]>
Commit | Line | Data |
---|---|---|
f2b1d720 MK |
1 | .\" Copyright (c) 2012, Vincent Weaver |
2 | .\" | |
1dd72f9c | 3 | .\" %%%LICENSE_START(GPLv2+_DOC_FULL) |
f2b1d720 MK |
4 | .\" This is free documentation; you can redistribute it and/or |
5 | .\" modify it under the terms of the GNU General Public License as | |
6 | .\" published by the Free Software Foundation; either version 2 of | |
7 | .\" the License, or (at your option) any later version. | |
8 | .\" | |
9 | .\" The GNU General Public License's references to "object code" | |
10 | .\" and "executables" are to be interpreted as the output of any | |
11 | .\" document formatting or typesetting system, including | |
12 | .\" intermediate and printed output. | |
13 | .\" | |
14 | .\" This manual is distributed in the hope that it will be useful, | |
15 | .\" but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | .\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | .\" GNU General Public License for more details. | |
18 | .\" | |
19 | .\" You should have received a copy of the GNU General Public | |
20 | .\" License along with this manual; if not, see | |
21 | .\" <http://www.gnu.org/licenses/>. | |
6a8d8745 | 22 | .\" %%%LICENSE_END |
f2b1d720 MK |
23 | .\" |
24 | .\" This document is based on the perf_event.h header file, the | |
25 | .\" tools/perf/design.txt file, and a lot of bitter experience. | |
26 | .\" | |
35deeb87 | 27 | .TH PERF_EVENT_OPEN 2 2016-12-12 "Linux" "Linux Programmer's Manual" |
f2b1d720 MK |
28 | .SH NAME |
29 | perf_event_open \- set up performance monitoring | |
30 | .SH SYNOPSIS | |
31 | .nf | |
32 | .B #include <linux/perf_event.h> | |
33 | .B #include <linux/hw_breakpoint.h> | |
34 | .sp | |
35 | .BI "int perf_event_open(struct perf_event_attr *" attr , | |
36 | .BI " pid_t " pid ", int " cpu ", int " group_fd , | |
37 | .BI " unsigned long " flags ); | |
38 | .fi | |
39 | ||
40 | .IR Note : | |
41 | There is no glibc wrapper for this system call; see NOTES. | |
42 | .SH DESCRIPTION | |
43 | Given a list of parameters, | |
44 | .BR perf_event_open () | |
45 | returns a file descriptor, for use in subsequent system calls | |
46 | .RB ( read "(2), " mmap "(2), " prctl "(2), " fcntl "(2), etc.)." | |
47 | .PP | |
48 | A call to | |
49 | .BR perf_event_open () | |
50 | creates a file descriptor that allows measuring performance | |
51 | information. | |
52 | Each file descriptor corresponds to one | |
53 | event that is measured; these can be grouped together | |
54 | to measure multiple events simultaneously. | |
55 | .PP | |
56 | Events can be enabled and disabled in two ways: via | |
57 | .BR ioctl (2) | |
58 | and via | |
0fe9e4b1 | 59 | .BR prctl (2). |
f2b1d720 MK |
60 | When an event is disabled it does not count or generate overflows but does |
61 | continue to exist and maintain its count value. | |
62 | .PP | |
63 | Events come in two flavors: counting and sampled. | |
64 | A | |
65 | .I counting | |
66 | event is one that is used for counting the aggregate number of events | |
67 | that occur. | |
68 | In general, counting event results are gathered with a | |
69 | .BR read (2) | |
70 | call. | |
71 | A | |
72 | .I sampling | |
73 | event periodically writes measurements to a buffer that can then | |
74 | be accessed via | |
0fe9e4b1 | 75 | .BR mmap (2). |
f2b1d720 MK |
76 | .SS Arguments |
77 | .P | |
f2b1d720 | 78 | The |
a02a1737 | 79 | .I pid |
f2b1d720 | 80 | and |
a02a1737 VW |
81 | .I cpu |
82 | arguments allow specifying which process and CPU to monitor: | |
83 | .TP | |
f2d15dc9 | 84 | .BR "pid == 0" " and " "cpu == \-1" |
ee7b0cbf | 85 | This measures the calling process/thread on any CPU. |
a02a1737 | 86 | .TP |
f2d15dc9 | 87 | .BR "pid == 0" " and " "cpu >= 0" |
ee7b0cbf | 88 | This measures the calling process/thread only |
a02a1737 VW |
89 | when running on the specified CPU. |
90 | .TP | |
f2d15dc9 | 91 | .BR "pid > 0" " and " "cpu == \-1" |
a02a1737 VW |
92 | This measures the specified process/thread on any CPU. |
93 | .TP | |
f2d15dc9 | 94 | .BR "pid > 0" " and " "cpu >= 0" |
a02a1737 VW |
95 | This measures the specified process/thread only |
96 | when running on the specified CPU. | |
97 | .TP | |
f2d15dc9 | 98 | .BR "pid == \-1" " and " "cpu >= 0" |
a02a1737 | 99 | This measures all processes/threads on the specified CPU. |
ce88f77b | 100 | This requires |
f2b1d720 MK |
101 | .B CAP_SYS_ADMIN |
102 | capability or a | |
103 | .I /proc/sys/kernel/perf_event_paranoid | |
104 | value of less than 1. | |
a02a1737 | 105 | .TP |
ce88f77b | 106 | .BR "pid == \-1" " and " "cpu == \-1" |
a02a1737 | 107 | This setting is invalid and will return an error. |
f2b1d720 | 108 | .P |
13ec13dc MK |
109 | When |
110 | .I pid | |
111 | is greater than zero, permission to perform this system call | |
112 | is governed by a ptrace access mode | |
113 | .B PTRACE_MODE_READ_REALCREDS | |
114 | check; see | |
115 | .BR ptrace (2). | |
116 | ||
f2b1d720 MK |
117 | The |
118 | .I group_fd | |
119 | argument allows event groups to be created. | |
120 | An event group has one event which is the group leader. | |
121 | The leader is created first, with | |
122 | .IR group_fd " = \-1." | |
123 | The rest of the group members are created with subsequent | |
124 | .BR perf_event_open () | |
125 | calls with | |
126 | .IR group_fd | |
bec6277e | 127 | being set to the file descriptor of the group leader. |
f2b1d720 MK |
128 | (A single event on its own is created with |
129 | .IR group_fd " = \-1" | |
130 | and is considered to be a group with only 1 member.) | |
33a0ccb2 | 131 | An event group is scheduled onto the CPU as a unit: it will |
d1007d14 | 132 | be put onto the CPU only if all of the events in the group can be put onto |
f2b1d720 MK |
133 | the CPU. |
134 | This means that the values of the member events can be | |
ce88f77b | 135 | meaningfully compared\(emadded, divided (to get ratios), and so on\(emwith each |
f2b1d720 MK |
136 | other, since they have counted events for the same set of executed |
137 | instructions. | |
138 | .P | |
139 | The | |
140 | .I flags | |
08e325e8 | 141 | argument is formed by ORing together zero or more of the following values: |
f2b1d720 | 142 | .TP |
60dafbc1 MK |
143 | .BR PERF_FLAG_FD_CLOEXEC " (since Linux 3.14)" |
144 | .\" commit a21b0b354d4ac39be691f51c53562e2c24443d9e | |
e9b1ab78 MK |
145 | This flag enables the close-on-exec flag for the created |
146 | event file descriptor, | |
147 | so that the file descriptor is automatically closed on | |
148 | .BR execve (2). | |
8bad22e5 MK |
149 | Setting the close-on-exec flags at creation time, rather than later with |
150 | .BR fcntl (2), | |
e9b1ab78 MK |
151 | avoids potential race conditions where the calling thread invokes |
152 | .BR perf_event_open () | |
a61dba34 MK |
153 | and |
154 | .BR fcntl (2) | |
e9b1ab78 MK |
155 | at the same time as another thread calls |
156 | .BR fork (2) | |
157 | then | |
158 | .BR execve (2). | |
159 | .TP | |
f2b1d720 | 160 | .BR PERF_FLAG_FD_NO_GROUP |
31266c04 VW |
161 | This flag tells the event to ignore the |
162 | .IR group_fd | |
163 | parameter except for the purpose of setting up output redirection | |
164 | using the | |
165 | .B PERF_FLAG_FD_OUTPUT | |
166 | flag. | |
f2b1d720 | 167 | .TP |
3117263f | 168 | .BR PERF_FLAG_FD_OUTPUT " (broken since Linux 2.6.35)" |
747a6e7c | 169 | .\" commit ac9721f3f54b27a16c7e1afb2481e7ee95a70318 |
31266c04 VW |
170 | This flag re-routes the event's sampled output to instead |
171 | be included in the mmap buffer of the event specified by | |
172 | .IR group_fd . | |
f2b1d720 | 173 | .TP |
3117263f | 174 | .BR PERF_FLAG_PID_CGROUP " (since Linux 2.6.39)" |
60dafbc1 | 175 | .\" commit e5d1367f17ba6a6fed5fd8b74e4d5720923e0c25 |
f2b1d720 MK |
176 | This flag activates per-container system-wide monitoring. |
177 | A container | |
ce88f77b | 178 | is an abstraction that isolates a set of resources for finer-grained |
699893d8 | 179 | control (CPUs, memory, etc.). |
f2b1d720 MK |
180 | In this mode, the event is measured |
181 | only if the thread running on the monitored CPU belongs to the designated | |
182 | container (cgroup). | |
183 | The cgroup is identified by passing a file descriptor | |
184 | opened on its directory in the cgroupfs filesystem. | |
185 | For instance, if the | |
186 | cgroup to monitor is called | |
187 | .IR test , | |
188 | then a file descriptor opened on | |
189 | .I /dev/cgroup/test | |
190 | (assuming cgroupfs is mounted on | |
191 | .IR /dev/cgroup ) | |
192 | must be passed as the | |
193 | .I pid | |
194 | parameter. | |
33a0ccb2 | 195 | cgroup monitoring is available only |
f2b1d720 MK |
196 | for system-wide events and may therefore require extra permissions. |
197 | .P | |
198 | The | |
199 | .I perf_event_attr | |
200 | structure provides detailed configuration information | |
201 | for the event being created. | |
202 | ||
203 | .in +4n | |
204 | .nf | |
205 | struct perf_event_attr { | |
da8bd8a4 MK |
206 | __u32 type; /* Type of event */ |
207 | __u32 size; /* Size of attribute structure */ | |
208 | __u64 config; /* Type-specific configuration */ | |
f2b1d720 MK |
209 | |
210 | union { | |
211 | __u64 sample_period; /* Period of sampling */ | |
212 | __u64 sample_freq; /* Frequency of sampling */ | |
213 | }; | |
214 | ||
ce88f77b MK |
215 | __u64 sample_type; /* Specifies values included in sample */ |
216 | __u64 read_format; /* Specifies values returned in read */ | |
217 | ||
218 | __u64 disabled : 1, /* off by default */ | |
219 | inherit : 1, /* children inherit it */ | |
220 | pinned : 1, /* must always be on PMU */ | |
221 | exclusive : 1, /* only group on PMU */ | |
222 | exclude_user : 1, /* don't count user */ | |
223 | exclude_kernel : 1, /* don't count kernel */ | |
224 | exclude_hv : 1, /* don't count hypervisor */ | |
225 | exclude_idle : 1, /* don't count when idle */ | |
226 | mmap : 1, /* include mmap data */ | |
227 | comm : 1, /* include comm data */ | |
228 | freq : 1, /* use freq, not period */ | |
229 | inherit_stat : 1, /* per task counts */ | |
230 | enable_on_exec : 1, /* next exec enables */ | |
231 | task : 1, /* trace fork/exit */ | |
232 | watermark : 1, /* wakeup_watermark */ | |
233 | precise_ip : 2, /* skid constraint */ | |
234 | mmap_data : 1, /* non-exec mmap data */ | |
235 | sample_id_all : 1, /* sample_type all events */ | |
236 | exclude_host : 1, /* don't count in host */ | |
237 | exclude_guest : 1, /* don't count in guest */ | |
238 | exclude_callchain_kernel : 1, | |
239 | /* exclude kernel callchains */ | |
240 | exclude_callchain_user : 1, | |
241 | /* exclude user callchains */ | |
9bfc542b | 242 | mmap2 : 1, /* include mmap with inode data */ |
dc9ec146 MK |
243 | comm_exec : 1, /* flag comm events that are |
244 | due to exec */ | |
6bd5186a | 245 | use_clockid : 1, /* use clockid for time fields */ |
9277a75d | 246 | context_switch : 1, /* context switch data */ |
6bd5186a | 247 | |
9277a75d | 248 | __reserved_1 : 37; |
f2b1d720 MK |
249 | |
250 | union { | |
251 | __u32 wakeup_events; /* wakeup every n events */ | |
7db515ef | 252 | __u32 wakeup_watermark; /* bytes before wakeup */ |
f2b1d720 MK |
253 | }; |
254 | ||
255 | __u32 bp_type; /* breakpoint type */ | |
256 | ||
257 | union { | |
258 | __u64 bp_addr; /* breakpoint address */ | |
259 | __u64 config1; /* extension of config */ | |
260 | }; | |
261 | ||
262 | union { | |
263 | __u64 bp_len; /* breakpoint length */ | |
264 | __u64 config2; /* extension of config1 */ | |
265 | }; | |
ce88f77b MK |
266 | __u64 branch_sample_type; /* enum perf_branch_sample_type */ |
267 | __u64 sample_regs_user; /* user regs to dump on samples */ | |
268 | __u32 sample_stack_user; /* size of stack to dump on | |
7db515ef | 269 | samples */ |
6bd5186a | 270 | __s32 clockid; /* clock to use for time fields */ |
f5281dfd | 271 | __u64 sample_regs_intr; /* regs to dump on samples */ |
cdc52f4a | 272 | __u32 aux_watermark; /* aux bytes before wakeup */ |
fd133d5d VW |
273 | __u16 sample_max_stack; /* max frames in callchain */ |
274 | __u16 __reserved_2; /* align to u64 */ | |
cdc52f4a | 275 | |
f2b1d720 MK |
276 | }; |
277 | .fi | |
278 | .in | |
279 | ||
280 | The fields of the | |
281 | .I perf_event_attr | |
282 | structure are described in more detail below: | |
f2b1d720 MK |
283 | .TP |
284 | .I type | |
285 | This field specifies the overall event type. | |
286 | It has one of the following values: | |
287 | .RS | |
288 | .TP | |
289 | .B PERF_TYPE_HARDWARE | |
290 | This indicates one of the "generalized" hardware events provided | |
291 | by the kernel. | |
292 | See the | |
293 | .I config | |
294 | field definition for more details. | |
295 | .TP | |
296 | .B PERF_TYPE_SOFTWARE | |
297 | This indicates one of the software-defined events provided by the kernel | |
298 | (even if no hardware support is available). | |
299 | .TP | |
300 | .B PERF_TYPE_TRACEPOINT | |
301 | This indicates a tracepoint | |
302 | provided by the kernel tracepoint infrastructure. | |
303 | .TP | |
304 | .B PERF_TYPE_HW_CACHE | |
305 | This indicates a hardware cache event. | |
306 | This has a special encoding, described in the | |
307 | .I config | |
308 | field definition. | |
309 | .TP | |
310 | .B PERF_TYPE_RAW | |
311 | This indicates a "raw" implementation-specific event in the | |
312 | .IR config " field." | |
313 | .TP | |
31c1f2b0 | 314 | .BR PERF_TYPE_BREAKPOINT " (since Linux 2.6.33)" |
60dafbc1 | 315 | .\" commit 24f1e32c60c45c89a997c73395b69c8af6f0a84e |
f2b1d720 MK |
316 | This indicates a hardware breakpoint as provided by the CPU. |
317 | Breakpoints can be read/write accesses to an address as well as | |
318 | execution of an instruction address. | |
319 | .TP | |
320 | .RB "dynamic PMU" | |
747a6e7c VW |
321 | Since Linux 2.6.38, |
322 | .\" commit 2e80a82a49c4c7eca4e35734380f28298ba5db19 | |
7db515ef | 323 | .BR perf_event_open () |
f2b1d720 MK |
324 | can support multiple PMUs. |
325 | To enable this, a value exported by the kernel can be used in the | |
326 | .I type | |
327 | field to indicate which PMU to use. | |
328 | The value to use can be found in the sysfs filesystem: | |
329 | there is a subdirectory per PMU instance under | |
330 | .IR /sys/bus/event_source/devices . | |
7d182bb6 | 331 | In each subdirectory there is a |
f2b1d720 MK |
332 | .I type |
333 | file whose content is an integer that can be used in the | |
334 | .I type | |
335 | field. | |
336 | For instance, | |
337 | .I /sys/bus/event_source/devices/cpu/type | |
338 | contains the value for the core CPU PMU, which is usually 4. | |
339 | .RE | |
f2b1d720 MK |
340 | .TP |
341 | .I "size" | |
342 | The size of the | |
343 | .I perf_event_attr | |
344 | structure for forward/backward compatibility. | |
345 | Set this using | |
346 | .I sizeof(struct perf_event_attr) | |
347 | to allow the kernel to see | |
348 | the struct size at the time of compilation. | |
349 | ||
350 | The related define | |
351 | .B PERF_ATTR_SIZE_VER0 | |
352 | is set to 64; this was the size of the first published struct. | |
353 | .B PERF_ATTR_SIZE_VER1 | |
354 | is 72, corresponding to the addition of breakpoints in Linux 2.6.33. | |
747a6e7c VW |
355 | .\" commit cb5d76999029ae7a517cb07dfa732c1b5a934fc2 |
356 | .\" this was added much later when PERF_ATTR_SIZE_VER2 happened | |
357 | .\" but the actual attr_size had increased in 2.6.33 | |
f2b1d720 MK |
358 | .B PERF_ATTR_SIZE_VER2 |
359 | is 80 corresponding to the addition of branch sampling in Linux 3.4. | |
747a6e7c | 360 | .\" commit cb5d76999029ae7a517cb07dfa732c1b5a934fc2 |
d2a6be2f | 361 | .B PERF_ATTR_SIZE_VER3 |
f2b1d720 | 362 | is 96 corresponding to the addition |
7ede2f66 DP |
363 | of |
364 | .I sample_regs_user | |
365 | and | |
366 | .I sample_stack_user | |
367 | in Linux 3.7. | |
747a6e7c | 368 | .\" commit 1659d129ed014b715b0b2120e6fd929bdd33ed03 |
f5281dfd VW |
369 | .B PERF_ATTR_SIZE_VER4 |
370 | is 104 corresponding to the addition of | |
371 | .I sample_regs_intr | |
372 | in Linux 3.19. | |
373 | .\" commit 60e2364e60e86e81bc6377f49779779e6120977f | |
cdc52f4a VW |
374 | .B PERF_ATTR_SIZE_VER5 |
375 | is 112 corresponding to the addition of | |
2050c098 | 376 | .I aux_watermark |
cdc52f4a VW |
377 | in Linux 4.1. |
378 | .\" commit 1a5941312414c71dece6717da9a0fa1303127afa | |
f2b1d720 MK |
379 | .TP |
380 | .I "config" | |
381 | This specifies which event you want, in conjunction with | |
382 | the | |
383 | .I type | |
384 | field. | |
385 | The | |
386 | .IR config1 " and " config2 | |
387 | fields are also taken into account in cases where 64 bits is not | |
388 | enough to fully specify the event. | |
389 | The encoding of these fields are event dependent. | |
390 | ||
f2b1d720 MK |
391 | There are various ways to set the |
