Commit | Line | Data |
---|---|---|
0793a61d | 1 | /* |
57c0c15b | 2 | * Performance events: |
0793a61d | 3 | * |
a308444c IM |
4 | * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de> |
5 | * Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar | |
6 | * Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra | |
0793a61d | 7 | * |
57c0c15b | 8 | * Data type definitions, declarations, prototypes. |
0793a61d | 9 | * |
a308444c | 10 | * Started by: Thomas Gleixner and Ingo Molnar |
0793a61d | 11 | * |
57c0c15b | 12 | * For licencing details see kernel-base/COPYING |
0793a61d | 13 | */ |
cdd6c482 IM |
14 | #ifndef _LINUX_PERF_EVENT_H |
15 | #define _LINUX_PERF_EVENT_H | |
0793a61d | 16 | |
f3dfd265 PM |
17 | #include <linux/types.h> |
18 | #include <linux/ioctl.h> | |
9aaa131a | 19 | #include <asm/byteorder.h> |
0793a61d TG |
20 | |
21 | /* | |
9f66a381 IM |
22 | * User-space ABI bits: |
23 | */ | |
24 | ||
25 | /* | |
0d48696f | 26 | * attr.type |
0793a61d | 27 | */ |
1c432d89 | 28 | enum perf_type_id { |
a308444c IM |
29 | PERF_TYPE_HARDWARE = 0, |
30 | PERF_TYPE_SOFTWARE = 1, | |
31 | PERF_TYPE_TRACEPOINT = 2, | |
32 | PERF_TYPE_HW_CACHE = 3, | |
33 | PERF_TYPE_RAW = 4, | |
24f1e32c | 34 | PERF_TYPE_BREAKPOINT = 5, |
b8e83514 | 35 | |
a308444c | 36 | PERF_TYPE_MAX, /* non-ABI */ |
b8e83514 | 37 | }; |
6c594c21 | 38 | |
b8e83514 | 39 | /* |
cdd6c482 IM |
40 | * Generalized performance event event_id types, used by the |
41 | * attr.event_id parameter of the sys_perf_event_open() | |
a308444c | 42 | * syscall: |
b8e83514 | 43 | */ |
1c432d89 | 44 | enum perf_hw_id { |
9f66a381 | 45 | /* |
b8e83514 | 46 | * Common hardware events, generalized by the kernel: |
9f66a381 | 47 | */ |
f4dbfa8f PZ |
48 | PERF_COUNT_HW_CPU_CYCLES = 0, |
49 | PERF_COUNT_HW_INSTRUCTIONS = 1, | |
50 | PERF_COUNT_HW_CACHE_REFERENCES = 2, | |
51 | PERF_COUNT_HW_CACHE_MISSES = 3, | |
52 | PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4, | |
53 | PERF_COUNT_HW_BRANCH_MISSES = 5, | |
54 | PERF_COUNT_HW_BUS_CYCLES = 6, | |
55 | ||
a308444c | 56 | PERF_COUNT_HW_MAX, /* non-ABI */ |
b8e83514 | 57 | }; |
e077df4f | 58 | |
8326f44d | 59 | /* |
cdd6c482 | 60 | * Generalized hardware cache events: |
8326f44d | 61 | * |
8be6e8f3 | 62 | * { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x |
8326f44d IM |
63 | * { read, write, prefetch } x |
64 | * { accesses, misses } | |
65 | */ | |
1c432d89 | 66 | enum perf_hw_cache_id { |
a308444c IM |
67 | PERF_COUNT_HW_CACHE_L1D = 0, |
68 | PERF_COUNT_HW_CACHE_L1I = 1, | |
69 | PERF_COUNT_HW_CACHE_LL = 2, | |
70 | PERF_COUNT_HW_CACHE_DTLB = 3, | |
71 | PERF_COUNT_HW_CACHE_ITLB = 4, | |
72 | PERF_COUNT_HW_CACHE_BPU = 5, | |
73 | ||
74 | PERF_COUNT_HW_CACHE_MAX, /* non-ABI */ | |
8326f44d IM |
75 | }; |
76 | ||
1c432d89 | 77 | enum perf_hw_cache_op_id { |
a308444c IM |
78 | PERF_COUNT_HW_CACHE_OP_READ = 0, |
79 | PERF_COUNT_HW_CACHE_OP_WRITE = 1, | |
80 | PERF_COUNT_HW_CACHE_OP_PREFETCH = 2, | |
8326f44d | 81 | |
a308444c | 82 | PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */ |
8326f44d IM |
83 | }; |
84 | ||
1c432d89 PZ |
85 | enum perf_hw_cache_op_result_id { |
86 | PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0, | |
87 | PERF_COUNT_HW_CACHE_RESULT_MISS = 1, | |
8326f44d | 88 | |
a308444c | 89 | PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */ |
8326f44d IM |
90 | }; |
91 | ||
b8e83514 | 92 | /* |
cdd6c482 IM |
93 | * Special "software" events provided by the kernel, even if the hardware |
94 | * does not support performance events. These events measure various | |
b8e83514 PZ |
95 | * physical and sw events of the kernel (and allow the profiling of them as |
96 | * well): | |
97 | */ | |
1c432d89 | 98 | enum perf_sw_ids { |
a308444c IM |
99 | PERF_COUNT_SW_CPU_CLOCK = 0, |
100 | PERF_COUNT_SW_TASK_CLOCK = 1, | |
101 | PERF_COUNT_SW_PAGE_FAULTS = 2, | |
102 | PERF_COUNT_SW_CONTEXT_SWITCHES = 3, | |
103 | PERF_COUNT_SW_CPU_MIGRATIONS = 4, | |
104 | PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, | |
105 | PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, | |
f7d79860 AB |
106 | PERF_COUNT_SW_ALIGNMENT_FAULTS = 7, |
107 | PERF_COUNT_SW_EMULATION_FAULTS = 8, | |
a308444c IM |
108 | |
109 | PERF_COUNT_SW_MAX, /* non-ABI */ | |
0793a61d TG |
110 | }; |
111 | ||
8a057d84 | 112 | /* |
0d48696f | 113 | * Bits that can be set in attr.sample_type to request information |
8a057d84 PZ |
114 | * in the overflow packets. |
115 | */ | |
cdd6c482 | 116 | enum perf_event_sample_format { |
a308444c IM |
117 | PERF_SAMPLE_IP = 1U << 0, |
118 | PERF_SAMPLE_TID = 1U << 1, | |
119 | PERF_SAMPLE_TIME = 1U << 2, | |
120 | PERF_SAMPLE_ADDR = 1U << 3, | |
3dab77fb | 121 | PERF_SAMPLE_READ = 1U << 4, |
a308444c IM |
122 | PERF_SAMPLE_CALLCHAIN = 1U << 5, |
123 | PERF_SAMPLE_ID = 1U << 6, | |
124 | PERF_SAMPLE_CPU = 1U << 7, | |
125 | PERF_SAMPLE_PERIOD = 1U << 8, | |
7f453c24 | 126 | PERF_SAMPLE_STREAM_ID = 1U << 9, |
3a43ce68 | 127 | PERF_SAMPLE_RAW = 1U << 10, |
974802ea | 128 | |
f413cdb8 | 129 | PERF_SAMPLE_MAX = 1U << 11, /* non-ABI */ |
8a057d84 PZ |
130 | }; |
131 | ||
53cfbf59 | 132 | /* |
cdd6c482 | 133 | * The format of the data returned by read() on a perf event fd, |
3dab77fb PZ |
134 | * as specified by attr.read_format: |
135 | * | |
136 | * struct read_format { | |
57c0c15b IM |
137 | * { u64 value; |
138 | * { u64 time_enabled; } && PERF_FORMAT_ENABLED | |
139 | * { u64 time_running; } && PERF_FORMAT_RUNNING | |
140 | * { u64 id; } && PERF_FORMAT_ID | |
141 | * } && !PERF_FORMAT_GROUP | |
3dab77fb | 142 | * |
57c0c15b IM |
143 | * { u64 nr; |
144 | * { u64 time_enabled; } && PERF_FORMAT_ENABLED | |
145 | * { u64 time_running; } && PERF_FORMAT_RUNNING | |
146 | * { u64 value; | |
147 | * { u64 id; } && PERF_FORMAT_ID | |
148 | * } cntr[nr]; | |
149 | * } && PERF_FORMAT_GROUP | |
3dab77fb | 150 | * }; |
53cfbf59 | 151 | */ |
cdd6c482 | 152 | enum perf_event_read_format { |
