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 symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
37 'param1', 'param2', etc are defined as formats for the PMU in
38 /sys/bus/event_source/devices/<pmu>/format/*.
40 - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
42 where M, N, K are numbers (in decimal, hex, octal format). Acceptable
43 values for each of 'config', 'config1' and 'config2' are defined by
44 corresponding entries in /sys/bus/event_source/devices/<pmu>/format/*
45 param1 and param2 are defined as formats for the PMU in:
46 /sys/bus/event_source/devices/<pmu>/format/*
48 There are also some parameters which are not defined in .../<pmu>/format/*.
49 These params can be used to overload default config values per event.
50 Here are some common parameters:
51 - 'period': Set event sampling period
52 - 'freq': Set event sampling frequency
53 - 'time': Disable/enable time stamping. Acceptable values are 1 for
54 enabling time stamping. 0 for disabling time stamping.
56 - 'call-graph': Disable/enable callgraph. Acceptable str are "fp" for
57 FP mode, "dwarf" for DWARF mode, "lbr" for LBR mode and
58 "no" for disable callgraph.
59 - 'stack-size': user stack size for dwarf mode
60 - 'name' : User defined event name. Single quotes (') may be used to
61 escape symbols in the name from parsing by shell and tool
62 like this: name=\'CPU_CLK_UNHALTED.THREAD:cmask=0x1\'.
63 - 'aux-output': Generate AUX records instead of events. This requires
64 that an AUX area event is also provided.
65 - 'aux-sample-size': Set sample size for AUX area sampling. If the
66 '--aux-sample' option has been used, set aux-sample-size=0 to disable
67 AUX area sampling for the event.
69 See the linkperf:perf-list[1] man page for more parameters.
71 Note: If user explicitly sets options which conflict with the params,
72 the value set by the parameters will be overridden.
74 Also not defined in .../<pmu>/format/* are PMU driver specific
75 configuration parameters. Any configuration parameter preceded by
76 the letter '@' is not interpreted in user space and sent down directly
77 to the PMU driver. For example:
79 perf record -e some_event/@cfg1,@cfg2=config/ ...
81 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
82 with the event for further processing. There is no restriction on
83 what the configuration parameters are, as long as their semantic is
84 understood and supported by the PMU driver.
86 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
87 where addr is the address in memory you want to break in.
88 Access is the memory access type (read, write, execute) it can
89 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
90 number of bytes from specified addr, which the breakpoint will cover.
91 If you want to profile read-write accesses in 0x1000, just set
93 If you want to profile write accesses in [0x1000~1008), just set
96 - a BPF source file (ending in .c) or a precompiled object file (ending
97 in .o) selects one or more BPF events.
98 The BPF program can attach to various perf events based on the ELF section
101 When processing a '.c' file, perf searches an installed LLVM to compile it
102 into an object file first. Optional clang options can be passed via the
103 '--clang-opt' command line option, e.g.:
105 perf record --clang-opt "-DLINUX_VERSION_CODE=0x50000" \
106 -e tests/bpf-script-example.c
108 Note: '--clang-opt' must be placed before '--event/-e'.
110 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
111 Each event is separated by commas and the group should be quoted to
112 prevent the shell interpretation. You also need to use --group on
113 "perf report" to view group events together.
116 Event filter. This option should follow an event selector (-e) which
117 selects either tracepoint event(s) or a hardware trace PMU
118 (e.g. Intel PT or CoreSight).
122 In the case of tracepoints, multiple '--filter' options are combined
127 A hardware trace PMU advertises its ability to accept a number of
128 address filters by specifying a non-zero value in
129 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
131 Address filters have the format:
133 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
136 - 'filter': defines a region that will be traced.
137 - 'start': defines an address at which tracing will begin.
138 - 'stop': defines an address at which tracing will stop.
139 - 'tracestop': defines a region in which tracing will stop.
141 <file name> is the name of the object file, <start> is the offset to the
142 code to trace in that file, and <size> is the size of the region to
143 trace. 'start' and 'stop' filters need not specify a <size>.
145 If no object file is specified then the kernel is assumed, in which case
146 the start address must be a current kernel memory address.
