PERF_SAMPLE_BRANCH_NO_TX = 1U << PERF_SAMPLE_BRANCH_NO_TX_SHIFT,
  196. PERF_SAMPLE_BRANCH_COND = 1U << PERF_SAMPLE_BRANCH_COND_SHIFT,
  197. PERF_SAMPLE_BRANCH_CALL_STACK = 1U << PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT,
  198. PERF_SAMPLE_BRANCH_IND_JUMP = 1U << PERF_SAMPLE_BRANCH_IND_JUMP_SHIFT,
  199. PERF_SAMPLE_BRANCH_CALL = 1U << PERF_SAMPLE_BRANCH_CALL_SHIFT,
  200. PERF_SAMPLE_BRANCH_NO_FLAGS = 1U << PERF_SAMPLE_BRANCH_NO_FLAGS_SHIFT,
  201. PERF_SAMPLE_BRANCH_NO_CYCLES = 1U << PERF_SAMPLE_BRANCH_NO_CYCLES_SHIFT,
  202. PERF_SAMPLE_BRANCH_TYPE_SAVE =
  203. 1U << PERF_SAMPLE_BRANCH_TYPE_SAVE_SHIFT,
  204. PERF_SAMPLE_BRANCH_HW_INDEX = 1U << PERF_SAMPLE_BRANCH_HW_INDEX_SHIFT,
  205. PERF_SAMPLE_BRANCH_PRIV_SAVE = 1U << PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT,
  206. PERF_SAMPLE_BRANCH_COUNTERS = 1U << PERF_SAMPLE_BRANCH_COUNTERS_SHIFT,
  207. PERF_SAMPLE_BRANCH_MAX = 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT,
  208. };
  209. /*
  210. * Common flow change classification
  211. */
  212. enum {
  213. PERF_BR_UNKNOWN = 0, /* unknown */
  214. PERF_BR_COND = 1, /* conditional */
  215. PERF_BR_UNCOND = 2, /* unconditional */
  216. PERF_BR_IND = 3, /* indirect */
  217. PERF_BR_CALL = 4, /* function call */
  218. PERF_BR_IND_CALL = 5, /* indirect function call */
  219. PERF_BR_RET = 6, /* function return */
  220. PERF_BR_SYSCALL = 7, /* syscall */
  221. PERF_BR_SYSRET = 8, /* syscall return */
  222. PERF_BR_COND_CALL = 9, /* conditional function call */
  223. PERF_BR_COND_RET = 10, /* conditional function return */
  224. PERF_BR_ERET = 11, /* exception return */
  225. PERF_BR_IRQ = 12, /* irq */
  226. PERF_BR_SERROR = 13, /* system error */
  227. PERF_BR_NO_TX = 14, /* not in transaction */
  228. PERF_BR_EXTEND_ABI = 15, /* extend ABI */
  229. PERF_BR_MAX,
  230. };
  231. /*
  232. * Common branch speculation outcome classification
  233. */
  234. enum {
  235. PERF_BR_SPEC_NA = 0, /* Not available */
  236. PERF_BR_SPEC_WRONG_PATH = 1, /* Speculative but on wrong path */
  237. PERF_BR_NON_SPEC_CORRECT_PATH = 2, /* Non-speculative but on correct path */
  238. PERF_BR_SPEC_CORRECT_PATH = 3, /* Speculative and on correct path */
  239. PERF_BR_SPEC_MAX,
  240. };
  241. enum {
  242. PERF_BR_NEW_FAULT_ALGN = 0, /* Alignment fault */
  243. PERF_BR_NEW_FAULT_DATA = 1, /* Data fault */
  244. PERF_BR_NEW_FAULT_INST = 2, /* Inst fault */
  245. PERF_BR_NEW_ARCH_1 = 3, /* Architecture specific */
  246. PERF_BR_NEW_ARCH_2 = 4, /* Architecture specific */
  247. PERF_BR_NEW_ARCH_3 = 5, /* Architecture specific */
  248. PERF_BR_NEW_ARCH_4 = 6, /* Architecture specific */
  249. PERF_BR_NEW_ARCH_5 = 7, /* Architecture specific */
  250. PERF_BR_NEW_MAX,
  251. };
  252. enum {
  253. PERF_BR_PRIV_UNKNOWN = 0,
  254. PERF_BR_PRIV_USER = 1,
  255. PERF_BR_PRIV_KERNEL = 2,
  256. PERF_BR_PRIV_HV = 3,
  257. };
  258. #define PERF_BR_ARM64_FIQ PERF_BR_NEW_ARCH_1
  259. #define PERF_BR_ARM64_DEBUG_HALT PERF_BR_NEW_ARCH_2
  260. #define PERF_BR_ARM64_DEBUG_EXIT PERF_BR_NEW_ARCH_3
  261. #define PERF_BR_ARM64_DEBUG_INST PERF_BR_NEW_ARCH_4
  262. #define PERF_BR_ARM64_DEBUG_DATA PERF_BR_NEW_ARCH_5
  263. #define PERF_SAMPLE_BRANCH_PLM_ALL \
  264. (PERF_SAMPLE_BRANCH_USER|\
  265. PERF_SAMPLE_BRANCH_KERNEL|\
  266. PERF_SAMPLE_BRANCH_HV)
  267. /*
  268. * Values to determine ABI of the registers dump.
  269. */
  270. enum perf_sample_regs_abi {
  271. PERF_SAMPLE_REGS_ABI_NONE = 0,
  272. PERF_SAMPLE_REGS_ABI_32 = 1,
  273. PERF_SAMPLE_REGS_ABI_64 = 2,
  274. };
  275. /*
  276. * Values for the memory transaction event qualifier, mostly for
  277. * abort events. Multiple bits can be set.
