Loading...
1perf-record(1)
2==============
3
4NAME
5----
6perf-record - Run a command and record its profile into perf.data
7
8SYNOPSIS
9--------
10[verse]
11'perf record' [-e <EVENT> | --event=EVENT] [-a] <command>
12'perf record' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
13
14DESCRIPTION
15-----------
16This command runs a command and gathers a performance counter profile
17from it, into perf.data - without displaying anything.
18
19This file can then be inspected later on, using 'perf report'.
20
21
22OPTIONS
23-------
24<command>...::
25 Any command you can specify in a shell.
26
27-e::
28--event=::
29 Select the PMU event. Selection can be:
30
31 - a symbolic event name (use 'perf list' to list all events)
32
33 - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
34 hexadecimal event descriptor.
35
36 - a symbolic or raw PMU event followed by an optional colon
37 and a list of event modifiers, e.g., cpu-cycles:p. See the
38 linkperf:perf-list[1] man page for details on event modifiers.
39
40 - a symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
41 'param1', 'param2', etc are defined as formats for the PMU in
42 /sys/bus/event_source/devices/<pmu>/format/*.
43
44 - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
45
46 where M, N, K are numbers (in decimal, hex, octal format). Acceptable
47 values for each of 'config', 'config1' and 'config2' are defined by
48 corresponding entries in /sys/bus/event_source/devices/<pmu>/format/*
49 param1 and param2 are defined as formats for the PMU in:
50 /sys/bus/event_source/devices/<pmu>/format/*
51
52 There are also some parameters which are not defined in .../<pmu>/format/*.
53 These params can be used to overload default config values per event.
54 Here are some common parameters:
55 - 'period': Set event sampling period
56 - 'freq': Set event sampling frequency
57 - 'time': Disable/enable time stamping. Acceptable values are 1 for
58 enabling time stamping. 0 for disabling time stamping.
59 The default is 1.
60 - 'call-graph': Disable/enable callgraph. Acceptable str are "fp" for
61 FP mode, "dwarf" for DWARF mode, "lbr" for LBR mode and
62 "no" for disable callgraph.
63 - 'stack-size': user stack size for dwarf mode
64 - 'name' : User defined event name. Single quotes (') may be used to
65 escape symbols in the name from parsing by shell and tool
66 like this: name=\'CPU_CLK_UNHALTED.THREAD:cmask=0x1\'.
67 - 'aux-output': Generate AUX records instead of events. This requires
68 that an AUX area event is also provided.
69 - 'aux-sample-size': Set sample size for AUX area sampling. If the
70 '--aux-sample' option has been used, set aux-sample-size=0 to disable
71 AUX area sampling for the event.
72
73 See the linkperf:perf-list[1] man page for more parameters.
74
75 Note: If user explicitly sets options which conflict with the params,
76 the value set by the parameters will be overridden.
77
78 Also not defined in .../<pmu>/format/* are PMU driver specific
79 configuration parameters. Any configuration parameter preceded by
80 the letter '@' is not interpreted in user space and sent down directly
81 to the PMU driver. For example:
82
83 perf record -e some_event/@cfg1,@cfg2=config/ ...
84
85 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
86 with the event for further processing. There is no restriction on
87 what the configuration parameters are, as long as their semantic is
88 understood and supported by the PMU driver.
89
90 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
91 where addr is the address in memory you want to break in.
92 Access is the memory access type (read, write, execute) it can
93 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
94 number of bytes from specified addr, which the breakpoint will cover.
95 If you want to profile read-write accesses in 0x1000, just set
96 'mem:0x1000:rw'.
97 If you want to profile write accesses in [0x1000~1008), just set
98 'mem:0x1000/8:w'.
99
100 - a BPF source file (ending in .c) or a precompiled object file (ending
101 in .o) selects one or more BPF events.
102 The BPF program can attach to various perf events based on the ELF section
103 names.
104
105 When processing a '.c' file, perf searches an installed LLVM to compile it
106 into an object file first. Optional clang options can be passed via the
107 '--clang-opt' command line option, e.g.:
108
109 perf record --clang-opt "-DLINUX_VERSION_CODE=0x50000" \
110 -e tests/bpf-script-example.c
111
112 Note: '--clang-opt' must be placed before '--event/-e'.
113
114 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
115 Each event is separated by commas and the group should be quoted to
116 prevent the shell interpretation. You also need to use --group on
117 "perf report" to view group events together.
118
119--filter=<filter>::
120 Event filter. This option should follow an event selector (-e) which
121 selects either tracepoint event(s) or a hardware trace PMU
122 (e.g. Intel PT or CoreSight).
123
124 - tracepoint filters
125
126 In the case of tracepoints, multiple '--filter' options are combined
127 using '&&'.
128
129 - address filters
130
131 A hardware trace PMU advertises its ability to accept a number of
132 address filters by specifying a non-zero value in
133 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
134
135 Address filters have the format:
136
137 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
138
139 Where:
140 - 'filter': defines a region that will be traced.
141 - 'start': defines an address at which tracing will begin.
142 - 'stop': defines an address at which tracing will stop.
143 - 'tracestop': defines a region in which tracing will stop.
144
145 <file name> is the name of the object file, <start> is the offset to the
146 code to trace in that file, and <size> is the size of the region to
147 trace. 'start' and 'stop' filters need not specify a <size>.
148
149 If no object file is specified then the kernel is assumed, in which case
150 the start address must be a current kernel memory address.
151
152 <start> can also be specified by providing the name of a symbol. If the
153 symbol name is not unique, it can be disambiguated by inserting #n where
154 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
155 select only a global symbol. <size> can also be specified by providing
156 the name of a symbol, in which case the size is calculated to the end
157 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
158 omitted and <start> is a symbol, then the size is calculated to the end
159 of that symbol.
160
161 If <size> is omitted and <start> is '*', then the start and size will
162 be calculated from the first and last symbols, i.e. to trace the whole
163 file.
164
165 If symbol names (or '*') are provided, they must be surrounded by white
166 space.
167
168 The filter passed to the kernel is not necessarily the same as entered.
169 To see the filter that is passed, use the -v option.
170
171 The kernel may not be able to configure a trace region if it is not
172 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
173 examined to determine if that is a possibility.
174
175 Multiple filters can be separated with space or comma.
176
177--exclude-perf::
178 Don't record events issued by perf itself. This option should follow
179 an event selector (-e) which selects tracepoint event(s). It adds a
180 filter expression 'common_pid != $PERFPID' to filters. If other
181 '--filter' exists, the new filter expression will be combined with
182 them by '&&'.