392 | .I config | |
393 | field that are dependent on the value of the previously | |
394 | described | |
395 | .I type | |
396 | field. | |
397 | What follows are various possible settings for | |
398 | .I config | |
399 | separated out by | |
400 | .IR type . | |
401 | ||
402 | If | |
403 | .I type | |
404 | is | |
405 | .BR PERF_TYPE_HARDWARE , | |
406 | we are measuring one of the generalized hardware CPU events. | |
407 | Not all of these are available on all platforms. | |
408 | Set | |
409 | .I config | |
410 | to one of the following: | |
411 | .RS 12 | |
412 | .TP | |
413 | .B PERF_COUNT_HW_CPU_CYCLES | |
414 | Total cycles. | |
2b538c3e | 415 | Be wary of what happens during CPU frequency scaling. |
f2b1d720 MK |
416 | .TP |
417 | .B PERF_COUNT_HW_INSTRUCTIONS | |
418 | Retired instructions. | |
419 | Be careful, these can be affected by various | |
2b538c3e | 420 | issues, most notably hardware interrupt counts. |
f2b1d720 MK |
421 | .TP |
422 | .B PERF_COUNT_HW_CACHE_REFERENCES | |
423 | Cache accesses. | |
424 | Usually this indicates Last Level Cache accesses but this may | |
425 | vary depending on your CPU. | |
426 | This may include prefetches and coherency messages; again this | |
427 | depends on the design of your CPU. | |
428 | .TP | |
429 | .B PERF_COUNT_HW_CACHE_MISSES | |
430 | Cache misses. | |
431 | Usually this indicates Last Level Cache misses; this is intended to be | |
432 | used in conjunction with the | |
433 | .B PERF_COUNT_HW_CACHE_REFERENCES | |
434 | event to calculate cache miss rates. | |
435 | .TP | |
436 | .B PERF_COUNT_HW_BRANCH_INSTRUCTIONS | |
437 | Retired branch instructions. | |
747a6e7c | 438 | Prior to Linux 2.6.35, this used |
f2b1d720 | 439 | the wrong event on AMD processors. |
747a6e7c | 440 | .\" commit f287d332ce835f77a4f5077d2c0ef1e3f9ea42d2 |
f2b1d720 MK |
441 | .TP |
442 | .B PERF_COUNT_HW_BRANCH_MISSES | |
443 | Mispredicted branch instructions. | |
444 | .TP | |
445 | .B PERF_COUNT_HW_BUS_CYCLES | |
446 | Bus cycles, which can be different from total cycles. | |
447 | .TP | |
31c1f2b0 | 448 | .BR PERF_COUNT_HW_STALLED_CYCLES_FRONTEND " (since Linux 3.0)" |
747a6e7c | 449 | .\" commit 8f62242246351b5a4bc0c1f00c0c7003edea128a |
f2b1d720 MK |
450 | Stalled cycles during issue. |
451 | .TP | |
31c1f2b0 | 452 | .BR PERF_COUNT_HW_STALLED_CYCLES_BACKEND " (since Linux 3.0)" |
747a6e7c | 453 | .\" commit 8f62242246351b5a4bc0c1f00c0c7003edea128a |
f2b1d720 MK |
454 | Stalled cycles during retirement. |
455 | .TP | |
31c1f2b0 | 456 | .BR PERF_COUNT_HW_REF_CPU_CYCLES " (since Linux 3.3)" |
60dafbc1 | 457 | .\" commit c37e17497e01fc0f5d2d6feb5723b210b3ab8890 |
f2b1d720 MK |
458 | Total cycles; not affected by CPU frequency scaling. |
459 | .RE | |
460 | .IP | |
461 | If | |
462 | .I type | |
463 | is | |
464 | .BR PERF_TYPE_SOFTWARE , | |
465 | we are measuring software events provided by the kernel. | |
466 | Set | |
467 | .I config | |
468 | to one of the following: | |
469 | .RS 12 | |
470 | .TP | |
471 | .B PERF_COUNT_SW_CPU_CLOCK | |
472 | This reports the CPU clock, a high-resolution per-CPU timer. | |
473 | .TP | |
474 | .B PERF_COUNT_SW_TASK_CLOCK | |
475 | This reports a clock count specific to the task that is running. | |
476 | .TP | |
477 | .B PERF_COUNT_SW_PAGE_FAULTS | |
478 | This reports the number of page faults. | |
479 | .TP | |
480 | .B PERF_COUNT_SW_CONTEXT_SWITCHES | |
481 | This counts context switches. | |
482 | Until Linux 2.6.34, these were all reported as user-space | |
483 | events, after that they are reported as happening in the kernel. | |
747a6e7c | 484 | .\" commit e49a5bd38159dfb1928fd25b173bc9de4bbadb21 |
f2b1d720 MK |
485 | .TP |
486 | .B PERF_COUNT_SW_CPU_MIGRATIONS | |
487 | This reports the number of times the process | |
488 | has migrated to a new CPU. | |
489 | .TP | |
490 | .B PERF_COUNT_SW_PAGE_FAULTS_MIN | |
491 | This counts the number of minor page faults. | |
492 | These did not require disk I/O to handle. | |
493 | .TP | |
494 | .B PERF_COUNT_SW_PAGE_FAULTS_MAJ | |
495 | This counts the number of major page faults. | |
496 | These required disk I/O to handle. | |
497 | .TP | |
31c1f2b0 | 498 | .BR PERF_COUNT_SW_ALIGNMENT_FAULTS " (since Linux 2.6.33)" |
60dafbc1 | 499 | .\" commit f7d7986060b2890fc26db6ab5203efbd33aa2497 |
f2b1d720 MK |
500 | This counts the number of alignment faults. |
501 | These happen when unaligned memory accesses happen; the kernel | |
502 | can handle these but it reduces performance. | |
33a0ccb2 | 503 | This happens only on some architectures (never on x86). |
f2b1d720 | 504 | .TP |
31c1f2b0 | 505 | .BR PERF_COUNT_SW_EMULATION_FAULTS " (since Linux 2.6.33)" |
60dafbc1 | 506 | .\" commit f7d7986060b2890fc26db6ab5203efbd33aa2497 |
f2b1d720 MK |
507 | This counts the number of emulation faults. |
508 | The kernel sometimes traps on unimplemented instructions | |
7db515ef | 509 | and emulates them for user space. |
f2b1d720 | 510 | This can negatively impact performance. |
dab38455 | 511 | .TP |
31c1f2b0 | 512 | .BR PERF_COUNT_SW_DUMMY " (since Linux 3.12)" |
60dafbc1 | 513 | .\" commit fa0097ee690693006ab1aea6c01ad3c851b65c77 |
dab38455 VW |
514 | This is a placeholder event that counts nothing. |
515 | Informational sample record types such as mmap or comm | |
516 | must be associated with an active event. | |
517 | This dummy event allows gathering such records without requiring | |
518 | a counting event. | |
f2b1d720 | 519 | .RE |
f2b1d720 | 520 | |
f2b1d720 MK |
521 | .RS |
522 | If | |
523 | .I type | |
524 | is | |
525 | .BR PERF_TYPE_TRACEPOINT , | |
526 | then we are measuring kernel tracepoints. | |
527 | The value to use in | |
528 | .I config | |
529 | can be obtained from under debugfs | |
530 | .I tracing/events/*/*/id | |
531 | if ftrace is enabled in the kernel. | |
f2b1d720 | 532 | .RE |
1f22e274 | 533 | |
f2b1d720 MK |
534 | .RS |
535 | If | |
536 | .I type | |
537 | is | |
538 | .BR PERF_TYPE_HW_CACHE , | |
539 | then we are measuring a hardware CPU cache event. | |
540 | To calculate the appropriate | |
541 | .I config | |
542 | value use the following equation: | |
543 | .RS 4 | |
544 | .nf | |
545 | ||
546 | (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) | | |
547 | (perf_hw_cache_op_result_id << 16) | |
548 | .fi | |
549 | .P | |
550 | where | |
551 | .I perf_hw_cache_id | |
552 | is one of: | |
7db515ef | 553 | .RS 4 |
f2b1d720 MK |
554 | .TP |
555 | .B PERF_COUNT_HW_CACHE_L1D | |
556 | for measuring Level 1 Data Cache | |
557 | .TP | |
558 | .B PERF_COUNT_HW_CACHE_L1I | |
559 | for measuring Level 1 Instruction Cache | |
560 | .TP | |
561 | .B PERF_COUNT_HW_CACHE_LL | |
562 | for measuring Last-Level Cache | |
563 | .TP | |
564 | .B PERF_COUNT_HW_CACHE_DTLB | |
565 | for measuring the Data TLB | |
566 | .TP | |
567 | .B PERF_COUNT_HW_CACHE_ITLB | |
568 | for measuring the Instruction TLB | |
569 | .TP | |
570 | .B PERF_COUNT_HW_CACHE_BPU | |
571 | for measuring the branch prediction unit | |
572 | .TP | |
5a69ce9c MK |
573 | .BR PERF_COUNT_HW_CACHE_NODE " (since Linux 3.1)" |
574 | .\" commit 89d6c0b5bdbb1927775584dcf532d98b3efe1477 | |
f2b1d720 MK |
575 | for measuring local memory accesses |
576 | .RE | |
f2b1d720 MK |
577 | .P |
578 | and | |
579 | .I perf_hw_cache_op_id | |
4af27572 | 580 | is one of: |
7db515ef | 581 | .RS 4 |
f2b1d720 MK |
582 | .TP |
583 | .B PERF_COUNT_HW_CACHE_OP_READ | |
584 | for read accesses | |
585 | .TP | |
586 | .B PERF_COUNT_HW_CACHE_OP_WRITE | |
587 | for write accesses | |
588 | .TP | |
589 | .B PERF_COUNT_HW_CACHE_OP_PREFETCH | |
590 | for prefetch accesses | |
591 | .RE | |
f2b1d720 MK |
592 | .P |
593 | and | |
594 | .I perf_hw_cache_op_result_id | |
4af27572 | 595 | is one of: |
7db515ef | 596 | .RS 4 |
f2b1d720 MK |
597 | .TP |
598 | .B PERF_COUNT_HW_CACHE_RESULT_ACCESS | |
599 | to measure accesses | |
600 | .TP | |
601 | .B PERF_COUNT_HW_CACHE_RESULT_MISS | |
602 | to measure misses | |
603 | .RE | |
604 | .RE | |
605 | ||
606 | If | |
607 | .I type | |
608 | is | |
609 | .BR PERF_TYPE_RAW , | |
610 | then a custom "raw" | |
611 | .I config | |
612 | value is needed. | |
613 | Most CPUs support events that are not covered by the "generalized" events. | |
614 | These are implementation defined; see your CPU manual (for example | |
615 | the Intel Volume 3B documentation or the AMD BIOS and Kernel Developer | |
616 | Guide). | |
617 | The libpfm4 library can be used to translate from the name in the | |
618 | architectural manuals to the raw hex value | |
619 | .BR perf_event_open () | |
620 | expects in this field. | |
621 | ||
622 | If | |
623 | .I type | |
624 | is | |
625 | .BR PERF_TYPE_BREAKPOINT , | |
626 | then leave | |
627 | .I config | |
628 | set to zero. | |
629 | Its parameters are set in other places. | |
630 | .RE | |
631 | .TP | |
632 | .IR sample_period ", " sample_freq | |
21977c9d | 633 | A "sampling" event is one that generates an overflow notification |
f2b1d720 MK |
634 | every N events, where N is given by |
635 | .IR sample_period . | |
21977c9d | 636 | A sampling event has |
f2b1d720 | 637 | .IR sample_period " > 0." |
21977c9d | 638 | When an overflow occurs, requested data is recorded |
f2b1d720 MK |
639 | in the mmap buffer. |
640 | The | |
641 | .I sample_type | |
21977c9d | 642 | field controls what data is recorded on each overflow. |
f2b1d720 MK |
643 | |
644 | .I sample_freq | |
645 | can be used if you wish to use frequency rather than period. | |
37bee118 | 646 | In this case, you set the |
f2b1d720 MK |
647 | .I freq |
648 | flag. | |
649 | The kernel will adjust the sampling period | |
650 | to try and achieve the desired rate. | |
651 | The rate of adjustment is a | |
652 | timer tick. | |
f2b1d720 MK |
653 | .TP |
654 | .I "sample_type" | |
655 | The various bits in this field specify which values to include | |
656 | in the sample. | |
657 | They will be recorded in a ring-buffer, | |
ad73a2cc | 658 | which is available to user space using |
f2b1d720 MK |
659 | .BR mmap (2). |
660 | The order in which the values are saved in the | |
661 | sample are documented in the MMAP Layout subsection below; | |
662 | it is not the | |
663 | .I "enum perf_event_sample_format" | |
664 | order. | |
665 | .RS | |
666 | .TP | |
667 | .B PERF_SAMPLE_IP | |
668 | Records instruction pointer. | |
669 | .TP | |
670 | .B PERF_SAMPLE_TID | |
7db515ef | 671 | Records the process and thread IDs. |
f2b1d720 MK |
672 | .TP |
673 | .B PERF_SAMPLE_TIME | |
674 | Records a timestamp. | |
675 | .TP | |
676 | .B PERF_SAMPLE_ADDR | |
677 | Records an address, if applicable. | |
678 | .TP | |
679 | .B PERF_SAMPLE_READ | |
680 | Record counter values for all events in a group, not just the group leader. | |
681 | .TP | |
682 | .B PERF_SAMPLE_CALLCHAIN | |
683 | Records the callchain (stack backtrace). | |
684 | .TP | |
685 | .B PERF_SAMPLE_ID | |
686 | Records a unique ID for the opened event's group leader. | |
687 | .TP | |
688 | .B PERF_SAMPLE_CPU | |
689 | Records CPU number. | |
690 | .TP | |
691 | .B PERF_SAMPLE_PERIOD | |
692 | Records the current sampling period. | |
693 | .TP | |
694 | .B PERF_SAMPLE_STREAM_ID | |
695 | Records a unique ID for the opened event. | |
696 | Unlike | |
697 | .B PERF_SAMPLE_ID | |
698 | the actual ID is returned, not the group leader. | |
8859d3a9 DP |
699 | This ID is the same as the one returned by |
700 | .BR PERF_FORMAT_ID . | |
f2b1d720 MK |
701 | .TP |
702 | .B PERF_SAMPLE_RAW | |
703 | Records additional data, if applicable. | |
704 | Usually returned by tracepoint events. | |
705 | .TP | |
31c1f2b0 | 706 | .BR PERF_SAMPLE_BRANCH_STACK " (since Linux 3.4)" |
60dafbc1 | 707 | .\" commit bce38cd53e5ddba9cb6d708c4ef3d04a4016ec7e |
045bf4d3 VW |
708 | This provides a record of recent branches, as provided |
709 | by CPU branch sampling hardware (such as Intel Last Branch Record). | |
710 | Not all hardware supports this feature. | |
711 | ||
712 | See the | |
713 | .I branch_sample_type | |
714 | field for how to filter which branches are reported. | |
f2b1d720 | 715 | .TP |
31c1f2b0 | 716 | .BR PERF_SAMPLE_REGS_USER " (since Linux 3.7)" |
60dafbc1 | 717 | .\" commit 4018994f3d8785275ef0e7391b75c3462c029e56 |
d1007d14 VW |
718 | Records the current user-level CPU register state |
719 | (the values in the process before the kernel was called). | |
f2b1d720 | 720 | .TP |
31c1f2b0 | 721 | .BR PERF_SAMPLE_STACK_USER " (since Linux 3.7)" |
60dafbc1 | 722 | .\" commit c5ebcedb566ef17bda7b02686e0d658a7bb42ee7 |
d1007d14 VW |
723 | Records the user level stack, allowing stack unwinding. |
724 | .TP | |
31c1f2b0 | 725 | .BR PERF_SAMPLE_WEIGHT " (since Linux 3.10)" |
60dafbc1 | 726 | .\" commit c3feedf2aaf9ac8bad6f19f5d21e4ee0b4b87e9c |
d1007d14 | 727 | Records a hardware provided weight value that expresses how |
51700fd7 | 728 | costly the sampled event was. |
d1007d14 VW |
729 | This allows the hardware to highlight expensive events in |
730 | a profile. | |
731 | .TP | |
31c1f2b0 | 732 | .BR PERF_SAMPLE_DATA_SRC " (since Linux 3.10)" |
60dafbc1 | 733 | .\" commit d6be9ad6c960f43800a6f118932bc8a5a4eadcd1 |
d1007d14 VW |
734 | Records the data source: where in the memory hierarchy |
735 | the data associated with the sampled instruction came from. | |
6170255e | 736 | This is available only if the underlying hardware |
d1007d14 | 737 | supports this feature. |
7480dabb | 738 | .TP |
31c1f2b0 | 739 | .BR PERF_SAMPLE_IDENTIFIER " (since Linux 3.12)" |
60dafbc1 | 740 | .\" commit ff3d527cebc1fa3707c617bfe9e74f53fcfb0955 |
8859d3a9 DP |
741 | Places the |
742 | .B SAMPLE_ID | |
743 | value in a fixed position in the record, | |
7480dabb VW |
744 | either at the beginning (for sample events) or at the end |
745 | (if a non-sample event). | |
746 | ||
747 | This was necessary because a sample stream may have | |
748 | records from various different event sources with different | |
749 | .I sample_type | |
750 | settings. | |
e9bd9b2c | 751 | Parsing the event stream properly was not possible because the |
8859d3a9 DP |
752 | format of the record was needed to find |
753 | .BR SAMPLE_ID , | |
754 | but | |
27f52b52 | 755 | the format could not be found without knowing what |
7480dabb VW |
756 | event the sample belonged to (causing a circular |
757 | dependency). | |
758 | ||
e41c36b2 | 759 | The |
7480dabb VW |
760 | .B PERF_SAMPLE_IDENTIFIER |
761 | setting makes the event stream always parsable | |
8859d3a9 DP |
762 | by putting |
763 | .B SAMPLE_ID | |
764 | in a fixed location, even though | |
765 | it means having duplicate | |
766 | .B SAMPLE_ID | |
767 | values in records. | |
1e043959 | 768 | .TP |
60dafbc1 MK |
769 | .BR PERF_SAMPLE_TRANSACTION " (since Linux 3.13)" |
770 | .\" commit fdfbbd07e91f8fe387140776f3fd94605f0c89e5 | |
84fc2a6e | 771 | Records reasons for transactional memory abort events |
1e043959 VW |
772 | (for example, from Intel TSX transactional memory support). |
773 | ||
774 | The | |
775 | .I precise_ip | |
b3f39642 | 776 | setting must be greater than 0 and a transactional memory abort |
1e043959 | 777 | event must be measured or no values will be recorded. |
84fc2a6e MK |
778 | Also note that some perf_event measurements, such as sampled |
779 | cycle counting, may cause extraneous aborts (by causing an | |
1e043959 | 780 | interrupt during a transaction). |
f5281dfd VW |
781 | .TP |
782 | .BR PERF_SAMPLE_REGS_INTR " (since Linux 3.19)" | |
783 | .\" commit 60e2364e60e86e81bc6377f49779779e6120977f | |
784 | Records a subset of the current CPU register state | |
785 | as specified by | |
786 | .IR sample_regs_intr . | |
787 | Unlike | |
788 | .B PERF_SAMPLE_REGS_USER | |
789 | the register values will return kernel register | |
790 | state if the overflow happened while kernel | |
791 | code is running. | |
792 | If the CPU supports hardware sampling of | |
b01ae37b | 793 | register state (i.e., PEBS on Intel x86) and |
f5281dfd VW |
794 | .I precise_ip |
795 | is set higher than zero then the register | |
796 | values returned are those captured by | |
797 | hardware at the time of the sampled | |
798 | instruction's retirement. | |
f2b1d720 | 799 | .RE |
f2b1d720 MK |
800 | .TP |
801 | .IR "read_format" | |
802 | This field specifies the format of the data returned by | |
803 | .BR read (2) | |
804 | on a | |
7db515ef | 805 | .BR perf_event_open () |
f2b1d720 MK |
806 | file descriptor. |
807 | .RS | |
808 | .TP | |
809 | .B PERF_FORMAT_TOTAL_TIME_ENABLED | |
7ede2f66 DP |
810 | Adds the 64-bit |
811 | .I time_enabled | |
812 | field. | |
f2b1d720 MK |
813 | This can be used to calculate estimated totals if |
814 | the PMU is overcommitted and multiplexing is happening. | |
815 | .TP | |
816 | .B PERF_FORMAT_TOTAL_TIME_RUNNING | |
7ede2f66 DP |
817 | Adds the 64-bit |
818 | .I time_running | |
819 | field. | |