  278. */
  279. enum {
  280. PERF_TXN_ELISION = (1 << 0), /* From elision */
  281. PERF_TXN_TRANSACTION = (1 << 1), /* From transaction */
  282. PERF_TXN_SYNC = (1 << 2), /* Instruction is related */
  283. PERF_TXN_ASYNC = (1 << 3), /* Instruction not related */
  284. PERF_TXN_RETRY = (1 << 4), /* Retry possible */
  285. PERF_TXN_CONFLICT = (1 << 5), /* Conflict abort */
  286. PERF_TXN_CAPACITY_WRITE = (1 << 6), /* Capacity write abort */
  287. PERF_TXN_CAPACITY_READ = (1 << 7), /* Capacity read abort */
  288. PERF_TXN_MAX = (1 << 8), /* non-ABI */
  289. /* bits 32..63 are reserved for the abort code */
  290. PERF_TXN_ABORT_MASK = (0xffffffffULL << 32),
  291. PERF_TXN_ABORT_SHIFT = 32,
  292. };
  293. /*
  294. * The format of the data returned by read() on a perf event fd,
  295. * as specified by attr.read_format:
  296. *
  297. * struct read_format {
  298. * { u64 value;
  299. * { u64 time_enabled; } && PERF_FORMAT_TOTAL_TIME_ENABLED
  300. * { u64 time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  301. * { u64 id; } && PERF_FORMAT_ID
  302. * { u64 lost; } && PERF_FORMAT_LOST
  303. * } && !PERF_FORMAT_GROUP
  304. *
  305. * { u64 nr;
  306. * { u64 time_enabled; } && PERF_FORMAT_TOTAL_TIME_ENABLED
  307. * { u64 time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  308. * { u64 value;
  309. * { u64 id; } && PERF_FORMAT_ID
  310. * { u64 lost; } && PERF_FORMAT_LOST
  311. * } cntr[nr];
  312. * } && PERF_FORMAT_GROUP
  313. * };
  314. */
  315. enum perf_event_read_format {
  316. PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,
  317. PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,
  318. PERF_FORMAT_ID = 1U << 2,
  319. PERF_FORMAT_GROUP = 1U << 3,
  320. PERF_FORMAT_LOST = 1U << 4,
  321. PERF_FORMAT_MAX = 1U << 5, /* non-ABI */
  322. };
  323. #define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */
  324. #define PERF_ATTR_SIZE_VER1 72 /* add: config2 */
  325. #define PERF_ATTR_SIZE_VER2 80 /* add: branch_sample_type */
  326. #define PERF_ATTR_SIZE_VER3 96 /* add: sample_regs_user */
  327. /* add: sample_stack_user */
  328. #define PERF_ATTR_SIZE_VER4 104 /* add: sample_regs_intr */
  329. #define PERF_ATTR_SIZE_VER5 112 /* add: aux_watermark */
  330. #define PERF_ATTR_SIZE_VER6 120 /* add: aux_sample_size */
  331. #define PERF_ATTR_SIZE_VER7 128 /* add: sig_data */
  332. #define PERF_ATTR_SIZE_VER8 136 /* add: config3 */
  333. /*
  334. * Hardware event_id to monitor via a performance monitoring event:
  335. *
  336. * @sample_max_stack: Max number of frame pointers in a callchain,
  337. * should be < /proc/sys/kernel/perf_event_max_stack
  338. */
  339. struct perf_event_attr {
  340. /*
  341. * Major type: hardware/software/tracepoint/etc.
  342. */
  343. __u32 type;
  344. /*
  345. * Size of the attr structure, for fwd/bwd compat.
  346. */
  347. __u32 size;
  348. /*
  349. * Type specific configuration information.
  350. */
  351. __u64 config;
  352. union {
  353. __u64 sample_period;
  354. __u64 sample_freq;
  355. };
  356. __u64 sample_type;
  357. __u64 read_format;
  358. __u64 disabled : 1, /* off by default */
  359. inherit : 1, /* children inherit it */
  360. pinned : 1, /* must always be on PMU */
  361. exclusive : 1, /* only group on PMU */
  362. exclude_user : 1, /* don't count user */
  363. exclude_kernel : 1, /* ditto kernel */
  364. exclude_hv : 1, /* ditto hypervisor */
  365. exclude_idle : 1, /* don't count when idle */
  366. mmap : 1, /* include mmap data */
  367. comm : 1, /* include comm data */
  368. freq : 1, /* use freq, not period */
  369. inherit_stat : 1, /* per task counts */
  370. enable_on_exec : 1, /* next exec enables */
  371. task : 1, /* trace fork/exit */
  372. watermark : 1, /* wakeup_watermark */
  373. /*
  374. * precise_ip:
  375. *
  376. * 0 - SAMPLE_IP can have arbitrary skid
  377. * 1 - SAMPLE_IP must have constant skid
  378. * 2 - SAMPLE_IP requested to have 0 skid
  379. * 3 - SAMPLE_IP must have 0 skid
  380. *
  381. * See also PERF_RECORD_MISC_EXACT_IP
  382. */
  383. precise_ip : 2, /* skid constraint */
  384. mmap_data : 1, /* non-exec mmap data */
  385. sample_id_all : 1, /* sample_type all events */
  386. exclude_host : 1, /* don't count in host */
  387. exclude_guest : 1, /* don't count in guest */
  388. exclude_callchain_kernel : 1, /* exclude kernel callchains */
  389. exclude_callchain_user : 1, /* exclude user callchains */
  390. mmap2 : 1, /* include mmap with inode data */
  391. comm_exec : 1, /* flag comm events that are due to an exec */
  392. use_clockid : 1, /* use @clockid for time fields */
  393. context_switch : 1, /* context switch data */
  394. write_backward : 1, /* Write ring buffer from end to beginning */
  395. namespaces : 1, /* include namespaces data */
  396. ksymbol : 1, /* include ksymbol events */
  397. bpf_event : 1, /* include bpf events */
  398. aux_output : 1, /* generate AUX records instead of events */
  399. cgroup : 1, /* include cgroup events */
  400. text_poke : 1, /* include text poke events */
  401. build_id : 1, /* use build id in mmap2 events */
  402. inherit_thread : 1, /* children only inherit if cloned with CLONE_THREAD */
  403. remove_on_exec : 1, /* event is removed from task on exec */
  404. sigtrap : 1, /* send synchronous SIGTRAP on event */
  405. __reserved_1 : 26;
  406. union {
  407. __u32 wakeup_events; /* wakeup every n events */
  408. __u32 wakeup_watermark; /* bytes before wakeup */
  409. };
  410. __u32 bp_type;
  411. union {
  412. __u64 bp_addr;
  413. __u64 kprobe_func; /* for perf_kprobe */
  414. __u64 uprobe_path; /* for perf_uprobe */
  415. __u64 config1; /* extension of config */
  416. };
  417. union {
  418. __u64 bp_len;
  419. __u64 kprobe_addr; /* when kprobe_func == NULL */
  420. __u64 probe_offset; /* for perf_[k,u]probe */
  421. __u64 config2; /* extension of config1 */
  422. };
  423. __u64 branch_sample_type; /* enum perf_branch_sample_type */
  424. /*
  425. * Defines set of user regs to dump on samples.
  426. * See asm/perf_regs.h for details.
  427. */
  428. __u64 sample_regs_user;
  429. /*
  430. * Defines size of the user stack to dump on samples.
  431. */
  432. __u32 sample_stack_user;
  433. __s32 clockid;
  434. /*
  435. * Defines set of regs to dump for each sample
  436. * state captured on:
  437. * - precise = 0: PMU interrupt
  438. * - precise > 0: sampled instruction
  439. *
  440. * See asm/perf_regs.h for details.