183
184-a::
185--all-cpus::
186 System-wide collection from all CPUs (default if no target is specified).
187
188-p::
189--pid=::
190 Record events on existing process ID (comma separated list).
191
192-t::
193--tid=::
194 Record events on existing thread ID (comma separated list).
195 This option also disables inheritance by default. Enable it by adding
196 --inherit.
197
198-u::
199--uid=::
200 Record events in threads owned by uid. Name or number.
201
202-r::
203--realtime=::
204 Collect data with this RT SCHED_FIFO priority.
205
206--no-buffering::
207 Collect data without buffering.
208
209-c::
210--count=::
211 Event period to sample.
212
213-o::
214--output=::
215 Output file name.
216
217-i::
218--no-inherit::
219 Child tasks do not inherit counters.
220
221-F::
222--freq=::
223 Profile at this frequency. Use 'max' to use the currently maximum
224 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
225 sysctl. Will throttle down to the currently maximum allowed frequency.
226 See --strict-freq.
227
228--strict-freq::
229 Fail if the specified frequency can't be used.
230
231-m::
232--mmap-pages=::
233 Number of mmap data pages (must be a power of two) or size
234 specification with appended unit character - B/K/M/G. The
235 size is rounded up to have nearest pages power of two value.
236 Also, by adding a comma, the number of mmap pages for AUX
237 area tracing can be specified.
238
239--group::
240 Put all events in a single event group. This precedes the --event
241 option and remains only for backward compatibility. See --event.
242
243-g::
244 Enables call-graph (stack chain/backtrace) recording for both
245 kernel space and user space.
246
247--call-graph::
248 Setup and enable call-graph (stack chain/backtrace) recording,
249 implies -g. Default is "fp" (for user space).
250
251 The unwinding method used for kernel space is dependent on the
252 unwinder used by the active kernel configuration, i.e
253 CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)
254
255 Any option specified here controls the method used for user space.
256
257 Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
258 Call Frame Information) or "lbr" (Hardware Last Branch Record
259 facility).
260
261 In some systems, where binaries are build with gcc
262 --fomit-frame-pointer, using the "fp" method will produce bogus
263 call graphs, using "dwarf", if available (perf tools linked to
264 the libunwind or libdw library) should be used instead.
265 Using the "lbr" method doesn't require any compiler options. It
266 will produce call graphs from the hardware LBR registers. The
267 main limitation is that it is only available on new Intel
268 platforms, such as Haswell. It can only get user call chain. It
269 doesn't work with branch stack sampling at the same time.
270
271 When "dwarf" recording is used, perf also records (user) stack dump
272 when sampled. Default size of the stack dump is 8192 (bytes).
273 User can change the size by passing the size after comma like
274 "--call-graph dwarf,4096".
275
276-q::
277--quiet::
278 Don't print any message, useful for scripting.
279
280-v::
281--verbose::
282 Be more verbose (show counter open errors, etc).
283
284-s::
285--stat::
286 Record per-thread event counts. Use it with 'perf report -T' to see
287 the values.
288
289-d::
290--data::
291 Record the sample virtual addresses.
292
293--phys-data::
294 Record the sample physical addresses.
295
296--data-page-size::
297 Record the sampled data address data page size.
298
299--code-page-size::
300 Record the sampled code address (ip) page size
301
302-T::
303--timestamp::
304 Record the sample timestamps. Use it with 'perf report -D' to see the
305 timestamps, for instance.
306
307-P::
308--period::
309 Record the sample period.
310
311--sample-cpu::
312 Record the sample cpu.
313
314-n::
315--no-samples::
316 Don't sample.
317
318-R::
319--raw-samples::
320Collect raw sample records from all opened counters (default for tracepoint counters).
321
322-C::
323--cpu::
324Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
325comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
326In per-thread mode with inheritance mode on (default), samples are captured only when
327the thread executes on the designated CPUs. Default is to monitor all CPUs.
328
329-B::
330--no-buildid::
331Do not save the build ids of binaries in the perf.data files. This skips
332post processing after recording, which sometimes makes the final step in
333the recording process to take a long time, as it needs to process all
334events looking for mmap records. The downside is that it can misresolve
335symbols if the workload binaries used when recording get locally rebuilt
336or upgraded, because the only key available in this case is the
337pathname. You can also set the "record.build-id" config variable to
338'skip to have this behaviour permanently.
339
340-N::
341--no-buildid-cache::
342Do not update the buildid cache. This saves some overhead in situations
343where the information in the perf.data file (which includes buildids)
344is sufficient. You can also set the "record.build-id" config variable to
345'no-cache' to have the same effect.
346
347-G name,...::
348--cgroup name,...::
349monitor only in the container (cgroup) called "name". This option is available only
350in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
351container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
352can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
353to first event, second cgroup to second event and so on. It is possible to provide
354an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
355corresponding events, i.e., they always refer to events defined earlier on the command
356line. If the user wants to track multiple events for a specific cgroup, the user can
357use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
358
359If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
360command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
361
362-b::
363--branch-any::
364Enable taken branch stack sampling. Any type of taken branch may be sampled.
365This is a shortcut for --branch-filter any. See --branch-filter for more infos.
366
367-j::
368--branch-filter::
369Enable taken branch stack sampling. Each sample captures a series of consecutive
370taken branches. The number of branches captured with each sample depends on the
371underlying hardware, the type of branches of interest, and the executed code.
372It is possible to select the types of branches captured by enabling filters. The
373following filters are defined:
374
375 - any: any type of branches
376 - any_call: any function call or system call
377 - any_ret: any function return or system call return
378 - ind_call: any indirect branch
379 - call: direct calls, including far (to/from kernel) calls
380 - u: only when the branch target is at the user level
381 - k: only when the branch target is in the kernel
382 - hv: only when the target is at the hypervisor level
383 - in_tx: only when the target is in a hardware transaction
384 - no_tx: only when the target is not in a hardware transaction
385 - abort_tx: only when the target is a hardware transaction abort
386 - cond: conditional branches
387 - save_type: save branch type during sampling in case binary is not available later
388
389+
390The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
391The privilege levels may be omitted, in which case, the privilege levels of the associated
392event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
393levels are subject to permissions. When sampling on multiple events, branch stack sampling
394is enabled for all the sampling events. The sampled branch type is the same for all events.
395The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
396Note that this feature may not be available on all processors.
397
398--weight::
399Enable weightened sampling. An additional weight is recorded per sample and can be
400displayed with the weight and local_weight sort keys. This currently works for TSX
401abort events and some memory events in precise mode on modern Intel CPUs.
