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\'.
64 See the linkperf:perf-list[1] man page for more parameters.
66 Note: If user explicitly sets options which conflict with the params,
67 the value set by the parameters will be overridden.
69 Also not defined in .../<pmu>/format/* are PMU driver specific
70 configuration parameters. Any configuration parameter preceded by
71 the letter '@' is not interpreted in user space and sent down directly
72 to the PMU driver. For example:
74 perf record -e some_event/@cfg1,@cfg2=config/ ...
76 will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
77 with the event for further processing. There is no restriction on
78 what the configuration parameters are, as long as their semantic is
79 understood and supported by the PMU driver.
81 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
82 where addr is the address in memory you want to break in.
83 Access is the memory access type (read, write, execute) it can
84 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
85 number of bytes from specified addr, which the breakpoint will cover.
86 If you want to profile read-write accesses in 0x1000, just set
88 If you want to profile write accesses in [0x1000~1008), just set
91 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
92 Each event is separated by commas and the group should be quoted to
93 prevent the shell interpretation. You also need to use --group on
94 "perf report" to view group events together.
97 Event filter. This option should follow an event selector (-e) which
98 selects either tracepoint event(s) or a hardware trace PMU
99 (e.g. Intel PT or CoreSight).
103 In the case of tracepoints, multiple '--filter' options are combined
108 A hardware trace PMU advertises its ability to accept a number of
109 address filters by specifying a non-zero value in
110 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
112 Address filters have the format:
114 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
117 - 'filter': defines a region that will be traced.
118 - 'start': defines an address at which tracing will begin.
119 - 'stop': defines an address at which tracing will stop.
120 - 'tracestop': defines a region in which tracing will stop.
122 <file name> is the name of the object file, <start> is the offset to the
123 code to trace in that file, and <size> is the size of the region to
124 trace. 'start' and 'stop' filters need not specify a <size>.
126 If no object file is specified then the kernel is assumed, in which case
127 the start address must be a current kernel memory address.
129 <start> can also be specified by providing the name of a symbol. If the
130 symbol name is not unique, it can be disambiguated by inserting #n where
131 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
132 select only a global symbol. <size> can also be specified by providing
133 the name of a symbol, in which case the size is calculated to the end
134 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
135 omitted and <start> is a symbol, then the size is calculated to the end
138 If <size> is omitted and <start> is '*', then the start and size will
139 be calculated from the first and last symbols, i.e. to trace the whole
142 If symbol names (or '*') are provided, they must be surrounded by white
145 The filter passed to the kernel is not necessarily the same as entered.
146 To see the filter that is passed, use the -v option.
148 The kernel may not be able to configure a trace region if it is not
149 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
150 examined to determine if that is a possibility.
152 Multiple filters can be separated with space or comma.
155 Don't record events issued by perf itself. This option should follow
156 an event selector (-e) which selects tracepoint event(s). It adds a
157 filter expression 'common_pid != $PERFPID' to filters. If other
158 '--filter' exists, the new filter expression will be combined with
163 System-wide collection from all CPUs (default if no target is specified).
167 Record events on existing process ID (comma separated list).
171 Record events on existing thread ID (comma separated list).
172 This option also disables inheritance by default. Enable it by adding
177 Record events in threads owned by uid. Name or number.
181 Collect data with this RT SCHED_FIFO priority.
184 Collect data without buffering.
188 Event period to sample.
196 Child tasks do not inherit counters.
200 Profile at this frequency. Use 'max' to use the currently maximum
201 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
202 sysctl. Will throttle down to the currently maximum allowed frequency.
206 Fail if the specified frequency can't be used.
210 Number of mmap data pages (must be a power of two) or size
211 specification with appended unit character - B/K/M/G. The
212 size is rounded up to have nearest pages power of two value.
213 Also, by adding a comma, the number of mmap pages for AUX
214 area tracing can be specified.
217 Put all events in a single event group. This precedes the --event
218 option and remains only for backward compatibility. See --event.
221 Enables call-graph (stack chain/backtrace) recording.
224 Setup and enable call-graph (stack chain/backtrace) recording,
225 implies -g. Default is "fp".