a308444c IM |
153 | PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0, |
154 | PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1, | |
155 | PERF_FORMAT_ID = 1U << 2, | |
3dab77fb | 156 | PERF_FORMAT_GROUP = 1U << 3, |
974802ea | 157 | |
57c0c15b | 158 | PERF_FORMAT_MAX = 1U << 4, /* non-ABI */ |
53cfbf59 PM |
159 | }; |
160 | ||
974802ea PZ |
161 | #define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */ |
162 | ||
9f66a381 | 163 | /* |
cdd6c482 | 164 | * Hardware event_id to monitor via a performance monitoring event: |
9f66a381 | 165 | */ |
cdd6c482 | 166 | struct perf_event_attr { |
974802ea | 167 | |
f4a2deb4 | 168 | /* |
a21ca2ca IM |
169 | * Major type: hardware/software/tracepoint/etc. |
170 | */ | |
171 | __u32 type; | |
974802ea PZ |
172 | |
173 | /* | |
174 | * Size of the attr structure, for fwd/bwd compat. | |
175 | */ | |
176 | __u32 size; | |
a21ca2ca IM |
177 | |
178 | /* | |
179 | * Type specific configuration information. | |
f4a2deb4 PZ |
180 | */ |
181 | __u64 config; | |
9f66a381 | 182 | |
60db5e09 | 183 | union { |
b23f3325 PZ |
184 | __u64 sample_period; |
185 | __u64 sample_freq; | |
60db5e09 PZ |
186 | }; |
187 | ||
b23f3325 PZ |
188 | __u64 sample_type; |
189 | __u64 read_format; | |
9f66a381 | 190 | |
2743a5b0 | 191 | __u64 disabled : 1, /* off by default */ |
0475f9ea PM |
192 | inherit : 1, /* children inherit it */ |
193 | pinned : 1, /* must always be on PMU */ | |
194 | exclusive : 1, /* only group on PMU */ | |
195 | exclude_user : 1, /* don't count user */ | |
196 | exclude_kernel : 1, /* ditto kernel */ | |
197 | exclude_hv : 1, /* ditto hypervisor */ | |
2743a5b0 | 198 | exclude_idle : 1, /* don't count when idle */ |
0a4a9391 | 199 | mmap : 1, /* include mmap data */ |
8d1b2d93 | 200 | comm : 1, /* include comm data */ |
60db5e09 | 201 | freq : 1, /* use freq, not period */ |
bfbd3381 | 202 | inherit_stat : 1, /* per task counts */ |
57e7986e | 203 | enable_on_exec : 1, /* next exec enables */ |
9f498cc5 | 204 | task : 1, /* trace fork/exit */ |
2667de81 | 205 | watermark : 1, /* wakeup_watermark */ |
ab608344 PZ |
206 | /* |
207 | * precise_ip: | |
208 | * | |
209 | * 0 - SAMPLE_IP can have arbitrary skid | |
210 | * 1 - SAMPLE_IP must have constant skid | |
211 | * 2 - SAMPLE_IP requested to have 0 skid | |
212 | * 3 - SAMPLE_IP must have 0 skid | |
213 | * | |
214 | * See also PERF_RECORD_MISC_EXACT_IP | |
215 | */ | |
216 | precise_ip : 2, /* skid constraint */ | |
3af9e859 | 217 | mmap_data : 1, /* non-exec mmap data */ |
c980d109 | 218 | sample_id_all : 1, /* sample_type all events */ |
ab608344 | 219 | |
c980d109 | 220 | __reserved_1 : 45; |
2743a5b0 | 221 | |
2667de81 PZ |
222 | union { |
223 | __u32 wakeup_events; /* wakeup every n events */ | |
224 | __u32 wakeup_watermark; /* bytes before wakeup */ | |
225 | }; | |
24f1e32c | 226 | |
f13c12c6 | 227 | __u32 bp_type; |
cd757645 MS |
228 | __u64 bp_addr; |
229 | __u64 bp_len; | |
eab656ae TG |
230 | }; |
231 | ||
d859e29f | 232 | /* |
cdd6c482 | 233 | * Ioctls that can be done on a perf event fd: |
d859e29f | 234 | */ |
cdd6c482 | 235 | #define PERF_EVENT_IOC_ENABLE _IO ('$', 0) |
57c0c15b IM |
236 | #define PERF_EVENT_IOC_DISABLE _IO ('$', 1) |
237 | #define PERF_EVENT_IOC_REFRESH _IO ('$', 2) | |
cdd6c482 | 238 | #define PERF_EVENT_IOC_RESET _IO ('$', 3) |
4c49b128 | 239 | #define PERF_EVENT_IOC_PERIOD _IOW('$', 4, __u64) |
cdd6c482 | 240 | #define PERF_EVENT_IOC_SET_OUTPUT _IO ('$', 5) |
6fb2915d | 241 | #define PERF_EVENT_IOC_SET_FILTER _IOW('$', 6, char *) |
cdd6c482 IM |
242 | |
243 | enum perf_event_ioc_flags { | |
3df5edad PZ |
244 | PERF_IOC_FLAG_GROUP = 1U << 0, |
245 | }; | |
d859e29f | 246 | |
37d81828 PM |
247 | /* |
248 | * Structure of the page that can be mapped via mmap | |
249 | */ | |
cdd6c482 | 250 | struct perf_event_mmap_page { |
37d81828 PM |
251 | __u32 version; /* version number of this structure */ |
252 | __u32 compat_version; /* lowest version this is compat with */ | |
38ff667b PZ |
253 | |
254 | /* | |
cdd6c482 | 255 | * Bits needed to read the hw events in user-space. |
38ff667b | 256 | * |
92f22a38 PZ |
257 | * u32 seq; |
258 | * s64 count; | |
38ff667b | 259 | * |
a2e87d06 PZ |
260 | * do { |
261 | * seq = pc->lock; | |
38ff667b | 262 | * |
a2e87d06 PZ |
263 | * barrier() |
264 | * if (pc->index) { | |
265 | * count = pmc_read(pc->index - 1); | |
266 | * count += pc->offset; | |
267 | * } else | |
268 | * goto regular_read; | |
38ff667b | 269 | * |
a2e87d06 PZ |
270 | * barrier(); |
271 | * } while (pc->lock != seq); | |
38ff667b | 272 | * |
92f22a38 PZ |
273 | * NOTE: for obvious reason this only works on self-monitoring |
274 | * processes. | |
38ff667b | 275 | */ |
37d81828 | 276 | __u32 lock; /* seqlock for synchronization */ |
cdd6c482 IM |
277 | __u32 index; /* hardware event identifier */ |
278 | __s64 offset; /* add to hardware event value */ | |
279 | __u64 time_enabled; /* time event active */ | |
280 | __u64 time_running; /* time event on cpu */ | |
7b732a75 | 281 | |
41f95331 PZ |
282 | /* |
283 | * Hole for extension of the self monitor capabilities | |
284 | */ | |
285 | ||
7f8b4e4e | 286 | __u64 __reserved[123]; /* align to 1k */ |
41f95331 | 287 | |
38ff667b PZ |
288 | /* |
289 | * Control data for the mmap() data buffer. | |
290 | * | |
43a21ea8 PZ |
291 | * User-space reading the @data_head value should issue an rmb(), on |
292 | * SMP capable platforms, after reading this value -- see | |
cdd6c482 | 293 | * perf_event_wakeup(). |
43a21ea8 PZ |
294 | * |
295 | * When the mapping is PROT_WRITE the @data_tail value should be | |
296 | * written by userspace to reflect the last read data. In this case | |
297 | * the kernel will not over-write unread data. | |
38ff667b | 298 | */ |
8e3747c1 | 299 | __u64 data_head; /* head in the data section */ |
43a21ea8 | 300 | __u64 data_tail; /* user-space written tail */ |
37d81828 PM |
301 | }; |
302 | ||