f2b1d720 | 820 | This can be used to calculate estimated totals if |
3d1ee497 | 821 | the PMU is overcommitted and multiplexing is happening. |
f2b1d720 MK |
822 | .TP |
823 | .B PERF_FORMAT_ID | |
824 | Adds a 64-bit unique value that corresponds to the event group. | |
825 | .TP | |
826 | .B PERF_FORMAT_GROUP | |
827 | Allows all counter values in an event group to be read with one read. | |
828 | .RE | |
f2b1d720 MK |
829 | .TP |
830 | .IR "disabled" | |
831 | The | |
832 | .I disabled | |
833 | bit specifies whether the counter starts out disabled or enabled. | |
834 | If disabled, the event can later be enabled by | |
835 | .BR ioctl (2), | |
836 | .BR prctl (2), | |
837 | or | |
838 | .IR enable_on_exec . | |
406650db VW |
839 | |
840 | When creating an event group, typically the group leader is initialized | |
841 | with | |
842 | .I disabled | |
843 | set to 1 and any child events are initialized with | |
844 | .I disabled | |
845 | set to 0. | |
846 | Despite | |
847 | .I disabled | |
848 | being 0, the child events will not start until the group leader | |
849 | is enabled. | |
f2b1d720 MK |
850 | .TP |
851 | .IR "inherit" | |
852 | The | |
853 | .I inherit | |
854 | bit specifies that this counter should count events of child | |
855 | tasks as well as the task specified. | |
33a0ccb2 | 856 | This applies only to new children, not to any existing children at |
f2b1d720 MK |
857 | the time the counter is created (nor to any new children of |
858 | existing children). | |
859 | ||
860 | Inherit does not work for some combinations of | |
4b3a5f01 MK |
861 | .IR read_format |
862 | values, such as | |
f2b1d720 | 863 | .BR PERF_FORMAT_GROUP . |
f2b1d720 MK |
864 | .TP |
865 | .IR "pinned" | |
866 | The | |
867 | .I pinned | |
868 | bit specifies that the counter should always be on the CPU if at all | |
869 | possible. | |
33a0ccb2 | 870 | It applies only to hardware counters and only to group leaders. |
f2b1d720 MK |
871 | If a pinned counter cannot be put onto the CPU (e.g., because there are |
872 | not enough hardware counters or because of a conflict with some other | |
873 | event), then the counter goes into an 'error' state, where reads | |
874 | return end-of-file (i.e., | |
875 | .BR read (2) | |
876 | returns 0) until the counter is subsequently enabled or disabled. | |
f2b1d720 MK |
877 | .TP |
878 | .IR "exclusive" | |
879 | The | |
880 | .I exclusive | |
881 | bit specifies that when this counter's group is on the CPU, | |
882 | it should be the only group using the CPU's counters. | |
883 | In the future this may allow monitoring programs to | |
884 | support PMU features that need to run alone so that they do not | |
885 | disrupt other hardware counters. | |
bea10c8c VW |
886 | |
887 | Note that many unexpected situations may prevent events with the | |
888 | .I exclusive | |
d3532647 | 889 | bit set from ever running. |
bea10c8c | 890 | This includes any users running a system-wide |
d3532647 | 891 | measurement as well as any kernel use of the performance counters |
bea10c8c | 892 | (including the commonly enabled NMI Watchdog Timer interface). |
f2b1d720 MK |
893 | .TP |
894 | .IR "exclude_user" | |
ad73a2cc | 895 | If this bit is set, the count excludes events that happen in user space. |
f2b1d720 MK |
896 | .TP |
897 | .IR "exclude_kernel" | |
edb3e316 | 898 | If this bit is set, the count excludes events that happen in kernel space. |
f2b1d720 MK |
899 | .TP |
900 | .IR "exclude_hv" | |
901 | If this bit is set, the count excludes events that happen in the | |
902 | hypervisor. | |
903 | This is mainly for PMUs that have built-in support for handling this | |
904 | (such as POWER). | |
905 | Extra support is needed for handling hypervisor measurements on most | |
906 | machines. | |
f2b1d720 MK |
907 | .TP |
908 | .IR "exclude_idle" | |
909 | If set, don't count when the CPU is idle. | |
f2b1d720 MK |
910 | .TP |
911 | .IR "mmap" | |
912 | The | |
913 | .I mmap | |
75ee11e5 | 914 | bit enables generation of |
cd7c700a | 915 | .B PERF_RECORD_MMAP |
75ee11e5 VW |
916 | samples for every |
917 | .BR mmap (2) | |
918 | call that has | |
cd7c700a | 919 | .B PROT_EXEC |
75ee11e5 VW |
920 | set. |
921 | This allows tools to notice new executable code being mapped into | |
922 | a program (dynamic shared libraries for example) | |
923 | so that addresses can be mapped back to the original code. | |
f2b1d720 MK |
924 | .TP |
925 | .IR "comm" | |
926 | The | |
927 | .I comm | |
928 | bit enables tracking of process command name as modified by the | |
cd7c700a | 929 | .BR exec (2) |
f2b1d720 | 930 | and |
cd7c700a | 931 | .BR prctl (PR_SET_NAME) |
49bc411c VW |
932 | system calls as well as writing to |
933 | .IR /proc/self/comm . | |
790ee6d6 | 934 | If the |
49bc411c | 935 | .I comm_exec |
790ee6d6 | 936 | flag is also successfully set (possible since Linux 3.16), |
747a6e7c | 937 | .\" commit 82b897782d10fcc4930c9d4a15b175348fdd2871 |
49bc411c VW |
938 | then the misc flag |
939 | .B PERF_RECORD_MISC_COMM_EXEC | |
940 | can be used to differentiate the | |
941 | .BR exec (2) | |
942 | case from the others. | |
f2b1d720 MK |
943 | .TP |
944 | .IR "freq" | |
945 | If this bit is set, then | |
946 | .I sample_frequency | |
947 | not | |
948 | .I sample_period | |
949 | is used when setting up the sampling interval. | |
f2b1d720 MK |
950 | .TP |
951 | .IR "inherit_stat" | |
952 | This bit enables saving of event counts on context switch for | |
953 | inherited tasks. | |
33a0ccb2 | 954 | This is meaningful only if the |
f2b1d720 MK |
955 | .I inherit |
956 | field is set. | |
f2b1d720 MK |
957 | .TP |
958 | .IR "enable_on_exec" | |
959 | If this bit is set, a counter is automatically | |
960 | enabled after a call to | |
961 | .BR exec (2). | |
f2b1d720 MK |
962 | .TP |
963 | .IR "task" | |
964 | If this bit is set, then | |
965 | fork/exit notifications are included in the ring buffer. | |
f2b1d720 MK |
966 | .TP |
967 | .IR "watermark" | |
21977c9d | 968 | If set, have an overflow notification happen when we cross the |
f2b1d720 MK |
969 | .I wakeup_watermark |
970 | boundary. | |
21977c9d | 971 | Otherwise, overflow notifications happen after |
f2b1d720 MK |
972 | .I wakeup_events |
973 | samples. | |
f2b1d720 | 974 | .TP |
31c1f2b0 | 975 | .IR "precise_ip" " (since Linux 2.6.35)" |
747a6e7c | 976 | .\" commit ab608344bcbde4f55ec4cd911b686b0ce3eae076 |
f2b1d720 MK |
977 | This controls the amount of skid. |
978 | Skid is how many instructions | |
979 | execute between an event of interest happening and the kernel | |
980 | being able to stop and record the event. | |
981 | Smaller skid is | |
982 | better and allows more accurate reporting of which events | |
983 | correspond to which instructions, but hardware is often limited | |
984 | with how small this can be. | |
985 | ||
5d73bc3f | 986 | The possible values of this field are the following: |
f2b1d720 | 987 | .RS |
dc9ec146 | 988 | .IP 0 3 |
f2b1d720 | 989 | .B SAMPLE_IP |
2b538c3e | 990 | can have arbitrary skid. |
dc9ec146 | 991 | .IP 1 |
f2b1d720 | 992 | .B SAMPLE_IP |
2b538c3e | 993 | must have constant skid. |
dc9ec146 | 994 | .IP 2 |
f2b1d720 | 995 | .B SAMPLE_IP |
2b538c3e | 996 | requested to have 0 skid. |
dc9ec146 | 997 | .IP 3 |
f2b1d720 MK |
998 | .B SAMPLE_IP |
999 | must have 0 skid. | |
5d73bc3f | 1000 | See also the description of |
f2b1d720 MK |
1001 | .BR PERF_RECORD_MISC_EXACT_IP . |
1002 | .RE | |
f2b1d720 | 1003 | .TP |
31c1f2b0 | 1004 | .IR "mmap_data" " (since Linux 2.6.36)" |
747a6e7c | 1005 | .\" commit 3af9e859281bda7eb7c20b51879cf43aa788ac2e |
b01ae37b | 1006 | This is the counterpart of the |
f2b1d720 | 1007 | .I mmap |
75ee11e5 VW |
1008 | field. |
1009 | This enables generation of | |
cd7c700a | 1010 | .B PERF_RECORD_MMAP |
75ee11e5 VW |
1011 | samples for |
1012 | .BR mmap (2) | |
1013 | calls that do not have | |
cd7c700a | 1014 | .B PROT_EXEC |
75ee11e5 | 1015 | set (for example data and SysV shared memory). |
f2b1d720 | 1016 | .TP |
31c1f2b0 | 1017 | .IR "sample_id_all" " (since Linux 2.6.38)" |
747a6e7c | 1018 | .\" commit c980d1091810df13f21aabbce545fd98f545bbf7 |
7480dabb | 1019 | If set, then TID, TIME, ID, STREAM_ID, and CPU can |
f2b1d720 MK |
1020 | additionally be included in |
1021 | .RB non- PERF_RECORD_SAMPLE s | |
1022 | if the corresponding | |
1023 | .I sample_type | |
1024 | is selected. | |
7480dabb | 1025 | |
e9bd9b2c | 1026 | If |
7480dabb | 1027 | .B PERF_SAMPLE_IDENTIFIER |
37bee118 | 1028 | is specified, then an additional ID value is included |
7480dabb VW |
1029 | as the last value to ease parsing the record stream. |
1030 | This may lead to the | |
e9bd9b2c | 1031 | .I id |
7480dabb VW |
1032 | value appearing twice. |
1033 | ||
1034 | The layout is described by this pseudo-structure: | |
dc9ec146 | 1035 | |
7480dabb VW |
1036 | .in +4n |
1037 | .nf | |
1038 | struct sample_id { | |
26d5cd2f MK |
1039 | { u32 pid, tid; } /* if PERF_SAMPLE_TID set */ |
1040 | { u64 time; } /* if PERF_SAMPLE_TIME set */ | |
1041 | { u64 id; } /* if PERF_SAMPLE_ID set */ | |
7480dabb | 1042 | { u64 stream_id;} /* if PERF_SAMPLE_STREAM_ID set */ |
26d5cd2f | 1043 | { u32 cpu, res; } /* if PERF_SAMPLE_CPU set */ |
7480dabb VW |
1044 | { u64 id; } /* if PERF_SAMPLE_IDENTIFIER set */ |
1045 | }; | |
1046 | .fi | |
f2b1d720 | 1047 | .TP |
31c1f2b0 | 1048 | .IR "exclude_host" " (since Linux 3.2)" |
747a6e7c | 1049 | .\" commit a240f76165e6255384d4bdb8139895fac7988799 |
e38fb93e | 1050 | When conducting measurements that include processes running |
5d73bc3f MK |
1051 | VM instances (i.e., have executed a |
1052 | .B KVM_RUN | |
1053 | .BR ioctl (2)), | |
1054 | only measure events happening inside a guest instance. | |
e38fb93e VW |
1055 | This is only meaningful outside the guests; this setting does |
1056 | not change counts gathered inside of a guest. | |
34d4e61d | 1057 | Currently, this functionality is x86 only. |
f2b1d720 | 1058 | .TP |
31c1f2b0 | 1059 | .IR "exclude_guest" " (since Linux 3.2)" |
747a6e7c | 1060 | .\" commit a240f76165e6255384d4bdb8139895fac7988799 |
e38fb93e | 1061 | When conducting measurements that include processes running |
5d73bc3f MK |
1062 | VM instances (i.e., have executed a |
1063 | .B KVM_RUN | |
1064 | .BR ioctl (2)), | |
1065 | do not measure events happening inside guest instances. | |
e38fb93e VW |
1066 | This is only meaningful outside the guests; this setting does |
1067 | not change counts gathered inside of a guest. | |
34d4e61d | 1068 | Currently, this functionality is x86 only. |
f2b1d720 | 1069 | .TP |
31c1f2b0 | 1070 | .IR "exclude_callchain_kernel" " (since Linux 3.7)" |
747a6e7c | 1071 | .\" commit d077526485d5c9b12fe85d0b2b3b7041e6bc5f91 |
f2b1d720 | 1072 | Do not include kernel callchains. |
f2b1d720 | 1073 | .TP |
31c1f2b0 | 1074 | .IR "exclude_callchain_user" " (since Linux 3.7)" |
747a6e7c | 1075 | .\" commit d077526485d5c9b12fe85d0b2b3b7041e6bc5f91 |
f2b1d720 | 1076 | Do not include user callchains. |
f2b1d720 | 1077 | .TP |
9bfc542b | 1078 | .IR "mmap2" " (since Linux 3.16)" |
747a6e7c VW |
1079 | .\" commit 13d7a2410fa637f450a29ecb515ac318ee40c741 |
1080 | .\" This is tricky; was committed during 3.12 development | |
1081 | .\" but right before release was disabled. | |
1082 | .\" So while you could select mmap2 starting with 3.12 | |
1083 | .\" it did not work until 3.16 | |
1084 | .\" commit a5a5ba72843dd05f991184d6cb9a4471acce1005 | |
9bfc542b VW |
1085 | Generate an extended executable mmap record that contains enough |
1086 | additional information to uniquely identify shared mappings. | |
1087 | The | |
1088 | .I mmap | |
1089 | flag must also be set for this to work. | |
1090 | .TP | |
49bc411c | 1091 | .IR "comm_exec" " (since Linux 3.16)" |
747a6e7c | 1092 | .\" commit 82b897782d10fcc4930c9d4a15b175348fdd2871 |
5ab35ae5 | 1093 | This is purely a feature-detection flag, it does not change |
49bc411c | 1094 | kernel behavior. |
5ab35ae5 | 1095 | If this flag can successfully be set, then, when |
49bc411c | 1096 | .I comm |
5ab35ae5 | 1097 | is enabled, the |
49bc411c VW |
1098 | .B PERF_RECORD_MISC_COMM_EXEC |
1099 | flag will be set in the | |
1100 | .I misc | |
1101 | field of a comm record header if the rename event being | |
1102 | reported was caused by a call to | |
1103 | .BR exec (2). | |
1104 | This allows tools to distinguish between the various | |
1105 | types of process renaming. | |
1106 | .TP | |
6bd5186a VW |
1107 | .IR "use_clockid" " (since Linux 4.1)" |
1108 | .\" commit 34f439278cef7b1177f8ce24f9fc81dfc6221d3b | |
1109 | This allows selecting which internal Linux clock to use | |
1110 | when generating timestamps via the | |
1111 | .I clockid | |
1112 | field. | |
1113 | This can make it easier to correlate perf sample times with | |
1114 | timestamps generated by other tools. | |
1115 | .TP | |
9277a75d VW |
1116 | .IR "context_switch" " (since Linux 4.3)" |
1117 | .\" commit 45ac1403f564f411c6a383a2448688ba8dd705a4 | |
1118 | This enables the generation of | |
1119 | .B PERF_RECORD_SWITCH | |
1120 | records when a context switch occurs. | |
1121 | It also enables the generation of | |
1122 | .B PERF_RECORD_SWITCH_CPU_WIDE | |
d5a24378 | 1123 | records when sampling in CPU-wide mode. |
9277a75d VW |
1124 | This functionality is in addition to existing tracepoint and |
1125 | software events for measuring context switches. | |
54905b0f MK |
1126 | The advantage of this method is that it will give full |
1127 | information even with strict | |
9277a75d VW |
1128 | .I perf_event_paranoid |
1129 | settings. | |
1130 | .TP | |
f2b1d720 MK |
1131 | .IR "wakeup_events" ", " "wakeup_watermark" |
1132 | This union sets how many samples | |
1133 | .RI ( wakeup_events ) | |
1134 | or bytes | |
1135 | .RI ( wakeup_watermark ) | |
21977c9d | 1136 | happen before an overflow notification happens. |
f2b1d720 MK |
1137 | Which one is used is selected by the |
1138 | .I watermark | |
cb8a928f | 1139 | bit flag. |
751c0f1a VW |
1140 | |
1141 | .I wakeup_events | |
6170255e | 1142 | counts only |
751c0f1a | 1143 | .B PERF_RECORD_SAMPLE |
51700fd7 | 1144 | record types. |
21977c9d | 1145 | To receive overflow notification for all |
751c0f1a | 1146 | .B PERF_RECORD |
21977c9d | 1147 | types choose watermark and set |
751c0f1a VW |
1148 | .I wakeup_watermark |
1149 | to 1. | |
21977c9d | 1150 | |
fc79d996 | 1151 | Prior to Linux 3.0, setting |
747a6e7c | 1152 | .\" commit f506b3dc0ec454a16d40cab9ee5d75435b39dc50 |
21977c9d VW |
1153 | .I wakeup_events |
1154 | to 0 resulted in no overflow notifications; | |
1155 | more recent kernels treat 0 the same as 1. | |
f2b1d720 | 1156 | .TP |
31c1f2b0 | 1157 | .IR "bp_type" " (since Linux 2.6.33)" |
747a6e7c | 1158 | .\" commit 24f1e32c60c45c89a997c73395b69c8af6f0a84e |
f2b1d720 MK |
1159 | This chooses the breakpoint type. |
1160 | It is one of: | |
1161 | .RS | |
1162 | .TP | |
1163 | .BR HW_BREAKPOINT_EMPTY | |
2b538c3e | 1164 | No breakpoint. |
f2b1d720 MK |
1165 | .TP |
1166 | .BR HW_BREAKPOINT_R | |
2b538c3e | 1167 | Count when we read the memory location. |
f2b1d720 MK |
1168 | .TP |
1169 | .BR HW_BREAKPOINT_W | |
2b538c3e | 1170 | Count when we write the memory location. |
f2b1d720 MK |
1171 | .TP |
1172 | .BR HW_BREAKPOINT_RW | |
2b538c3e | 1173 | Count when we read or write the memory location. |
f2b1d720 MK |
1174 | .TP |
1175 | .BR HW_BREAKPOINT_X | |
2b538c3e | 1176 | Count when we execute code at the memory location. |
f2b1d720 | 1177 | .LP |
7db515ef | 1178 | The values can be combined via a bitwise or, but the |
f2b1d720 MK |
1179 | combination of |
1180 | .B HW_BREAKPOINT_R | |
1181 | or | |
1182 | .B HW_BREAKPOINT_W | |
1183 | with | |
1184 | .B HW_BREAKPOINT_X | |
1185 | is not allowed. | |
1186 | .RE | |
f2b1d720 | 1187 | .TP |
31c1f2b0 | 1188 | .IR "bp_addr" " (since Linux 2.6.33)" |
747a6e7c | 1189 | .\" commit 24f1e32c60c45c89a997c73395b69c8af6f0a84e |
5d73bc3f | 1190 | This is the address of the breakpoint. |
4b3a5f01 MK |
1191 | For execution breakpoints, this is the memory address of the instruction |
1192 | of interest; for read and write breakpoints, it is the memory address | |
f2b1d720 | 1193 | of the memory location of interest. |
f2b1d720 | 1194 | .TP |
31c1f2b0 | 1195 | .IR "config1" " (since Linux 2.6.39)" |
747a6e7c | 1196 | .\" commit a7e3ed1e470116c9d12c2f778431a481a6be8ab6 |
f2b1d720 MK |
1197 | .I config1 |
1198 | is used for setting events that need an extra register or otherwise | |
1199 | do not fit in the regular config field. | |
1200 | Raw OFFCORE_EVENTS on Nehalem/Westmere/SandyBridge use this field | |
4b3a5f01 | 1201 | on Linux 3.3 and later kernels. |
f2b1d720 | 1202 | .TP |
31c1f2b0 | 1203 | .IR "bp_len" " (since Linux 2.6.33)" |
747a6e7c | 1204 | .\" commit 24f1e32c60c45c89a997c73395b69c8af6f0a84e |
