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
65 See the linkperf:perf-list[1] man page for more parameters.
67 Note: If user explicitly sets options which conflict with the params,
68 the value set by the parameters will be overridden.
70 Also not defined in .../<pmu>/format/* are PMU driver specific
71 configuration parameters. Any configuration parameter preceded by
72 the letter '@' is not interpreted in user space and sent down directly
73 to the PMU driver. For example:
75 perf record -e some_event/@cfg1,@cfg2=config/ ...
77 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
78 with the event for further processing. There is no restriction on
79 what the configuration parameters are, as long as their semantic is
80 understood and supported by the PMU driver.
82 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
83 where addr is the address in memory you want to break in.
84 Access is the memory access type (read, write, execute) it can
85 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
86 number of bytes from specified addr, which the breakpoint will cover.
87 If you want to profile read-write accesses in 0x1000, just set
89 If you want to profile write accesses in [0x1000~1008), just set
92 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
93 Each event is separated by commas and the group should be quoted to
94 prevent the shell interpretation. You also need to use --group on
95 "perf report" to view group events together.
98 Event filter. This option should follow a event selector (-e) which
99 selects either tracepoint event(s) or a hardware trace PMU
100 (e.g. Intel PT or CoreSight).
104 In the case of tracepoints, multiple '--filter' options are combined
109 A hardware trace PMU advertises its ability to accept a number of
110 address filters by specifying a non-zero value in
111 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
113 Address filters have the format:
115 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
118 - 'filter': defines a region that will be traced.
119 - 'start': defines an address at which tracing will begin.
120 - 'stop': defines an address at which tracing will stop.
121 - 'tracestop': defines a region in which tracing will stop.
123 <file name> is the name of the object file, <start> is the offset to the
124 code to trace in that file, and <size> is the size of the region to
125 trace. 'start' and 'stop' filters need not specify a <size>.
127 If no object file is specified then the kernel is assumed, in which case
128 the start address must be a current kernel memory address.
130 <start> can also be specified by providing the name of a symbol. If the
131 symbol name is not unique, it can be disambiguated by inserting #n where
132 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
133 select only a global symbol. <size> can also be specified by providing
134 the name of a symbol, in which case the size is calculated to the end
135 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
136 omitted and <start> is a symbol, then the size is calculated to the end
139 If <size> is omitted and <start> is '*', then the start and size will
140 be calculated from the first and last symbols, i.e. to trace the whole
143 If symbol names (or '*') are provided, they must be surrounded by white
146 The filter passed to the kernel is not necessarily the same as entered.
147 To see the filter that is passed, use the -v option.
149 The kernel may not be able to configure a trace region if it is not
150 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
151 examined to determine if that is a possibility.
153 Multiple filters can be separated with space or comma.
156 Don't record events issued by perf itself. This option should follow
157 a event selector (-e) which selects tracepoint event(s). It adds a
158 filter expression 'common_pid != $PERFPID' to filters. If other
159 '--filter' exists, the new filter expression will be combined with
164 System-wide collection from all CPUs (default if no target is specified).
168 Record events on existing process ID (comma separated list).
172 Record events on existing thread ID (comma separated list).
173 This option also disables inheritance by default. Enable it by adding
178 Record events in threads owned by uid. Name or number.
182 Collect data with this RT SCHED_FIFO priority.
185 Collect data without buffering.
189 Event period to sample.
197 Child tasks do not inherit counters.
200 Profile at this frequency.
204 Number of mmap data pages (must be a power of two) or size
205 specification with appended unit character - B/K/M/G. The
206 size is rounded up to have nearest pages power of two value.
207 Also, by adding a comma, the number of mmap pages for AUX
208 area tracing can be specified.
211 Put all events in a single event group. This precedes the --event
212 option and remains only for backward compatibility. See --event.
215 Enables call-graph (stack chain/backtrace) recording.
218 Setup and enable call-graph (stack chain/backtrace) recording,
219 implies -g. Default is "fp".
