6 perf-record - Run a command and record its profile into perf.data
11 'perf record' [-e <EVENT> | --event=EVENT] [-l] [-a] <command>
12 'perf record' [-e <EVENT> | --event=EVENT] [-l] [-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 params which are not defined in .../<pmu>/format/*.
53 These params can be used to overload default config values per event.
54 Here is a list of the params.
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 Note: If user explicitly sets options which conflict with the params,
65 the value set by the params will be overridden.
67 Also not defined in .../<pmu>/format/* are PMU driver specific
68 configuration parameters. Any configuration parameter preceded by
69 the letter '@' is not interpreted in user space and sent down directly
70 to the PMU driver. For example:
72 perf record -e some_event/@cfg1,@cfg2=config/ ...
74 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
75 with the event for further processing. There is no restriction on
76 what the configuration parameters are, as long as their semantic is
77 understood and supported by the PMU driver.
79 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
80 where addr is the address in memory you want to break in.
81 Access is the memory access type (read, write, execute) it can
82 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
83 number of bytes from specified addr, which the breakpoint will cover.
84 If you want to profile read-write accesses in 0x1000, just set
86 If you want to profile write accesses in [0x1000~1008), just set
89 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
90 Each event is separated by commas and the group should be quoted to
91 prevent the shell interpretation. You also need to use --group on
92 "perf report" to view group events together.
95 Event filter. This option should follow a event selector (-e) which
96 selects either tracepoint event(s) or a hardware trace PMU
97 (e.g. Intel PT or CoreSight).
101 In the case of tracepoints, multiple '--filter' options are combined
106 A hardware trace PMU advertises its ability to accept a number of
107 address filters by specifying a non-zero value in
108 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
110 Address filters have the format:
112 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
115 - 'filter': defines a region that will be traced.
116 - 'start': defines an address at which tracing will begin.
117 - 'stop': defines an address at which tracing will stop.
118 - 'tracestop': defines a region in which tracing will stop.
120 <file name> is the name of the object file, <start> is the offset to the
121 code to trace in that file, and <size> is the size of the region to
122 trace. 'start' and 'stop' filters need not specify a <size>.
124 If no object file is specified then the kernel is assumed, in which case
125 the start address must be a current kernel memory address.
127 <start> can also be specified by providing the name of a symbol. If the
128 symbol name is not unique, it can be disambiguated by inserting #n where
129 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
130 select only a global symbol. <size> can also be specified by providing
131 the name of a symbol, in which case the size is calculated to the end
132 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
133 omitted and <start> is a symbol, then the size is calculated to the end
136 If <size> is omitted and <start> is '*', then the start and size will
137 be calculated from the first and last symbols, i.e. to trace the whole
140 If symbol names (or '*') are provided, they must be surrounded by white
143 The filter passed to the kernel is not necessarily the same as entered.
144 To see the filter that is passed, use the -v option.
146 The kernel may not be able to configure a trace region if it is not
147 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
148 examined to determine if that is a possibility.
150 Multiple filters can be separated with space or comma.
153 Don't record events issued by perf itself. This option should follow
154 a event selector (-e) which selects tracepoint event(s). It adds a
155 filter expression 'common_pid != $PERFPID' to filters. If other
156 '--filter' exists, the new filter expression will be combined with
161 System-wide collection from all CPUs.
165 Record events on existing process ID (comma separated list).
169 Record events on existing thread ID (comma separated list).
170 This option also disables inheritance by default. Enable it by adding
175 Record events in threads owned by uid. Name or number.
179 Collect data with this RT SCHED_FIFO priority.
182 Collect data without buffering.
186 Event period to sample.
194 Child tasks do not inherit counters.
197 Profile at this frequency.
201 Number of mmap data pages (must be a power of two) or size
202 specification with appended unit character - B/K/M/G. The
203 size is rounded up to have nearest pages power of two value.
204 Also, by adding a comma, the number of mmap pages for AUX
205 area tracing can be specified.
208 Put all events in a single event group. This precedes the --event
209 option and remains only for backward compatibility. See --event.
212 Enables call-graph (stack chain/backtrace) recording.
215 Setup and enable call-graph (stack chain/backtrace) recording,
216 implies -g. Default is "fp".
218 Allows specifying "fp" (frame pointer) or "dwarf"
219 (DWARF's CFI - Call Frame Information) or "lbr"
220 (Hardware Last Branch Record facility) as the method to collect
221 the information used to show the call graphs.
223 In some systems, where binaries are build with gcc
224 --fomit-frame-pointer, using the "fp" method will produce bogus
225 call graphs, using "dwarf", if available (perf tools linked to
226 the libunwind or libdw library) should be used instead.
227 Using the "lbr" method doesn't require any compiler options. It
228 will produce call graphs from the hardware LBR registers. The
229 main limition is that it is only available on new Intel
230 platforms, such as Haswell. It can only get user call chain. It
231 doesn't work with branch stack sampling at the same time.
233 When "dwarf" recording is used, perf also records (user) stack dump
234 when sampled. Default size of the stack dump is 8192 (bytes).
235 User can change the size by passing the size after comma like
236 "--call-graph dwarf,4096".
240 Don't print any message, useful for scripting.
244 Be more verbose (show counter open errors, etc).
248 Record per-thread event counts. Use it with 'perf report -T' to see
253 Record the sample addresses.
257 Record the sample timestamps. Use it with 'perf report -D' to see the
258 timestamps, for instance.
262 Record the sample period.
265 Record the sample cpu.
273 Collect raw sample records from all opened counters (default for tracepoint counters).
277 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
278 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
279 In per-thread mode with inheritance mode on (default), samples are captured only when
280 the thread executes on the designated CPUs. Default is to monitor all CPUs.