f2b1d720 MK |
1205 | .I bp_len |
1206 | is the length of the breakpoint being measured if | |
1207 | .I type | |
1208 | is | |
1209 | .BR PERF_TYPE_BREAKPOINT . | |
1210 | Options are | |
1211 | .BR HW_BREAKPOINT_LEN_1 , | |
1212 | .BR HW_BREAKPOINT_LEN_2 , | |
1213 | .BR HW_BREAKPOINT_LEN_4 , | |
4b3a5f01 | 1214 | and |
f2b1d720 MK |
1215 | .BR HW_BREAKPOINT_LEN_8 . |
1216 | For an execution breakpoint, set this to | |
1217 | .IR sizeof(long) . | |
f2b1d720 | 1218 | .TP |
31c1f2b0 | 1219 | .IR "config2" " (since Linux 2.6.39)" |
747a6e7c | 1220 | .\" commit a7e3ed1e470116c9d12c2f778431a481a6be8ab6 |
f2b1d720 MK |
1221 | .I config2 |
1222 | is a further extension of the | |
1223 | .I config1 | |
1224 | field. | |
f2b1d720 | 1225 | .TP |
31c1f2b0 | 1226 | .IR "branch_sample_type" " (since Linux 3.4)" |
747a6e7c | 1227 | .\" commit bce38cd53e5ddba9cb6d708c4ef3d04a4016ec7e |
8a94e783 | 1228 | If |
045bf4d3 VW |
1229 | .B PERF_SAMPLE_BRANCH_STACK |
1230 | is enabled, then this specifies what branches to include | |
1231 | in the branch record. | |
e3c9782b VW |
1232 | |
1233 | The first part of the value is the privilege level, which | |
4b3a5f01 | 1234 | is a combination of one of the values listed below. |
045bf4d3 VW |
1235 | If the user does not set privilege level explicitly, the kernel |
1236 | will use the event's privilege level. | |
1237 | Event and branch privilege levels do not have to match. | |
f2b1d720 MK |
1238 | .RS |
1239 | .TP | |
1240 | .B PERF_SAMPLE_BRANCH_USER | |
33d6e2c7 | 1241 | Branch target is in user space. |
f2b1d720 MK |
1242 | .TP |
1243 | .B PERF_SAMPLE_BRANCH_KERNEL | |
33d6e2c7 | 1244 | Branch target is in kernel space. |
f2b1d720 MK |
1245 | .TP |
1246 | .B PERF_SAMPLE_BRANCH_HV | |
33d6e2c7 | 1247 | Branch target is in hypervisor. |
e3c9782b VW |
1248 | .TP |
1249 | .B PERF_SAMPLE_BRANCH_PLM_ALL | |
1250 | A convenience value that is the three preceding values ORed together. | |
e3c9782b VW |
1251 | .P |
1252 | In addition to the privilege value, at least one or more of the | |
1253 | following bits must be set. | |
f2b1d720 MK |
1254 | .TP |
1255 | .B PERF_SAMPLE_BRANCH_ANY | |
33d6e2c7 | 1256 | Any branch type. |
f2b1d720 MK |
1257 | .TP |
1258 | .B PERF_SAMPLE_BRANCH_ANY_CALL | |
c6e5df74 | 1259 | Any call branch (includes direct calls, indirect calls, and far jumps). |
f2b1d720 | 1260 | .TP |
e3c9782b | 1261 | .B PERF_SAMPLE_BRANCH_IND_CALL |
33d6e2c7 | 1262 | Indirect calls. |
f2b1d720 | 1263 | .TP |
c6e5df74 VW |
1264 | .BR PERF_SAMPLE_BRANCH_CALL " (since Linux 4.4)" |
1265 | .\" commit c229bf9dc179d2023e185c0f705bdf68484c1e73 | |
1266 | Direct calls. | |
1267 | .TP | |
1268 | .B PERF_SAMPLE_BRANCH_ANY_RETURN | |
1269 | Any return branch. | |
1270 | .TP | |
dde354c9 VW |
1271 | .BR PERF_SAMPLE_BRANCH_IND_JUMP " (since Linux 4.2)" |
1272 | .\" commit c9fdfa14c3792c0160849c484e83aa57afd80ccc | |
1273 | Indirect jumps. | |
1274 | .TP | |
aea60aad | 1275 | .BR PERF_SAMPLE_BRANCH_COND " (since Linux 3.16)" |
60dafbc1 | 1276 | .\" commit bac52139f0b7ab31330e98fd87fc5a2664951050 |
aea60aad VW |
1277 | Conditional branches. |
1278 | .TP | |
31c1f2b0 | 1279 | .BR PERF_SAMPLE_BRANCH_ABORT_TX " (since Linux 3.11)" |
60dafbc1 | 1280 | .\" commit 135c5612c460f89657c4698fe2ea753f6f667963 |
33d6e2c7 | 1281 | Transactional memory aborts. |
e3c9782b | 1282 | .TP |
31c1f2b0 | 1283 | .BR PERF_SAMPLE_BRANCH_IN_TX " (since Linux 3.11)" |
60dafbc1 | 1284 | .\" commit 135c5612c460f89657c4698fe2ea753f6f667963 |
33d6e2c7 | 1285 | Branch in transactional memory transaction. |
e3c9782b | 1286 | .TP |
31c1f2b0 | 1287 | .BR PERF_SAMPLE_BRANCH_NO_TX " (since Linux 3.11)" |
60dafbc1 | 1288 | .\" commit 135c5612c460f89657c4698fe2ea753f6f667963 |
33d6e2c7 | 1289 | Branch not in transactional memory transaction. |
bb7e6ff0 VW |
1290 | .BR PERF_SAMPLE_BRANCH_CALL_STACK " (since Linux 4.1)" |
1291 | .\" commit 2c44b1936bb3b135a3fac8b3493394d42e51cf70 | |
95655a22 | 1292 | Branch is part of a hardware-generated call stack. |
bb7e6ff0 VW |
1293 | This requires hardware support, currently only found |
1294 | on Intel x86 Haswell or newer. | |
f2b1d720 | 1295 | .RE |
f2b1d720 | 1296 | .TP |
31c1f2b0 | 1297 | .IR "sample_regs_user" " (since Linux 3.7)" |
747a6e7c | 1298 | .\" commit 4018994f3d8785275ef0e7391b75c3462c029e56 |
4651e412 | 1299 | This bit mask defines the set of user CPU registers to dump on samples. |
76c637e1 | 1300 | The layout of the register mask is architecture-specific and |
4b3a5f01 | 1301 | is described in the kernel header file |
d1007d14 | 1302 | .IR arch/ARCH/include/uapi/asm/perf_regs.h . |
f2b1d720 | 1303 | .TP |
31c1f2b0 | 1304 | .IR "sample_stack_user" " (since Linux 3.7)" |
747a6e7c | 1305 | .\" commit c5ebcedb566ef17bda7b02686e0d658a7bb42ee7 |
d1007d14 VW |
1306 | This defines the size of the user stack to dump if |
1307 | .B PERF_SAMPLE_STACK_USER | |
1308 | is specified. | |
6bd5186a VW |
1309 | .TP |
1310 | .IR "clockid" " (since Linux 4.1)" | |
1311 | .\" commit 34f439278cef7b1177f8ce24f9fc81dfc6221d3b | |
1312 | If | |
1313 | .I use_clockid | |
1314 | is set, then this field selects which internal Linux timer to | |
1315 | use for timestamps. | |
1316 | The available timers are defined in | |
1317 | .IR linux/time.h , | |
1318 | with | |
95655a22 MK |
1319 | .BR CLOCK_MONOTONIC , |
1320 | .BR CLOCK_MONOTONIC_RAW , | |
1321 | .BR CLOCK_REALTIME , | |
1322 | .BR CLOCK_BOOTTIME , | |
1323 | and | |
1324 | .B CLOCK_TAI | |
6bd5186a | 1325 | currently supported. |
cdc52f4a VW |
1326 | .TP |
1327 | .IR "aux_watermark" " (since Linux 4.1)" | |
1328 | .\" commit 1a5941312414c71dece6717da9a0fa1303127afa | |
1329 | This specifies how much data is required to trigger a | |
1330 | .B PERF_RECORD_AUX | |
1331 | sample. | |
fd133d5d VW |
1332 | .TP |
1333 | .IR "sample_max_stack" " (since Linux 4.8)" | |
1334 | .\" commit 97c79a38cd454602645f0470ffb444b3b75ce574 | |
1335 | When | |
1336 | .I sample_type | |
1337 | includes | |
5dd3feec | 1338 | .BR PERF_SAMPLE_CALLCHAIN , |
4b3a5f01 | 1339 | this field specifies how many stack frames to report when |
fd133d5d | 1340 | generating the callchain. |
73d8cece | 1341 | .SS Reading results |
f2b1d720 | 1342 | Once a |
7db515ef | 1343 | .BR perf_event_open () |
3d1ee497 | 1344 | file descriptor has been opened, the values |
f2b1d720 MK |
1345 | of the events can be read from the file descriptor. |
1346 | The values that are there are specified by the | |
1347 | .I read_format | |
7db515ef MK |
1348 | field in the |
1349 | .I attr | |
1350 | structure at open time. | |
f2b1d720 MK |
1351 | |
1352 | If you attempt to read into a buffer that is not big enough to hold the | |
4b3a5f01 | 1353 | data, the error |
f2b1d720 | 1354 | .B ENOSPC |
4b3a5f01 | 1355 | results. |
f2b1d720 MK |
1356 | |
1357 | Here is the layout of the data returned by a read: | |
e525b89f | 1358 | .IP * 2 |
f2b1d720 MK |
1359 | If |
1360 | .B PERF_FORMAT_GROUP | |
1361 | was specified to allow reading all events in a group at once: | |
1362 | ||
1363 | .in +4n | |
1364 | .nf | |
1365 | struct read_format { | |
e525b89f MK |
1366 | u64 nr; /* The number of events */ |
1367 | u64 time_enabled; /* if PERF_FORMAT_TOTAL_TIME_ENABLED */ | |
1368 | u64 time_running; /* if PERF_FORMAT_TOTAL_TIME_RUNNING */ | |
e307112d | 1369 | struct { |
e525b89f MK |
1370 | u64 value; /* The value of the event */ |
1371 | u64 id; /* if PERF_FORMAT_ID */ | |
f2b1d720 MK |
1372 | } values[nr]; |
1373 | }; | |
1374 | .fi | |
1375 | .in | |
e525b89f | 1376 | .IP * |
f2b1d720 MK |
1377 | If |
1378 | .B PERF_FORMAT_GROUP | |
1379 | was | |
1380 | .I not | |
e525b89f | 1381 | specified: |
f2b1d720 MK |
1382 | |
1383 | .in +4n | |
1384 | .nf | |
1385 | struct read_format { | |
1386 | u64 value; /* The value of the event */ | |
1387 | u64 time_enabled; /* if PERF_FORMAT_TOTAL_TIME_ENABLED */ | |
1388 | u64 time_running; /* if PERF_FORMAT_TOTAL_TIME_RUNNING */ | |
1389 | u64 id; /* if PERF_FORMAT_ID */ | |
1390 | }; | |
1391 | .fi | |
1392 | .in | |
e525b89f MK |
1393 | .PP |
1394 | The values read are as follows: | |
f2b1d720 MK |
1395 | .TP |
1396 | .I nr | |
1397 | The number of events in this file descriptor. | |
fcc4f4f4 | 1398 | Available only if |
f2b1d720 MK |
1399 | .B PERF_FORMAT_GROUP |
1400 | was specified. | |
f2b1d720 MK |
1401 | .TP |
1402 | .IR time_enabled ", " time_running | |
1403 | Total time the event was enabled and running. | |
4b3a5f01 | 1404 | Normally these values are the same. |
37bee118 MK |
1405 | If more events are started, |
1406 | then available counter slots on the PMU, then multiplexing | |
33a0ccb2 | 1407 | happens and events run only part of the time. |
37bee118 | 1408 | In that case, the |
f2b1d720 MK |
1409 | .I time_enabled |
1410 | and | |
1411 | .I time running | |
1412 | values can be used to scale an estimated value for the count. | |
f2b1d720 MK |
1413 | .TP |
1414 | .I value | |
1415 | An unsigned 64-bit value containing the counter result. | |
f2b1d720 MK |
1416 | .TP |
1417 | .I id | |
4b3a5f01 | 1418 | A globally unique value for this particular event; only present if |
f2b1d720 | 1419 | .B PERF_FORMAT_ID |
e525b89f MK |
1420 | was specified in |
1421 | .IR read_format . | |
73d8cece | 1422 | .SS MMAP layout |
f2b1d720 | 1423 | When using |
7db515ef | 1424 | .BR perf_event_open () |
f2b1d720 MK |
1425 | in sampled mode, asynchronous events |
1426 | (like counter overflow or | |
1427 | .B PROT_EXEC | |
1428 | mmap tracking) | |
1429 | are logged into a ring-buffer. | |
1430 | This ring-buffer is created and accessed through | |
1431 | .BR mmap (2). | |
1432 | ||
1433 | The mmap size should be 1+2^n pages, where the first page is a | |
1434 | metadata page | |
e525b89f | 1435 | .RI ( "struct perf_event_mmap_page" ) |
f2b1d720 MK |
1436 | that contains various |
1437 | bits of information such as where the ring-buffer head is. | |
1438 | ||
95655a22 | 1439 | Before kernel 2.6.39, there is a bug that means you must allocate an mmap |
f2b1d720 MK |
1440 | ring buffer when sampling even if you do not plan to access it. |
1441 | ||
1442 | The structure of the first metadata mmap page is as follows: | |
1443 | ||
1444 | .in +4n | |
1445 | .nf | |
1446 | struct perf_event_mmap_page { | |
ce88f77b MK |
1447 | __u32 version; /* version number of this structure */ |
1448 | __u32 compat_version; /* lowest version this is compat with */ | |
1449 | __u32 lock; /* seqlock for synchronization */ | |
1450 | __u32 index; /* hardware counter identifier */ | |
1451 | __s64 offset; /* add to hardware counter value */ | |
1452 | __u64 time_enabled; /* time event active */ | |
1453 | __u64 time_running; /* time event on CPU */ | |
f2b1d720 MK |
1454 | union { |
1455 | __u64 capabilities; | |
135cba8b | 1456 | struct { |
ce88f77b MK |
1457 | __u64 cap_usr_time / cap_usr_rdpmc / cap_bit0 : 1, |
1458 | cap_bit0_is_deprecated : 1, | |
1459 | cap_user_rdpmc : 1, | |
1460 | cap_user_time : 1, | |
1461 | cap_user_time_zero : 1, | |
135cba8b | 1462 | }; |
f2b1d720 | 1463 | }; |
ce88f77b MK |
1464 | __u16 pmc_width; |
1465 | __u16 time_shift; | |
1466 | __u32 time_mult; | |
1467 | __u64 time_offset; | |
1468 | __u64 __reserved[120]; /* Pad to 1k */ | |
1469 | __u64 data_head; /* head in the data section */ | |
1470 | __u64 data_tail; /* user-space written tail */ | |
21d9849a VW |
1471 | __u64 data_offset; /* where the buffer starts */ |
1472 | __u64 data_size; /* data buffer size */ | |
4e47c6e5 VW |
1473 | __u64 aux_head; |
1474 | __u64 aux_tail; | |
1475 | __u64 aux_offset; | |
1476 | __u64 aux_size; | |
21d9849a | 1477 | |
f2b1d720 MK |
1478 | } |
1479 | .fi | |
1480 | .in | |
1481 | ||
ce88f77b | 1482 | The following list describes the fields in the |
f2b1d720 | 1483 | .I perf_event_mmap_page |
e525b89f | 1484 | structure in more detail: |
f2b1d720 MK |
1485 | .TP |
1486 | .I version | |
1487 | Version number of this structure. | |
f2b1d720 MK |
1488 | .TP |
1489 | .I compat_version | |
1490 | The lowest version this is compatible with. | |
f2b1d720 MK |
1491 | .TP |
1492 | .I lock | |
1493 | A seqlock for synchronization. | |
f2b1d720 MK |
1494 | .TP |
1495 | .I index | |
1496 | A unique hardware counter identifier. | |
f2b1d720 MK |
1497 | .TP |
1498 | .I offset | |
135cba8b VW |
1499 | When using rdpmc for reads this offset value |
1500 | must be added to the one returned by rdpmc to get | |
1501 | the current total event count. | |
f2b1d720 MK |
1502 | .TP |
1503 | .I time_enabled | |
1504 | Time the event was active. | |
f2b1d720 MK |
1505 | .TP |
1506 | .I time_running | |
1507 | Time the event was running. | |
f2b1d720 | 1508 | .TP |
31c1f2b0 | 1509 | .IR cap_usr_time " / " cap_usr_rdpmc " / " cap_bit0 " (since Linux 3.4)" |
747a6e7c | 1510 | .\" commit c7206205d00ab375839bd6c7ddb247d600693c09 |
e9bd9b2c | 1511 | There was a bug in the definition of |
f2b1d720 | 1512 | .I cap_usr_time |
135cba8b VW |
1513 | and |
1514 | .I cap_usr_rdpmc | |
1515 | from Linux 3.4 until Linux 3.11. | |
1516 | Both bits were defined to point to the same location, so it was | |
e9bd9b2c | 1517 | impossible to know if |
135cba8b VW |
1518 | .I cap_usr_time |
1519 | or | |
1520 | .I cap_usr_rdpmc | |
1521 | were actually set. | |
1522 | ||
4010bc07 | 1523 | Starting with Linux 3.12, these are renamed to |
747a6e7c | 1524 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
135cba8b | 1525 | .I cap_bit0 |
e41c36b2 | 1526 | and you should use the |
135cba8b VW |
1527 | .I cap_user_time |
1528 | and | |
1529 | .I cap_user_rdpmc | |
1530 | fields instead. | |
f2b1d720 | 1531 | .TP |
31c1f2b0 | 1532 | .IR cap_bit0_is_deprecated " (since Linux 3.12)" |
747a6e7c | 1533 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
37bee118 | 1534 | If set, this bit indicates that the kernel supports |
135cba8b VW |
1535 | the properly separated |
1536 | .I cap_user_time | |
1537 | and | |
1538 | .I cap_user_rdpmc | |
1539 | bits. | |
1540 | ||
1541 | If not-set, it indicates an older kernel where | |
1542 | .I cap_usr_time | |
1543 | and | |
f2b1d720 | 1544 | .I cap_usr_rdpmc |
135cba8b VW |
1545 | map to the same bit and thus both features should |
1546 | be used with caution. | |
135cba8b | 1547 | .TP |
31c1f2b0 | 1548 | .IR cap_user_rdpmc " (since Linux 3.12)" |
747a6e7c | 1549 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
f2b1d720 MK |
1550 | If the hardware supports user-space read of performance counters |
1551 | without syscall (this is the "rdpmc" instruction on x86), then | |
1552 | the following code can be used to do a read: | |
1553 | ||
1554 | .in +4n | |
1555 | .nf | |
1556 | u32 seq, time_mult, time_shift, idx, width; | |
1557 | u64 count, enabled, running; | |
1558 | u64 cyc, time_offset; | |
f2b1d720 MK |
1559 | |
1560 | do { | |
1561 | seq = pc\->lock; | |
1562 | barrier(); | |
1563 | enabled = pc\->time_enabled; | |
1564 | running = pc\->time_running; | |
1565 | ||
1566 | if (pc\->cap_usr_time && enabled != running) { | |
1567 | cyc = rdtsc(); | |
1568 | time_offset = pc\->time_offset; | |
1569 | time_mult = pc\->time_mult; | |
1570 | time_shift = pc\->time_shift; | |
1571 | } | |
1572 | ||
1573 | idx = pc\->index; | |
1574 | count = pc\->offset; | |
1575 | ||
1576 | if (pc\->cap_usr_rdpmc && idx) { | |
1577 | width = pc\->pmc_width; | |
135cba8b | 1578 | count += rdpmc(idx \- 1); |
f2b1d720 MK |
1579 | } |
1580 | ||
1581 | barrier(); | |
1582 | } while (pc\->lock != seq); | |
1583 | .fi | |
1584 | .in | |
f2b1d720 | 1585 | .TP |
cc19ea28 | 1586 | .IR cap_user_time " (since Linux 3.12)" |
747a6e7c | 1587 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
7d182bb6 | 1588 | This bit indicates the hardware has a constant, nonstop |
135cba8b VW |
1589 | timestamp counter (TSC on x86). |
1590 | .TP | |
31c1f2b0 | 1591 | .IR cap_user_time_zero " (since Linux 3.12)" |
747a6e7c | 1592 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
135cba8b VW |
1593 | Indicates the presence of |
1594 | .I time_zero | |
1595 | which allows mapping timestamp values to | |
1596 | the hardware clock. | |
1597 | .TP | |
f2b1d720 MK |
1598 | .I pmc_width |
1599 | If | |
1600 | .IR cap_usr_rdpmc , | |
1601 | this field provides the bit-width of the value | |
1602 | read using the rdpmc or equivalent instruction. | |
1603 | This can be used to sign extend the result like: | |
1604 | ||
1605 | .in +4n | |
1606 | .nf | |
1607 | pmc <<= 64 \- pmc_width; | |
1608 | pmc >>= 64 \- pmc_width; // signed shift right | |
1609 | count += pmc; | |
1610 | .fi | |
1611 | .in | |
f2b1d720 MK |
1612 | .TP |
1613 | .IR time_shift ", " time_mult ", " time_offset | |
1614 | ||
1615 | If | |
1616 | .IR cap_usr_time , | |
1617 | these fields can be used to compute the time | |
4b3a5f01 MK |
1618 | delta since |
1619 | .I time_enabled | |
1620 | (in nanoseconds) using rdtsc or similar. | |