  441. */
  442. __u64 sample_regs_intr;
  443. /*
  444. * Wakeup watermark for AUX area
  445. */
  446. __u32 aux_watermark;
  447. __u16 sample_max_stack;
  448. __u16 __reserved_2;
  449. __u32 aux_sample_size;
  450. __u32 __reserved_3;
  451. /*
  452. * User provided data if sigtrap=1, passed back to user via
  453. * siginfo_t::si_perf_data, e.g. to permit user to identify the event.
  454. * Note, siginfo_t::si_perf_data is long-sized, and sig_data will be
  455. * truncated accordingly on 32 bit architectures.
  456. */
  457. __u64 sig_data;
  458. __u64 config3; /* extension of config2 */
  459. };
  460. /*
  461. * Structure used by below PERF_EVENT_IOC_QUERY_BPF command
  462. * to query bpf programs attached to the same perf tracepoint
  463. * as the given perf event.
  464. */
  465. struct perf_event_query_bpf {
  466. /*
  467. * The below ids array length
  468. */
  469. __u32 ids_len;
  470. /*
  471. * Set by the kernel to indicate the number of
  472. * available programs
  473. */
  474. __u32 prog_cnt;
  475. /*
  476. * User provided buffer to store program ids
  477. */
  478. __u32 ids[];
  479. };
  480. /*
  481. * Ioctls that can be done on a perf event fd:
  482. */
  483. #define PERF_EVENT_IOC_ENABLE _IO ('$', 0)
  484. #define PERF_EVENT_IOC_DISABLE _IO ('$', 1)
  485. #define PERF_EVENT_IOC_REFRESH _IO ('$', 2)
  486. #define PERF_EVENT_IOC_RESET _IO ('$', 3)
  487. #define PERF_EVENT_IOC_PERIOD _IOW('$', 4, __u64)
  488. #define PERF_EVENT_IOC_SET_OUTPUT _IO ('$', 5)
  489. #define PERF_EVENT_IOC_SET_FILTER _IOW('$', 6, char *)
  490. #define PERF_EVENT_IOC_ID _IOR('$', 7, __u64 *)
  491. #define PERF_EVENT_IOC_SET_BPF _IOW('$', 8, __u32)
  492. #define PERF_EVENT_IOC_PAUSE_OUTPUT _IOW('$', 9, __u32)
  493. #define PERF_EVENT_IOC_QUERY_BPF _IOWR('$', 10, struct perf_event_query_bpf *)
  494. #define PERF_EVENT_IOC_MODIFY_ATTRIBUTES _IOW('$', 11, struct perf_event_attr *)
  495. enum perf_event_ioc_flags {
  496. PERF_IOC_FLAG_GROUP = 1U << 0,
  497. };
  498. /*
  499. * Structure of the page that can be mapped via mmap
  500. */
  501. struct perf_event_mmap_page {
  502. __u32 version; /* version number of this structure */
  503. __u32 compat_version; /* lowest version this is compat with */
  504. /*
  505. * Bits needed to read the hw events in user-space.
  506. *
  507. * u32 seq, time_mult, time_shift, index, width;
  508. * u64 count, enabled, running;
  509. * u64 cyc, time_offset;
  510. * s64 pmc = 0;
  511. *
  512. * do {
  513. * seq = pc->lock;
  514. * barrier()
  515. *
  516. * enabled = pc->time_enabled;
  517. * running = pc->time_running;
  518. *
  519. * if (pc->cap_usr_time && enabled != running) {
  520. * cyc = rdtsc();
  521. * time_offset = pc->time_offset;
  522. * time_mult = pc->time_mult;
  523. * time_shift = pc->time_shift;
  524. * }
  525. *
  526. * index = pc->index;
  527. * count = pc->offset;
  528. * if (pc->cap_user_rdpmc && index) {
  529. * width = pc->pmc_width;
  530. * pmc = rdpmc(index - 1);
  531. * }
  532. *
  533. * barrier();
  534. * } while (pc->lock != seq);
  535. *
  536. * NOTE: for obvious reason this only works on self-monitoring
  537. * processes.
  538. */
  539. __u32 lock; /* seqlock for synchronization */
  540. __u32 index; /* hardware event identifier */
  541. __s64 offset; /* add to hardware event value */
  542. __u64 time_enabled; /* time event active */
  543. __u64 time_running; /* time event on cpu */
  544. union {
  545. __u64 capabilities;
  546. struct {
  547. __u64 cap_bit0 : 1, /* Always 0, deprecated, see commit 860f085b74e9 */
  548. cap_bit0_is_deprecated : 1, /* Always 1, signals that bit 0 is zero */
  549. cap_user_rdpmc : 1, /* The RDPMC instruction can be used to read counts */
  550. cap_user_time : 1, /* The time_{shift,mult,offset} fields are used */
  551. cap_user_time_zero : 1, /* The time_zero field is used */
  552. cap_user_time_short : 1, /* the time_{cycle,mask} fields are used */
  553. cap_____res : 58;
  554. };
  555. };
  556. /*
  557. * If cap_user_rdpmc this field provides the bit-width of the value
  558. * read using the rdpmc() or equivalent instruction. This can be used
  559. * to sign extend the result like:
  560. *
  561. * pmc <<= 64 - width;
  562. * pmc >>= 64 - width; // signed shift right
  563. * count += pmc;
  564. */
  565. __u16 pmc_width;
  566. /*
  567. * If cap_usr_time the below fields can be used to compute the time
  568. * delta since time_enabled (in ns) using rdtsc or similar.
  569. *
  570. * u64 quot, rem;
  571. * u64 delta;
  572. *
  573. * quot = (cyc >> time_shift);
  574. * rem = cyc & (((u64)1 << time_shift) - 1);
  575. * delta = time_offset + quot * time_mult +
  576. * ((rem * time_mult) >> time_shift);
  577. *
  578. * Where time_offset,time_mult,time_shift and cyc are read in the
  579. * seqcount loop described above. This delta can then be added to
  580. * enabled and possible running (if index), improving the scaling:
  581. *
  582. * enabled += delta;
  583. * if (index)
  584. * running += delta;
  585. *
  586. * quot = count / running;
  587. * rem = count % running;
  588. * count = quot * enabled + (rem * enabled) / running;
  589. */
  590. __u16 time_shift;
  591. __u32 time_mult;
  592. __u64 time_offset;
  593. /*
  594. * If cap_usr_time_zero, the hardware clock (e.g. TSC) can be calculated
  595. * from sample timestamps.