39447b38 | 303 | #define PERF_RECORD_MISC_CPUMODE_MASK (7 << 0) |
184f412c | 304 | #define PERF_RECORD_MISC_CPUMODE_UNKNOWN (0 << 0) |
cdd6c482 IM |
305 | #define PERF_RECORD_MISC_KERNEL (1 << 0) |
306 | #define PERF_RECORD_MISC_USER (2 << 0) | |
307 | #define PERF_RECORD_MISC_HYPERVISOR (3 << 0) | |
39447b38 ZY |
308 | #define PERF_RECORD_MISC_GUEST_KERNEL (4 << 0) |
309 | #define PERF_RECORD_MISC_GUEST_USER (5 << 0) | |
6fab0192 | 310 | |
ab608344 PZ |
311 | /* |
312 | * Indicates that the content of PERF_SAMPLE_IP points to | |
313 | * the actual instruction that triggered the event. See also | |
314 | * perf_event_attr::precise_ip. | |
315 | */ | |
316 | #define PERF_RECORD_MISC_EXACT_IP (1 << 14) | |
ef21f683 PZ |
317 | /* |
318 | * Reserve the last bit to indicate some extended misc field | |
319 | */ | |
320 | #define PERF_RECORD_MISC_EXT_RESERVED (1 << 15) | |
321 | ||
5c148194 PZ |
322 | struct perf_event_header { |
323 | __u32 type; | |
6fab0192 PZ |
324 | __u16 misc; |
325 | __u16 size; | |
5c148194 PZ |
326 | }; |
327 | ||
328 | enum perf_event_type { | |
5ed00415 | 329 | |
0c593b34 | 330 | /* |
c980d109 ACM |
331 | * If perf_event_attr.sample_id_all is set then all event types will |
332 | * have the sample_type selected fields related to where/when | |
333 | * (identity) an event took place (TID, TIME, ID, CPU, STREAM_ID) | |
334 | * described in PERF_RECORD_SAMPLE below, it will be stashed just after | |
335 | * the perf_event_header and the fields already present for the existing | |
336 | * fields, i.e. at the end of the payload. That way a newer perf.data | |
337 | * file will be supported by older perf tools, with these new optional | |
338 | * fields being ignored. | |
339 | * | |
0c593b34 PZ |
340 | * The MMAP events record the PROT_EXEC mappings so that we can |
341 | * correlate userspace IPs to code. They have the following structure: | |
342 | * | |
343 | * struct { | |
0127c3ea | 344 | * struct perf_event_header header; |
0c593b34 | 345 | * |
0127c3ea IM |
346 | * u32 pid, tid; |
347 | * u64 addr; | |
348 | * u64 len; | |
349 | * u64 pgoff; | |
350 | * char filename[]; | |
0c593b34 PZ |
351 | * }; |
352 | */ | |
cdd6c482 | 353 | PERF_RECORD_MMAP = 1, |
0a4a9391 | 354 | |
43a21ea8 PZ |
355 | /* |
356 | * struct { | |
57c0c15b IM |
357 | * struct perf_event_header header; |
358 | * u64 id; | |
359 | * u64 lost; | |
43a21ea8 PZ |
360 | * }; |
361 | */ | |
cdd6c482 | 362 | PERF_RECORD_LOST = 2, |
43a21ea8 | 363 | |
8d1b2d93 PZ |
364 | /* |
365 | * struct { | |
0127c3ea | 366 | * struct perf_event_header header; |
8d1b2d93 | 367 | * |
0127c3ea IM |
368 | * u32 pid, tid; |
369 | * char comm[]; | |
8d1b2d93 PZ |
370 | * }; |
371 | */ | |
cdd6c482 | 372 | PERF_RECORD_COMM = 3, |
8d1b2d93 | 373 | |
9f498cc5 PZ |
374 | /* |
375 | * struct { | |
376 | * struct perf_event_header header; | |
377 | * u32 pid, ppid; | |
378 | * u32 tid, ptid; | |
393b2ad8 | 379 | * u64 time; |
9f498cc5 PZ |
380 | * }; |
381 | */ | |
cdd6c482 | 382 | PERF_RECORD_EXIT = 4, |
9f498cc5 | 383 | |
26b119bc PZ |
384 | /* |
385 | * struct { | |
0127c3ea IM |
386 | * struct perf_event_header header; |
387 | * u64 time; | |
689802b2 | 388 | * u64 id; |
7f453c24 | 389 | * u64 stream_id; |
a78ac325 PZ |
390 | * }; |
391 | */ | |
184f412c IM |
392 | PERF_RECORD_THROTTLE = 5, |
393 | PERF_RECORD_UNTHROTTLE = 6, | |
a78ac325 | 394 | |
60313ebe PZ |
395 | /* |
396 | * struct { | |
a21ca2ca IM |
397 | * struct perf_event_header header; |
398 | * u32 pid, ppid; | |
9f498cc5 | 399 | * u32 tid, ptid; |
a6f10a2f | 400 | * u64 time; |
60313ebe PZ |
401 | * }; |
402 | */ | |
cdd6c482 | 403 | PERF_RECORD_FORK = 7, |
60313ebe | 404 | |
38b200d6 PZ |
405 | /* |
406 | * struct { | |
184f412c IM |
407 | * struct perf_event_header header; |
408 | * u32 pid, tid; | |
3dab77fb | 409 | * |
184f412c | 410 | * struct read_format values; |
38b200d6 PZ |
411 | * }; |
412 | */ | |
cdd6c482 | 413 | PERF_RECORD_READ = 8, |
38b200d6 | 414 | |
8a057d84 | 415 | /* |
0c593b34 | 416 | * struct { |
0127c3ea | 417 | * struct perf_event_header header; |
0c593b34 | 418 | * |
43a21ea8 PZ |
419 | * { u64 ip; } && PERF_SAMPLE_IP |
420 | * { u32 pid, tid; } && PERF_SAMPLE_TID | |
421 | * { u64 time; } && PERF_SAMPLE_TIME | |
422 | * { u64 addr; } && PERF_SAMPLE_ADDR | |
e6e18ec7 | 423 | * { u64 id; } && PERF_SAMPLE_ID |
7f453c24 | 424 | * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID |
43a21ea8 | 425 | * { u32 cpu, res; } && PERF_SAMPLE_CPU |
57c0c15b | 426 | * { u64 period; } && PERF_SAMPLE_PERIOD |
0c593b34 | 427 | * |
3dab77fb | 428 | * { struct read_format values; } && PERF_SAMPLE_READ |
0c593b34 | 429 | * |
f9188e02 | 430 | * { u64 nr, |
43a21ea8 | 431 | * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN |
3dab77fb | 432 | * |
57c0c15b IM |
433 | * # |
434 | * # The RAW record below is opaque data wrt the ABI | |
435 | * # | |
436 | * # That is, the ABI doesn't make any promises wrt to | |
437 | * # the stability of its content, it may vary depending | |
438 | * # on event, hardware, kernel version and phase of | |
439 | * # the moon. | |
440 | * # | |
441 | * # In other words, PERF_SAMPLE_RAW contents are not an ABI. | |
442 | * # | |
3dab77fb | 443 | * |
a044560c PZ |
444 | * { u32 size; |
445 | * char data[size];}&& PERF_SAMPLE_RAW | |
0c593b34 | 446 | * }; |
8a057d84 | 447 | */ |
184f412c | 448 | PERF_RECORD_SAMPLE = 9, |
e6e18ec7 | 449 | |
cdd6c482 | 450 | PERF_RECORD_MAX, /* non-ABI */ |
5c148194 PZ |
451 | }; |
452 | ||
f9188e02 PZ |
453 | enum perf_callchain_context { |
454 | PERF_CONTEXT_HV = (__u64)-32, | |
455 | PERF_CONTEXT_KERNEL = (__u64)-128, | |
456 | PERF_CONTEXT_USER = (__u64)-512, | |
7522060c | 457 | |
f9188e02 PZ |
458 | PERF_CONTEXT_GUEST = (__u64)-2048, |
459 | PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176, | |
460 | PERF_CONTEXT_GUEST_USER = (__u64)-2560, | |
461 | ||
462 | PERF_CONTEXT_MAX = (__u64)-4095, | |
7522060c IM |