221 Allows specifying "fp" (frame pointer) or "dwarf"
222 (DWARF's CFI - Call Frame Information) or "lbr"
223 (Hardware Last Branch Record facility) as the method to collect
224 the information used to show the call graphs.
226 In some systems, where binaries are build with gcc
227 --fomit-frame-pointer, using the "fp" method will produce bogus
228 call graphs, using "dwarf", if available (perf tools linked to
229 the libunwind or libdw library) should be used instead.
230 Using the "lbr" method doesn't require any compiler options. It
231 will produce call graphs from the hardware LBR registers. The
232 main limitation is that it is only available on new Intel
233 platforms, such as Haswell. It can only get user call chain. It
234 doesn't work with branch stack sampling at the same time.
236 When "dwarf" recording is used, perf also records (user) stack dump
237 when sampled. Default size of the stack dump is 8192 (bytes).
238 User can change the size by passing the size after comma like
239 "--call-graph dwarf,4096".
243 Don't print any message, useful for scripting.
247 Be more verbose (show counter open errors, etc).
251 Record per-thread event counts. Use it with 'perf report -T' to see
256 Record the sample virtual addresses.
259 Record the sample physical addresses.
263 Record the sample timestamps. Use it with 'perf report -D' to see the
264 timestamps, for instance.
268 Record the sample period.
271 Record the sample cpu.
279 Collect raw sample records from all opened counters (default for tracepoint counters).
283 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
284 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
285 In per-thread mode with inheritance mode on (default), samples are captured only when
286 the thread executes on the designated CPUs. Default is to monitor all CPUs.
290 Do not save the build ids of binaries in the perf.data files. This skips
291 post processing after recording, which sometimes makes the final step in
292 the recording process to take a long time, as it needs to process all
293 events looking for mmap records. The downside is that it can misresolve
294 symbols if the workload binaries used when recording get locally rebuilt
295 or upgraded, because the only key available in this case is the
296 pathname. You can also set the "record.build-id" config variable to
297 'skip to have this behaviour permanently.
301 Do not update the buildid cache. This saves some overhead in situations
302 where the information in the perf.data file (which includes buildids)
303 is sufficient. You can also set the "record.build-id" config variable to
304 'no-cache' to have the same effect.
308 monitor only in the container (cgroup) called "name". This option is available only
309 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
310 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
311 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
312 to first event, second cgroup to second event and so on. It is possible to provide
313 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
314 corresponding events, i.e., they always refer to events defined earlier on the command
319 Enable taken branch stack sampling. Any type of taken branch may be sampled.
320 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
324 Enable taken branch stack sampling. Each sample captures a series of consecutive
325 taken branches. The number of branches captured with each sample depends on the
326 underlying hardware, the type of branches of interest, and the executed code.
327 It is possible to select the types of branches captured by enabling filters. The
328 following filters are defined:
330 - any: any type of branches
331 - any_call: any function call or system call
332 - any_ret: any function return or system call return
333 - ind_call: any indirect branch
334 - call: direct calls, including far (to/from kernel) calls
335 - u: only when the branch target is at the user level
336 - k: only when the branch target is in the kernel
337 - hv: only when the target is at the hypervisor level
338 - in_tx: only when the target is in a hardware transaction
339 - no_tx: only when the target is not in a hardware transaction
340 - abort_tx: only when the target is a hardware transaction abort
341 - cond: conditional branches
342 - save_type: save branch type during sampling in case binary is not available later
345 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
346 The privilege levels may be omitted, in which case, the privilege levels of the associated
347 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
348 levels are subject to permissions. When sampling on multiple events, branch stack sampling
349 is enabled for all the sampling events. The sampled branch type is the same for all events.
350 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
351 Note that this feature may not be available on all processors.
354 Enable weightened sampling. An additional weight is recorded per sample and can be
355 displayed with the weight and local_weight sort keys. This currently works for TSX
356 abort events and some memory events in precise mode on modern Intel CPUs.
