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_sources/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_sources/devices/<pmu>/format/*
45 param1 and param2 are defined as formats for the PMU in:
46 /sys/bus/event_sources/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 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
64 where addr is the address in memory you want to break in.
65 Access is the memory access type (read, write, execute) it can
66 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
67 number of bytes from specified addr, which the breakpoint will cover.
68 If you want to profile read-write accesses in 0x1000, just set
70 If you want to profile write accesses in [0x1000~1008), just set
73 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
74 Each event is separated by commas and the group should be quoted to
75 prevent the shell interpretation. You also need to use --group on
76 "perf report" to view group events together.
79 Event filter. This option should follow a event selector (-e) which
80 selects tracepoint event(s). Multiple '--filter' options are combined
84 Don't record events issued by perf itself. This option should follow
85 a event selector (-e) which selects tracepoint event(s). It adds a
86 filter expression 'common_pid != $PERFPID' to filters. If other
87 '--filter' exists, the new filter expression will be combined with
92 System-wide collection from all CPUs.
96 Record events on existing process ID (comma separated list).
100 Record events on existing thread ID (comma separated list).
101 This option also disables inheritance by default. Enable it by adding
106 Record events in threads owned by uid. Name or number.
110 Collect data with this RT SCHED_FIFO priority.
113 Collect data without buffering.
117 Event period to sample.
125 Child tasks do not inherit counters.
128 Profile at this frequency.
132 Number of mmap data pages (must be a power of two) or size
133 specification with appended unit character - B/K/M/G. The
134 size is rounded up to have nearest pages power of two value.
135 Also, by adding a comma, the number of mmap pages for AUX
136 area tracing can be specified.
139 Put all events in a single event group. This precedes the --event
140 option and remains only for backward compatibility. See --event.
143 Enables call-graph (stack chain/backtrace) recording.
146 Setup and enable call-graph (stack chain/backtrace) recording,
149 Allows specifying "fp" (frame pointer) or "dwarf"
150 (DWARF's CFI - Call Frame Information) or "lbr"
151 (Hardware Last Branch Record facility) as the method to collect
152 the information used to show the call graphs.
154 In some systems, where binaries are build with gcc
155 --fomit-frame-pointer, using the "fp" method will produce bogus
156 call graphs, using "dwarf", if available (perf tools linked to
157 the libunwind library) should be used instead.
158 Using the "lbr" method doesn't require any compiler options. It
159 will produce call graphs from the hardware LBR registers. The
160 main limition is that it is only available on new Intel
161 platforms, such as Haswell. It can only get user call chain. It
162 doesn't work with branch stack sampling at the same time.
166 Don't print any message, useful for scripting.
170 Be more verbose (show counter open errors, etc).
174 Record per-thread event counts. Use it with 'perf report -T' to see
179 Record the sample addresses.
183 Record the sample timestamps. Use it with 'perf report -D' to see the
184 timestamps, for instance.
188 Record the sample period.
196 Collect raw sample records from all opened counters (default for tracepoint counters).
200 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
201 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
202 In per-thread mode with inheritance mode on (default), samples are captured only when
203 the thread executes on the designated CPUs. Default is to monitor all CPUs.
207 Do not update the buildid cache. This saves some overhead in situations
208 where the information in the perf.data file (which includes buildids)
213 monitor only in the container (cgroup) called "name". This option is available only
214 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
215 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
216 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
217 to first event, second cgroup to second event and so on. It is possible to provide
218 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
219 corresponding events, i.e., they always refer to events defined earlier on the command
224 Enable taken branch stack sampling. Any type of taken branch may be sampled.
225 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
229 Enable taken branch stack sampling. Each sample captures a series of consecutive
230 taken branches. The number of branches captured with each sample depends on the
231 underlying hardware, the type of branches of interest, and the executed code.
232 It is possible to select the types of branches captured by enabling filters. The
233 following filters are defined:
235 - any: any type of branches
236 - any_call: any function call or system call
237 - any_ret: any function return or system call return
238 - ind_call: any indirect branch
239 - u: only when the branch target is at the user level
240 - k: only when the branch target is in the kernel
241 - hv: only when the target is at the hypervisor level
242 - in_tx: only when the target is in a hardware transaction
243 - no_tx: only when the target is not in a hardware transaction
244 - abort_tx: only when the target is a hardware transaction abort
245 - cond: conditional branches
248 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
249 The privilege levels may be omitted, in which case, the privilege levels of the associated
250 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
251 levels are subject to permissions. When sampling on multiple events, branch stack sampling
252 is enabled for all the sampling events. The sampled branch type is the same for all events.
253 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
254 Note that this feature may not be available on all processors.
257 Enable weightened sampling. An additional weight is recorded per sample and can be
258 displayed with the weight and local_weight sort keys. This currently works for TSX
259 abort events and some memory events in precise mode on modern Intel CPUs.
262 Record transaction flags for transaction related events.
265 Use per-thread mmaps. By default per-cpu mmaps are created. This option
266 overrides that and uses per-thread mmaps. A side-effect of that is that
267 inheritance is automatically disabled. --per-thread is ignored with a warning
268 if combined with -a or -C options.
272 After starting the program, wait msecs before measuring. This is useful to
273 filter out the startup phase of the program, which is often very different.
277 Capture machine state (registers) at interrupt, i.e., on counter overflows for
278 each sample. List of captured registers depends on the architecture. This option
279 is off by default. It is possible to select the registers to sample using their
280 symbolic names, e.g. on x86, ax, si. To list the available registers use
281 --intr-regs=\?. To name registers, pass a comma separated list such as
282 --intr-regs=ax,bx. The list of register is architecture dependent.
286 Record running and enabled time for read events (:S)
290 Sets the clock id to use for the various time fields in the perf_event_type
291 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
292 CLOCK_MONOTONIC_RAW are supported, some events might also allow
293 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
297 Select AUX area tracing Snapshot Mode. This option is valid only with an
298 AUX area tracing event. Optionally the number of bytes to capture per
299 snapshot can be specified. In Snapshot Mode, trace data is captured only when
300 signal SIGUSR2 is received.
303 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
304 because the file may be huge. A time out is needed in such cases.
305 This option sets the time out limit. The default value is 500 ms.
308 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
309 PERF_RECORD_SWITCH_CPU_WIDE.
313 linkperf:perf-stat[1], linkperf:perf-list[1]