402
403--namespaces::
404Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.
405
406--all-cgroups::
407Record events of type PERF_RECORD_CGROUP. This enables 'cgroup' sort key.
408
409--transaction::
410Record transaction flags for transaction related events.
411
412--per-thread::
413Use per-thread mmaps. By default per-cpu mmaps are created. This option
414overrides that and uses per-thread mmaps. A side-effect of that is that
415inheritance is automatically disabled. --per-thread is ignored with a warning
416if combined with -a or -C options.
417
418-D::
419--delay=::
420After starting the program, wait msecs before measuring (-1: start with events
421disabled). This is useful to filter out the startup phase of the program, which
422is often very different.
423
424-I::
425--intr-regs::
426Capture machine state (registers) at interrupt, i.e., on counter overflows for
427each sample. List of captured registers depends on the architecture. This option
428is off by default. It is possible to select the registers to sample using their
429symbolic names, e.g. on x86, ax, si. To list the available registers use
430--intr-regs=\?. To name registers, pass a comma separated list such as
431--intr-regs=ax,bx. The list of register is architecture dependent.
432
433--user-regs::
434Similar to -I, but capture user registers at sample time. To list the available
435user registers use --user-regs=\?.
436
437--running-time::
438Record running and enabled time for read events (:S)
439
440-k::
441--clockid::
442Sets the clock id to use for the various time fields in the perf_event_type
443records. See clock_gettime(). In particular CLOCK_MONOTONIC and
444CLOCK_MONOTONIC_RAW are supported, some events might also allow
445CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
446
447-S::
448--snapshot::
449Select AUX area tracing Snapshot Mode. This option is valid only with an
450AUX area tracing event. Optionally, certain snapshot capturing parameters
451can be specified in a string that follows this option:
452 'e': take one last snapshot on exit; guarantees that there is at least one
453 snapshot in the output file;
454 <size>: if the PMU supports this, specify the desired snapshot size.
455
456In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
457and on exit if the above 'e' option is given.
458
459--aux-sample[=OPTIONS]::
460Select AUX area sampling. At least one of the events selected by the -e option
461must be an AUX area event. Samples on other events will be created containing
462data from the AUX area. Optionally sample size may be specified, otherwise it
463defaults to 4KiB.
464
465--proc-map-timeout::
466When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
467because the file may be huge. A time out is needed in such cases.
468This option sets the time out limit. The default value is 500 ms.
469
470--switch-events::
471Record context switch events i.e. events of type PERF_RECORD_SWITCH or
472PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT or CoreSight)
473switch events will be enabled automatically, which can be suppressed by
474by the option --no-switch-events.
475
476--clang-path=PATH::
477Path to clang binary to use for compiling BPF scriptlets.
478(enabled when BPF support is on)
479
480--clang-opt=OPTIONS::
481Options passed to clang when compiling BPF scriptlets.
482(enabled when BPF support is on)
483
484--vmlinux=PATH::
485Specify vmlinux path which has debuginfo.
486(enabled when BPF prologue is on)
487
488--buildid-all::
489Record build-id of all DSOs regardless whether it's actually hit or not.
490
491--buildid-mmap::
492Record build ids in mmap2 events, disables build id cache (implies --no-buildid).
493
494--aio[=n]::
495Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
496Asynchronous mode is supported only when linking Perf tool with libc library
497providing implementation for Posix AIO API.
498
499--affinity=mode::
500Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
501 node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
502 cpu - thread affinity mask is set to cpu of the processed mmap buffer
503
504--mmap-flush=number::
505
506Specify minimal number of bytes that is extracted from mmap data pages and
507processed for output. One can specify the number using B/K/M/G suffixes.
508
509The maximal allowed value is a quarter of the size of mmaped data pages.
510
511The default option value is 1 byte which means that every time that the output
512writing thread finds some new data in the mmaped buffer the data is extracted,
513possibly compressed (-z) and written to the output, perf.data or pipe.
514
515Larger data chunks are compressed more effectively in comparison to smaller
516chunks so extraction of larger chunks from the mmap data pages is preferable
517from the perspective of output size reduction.
518
519Also at some cases executing less output write syscalls with bigger data size
520can take less time than executing more output write syscalls with smaller data
521size thus lowering runtime profiling overhead.
522
523-z::
524--compression-level[=n]::
525Produce compressed trace using specified level n (default: 1 - fastest compression,
52622 - smallest trace)
527
528--all-kernel::
529Configure all used events to run in kernel space.
530
531--all-user::
532Configure all used events to run in user space.
533
534--kernel-callchains::
535Collect callchains only from kernel space. I.e. this option sets
536perf_event_attr.exclude_callchain_user to 1.
537
538--user-callchains::
539Collect callchains only from user space. I.e. this option sets
540perf_event_attr.exclude_callchain_kernel to 1.
541
542Don't use both --kernel-callchains and --user-callchains at the same time or no
543callchains will be collected.
544
545--timestamp-filename
546Append timestamp to output file name.
547
548--timestamp-boundary::
549Record timestamp boundary (time of first/last samples).
550
551--switch-output[=mode]::
552Generate multiple perf.data files, timestamp prefixed, switching to a new one
553based on 'mode' value:
554 "signal" - when receiving a SIGUSR2 (default value) or
555 <size> - when reaching the size threshold, size is expected to
556 be a number with appended unit character - B/K/M/G
557 <time> - when reaching the time threshold, size is expected to
558 be a number with appended unit character - s/m/h/d
559
560 Note: the precision of the size threshold hugely depends
561 on your configuration - the number and size of your ring
562 buffers (-m). It is generally more precise for higher sizes
563 (like >5M), for lower values expect different sizes.
564
565A possible use case is to, given an external event, slice the perf.data file
566that gets then processed, possibly via a perf script, to decide if that
567particular perf.data snapshot should be kept or not.
568
569Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
570The reason for the latter two is to reduce the data file switching
571overhead. You can still switch them on with:
572
573 --switch-output --no-no-buildid --no-no-buildid-cache
574
575--switch-output-event::
576Events that will cause the switch of the perf.data file, auto-selecting
577--switch-output=signal, the results are similar as internally the side band
578thread will also send a SIGUSR2 to the main one.
579
580Uses the same syntax as --event, it will just not be recorded, serving only to
581switch the perf.data file as soon as the --switch-output event is processed by
582a separate sideband thread.
583
584This sideband thread is also used to other purposes, like processing the
585PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
586information, etc.
587
588--switch-max-files=N::
589
590When rotating perf.data with --switch-output, only keep N files.
