6 perf-record - Run a command and record its profile into perf.data
11 'perf record' [-e <EVENT> | --event=EVENT] [-a] <command>
12 'perf record' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
16 This command runs a command and gathers a performance counter profile
17 from it, into perf.data - without displaying anything.
19 This file can then be inspected later on, using 'perf report'.
25 Any command you can specify in a shell.
29 Select the PMU event. Selection can be:
31 - a symbolic event name (use 'perf list' to list all events)
33 - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
34 hexadecimal event descriptor.
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.
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/*.
44 - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
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/*
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.
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.
73 See the linkperf:perf-list[1] man page for more parameters.
75 Note: If user explicitly sets options which conflict with the params,
76 the value set by the parameters will be overridden.
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:
83 perf record -e some_event/@cfg1,@cfg2=config/ ...
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.
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
97 If you want to profile write accesses in [0x1000~1008), just set
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
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.:
109 perf record --clang-opt "-DLINUX_VERSION_CODE=0x50000" \
110 -e tests/bpf-script-example.c
112 Note: '--clang-opt' must be placed before '--event/-e'.
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.
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).
126 In the case of tracepoints, multiple '--filter' options are combined
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.
135 Address filters have the format:
137 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
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.
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>.
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.
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
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
165 If symbol names (or '*') are provided, they must be surrounded by white
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.
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.
175 Multiple filters can be separated with space or comma.
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
186 System-wide collection from all CPUs (default if no target is specified).
190 Record events on existing process ID (comma separated list).
194 Record events on existing thread ID (comma separated list).
195 This option also disables inheritance by default. Enable it by adding
200 Record events in threads owned by uid. Name or number.
204 Collect data with this RT SCHED_FIFO priority.
207 Collect data without buffering.
211 Event period to sample.
219 Child tasks do not inherit counters.
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.
229 Fail if the specified frequency can't be used.
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.
240 Put all events in a single event group. This precedes the --event
241 option and remains only for backward compatibility. See --event.
244 Enables call-graph (stack chain/backtrace) recording for both
245 kernel space and user space.
248 Setup and enable call-graph (stack chain/backtrace) recording,
249 implies -g. Default is "fp" (for user space).
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)
255 Any option specified here controls the method used for user space.
257 Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
258 Call Frame Information) or "lbr" (Hardware Last Branch Record
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.
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".
278 Don't print any message, useful for scripting.
282 Be more verbose (show counter open errors, etc).
286 Record per-thread event counts. Use it with 'perf report -T' to see
291 Record the sample virtual addresses.
294 Record the sample physical addresses.
297 Record the sampled data address data page size.
300 Record the sampled code address (ip) page size
304 Record the sample timestamps. Use it with 'perf report -D' to see the
305 timestamps, for instance.
309 Record the sample period.
312 Record the sample cpu.
320 Collect raw sample records from all opened counters (default for tracepoint counters).
324 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
325 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
326 In per-thread mode with inheritance mode on (default), samples are captured only when
327 the thread executes on the designated CPUs. Default is to monitor all CPUs.
331 Do not save the build ids of binaries in the perf.data files. This skips
332 post processing after recording, which sometimes makes the final step in
333 the recording process to take a long time, as it needs to process all
334 events looking for mmap records. The downside is that it can misresolve
335 symbols if the workload binaries used when recording get locally rebuilt
336 or upgraded, because the only key available in this case is the
337 pathname. You can also set the "record.build-id" config variable to
338 'skip to have this behaviour permanently.
342 Do not update the buildid cache. This saves some overhead in situations
343 where the information in the perf.data file (which includes buildids)
344 is sufficient. You can also set the "record.build-id" config variable to
345 'no-cache' to have the same effect.