227 Allows specifying "fp" (frame pointer) or "dwarf"
228 (DWARF's CFI - Call Frame Information) or "lbr"
229 (Hardware Last Branch Record facility) as the method to collect
230 the information used to show the call graphs.
232 In some systems, where binaries are build with gcc
233 --fomit-frame-pointer, using the "fp" method will produce bogus
234 call graphs, using "dwarf", if available (perf tools linked to
235 the libunwind or libdw library) should be used instead.
236 Using the "lbr" method doesn't require any compiler options. It
237 will produce call graphs from the hardware LBR registers. The
238 main limitation is that it is only available on new Intel
239 platforms, such as Haswell. It can only get user call chain. It
240 doesn't work with branch stack sampling at the same time.
242 When "dwarf" recording is used, perf also records (user) stack dump
243 when sampled. Default size of the stack dump is 8192 (bytes).
244 User can change the size by passing the size after comma like
245 "--call-graph dwarf,4096".
249 Don't print any message, useful for scripting.
253 Be more verbose (show counter open errors, etc).
257 Record per-thread event counts. Use it with 'perf report -T' to see
262 Record the sample virtual addresses.
265 Record the sample physical addresses.
269 Record the sample timestamps. Use it with 'perf report -D' to see the
270 timestamps, for instance.
274 Record the sample period.
277 Record the sample cpu.
285 Collect raw sample records from all opened counters (default for tracepoint counters).
289 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
290 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
291 In per-thread mode with inheritance mode on (default), samples are captured only when
292 the thread executes on the designated CPUs. Default is to monitor all CPUs.
296 Do not save the build ids of binaries in the perf.data files. This skips
297 post processing after recording, which sometimes makes the final step in
298 the recording process to take a long time, as it needs to process all
299 events looking for mmap records. The downside is that it can misresolve
300 symbols if the workload binaries used when recording get locally rebuilt
301 or upgraded, because the only key available in this case is the
302 pathname. You can also set the "record.build-id" config variable to
303 'skip to have this behaviour permanently.
307 Do not update the buildid cache. This saves some overhead in situations
308 where the information in the perf.data file (which includes buildids)
309 is sufficient. You can also set the "record.build-id" config variable to
310 'no-cache' to have the same effect.
314 monitor only in the container (cgroup) called "name". This option is available only
315 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
316 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
317 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
318 to first event, second cgroup to second event and so on. It is possible to provide
319 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
320 corresponding events, i.e., they always refer to events defined earlier on the command
321 line. If the user wants to track multiple events for a specific cgroup, the user can
322 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
324 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
325 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
329 Enable taken branch stack sampling. Any type of taken branch may be sampled.
330 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
334 Enable taken branch stack sampling. Each sample captures a series of consecutive
335 taken branches. The number of branches captured with each sample depends on the
336 underlying hardware, the type of branches of interest, and the executed code.
337 It is possible to select the types of branches captured by enabling filters. The
338 following filters are defined:
340 - any: any type of branches
341 - any_call: any function call or system call
342 - any_ret: any function return or system call return
343 - ind_call: any indirect branch
344 - call: direct calls, including far (to/from kernel) calls
345 - u: only when the branch target is at the user level
346 - k: only when the branch target is in the kernel
347 - hv: only when the target is at the hypervisor level
348 - in_tx: only when the target is in a hardware transaction
349 - no_tx: only when the target is not in a hardware transaction
350 - abort_tx: only when the target is a hardware transaction abort
351 - cond: conditional branches
352 - save_type: save branch type during sampling in case binary is not available later
355 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
356 The privilege levels may be omitted, in which case, the privilege levels of the associated
357 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
358 levels are subject to permissions. When sampling on multiple events, branch stack sampling
359 is enabled for all the sampling events. The sampled branch type is the same for all events.
360 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
361 Note that this feature may not be available on all processors.
364 Enable weightened sampling. An additional weight is recorded per sample and can be
365 displayed with the weight and local_weight sort keys. This currently works for TSX
366 abort events and some memory events in precise mode on modern Intel CPUs.