591
592--dry-run::
593Parse options then exit. --dry-run can be used to detect errors in cmdline
594options.
595
596'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
597in config file is set to true.
598
599--tail-synthesize::
600Instead of collecting non-sample events (for example, fork, comm, mmap) at
601the beginning of record, collect them during finalizing an output file.
602The collected non-sample events reflects the status of the system when
603record is finished.
604
605--overwrite::
606Makes all events use an overwritable ring buffer. An overwritable ring
607buffer works like a flight recorder: when it gets full, the kernel will
608overwrite the oldest records, that thus will never make it to the
609perf.data file.
610
611When '--overwrite' and '--switch-output' are used perf records and drops
612events until it receives a signal, meaning that something unusual was
613detected that warrants taking a snapshot of the most current events,
614those fitting in the ring buffer at that moment.
615
616'overwrite' attribute can also be set or canceled for an event using
617config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
618
619Implies --tail-synthesize.
620
621--kcore::
622Make a copy of /proc/kcore and place it into a directory with the perf data file.
623
624--max-size=<size>::
625Limit the sample data max size, <size> is expected to be a number with
626appended unit character - B/K/M/G
627
628--num-thread-synthesize::
629 The number of threads to run when synthesizing events for existing processes.
630 By default, the number of threads equals 1.
631
632ifdef::HAVE_LIBPFM[]
633--pfm-events events::
634Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
635including support for event filters. For example '--pfm-events
636inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
637option using the comma separator. Hardware events and generic hardware
638events cannot be mixed together. The latter must be used with the -e
639option. The -e option and this one can be mixed and matched. Events
640can be grouped using the {} notation.
641endif::HAVE_LIBPFM[]
642
643--control=fifo:ctl-fifo[,ack-fifo]::
644--control=fd:ctl-fd[,ack-fd]::
645ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
646Listen on ctl-fd descriptor for command to control measurement.
647
648Available commands:
649 'enable' : enable events
650 'disable' : disable events
651 'enable name' : enable event 'name'
652 'disable name' : disable event 'name'
653 'snapshot' : AUX area tracing snapshot).
654 'stop' : stop perf record
655 'ping' : ping
656
657 'evlist [-v|-g|-F] : display all events
658 -F Show just the sample frequency used for each event.
659 -v Show all fields.
660 -g Show event group information.
661
662Measurements can be started with events disabled using --delay=-1 option. Optionally
663send control command completion ('ack\n') to ack-fd descriptor to synchronize with the
664controlling process. Example of bash shell script to enable and disable events during
665measurements:
666
667 #!/bin/bash
668
669 ctl_dir=/tmp/
670
671 ctl_fifo=${ctl_dir}perf_ctl.fifo
672 test -p ${ctl_fifo} && unlink ${ctl_fifo}
673 mkfifo ${ctl_fifo}
674 exec {ctl_fd}<>${ctl_fifo}
675
676 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
677 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
678 mkfifo ${ctl_ack_fifo}
679 exec {ctl_fd_ack}<>${ctl_ack_fifo}
680
681 perf record -D -1 -e cpu-cycles -a \
682 --control fd:${ctl_fd},${ctl_fd_ack} \
683 -- sleep 30 &
684 perf_pid=$!
685
686 sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
687 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
688
689 exec {ctl_fd_ack}>&-
690 unlink ${ctl_ack_fifo}
691
692 exec {ctl_fd}>&-
693 unlink ${ctl_fifo}
694
695 wait -n ${perf_pid}
696 exit $?
697
698include::intel-hybrid.txt[]
699
700SEE ALSO
701--------
702linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]
1perf-record(1)
2==============
3
4NAME
5----
6perf-record - Run a command and record its profile into perf.data
7
8SYNOPSIS
9--------
10[verse]
11'perf record' [-e <EVENT> | --event=EVENT] [-a] <command>
12'perf record' [-e <EVENT> | --event=EVENT] [-a] \-- <command> [<options>]
13
14DESCRIPTION
15-----------
16This command runs a command and gathers a performance counter profile
17from it, into perf.data - without displaying anything.
18
19This file can then be inspected later on, using 'perf report'.
20
21
22OPTIONS
23-------
24<command>...::
25 Any command you can specify in a shell.
26
27-e::
28--event=::
29 Select the PMU event. Selection can be:
30
31 - a symbolic event name (use 'perf list' to list all events)
32
33 - a raw PMU event in the form of rN where N is a hexadecimal value
34 that represents the raw register encoding with the layout of the
35 event control registers as described by entries in
36 /sys/bus/event_source/devices/cpu/format/*.
37
38 - a symbolic or raw PMU event followed by an optional colon
39 and a list of event modifiers, e.g., cpu-cycles:p. See the
40 linkperf:perf-list[1] man page for details on event modifiers.
41
42 - a symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
43 'param1', 'param2', etc are defined as formats for the PMU in
44 /sys/bus/event_source/devices/<pmu>/format/*.
45
46 - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
47
48 where M, N, K are numbers (in decimal, hex, octal format). Acceptable
49 values for each of 'config', 'config1' and 'config2' are defined by
50 corresponding entries in /sys/bus/event_source/devices/<pmu>/format/*
51 param1 and param2 are defined as formats for the PMU in:
52 /sys/bus/event_source/devices/<pmu>/format/*
53
54 There are also some parameters which are not defined in .../<pmu>/format/*.
55 These params can be used to overload default config values per event.
56 Here are some common parameters:
57 - 'period': Set event sampling period
58 - 'freq': Set event sampling frequency
59 - 'time': Disable/enable time stamping. Acceptable values are 1 for
60 enabling time stamping. 0 for disabling time stamping.
61 The default is 1.
62 - 'call-graph': Disable/enable callgraph. Acceptable str are "fp" for
63 FP mode, "dwarf" for DWARF mode, "lbr" for LBR mode and
64 "no" for disable callgraph.
65 - 'stack-size': user stack size for dwarf mode
66 - 'name' : User defined event name. Single quotes (') may be used to
67 escape symbols in the name from parsing by shell and tool
68 like this: name=\'CPU_CLK_UNHALTED.THREAD:cmask=0x1\'.
69 - 'aux-output': Generate AUX records instead of events. This requires
70 that an AUX area event is also provided.
71 - 'aux-sample-size': Set sample size for AUX area sampling. If the
72 '--aux-sample' option has been used, set aux-sample-size=0 to disable
73 AUX area sampling for the event.
74
75 See the linkperf:perf-list[1] man page for more parameters.
76
77 Note: If user explicitly sets options which conflict with the params,
78 the value set by the parameters will be overridden.