359 Record events of type PERF_RECORD_NAMESPACES.
362 Record transaction flags for transaction related events.
365 Use per-thread mmaps. By default per-cpu mmaps are created. This option
366 overrides that and uses per-thread mmaps. A side-effect of that is that
367 inheritance is automatically disabled. --per-thread is ignored with a warning
368 if combined with -a or -C options.
372 After starting the program, wait msecs before measuring. This is useful to
373 filter out the startup phase of the program, which is often very different.
377 Capture machine state (registers) at interrupt, i.e., on counter overflows for
378 each sample. List of captured registers depends on the architecture. This option
379 is off by default. It is possible to select the registers to sample using their
380 symbolic names, e.g. on x86, ax, si. To list the available registers use
381 --intr-regs=\?. To name registers, pass a comma separated list such as
382 --intr-regs=ax,bx. The list of register is architecture dependent.
386 Record running and enabled time for read events (:S)
390 Sets the clock id to use for the various time fields in the perf_event_type
391 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
392 CLOCK_MONOTONIC_RAW are supported, some events might also allow
393 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
397 Select AUX area tracing Snapshot Mode. This option is valid only with an
398 AUX area tracing event. Optionally the number of bytes to capture per
399 snapshot can be specified. In Snapshot Mode, trace data is captured only when
400 signal SIGUSR2 is received.
403 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
404 because the file may be huge. A time out is needed in such cases.
405 This option sets the time out limit. The default value is 500 ms.
408 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
409 PERF_RECORD_SWITCH_CPU_WIDE.
412 Path to clang binary to use for compiling BPF scriptlets.
413 (enabled when BPF support is on)
415 --clang-opt=OPTIONS::
416 Options passed to clang when compiling BPF scriptlets.
417 (enabled when BPF support is on)
420 Specify vmlinux path which has debuginfo.
421 (enabled when BPF prologue is on)
424 Record build-id of all DSOs regardless whether it's actually hit or not.
427 Configure all used events to run in kernel space.
430 Configure all used events to run in user space.
433 Append timestamp to output file name.
435 --switch-output[=mode]::
436 Generate multiple perf.data files, timestamp prefixed, switching to a new one
437 based on 'mode' value:
438 "signal" - when receiving a SIGUSR2 (default value) or
439 <size> - when reaching the size threshold, size is expected to
440 be a number with appended unit character - B/K/M/G
441 <time> - when reaching the time threshold, size is expected to
442 be a number with appended unit character - s/m/h/d
444 Note: the precision of the size threshold hugely depends
445 on your configuration - the number and size of your ring
446 buffers (-m). It is generally more precise for higher sizes
447 (like >5M), for lower values expect different sizes.
449 A possible use case is to, given an external event, slice the perf.data file
450 that gets then processed, possibly via a perf script, to decide if that
451 particular perf.data snapshot should be kept or not.
453 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
454 The reason for the latter two is to reduce the data file switching
455 overhead. You can still switch them on with:
457 --switch-output --no-no-buildid --no-no-buildid-cache
460 Parse options then exit. --dry-run can be used to detect errors in cmdline
463 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
464 in config file is set to true.
467 Instead of collecting non-sample events (for example, fork, comm, mmap) at
468 the beginning of record, collect them during finalizing an output file.
469 The collected non-sample events reflects the status of the system when
473 Makes all events use an overwritable ring buffer. An overwritable ring
474 buffer works like a flight recorder: when it gets full, the kernel will
475 overwrite the oldest records, that thus will never make it to the
478 When '--overwrite' and '--switch-output' are used perf records and drops
479 events until it receives a signal, meaning that something unusual was
480 detected that warrants taking a snapshot of the most current events,
481 those fitting in the ring buffer at that moment.
483 'overwrite' attribute can also be set or canceled for an event using
484 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
486 Implies --tail-synthesize.
490 linkperf:perf-stat[1], linkperf:perf-list[1]