369 Record events of type PERF_RECORD_NAMESPACES.
372 Record transaction flags for transaction related events.
375 Use per-thread mmaps. By default per-cpu mmaps are created. This option
376 overrides that and uses per-thread mmaps. A side-effect of that is that
377 inheritance is automatically disabled. --per-thread is ignored with a warning
378 if combined with -a or -C options.
382 After starting the program, wait msecs before measuring. This is useful to
383 filter out the startup phase of the program, which is often very different.
387 Capture machine state (registers) at interrupt, i.e., on counter overflows for
388 each sample. List of captured registers depends on the architecture. This option
389 is off by default. It is possible to select the registers to sample using their
390 symbolic names, e.g. on x86, ax, si. To list the available registers use
391 --intr-regs=\?. To name registers, pass a comma separated list such as
392 --intr-regs=ax,bx. The list of register is architecture dependent.
395 Capture user registers at sample time. Same arguments as -I.
398 Record running and enabled time for read events (:S)
402 Sets the clock id to use for the various time fields in the perf_event_type
403 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
404 CLOCK_MONOTONIC_RAW are supported, some events might also allow
405 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
409 Select AUX area tracing Snapshot Mode. This option is valid only with an
410 AUX area tracing event. Optionally the number of bytes to capture per
411 snapshot can be specified. In Snapshot Mode, trace data is captured only when
412 signal SIGUSR2 is received.
415 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
416 because the file may be huge. A time out is needed in such cases.
417 This option sets the time out limit. The default value is 500 ms.
420 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
421 PERF_RECORD_SWITCH_CPU_WIDE.
424 Path to clang binary to use for compiling BPF scriptlets.
425 (enabled when BPF support is on)
427 --clang-opt=OPTIONS::
428 Options passed to clang when compiling BPF scriptlets.
429 (enabled when BPF support is on)
432 Specify vmlinux path which has debuginfo.
433 (enabled when BPF prologue is on)
436 Record build-id of all DSOs regardless whether it's actually hit or not.
439 Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
440 Asynchronous mode is supported only when linking Perf tool with libc library
441 providing implementation for Posix AIO API.
444 Configure all used events to run in kernel space.
447 Configure all used events to run in user space.
450 Append timestamp to output file name.
452 --timestamp-boundary::
453 Record timestamp boundary (time of first/last samples).
455 --switch-output[=mode]::
456 Generate multiple perf.data files, timestamp prefixed, switching to a new one
457 based on 'mode' value:
458 "signal" - when receiving a SIGUSR2 (default value) or
459 <size> - when reaching the size threshold, size is expected to
460 be a number with appended unit character - B/K/M/G
461 <time> - when reaching the time threshold, size is expected to
462 be a number with appended unit character - s/m/h/d
464 Note: the precision of the size threshold hugely depends
465 on your configuration - the number and size of your ring
466 buffers (-m). It is generally more precise for higher sizes
467 (like >5M), for lower values expect different sizes.
469 A possible use case is to, given an external event, slice the perf.data file
470 that gets then processed, possibly via a perf script, to decide if that
471 particular perf.data snapshot should be kept or not.
473 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
474 The reason for the latter two is to reduce the data file switching
475 overhead. You can still switch them on with:
477 --switch-output --no-no-buildid --no-no-buildid-cache
480 Parse options then exit. --dry-run can be used to detect errors in cmdline
483 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
484 in config file is set to true.
487 Instead of collecting non-sample events (for example, fork, comm, mmap) at
488 the beginning of record, collect them during finalizing an output file.
489 The collected non-sample events reflects the status of the system when
493 Makes all events use an overwritable ring buffer. An overwritable ring
494 buffer works like a flight recorder: when it gets full, the kernel will
495 overwrite the oldest records, that thus will never make it to the
498 When '--overwrite' and '--switch-output' are used perf records and drops
499 events until it receives a signal, meaning that something unusual was
500 detected that warrants taking a snapshot of the most current events,
501 those fitting in the ring buffer at that moment.
503 'overwrite' attribute can also be set or canceled for an event using
504 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
506 Implies --tail-synthesize.
510 linkperf:perf-stat[1], linkperf:perf-list[1]
projects / linux / blob