463 | }; |
464 | ||
a4be7c27 PZ |
465 | #define PERF_FLAG_FD_NO_GROUP (1U << 0) |
466 | #define PERF_FLAG_FD_OUTPUT (1U << 1) | |
e5d1367f | 467 | #define PERF_FLAG_PID_CGROUP (1U << 2) /* pid=cgroup id, per-cpu mode only */ |
a4be7c27 | 468 | |
f3dfd265 | 469 | #ifdef __KERNEL__ |
9f66a381 | 470 | /* |
f3dfd265 | 471 | * Kernel-internal data types and definitions: |
9f66a381 IM |
472 | */ |
473 | ||
cdd6c482 | 474 | #ifdef CONFIG_PERF_EVENTS |
e5d1367f | 475 | # include <linux/cgroup.h> |
cdd6c482 | 476 | # include <asm/perf_event.h> |
7be79236 | 477 | # include <asm/local64.h> |
f3dfd265 PM |
478 | #endif |
479 | ||
39447b38 ZY |
480 | struct perf_guest_info_callbacks { |
481 | int (*is_in_guest) (void); | |
482 | int (*is_user_mode) (void); | |
483 | unsigned long (*get_guest_ip) (void); | |
484 | }; | |
485 | ||
2ff6cfd7 AB |
486 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
487 | #include <asm/hw_breakpoint.h> | |
488 | #endif | |
489 | ||
f3dfd265 PM |
490 | #include <linux/list.h> |
491 | #include <linux/mutex.h> | |
492 | #include <linux/rculist.h> | |
493 | #include <linux/rcupdate.h> | |
494 | #include <linux/spinlock.h> | |
d6d020e9 | 495 | #include <linux/hrtimer.h> |
3c446b3d | 496 | #include <linux/fs.h> |
709e50cf | 497 | #include <linux/pid_namespace.h> |
906010b2 | 498 | #include <linux/workqueue.h> |
5331d7b8 | 499 | #include <linux/ftrace.h> |
85cfabbc | 500 | #include <linux/cpu.h> |
e360adbe | 501 | #include <linux/irq_work.h> |
82cd6def | 502 | #include <linux/jump_label_ref.h> |
f3dfd265 | 503 | #include <asm/atomic.h> |
fa588151 | 504 | #include <asm/local.h> |
f3dfd265 | 505 | |
f9188e02 PZ |
506 | #define PERF_MAX_STACK_DEPTH 255 |
507 | ||
508 | struct perf_callchain_entry { | |
509 | __u64 nr; | |
510 | __u64 ip[PERF_MAX_STACK_DEPTH]; | |
511 | }; | |
512 | ||
3a43ce68 FW |
513 | struct perf_raw_record { |
514 | u32 size; | |
515 | void *data; | |
f413cdb8 FW |
516 | }; |
517 | ||
caff2bef PZ |
518 | struct perf_branch_entry { |
519 | __u64 from; | |
520 | __u64 to; | |
521 | __u64 flags; | |
522 | }; | |
523 | ||
524 | struct perf_branch_stack { | |
525 | __u64 nr; | |
526 | struct perf_branch_entry entries[0]; | |
527 | }; | |
528 | ||
f3dfd265 PM |
529 | struct task_struct; |
530 | ||
0793a61d | 531 | /** |
cdd6c482 | 532 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 533 | */ |
cdd6c482 IM |
534 | struct hw_perf_event { |
535 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
536 | union { |
537 | struct { /* hardware */ | |
a308444c | 538 | u64 config; |
447a194b | 539 | u64 last_tag; |
a308444c | 540 | unsigned long config_base; |
cdd6c482 | 541 | unsigned long event_base; |
a308444c | 542 | int idx; |
447a194b | 543 | int last_cpu; |
d6d020e9 | 544 | }; |
721a669b | 545 | struct { /* software */ |
a308444c | 546 | struct hrtimer hrtimer; |
d6d020e9 | 547 | }; |
24f1e32c | 548 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 FW |
549 | struct { /* breakpoint */ |
550 | struct arch_hw_breakpoint info; | |
551 | struct list_head bp_list; | |
d580ff86 PZ |
552 | /* |
553 | * Crufty hack to avoid the chicken and egg | |
554 | * problem hw_breakpoint has with context | |
555 | * creation and event initalization. | |
556 | */ | |
557 | struct task_struct *bp_target; | |
45a73372 | 558 | }; |
24f1e32c | 559 | #endif |
d6d020e9 | 560 | }; |
a4eaf7f1 | 561 | int state; |
e7850595 | 562 | local64_t prev_count; |
b23f3325 | 563 | u64 sample_period; |
9e350de3 | 564 | u64 last_period; |
e7850595 | 565 | local64_t period_left; |
60db5e09 | 566 | u64 interrupts; |
6a24ed6c | 567 | |
abd50713 PZ |
568 | u64 freq_time_stamp; |
569 | u64 freq_count_stamp; | |
ee06094f | 570 | #endif |
0793a61d TG |
571 | }; |
572 | ||
a4eaf7f1 PZ |
573 | /* |
574 | * hw_perf_event::state flags | |
575 | */ | |
576 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
577 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
578 | #define PERF_HES_ARCH 0x04 | |
579 | ||
cdd6c482 | 580 | struct perf_event; |
621a01ea | 581 | |
8d2cacbb PZ |
582 | /* |
583 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
584 | */ | |
585 | #define PERF_EVENT_TXN 0x1 | |
6bde9b6c | 586 | |
621a01ea | 587 | /** |
4aeb0b42 | 588 | * struct pmu - generic performance monitoring unit |
621a01ea | 589 | */ |
4aeb0b42 | 590 | struct pmu { |
b0a873eb PZ |
591 | struct list_head entry; |
592 | ||
abe43400 | 593 | struct device *dev; |
2e80a82a PZ |
594 | char *name; |
595 | int type; | |
596 | ||
108b02cf PZ |
597 | int * __percpu pmu_disable_count; |
598 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
8dc85d54 | 599 | int task_ctx_nr; |
6bde9b6c LM |
600 | |
601 | /* | |
a4eaf7f1 PZ |
602 | * Fully disable/enable this PMU, can be used to protect from the PMI |
603 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 604 | */ |
ad5133b7 PZ |
605 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
606 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 607 | |
8d2cacbb | 608 | /* |
a4eaf7f1 | 609 | * Try and initialize the event for this PMU. |
24cd7f54 | 610 | * Should return -ENOENT when the @event doesn't match this PMU. |
8d2cacbb | 611 | */ |
b0a873eb PZ |
612 | int (*event_init) (struct perf_event *event); |
613 | ||
a4eaf7f1 PZ |
614 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
615 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
616 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
617 | ||
8d2cacbb | 618 | /* |
a4eaf7f1 PZ |
619 | * Adds/Removes a counter to/from the PMU, can be done inside |
620 | * a transaction, see the ->*_txn() methods. | |
621 | */ | |
622 | int (*add) (struct perf_event *event, int flags); | |
623 | void (*del) (struct perf_event *event, int flags); | |
624 | ||
625 | /* | |
626 | * Starts/Stops a counter present on the PMU. The PMI handler | |
627 | * should stop the counter when perf_event_overflow() returns | |
628 | * !0. ->start() will be used to continue. | |