284 Do not save the build ids of binaries in the perf.data files. This skips
285 post processing after recording, which sometimes makes the final step in
286 the recording process to take a long time, as it needs to process all
287 events looking for mmap records. The downside is that it can misresolve
288 symbols if the workload binaries used when recording get locally rebuilt
289 or upgraded, because the only key available in this case is the
290 pathname. You can also set the "record.build-id" config variable to
291 'skip to have this behaviour permanently.
295 Do not update the buildid cache. This saves some overhead in situations
296 where the information in the perf.data file (which includes buildids)
297 is sufficient. You can also set the "record.build-id" config variable to
298 'no-cache' to have the same effect.
302 monitor only in the container (cgroup) called "name". This option is available only
303 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
304 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
305 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
306 to first event, second cgroup to second event and so on. It is possible to provide
307 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
308 corresponding events, i.e., they always refer to events defined earlier on the command
313 Enable taken branch stack sampling. Any type of taken branch may be sampled.
314 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
318 Enable taken branch stack sampling. Each sample captures a series of consecutive
319 taken branches. The number of branches captured with each sample depends on the
320 underlying hardware, the type of branches of interest, and the executed code.
321 It is possible to select the types of branches captured by enabling filters. The
322 following filters are defined:
324 - any: any type of branches
325 - any_call: any function call or system call
326 - any_ret: any function return or system call return
327 - ind_call: any indirect branch
328 - call: direct calls, including far (to/from kernel) calls
329 - u: only when the branch target is at the user level
330 - k: only when the branch target is in the kernel
331 - hv: only when the target is at the hypervisor level
332 - in_tx: only when the target is in a hardware transaction
333 - no_tx: only when the target is not in a hardware transaction
334 - abort_tx: only when the target is a hardware transaction abort
335 - cond: conditional branches
338 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
339 The privilege levels may be omitted, in which case, the privilege levels of the associated
340 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
341 levels are subject to permissions. When sampling on multiple events, branch stack sampling
342 is enabled for all the sampling events. The sampled branch type is the same for all events.
343 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
344 Note that this feature may not be available on all processors.
347 Enable weightened sampling. An additional weight is recorded per sample and can be
348 displayed with the weight and local_weight sort keys. This currently works for TSX
349 abort events and some memory events in precise mode on modern Intel CPUs.
352 Record transaction flags for transaction related events.
355 Use per-thread mmaps. By default per-cpu mmaps are created. This option
356 overrides that and uses per-thread mmaps. A side-effect of that is that
357 inheritance is automatically disabled. --per-thread is ignored with a warning
358 if combined with -a or -C options.
362 After starting the program, wait msecs before measuring. This is useful to
363 filter out the startup phase of the program, which is often very different.
367 Capture machine state (registers) at interrupt, i.e., on counter overflows for
368 each sample. List of captured registers depends on the architecture. This option
369 is off by default. It is possible to select the registers to sample using their
370 symbolic names, e.g. on x86, ax, si. To list the available registers use
371 --intr-regs=\?. To name registers, pass a comma separated list such as
372 --intr-regs=ax,bx. The list of register is architecture dependent.
376 Record running and enabled time for read events (:S)
380 Sets the clock id to use for the various time fields in the perf_event_type
381 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
382 CLOCK_MONOTONIC_RAW are supported, some events might also allow
383 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
387 Select AUX area tracing Snapshot Mode. This option is valid only with an
388 AUX area tracing event. Optionally the number of bytes to capture per
389 snapshot can be specified. In Snapshot Mode, trace data is captured only when
390 signal SIGUSR2 is received.
393 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
394 because the file may be huge. A time out is needed in such cases.
395 This option sets the time out limit. The default value is 500 ms.
398 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
399 PERF_RECORD_SWITCH_CPU_WIDE.
402 Path to clang binary to use for compiling BPF scriptlets.
403 (enabled when BPF support is on)
405 --clang-opt=OPTIONS::
406 Options passed to clang when compiling BPF scriptlets.
407 (enabled when BPF support is on)
410 Specify vmlinux path which has debuginfo.
411 (enabled when BPF prologue is on)
414 Record build-id of all DSOs regardless whether it's actually hit or not.
417 Configure all used events to run in kernel space.
420 Configure all used events to run in user space.
423 Append timestamp to output file name.
426 Generate multiple perf.data files, timestamp prefixed, switching to a new one
427 when receiving a SIGUSR2.
429 A possible use case is to, given an external event, slice the perf.data file
430 that gets then processed, possibly via a perf script, to decide if that
431 particular perf.data snapshot should be kept or not.
433 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
436 Parse options then exit. --dry-run can be used to detect errors in cmdline
439 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
440 in config file is set to true.
443 Instead of collecting non-sample events (for example, fork, comm, mmap) at
444 the beginning of record, collect them during finalizing an output file.
445 The collected non-sample events reflects the status of the system when
449 Makes all events use an overwritable ring buffer. An overwritable ring
450 buffer works like a flight recorder: when it gets full, the kernel will
451 overwrite the oldest records, that thus will never make it to the
454 When '--overwrite' and '--switch-output' are used perf records and drops
455 events until it receives a signal, meaning that something unusual was
456 detected that warrants taking a snapshot of the most current events,
457 those fitting in the ring buffer at that moment.
459 'overwrite' attribute can also be set or canceled for an event using
460 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
462 Implies --tail-synthesize.
466 linkperf:perf-stat[1], linkperf:perf-list[1]