349 monitor only in the container (cgroup) called "name". This option is available only
350 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
351 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
352 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
353 to first event, second cgroup to second event and so on. It is possible to provide
354 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
355 corresponding events, i.e., they always refer to events defined earlier on the command
356 line. If the user wants to track multiple events for a specific cgroup, the user can
357 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
359 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
360 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
364 Enable taken branch stack sampling. Any type of taken branch may be sampled.
365 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
369 Enable taken branch stack sampling. Each sample captures a series of consecutive
370 taken branches. The number of branches captured with each sample depends on the
371 underlying hardware, the type of branches of interest, and the executed code.
372 It is possible to select the types of branches captured by enabling filters. The
373 following filters are defined:
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
390 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
391 The privilege levels may be omitted, in which case, the privilege levels of the associated
392 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
393 levels are subject to permissions. When sampling on multiple events, branch stack sampling
394 is enabled for all the sampling events. The sampled branch type is the same for all events.
395 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
396 Note that this feature may not be available on all processors.
399 Enable weightened sampling. An additional weight is recorded per sample and can be
400 displayed with the weight and local_weight sort keys. This currently works for TSX
401 abort events and some memory events in precise mode on modern Intel CPUs.
404 Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.
407 Record events of type PERF_RECORD_CGROUP. This enables 'cgroup' sort key.
410 Record transaction flags for transaction related events.
413 Use per-thread mmaps. By default per-cpu mmaps are created. This option
414 overrides that and uses per-thread mmaps. A side-effect of that is that
415 inheritance is automatically disabled. --per-thread is ignored with a warning
416 if combined with -a or -C options.
420 After starting the program, wait msecs before measuring (-1: start with events
421 disabled). This is useful to filter out the startup phase of the program, which
422 is often very different.
426 Capture machine state (registers) at interrupt, i.e., on counter overflows for
427 each sample. List of captured registers depends on the architecture. This option
428 is off by default. It is possible to select the registers to sample using their
429 symbolic 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.
434 Similar to -I, but capture user registers at sample time. To list the available
435 user registers use --user-regs=\?.
438 Record running and enabled time for read events (:S)
442 Sets the clock id to use for the various time fields in the perf_event_type
443 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
444 CLOCK_MONOTONIC_RAW are supported, some events might also allow
445 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
449 Select AUX area tracing Snapshot Mode. This option is valid only with an
450 AUX area tracing event. Optionally, certain snapshot capturing parameters
451 can 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.
456 In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
457 and on exit if the above 'e' option is given.
459 --aux-sample[=OPTIONS]::
460 Select AUX area sampling. At least one of the events selected by the -e option
461 must be an AUX area event. Samples on other events will be created containing
462 data from the AUX area. Optionally sample size may be specified, otherwise it
466 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
467 because the file may be huge. A time out is needed in such cases.
468 This option sets the time out limit. The default value is 500 ms.
471 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
472 PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT or CoreSight)
473 switch events will be enabled automatically, which can be suppressed by
474 by the option --no-switch-events.
477 Path to clang binary to use for compiling BPF scriptlets.
478 (enabled when BPF support is on)
480 --clang-opt=OPTIONS::
481 Options passed to clang when compiling BPF scriptlets.
482 (enabled when BPF support is on)
485 Specify vmlinux path which has debuginfo.
486 (enabled when BPF prologue is on)
489 Record build-id of all DSOs regardless whether it's actually hit or not.
492 Record build ids in mmap2 events, disables build id cache (implies --no-buildid).
495 Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
496 Asynchronous mode is supported only when linking Perf tool with libc library
497 providing implementation for Posix AIO API.
500 Set 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
504 --mmap-flush=number::
506 Specify minimal number of bytes that is extracted from mmap data pages and
507 processed for output. One can specify the number using B/K/M/G suffixes.
509 The maximal allowed value is a quarter of the size of mmaped data pages.
511 The default option value is 1 byte which means that every time that the output
512 writing thread finds some new data in the mmaped buffer the data is extracted,
513 possibly compressed (-z) and written to the output, perf.data or pipe.
515 Larger data chunks are compressed more effectively in comparison to smaller
516 chunks so extraction of larger chunks from the mmap data pages is preferable
517 from the perspective of output size reduction.