629 | */ | |
630 | void (*start) (struct perf_event *event, int flags); | |
631 | void (*stop) (struct perf_event *event, int flags); | |
632 | ||
633 | /* | |
634 | * Updates the counter value of the event. | |
635 | */ | |
cdd6c482 | 636 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
637 | |
638 | /* | |
24cd7f54 PZ |
639 | * Group events scheduling is treated as a transaction, add |
640 | * group events as a whole and perform one schedulability test. | |
641 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
642 | * |
643 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 644 | * do schedulability tests. |
8d2cacbb | 645 | */ |
ad5133b7 | 646 | void (*start_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 647 | /* |
a4eaf7f1 | 648 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
649 | * then ->commit_txn() is required to perform one. On success |
650 | * the transaction is closed. On error the transaction is kept | |
651 | * open until ->cancel_txn() is called. | |
652 | */ | |
ad5133b7 | 653 | int (*commit_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 654 | /* |
a4eaf7f1 PZ |
655 | * Will cancel the transaction, assumes ->del() is called |
656 | * for each successfull ->add() during the transaction. | |
8d2cacbb | 657 | */ |
ad5133b7 | 658 | void (*cancel_txn) (struct pmu *pmu); /* optional */ |
621a01ea IM |
659 | }; |
660 | ||
6a930700 | 661 | /** |
cdd6c482 | 662 | * enum perf_event_active_state - the states of a event |
6a930700 | 663 | */ |
cdd6c482 | 664 | enum perf_event_active_state { |
57c0c15b | 665 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
666 | PERF_EVENT_STATE_OFF = -1, |
667 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 668 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
669 | }; |
670 | ||
9b51f66d IM |
671 | struct file; |
672 | ||
d57e34fd PZ |
673 | #define PERF_BUFFER_WRITABLE 0x01 |
674 | ||
ca5135e6 | 675 | struct perf_buffer { |
ac9721f3 | 676 | atomic_t refcount; |
7b732a75 | 677 | struct rcu_head rcu_head; |
906010b2 PZ |
678 | #ifdef CONFIG_PERF_USE_VMALLOC |
679 | struct work_struct work; | |
3cafa9fb | 680 | int page_order; /* allocation order */ |
906010b2 | 681 | #endif |
8740f941 | 682 | int nr_pages; /* nr of data pages */ |
43a21ea8 | 683 | int writable; /* are we writable */ |
8740f941 | 684 | |
c33a0bc4 | 685 | atomic_t poll; /* POLL_ for wakeups */ |
8740f941 | 686 | |
fa588151 PZ |
687 | local_t head; /* write position */ |
688 | local_t nest; /* nested writers */ | |
689 | local_t events; /* event limit */ | |
adb8e118 | 690 | local_t wakeup; /* wakeup stamp */ |
fa588151 | 691 | local_t lost; /* nr records lost */ |
ef60777c | 692 | |
2667de81 PZ |
693 | long watermark; /* wakeup watermark */ |
694 | ||
57c0c15b | 695 | struct perf_event_mmap_page *user_page; |
0127c3ea | 696 | void *data_pages[0]; |
7b732a75 PZ |
697 | }; |
698 | ||
453f19ee PZ |
699 | struct perf_sample_data; |
700 | ||
b326e956 FW |
701 | typedef void (*perf_overflow_handler_t)(struct perf_event *, int, |
702 | struct perf_sample_data *, | |
703 | struct pt_regs *regs); | |
704 | ||
d6f962b5 FW |
705 | enum perf_group_flag { |
706 | PERF_GROUP_SOFTWARE = 0x1, | |
707 | }; | |
708 | ||
76e1d904 FW |
709 | #define SWEVENT_HLIST_BITS 8 |
710 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
711 | ||
712 | struct swevent_hlist { | |
713 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; | |
714 | struct rcu_head rcu_head; | |
715 | }; | |
716 | ||
8a49542c PZ |
717 | #define PERF_ATTACH_CONTEXT 0x01 |
718 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 719 | #define PERF_ATTACH_TASK 0x04 |
8a49542c | 720 | |
e5d1367f SE |
721 | #ifdef CONFIG_CGROUP_PERF |
722 | /* | |
723 | * perf_cgroup_info keeps track of time_enabled for a cgroup. | |
724 | * This is a per-cpu dynamically allocated data structure. | |
725 | */ | |
726 | struct perf_cgroup_info { | |
727 | u64 time; | |
728 | u64 timestamp; | |
729 | }; | |
730 | ||
731 | struct perf_cgroup { | |
732 | struct cgroup_subsys_state css; | |
733 | struct perf_cgroup_info *info; /* timing info, one per cpu */ | |
734 | }; | |
735 | #endif | |
736 | ||
0793a61d | 737 | /** |
cdd6c482 | 738 | * struct perf_event - performance event kernel representation: |
0793a61d | 739 | */ |
cdd6c482 IM |
740 | struct perf_event { |
741 | #ifdef CONFIG_PERF_EVENTS | |
65abc865 | 742 | struct list_head group_entry; |
592903cd | 743 | struct list_head event_entry; |
04289bb9 | 744 | struct list_head sibling_list; |
76e1d904 | 745 | struct hlist_node hlist_entry; |
0127c3ea | 746 | int nr_siblings; |
d6f962b5 | 747 | int group_flags; |
cdd6c482 | 748 | struct perf_event *group_leader; |
a4eaf7f1 | 749 | struct pmu *pmu; |
04289bb9 | 750 | |
cdd6c482 | 751 | enum perf_event_active_state state; |
8a49542c | 752 | unsigned int attach_state; |
e7850595 | 753 | local64_t count; |
a6e6dea6 | 754 | atomic64_t child_count; |
ee06094f | 755 | |
53cfbf59 | 756 | /* |
cdd6c482 | 757 | * These are the total time in nanoseconds that the event |
53cfbf59 | 758 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 759 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
760 | * and running (scheduled onto the CPU), respectively. |
761 | * | |
762 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 763 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
764 | */ |
765 | u64 total_time_enabled; | |
766 | u64 total_time_running; | |
767 | ||
768 | /* | |
769 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 770 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
771 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
772 | * in time. | |
cdd6c482 IM |
773 | * tstamp_enabled: the notional time when the event was enabled |
774 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 775 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 776 | * event was scheduled off. |
53cfbf59 PM |
777 | */ |