148 <start> can also be specified by providing the name of a symbol. If the
149 symbol name is not unique, it can be disambiguated by inserting #n where
150 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
151 select only a global symbol. <size> can also be specified by providing
152 the name of a symbol, in which case the size is calculated to the end
153 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
154 omitted and <start> is a symbol, then the size is calculated to the end
157 If <size> is omitted and <start> is '*', then the start and size will
158 be calculated from the first and last symbols, i.e. to trace the whole
161 If symbol names (or '*') are provided, they must be surrounded by white
164 The filter passed to the kernel is not necessarily the same as entered.
165 To see the filter that is passed, use the -v option.
167 The kernel may not be able to configure a trace region if it is not
168 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
169 examined to determine if that is a possibility.
171 Multiple filters can be separated with space or comma.
174 Don't record events issued by perf itself. This option should follow
175 an event selector (-e) which selects tracepoint event(s). It adds a
176 filter expression 'common_pid != $PERFPID' to filters. If other
177 '--filter' exists, the new filter expression will be combined with
182 System-wide collection from all CPUs (default if no target is specified).
186 Record events on existing process ID (comma separated list).
190 Record events on existing thread ID (comma separated list).
191 This option also disables inheritance by default. Enable it by adding
196 Record events in threads owned by uid. Name or number.
200 Collect data with this RT SCHED_FIFO priority.
203 Collect data without buffering.
207 Event period to sample.
215 Child tasks do not inherit counters.
219 Profile at this frequency. Use 'max' to use the currently maximum
220 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
221 sysctl. Will throttle down to the currently maximum allowed frequency.
225 Fail if the specified frequency can't be used.
229 Number of mmap data pages (must be a power of two) or size
230 specification with appended unit character - B/K/M/G. The
231 size is rounded up to have nearest pages power of two value.
232 Also, by adding a comma, the number of mmap pages for AUX
233 area tracing can be specified.
236 Put all events in a single event group. This precedes the --event
237 option and remains only for backward compatibility. See --event.
240 Enables call-graph (stack chain/backtrace) recording.
243 Setup and enable call-graph (stack chain/backtrace) recording,
244 implies -g. Default is "fp".
246 Allows specifying "fp" (frame pointer) or "dwarf"
247 (DWARF's CFI - Call Frame Information) or "lbr"
248 (Hardware Last Branch Record facility) as the method to collect
249 the information used to show the call graphs.
251 In some systems, where binaries are build with gcc
252 --fomit-frame-pointer, using the "fp" method will produce bogus
253 call graphs, using "dwarf", if available (perf tools linked to
254 the libunwind or libdw library) should be used instead.
255 Using the "lbr" method doesn't require any compiler options. It
256 will produce call graphs from the hardware LBR registers. The
257 main limitation is that it is only available on new Intel
258 platforms, such as Haswell. It can only get user call chain. It
259 doesn't work with branch stack sampling at the same time.
261 When "dwarf" recording is used, perf also records (user) stack dump
262 when sampled. Default size of the stack dump is 8192 (bytes).
263 User can change the size by passing the size after comma like
264 "--call-graph dwarf,4096".
268 Don't print any message, useful for scripting.
272 Be more verbose (show counter open errors, etc).
276 Record per-thread event counts. Use it with 'perf report -T' to see
281 Record the sample virtual addresses.
284 Record the sample physical addresses.
288 Record the sample timestamps. Use it with 'perf report -D' to see the
289 timestamps, for instance.
293 Record the sample period.
296 Record the sample cpu.
304 Collect raw sample records from all opened counters (default for tracepoint counters).
308 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
309 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
310 In per-thread mode with inheritance mode on (default), samples are captured only when
311 the thread executes on the designated CPUs. Default is to monitor all CPUs.
315 Do not save the build ids of binaries in the perf.data files. This skips
316 post processing after recording, which sometimes makes the final step in
317 the recording process to take a long time, as it needs to process all
318 events looking for mmap records. The downside is that it can misresolve
319 symbols if the workload binaries used when recording get locally rebuilt
320 or upgraded, because the only key available in this case is the
321 pathname. You can also set the "record.build-id" config variable to
322 'skip to have this behaviour permanently.
