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