778 | u64 tstamp_enabled; | |
779 | u64 tstamp_running; | |
780 | u64 tstamp_stopped; | |
781 | ||
eed01528 SE |
782 | /* |
783 | * timestamp shadows the actual context timing but it can | |
784 | * be safely used in NMI interrupt context. It reflects the | |
785 | * context time as it was when the event was last scheduled in. | |
786 | * | |
787 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
788 | * compute ctx_time for a sample, simply add perf_clock(). | |
789 | */ | |
790 | u64 shadow_ctx_time; | |
791 | ||
24f1e32c | 792 | struct perf_event_attr attr; |
c320c7b7 | 793 | u16 header_size; |
6844c09d | 794 | u16 id_header_size; |
c320c7b7 | 795 | u16 read_size; |
cdd6c482 | 796 | struct hw_perf_event hw; |
0793a61d | 797 | |
cdd6c482 | 798 | struct perf_event_context *ctx; |
9b51f66d | 799 | struct file *filp; |
0793a61d | 800 | |
53cfbf59 PM |
801 | /* |
802 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 803 | * events have been enabled and running, respectively. |
53cfbf59 PM |
804 | */ |
805 | atomic64_t child_total_time_enabled; | |
806 | atomic64_t child_total_time_running; | |
807 | ||
0793a61d | 808 | /* |
d859e29f | 809 | * Protect attach/detach and child_list: |
0793a61d | 810 | */ |
fccc714b PZ |
811 | struct mutex child_mutex; |
812 | struct list_head child_list; | |
cdd6c482 | 813 | struct perf_event *parent; |
0793a61d TG |
814 | |
815 | int oncpu; | |
816 | int cpu; | |
817 | ||
082ff5a2 PZ |
818 | struct list_head owner_entry; |
819 | struct task_struct *owner; | |
820 | ||
7b732a75 PZ |
821 | /* mmap bits */ |
822 | struct mutex mmap_mutex; | |
823 | atomic_t mmap_count; | |
ac9721f3 PZ |
824 | int mmap_locked; |
825 | struct user_struct *mmap_user; | |
ca5135e6 | 826 | struct perf_buffer *buffer; |
37d81828 | 827 | |
7b732a75 | 828 | /* poll related */ |
0793a61d | 829 | wait_queue_head_t waitq; |
3c446b3d | 830 | struct fasync_struct *fasync; |
79f14641 PZ |
831 | |
832 | /* delayed work for NMIs and such */ | |
833 | int pending_wakeup; | |
4c9e2542 | 834 | int pending_kill; |
79f14641 | 835 | int pending_disable; |
e360adbe | 836 | struct irq_work pending; |
592903cd | 837 | |
79f14641 PZ |
838 | atomic_t event_limit; |
839 | ||
cdd6c482 | 840 | void (*destroy)(struct perf_event *); |
592903cd | 841 | struct rcu_head rcu_head; |
709e50cf PZ |
842 | |
843 | struct pid_namespace *ns; | |
8e5799b1 | 844 | u64 id; |
6fb2915d | 845 | |
b326e956 | 846 | perf_overflow_handler_t overflow_handler; |
453f19ee | 847 | |
07b139c8 | 848 | #ifdef CONFIG_EVENT_TRACING |
1c024eca | 849 | struct ftrace_event_call *tp_event; |
6fb2915d | 850 | struct event_filter *filter; |
ee06094f | 851 | #endif |
6fb2915d | 852 | |
e5d1367f SE |
853 | #ifdef CONFIG_CGROUP_PERF |
854 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
855 | int cgrp_defer_enabled; | |
856 | #endif | |
857 | ||
6fb2915d | 858 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
859 | }; |
860 | ||
b04243ef PZ |
861 | enum perf_event_context_type { |
862 | task_context, | |
863 | cpu_context, | |
864 | }; | |
865 | ||
0793a61d | 866 | /** |
cdd6c482 | 867 | * struct perf_event_context - event context structure |
0793a61d | 868 | * |
cdd6c482 | 869 | * Used as a container for task events and CPU events as well: |
0793a61d | 870 | */ |
cdd6c482 | 871 | struct perf_event_context { |
b04243ef | 872 | enum perf_event_context_type type; |
108b02cf | 873 | struct pmu *pmu; |
0793a61d | 874 | /* |
cdd6c482 | 875 | * Protect the states of the events in the list, |
d859e29f | 876 | * nr_active, and the list: |
0793a61d | 877 | */ |
e625cce1 | 878 | raw_spinlock_t lock; |
d859e29f | 879 | /* |
cdd6c482 | 880 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
881 | * is sufficient to ensure the list doesn't change; to change |
882 | * the list you need to lock both the mutex and the spinlock. | |
883 | */ | |
a308444c | 884 | struct mutex mutex; |
04289bb9 | 885 | |
889ff015 FW |
886 | struct list_head pinned_groups; |
887 | struct list_head flexible_groups; | |
a308444c | 888 | struct list_head event_list; |
cdd6c482 | 889 | int nr_events; |
a308444c IM |
890 | int nr_active; |
891 | int is_active; | |
bfbd3381 | 892 | int nr_stat; |
dddd3379 | 893 | int rotate_disable; |
a308444c IM |
894 | atomic_t refcount; |
895 | struct task_struct *task; | |
53cfbf59 PM |
896 | |
897 | /* | |
4af4998b | 898 | * Context clock, runs when context enabled. |
53cfbf59 | 899 | */ |
a308444c IM |
900 | u64 time; |
901 | u64 timestamp; | |
564c2b21 PM |
902 | |
903 | /* | |
904 | * These fields let us detect when two contexts have both | |
905 | * been cloned (inherited) from a common ancestor. | |
906 | */ | |
cdd6c482 | 907 | struct perf_event_context *parent_ctx; |
a308444c IM |
908 | u64 parent_gen; |
909 | u64 generation; | |
910 | int pin_count; | |
911 | struct rcu_head rcu_head; | |
e5d1367f | 912 | int nr_cgroups; /* cgroup events present */ |
0793a61d TG |
913 | }; |
914 | ||
7ae07ea3 FW |
915 | /* |
916 | * Number of contexts where an event can trigger: | |
917 | * task, softirq, hardirq, nmi. | |
918 | */ | |
919 | #define PERF_NR_CONTEXTS 4 | |
920 | ||
0793a61d | 921 | /** |
cdd6c482 | 922 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
923 | */ |
924 | struct perf_cpu_context { | |
cdd6c482 IM |
925 | struct perf_event_context ctx; |
926 | struct perf_event_context *task_ctx; | |
0793a61d | 927 | int active_oncpu; |
3b6f9e5c | 928 | int exclusive; |
e9d2b064 PZ |
929 | struct list_head rotation_list; |
930 | int jiffies_interval; | |
51676957 | 931 | struct pmu *active_pmu; |
e5d1367f SE |
932 | #ifdef CONFIG_CGROUP_PERF |
933 | struct perf_cgroup *cgrp; | |
934 | #endif | |
0793a61d TG |
935 | }; |
936 | ||
5622f295 | 937 | struct perf_output_handle { |
57c0c15b | 938 | struct perf_event *event; |
ca5135e6 | 939 | struct perf_buffer *buffer; |
6d1acfd5 | 940 | unsigned long wakeup; |
5d967a8b PZ |
941 | unsigned long size; |
942 | void *addr; | |