326 Do not update the buildid cache. This saves some overhead in situations
327 where the information in the perf.data file (which includes buildids)
328 is sufficient. You can also set the "record.build-id" config variable to
329 'no-cache' to have the same effect.
333 monitor only in the container (cgroup) called "name". This option is available only
334 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
335 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
336 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
337 to first event, second cgroup to second event and so on. It is possible to provide
338 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
339 corresponding events, i.e., they always refer to events defined earlier on the command
340 line. If the user wants to track multiple events for a specific cgroup, the user can
341 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
343 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
344 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
348 Enable taken branch stack sampling. Any type of taken branch may be sampled.
349 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
353 Enable taken branch stack sampling. Each sample captures a series of consecutive
354 taken branches. The number of branches captured with each sample depends on the
355 underlying hardware, the type of branches of interest, and the executed code.
356 It is possible to select the types of branches captured by enabling filters. The
357 following filters are defined:
359 - any: any type of branches
360 - any_call: any function call or system call
361 - any_ret: any function return or system call return
362 - ind_call: any indirect branch
363 - call: direct calls, including far (to/from kernel) calls
364 - u: only when the branch target is at the user level
365 - k: only when the branch target is in the kernel
366 - hv: only when the target is at the hypervisor level
367 - in_tx: only when the target is in a hardware transaction
368 - no_tx: only when the target is not in a hardware transaction
369 - abort_tx: only when the target is a hardware transaction abort
370 - cond: conditional branches
371 - save_type: save branch type during sampling in case binary is not available later
374 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
375 The privilege levels may be omitted, in which case, the privilege levels of the associated
376 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
377 levels are subject to permissions. When sampling on multiple events, branch stack sampling
378 is enabled for all the sampling events. The sampled branch type is the same for all events.
379 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
380 Note that this feature may not be available on all processors.
383 Enable weightened sampling. An additional weight is recorded per sample and can be
384 displayed with the weight and local_weight sort keys. This currently works for TSX
385 abort events and some memory events in precise mode on modern Intel CPUs.
388 Record events of type PERF_RECORD_NAMESPACES.
391 Record transaction flags for transaction related events.
394 Use per-thread mmaps. By default per-cpu mmaps are created. This option
395 overrides that and uses per-thread mmaps. A side-effect of that is that
396 inheritance is automatically disabled. --per-thread is ignored with a warning
397 if combined with -a or -C options.
401 After starting the program, wait msecs before measuring. This is useful to
402 filter out the startup phase of the program, which is often very different.
406 Capture machine state (registers) at interrupt, i.e., on counter overflows for
407 each sample. List of captured registers depends on the architecture. This option
408 is off by default. It is possible to select the registers to sample using their
409 symbolic names, e.g. on x86, ax, si. To list the available registers use
410 --intr-regs=\?. To name registers, pass a comma separated list such as
411 --intr-regs=ax,bx. The list of register is architecture dependent.
414 Similar to -I, but capture user registers at sample time. To list the available
415 user registers use --user-regs=\?.
418 Record running and enabled time for read events (:S)
422 Sets the clock id to use for the various time fields in the perf_event_type
423 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
424 CLOCK_MONOTONIC_RAW are supported, some events might also allow
425 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
429 Select AUX area tracing Snapshot Mode. This option is valid only with an
430 AUX area tracing event. Optionally, certain snapshot capturing parameters
431 can be specified in a string that follows this option:
432 'e': take one last snapshot on exit; guarantees that there is at least one
433 snapshot in the output file;
434 <size>: if the PMU supports this, specify the desired snapshot size.
436 In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
437 and on exit if the above 'e' option is given.
439 --aux-sample[=OPTIONS]::
440 Select AUX area sampling. At least one of the events selected by the -e option
441 must be an AUX area event. Samples on other events will be created containing
442 data from the AUX area. Optionally sample size may be specified, otherwise it
446 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
447 because the file may be huge. A time out is needed in such cases.