f2b1d720 MK |
1621 | .nf |
1622 | ||
1623 | u64 quot, rem; | |
1624 | u64 delta; | |
1625 | quot = (cyc >> time_shift); | |
988688f6 | 1626 | rem = cyc & (((u64)1 << time_shift) \- 1); |
f2b1d720 MK |
1627 | delta = time_offset + quot * time_mult + |
1628 | ((rem * time_mult) >> time_shift); | |
1629 | .fi | |
1630 | ||
7db515ef MK |
1631 | Where |
1632 | .IR time_offset , | |
1633 | .IR time_mult , | |
1634 | .IR time_shift , | |
1635 | and | |
1636 | .IR cyc | |
1637 | are read in the | |
f2b1d720 MK |
1638 | seqcount loop described above. |
1639 | This delta can then be added to | |
1640 | enabled and possible running (if idx), improving the scaling: | |
1641 | .nf | |
1642 | ||
1643 | enabled += delta; | |
1644 | if (idx) | |
1645 | running += delta; | |
1646 | quot = count / running; | |
1647 | rem = count % running; | |
1648 | count = quot * enabled + (rem * enabled) / running; | |
1649 | .fi | |
f2b1d720 | 1650 | .TP |
31c1f2b0 | 1651 | .IR time_zero " (since Linux 3.12)" |
747a6e7c | 1652 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
135cba8b | 1653 | |
e9bd9b2c | 1654 | If |
135cba8b | 1655 | .I cap_usr_time_zero |
37bee118 | 1656 | is set, then the hardware clock (the TSC timestamp counter on x86) |
135cba8b VW |
1657 | can be calculated from the |
1658 | .IR time_zero ", " time_mult ", and " time_shift " values:" | |
ce88f77b | 1659 | |
135cba8b VW |
1660 | .nf |
1661 | time = timestamp - time_zero; | |
1662 | quot = time / time_mult; | |
1663 | rem = time % time_mult; | |
1664 | cyc = (quot << time_shift) + (rem << time_shift) / time_mult; | |
1665 | .fi | |
ce88f77b | 1666 | |
135cba8b | 1667 | And vice versa: |
ce88f77b | 1668 | |
135cba8b VW |
1669 | .nf |
1670 | quot = cyc >> time_shift; | |
988688f6 | 1671 | rem = cyc & (((u64)1 << time_shift) - 1); |
135cba8b VW |
1672 | timestamp = time_zero + quot * time_mult + |
1673 | ((rem * time_mult) >> time_shift); | |
1674 | .fi | |
1675 | .TP | |
f2b1d720 MK |
1676 | .I data_head |
1677 | This points to the head of the data section. | |
7db515ef MK |
1678 | The value continuously increases, it does not wrap. |
1679 | The value needs to be manually wrapped by the size of the mmap buffer | |
f2b1d720 MK |
1680 | before accessing the samples. |
1681 | ||
ce88f77b MK |
1682 | On SMP-capable platforms, after reading the |
1683 | .I data_head | |
1684 | value, | |
ad73a2cc | 1685 | user space should issue an rmb(). |
f2b1d720 | 1686 | .TP |
fecd584f | 1687 | .I data_tail |
f2b1d720 MK |
1688 | When the mapping is |
1689 | .BR PROT_WRITE , | |
7db515ef MK |
1690 | the |
1691 | .I data_tail | |
1692 | value should be written by user space to reflect the last read data. | |
31020de9 | 1693 | In this case, the kernel will not overwrite unread data. |
21d9849a VW |
1694 | .TP |
1695 | .IR data_offset " (since Linux 4.1)" | |
1696 | .\" commit e8c6deac69629c0cb97c3d3272f8631ef17f8f0f | |
1697 | Contains the offset of the location in the mmap buffer | |
1698 | where perf sample data begins. | |
1699 | .TP | |
1700 | .IR data_size " (since Linux 4.1)" | |
1701 | .\" commit e8c6deac69629c0cb97c3d3272f8631ef17f8f0f | |
1702 | Contains the size of the perf sample region within | |
1703 | the mmap buffer. | |
4e47c6e5 VW |
1704 | .TP |
1705 | .IR aux_head ", " aux_tail ", " aux_offset ", " aux_size " (since Linux 4.1) | |
1706 | .\" commit 45bfb2e50471abbbfd83d40d28c986078b0d24ff | |
95655a22 MK |
1707 | The AUX region allows mmaping a separate sample buffer for |
1708 | high-bandwidth data streams (separate from the main perf sample buffer). | |
1709 | An example of a high-bandwidth stream is instruction tracing support, | |
4e47c6e5 VW |
1710 | as is found in newer Intel processors. |
1711 | ||
1712 | To set up an AUX area, first | |
1713 | .I aux_offset | |
1714 | needs to be set with an offset greater than | |
1715 | .IR data_offset + data_size | |
1716 | and | |
1717 | .I aux_size | |
1718 | needs to be set to the desired buffer size. | |
1719 | The desired offset and size must be page aligned, and the size | |
1720 | must be a power of two. | |
1721 | These values are then passed to mmap in order to map the AUX buffer. | |
95655a22 MK |
1722 | Pages in the AUX buffer are included as part of the |
1723 | .BR RLIMIT_MEMLOCK | |
1724 | resource limit (see | |
1725 | .BR setrlimit (2)), | |
1726 | and also as part of the | |
4e47c6e5 VW |
1727 | .I perf_event_mlock_kb |
1728 | allowance. | |
1729 | ||
95655a22 | 1730 | By default, the AUX buffer will be truncated if it will not fit |
b1355f6a VW |
1731 | in the available space in the ring buffer. |
1732 | If the AUX buffer is mapped as a read only buffer, then it will | |
1733 | operate in ring buffer mode where old data will be overwritten | |
1734 | by new. | |
95655a22 | 1735 | In overwrite mode, it might not be possible to infer where the |
b1355f6a VW |
1736 | new data began, and it is the consumer's job to disable |
1737 | measurement while reading to avoid possible data races. | |
1738 | ||
4e47c6e5 VW |
1739 | The |
1740 | .IR aux_head " and " aux_tail | |
1741 | ring buffer pointers have the same behavior and ordering | |
1742 | rules as the previous described | |
1743 | .IR data_head " and " data_tail . | |
e525b89f | 1744 | .PP |
f2b1d720 MK |
1745 | The following 2^n ring-buffer pages have the layout described below. |
1746 | ||
1747 | If | |
1748 | .I perf_event_attr.sample_id_all | |
1749 | is set, then all event types will | |
1750 | have the sample_type selected fields related to where/when (identity) | |
1751 | an event took place (TID, TIME, ID, CPU, STREAM_ID) described in | |
1752 | .B PERF_RECORD_SAMPLE | |
1753 | below, it will be stashed just after the | |
7db515ef MK |
1754 | .I perf_event_header |
1755 | and the fields already present for the existing | |
3d1ee497 | 1756 | fields, that is, at the end of the payload. |
4b3a5f01 MK |
1757 | This allows a newer perf.data |
1758 | file to be supported by older perf tools, with the new optional | |
f2b1d720 MK |
1759 | fields being ignored. |
1760 | ||
1761 | The mmap values start with a header: | |
1762 | ||
1763 | .in +4n | |
1764 | .nf | |
1765 | struct perf_event_header { | |
1766 | __u32 type; | |
1767 | __u16 misc; | |
1768 | __u16 size; | |
1769 | }; | |
1770 | .fi | |
1771 | .in | |
1772 | ||
1773 | Below, we describe the | |
1774 | .I perf_event_header | |
1775 | fields in more detail. | |
4047bc6c MK |
1776 | For ease of reading, |
1777 | the fields with shorter descriptions are presented first. | |
1778 | .TP | |
1779 | .I size | |
1780 | This indicates the size of the record. | |
1781 | .TP | |
1782 | .I misc | |
1783 | The | |
1784 | .I misc | |
1785 | field contains additional information about the sample. | |
1786 | ||
1787 | The CPU mode can be determined from this value by masking with | |
1788 | .B PERF_RECORD_MISC_CPUMODE_MASK | |
1789 | and looking for one of the following (note these are not | |
1790 | bit masks, only one can be set at a time): | |
1791 | .RS | |
1792 | .TP | |
1793 | .B PERF_RECORD_MISC_CPUMODE_UNKNOWN | |
1794 | Unknown CPU mode. | |
1795 | .TP | |
1796 | .B PERF_RECORD_MISC_KERNEL | |
1797 | Sample happened in the kernel. | |
1798 | .TP | |
1799 | .B PERF_RECORD_MISC_USER | |
1800 | Sample happened in user code. | |
1801 | .TP | |
1802 | .B PERF_RECORD_MISC_HYPERVISOR | |
1803 | Sample happened in the hypervisor. | |
1804 | .TP | |
747a6e7c | 1805 | .BR PERF_RECORD_MISC_GUEST_KERNEL " (since Linux 2.6.35)" |
60dafbc1 | 1806 | .\" commit 39447b386c846bbf1c56f6403c5282837486200f |
4047bc6c MK |
1807 | Sample happened in the guest kernel. |
1808 | .TP | |
747a6e7c | 1809 | .B PERF_RECORD_MISC_GUEST_USER " (since Linux 2.6.35)" |
60dafbc1 | 1810 | .\" commit 39447b386c846bbf1c56f6403c5282837486200f |
4047bc6c MK |
1811 | Sample happened in guest user code. |
1812 | .RE | |
1813 | ||
1814 | .RS | |
d5a24378 MK |
1815 | Since the following three statuses are generated by |
1816 | different record types, they alias to the same bit: | |
4047bc6c | 1817 | .TP |
60dafbc1 MK |
1818 | .BR PERF_RECORD_MISC_MMAP_DATA " (since Linux 3.10)" |
1819 | .\" commit 2fe85427e3bf65d791700d065132772fc26e4d75 | |
4047bc6c MK |
1820 | This is set when the mapping is not executable; |
1821 | otherwise the mapping is executable. | |
1822 | .TP | |
60dafbc1 MK |
1823 | .BR PERF_RECORD_MISC_COMM_EXEC " (since Linux 3.16)" |
1824 | .\" commit 82b897782d10fcc4930c9d4a15b175348fdd2871 | |
49bc411c VW |
1825 | This is set for a |
1826 | .B PERF_RECORD_COMM | |
1827 | record on kernels more recent than Linux 3.16 | |
1828 | if a process name change was caused by an | |
1829 | .BR exec (2) | |
1830 | system call. | |
9277a75d VW |
1831 | .TP |
1832 | .BR PERF_RECORD_MISC_SWITCH_OUT " (since Linux 4.3)" | |
1833 | .\" commit 45ac1403f564f411c6a383a2448688ba8dd705a4 | |
1834 | When a | |
d5a24378 MK |
1835 | .BR PERF_RECORD_SWITCH |
1836 | or | |
1837 | .BR PERF_RECORD_SWITCH_CPU_WIDE | |
1838 | record is generated, this bit indicates that the | |
9277a75d | 1839 | context switch is away from the current process |
d5a24378 | 1840 | (instead of into the current process). |
9277a75d VW |
1841 | .RE |
1842 | ||
1843 | .RS | |
1844 | In addition, the following bits can be set: | |
49bc411c | 1845 | .TP |
4047bc6c MK |
1846 | .B PERF_RECORD_MISC_EXACT_IP |
1847 | This indicates that the content of | |
1848 | .B PERF_SAMPLE_IP | |
1849 | points | |
1850 | to the actual instruction that triggered the event. | |
1851 | See also | |
1852 | .IR perf_event_attr.precise_ip . | |
1853 | .TP | |
60dafbc1 MK |
1854 | .BR PERF_RECORD_MISC_EXT_RESERVED " (since Linux 2.6.35)" |
1855 | .\" commit 1676b8a077c352085d52578fb4f29350b58b6e74 | |
4047bc6c | 1856 | This indicates there is extended data available (currently not used). |
ffbc7c02 VW |
1857 | .TP |
1858 | .B PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT | |
1859 | .\" commit 930e6fcd2bcce9bcd9d4aa7e755678d33f3fe6f4 | |
1860 | This bit is not set by the kernel. | |
141efa1b MK |
1861 | It is reserved for the user-space perf utility to indicate that |
1862 | .I /proc/i[pid]/maps | |
1863 | parsing was taking too long and was stopped, and thus the mmap | |
ffbc7c02 | 1864 | records may be truncated. |
4047bc6c | 1865 | .RE |
f2b1d720 MK |
1866 | .TP |
1867 | .I type | |
1868 | The | |
1869 | .I type | |
1870 | value is one of the below. | |
1871 | The values in the corresponding record (that follows the header) | |
1872 | depend on the | |
1873 | .I type | |
1874 | selected as shown. | |
f2b1d720 | 1875 | .RS |
7db515ef | 1876 | .TP 4 |
f2b1d720 MK |
1877 | .B PERF_RECORD_MMAP |
1878 | The MMAP events record the | |
1879 | .B PROT_EXEC | |
1880 | mappings so that we can correlate | |
ad73a2cc | 1881 | user-space IPs to code. |
f2b1d720 MK |
1882 | They have the following structure: |
1883 | ||
1884 | .in +4n | |
1885 | .nf | |
1886 | struct { | |
1887 | struct perf_event_header header; | |
1888 | u32 pid, tid; | |
1889 | u64 addr; | |
1890 | u64 len; | |
1891 | u64 pgoff; | |
1892 | char filename[]; | |
1893 | }; | |
1894 | .fi | |
1895 | .in | |
9bfc542b VW |
1896 | .RS |
1897 | .TP | |
1898 | .I pid | |
3a058284 | 1899 | is the process ID. |
9bfc542b VW |
1900 | .TP |
1901 | .I tid | |
3a058284 | 1902 | is the thread ID. |
9bfc542b VW |
1903 | .TP |
1904 | .I addr | |
1905 | is the address of the allocated memory. | |
1906 | .I len | |
1907 | is the length of the allocated memory. | |
1908 | .I pgoff | |
1909 | is the page offset of the allocated memory. | |
1910 | .I filename | |
1911 | is a string describing the backing of the allocated memory. | |
1912 | .RE | |
f2b1d720 MK |
1913 | .TP |
1914 | .B PERF_RECORD_LOST | |
1915 | This record indicates when events are lost. | |
1916 | ||
1917 | .in +4n | |
1918 | .nf | |
1919 | struct { | |
1920 | struct perf_event_header header; | |
7a10da70 MK |
1921 | u64 id; |
1922 | u64 lost; | |
7480dabb | 1923 | struct sample_id sample_id; |
f2b1d720 MK |
1924 | }; |
1925 | .fi | |
1926 | .in | |
f2b1d720 MK |
1927 | .RS |
1928 | .TP | |
1929 | .I id | |
1930 | is the unique event ID for the samples that were lost. | |
1931 | .TP | |
1932 | .I lost | |
1933 | is the number of events that were lost. | |
1934 | .RE | |
f2b1d720 MK |
1935 | .TP |
1936 | .B PERF_RECORD_COMM | |
1937 | This record indicates a change in the process name. | |
1938 | ||
1939 | .in +4n | |
1940 | .nf | |
1941 | struct { | |
1942 | struct perf_event_header header; | |
7a10da70 MK |
1943 | u32 pid; |
1944 | u32 tid; | |
1945 | char comm[]; | |
7480dabb | 1946 | struct sample_id sample_id; |
f2b1d720 MK |
1947 | }; |
1948 | .fi | |
1949 | .in | |
49bc411c VW |
1950 | .RS |
1951 | .TP | |
1952 | .I pid | |
5ab35ae5 | 1953 | is the process ID. |
49bc411c VW |
1954 | .TP |
1955 | .I tid | |
5ab35ae5 | 1956 | is the thread ID. |
49bc411c VW |
1957 | .TP |
1958 | .I comm | |
1959 | is a string containing the new name of the process. | |
1960 | .RE | |
f2b1d720 MK |
1961 | .TP |
1962 | .B PERF_RECORD_EXIT | |
1963 | This record indicates a process exit event. | |
1964 | ||
1965 | .in +4n | |
1966 | .nf | |
1967 | struct { | |
1968 | struct perf_event_header header; | |
7a10da70 MK |
1969 | u32 pid, ppid; |
1970 | u32 tid, ptid; | |
1971 | u64 time; | |
7480dabb | 1972 | struct sample_id sample_id; |
f2b1d720 MK |
1973 | }; |
1974 | .fi | |
1975 | .in | |
f2b1d720 MK |
1976 | .TP |
1977 | .BR PERF_RECORD_THROTTLE ", " PERF_RECORD_UNTHROTTLE | |
1978 | This record indicates a throttle/unthrottle event. | |
1979 | ||
1980 | .in +4n | |
1981 | .nf | |
1982 | struct { | |
1983 | struct perf_event_header header; | |
7a10da70 MK |
1984 | u64 time; |
1985 | u64 id; | |
1986 | u64 stream_id; | |
7480dabb | 1987 | struct sample_id sample_id; |
f2b1d720 MK |
1988 | }; |
1989 | .fi | |
1990 | .in | |
f2b1d720 MK |
1991 | .TP |
1992 | .B PERF_RECORD_FORK | |
1993 | This record indicates a fork event. | |
1994 | ||
1995 | .in +4n | |
1996 | .nf | |
1997 | struct { | |
1998 | struct perf_event_header header; | |
7a10da70 MK |
1999 | u32 pid, ppid; |
2000 | u32 tid, ptid; | |
2001 | u64 time; | |
7480dabb | 2002 | struct sample_id sample_id; |
f2b1d720 MK |
2003 | }; |
2004 | .fi | |
2005 | .in | |
f2b1d720 MK |
2006 | .TP |
2007 | .B PERF_RECORD_READ | |
2008 | This record indicates a read event. | |
2009 | ||
2010 | .in +4n | |
2011 | .nf | |
2012 | struct { | |
2013 | struct perf_event_header header; | |
7a10da70 | 2014 | u32 pid, tid; |
f2b1d720 | 2015 | struct read_format values; |
7480dabb | 2016 | struct sample_id sample_id; |
f2b1d720 MK |
2017 | }; |
2018 | .fi | |
2019 | .in | |
f2b1d720 MK |
2020 | .TP |
2021 | .B PERF_RECORD_SAMPLE | |
2022 | This record indicates a sample. | |
2023 | ||
2024 | .in +4n | |
2025 | .nf | |
2026 | struct { | |
2027 | struct perf_event_header header; | |
880403e9 MK |
2028 | u64 sample_id; /* if PERF_SAMPLE_IDENTIFIER */ |
2029 | u64 ip; /* if PERF_SAMPLE_IP */ | |
2030 | u32 pid, tid; /* if PERF_SAMPLE_TID */ | |
2031 | u64 time; /* if PERF_SAMPLE_TIME */ | |
2032 | u64 addr; /* if PERF_SAMPLE_ADDR */ | |
2033 | u64 id; /* if PERF_SAMPLE_ID */ | |
2034 | u64 stream_id; /* if PERF_SAMPLE_STREAM_ID */ | |
2035 | u32 cpu, res; /* if PERF_SAMPLE_CPU */ | |
2036 | u64 period; /* if PERF_SAMPLE_PERIOD */ | |
2037 | struct read_format v; /* if PERF_SAMPLE_READ */ | |
2038 | u64 nr; /* if PERF_SAMPLE_CALLCHAIN */ | |
2039 | u64 ips[nr]; /* if PERF_SAMPLE_CALLCHAIN */ | |
2040 | u32 size; /* if PERF_SAMPLE_RAW */ | |
2041 | char data[size]; /* if PERF_SAMPLE_RAW */ | |
2042 | u64 bnr; /* if PERF_SAMPLE_BRANCH_STACK */ | |
7db515ef | 2043 | struct perf_branch_entry lbr[bnr]; |
880403e9 MK |
2044 | /* if PERF_SAMPLE_BRANCH_STACK */ |
2045 | u64 abi; /* if PERF_SAMPLE_REGS_USER */ | |
7a10da70 | 2046 | u64 regs[weight(mask)]; |
880403e9 MK |
2047 | /* if PERF_SAMPLE_REGS_USER */ |
2048 | u64 size; /* if PERF_SAMPLE_STACK_USER */ | |
2049 | char data[size]; /* if PERF_SAMPLE_STACK_USER */ | |
2050 | u64 dyn_size; /* if PERF_SAMPLE_STACK_USER && | |
2051 | size != 0 */ | |
2052 | u64 weight; /* if PERF_SAMPLE_WEIGHT */ | |
2053 | u64 data_src; /* if PERF_SAMPLE_DATA_SRC */ | |
2054 | u64 transaction; /* if PERF_SAMPLE_TRANSACTION */ | |
2055 | u64 abi; /* if PERF_SAMPLE_REGS_INTR */ | |
7a10da70 | 2056 | u64 regs[weight(mask)]; |
880403e9 | 2057 | /* if PERF_SAMPLE_REGS_INTR */ |
f2b1d720 MK |
2058 | }; |
2059 | .fi | |
4047bc6c MK |
2060 | .RS 4 |
2061 | .TP 4 | |
7480dabb VW |
2062 | .I sample_id |
2063 | If | |
2064 | .B PERF_SAMPLE_IDENTIFIER | |
2065 | is enabled, a 64-bit unique ID is included. | |
e9bd9b2c | 2066 | This is a duplication of the |
7480dabb VW |
2067 | .B PERF_SAMPLE_ID |
2068 | .I id | |
2069 | value, but included at the beginning of the sample | |
2070 | so parsers can easily obtain the value. | |