79
80 Also not defined in .../<pmu>/format/* are PMU driver specific
81 configuration parameters. Any configuration parameter preceded by
82 the letter '@' is not interpreted in user space and sent down directly
83 to the PMU driver. For example:
84
85 perf record -e some_event/@cfg1,@cfg2=config/ ...
86
87 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
88 with the event for further processing. There is no restriction on
89 what the configuration parameters are, as long as their semantic is
90 understood and supported by the PMU driver.
91
92 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
93 where addr is the address in memory you want to break in.
94 Access is the memory access type (read, write, execute) it can
95 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
96 number of bytes from specified addr, which the breakpoint will cover.
97 If you want to profile read-write accesses in 0x1000, just set
98 'mem:0x1000:rw'.
99 If you want to profile write accesses in [0x1000~1008), just set
100 'mem:0x1000/8:w'.
101
102 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
103 Each event is separated by commas and the group should be quoted to
104 prevent the shell interpretation. You also need to use --group on
105 "perf report" to view group events together.
106
107--filter=<filter>::
108 Event filter. This option should follow an event selector (-e).
109 If the event is a tracepoint, the filter string will be parsed by
110 the kernel. If the event is a hardware trace PMU (e.g. Intel PT
111 or CoreSight), it'll be processed as an address filter. Otherwise
112 it means a general filter using BPF which can be applied for any
113 kind of event.
114
115 - tracepoint filters
116
117 In the case of tracepoints, multiple '--filter' options are combined
118 using '&&'.
119
120 - address filters
121
122 A hardware trace PMU advertises its ability to accept a number of
123 address filters by specifying a non-zero value in
124 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
125
126 Address filters have the format:
127
128 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
129
130 Where:
131 - 'filter': defines a region that will be traced.
132 - 'start': defines an address at which tracing will begin.
133 - 'stop': defines an address at which tracing will stop.
134 - 'tracestop': defines a region in which tracing will stop.
135
136 <file name> is the name of the object file, <start> is the offset to the
137 code to trace in that file, and <size> is the size of the region to
138 trace. 'start' and 'stop' filters need not specify a <size>.
139
140 If no object file is specified then the kernel is assumed, in which case
141 the start address must be a current kernel memory address.
142
143 <start> can also be specified by providing the name of a symbol. If the
144 symbol name is not unique, it can be disambiguated by inserting #n where
145 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
146 select only a global symbol. <size> can also be specified by providing
147 the name of a symbol, in which case the size is calculated to the end
148 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
149 omitted and <start> is a symbol, then the size is calculated to the end
150 of that symbol.
151
152 If <size> is omitted and <start> is '*', then the start and size will
153 be calculated from the first and last symbols, i.e. to trace the whole
154 file.
155
156 If symbol names (or '*') are provided, they must be surrounded by white
157 space.
158
159 The filter passed to the kernel is not necessarily the same as entered.
160 To see the filter that is passed, use the -v option.
161
162 The kernel may not be able to configure a trace region if it is not
163 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
164 examined to determine if that is a possibility.
165
166 Multiple filters can be separated with space or comma.
167
168 - bpf filters
169
170 A BPF filter can access the sample data and make a decision based on the
171 data. Users need to set an appropriate sample type to use the BPF
172 filter. BPF filters need root privilege.
173
174 The sample data field can be specified in lower case letter. Multiple
175 filters can be separated with comma. For example,
176
177 --filter 'period > 1000, cpu == 1'
178 or
179 --filter 'mem_op == load || mem_op == store, mem_lvl > l1'
180
181 The former filter only accept samples with period greater than 1000 AND
182 CPU number is 1. The latter one accepts either load and store memory
183 operations but it should have memory level above the L1. Since the
184 mem_op and mem_lvl fields come from the (memory) data_source, it'd only
185 work with some events which set the data_source field.
186
187 Also user should request to collect that information (with -d option in
188 the above case). Otherwise, the following message will be shown.
189
190 $ sudo perf record -e cycles --filter 'mem_op == load'
191 Error: cycles event does not have PERF_SAMPLE_DATA_SRC
192 Hint: please add -d option to perf record.
193 failed to set filter "BPF" on event cycles with 22 (Invalid argument)
194
195 Essentially the BPF filter expression is:
196
197 <term> <operator> <value> (("," | "||") <term> <operator> <value>)*
198
199 The <term> can be one of:
200 ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
201 code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
202 p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
203 mem_dtlb, mem_blk, mem_hops, uid, gid
204
205 The <operator> can be one of:
206 ==, !=, >, >=, <, <=, &
207
208 The <value> can be one of:
209 <number> (for any term)
210 na, load, store, pfetch, exec (for mem_op)
211 l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
212 na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
213 remote (for mem_remote)
214 na, locked (for mem_locked)
215 na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
216 na, by_data, by_addr (for mem_blk)
217 hops0, hops1, hops2, hops3 (for mem_hops)
218
219--exclude-perf::
220 Don't record events issued by perf itself. This option should follow
221 an event selector (-e) which selects tracepoint event(s). It adds a
222 filter expression 'common_pid != $PERFPID' to filters. If other
223 '--filter' exists, the new filter expression will be combined with
224 them by '&&'.
225
226-a::
227--all-cpus::
228 System-wide collection from all CPUs (default if no target is specified).
229
230-p::
231--pid=::
232 Record events on existing process ID (comma separated list).
233
234-t::
235--tid=::
236 Record events on existing thread ID (comma separated list).
237 This option also disables inheritance by default. Enable it by adding
238 --inherit.
239
240-u::
241--uid=::
242 Record events in threads owned by uid. Name or number.
243
244-r::
245--realtime=::
246 Collect data with this RT SCHED_FIFO priority.
247
248--no-buffering::
249 Collect data without buffering.
250
251-c::
252--count=::
253 Event period to sample.
254
255-o::
256--output=::
257 Output file name.
258
259-i::
260--no-inherit::
261 Child tasks do not inherit counters.
262
263-F::
264--freq=::
265 Profile at this frequency. Use 'max' to use the currently maximum
266 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
267 sysctl. Will throttle down to the currently maximum allowed frequency.
268 See --strict-freq.
269
270--strict-freq::
271 Fail if the specified frequency can't be used.
272
273-m::
274--mmap-pages=::
275 Number of mmap data pages (must be a power of two) or size
276 specification in bytes with appended unit character - B/K/M/G.