  596. *
  597. * time = timestamp - time_zero;
  598. * quot = time / time_mult;
  599. * rem = time % time_mult;
  600. * cyc = (quot << time_shift) + (rem << time_shift) / time_mult;
  601. *
  602. * And vice versa:
  603. *
  604. * quot = cyc >> time_shift;
  605. * rem = cyc & (((u64)1 << time_shift) - 1);
  606. * timestamp = time_zero + quot * time_mult +
  607. * ((rem * time_mult) >> time_shift);
  608. */
  609. __u64 time_zero;
  610. __u32 size; /* Header size up to __reserved[] fields. */
  611. __u32 __reserved_1;
  612. /*
  613. * If cap_usr_time_short, the hardware clock is less than 64bit wide
  614. * and we must compute the 'cyc' value, as used by cap_usr_time, as:
  615. *
  616. * cyc = time_cycles + ((cyc - time_cycles) & time_mask)
  617. *
  618. * NOTE: this form is explicitly chosen such that cap_usr_time_short
  619. * is a correction on top of cap_usr_time, and code that doesn't
  620. * know about cap_usr_time_short still works under the assumption
  621. * the counter doesn't wrap.
  622. */
  623. __u64 time_cycles;
  624. __u64 time_mask;
  625. /*
  626. * Hole for extension of the self monitor capabilities
  627. */
  628. __u8 __reserved[116*8]; /* align to 1k. */
  629. /*
  630. * Control data for the mmap() data buffer.
  631. *
  632. * User-space reading the @data_head value should issue an smp_rmb(),
  633. * after reading this value.
  634. *
  635. * When the mapping is PROT_WRITE the @data_tail value should be
  636. * written by userspace to reflect the last read data, after issueing
  637. * an smp_mb() to separate the data read from the ->data_tail store.
  638. * In this case the kernel will not over-write unread data.
  639. *
  640. * See perf_output_put_handle() for the data ordering.
  641. *
  642. * data_{offset,size} indicate the location and size of the perf record
  643. * buffer within the mmapped area.
  644. */
  645. __u64 data_head; /* head in the data section */
  646. __u64 data_tail; /* user-space written tail */
  647. __u64 data_offset; /* where the buffer starts */
  648. __u64 data_size; /* data buffer size */
  649. /*
  650. * AUX area is defined by aux_{offset,size} fields that should be set
  651. * by the userspace, so that
  652. *
  653. * aux_offset >= data_offset + data_size
  654. *
  655. * prior to mmap()ing it. Size of the mmap()ed area should be aux_size.
  656. *
  657. * Ring buffer pointers aux_{head,tail} have the same semantics as
  658. * data_{head,tail} and same ordering rules apply.
  659. */
  660. __u64 aux_head;
  661. __u64 aux_tail;
  662. __u64 aux_offset;
  663. __u64 aux_size;
  664. };
  665. /*
  666. * The current state of perf_event_header::misc bits usage:
  667. * ('|' used bit, '-' unused bit)
  668. *
  669. * 012 CDEF
  670. * |||---------||||
  671. *
  672. * Where:
  673. * 0-2 CPUMODE_MASK
  674. *
  675. * C PROC_MAP_PARSE_TIMEOUT
  676. * D MMAP_DATA / COMM_EXEC / FORK_EXEC / SWITCH_OUT
  677. * E MMAP_BUILD_ID / EXACT_IP / SCHED_OUT_PREEMPT
  678. * F (reserved)
  679. */
  680. #define PERF_RECORD_MISC_CPUMODE_MASK (7 << 0)
  681. #define PERF_RECORD_MISC_CPUMODE_UNKNOWN (0 << 0)
  682. #define PERF_RECORD_MISC_KERNEL (1 << 0)
  683. #define PERF_RECORD_MISC_USER (2 << 0)
  684. #define PERF_RECORD_MISC_HYPERVISOR (3 << 0)
  685. #define PERF_RECORD_MISC_GUEST_KERNEL (4 << 0)
  686. #define PERF_RECORD_MISC_GUEST_USER (5 << 0)
  687. /*
  688. * Indicates that /proc/PID/maps parsing are truncated by time out.
  689. */
  690. #define PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT (1 << 12)
  691. /*
  692. * Following PERF_RECORD_MISC_* are used on different
  693. * events, so can reuse the same bit position:
  694. *
  695. * PERF_RECORD_MISC_MMAP_DATA - PERF_RECORD_MMAP* events
  696. * PERF_RECORD_MISC_COMM_EXEC - PERF_RECORD_COMM event
  697. * PERF_RECORD_MISC_FORK_EXEC - PERF_RECORD_FORK event (perf internal)
  698. * PERF_RECORD_MISC_SWITCH_OUT - PERF_RECORD_SWITCH* events
  699. */
  700. #define PERF_RECORD_MISC_MMAP_DATA (1 << 13)
  701. #define PERF_RECORD_MISC_COMM_EXEC (1 << 13)
  702. #define PERF_RECORD_MISC_FORK_EXEC (1 << 13)
  703. #define PERF_RECORD_MISC_SWITCH_OUT (1 << 13)
  704. /*
  705. * These PERF_RECORD_MISC_* flags below are safely reused
  706. * for the following events:
  707. *
  708. * PERF_RECORD_MISC_EXACT_IP - PERF_RECORD_SAMPLE of precise events
  709. * PERF_RECORD_MISC_SWITCH_OUT_PREEMPT - PERF_RECORD_SWITCH* events
  710. * PERF_RECORD_MISC_MMAP_BUILD_ID - PERF_RECORD_MMAP2 event
  711. *
  712. *
  713. * PERF_RECORD_MISC_EXACT_IP:
  714. * Indicates that the content of PERF_SAMPLE_IP points to
  715. * the actual instruction that triggered the event. See also
  716. * perf_event_attr::precise_ip.
  717. *
  718. * PERF_RECORD_MISC_SWITCH_OUT_PREEMPT:
  719. * Indicates that thread was preempted in TASK_RUNNING state.
  720. *
  721. * PERF_RECORD_MISC_MMAP_BUILD_ID:
  722. * Indicates that mmap2 event carries build id data.
  723. */
  724. #define PERF_RECORD_MISC_EXACT_IP (1 << 14)
  725. #define PERF_RECORD_MISC_SWITCH_OUT_PREEMPT (1 << 14)
  726. #define PERF_RECORD_MISC_MMAP_BUILD_ID (1 << 14)
  727. /*
  728. * Reserve the last bit to indicate some extended misc field
  729. */
  730. #define PERF_RECORD_MISC_EXT_RESERVED (1 << 15)
  731. struct perf_event_header {
  732. __u32 type;
  733. __u16 misc;
  734. __u16 size;
  735. };
  736. struct perf_ns_link_info {
  737. __u64 dev;
  738. __u64 ino;
  739. };
  740. enum {
  741. NET_NS_INDEX = 0,
  742. UTS_NS_INDEX = 1,
  743. IPC_NS_INDEX = 2,
  744. PID_NS_INDEX = 3,
  745. USER_NS_INDEX = 4,
  746. MNT_NS_INDEX = 5,
  747. CGROUP_NS_INDEX = 6,
  748. NR_NAMESPACES, /* number of available namespaces */
  749. };
  750. enum perf_event_type {
  751. /*
  752. * If perf_event_attr.sample_id_all is set then all event types will
  753. * have the sample_type selected fields related to where/when
  754. * (identity) an event took place (TID, TIME, ID, STREAM_ID, CPU,
  755. * IDENTIFIER) described in PERF_RECORD_SAMPLE below, it will be stashed
  756. * just after the perf_event_header and the fields already present for
  757. * the existing fields, i.e. at the end of the payload. That way a newer
  758. * perf.data file will be supported by older perf tools, with these new
  759. * optional fields being ignored.