943 | int page; | |
57c0c15b IM |
944 | int nmi; |
945 | int sample; | |
5622f295 MM |
946 | }; |
947 | ||
cdd6c482 | 948 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 949 | |
2e80a82a | 950 | extern int perf_pmu_register(struct pmu *pmu, char *name, int type); |
b0a873eb | 951 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 952 | |
3bf101ba | 953 | extern int perf_num_counters(void); |
84c79910 | 954 | extern const char *perf_pmu_name(void); |
82cd6def PZ |
955 | extern void __perf_event_task_sched_in(struct task_struct *task); |
956 | extern void __perf_event_task_sched_out(struct task_struct *task, struct task_struct *next); | |
cdd6c482 IM |
957 | extern int perf_event_init_task(struct task_struct *child); |
958 | extern void perf_event_exit_task(struct task_struct *child); | |
959 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 960 | extern void perf_event_delayed_put(struct task_struct *task); |
cdd6c482 | 961 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
962 | extern void perf_pmu_disable(struct pmu *pmu); |
963 | extern void perf_pmu_enable(struct pmu *pmu); | |
cdd6c482 IM |
964 | extern int perf_event_task_disable(void); |
965 | extern int perf_event_task_enable(void); | |
cdd6c482 | 966 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
967 | extern int perf_event_release_kernel(struct perf_event *event); |
968 | extern struct perf_event * | |
969 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
970 | int cpu, | |
38a81da2 | 971 | struct task_struct *task, |
b326e956 | 972 | perf_overflow_handler_t callback); |
59ed446f PZ |
973 | extern u64 perf_event_read_value(struct perf_event *event, |
974 | u64 *enabled, u64 *running); | |
5c92d124 | 975 | |
df1a132b | 976 | struct perf_sample_data { |
5622f295 MM |
977 | u64 type; |
978 | ||
979 | u64 ip; | |
980 | struct { | |
981 | u32 pid; | |
982 | u32 tid; | |
983 | } tid_entry; | |
984 | u64 time; | |
a308444c | 985 | u64 addr; |
5622f295 MM |
986 | u64 id; |
987 | u64 stream_id; | |
988 | struct { | |
989 | u32 cpu; | |
990 | u32 reserved; | |
991 | } cpu_entry; | |
a308444c | 992 | u64 period; |
5622f295 | 993 | struct perf_callchain_entry *callchain; |
3a43ce68 | 994 | struct perf_raw_record *raw; |
df1a132b PZ |
995 | }; |
996 | ||
dc1d628a PZ |
997 | static inline |
998 | void perf_sample_data_init(struct perf_sample_data *data, u64 addr) | |
999 | { | |
1000 | data->addr = addr; | |
1001 | data->raw = NULL; | |
1002 | } | |
1003 | ||
5622f295 MM |
1004 | extern void perf_output_sample(struct perf_output_handle *handle, |
1005 | struct perf_event_header *header, | |
1006 | struct perf_sample_data *data, | |
cdd6c482 | 1007 | struct perf_event *event); |
5622f295 MM |
1008 | extern void perf_prepare_sample(struct perf_event_header *header, |
1009 | struct perf_sample_data *data, | |
cdd6c482 | 1010 | struct perf_event *event, |
5622f295 MM |
1011 | struct pt_regs *regs); |
1012 | ||
cdd6c482 | 1013 | extern int perf_event_overflow(struct perf_event *event, int nmi, |
5622f295 MM |
1014 | struct perf_sample_data *data, |
1015 | struct pt_regs *regs); | |
df1a132b | 1016 | |
6c7e550f FBH |
1017 | static inline bool is_sampling_event(struct perf_event *event) |
1018 | { | |
1019 | return event->attr.sample_period != 0; | |
1020 | } | |
1021 | ||
3b6f9e5c | 1022 | /* |
cdd6c482 | 1023 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 1024 | */ |
cdd6c482 | 1025 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 1026 | { |
89a1e187 | 1027 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
1028 | } |
1029 | ||
cdd6c482 | 1030 | extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 1031 | |
cdd6c482 | 1032 | extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64); |
f29ac756 | 1033 | |
b0f82b81 FW |
1034 | #ifndef perf_arch_fetch_caller_regs |
1035 | static inline void | |
5cfaf214 | 1036 | perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 1037 | #endif |
5331d7b8 FW |
1038 | |
1039 | /* | |
1040 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
1041 | * the nth caller. We only need a few of the regs: | |
1042 | * - ip for PERF_SAMPLE_IP | |
1043 | * - cs for user_mode() tests | |
1044 | * - bp for callchains | |
1045 | * - eflags, for future purposes, just in case | |
1046 | */ | |
b0f82b81 | 1047 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 1048 | { |
5331d7b8 FW |
1049 | memset(regs, 0, sizeof(*regs)); |
1050 | ||
b0f82b81 | 1051 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
1052 | } |
1053 | ||
7e54a5a0 | 1054 | static __always_inline void |
e49a5bd3 FW |
1055 | perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) |
1056 | { | |
7e54a5a0 PZ |
1057 | struct pt_regs hot_regs; |
1058 | ||
1059 | JUMP_LABEL(&perf_swevent_enabled[event_id], have_event); | |
1060 | return; | |
1061 | ||
1062 | have_event: | |
1063 | if (!regs) { | |
1064 | perf_fetch_caller_regs(&hot_regs); | |
1065 | regs = &hot_regs; | |
e49a5bd3 | 1066 | } |
7e54a5a0 | 1067 | __perf_sw_event(event_id, nr, nmi, regs, addr); |
e49a5bd3 FW |
1068 | } |
1069 | ||
e5d1367f | 1070 | extern atomic_t perf_sched_events; |
ee6dcfa4 PZ |
1071 | |
1072 | static inline void perf_event_task_sched_in(struct task_struct *task) | |
1073 | { | |
e5d1367f | 1074 | COND_STMT(&perf_sched_events, __perf_event_task_sched_in(task)); |
ee6dcfa4 PZ |
1075 | } |
1076 | ||
1077 | static inline | |
1078 | void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next) | |
1079 | { | |
1080 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0); | |
1081 | ||
e5d1367f | 1082 | COND_STMT(&perf_sched_events, __perf_event_task_sched_out(task, next)); |
ee6dcfa4 PZ |
1083 | } |
1084 | ||
3af9e859 | 1085 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 1086 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
1087 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
1088 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 1089 | |