277 The size is rounded up to the nearest power-of-two page value.
278 By adding a comma, an additional parameter with the same
279 semantics used for the normal mmap areas can be specified for
280 AUX tracing area.
281
282-g::
283 Enables call-graph (stack chain/backtrace) recording for both
284 kernel space and user space.
285
286--call-graph::
287 Setup and enable call-graph (stack chain/backtrace) recording,
288 implies -g. Default is "fp" (for user space).
289
290 The unwinding method used for kernel space is dependent on the
291 unwinder used by the active kernel configuration, i.e
292 CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)
293
294 Any option specified here controls the method used for user space.
295
296 Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
297 Call Frame Information) or "lbr" (Hardware Last Branch Record
298 facility).
299
300 In some systems, where binaries are build with gcc
301 --fomit-frame-pointer, using the "fp" method will produce bogus
302 call graphs, using "dwarf", if available (perf tools linked to
303 the libunwind or libdw library) should be used instead.
304 Using the "lbr" method doesn't require any compiler options. It
305 will produce call graphs from the hardware LBR registers. The
306 main limitation is that it is only available on new Intel
307 platforms, such as Haswell. It can only get user call chain. It
308 doesn't work with branch stack sampling at the same time.
309
310 When "dwarf" recording is used, perf also records (user) stack dump
311 when sampled. Default size of the stack dump is 8192 (bytes).
312 User can change the size by passing the size after comma like
313 "--call-graph dwarf,4096".
314
315 When "fp" recording is used, perf tries to save stack entries
316 up to the number specified in sysctl.kernel.perf_event_max_stack
317 by default. User can change the number by passing it after comma
318 like "--call-graph fp,32".
319
320-q::
321--quiet::
322 Don't print any warnings or messages, useful for scripting.
323
324-v::
325--verbose::
326 Be more verbose (show counter open errors, etc).
327
328-s::
329--stat::
330 Record per-thread event counts. Use it with 'perf report -T' to see
331 the values.
332
333-d::
334--data::
335 Record the sample virtual addresses.
336
337--phys-data::
338 Record the sample physical addresses.
339
340--data-page-size::
341 Record the sampled data address data page size.
342
343--code-page-size::
344 Record the sampled code address (ip) page size
345
346-T::
347--timestamp::
348 Record the sample timestamps. Use it with 'perf report -D' to see the
349 timestamps, for instance.
350
351-P::
352--period::
353 Record the sample period.
354
355--sample-cpu::
356 Record the sample cpu.
357
358--sample-identifier::
359 Record the sample identifier i.e. PERF_SAMPLE_IDENTIFIER bit set in
360 the sample_type member of the struct perf_event_attr argument to the
361 perf_event_open system call.
362
363-n::
364--no-samples::
365 Don't sample.
366
367-R::
368--raw-samples::
369Collect raw sample records from all opened counters (default for tracepoint counters).
370
371-C::
372--cpu::
373Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
374comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
375In per-thread mode with inheritance mode on (default), samples are captured only when
376the thread executes on the designated CPUs. Default is to monitor all CPUs.
377
378User space tasks can migrate between CPUs, so when tracing selected CPUs,
379a dummy event is created to track sideband for all CPUs.
380
381-B::
382--no-buildid::
383Do not save the build ids of binaries in the perf.data files. This skips
384post processing after recording, which sometimes makes the final step in
385the recording process to take a long time, as it needs to process all
386events looking for mmap records. The downside is that it can misresolve
387symbols if the workload binaries used when recording get locally rebuilt
388or upgraded, because the only key available in this case is the
389pathname. You can also set the "record.build-id" config variable to
390'skip to have this behaviour permanently.
391
392-N::
393--no-buildid-cache::
394Do not update the buildid cache. This saves some overhead in situations
395where the information in the perf.data file (which includes buildids)
396is sufficient. You can also set the "record.build-id" config variable to
397'no-cache' to have the same effect.
398
399-G name,...::
400--cgroup name,...::
401monitor only in the container (cgroup) called "name". This option is available only
402in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
403container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
404can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
405to first event, second cgroup to second event and so on. It is possible to provide
406an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
407corresponding events, i.e., they always refer to events defined earlier on the command
408line. If the user wants to track multiple events for a specific cgroup, the user can
409use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
410
411If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
412command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
413
414-b::
415--branch-any::
416Enable taken branch stack sampling. Any type of taken branch may be sampled.
417This is a shortcut for --branch-filter any. See --branch-filter for more infos.
418
419-j::
420--branch-filter::
421Enable taken branch stack sampling. Each sample captures a series of consecutive
422taken branches. The number of branches captured with each sample depends on the
423underlying hardware, the type of branches of interest, and the executed code.
424It is possible to select the types of branches captured by enabling filters. The
425following filters are defined:
426
427 - any: any type of branches
428 - any_call: any function call or system call
429 - any_ret: any function return or system call return
430 - ind_call: any indirect branch
431 - ind_jmp: any indirect jump
432 - call: direct calls, including far (to/from kernel) calls
433 - u: only when the branch target is at the user level
434 - k: only when the branch target is in the kernel
435 - hv: only when the target is at the hypervisor level
436 - in_tx: only when the target is in a hardware transaction
437 - no_tx: only when the target is not in a hardware transaction
438 - abort_tx: only when the target is a hardware transaction abort
439 - cond: conditional branches
440 - call_stack: save call stack
441 - no_flags: don't save branch flags e.g prediction, misprediction etc
442 - no_cycles: don't save branch cycles
443 - hw_index: save branch hardware index
444 - save_type: save branch type during sampling in case binary is not available later
445 For the platforms with Intel Arch LBR support (12th-Gen+ client or
446 4th-Gen Xeon+ server), the save branch type is unconditionally enabled
447 when the taken branch stack sampling is enabled.
448 - priv: save privilege state during sampling in case binary is not available later
449 - counter: save occurrences of the event since the last branch entry. Currently, the
450 feature is only supported by a newer CPU, e.g., Intel Sierra Forest and
451 later platforms. An error out is expected if it's used on the unsupported
452 kernel or CPUs.
453
454+
455The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
456The privilege levels may be omitted, in which case, the privilege levels of the associated
457event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
458levels are subject to permissions. When sampling on multiple events, branch stack sampling
459is enabled for all the sampling events. The sampled branch type is the same for all events.
460The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
461Note that this feature may not be available on all processors.
462
463-W::
464--weight::
465Enable weightened sampling. An additional weight is recorded per sample and can be
466displayed with the weight and local_weight sort keys. This currently works for TSX
467abort events and some memory events in precise mode on modern Intel CPUs.
468
469--namespaces::
470Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.