  760. *
  761. * struct sample_id {
  762. * { u32 pid, tid; } && PERF_SAMPLE_TID
  763. * { u64 time; } && PERF_SAMPLE_TIME
  764. * { u64 id; } && PERF_SAMPLE_ID
  765. * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID
  766. * { u32 cpu, res; } && PERF_SAMPLE_CPU
  767. * { u64 id; } && PERF_SAMPLE_IDENTIFIER
  768. * } && perf_event_attr::sample_id_all
  769. *
  770. * Note that PERF_SAMPLE_IDENTIFIER duplicates PERF_SAMPLE_ID. The
  771. * advantage of PERF_SAMPLE_IDENTIFIER is that its position is fixed
  772. * relative to header.size.
  773. */
  774. /*
  775. * The MMAP events record the PROT_EXEC mappings so that we can
  776. * correlate userspace IPs to code. They have the following structure:
  777. *
  778. * struct {
  779. * struct perf_event_header header;
  780. *
  781. * u32 pid, tid;
  782. * u64 addr;
  783. * u64 len;
  784. * u64 pgoff;
  785. * char filename[];
  786. * struct sample_id sample_id;
  787. * };
  788. */
  789. PERF_RECORD_MMAP = 1,
  790. /*
  791. * struct {
  792. * struct perf_event_header header;
  793. * u64 id;
  794. * u64 lost;
  795. * struct sample_id sample_id;
  796. * };
  797. */
  798. PERF_RECORD_LOST = 2,
  799. /*
  800. * struct {
  801. * struct perf_event_header header;
  802. *
  803. * u32 pid, tid;
  804. * char comm[];
  805. * struct sample_id sample_id;
  806. * };
  807. */
  808. PERF_RECORD_COMM = 3,
  809. /*
  810. * struct {
  811. * struct perf_event_header header;
  812. * u32 pid, ppid;
  813. * u32 tid, ptid;
  814. * u64 time;
  815. * struct sample_id sample_id;
  816. * };
  817. */
  818. PERF_RECORD_EXIT = 4,
  819. /*
  820. * struct {
  821. * struct perf_event_header header;
  822. * u64 time;
  823. * u64 id;
  824. * u64 stream_id;
  825. * struct sample_id sample_id;
  826. * };
  827. */
  828. PERF_RECORD_THROTTLE = 5,
  829. PERF_RECORD_UNTHROTTLE = 6,
  830. /*
  831. * struct {
  832. * struct perf_event_header header;
  833. * u32 pid, ppid;
  834. * u32 tid, ptid;
  835. * u64 time;
  836. * struct sample_id sample_id;
  837. * };
  838. */
  839. PERF_RECORD_FORK = 7,
  840. /*
  841. * struct {
  842. * struct perf_event_header header;
  843. * u32 pid, tid;
  844. *
  845. * struct read_format values;
  846. * struct sample_id sample_id;
  847. * };
  848. */
  849. PERF_RECORD_READ = 8,
  850. /*
  851. * struct {
  852. * struct perf_event_header header;
  853. *
  854. * #
  855. * # Note that PERF_SAMPLE_IDENTIFIER duplicates PERF_SAMPLE_ID.
  856. * # The advantage of PERF_SAMPLE_IDENTIFIER is that its position
  857. * # is fixed relative to header.
  858. * #
  859. *
  860. * { u64 id; } && PERF_SAMPLE_IDENTIFIER
  861. * { u64 ip; } && PERF_SAMPLE_IP
  862. * { u32 pid, tid; } && PERF_SAMPLE_TID
  863. * { u64 time; } && PERF_SAMPLE_TIME
  864. * { u64 addr; } && PERF_SAMPLE_ADDR
  865. * { u64 id; } && PERF_SAMPLE_ID
  866. * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID
  867. * { u32 cpu, res; } && PERF_SAMPLE_CPU
  868. * { u64 period; } && PERF_SAMPLE_PERIOD
  869. *
  870. * { struct read_format values; } && PERF_SAMPLE_READ
  871. *
  872. * { u64 nr,
  873. * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
  874. *
  875. * #
  876. * # The RAW record below is opaque data wrt the ABI
  877. * #
  878. * # That is, the ABI doesn't make any promises wrt to
  879. * # the stability of its content, it may vary depending
  880. * # on event, hardware, kernel version and phase of
  881. * # the moon.
  882. * #
  883. * # In other words, PERF_SAMPLE_RAW contents are not an ABI.
  884. * #
  885. *
  886. * { u32 size;
  887. * char data[size];}&& PERF_SAMPLE_RAW
  888. *
  889. * { u64 nr;
  890. * { u64 hw_idx; } && PERF_SAMPLE_BRANCH_HW_INDEX
  891. * { u64 from, to, flags } lbr[nr];
  892. * #
  893. * # The format of the counters is decided by the
  894. * # "branch_counter_nr" and "branch_counter_width",
  895. * # which are defined in the ABI.