519 Also at some cases executing less output write syscalls with bigger data size
520 can take less time than executing more output write syscalls with smaller data
521 size thus lowering runtime profiling overhead.
524 --compression-level[=n]::
525 Produce compressed trace using specified level n (default: 1 - fastest compression,
529 Configure all used events to run in kernel space.
532 Configure all used events to run in user space.
534 --kernel-callchains::
535 Collect callchains only from kernel space. I.e. this option sets
536 perf_event_attr.exclude_callchain_user to 1.
539 Collect callchains only from user space. I.e. this option sets
540 perf_event_attr.exclude_callchain_kernel to 1.
542 Don't use both --kernel-callchains and --user-callchains at the same time or no
543 callchains will be collected.
546 Append timestamp to output file name.
548 --timestamp-boundary::
549 Record timestamp boundary (time of first/last samples).
551 --switch-output[=mode]::
552 Generate multiple perf.data files, timestamp prefixed, switching to a new one
553 based 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
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.
565 A possible use case is to, given an external event, slice the perf.data file
566 that gets then processed, possibly via a perf script, to decide if that
567 particular perf.data snapshot should be kept or not.
569 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
570 The reason for the latter two is to reduce the data file switching
571 overhead. You can still switch them on with:
573 --switch-output --no-no-buildid --no-no-buildid-cache
575 --switch-output-event::
576 Events 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
578 thread will also send a SIGUSR2 to the main one.
580 Uses the same syntax as --event, it will just not be recorded, serving only to
581 switch the perf.data file as soon as the --switch-output event is processed by
582 a separate sideband thread.
584 This sideband thread is also used to other purposes, like processing the
585 PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
588 --switch-max-files=N::
590 When rotating perf.data with --switch-output, only keep N files.
593 Parse options then exit. --dry-run can be used to detect errors in cmdline
596 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
597 in config file is set to true.
600 Instead of collecting non-sample events (for example, fork, comm, mmap) at
601 the beginning of record, collect them during finalizing an output file.
602 The collected non-sample events reflects the status of the system when
606 Makes all events use an overwritable ring buffer. An overwritable ring
607 buffer works like a flight recorder: when it gets full, the kernel will
608 overwrite the oldest records, that thus will never make it to the
611 When '--overwrite' and '--switch-output' are used perf records and drops
612 events until it receives a signal, meaning that something unusual was
613 detected that warrants taking a snapshot of the most current events,
614 those fitting in the ring buffer at that moment.
616 'overwrite' attribute can also be set or canceled for an event using
617 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
619 Implies --tail-synthesize.
622 Make a copy of /proc/kcore and place it into a directory with the perf data file.
625 Limit the sample data max size, <size> is expected to be a number with
626 appended unit character - B/K/M/G
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.
633 --pfm-events events::
634 Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
635 including support for event filters. For example '--pfm-events
636 inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
637 option using the comma separator. Hardware events and generic hardware
638 events cannot be mixed together. The latter must be used with the -e
639 option. The -e option and this one can be mixed and matched. Events
640 can be grouped using the {} notation.
643 --control=fifo:ctl-fifo[,ack-fifo]::
644 --control=fd:ctl-fd[,ack-fd]::
645 ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
646 Listen on ctl-fd descriptor for command to control measurement.
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).
655 Measurements can be started with events disabled using --delay=-1 option. Optionally
656 send control command completion ('ack\n') to ack-fd descriptor to synchronize with the
657 controlling process. Example of bash shell script to enable and disable events during
664 ctl_fifo=${ctl_dir}perf_ctl.fifo
665 test -p ${ctl_fifo} && unlink ${ctl_fifo}
667 exec {ctl_fd}<>${ctl_fifo}
669 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
670 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
671 mkfifo ${ctl_ack_fifo}
672 exec {ctl_fd_ack}<>${ctl_ack_fifo}
674 perf record -D -1 -e cpu-cycles -a \
675 --control fd:${ctl_fd},${ctl_fd_ack} \
679 sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
680 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
683 unlink ${ctl_ack_fifo}
694 linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]