2071 | .TP | |
f2b1d720 | 2072 | .I ip |
7db515ef MK |
2073 | If |
2074 | .B PERF_SAMPLE_IP | |
2075 | is enabled, then a 64-bit instruction | |
f2b1d720 | 2076 | pointer value is included. |
f2b1d720 | 2077 | .TP |
7db515ef MK |
2078 | .IR pid ", " tid |
2079 | If | |
2080 | .B PERF_SAMPLE_TID | |
2081 | is enabled, then a 32-bit process ID | |
2082 | and 32-bit thread ID are included. | |
f2b1d720 MK |
2083 | .TP |
2084 | .I time | |
7db515ef MK |
2085 | If |
2086 | .B PERF_SAMPLE_TIME | |
2087 | is enabled, then a 64-bit timestamp | |
f2b1d720 MK |
2088 | is included. |
2089 | This is obtained via local_clock() which is a hardware timestamp | |
2090 | if available and the jiffies value if not. | |
f2b1d720 MK |
2091 | .TP |
2092 | .I addr | |
7db515ef MK |
2093 | If |
2094 | .B PERF_SAMPLE_ADDR | |
2095 | is enabled, then a 64-bit address is included. | |
f2b1d720 MK |
2096 | This is usually the address of a tracepoint, |
2097 | breakpoint, or software event; otherwise the value is 0. | |
f2b1d720 MK |
2098 | .TP |
2099 | .I id | |
7db515ef MK |
2100 | If |
2101 | .B PERF_SAMPLE_ID | |
2102 | is enabled, a 64-bit unique ID is included. | |
f2b1d720 | 2103 | If the event is a member of an event group, the group leader ID is returned. |
7db515ef MK |
2104 | This ID is the same as the one returned by |
2105 | .BR PERF_FORMAT_ID . | |
f2b1d720 MK |
2106 | .TP |
2107 | .I stream_id | |
7db515ef MK |
2108 | If |
2109 | .B PERF_SAMPLE_STREAM_ID | |
2110 | is enabled, a 64-bit unique ID is included. | |
f2b1d720 MK |
2111 | Unlike |
2112 | .B PERF_SAMPLE_ID | |
2113 | the actual ID is returned, not the group leader. | |
7db515ef MK |
2114 | This ID is the same as the one returned by |
2115 | .BR PERF_FORMAT_ID . | |
f2b1d720 | 2116 | .TP |
7db515ef MK |
2117 | .IR cpu ", " res |
2118 | If | |
2119 | .B PERF_SAMPLE_CPU | |
2120 | is enabled, this is a 32-bit value indicating | |
f2b1d720 MK |
2121 | which CPU was being used, in addition to a reserved (unused) |
2122 | 32-bit value. | |
f2b1d720 MK |
2123 | .TP |
2124 | .I period | |
7db515ef MK |
2125 | If |
2126 | .B PERF_SAMPLE_PERIOD | |
2127 | is enabled, a 64-bit value indicating | |
f2b1d720 | 2128 | the current sampling period is written. |
f2b1d720 MK |
2129 | .TP |
2130 | .I v | |
7db515ef MK |
2131 | If |
2132 | .B PERF_SAMPLE_READ | |
2133 | is enabled, a structure of type read_format | |
f2b1d720 MK |
2134 | is included which has values for all events in the event group. |
2135 | The values included depend on the | |
2136 | .I read_format | |
7db515ef MK |
2137 | value used at |
2138 | .BR perf_event_open () | |
2139 | time. | |
f2b1d720 | 2140 | .TP |
7db515ef MK |
2141 | .IR nr ", " ips[nr] |
2142 | If | |
2143 | .B PERF_SAMPLE_CALLCHAIN | |
2144 | is enabled, then a 64-bit number is included | |
f2b1d720 | 2145 | which indicates how many following 64-bit instruction pointers will |
7db515ef MK |
2146 | follow. |
2147 | This is the current callchain. | |
f2b1d720 | 2148 | .TP |
7ede2f66 | 2149 | .IR size ", " data[size] |
7db515ef MK |
2150 | If |
2151 | .B PERF_SAMPLE_RAW | |
2152 | is enabled, then a 32-bit value indicating size | |
f2b1d720 MK |
2153 | is included followed by an array of 8-bit values of length size. |
2154 | The values are padded with 0 to have 64-bit alignment. | |
2155 | ||
2156 | This RAW record data is opaque with respect to the ABI. | |
2157 | The ABI doesn't make any promises with respect to the stability | |
2158 | of its content, it may vary depending | |
2159 | on event, hardware, and kernel version. | |
f2b1d720 | 2160 | .TP |
7db515ef MK |
2161 | .IR bnr ", " lbr[bnr] |
2162 | If | |
2163 | .B PERF_SAMPLE_BRANCH_STACK | |
2164 | is enabled, then a 64-bit value indicating | |
2165 | the number of records is included, followed by | |
2166 | .I bnr | |
2167 | .I perf_branch_entry | |
045bf4d3 VW |
2168 | structures which each include the fields: |
2169 | .RS | |
2170 | .TP | |
2171 | .I from | |
2b538c3e | 2172 | This indicates the source instruction (may not be a branch). |
045bf4d3 VW |
2173 | .TP |
2174 | .I to | |
2b538c3e | 2175 | The branch target. |
045bf4d3 VW |
2176 | .TP |
2177 | .I mispred | |
2b538c3e | 2178 | The branch target was mispredicted. |
045bf4d3 VW |
2179 | .TP |
2180 | .I predicted | |
2b538c3e | 2181 | The branch target was predicted. |
e3c9782b | 2182 | .TP |
31c1f2b0 | 2183 | .IR in_tx " (since Linux 3.11)" |
747a6e7c | 2184 | .\" commit 135c5612c460f89657c4698fe2ea753f6f667963 |
2b538c3e | 2185 | The branch was in a transactional memory transaction. |
e3c9782b | 2186 | .TP |
31c1f2b0 | 2187 | .IR abort " (since Linux 3.11)" |
747a6e7c | 2188 | .\" commit 135c5612c460f89657c4698fe2ea753f6f667963 |
2b538c3e | 2189 | The branch was in an aborted transactional memory transaction. |
96919592 VW |
2190 | .TP |
2191 | .IR cycles " (since Linux 4.3)" | |
2192 | .\" commit 71ef3c6b9d4665ee7afbbe4c208a98917dcfc32f | |
2193 | This reports the number of cycles elapsed since the | |
2194 | previous branch stack update. | |
e3c9782b | 2195 | .P |
045bf4d3 VW |
2196 | The entries are from most to least recent, so the first entry |
2197 | has the most recent branch. | |
2198 | ||
8a94e783 | 2199 | Support for |
dceb9af6 MK |
2200 | .IR mispred , |
2201 | .IR predicted , | |
2202 | and | |
2203 | .IR cycles | |
96919592 | 2204 | is optional; if not supported, those |
045bf4d3 VW |
2205 | values will be 0. |
2206 | ||
e3c9782b VW |
2207 | The type of branches recorded is specified by the |
2208 | .I branch_sample_type | |
2209 | field. | |
2210 | .RE | |
f2b1d720 | 2211 | .TP |
7db515ef MK |
2212 | .IR abi ", " regs[weight(mask)] |
2213 | If | |
2214 | .B PERF_SAMPLE_REGS_USER | |
d1007d14 | 2215 | is enabled, then the user CPU registers are recorded. |
f2b1d720 MK |
2216 | |
2217 | The | |
2218 | .I abi | |
2219 | field is one of | |
2220 | .BR PERF_SAMPLE_REGS_ABI_NONE ", " PERF_SAMPLE_REGS_ABI_32 " or " | |
7db515ef | 2221 | .BR PERF_SAMPLE_REGS_ABI_64 . |
d1007d14 VW |
2222 | |
2223 | The | |
2224 | .I regs | |
2225 | field is an array of the CPU registers that were specified by | |
2226 | the | |
2227 | .I sample_regs_user | |
2228 | attr field. | |
2229 | The number of values is the number of bits set in the | |
51700fd7 | 2230 | .I sample_regs_user |
4651e412 | 2231 | bit mask. |
f2b1d720 | 2232 | .TP |
7db515ef MK |
2233 | .IR size ", " data[size] ", " dyn_size |
2234 | If | |
2235 | .B PERF_SAMPLE_STACK_USER | |
02ca78a0 VW |
2236 | is enabled, then the user stack is recorded. |
2237 | This can be used to generate stack backtraces. | |
d1007d14 VW |
2238 | .I size |
2239 | is the size requested by the user in | |
02ca78a0 | 2240 | .I sample_stack_user |
d1007d14 VW |
2241 | or else the maximum record size. |
2242 | .I data | |
02ca78a0 VW |
2243 | is the stack data (a raw dump of the memory pointed to by the |
2244 | stack pointer at the time of sampling). | |
d1007d14 VW |
2245 | .I dyn_size |
2246 | is the amount of data actually dumped (can be less than | |
460e3d7a | 2247 | .IR size ). |
4dc411dd KF |
2248 | Note that |
2249 | .I dyn_size | |
2250 | is omitted if | |
2251 | .I size | |
2252 | is 0. | |
d1007d14 | 2253 | .TP |
51700fd7 | 2254 | .I weight |
d1007d14 VW |
2255 | If |
2256 | .B PERF_SAMPLE_WEIGHT | |
7de4a1e3 | 2257 | is enabled, then a 64-bit value provided by the hardware |
d1007d14 VW |
2258 | is recorded that indicates how costly the event was. |
2259 | This allows expensive events to stand out more clearly | |
2260 | in profiles. | |
2261 | .TP | |
2262 | .I data_src | |
51700fd7 | 2263 | If |
d1007d14 | 2264 | .B PERF_SAMPLE_DATA_SRC |
7de4a1e3 | 2265 | is enabled, then a 64-bit value is recorded that is made up of |
d1007d14 VW |
2266 | the following fields: |
2267 | .RS | |
2b538c3e | 2268 | .TP 4 |
d1007d14 | 2269 | .I mem_op |
2b538c3e MK |
2270 | Type of opcode, a bitwise combination of: |
2271 | ||
2272 | .PD 0 | |
2273 | .RS | |
2274 | .TP 24 | |
d1007d14 | 2275 | .B PERF_MEM_OP_NA |
2b538c3e MK |
2276 | Not available |
2277 | .TP | |
d1007d14 | 2278 | .B PERF_MEM_OP_LOAD |
2b538c3e MK |
2279 | Load instruction |
2280 | .TP | |
d1007d14 | 2281 | .B PERF_MEM_OP_STORE |
2b538c3e MK |
2282 | Store instruction |
2283 | .TP | |
d1007d14 | 2284 | .B PERF_MEM_OP_PFETCH |
2b538c3e MK |
2285 | Prefetch |
2286 | .TP | |
d1007d14 | 2287 | .B PERF_MEM_OP_EXEC |
2b538c3e MK |
2288 | Executable code |
2289 | .RE | |
2290 | .PD | |
d1007d14 VW |
2291 | .TP |
2292 | .I mem_lvl | |
bc9d90b5 | 2293 | Memory hierarchy level hit or miss, a bitwise combination of |
ef4f4031 | 2294 | the following, shifted left by |
bc9d90b5 | 2295 | .BR PERF_MEM_LVL_SHIFT : |
2b538c3e MK |
2296 | |
2297 | .PD 0 | |
2298 | .RS | |
2299 | .TP 24 | |
d1007d14 | 2300 | .B PERF_MEM_LVL_NA |
2b538c3e MK |
2301 | Not available |
2302 | .TP | |
d1007d14 | 2303 | .B PERF_MEM_LVL_HIT |
2b538c3e MK |
2304 | Hit |
2305 | .TP | |
d1007d14 | 2306 | .B PERF_MEM_LVL_MISS |
2b538c3e MK |
2307 | Miss |
2308 | .TP | |
d1007d14 | 2309 | .B PERF_MEM_LVL_L1 |
2b538c3e MK |
2310 | Level 1 cache |
2311 | .TP | |
d1007d14 | 2312 | .B PERF_MEM_LVL_LFB |
2b538c3e MK |
2313 | Line fill buffer |
2314 | .TP | |
d1007d14 | 2315 | .B PERF_MEM_LVL_L2 |
2b538c3e MK |
2316 | Level 2 cache |
2317 | .TP | |
d1007d14 | 2318 | .B PERF_MEM_LVL_L3 |
2b538c3e MK |
2319 | Level 3 cache |
2320 | .TP | |
d1007d14 | 2321 | .B PERF_MEM_LVL_LOC_RAM |
2b538c3e MK |
2322 | Local DRAM |
2323 | .TP | |
d1007d14 | 2324 | .B PERF_MEM_LVL_REM_RAM1 |
2b538c3e MK |
2325 | Remote DRAM 1 hop |
2326 | .TP | |
d1007d14 | 2327 | .B PERF_MEM_LVL_REM_RAM2 |
2b538c3e MK |
2328 | Remote DRAM 2 hops |
2329 | .TP | |
d1007d14 | 2330 | .B PERF_MEM_LVL_REM_CCE1 |
2b538c3e MK |
2331 | Remote cache 1 hop |
2332 | .TP | |
d1007d14 | 2333 | .B PERF_MEM_LVL_REM_CCE2 |
2b538c3e MK |
2334 | Remote cache 2 hops |
2335 | .TP | |
d1007d14 | 2336 | .B PERF_MEM_LVL_IO |
2b538c3e MK |
2337 | I/O memory |
2338 | .TP | |
d1007d14 | 2339 | .B PERF_MEM_LVL_UNC |
2b538c3e MK |
2340 | Uncached memory |
2341 | .RE | |
2342 | .PD | |
d1007d14 VW |
2343 | .TP |
2344 | .I mem_snoop | |
bc9d90b5 VW |
2345 | Snoop mode, a bitwise combination of the following, shifted left by |
2346 | .BR PERF_MEM_SNOOP_SHIFT : | |
2b538c3e MK |
2347 | |
2348 | .PD 0 | |
2349 | .RS | |
2350 | .TP 24 | |
d1007d14 | 2351 | .B PERF_MEM_SNOOP_NA |
2b538c3e MK |
2352 | Not available |
2353 | .TP | |
d1007d14 | 2354 | .B PERF_MEM_SNOOP_NONE |
2b538c3e MK |
2355 | No snoop |
2356 | .TP | |
d1007d14 | 2357 | .B PERF_MEM_SNOOP_HIT |
2b538c3e MK |
2358 | Snoop hit |
2359 | .TP | |
d1007d14 | 2360 | .B PERF_MEM_SNOOP_MISS |
2b538c3e MK |
2361 | Snoop miss |
2362 | .TP | |
d1007d14 | 2363 | .B PERF_MEM_SNOOP_HITM |
2b538c3e MK |
2364 | Snoop hit modified |
2365 | .RE | |
2366 | .PD | |
d1007d14 VW |
2367 | .TP |
2368 | .I mem_lock | |
bc9d90b5 VW |
2369 | Lock instruction, a bitwise combination of the following, shifted left by |
2370 | .BR PERF_MEM_LOCK_SHIFT : | |
2b538c3e MK |
2371 | |
2372 | .PD 0 | |
2373 | .RS | |
2374 | .TP 24 | |
d1007d14 | 2375 | .B PERF_MEM_LOCK_NA |
2b538c3e MK |
2376 | Not available |
2377 | .TP | |
d1007d14 | 2378 | .B PERF_MEM_LOCK_LOCKED |
2b538c3e MK |
2379 | Locked transaction |
2380 | .RE | |
2381 | .PD | |
d1007d14 VW |
2382 | .TP |
2383 | .I mem_dtlb | |
bc9d90b5 VW |
2384 | TLB access hit or miss, a bitwise combination of the following, shifted |
2385 | left by | |
2386 | .BR PERF_MEM_TLB_SHIFT : | |
2b538c3e MK |
2387 | |
2388 | .PD 0 | |
2389 | .RS | |
2390 | .TP 24 | |
d1007d14 | 2391 | .B PERF_MEM_TLB_NA |
2b538c3e MK |
2392 | Not available |
2393 | .TP | |
d1007d14 | 2394 | .B PERF_MEM_TLB_HIT |
2b538c3e MK |
2395 | Hit |
2396 | .TP | |
d1007d14 | 2397 | .B PERF_MEM_TLB_MISS |
2b538c3e MK |
2398 | Miss |
2399 | .TP | |
d1007d14 | 2400 | .B PERF_MEM_TLB_L1 |
2b538c3e MK |
2401 | Level 1 TLB |
2402 | .TP | |
d1007d14 | 2403 | .B PERF_MEM_TLB_L2 |
2b538c3e MK |
2404 | Level 2 TLB |
2405 | .TP | |
d1007d14 | 2406 | .B PERF_MEM_TLB_WK |
2b538c3e MK |
2407 | Hardware walker |
2408 | .TP | |
d1007d14 | 2409 | .B PERF_MEM_TLB_OS |
2b538c3e MK |
2410 | OS fault handler |
2411 | .RE | |
2412 | .PD | |
d1007d14 | 2413 | .RE |
1e043959 VW |
2414 | .TP |
2415 | .I transaction | |
2416 | If the | |
2417 | .B PERF_SAMPLE_TRANSACTION | |
37bee118 | 2418 | flag is set, then a 64-bit field is recorded describing |
1e043959 VW |
2419 | the sources of any transactional memory aborts. |
2420 | ||
2421 | The field is a bitwise combination of the following values: | |
2422 | .RS | |
2423 | .TP | |
2424 | .B PERF_TXN_ELISION | |
b3f39642 | 2425 | Abort from an elision type transaction (Intel-CPU-specific). |
1e043959 VW |
2426 | .TP |
2427 | .B PERF_TXN_TRANSACTION | |
b3f39642 | 2428 | Abort from a generic transaction. |
1e043959 VW |
2429 | .TP |
2430 | .B PERF_TXN_SYNC | |
b3f39642 | 2431 | Synchronous abort (related to the reported instruction). |
1e043959 VW |
2432 | .TP |
2433 | .B PERF_TXN_ASYNC | |
b3f39642 | 2434 | Asynchronous abort (not related to the reported instruction). |
1e043959 VW |
2435 | .TP |
2436 | .B PERF_TXN_RETRY | |
053a3e08 | 2437 | Retryable abort (retrying the transaction may have succeeded). |
1e043959 VW |
2438 | .TP |
2439 | .B PERF_TXN_CONFLICT | |
b3f39642 | 2440 | Abort due to memory conflicts with other threads. |
1e043959 VW |
2441 | .TP |
2442 | .B PERF_TXN_CAPACITY_WRITE | |
b3f39642 | 2443 | Abort due to write capacity overflow. |
1e043959 VW |
2444 | .TP |
2445 | .B PERF_TXN_CAPACITY_READ | |
b3f39642 | 2446 | Abort due to read capacity overflow. |
1e043959 | 2447 | .RE |
b3f39642 MK |
2448 | .IP |
2449 | In addition, a user-specified abort code can be obtained from | |
2450 | the high 32 bits of the field by shifting right by | |
1e043959 | 2451 | .B PERF_TXN_ABORT_SHIFT |
4b3a5f01 | 2452 | and masking with the value |
1e043959 | 2453 | .BR PERF_TXN_ABORT_MASK . |
f5281dfd VW |
2454 | .TP |
2455 | .IR abi ", " regs[weight(mask)] | |
2456 | If | |
2457 | .B PERF_SAMPLE_REGS_INTR | |
2458 | is enabled, then the user CPU registers are recorded. | |
2459 | ||
2460 | The | |
2461 | .I abi | |
2462 | field is one of | |
4b3a5f01 MK |
2463 | .BR PERF_SAMPLE_REGS_ABI_NONE , |
2464 | .BR PERF_SAMPLE_REGS_ABI_32 , | |
2465 | or | |
f5281dfd VW |
2466 | .BR PERF_SAMPLE_REGS_ABI_64 . |
2467 | ||
2468 | The | |
2469 | .I regs | |
2470 | field is an array of the CPU registers that were specified by | |
2471 | the | |
2472 | .I sample_regs_intr | |
2473 | attr field. | |
2474 | The number of values is the number of bits set in the | |
2475 | .I sample_regs_intr | |
2476 | bit mask. | |
f2b1d720 | 2477 | .RE |
9bfc542b VW |
2478 | .TP |
2479 | .B PERF_RECORD_MMAP2 | |
2480 | This record includes extended information on | |
2481 | .BR mmap (2) | |
2482 | calls returning executable mappings. | |
2483 | The format is similar to that of the | |
2484 | .B PERF_RECORD_MMAP | |
3a058284 | 2485 | record, but includes extra values that allow uniquely identifying |
9bfc542b | 2486 | shared mappings. |
3a058284 | 2487 | |
9bfc542b VW |
2488 | .in +4n |
2489 | .nf | |
2490 | struct { | |
2491 | struct perf_event_header header; | |
7a10da70 MK |
2492 | u32 pid; |
2493 | u32 tid; | |
2494 | u64 addr; | |
2495 | u64 len; | |
2496 | u64 pgoff; | |
2497 | u32 maj; | |
2498 | u32 min; | |
2499 | u64 ino; | |
2500 | u64 ino_generation; | |
2501 | u32 prot; | |
2502 | u32 flags; | |
2503 | char filename[]; | |
9bfc542b VW |
2504 | struct sample_id sample_id; |
2505 | }; | |
2506 | .fi | |
2507 | .RS | |
2508 | .TP | |
2509 | .I pid | |
3a058284 | 2510 | is the process ID. |
9bfc542b VW |
2511 | .TP |
2512 | .I tid | |
3a058284 | 2513 | is the thread ID. |
9bfc542b VW |
2514 | .TP |
2515 | .I addr | |
2516 | is the address of the allocated memory. | |
2517 | .TP | |
2518 | .I len | |
2519 | is the length of the allocated memory. | |
2520 | .TP | |
2521 | .I pgoff | |
2522 | is the page offset of the allocated memory. | |
2523 | .TP | |
2524 | .I maj | |
3a058284 | 2525 | is the major ID of the underlying device. |
9bfc542b VW |
2526 | .TP |
2527 | .I min | |
3a058284 | 2528 | is the minor ID of the underlying device. |
9bfc542b VW |
2529 | .TP |
2530 | .I ino | |
3a058284 | 2531 | is the inode number. |
9bfc542b VW |
2532 | .TP |
2533 | .I ino_generation | |
2534 | is the inode generation. | |
2535 | .TP | |
2536 | .I prot | |
2537 | is the protection information. | |
2538 | .TP | |
2539 | .I flags | |
2540 | is the flags information. | |
2541 | .TP | |
2542 | .I filename | |
2543 | is a string describing the backing of the allocated memory. | |
2544 | .RE | |
1fda209c VW |
2545 | .TP |
2546 | .BR PERF_RECORD_AUX " (since Linux 4.1)" | |
2547 | \" commit 68db7e98c3a6ebe7284b6cf14906ed7c55f3f7f0 | |
2548 | This record reports that new data is available in the separate | |