  896. * #
  897. * { u64 counters; } cntr[nr] && PERF_SAMPLE_BRANCH_COUNTERS
  898. * } && PERF_SAMPLE_BRANCH_STACK
  899. *
  900. * { u64 abi; # enum perf_sample_regs_abi
  901. * u64 regs[weight(mask)]; } && PERF_SAMPLE_REGS_USER
  902. *
  903. * { u64 size;
  904. * char data[size];
  905. * u64 dyn_size; } && PERF_SAMPLE_STACK_USER
  906. *
  907. * { union perf_sample_weight
  908. * {
  909. * u64 full; && PERF_SAMPLE_WEIGHT
  910. * #if defined(__LITTLE_ENDIAN_BITFIELD)
  911. * struct {
  912. * u32 var1_dw;
  913. * u16 var2_w;
  914. * u16 var3_w;
  915. * } && PERF_SAMPLE_WEIGHT_STRUCT
  916. * #elif defined(__BIG_ENDIAN_BITFIELD)
  917. * struct {
  918. * u16 var3_w;
  919. * u16 var2_w;
  920. * u32 var1_dw;
  921. * } && PERF_SAMPLE_WEIGHT_STRUCT
  922. * #endif
  923. * }
  924. * }
  925. * { u64 data_src; } && PERF_SAMPLE_DATA_SRC
  926. * { u64 transaction; } && PERF_SAMPLE_TRANSACTION
  927. * { u64 abi; # enum perf_sample_regs_abi
  928. * u64 regs[weight(mask)]; } && PERF_SAMPLE_REGS_INTR
  929. * { u64 phys_addr;} && PERF_SAMPLE_PHYS_ADDR
  930. * { u64 size;
  931. * char data[size]; } && PERF_SAMPLE_AUX
  932. * { u64 data_page_size;} && PERF_SAMPLE_DATA_PAGE_SIZE
  933. * { u64 code_page_size;} && PERF_SAMPLE_CODE_PAGE_SIZE
  934. * };
  935. */
  936. PERF_RECORD_SAMPLE = 9,
  937. /*
  938. * The MMAP2 records are an augmented version of MMAP, they add
  939. * maj, min, ino numbers to be used to uniquely identify each mapping
  940. *
  941. * struct {
  942. * struct perf_event_header header;
  943. *
  944. * u32 pid, tid;
  945. * u64 addr;
  946. * u64 len;
  947. * u64 pgoff;
  948. * union {
  949. * struct {
  950. * u32 maj;
  951. * u32 min;
  952. * u64 ino;
  953. * u64 ino_generation;
  954. * };
  955. * struct {
  956. * u8 build_id_size;
  957. * u8 __reserved_1;
  958. * u16 __reserved_2;
  959. * u8 build_id[20];
  960. * };
  961. * };
  962. * u32 prot, flags;
  963. * char filename[];
  964. * struct sample_id sample_id;
  965. * };
  966. */
  967. PERF_RECORD_MMAP2 = 10,
  968. /*
  969. * Records that new data landed in the AUX buffer part.
  970. *
  971. * struct {
  972. * struct perf_event_header header;
  973. *
  974. * u64 aux_offset;
  975. * u64 aux_size;
  976. * u64 flags;
  977. * struct sample_id sample_id;
  978. * };
  979. */
  980. PERF_RECORD_AUX = 11,
  981. /*
  982. * Indicates that instruction trace has started
  983. *
  984. * struct {
  985. * struct perf_event_header header;
  986. * u32 pid;
  987. * u32 tid;
  988. * struct sample_id sample_id;
  989. * };
  990. */
  991. PERF_RECORD_ITRACE_START = 12,
  992. /*
  993. * Records the dropped/lost sample number.
  994. *
  995. * struct {
  996. * struct perf_event_header header;
  997. *
  998. * u64 lost;
  999. * struct sample_id sample_id;
  1000. * };
  1001. */
  1002. PERF_RECORD_LOST_SAMPLES = 13,
  1003. /*
  1004. * Records a context switch in or out (flagged by
  1005. * PERF_RECORD_MISC_SWITCH_OUT). See also
  1006. * PERF_RECORD_SWITCH_CPU_WIDE.
  1007. *
  1008. * struct {
  1009. * struct perf_event_header header;
  1010. * struct sample_id sample_id;
  1011. * };
  1012. */
  1013. PERF_RECORD_SWITCH = 14,
  1014. /*
  1015. * CPU-wide version of PERF_RECORD_SWITCH with next_prev_pid and
  1016. * next_prev_tid that are the next (switching out) or previous
  1017. * (switching in) pid/tid.
  1018. *
  1019. * struct {
  1020. * struct perf_event_header header;
  1021. * u32 next_prev_pid;
  1022. * u32 next_prev_tid;
  1023. * struct sample_id sample_id;
  1024. * };
  1025. */
  1026. PERF_RECORD_SWITCH_CPU_WIDE = 15,
  1027. /*
  1028. * struct {
  1029. * struct perf_event_header header;
  1030. * u32 pid;
  1031. * u32 tid;
  1032. * u64 nr_namespaces;
  1033. * { u64 dev, inode; } [nr_namespaces];
  1034. * struct sample_id sample_id;
  1035. * };
  1036. */
  1037. PERF_RECORD_NAMESPACES = 16,
  1038. /*
  1039. * Record ksymbol register/unregister events:
  1040. *
  1041. * struct {
  1042. * struct perf_event_header header;
  1043. * u64 addr;
  1044. * u32 len;
  1045. * u16 ksym_type;
  1046. * u16 flags;
  1047. * char name[];
  1048. * struct sample_id sample_id;
  1049. * };
  1050. */
  1051. PERF_RECORD_KSYMBOL = 17,
  1052. /*
  1053. * Record bpf events:
  1054. * enum perf_bpf_event_type {
  1055. * PERF_BPF_EVENT_UNKNOWN = 0,
  1056. * PERF_BPF_EVENT_PROG_LOAD = 1,
  1057. * PERF_BPF_EVENT_PROG_UNLOAD = 2,
  1058. * };
  1059. *
  1060. * struct {
  1061. * struct perf_event_header header;
  1062. * u16 type;
  1063. * u16 flags;
  1064. * u32 id;
  1065. * u8 tag[BPF_TAG_SIZE];
  1066. * struct sample_id sample_id;
  1067. * };
  1068. */
  1069. PERF_RECORD_BPF_EVENT = 18,
  1070. /*
  1071. * struct {
  1072. * struct perf_event_header header;
  1073. * u64 id;
  1074. * char path[];
  1075. * struct sample_id sample_id;
  1076. * };
  1077. */
  1078. PERF_RECORD_CGROUP = 19,
  1079. /*
  1080. * Records changes to kernel text i.e. self-modified code. 'old_len' is
  1081. * the number of old bytes, 'new_len' is the number of new bytes. Either
  1082. * 'old_len' or 'new_len' may be zero to indicate, for example, the
  1083. * addition or removal of a trampoline. 'bytes' contains the old bytes
  1084. * followed immediately by the new bytes.
  1085. *
  1086. * struct {
  1087. * struct perf_event_header header;
  1088. * u64 addr;
  1089. * u16 old_len;
  1090. * u16 new_len;
  1091. * u8 bytes[];
  1092. * struct sample_id sample_id;
  1093. * };
  1094. */
  1095. PERF_RECORD_TEXT_POKE = 20,
  1096. /*
  1097. * Data written to the AUX area by hardware due to aux_output, may need
  1098. * to be matched to the event by an architecture-specific hardware ID.
  1099. * This records the hardware ID, but requires sample_id to provide the
  1100. * event ID. e.g. Intel PT uses this record to disambiguate PEBS-via-PT
  1101. * records from multiple events.