471
472--all-cgroups::
473Record events of type PERF_RECORD_CGROUP. This enables 'cgroup' sort key.
474
475--transaction::
476Record transaction flags for transaction related events.
477
478--per-thread::
479Use per-thread mmaps. By default per-cpu mmaps are created. This option
480overrides that and uses per-thread mmaps. A side-effect of that is that
481inheritance is automatically disabled. --per-thread is ignored with a warning
482if combined with -a or -C options.
483
484-D::
485--delay=::
486After starting the program, wait msecs before measuring (-1: start with events
487disabled), or enable events only for specified ranges of msecs (e.g.
488-D 10-20,30-40 means wait 10 msecs, enable for 10 msecs, wait 10 msecs, enable
489for 10 msecs, then stop). Note, delaying enabling of events is useful to filter
490out the startup phase of the program, which is often very different.
491
492-I::
493--intr-regs::
494Capture machine state (registers) at interrupt, i.e., on counter overflows for
495each sample. List of captured registers depends on the architecture. This option
496is off by default. It is possible to select the registers to sample using their
497symbolic names, e.g. on x86, ax, si. To list the available registers use
498--intr-regs=\?. To name registers, pass a comma separated list such as
499--intr-regs=ax,bx. The list of register is architecture dependent.
500
501--user-regs::
502Similar to -I, but capture user registers at sample time. To list the available
503user registers use --user-regs=\?.
504
505--running-time::
506Record running and enabled time for read events (:S)
507
508-k::
509--clockid::
510Sets the clock id to use for the various time fields in the perf_event_type
511records. See clock_gettime(). In particular CLOCK_MONOTONIC and
512CLOCK_MONOTONIC_RAW are supported, some events might also allow
513CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
514
515-S::
516--snapshot::
517Select AUX area tracing Snapshot Mode. This option is valid only with an
518AUX area tracing event. Optionally, certain snapshot capturing parameters
519can be specified in a string that follows this option:
520
521 - 'e': take one last snapshot on exit; guarantees that there is at least one
522 snapshot in the output file;
523 - <size>: if the PMU supports this, specify the desired snapshot size.
524
525In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
526and on exit if the above 'e' option is given.
527
528--aux-sample[=OPTIONS]::
529Select AUX area sampling. At least one of the events selected by the -e option
530must be an AUX area event. Samples on other events will be created containing
531data from the AUX area. Optionally sample size may be specified, otherwise it
532defaults to 4KiB.
533
534--proc-map-timeout::
535When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
536because the file may be huge. A time out is needed in such cases.
537This option sets the time out limit. The default value is 500 ms.
538
539--switch-events::
540Record context switch events i.e. events of type PERF_RECORD_SWITCH or
541PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT, CoreSight or Arm SPE)
542switch events will be enabled automatically, which can be suppressed by
543by the option --no-switch-events.
544
545--vmlinux=PATH::
546Specify vmlinux path which has debuginfo.
547(enabled when BPF prologue is on)
548
549--buildid-all::
550Record build-id of all DSOs regardless whether it's actually hit or not.
551
552--buildid-mmap::
553Record build ids in mmap2 events, disables build id cache (implies --no-buildid).
554
555--aio[=n]::
556Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
557Asynchronous mode is supported only when linking Perf tool with libc library
558providing implementation for Posix AIO API.
559
560--affinity=mode::
561Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
562
563 - node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
564 - cpu - thread affinity mask is set to cpu of the processed mmap buffer
565
566--mmap-flush=number::
567
568Specify minimal number of bytes that is extracted from mmap data pages and
569processed for output. One can specify the number using B/K/M/G suffixes.
570
571The maximal allowed value is a quarter of the size of mmaped data pages.
572
573The default option value is 1 byte which means that every time that the output
574writing thread finds some new data in the mmaped buffer the data is extracted,
575possibly compressed (-z) and written to the output, perf.data or pipe.
576
577Larger data chunks are compressed more effectively in comparison to smaller
578chunks so extraction of larger chunks from the mmap data pages is preferable
579from the perspective of output size reduction.
580
581Also at some cases executing less output write syscalls with bigger data size
582can take less time than executing more output write syscalls with smaller data
583size thus lowering runtime profiling overhead.
584
585-z::
586--compression-level[=n]::
587Produce compressed trace using specified level n (default: 1 - fastest compression,
58822 - smallest trace)
589
590--all-kernel::
591Configure all used events to run in kernel space.
592
593--all-user::
594Configure all used events to run in user space.
595
596--kernel-callchains::
597Collect callchains only from kernel space. I.e. this option sets
598perf_event_attr.exclude_callchain_user to 1.
599
600--user-callchains::
601Collect callchains only from user space. I.e. this option sets
602perf_event_attr.exclude_callchain_kernel to 1.
603
604Don't use both --kernel-callchains and --user-callchains at the same time or no
605callchains will be collected.
606
607--timestamp-filename
608Append timestamp to output file name.
609
610--timestamp-boundary::
611Record timestamp boundary (time of first/last samples).
612
613--switch-output[=mode]::
614Generate multiple perf.data files, timestamp prefixed, switching to a new one
615based on 'mode' value:
616
617 - "signal" - when receiving a SIGUSR2 (default value) or
618 - <size> - when reaching the size threshold, size is expected to
619 be a number with appended unit character - B/K/M/G
620 - <time> - when reaching the time threshold, size is expected to
621 be a number with appended unit character - s/m/h/d
622
623 Note: the precision of the size threshold hugely depends
624 on your configuration - the number and size of your ring
625 buffers (-m). It is generally more precise for higher sizes
626 (like >5M), for lower values expect different sizes.
627
628A possible use case is to, given an external event, slice the perf.data file
629that gets then processed, possibly via a perf script, to decide if that
630particular perf.data snapshot should be kept or not.
631
632Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
633The reason for the latter two is to reduce the data file switching
634overhead. You can still switch them on with:
635
636 --switch-output --no-no-buildid --no-no-buildid-cache
637
638--switch-output-event::
639Events that will cause the switch of the perf.data file, auto-selecting
640--switch-output=signal, the results are similar as internally the side band
641thread will also send a SIGUSR2 to the main one.
642
643Uses the same syntax as --event, it will just not be recorded, serving only to
644switch the perf.data file as soon as the --switch-output event is processed by
645a separate sideband thread.
646
647This sideband thread is also used to other purposes, like processing the
648PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
649information, etc.
650
651--switch-max-files=N::
652
653When rotating perf.data with --switch-output, only keep N files.
654
655--dry-run::
656Parse options then exit. --dry-run can be used to detect errors in cmdline
657options.