2549 | AUX buffer region. | |
2550 | ||
2551 | .in +4n | |
2552 | .nf | |
2553 | struct { | |
2554 | struct perf_event_header header; | |
7a10da70 MK |
2555 | u64 aux_offset; |
2556 | u64 aux_size; | |
2557 | u64 flags; | |
1fda209c VW |
2558 | struct sample_id sample_id; |
2559 | }; | |
2560 | .fi | |
2561 | .RS | |
2562 | .TP | |
2563 | .I aux_offset | |
2564 | offset in the AUX mmap region where the new data begins. | |
2565 | .TP | |
2566 | .I aux_size | |
2567 | size of the data made available. | |
2568 | .TP | |
2569 | .I flags | |
95655a22 | 2570 | describes the AUX update. |
1fda209c VW |
2571 | .RS |
2572 | .TP | |
2573 | .B PERF_AUX_FLAG_TRUNCATED | |
95655a22 | 2574 | if set, then the data returned was truncated to fit the available |
1fda209c | 2575 | buffer size. |
b1355f6a VW |
2576 | .TP |
2577 | .B PERF_AUX_FLAG_OVERWRITE | |
2578 | .\" commit 2023a0d2829e521fe6ad6b9907f3f90bfbf57142 | |
95655a22 | 2579 | if set, then the data returned has overwritten previous data. |
1fda209c VW |
2580 | .RE |
2581 | .RE | |
6932aac3 VW |
2582 | .TP |
2583 | .BR PERF_RECORD_ITRACE_START " (since Linux 4.1)" | |
2584 | \" ec0d7729bbaed4b9d2d3fada693278e13a3d1368 | |
2585 | This record indicates which process has initiated an instruction | |
2586 | trace event, allowing tools to properly correlate the instruction | |
2587 | addresses in the AUX buffer with the proper executable. | |
2588 | ||
2589 | .in +4n | |
2590 | .nf | |
2591 | struct { | |
2592 | struct perf_event_header header; | |
7a10da70 MK |
2593 | u32 pid; |
2594 | u32 tid; | |
6932aac3 VW |
2595 | }; |
2596 | .fi | |
2597 | .RS | |
2598 | .TP | |
2599 | .I pid | |
95655a22 | 2600 | process ID of the thread starting an instruction trace. |
6932aac3 VW |
2601 | .TP |
2602 | .I tid | |
95655a22 | 2603 | thread ID of the thread starting an instruction trace. |
6932aac3 | 2604 | .RE |
46012ba3 DH |
2605 | .TP |
2606 | .BR PERF_RECORD_LOST_SAMPLES " (since Linux 4.2)" | |
2607 | \" f38b0dbb491a6987e198aa6b428db8692a6480f8 | |
2608 | When using hardware sampling (such as Intel PEBS) this record | |
4199d3a1 | 2609 | indicates some number of samples that may have been lost. |
46012ba3 DH |
2610 | |
2611 | .in +4n | |
2612 | .nf | |
2613 | struct { | |
2614 | struct perf_event_header header; | |
7a10da70 | 2615 | u64 lost; |
46012ba3 DH |
2616 | struct sample_id sample_id; |
2617 | }; | |
2618 | .fi | |
2619 | .RS | |
2620 | .TP | |
2621 | .I lost | |
2622 | the number of potentially lost samples. | |
2623 | .RE | |
9277a75d VW |
2624 | .TP |
2625 | .BR PERF_RECORD_SWITCH " (since Linux 4.3)" | |
2626 | \" commit 45ac1403f564f411c6a383a2448688ba8dd705a4 | |
2627 | This record indicates a context switch has happened. | |
2628 | The | |
2629 | .B PERF_RECORD_MISC_SWITCH_OUT | |
2630 | bit in the | |
2631 | .I misc | |
2632 | field indicates whether it was a context switch into | |
2633 | or away from the current process. | |
2634 | ||
2635 | .in +4n | |
2636 | .nf | |
2637 | struct { | |
2638 | struct perf_event_header header; | |
2639 | struct sample_id sample_id; | |
2640 | }; | |
2641 | .fi | |
2642 | .TP | |
2643 | .BR PERF_RECORD_SWITCH_CPU_WIDE " (since Linux 4.3)" | |
2644 | \" commit 45ac1403f564f411c6a383a2448688ba8dd705a4 | |
2645 | As with | |
2646 | .B PERF_RECORD_SWITCH | |
2647 | this record indicates a context switch has happened, | |
d5a24378 | 2648 | but it only occurs when sampling in CPU-wide mode |
9277a75d VW |
2649 | and provides additional information on the process |
2650 | being switched to/from. | |
2651 | The | |
2652 | .B PERF_RECORD_MISC_SWITCH_OUT | |
2653 | bit in the | |
2654 | .I misc | |
2655 | field indicates whether it was a context switch into | |
2656 | or away from the current process. | |
2657 | ||
2658 | .in +4n | |
2659 | .nf | |
2660 | struct { | |
2661 | struct perf_event_header header; | |
2662 | u32 next_prev_pid; | |
2663 | u32 next_prev_tid; | |
2664 | struct sample_id sample_id; | |
2665 | }; | |
2666 | .fi | |
2667 | .RS | |
2668 | .TP | |
2669 | .I next_prev_pid | |
d5a24378 | 2670 | The process ID of the previous (if switching in) |
9277a75d VW |
2671 | or next (if switching out) process on the CPU. |
2672 | .TP | |
2673 | .I next_prev_tid | |
d5a24378 | 2674 | The thread ID of the previous (if switching in) |
9277a75d VW |
2675 | or next (if switching out) thread on the CPU. |
2676 | .RE | |
f2b1d720 | 2677 | .RE |
21977c9d VW |
2678 | .SS Overflow handling |
2679 | Events can be set to notify when a threshold is crossed, | |
2680 | indicating an overflow. | |
2681 | Overflow conditions can be captured by monitoring the | |
2682 | event file descriptor with | |
f2b1d720 MK |
2683 | .BR poll (2), |
2684 | .BR select (2), | |
21977c9d | 2685 | or |
4b3a5f01 | 2686 | .BR epoll (7). |
6831ba6b MK |
2687 | Alternatively, the overflow events can be captured via sa signal handler, |
2688 | by enabling I/O signaling on the file descriptor; see the discussion of the | |
fc79d996 | 2689 | .BR F_SETOWN |
6831ba6b MK |
2690 | and |
2691 | .BR F_SETSIG | |
2692 | operations in | |
2693 | .BR fcntl (2). | |
f2b1d720 | 2694 | |
6170255e | 2695 | Overflows are generated only by sampling events |
f2b1d720 | 2696 | .RI ( sample_period |
7d182bb6 | 2697 | must have a nonzero value). |
f2b1d720 | 2698 | |
21977c9d | 2699 | There are two ways to generate overflow notifications. |
f2b1d720 MK |
2700 | |
2701 | The first is to set a | |
2702 | .I wakeup_events | |
2703 | or | |
2704 | .I wakeup_watermark | |
21977c9d | 2705 | value that will trigger if a certain number of samples |
f2b1d720 | 2706 | or bytes have been written to the mmap ring buffer. |
fc79d996 | 2707 | In this case, |
7db515ef | 2708 | .B POLL_IN |
21977c9d | 2709 | is indicated. |
f2b1d720 MK |
2710 | |
2711 | The other way is by use of the | |
7db515ef | 2712 | .B PERF_EVENT_IOC_REFRESH |
f2b1d720 MK |
2713 | ioctl. |
2714 | This ioctl adds to a counter that decrements each time the event overflows. | |
21977c9d | 2715 | When nonzero, |
7db515ef | 2716 | .B POLL_IN |
21977c9d VW |
2717 | is indicated, but |
2718 | once the counter reaches 0 | |
7db515ef | 2719 | .B POLL_HUP |
21977c9d | 2720 | is indicated and |
f2b1d720 MK |
2721 | the underlying event is disabled. |
2722 | ||
50e4319c VW |
2723 | Refreshing an event group leader refreshes all siblings and |
2724 | refreshing with a parameter of 0 currently enables infinite | |
2725 | refreshes; | |
2726 | these behaviors are unsupported and should not be relied on. | |
2727 | .\" See https://lkml.org/lkml/2011/5/24/337 | |
2728 | ||
4010bc07 | 2729 | Starting with Linux 3.18, |
747a6e7c | 2730 | .\" commit 179033b3e064d2cd3f5f9945e76b0a0f0fbf4883 |
21977c9d VW |
2731 | .B POLL_HUP |
2732 | is indicated if the event being monitored is attached to a different | |
2733 | process and that process exits. | |
73d8cece | 2734 | .SS rdpmc instruction |
f2b1d720 | 2735 | Starting with Linux 3.4 on x86, you can use the |
747a6e7c | 2736 | .\" commit c7206205d00ab375839bd6c7ddb247d600693c09 |
f2b1d720 MK |
2737 | .I rdpmc |
2738 | instruction to get low-latency reads without having to enter the kernel. | |
2739 | Note that using | |
2740 | .I rdpmc | |
2741 | is not necessarily faster than other methods for reading event values. | |
2742 | ||
2743 | Support for this can be detected with the | |
2744 | .I cap_usr_rdpmc | |
2745 | field in the mmap page; documentation on how | |
2746 | to calculate event values can be found in that section. | |
562c69f6 VW |
2747 | |
2748 | Originally, when rdpmc support was enabled, any process (not just ones | |
2749 | with an active perf event) could use the rdpmc instruction to access | |
2750 | the counters. | |
fc79d996 | 2751 | Starting with Linux 4.0, |
562c69f6 VW |
2752 | .\" 7911d3f7af14a614617e38245fedf98a724e46a9 |
2753 | rdpmc support is only allowed if an event is currently enabled | |
95655a22 | 2754 | in a process's context. |
562c69f6 VW |
2755 | To restore the old behavior, write the value 2 to |
2756 | .IR /sys/devices/cpu/rdpmc . | |
73d8cece | 2757 | .SS perf_event ioctl calls |
f2b1d720 MK |
2758 | .PP |
2759 | Various ioctls act on | |
7db515ef | 2760 | .BR perf_event_open () |
ce88f77b | 2761 | file descriptors: |
f2b1d720 MK |
2762 | .TP |
2763 | .B PERF_EVENT_IOC_ENABLE | |
ce88f77b | 2764 | This enables the individual event or event group specified by the |
7db515ef | 2765 | file descriptor argument. |
f2b1d720 | 2766 | |
51700fd7 | 2767 | If the |
8cc8b90d | 2768 | .B PERF_IOC_FLAG_GROUP |
51700fd7 | 2769 | bit is set in the ioctl argument, then all events in a group are |
dbc01ecd VW |
2770 | enabled, even if the event specified is not the group leader |
2771 | (but see BUGS). | |
f2b1d720 MK |
2772 | .TP |
2773 | .B PERF_EVENT_IOC_DISABLE | |
ce88f77b | 2774 | This disables the individual counter or event group specified by the |
7db515ef | 2775 | file descriptor argument. |
f2b1d720 MK |
2776 | |
2777 | Enabling or disabling the leader of a group enables or disables the | |
2778 | entire group; that is, while the group leader is disabled, none of the | |
2779 | counters in the group will count. | |
33a0ccb2 MK |
2780 | Enabling or disabling a member of a group other than the leader |
2781 | affects only that counter; disabling a non-leader | |
f2b1d720 MK |
2782 | stops that counter from counting but doesn't affect any other counter. |
2783 | ||
51700fd7 | 2784 | If the |
8cc8b90d | 2785 | .B PERF_IOC_FLAG_GROUP |
51700fd7 | 2786 | bit is set in the ioctl argument, then all events in a group are |
dbc01ecd VW |
2787 | disabled, even if the event specified is not the group leader |
2788 | (but see BUGS). | |
f2b1d720 MK |
2789 | .TP |
2790 | .B PERF_EVENT_IOC_REFRESH | |
2791 | Non-inherited overflow counters can use this | |
2792 | to enable a counter for a number of overflows specified by the argument, | |
2793 | after which it is disabled. | |
2794 | Subsequent calls of this ioctl add the argument value to the current | |
2795 | count. | |
21977c9d | 2796 | An overflow notification with |
7db515ef MK |
2797 | .B POLL_IN |
2798 | set will happen on each overflow until the | |
21977c9d VW |
2799 | count reaches 0; when that happens a notification with |
2800 | .B POLL_HUP | |
7db515ef | 2801 | set is sent and the event is disabled. |
f2b1d720 | 2802 | Using an argument of 0 is considered undefined behavior. |
f2b1d720 MK |
2803 | .TP |
2804 | .B PERF_EVENT_IOC_RESET | |
36127c0e | 2805 | Reset the event count specified by the |
6061d29f | 2806 | file descriptor argument to zero. |
33a0ccb2 | 2807 | This resets only the counts; there is no way to reset the |
f2b1d720 MK |
2808 | multiplexing |
2809 | .I time_enabled | |
2810 | or | |
2811 | .I time_running | |
2812 | values. | |
f2b1d720 | 2813 | |
51700fd7 | 2814 | If the |
8cc8b90d | 2815 | .B PERF_IOC_FLAG_GROUP |
51700fd7 | 2816 | bit is set in the ioctl argument, then all events in a group are |
dbc01ecd VW |
2817 | reset, even if the event specified is not the group leader |
2818 | (but see BUGS). | |
f2b1d720 MK |
2819 | .TP |
2820 | .B PERF_EVENT_IOC_PERIOD | |
e6cf5694 | 2821 | This updates the overflow period for the event. |
3f118a29 | 2822 | |
747a6e7c VW |
2823 | Since Linux 3.7 (on ARM) |
2824 | .\" commit 3581fe0ef37ce12ac7a4f74831168352ae848edc | |
2825 | and Linux 3.14 (all other architectures), | |
2826 | .\" commit bad7192b842c83e580747ca57104dd51fe08c223 | |
3f118a29 | 2827 | the new period takes effect immediately. |
ed81fdd9 | 2828 | On older kernels, the new period did not take effect until |
3f118a29 | 2829 | after the next overflow. |
f2b1d720 MK |
2830 | |
2831 | The argument is a pointer to a 64-bit value containing the | |
2832 | desired new period. | |
e6cf5694 | 2833 | |
fc79d996 | 2834 | Prior to Linux 2.6.36, |
747a6e7c VW |
2835 | .\" commit ad0cf3478de8677f720ee06393b3147819568d6a |
2836 | this ioctl always failed due to a bug | |
e6cf5694 | 2837 | in the kernel. |
f2b1d720 MK |
2838 | .TP |
2839 | .B PERF_EVENT_IOC_SET_OUTPUT | |
2840 | This tells the kernel to report event notifications to the specified | |
2841 | file descriptor rather than the default one. | |
2842 | The file descriptors must all be on the same CPU. | |
2843 | ||
2844 | The argument specifies the desired file descriptor, or \-1 if | |
2845 | output should be ignored. | |
f2b1d720 | 2846 | .TP |
31c1f2b0 | 2847 | .BR PERF_EVENT_IOC_SET_FILTER " (since Linux 2.6.33)" |
60dafbc1 | 2848 | .\" commit 6fb2915df7f0747d9044da9dbff5b46dc2e20830 |
f2b1d720 MK |
2849 | This adds an ftrace filter to this event. |
2850 | ||
2851 | The argument is a pointer to the desired ftrace filter. | |
a0dcc8dd | 2852 | .TP |
31c1f2b0 | 2853 | .BR PERF_EVENT_IOC_ID " (since Linux 3.12)" |
60dafbc1 | 2854 | .\" commit cf4957f17f2a89984915ea808876d9c82225b862 |
bec6277e | 2855 | This returns the event ID value for the given event file descriptor. |
a0dcc8dd VW |
2856 | |
2857 | The argument is a pointer to a 64-bit unsigned integer | |
2858 | to hold the result. | |
b0f7b411 VW |
2859 | .TP |
2860 | .BR PERF_EVENT_IOC_SET_BPF " (since Linux 4.1)" | |
2861 | .\" commit 2541517c32be2531e0da59dfd7efc1ce844644f5 | |
2862 | This allows attaching a Berkeley Packet Filter (BPF) | |
2863 | program to an existing kprobe tracepoint event. | |
2864 | You need | |
2865 | .B CAP_SYS_ADMIN | |
2866 | privileges to use this ioctl. | |
2867 | ||
2868 | The argument is a BPF program file descriptor that was created by | |
2869 | a previous | |
2870 | .BR bpf (2) | |
2871 | system call. | |
fc79d996 | 2872 | .SS Using prctl(2) |
f2b1d720 MK |
2873 | A process can enable or disable all the event groups that are |
2874 | attached to it using the | |
2875 | .BR prctl (2) | |
2876 | .B PR_TASK_PERF_EVENTS_ENABLE | |
2877 | and | |
2878 | .B PR_TASK_PERF_EVENTS_DISABLE | |
2879 | operations. | |
ee7b0cbf | 2880 | This applies to all counters on the calling process, whether created by |
f2b1d720 MK |
2881 | this process or by another, and does not affect any counters that this |
2882 | process has created on other processes. | |
33a0ccb2 | 2883 | It enables or disables only |
f2b1d720 | 2884 | the group leaders, not any other members in the groups. |
f2b1d720 | 2885 | .SS perf_event related configuration files |
dc9ec146 | 2886 | |
7db515ef MK |
2887 | Files in |
2888 | .I /proc/sys/kernel/ | |
7db515ef | 2889 | .RS 4 |
f2b1d720 | 2890 | .TP |
7db515ef | 2891 | .I /proc/sys/kernel/perf_event_paranoid |
f2b1d720 MK |
2892 | The |
2893 | .I perf_event_paranoid | |
2894 | file can be set to restrict access to the performance counters. | |
dc9ec146 MK |
2895 | |
2896 | .PD 0 | |
2b538c3e MK |
2897 | .RS |
2898 | .IP 2 4 | |
3eb95192 | 2899 | allow only user-space measurements (default since Linux 4.6). |
b5eb75f7 | 2900 | .\" default changed in commit 0161028b7c8aebef64194d3d73e43bc3b53b5c66 |
2b538c3e | 2901 | .IP 1 |
3eb95192 | 2902 | allow both kernel and user measurements (default before Linux 4.6). |
2b538c3e MK |
2903 | .IP 0 |
2904 | allow access to CPU-specific data but not raw tracepoint samples. | |
2905 | .IP \-1 | |
2906 | no restrictions. | |
2907 | .RE | |
dc9ec146 | 2908 | .PD |
2b538c3e | 2909 | .IP |
f2b1d720 MK |
2910 | The existence of the |
2911 | .I perf_event_paranoid | |
2912 | file is the official method for determining if a kernel supports | |
7db515ef | 2913 | .BR perf_event_open (). |
f2b1d720 MK |
2914 | .TP |
2915 | .I /proc/sys/kernel/perf_event_max_sample_rate | |
7db515ef MK |
2916 | This sets the maximum sample rate. |
2917 | Setting this too high can allow | |
f2b1d720 | 2918 | users to sample at a rate that impacts overall machine performance |
7db515ef MK |
2919 | and potentially lock up the machine. |
2920 | The default value is | |
f2b1d720 | 2921 | 100000 (samples per second). |
fd133d5d VW |
2922 | .TP |
2923 | .I /proc/sys/kernel/perf_event_max_stack | |
2924 | .\" Introduced in c5dfd78eb79851e278b7973031b9ca363da87a7e | |
5dd3feec | 2925 | This file sets the maximum depth of stack frame entries reported |
fd133d5d | 2926 | when generating a call trace. |
f2b1d720 MK |
2927 | .TP |
2928 | .I /proc/sys/kernel/perf_event_mlock_kb | |
ce88f77b MK |
2929 | Maximum number of pages an unprivileged user can |
2930 | .BR mlock (2). | |
f2b1d720 MK |
2931 | The default is 516 (kB). |
2932 | .RE | |
dc9ec146 | 2933 | |
7db515ef MK |
2934 | Files in |
2935 | .I /sys/bus/event_source/devices/ | |
dc9ec146 | 2936 | |
7db515ef | 2937 | .RS 4 |
ce88f77b | 2938 | Since Linux 2.6.34, the kernel supports having multiple PMUs |
f2b1d720 MK |
2939 | available for monitoring. |
2940 | Information on how to program these PMUs can be found under | |
2941 | .IR /sys/bus/event_source/devices/ . | |
2942 | Each subdirectory corresponds to a different PMU. | |