  1102. *
  1103. * struct {
  1104. * struct perf_event_header header;
  1105. * u64 hw_id;
  1106. * struct sample_id sample_id;
  1107. * };
  1108. */
  1109. PERF_RECORD_AUX_OUTPUT_HW_ID = 21,
  1110. PERF_RECORD_MAX, /* non-ABI */
  1111. };
  1112. enum perf_record_ksymbol_type {
  1113. PERF_RECORD_KSYMBOL_TYPE_UNKNOWN = 0,
  1114. PERF_RECORD_KSYMBOL_TYPE_BPF = 1,
  1115. /*
  1116. * Out of line code such as kprobe-replaced instructions or optimized
  1117. * kprobes or ftrace trampolines.
  1118. */
  1119. PERF_RECORD_KSYMBOL_TYPE_OOL = 2,
  1120. PERF_RECORD_KSYMBOL_TYPE_MAX /* non-ABI */
  1121. };
  1122. #define PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER (1 << 0)
  1123. enum perf_bpf_event_type {
  1124. PERF_BPF_EVENT_UNKNOWN = 0,
  1125. PERF_BPF_EVENT_PROG_LOAD = 1,
  1126. PERF_BPF_EVENT_PROG_UNLOAD = 2,
  1127. PERF_BPF_EVENT_MAX, /* non-ABI */
  1128. };
  1129. #define PERF_MAX_STACK_DEPTH 127
  1130. #define PERF_MAX_CONTEXTS_PER_STACK 8
  1131. enum perf_callchain_context {
  1132. PERF_CONTEXT_HV = (__u64)-32,
  1133. PERF_CONTEXT_KERNEL = (__u64)-128,
  1134. PERF_CONTEXT_USER = (__u64)-512,
  1135. PERF_CONTEXT_GUEST = (__u64)-2048,
  1136. PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176,
  1137. PERF_CONTEXT_GUEST_USER = (__u64)-2560,
  1138. PERF_CONTEXT_MAX = (__u64)-4095,
  1139. };
  1140. /**
  1141. * PERF_RECORD_AUX::flags bits
  1142. */
  1143. #define PERF_AUX_FLAG_TRUNCATED 0x01 /* record was truncated to fit */
  1144. #define PERF_AUX_FLAG_OVERWRITE 0x02 /* snapshot from overwrite mode */
  1145. #define PERF_AUX_FLAG_PARTIAL 0x04 /* record contains gaps */
  1146. #define PERF_AUX_FLAG_COLLISION 0x08 /* sample collided with another */
  1147. #define PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK 0xff00 /* PMU specific trace format type */
  1148. /* CoreSight PMU AUX buffer formats */
  1149. #define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT 0x0000 /* Default for backward compatibility */
  1150. #define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW 0x0100 /* Raw format of the source */
  1151. #define PERF_FLAG_FD_NO_GROUP (1UL << 0)
  1152. #define PERF_FLAG_FD_OUTPUT (1UL << 1)
  1153. #define PERF_FLAG_PID_CGROUP (1UL << 2) /* pid=cgroup id, per-cpu mode only */
  1154. #define PERF_FLAG_FD_CLOEXEC (1UL << 3) /* O_CLOEXEC */
  1155. #if defined(__LITTLE_ENDIAN_BITFIELD)
  1156. union perf_mem_data_src {
  1157. __u64 val;
  1158. struct {
  1159. __u64 mem_op:5, /* type of opcode */
  1160. mem_lvl:14, /* memory hierarchy level */
  1161. mem_snoop:5, /* snoop mode */
  1162. mem_lock:2, /* lock instr */
  1163. mem_dtlb:7, /* tlb access */
  1164. mem_lvl_num:4, /* memory hierarchy level number */
  1165. mem_remote:1, /* remote */
  1166. mem_snoopx:2, /* snoop mode, ext */
  1167. mem_blk:3, /* access blocked */
  1168. mem_hops:3, /* hop level */
  1169. mem_rsvd:18;
  1170. };
  1171. };
  1172. #elif defined(__BIG_ENDIAN_BITFIELD)
  1173. union perf_mem_data_src {
  1174. __u64 val;
  1175. struct {
  1176. __u64 mem_rsvd:18,
  1177. mem_hops:3, /* hop level */
  1178. mem_blk:3, /* access blocked */
  1179. mem_snoopx:2, /* snoop mode, ext */
  1180. mem_remote:1, /* remote */
  1181. mem_lvl_num:4, /* memory hierarchy level number */
  1182. mem_dtlb:7, /* tlb access */
  1183. mem_lock:2, /* lock instr */
  1184. mem_snoop:5, /* snoop mode */
  1185. mem_lvl:14, /* memory hierarchy level */
  1186. mem_op:5; /* type of opcode */
  1187. };
  1188. };
  1189. #else
  1190. #error "Unknown endianness"
  1191. #endif
  1192. /* type of opcode (load/store/prefetch,code) */
  1193. #define PERF_MEM_OP_NA 0x01 /* not available */
  1194. #define PERF_MEM_OP_LOAD 0x02 /* load instruction */
  1195. #define PERF_MEM_OP_STORE 0x04 /* store instruction */
  1196. #define PERF_MEM_OP_PFETCH 0x08 /* prefetch */
  1197. #define PERF_MEM_OP_EXEC 0x10 /* code (execution) */
  1198. #define PERF_MEM_OP_SHIFT 0
  1199. /*
  1200. * PERF_MEM_LVL_* namespace being depricated to some extent in the
  1201. * favour of newer composite PERF_MEM_{LVLNUM_,REMOTE_,SNOOPX_} fields.
  1202. * Supporting this namespace inorder to not break defined ABIs.