658
659'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
660in config file is set to true.
661
662--synth=TYPE::
663Collect and synthesize given type of events (comma separated). Note that
664this option controls the synthesis from the /proc filesystem which represent
665task status for pre-existing threads.
666
667Kernel (and some other) events are recorded regardless of the
668choice in this option. For example, --synth=no would have MMAP events for
669kernel and modules.
670
671Available types are:
672
673 - 'task' - synthesize FORK and COMM events for each task
674 - 'mmap' - synthesize MMAP events for each process (implies 'task')
675 - 'cgroup' - synthesize CGROUP events for each cgroup
676 - 'all' - synthesize all events (default)
677 - 'no' - do not synthesize any of the above events
678
679--tail-synthesize::
680Instead of collecting non-sample events (for example, fork, comm, mmap) at
681the beginning of record, collect them during finalizing an output file.
682The collected non-sample events reflects the status of the system when
683record is finished.
684
685--overwrite::
686Makes all events use an overwritable ring buffer. An overwritable ring
687buffer works like a flight recorder: when it gets full, the kernel will
688overwrite the oldest records, that thus will never make it to the
689perf.data file.
690
691When '--overwrite' and '--switch-output' are used perf records and drops
692events until it receives a signal, meaning that something unusual was
693detected that warrants taking a snapshot of the most current events,
694those fitting in the ring buffer at that moment.
695
696'overwrite' attribute can also be set or canceled for an event using
697config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
698
699Implies --tail-synthesize.
700
701--kcore::
702Make a copy of /proc/kcore and place it into a directory with the perf data file.
703
704--max-size=<size>::
705Limit the sample data max size, <size> is expected to be a number with
706appended unit character - B/K/M/G
707
708--num-thread-synthesize::
709 The number of threads to run when synthesizing events for existing processes.
710 By default, the number of threads equals 1.
711
712ifdef::HAVE_LIBPFM[]
713--pfm-events events::
714Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
715including support for event filters. For example '--pfm-events
716inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
717option using the comma separator. Hardware events and generic hardware
718events cannot be mixed together. The latter must be used with the -e
719option. The -e option and this one can be mixed and matched. Events
720can be grouped using the {} notation.
721endif::HAVE_LIBPFM[]
722
723--control=fifo:ctl-fifo[,ack-fifo]::
724--control=fd:ctl-fd[,ack-fd]::
725ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
726Listen on ctl-fd descriptor for command to control measurement.
727
728Available commands:
729
730 - 'enable' : enable events
731 - 'disable' : disable events
732 - 'enable name' : enable event 'name'
733 - 'disable name' : disable event 'name'
734 - 'snapshot' : AUX area tracing snapshot).
735 - 'stop' : stop perf record
736 - 'ping' : ping
737 - 'evlist [-v|-g|-F] : display all events
738
739 -F Show just the sample frequency used for each event.
740 -v Show all fields.
741 -g Show event group information.
742
743Measurements can be started with events disabled using --delay=-1 option. Optionally
744send control command completion ('ack\n') to ack-fd descriptor to synchronize with the
745controlling process. Example of bash shell script to enable and disable events during
746measurements:
747
748 #!/bin/bash
749
750 ctl_dir=/tmp/
751
752 ctl_fifo=${ctl_dir}perf_ctl.fifo
753 test -p ${ctl_fifo} && unlink ${ctl_fifo}
754 mkfifo ${ctl_fifo}
755 exec {ctl_fd}<>${ctl_fifo}
756
757 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
758 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
759 mkfifo ${ctl_ack_fifo}
760 exec {ctl_fd_ack}<>${ctl_ack_fifo}
761
762 perf record -D -1 -e cpu-cycles -a \
763 --control fd:${ctl_fd},${ctl_fd_ack} \
764 -- sleep 30 &
765 perf_pid=$!
766
767 sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
768 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
769
770 exec {ctl_fd_ack}>&-
771 unlink ${ctl_ack_fifo}
772
773 exec {ctl_fd}>&-
774 unlink ${ctl_fifo}
775
776 wait -n ${perf_pid}
777 exit $?
778
779--threads=<spec>::
780Write collected trace data into several data files using parallel threads.
781<spec> value can be user defined list of masks. Masks separated by colon
782define CPUs to be monitored by a thread and affinity mask of that thread
783is separated by slash:
784
785 <cpus mask 1>/<affinity mask 1>:<cpus mask 2>/<affinity mask 2>:...
786
787CPUs or affinity masks must not overlap with other corresponding masks.
788Invalid CPUs are ignored, but masks containing only invalid CPUs are not
789allowed.
790
791For example user specification like the following:
792
793 0,2-4/2-4:1,5-7/5-7
794
795specifies parallel threads layout that consists of two threads,
796the first thread monitors CPUs 0 and 2-4 with the affinity mask 2-4,
797the second monitors CPUs 1 and 5-7 with the affinity mask 5-7.
798
799<spec> value can also be a string meaning predefined parallel threads
800layout:
801
802 - cpu - create new data streaming thread for every monitored cpu
803 - core - create new thread to monitor CPUs grouped by a core
804 - package - create new thread to monitor CPUs grouped by a package
805 - numa - create new threed to monitor CPUs grouped by a NUMA domain
806
807Predefined layouts can be used on systems with large number of CPUs in
808order not to spawn multiple per-cpu streaming threads but still avoid LOST
809events in data directory files. Option specified with no or empty value
810defaults to CPU layout. Masks defined or provided by the option value are
811filtered through the mask provided by -C option.
812
813--debuginfod[=URLs]::
814 Specify debuginfod URL to be used when cacheing perf.data binaries,
815 it follows the same syntax as the DEBUGINFOD_URLS variable, like:
816
817 http://192.168.122.174:8002
818
819 If the URLs is not specified, the value of DEBUGINFOD_URLS
820 system environment variable is used.
821
822--off-cpu::
823 Enable off-cpu profiling with BPF. The BPF program will collect
824 task scheduling information with (user) stacktrace and save them
825 as sample data of a software event named "offcpu-time". The
826 sample period will have the time the task slept in nanoseconds.
827
828 Note that BPF can collect stack traces using frame pointer ("fp")
829 only, as of now. So the applications built without the frame
830 pointer might see bogus addresses.
831
832--setup-filter=<action>::
833 Prepare BPF filter to be used by regular users. The action should be
834 either "pin" or "unpin". The filter can be used after it's pinned.
835
836
837include::intel-hybrid.txt[]
838
839SEE ALSO
840--------
841linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]