f2b1d720 | 2943 | .TP |
31c1f2b0 | 2944 | .IR /sys/bus/event_source/devices/*/type " (since Linux 2.6.38)" |
747a6e7c | 2945 | .\" commit abe43400579d5de0078c2d3a760e6598e183f871 |
f2b1d720 MK |
2946 | This contains an integer that can be used in the |
2947 | .I type | |
ce88f77b MK |
2948 | field of |
2949 | .I perf_event_attr | |
2950 | to indicate that you wish to use this PMU. | |
f2b1d720 | 2951 | .TP |
562c69f6 | 2952 | .IR /sys/bus/event_source/devices/cpu/rdpmc " (since Linux 3.4)" |
747a6e7c | 2953 | .\" commit 0c9d42ed4cee2aa1dfc3a260b741baae8615744f |
8a94e783 | 2954 | If this file is 1, then direct user-space access to the |
e30dc77f VW |
2955 | performance counter registers is allowed via the rdpmc instruction. |
2956 | This can be disabled by echoing 0 to the file. | |
562c69f6 VW |
2957 | |
2958 | As of Linux 4.0 | |
2959 | .\" a66734297f78707ce39d756b656bfae861d53f62 | |
2960 | .\" 7911d3f7af14a614617e38245fedf98a724e46a9 | |
2961 | the behavior has changed, so that 1 now means only allow access | |
2962 | to processes with active perf events, with 2 indicating the old | |
2963 | allow-anyone-access behavior. | |
f2b1d720 | 2964 | .TP |
31c1f2b0 | 2965 | .IR /sys/bus/event_source/devices/*/format/ " (since Linux 3.4)" |
747a6e7c | 2966 | .\" commit 641cc938815dfd09f8fa1ec72deb814f0938ac33 |
7d182bb6 MK |
2967 | This subdirectory contains information on the architecture-specific |
2968 | subfields available for programming the various | |
f2b1d720 | 2969 | .I config |
ce88f77b MK |
2970 | fields in the |
2971 | .I perf_event_attr | |
2972 | struct. | |
e30dc77f VW |
2973 | |
2974 | The content of each file is the name of the config field, followed | |
2975 | by a colon, followed by a series of integer bit ranges separated by | |
2976 | commas. | |
8a94e783 | 2977 | For example, the file |
e30dc77f VW |
2978 | .I event |
2979 | may contain the value | |
2980 | .I config1:1,6-10,44 | |
2981 | which indicates that event is an attribute that occupies bits 1,6-10, and 44 | |
ce88f77b MK |
2982 | of |
2983 | .IR perf_event_attr::config1 . | |
e30dc77f | 2984 | .TP |
31c1f2b0 | 2985 | .IR /sys/bus/event_source/devices/*/events/ " (since Linux 3.4)" |
747a6e7c | 2986 | .\" commit 641cc938815dfd09f8fa1ec72deb814f0938ac33 |
7d182bb6 | 2987 | This subdirectory contains files with predefined events. |
f2b1d720 | 2988 | The contents are strings describing the event settings |
e30dc77f | 2989 | expressed in terms of the fields found in the previously mentioned |
f2b1d720 MK |
2990 | .I ./format/ |
2991 | directory. | |
2992 | These are not necessarily complete lists of all events supported by | |
2993 | a PMU, but usually a subset of events deemed useful or interesting. | |
e30dc77f VW |
2994 | |
2995 | The content of each file is a list of attribute names | |
8a94e783 MK |
2996 | separated by commas. |
2997 | Each entry has an optional value (either hex or decimal). | |
37bee118 | 2998 | If no value is specified, then it is assumed to be a single-bit |
e30dc77f VW |
2999 | field with a value of 1. |
3000 | An example entry may look like this: | |
699893d8 | 3001 | .IR event=0x2,inv,ldlat=3 . |
f2b1d720 MK |
3002 | .TP |
3003 | .I /sys/bus/event_source/devices/*/uevent | |
e30dc77f VW |
3004 | This file is the standard kernel device interface |
3005 | for injecting hotplug events. | |
3006 | .TP | |
31c1f2b0 | 3007 | .IR /sys/bus/event_source/devices/*/cpumask " (since Linux 3.7)" |
747a6e7c | 3008 | .\" commit 314d9f63f385096580e9e2a06eaa0745d92fe4ac |
699893d8 DP |
3009 | The |
3010 | .I cpumask | |
3011 | file contains a comma-separated list of integers that | |
3012 | indicate a representative CPU number for each socket (package) | |
e30dc77f VW |
3013 | on the motherboard. |
3014 | This is needed when setting up uncore or northbridge events, as | |
3015 | those PMUs present socket-wide events. | |
f2b1d720 | 3016 | .RE |
47297adb | 3017 | .SH RETURN VALUE |
f2b1d720 MK |
3018 | .BR perf_event_open () |
3019 | returns the new file descriptor, or \-1 if an error occurred | |
3020 | (in which case, | |
3021 | .I errno | |
3022 | is set appropriately). | |
3023 | .SH ERRORS | |
d8b7d950 VW |
3024 | The errors returned by |
3025 | .BR perf_event_open () | |
3026 | can be inconsistent, and may | |
3027 | vary across processor architectures and performance monitoring units. | |
f2b1d720 | 3028 | .TP |
82b09254 | 3029 | .B E2BIG |
ce88f77b MK |
3030 | Returned if the |
3031 | .I perf_event_attr | |
82b09254 VW |
3032 | .I size |
3033 | value is too small | |
3034 | (smaller than | |
3035 | .BR PERF_ATTR_SIZE_VER0 ), | |
3036 | too big (larger than the page size), | |
3037 | or larger than the kernel supports and the extra bytes are not zero. | |
3038 | When | |
3039 | .B E2BIG | |
ce88f77b MK |
3040 | is returned, the |
3041 | .I perf_event_attr | |
e9bd9b2c | 3042 | .I size |
d6af98f8 | 3043 | field is overwritten by the kernel to be the size of the structure |
82b09254 VW |
3044 | it was expecting. |
3045 | .TP | |
d8b7d950 | 3046 | .B EACCES |
27f0af8e VW |
3047 | Returned when the requested event requires |
3048 | .B CAP_SYS_ADMIN | |
3049 | permissions (or a more permissive perf_event paranoid setting). | |
3050 | Some common cases where an unprivileged process | |
3051 | may encounter this error: | |
3052 | attaching to a process owned by a different user; | |
2b23ecbd MK |
3053 | monitoring all processes on a given CPU (i.e., specifying the |
3054 | .I pid | |
3055 | argument as \-1); | |
079928f3 | 3056 | and not setting |
accec051 | 3057 | .I exclude_kernel |
079928f3 | 3058 | when the paranoid setting requires it. |
d8b7d950 VW |
3059 | .TP |
3060 | .B EBADF | |
3061 | Returned if the | |
3062 | .I group_fd | |
accec051 MK |
3063 | file descriptor is not valid, or, if |
3064 | .B PERF_FLAG_PID_CGROUP | |
3065 | is set, | |
d8b7d950 VW |
3066 | the cgroup file descriptor in |
3067 | .I pid | |
3068 | is not valid. | |
3069 | .TP | |
f27486cb VW |
3070 | .BR EBUSY " (since Linux 4.1)" |
3071 | .\" bed5b25ad9c8a2f5d735ef0bc746ec870c01c1b0 | |
3072 | Returned if another event already has exclusive | |
3073 | access to the PMU. | |
3074 | .TP | |
d8b7d950 VW |
3075 | .B EFAULT |
3076 | Returned if the | |
3077 | .I attr | |
3078 | pointer points at an invalid memory address. | |
3079 | .TP | |
f2b1d720 | 3080 | .B EINVAL |
d8b7d950 VW |
3081 | Returned if the specified event is invalid. |
3082 | There are many possible reasons for this. | |
3083 | A not-exhaustive list: | |
3084 | .I sample_freq | |
accec051 | 3085 | is higher than the maximum setting; |
d8b7d950 VW |
3086 | the |
3087 | .I cpu | |
accec051 | 3088 | to monitor does not exist; |
d8b7d950 | 3089 | .I read_format |
accec051 | 3090 | is out of range; |
d8b7d950 | 3091 | .I sample_type |
accec051 | 3092 | is out of range; |
d8b7d950 VW |
3093 | the |
3094 | .I flags | |
accec051 | 3095 | value is out of range; |
d8b7d950 VW |
3096 | .I exclusive |
3097 | or | |
3098 | .I pinned | |
accec051 | 3099 | set and the event is not a group leader; |
d8b7d950 VW |
3100 | the event |
3101 | .I config | |
accec051 MK |
3102 | values are out of range or set reserved bits; |
3103 | the generic event selected is not supported; or | |
d8b7d950 VW |
3104 | there is not enough room to add the selected event. |
3105 | .TP | |
3106 | .B EMFILE | |
3107 | Each opened event uses one file descriptor. | |
26c32fab MK |
3108 | If a large number of events are opened, |
3109 | the per-process limit on the number of open file descriptors will be reached, | |
3110 | and no more events can be created. | |
d8b7d950 VW |
3111 | .TP |
3112 | .B ENODEV | |
3113 | Returned when the event involves a feature not supported | |
accec051 | 3114 | by the current CPU. |
d8b7d950 VW |
3115 | .TP |
3116 | .B ENOENT | |
3117 | Returned if the | |
3118 | .I type | |
3119 | setting is not valid. | |
accec051 | 3120 | This error is also returned for |
d8b7d950 | 3121 | some unsupported generic events. |
f2b1d720 MK |
3122 | .TP |
3123 | .B ENOSPC | |
3124 | Prior to Linux 3.3, if there was not enough room for the event, | |
747a6e7c | 3125 | .\" commit aa2bc1ade59003a379ffc485d6da2d92ea3370a6 |
f2b1d720 MK |
3126 | .B ENOSPC |
3127 | was returned. | |
accec051 | 3128 | In Linux 3.3, this was changed to |
f2b1d720 MK |
3129 | .BR EINVAL . |
3130 | .B ENOSPC | |
d8b7d950 | 3131 | is still returned if you try to add more breakpoint events |
accec051 | 3132 | than supported by the hardware. |
d8b7d950 VW |
3133 | .TP |
3134 | .B ENOSYS | |
3135 | Returned if | |
3136 | .B PERF_SAMPLE_STACK_USER | |
3137 | is set in | |
3138 | .I sample_type | |
3139 | and it is not supported by hardware. | |
3140 | .TP | |
3141 | .B EOPNOTSUPP | |
3142 | Returned if an event requiring a specific hardware feature is | |
3143 | requested but there is no hardware support. | |
3144 | This includes requesting low-skid events if not supported, | |
3145 | branch tracing if it is not available, sampling if no PMU | |
3146 | interrupt is available, and branch stacks for software events. | |
3147 | .TP | |
fd133d5d VW |
3148 | .BR EOVERFLOW " (since Linux 4.8)" |
3149 | .\" 97c79a38cd454602645f0470ffb444b3b75ce574 | |
3150 | Returned if | |
3151 | .B PERF_SAMPLE_CALLCHAIN | |
3152 | is requested and | |
3153 | .I sample_max_stack | |
3154 | is larger than the maximum specified in | |
3155 | .IR /proc/sys/kernel/perf_event_max_stack . | |
3156 | .TP | |
d8b7d950 | 3157 | .B EPERM |
27f0af8e VW |
3158 | Returned on many (but not all) architectures when an unsupported |
3159 | .IR exclude_hv ", " exclude_idle ", " exclude_user ", or " exclude_kernel | |
3160 | setting is specified. | |
3161 | ||
3162 | It can also happen, as with | |
3163 | .BR EACCES , | |
3164 | when the requested event requires | |
3165 | .B CAP_SYS_ADMIN | |
3166 | permissions (or a more permissive perf_event paranoid setting). | |
3167 | This includes setting a breakpoint on a kernel address, | |
3168 | and (since Linux 3.13) setting a kernel function-trace tracepoint. | |
747a6e7c | 3169 | .\" commit a4e95fc2cbb31d70a65beffeaf8773f881328c34 |
d8b7d950 VW |
3170 | .TP |
3171 | .B ESRCH | |
3172 | Returned if attempting to attach to a process that does not exist. | |
f2b1d720 | 3173 | .SH VERSION |
f2b1d720 MK |
3174 | .BR perf_event_open () |
3175 | was introduced in Linux 2.6.31 but was called | |
747a6e7c | 3176 | .\" commit 0793a61d4df8daeac6492dbf8d2f3e5713caae5e |
ffd4dec0 | 3177 | .BR perf_counter_open (). |
f2b1d720 | 3178 | It was renamed in Linux 2.6.32. |
747a6e7c | 3179 | .\" commit cdd6c482c9ff9c55475ee7392ec8f672eddb7be6 |
f2b1d720 | 3180 | .SH CONFORMING TO |
7db515ef MK |
3181 | This |
3182 | .BR perf_event_open () | |
dc9ec146 | 3183 | system call Linux-specific |
f2b1d720 | 3184 | and should not be used in programs intended to be portable. |
f2b1d720 MK |
3185 | .SH NOTES |
3186 | Glibc does not provide a wrapper for this system call; call it using | |
3187 | .BR syscall (2). | |
7db515ef | 3188 | See the example below. |
f2b1d720 MK |
3189 | |
3190 | The official way of knowing if | |
7db515ef | 3191 | .BR perf_event_open () |
f2b1d720 MK |
3192 | support is enabled is checking |
3193 | for the existence of the file | |
7db515ef | 3194 | .IR /proc/sys/kernel/perf_event_paranoid . |
f2b1d720 | 3195 | .SH BUGS |
f2b1d720 MK |
3196 | The |
3197 | .B F_SETOWN_EX | |
3198 | option to | |
7db515ef | 3199 | .BR fcntl (2) |
f2b1d720 MK |
3200 | is needed to properly get overflow signals in threads. |
3201 | This was introduced in Linux 2.6.32. | |
747a6e7c | 3202 | .\" commit ba0a6c9f6fceed11c6a99e8326f0477fe383e6b5 |
f2b1d720 | 3203 | |
747a6e7c VW |
3204 | Prior to Linux 2.6.33 (at least for x86), |
3205 | .\" commit b690081d4d3f6a23541493f1682835c3cd5c54a1 | |
3206 | the kernel did not check | |
f2b1d720 MK |
3207 | if events could be scheduled together until read time. |
3208 | The same happens on all known kernels if the NMI watchdog is enabled. | |
3209 | This means to see if a given set of events works you have to | |
3210 | .BR perf_event_open (), | |
3211 | start, then read before you know for sure you | |
3212 | can get valid measurements. | |
3213 | ||
b5190152 MK |
3214 | Prior to Linux 2.6.34, |
3215 | .\" FIXME . cannot find a kernel commit for this one | |
3216 | event constraints were not enforced by the kernel. | |
f2b1d720 MK |
3217 | In that case, some events would silently return "0" if the kernel |
3218 | scheduled them in an improper counter slot. | |
3219 | ||
ce88f77b | 3220 | Prior to Linux 2.6.34, there was a bug when multiplexing where the |
f2b1d720 | 3221 | wrong results could be returned. |
747a6e7c | 3222 | .\" commit 45e16a6834b6af098702e5ea6c9a40de42ff77d8 |
f2b1d720 MK |
3223 | |
3224 | Kernels from Linux 2.6.35 to Linux 2.6.39 can quickly crash the kernel if | |
3225 | "inherit" is enabled and many threads are started. | |
747a6e7c | 3226 | .\" commit 38b435b16c36b0d863efcf3f07b34a6fac9873fd |
f2b1d720 MK |
3227 | |
3228 | Prior to Linux 2.6.35, | |
747a6e7c | 3229 | .\" commit 050735b08ca8a016bbace4445fa025b88fee770b |
f2b1d720 MK |
3230 | .B PERF_FORMAT_GROUP |
3231 | did not work with attached processes. | |
3232 | ||
f2b1d720 MK |
3233 | There is a bug in the kernel code between |
3234 | Linux 2.6.36 and Linux 3.0 that ignores the | |
3235 | "watermark" field and acts as if a wakeup_event | |
3236 | was chosen if the union has a | |
7d182bb6 | 3237 | nonzero value in it. |
747a6e7c | 3238 | .\" commit 4ec8363dfc1451f8c8f86825731fe712798ada02 |
f2b1d720 | 3239 | |
8a94e783 | 3240 | From Linux 2.6.31 to Linux 3.4, the |
dbc01ecd VW |
3241 | .B PERF_IOC_FLAG_GROUP |
3242 | ioctl argument was broken and would repeatedly operate | |
3243 | on the event specified rather than iterating across | |
3244 | all sibling events in a group. | |
747a6e7c | 3245 | .\" commit 724b6daa13e100067c30cfc4d1ad06629609dc4e |
dbc01ecd | 3246 | |
7205b8df | 3247 | From Linux 3.4 to Linux 3.11, the mmap |
747a6e7c | 3248 | .\" commit fa7315871046b9a4c48627905691dbde57e51033 |
135cba8b VW |
3249 | .I cap_usr_rdpmc |
3250 | and | |
3251 | .I cap_usr_time | |
3252 | bits mapped to the same location. | |
3253 | Code should migrate to the new | |
3254 | .I cap_user_rdpmc | |
3255 | and | |
3256 | .I cap_user_time | |
3257 | fields instead. | |
3258 | ||
7db515ef MK |
3259 | Always double-check your results! |
3260 | Various generalized events have had wrong values. | |
f2b1d720 MK |
3261 | For example, retired branches measured |
3262 | the wrong thing on AMD machines until Linux 2.6.35. | |
747a6e7c | 3263 | .\" commit f287d332ce835f77a4f5077d2c0ef1e3f9ea42d2 |
f2b1d720 MK |
3264 | .SH EXAMPLE |
3265 | The following is a short example that measures the total | |
7db515ef MK |
3266 | instruction count of a call to |
3267 | .BR printf (3). | |
f2b1d720 MK |
3268 | .nf |
3269 | ||
3270 | #include <stdlib.h> | |
3271 | #include <stdio.h> | |
3272 | #include <unistd.h> | |
3273 | #include <string.h> | |
3274 | #include <sys/ioctl.h> | |
3275 | #include <linux/perf_event.h> | |
3276 | #include <asm/unistd.h> | |
3277 | ||
571767ca | 3278 | static long |
7db515ef MK |
3279 | perf_event_open(struct perf_event_attr *hw_event, pid_t pid, |
3280 | int cpu, int group_fd, unsigned long flags) | |
f2b1d720 MK |
3281 | { |
3282 | int ret; | |
3283 | ||
7db515ef MK |
3284 | ret = syscall(__NR_perf_event_open, hw_event, pid, cpu, |
3285 | group_fd, flags); | |
f2b1d720 MK |
3286 | return ret; |
3287 | } | |
3288 | ||
f2b1d720 MK |
3289 | int |
3290 | main(int argc, char **argv) | |
3291 | { | |
f2b1d720 MK |
3292 | struct perf_event_attr pe; |
3293 | long long count; | |
3294 | int fd; | |
3295 | ||
3296 | memset(&pe, 0, sizeof(struct perf_event_attr)); | |
3297 | pe.type = PERF_TYPE_HARDWARE; | |
3298 | pe.size = sizeof(struct perf_event_attr); | |
3299 | pe.config = PERF_COUNT_HW_INSTRUCTIONS; | |
3300 | pe.disabled = 1; | |
3301 | pe.exclude_kernel = 1; | |
3302 | pe.exclude_hv = 1; | |
3303 | ||
3304 | fd = perf_event_open(&pe, 0, \-1, \-1, 0); | |
7db515ef | 3305 | if (fd == \-1) { |
f2b1d720 | 3306 | fprintf(stderr, "Error opening leader %llx\\n", pe.config); |
7db515ef | 3307 | exit(EXIT_FAILURE); |
f2b1d720 MK |
3308 | } |
3309 | ||
3310 | ioctl(fd, PERF_EVENT_IOC_RESET, 0); | |
3311 | ioctl(fd, PERF_EVENT_IOC_ENABLE, 0); | |
3312 | ||
3313 | printf("Measuring instruction count for this printf\\n"); | |
3314 | ||
3315 | ioctl(fd, PERF_EVENT_IOC_DISABLE, 0); | |
3316 | read(fd, &count, sizeof(long long)); | |
3317 | ||
3318 | printf("Used %lld instructions\\n", count); | |
3319 | ||
3320 | close(fd); | |
3321 | } | |
3322 | .fi | |
47297adb | 3323 | .SH SEE ALSO |
f2b1d720 MK |
3324 | .BR fcntl (2), |
3325 | .BR mmap (2), | |
3326 | .BR open (2), | |
3327 | .BR prctl (2), | |
3328 | .BR read (2) |