  1203. *
  1204. * memory hierarchy (memory level, hit or miss)
  1205. */
  1206. #define PERF_MEM_LVL_NA 0x01 /* not available */
  1207. #define PERF_MEM_LVL_HIT 0x02 /* hit level */
  1208. #define PERF_MEM_LVL_MISS 0x04 /* miss level */
  1209. #define PERF_MEM_LVL_L1 0x08 /* L1 */
  1210. #define PERF_MEM_LVL_LFB 0x10 /* Line Fill Buffer */
  1211. #define PERF_MEM_LVL_L2 0x20 /* L2 */
  1212. #define PERF_MEM_LVL_L3 0x40 /* L3 */
  1213. #define PERF_MEM_LVL_LOC_RAM 0x80 /* Local DRAM */
  1214. #define PERF_MEM_LVL_REM_RAM1 0x100 /* Remote DRAM (1 hop) */
  1215. #define PERF_MEM_LVL_REM_RAM2 0x200 /* Remote DRAM (2 hops) */
  1216. #define PERF_MEM_LVL_REM_CCE1 0x400 /* Remote Cache (1 hop) */
  1217. #define PERF_MEM_LVL_REM_CCE2 0x800 /* Remote Cache (2 hops) */
  1218. #define PERF_MEM_LVL_IO 0x1000 /* I/O memory */
  1219. #define PERF_MEM_LVL_UNC 0x2000 /* Uncached memory */
  1220. #define PERF_MEM_LVL_SHIFT 5
  1221. #define PERF_MEM_REMOTE_REMOTE 0x01 /* Remote */
  1222. #define PERF_MEM_REMOTE_SHIFT 37
  1223. #define PERF_MEM_LVLNUM_L1 0x01 /* L1 */
  1224. #define PERF_MEM_LVLNUM_L2 0x02 /* L2 */
  1225. #define PERF_MEM_LVLNUM_L3 0x03 /* L3 */
  1226. #define PERF_MEM_LVLNUM_L4 0x04 /* L4 */
  1227. #define PERF_MEM_LVLNUM_L2_MHB 0x05 /* L2 Miss Handling Buffer */
  1228. #define PERF_MEM_LVLNUM_MSC 0x06 /* Memory-side Cache */
  1229. /* 0x7 available */
  1230. #define PERF_MEM_LVLNUM_UNC 0x08 /* Uncached */
  1231. #define PERF_MEM_LVLNUM_CXL 0x09 /* CXL */
  1232. #define PERF_MEM_LVLNUM_IO 0x0a /* I/O */
  1233. #define PERF_MEM_LVLNUM_ANY_CACHE 0x0b /* Any cache */
  1234. #define PERF_MEM_LVLNUM_LFB 0x0c /* LFB / L1 Miss Handling Buffer */
  1235. #define PERF_MEM_LVLNUM_RAM 0x0d /* RAM */
  1236. #define PERF_MEM_LVLNUM_PMEM 0x0e /* PMEM */
  1237. #define PERF_MEM_LVLNUM_NA 0x0f /* N/A */
  1238. #define PERF_MEM_LVLNUM_SHIFT 33
  1239. /* snoop mode */
  1240. #define PERF_MEM_SNOOP_NA 0x01 /* not available */
  1241. #define PERF_MEM_SNOOP_NONE 0x02 /* no snoop */
  1242. #define PERF_MEM_SNOOP_HIT 0x04 /* snoop hit */
  1243. #define PERF_MEM_SNOOP_MISS 0x08 /* snoop miss */
  1244. #define PERF_MEM_SNOOP_HITM 0x10 /* snoop hit modified */
  1245. #define PERF_MEM_SNOOP_SHIFT 19
  1246. #define PERF_MEM_SNOOPX_FWD 0x01 /* forward */
  1247. #define PERF_MEM_SNOOPX_PEER 0x02 /* xfer from peer */
  1248. #define PERF_MEM_SNOOPX_SHIFT 38
  1249. /* locked instruction */
  1250. #define PERF_MEM_LOCK_NA 0x01 /* not available */
  1251. #define PERF_MEM_LOCK_LOCKED 0x02 /* locked transaction */
  1252. #define PERF_MEM_LOCK_SHIFT 24
  1253. /* TLB access */
  1254. #define PERF_MEM_TLB_NA 0x01 /* not available */
  1255. #define PERF_MEM_TLB_HIT 0x02 /* hit level */
  1256. #define PERF_MEM_TLB_MISS 0x04 /* miss level */
  1257. #define PERF_MEM_TLB_L1 0x08 /* L1 */
  1258. #define PERF_MEM_TLB_L2 0x10 /* L2 */
  1259. #define PERF_MEM_TLB_WK 0x20 /* Hardware Walker*/
  1260. #define PERF_MEM_TLB_OS 0x40 /* OS fault handler */
  1261. #define PERF_MEM_TLB_SHIFT 26
  1262. /* Access blocked */
  1263. #define PERF_MEM_BLK_NA 0x01 /* not available */
  1264. #define PERF_MEM_BLK_DATA 0x02 /* data could not be forwarded */
  1265. #define PERF_MEM_BLK_ADDR 0x04 /* address conflict */
  1266. #define PERF_MEM_BLK_SHIFT 40
  1267. /* hop level */
  1268. #define PERF_MEM_HOPS_0 0x01 /* remote core, same node */
  1269. #define PERF_MEM_HOPS_1 0x02 /* remote node, same socket */
  1270. #define PERF_MEM_HOPS_2 0x03 /* remote socket, same board */
  1271. #define PERF_MEM_HOPS_3 0x04 /* remote board */
  1272. /* 5-7 available */
  1273. #define PERF_MEM_HOPS_SHIFT 43
  1274. #define PERF_MEM_S(a, s) \
  1275. (((__u64)PERF_MEM_##a##_##s) << PERF_MEM_##a##_SHIFT)
  1276. /*
  1277. * single taken branch record layout:
  1278. *
  1279. * from: source instruction (may not always be a branch insn)
  1280. * to: branch target
  1281. * mispred: branch target was mispredicted
  1282. * predicted: branch target was predicted
  1283. *
  1284. * support for mispred, predicted is optional. In case it
  1285. * is not supported mispred = predicted = 0.
  1286. *
  1287. * in_tx: running in a hardware transaction
  1288. * abort: aborting a hardware transaction
  1289. * cycles: cycles from last branch (or 0 if not supported)
  1290. * type: branch type
  1291. * spec: branch speculation info (or 0 if not supported)
  1292. */
  1293. struct perf_branch_entry {
  1294. __u64 from;
  1295. __u64 to;
  1296. __u64 mispred:1, /* target mispredicted */
  1297. predicted:1,/* target predicted */
  1298. in_tx:1, /* in transaction */
  1299. abort:1, /* transaction abort */
  1300. cycles:16, /* cycle count to last branch */
  1301. type:4, /* branch type */
  1302. spec:2, /* branch speculation info */
  1303. new_type:4, /* additional branch type */
  1304. priv:3, /* privilege level */
  1305. reserved:31;
  1306. };
  1307. /* Size of used info bits in struct perf_branch_entry */
  1308. #define PERF_BRANCH_ENTRY_INFO_BITS_MAX 33
  1309. union perf_sample_weight {
  1310. __u64 full;
  1311. #if defined(__LITTLE_ENDIAN_BITFIELD)
  1312. struct {
  1313. __u32 var1_dw;
  1314. __u16 var2_w;
  1315. __u16 var3_w;
  1316. };
  1317. #elif defined(__BIG_ENDIAN_BITFIELD)
  1318. struct {
  1319. __u16 var3_w;
  1320. __u16 var2_w;
  1321. __u32 var1_dw;
  1322. };
  1323. #else
  1324. #error "Unknown endianness"
  1325. #endif
  1326. };
  1327. #endif /* _LINUX_PERF_EVENT_H */