448 This option sets the time out limit. The default value is 500 ms.
451 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
452 PERF_RECORD_SWITCH_CPU_WIDE.
455 Path to clang binary to use for compiling BPF scriptlets.
456 (enabled when BPF support is on)
458 --clang-opt=OPTIONS::
459 Options passed to clang when compiling BPF scriptlets.
460 (enabled when BPF support is on)
463 Specify vmlinux path which has debuginfo.
464 (enabled when BPF prologue is on)
467 Record build-id of all DSOs regardless whether it's actually hit or not.
470 Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
471 Asynchronous mode is supported only when linking Perf tool with libc library
472 providing implementation for Posix AIO API.
475 Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
476 node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
477 cpu - thread affinity mask is set to cpu of the processed mmap buffer
479 --mmap-flush=number::
481 Specify minimal number of bytes that is extracted from mmap data pages and
482 processed for output. One can specify the number using B/K/M/G suffixes.
484 The maximal allowed value is a quarter of the size of mmaped data pages.
486 The default option value is 1 byte which means that every time that the output
487 writing thread finds some new data in the mmaped buffer the data is extracted,
488 possibly compressed (-z) and written to the output, perf.data or pipe.
490 Larger data chunks are compressed more effectively in comparison to smaller
491 chunks so extraction of larger chunks from the mmap data pages is preferable
492 from the perspective of output size reduction.
494 Also at some cases executing less output write syscalls with bigger data size
495 can take less time than executing more output write syscalls with smaller data
496 size thus lowering runtime profiling overhead.
499 --compression-level[=n]::
500 Produce compressed trace using specified level n (default: 1 - fastest compression,
504 Configure all used events to run in kernel space.
507 Configure all used events to run in user space.
509 --kernel-callchains::
510 Collect callchains only from kernel space. I.e. this option sets
511 perf_event_attr.exclude_callchain_user to 1.
514 Collect callchains only from user space. I.e. this option sets
515 perf_event_attr.exclude_callchain_kernel to 1.
517 Don't use both --kernel-callchains and --user-callchains at the same time or no
518 callchains will be collected.
521 Append timestamp to output file name.
523 --timestamp-boundary::
524 Record timestamp boundary (time of first/last samples).
526 --switch-output[=mode]::
527 Generate multiple perf.data files, timestamp prefixed, switching to a new one
528 based on 'mode' value:
529 "signal" - when receiving a SIGUSR2 (default value) or
530 <size> - when reaching the size threshold, size is expected to
531 be a number with appended unit character - B/K/M/G
532 <time> - when reaching the time threshold, size is expected to
533 be a number with appended unit character - s/m/h/d
535 Note: the precision of the size threshold hugely depends
536 on your configuration - the number and size of your ring
537 buffers (-m). It is generally more precise for higher sizes
538 (like >5M), for lower values expect different sizes.
540 A possible use case is to, given an external event, slice the perf.data file
541 that gets then processed, possibly via a perf script, to decide if that
542 particular perf.data snapshot should be kept or not.
544 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
545 The reason for the latter two is to reduce the data file switching
546 overhead. You can still switch them on with:
548 --switch-output --no-no-buildid --no-no-buildid-cache
550 --switch-max-files=N::
552 When rotating perf.data with --switch-output, only keep N files.
555 Parse options then exit. --dry-run can be used to detect errors in cmdline
558 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
559 in config file is set to true.
562 Instead of collecting non-sample events (for example, fork, comm, mmap) at
563 the beginning of record, collect them during finalizing an output file.
564 The collected non-sample events reflects the status of the system when
568 Makes all events use an overwritable ring buffer. An overwritable ring
569 buffer works like a flight recorder: when it gets full, the kernel will
570 overwrite the oldest records, that thus will never make it to the
573 When '--overwrite' and '--switch-output' are used perf records and drops
574 events until it receives a signal, meaning that something unusual was
575 detected that warrants taking a snapshot of the most current events,
576 those fitting in the ring buffer at that moment.
578 'overwrite' attribute can also be set or canceled for an event using
579 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
581 Implies --tail-synthesize.
584 Make a copy of /proc/kcore and place it into a directory with the perf data file.
587 Limit the sample data max size, <size> is expected to be a number with
588 appended unit character - B/K/M/G
592 linkperf:perf-stat[1], linkperf:perf-list[1]
projects / linux-block.git / blob