cdd6c482 IM |
1090 | extern void perf_event_comm(struct task_struct *tsk); |
1091 | extern void perf_event_fork(struct task_struct *tsk); | |
8d1b2d93 | 1092 | |
56962b44 FW |
1093 | /* Callchains */ |
1094 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
1095 | ||
1096 | extern void perf_callchain_user(struct perf_callchain_entry *entry, | |
1097 | struct pt_regs *regs); | |
1098 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, | |
1099 | struct pt_regs *regs); | |
56962b44 | 1100 | |
394ee076 | 1101 | |
70791ce9 FW |
1102 | static inline void |
1103 | perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) | |
1104 | { | |
1105 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
1106 | entry->ip[entry->nr++] = ip; | |
1107 | } | |
394ee076 | 1108 | |
cdd6c482 IM |
1109 | extern int sysctl_perf_event_paranoid; |
1110 | extern int sysctl_perf_event_mlock; | |
1111 | extern int sysctl_perf_event_sample_rate; | |
1ccd1549 | 1112 | |
163ec435 PZ |
1113 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
1114 | void __user *buffer, size_t *lenp, | |
1115 | loff_t *ppos); | |
1116 | ||
320ebf09 PZ |
1117 | static inline bool perf_paranoid_tracepoint_raw(void) |
1118 | { | |
1119 | return sysctl_perf_event_paranoid > -1; | |
1120 | } | |
1121 | ||
1122 | static inline bool perf_paranoid_cpu(void) | |
1123 | { | |
1124 | return sysctl_perf_event_paranoid > 0; | |
1125 | } | |
1126 | ||
1127 | static inline bool perf_paranoid_kernel(void) | |
1128 | { | |
1129 | return sysctl_perf_event_paranoid > 1; | |
1130 | } | |
1131 | ||
cdd6c482 | 1132 | extern void perf_event_init(void); |
1c024eca PZ |
1133 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
1134 | int entry_size, struct pt_regs *regs, | |
ecc55f84 | 1135 | struct hlist_head *head, int rctx); |
24f1e32c | 1136 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 1137 | |
9d23a90a | 1138 | #ifndef perf_misc_flags |
cdd6c482 IM |
1139 | #define perf_misc_flags(regs) (user_mode(regs) ? PERF_RECORD_MISC_USER : \ |
1140 | PERF_RECORD_MISC_KERNEL) | |
9d23a90a PM |
1141 | #define perf_instruction_pointer(regs) instruction_pointer(regs) |
1142 | #endif | |
1143 | ||
5622f295 | 1144 | extern int perf_output_begin(struct perf_output_handle *handle, |
cdd6c482 | 1145 | struct perf_event *event, unsigned int size, |
5622f295 MM |
1146 | int nmi, int sample); |
1147 | extern void perf_output_end(struct perf_output_handle *handle); | |
1148 | extern void perf_output_copy(struct perf_output_handle *handle, | |
1149 | const void *buf, unsigned int len); | |
4ed7c92d PZ |
1150 | extern int perf_swevent_get_recursion_context(void); |
1151 | extern void perf_swevent_put_recursion_context(int rctx); | |
44234adc FW |
1152 | extern void perf_event_enable(struct perf_event *event); |
1153 | extern void perf_event_disable(struct perf_event *event); | |
e9d2b064 | 1154 | extern void perf_event_task_tick(void); |
0793a61d TG |
1155 | #else |
1156 | static inline void | |
49f47433 | 1157 | perf_event_task_sched_in(struct task_struct *task) { } |
0793a61d | 1158 | static inline void |
cdd6c482 | 1159 | perf_event_task_sched_out(struct task_struct *task, |
49f47433 | 1160 | struct task_struct *next) { } |
cdd6c482 IM |
1161 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
1162 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
1163 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 1164 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
57c0c15b | 1165 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
1166 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
1167 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
15dbf27c | 1168 | |
925d519a | 1169 | static inline void |
cdd6c482 | 1170 | perf_sw_event(u32 event_id, u64 nr, int nmi, |
78f13e95 | 1171 | struct pt_regs *regs, u64 addr) { } |
24f1e32c | 1172 | static inline void |
184f412c | 1173 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 1174 | |
39447b38 | 1175 | static inline int perf_register_guest_info_callbacks |
dcf46b94 | 1176 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1177 | static inline int perf_unregister_guest_info_callbacks |
dcf46b94 | 1178 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1179 | |
57c0c15b | 1180 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
cdd6c482 IM |
1181 | static inline void perf_event_comm(struct task_struct *tsk) { } |
1182 | static inline void perf_event_fork(struct task_struct *tsk) { } | |
1183 | static inline void perf_event_init(void) { } | |
184f412c | 1184 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 1185 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
44234adc FW |
1186 | static inline void perf_event_enable(struct perf_event *event) { } |
1187 | static inline void perf_event_disable(struct perf_event *event) { } | |
e9d2b064 | 1188 | static inline void perf_event_task_tick(void) { } |
0793a61d TG |
1189 | #endif |
1190 | ||
5622f295 MM |
1191 | #define perf_output_put(handle, x) \ |
1192 | perf_output_copy((handle), &(x), sizeof(x)) | |
1193 | ||
3f6da390 PZ |
1194 | /* |
1195 | * This has to have a higher priority than migration_notifier in sched.c. | |
1196 | */ | |
1197 | #define perf_cpu_notifier(fn) \ | |
1198 | do { \ | |
1199 | static struct notifier_block fn##_nb __cpuinitdata = \ | |
50a323b7 | 1200 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ |
3f6da390 PZ |
1201 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
1202 | (void *)(unsigned long)smp_processor_id()); \ | |
1203 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ | |
1204 | (void *)(unsigned long)smp_processor_id()); \ | |
1205 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ | |
1206 | (void *)(unsigned long)smp_processor_id()); \ | |
1207 | register_cpu_notifier(&fn##_nb); \ | |
1208 | } while (0) | |
1209 | ||
f3dfd265 | 1210 | #endif /* __KERNEL__ */ |
cdd6c482 | 1211 | #endif /* _LINUX_PERF_EVENT_H */ |