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_sources/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_sources/devices/<pmu>/format/*
49 param1 and param2 are defined as formats for the PMU in:
50 /sys/bus/event_sources/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 - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
68 where addr is the address in memory you want to break in.
69 Access is the memory access type (read, write, execute) it can
70 be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
71 number of bytes from specified addr, which the breakpoint will cover.
72 If you want to profile read-write accesses in 0x1000, just set
74 If you want to profile write accesses in [0x1000~1008), just set
77 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
78 Each event is separated by commas and the group should be quoted to
79 prevent the shell interpretation. You also need to use --group on
80 "perf report" to view group events together.
83 Event filter. This option should follow a event selector (-e) which
84 selects tracepoint event(s). Multiple '--filter' options are combined
88 Don't record events issued by perf itself. This option should follow
89 a event selector (-e) which selects tracepoint event(s). It adds a
90 filter expression 'common_pid != $PERFPID' to filters. If other
91 '--filter' exists, the new filter expression will be combined with
96 System-wide collection from all CPUs.
100 Record events on existing process ID (comma separated list).
104 Record events on existing thread ID (comma separated list).
105 This option also disables inheritance by default. Enable it by adding
110 Record events in threads owned by uid. Name or number.
114 Collect data with this RT SCHED_FIFO priority.
117 Collect data without buffering.
121 Event period to sample.
129 Child tasks do not inherit counters.
132 Profile at this frequency.
136 Number of mmap data pages (must be a power of two) or size
137 specification with appended unit character - B/K/M/G. The
138 size is rounded up to have nearest pages power of two value.
139 Also, by adding a comma, the number of mmap pages for AUX
140 area tracing can be specified.
143 Put all events in a single event group. This precedes the --event
144 option and remains only for backward compatibility. See --event.
147 Enables call-graph (stack chain/backtrace) recording.
150 Setup and enable call-graph (stack chain/backtrace) recording,
151 implies -g. Default is "fp".
153 Allows specifying "fp" (frame pointer) or "dwarf"
154 (DWARF's CFI - Call Frame Information) or "lbr"
155 (Hardware Last Branch Record facility) as the method to collect
156 the information used to show the call graphs.
158 In some systems, where binaries are build with gcc
159 --fomit-frame-pointer, using the "fp" method will produce bogus
160 call graphs, using "dwarf", if available (perf tools linked to
161 the libunwind or libdw library) should be used instead.
162 Using the "lbr" method doesn't require any compiler options. It
163 will produce call graphs from the hardware LBR registers. The
164 main limition is that it is only available on new Intel
165 platforms, such as Haswell. It can only get user call chain. It
166 doesn't work with branch stack sampling at the same time.
168 When "dwarf" recording is used, perf also records (user) stack dump
169 when sampled. Default size of the stack dump is 8192 (bytes).
170 User can change the size by passing the size after comma like
171 "--call-graph dwarf,4096".
175 Don't print any message, useful for scripting.
179 Be more verbose (show counter open errors, etc).
183 Record per-thread event counts. Use it with 'perf report -T' to see
188 Record the sample addresses.
192 Record the sample timestamps. Use it with 'perf report -D' to see the
193 timestamps, for instance.
197 Record the sample period.
205 Collect raw sample records from all opened counters (default for tracepoint counters).
209 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
210 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
211 In per-thread mode with inheritance mode on (default), samples are captured only when
212 the thread executes on the designated CPUs. Default is to monitor all CPUs.
216 Do not update the buildid cache. This saves some overhead in situations
217 where the information in the perf.data file (which includes buildids)
222 monitor only in the container (cgroup) called "name". This option is available only
223 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
224 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
225 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
226 to first event, second cgroup to second event and so on. It is possible to provide
227 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
228 corresponding events, i.e., they always refer to events defined earlier on the command
233 Enable taken branch stack sampling. Any type of taken branch may be sampled.
234 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
238 Enable taken branch stack sampling. Each sample captures a series of consecutive
239 taken branches. The number of branches captured with each sample depends on the
240 underlying hardware, the type of branches of interest, and the executed code.
241 It is possible to select the types of branches captured by enabling filters. The
242 following filters are defined:
244 - any: any type of branches
245 - any_call: any function call or system call
246 - any_ret: any function return or system call return
247 - ind_call: any indirect branch
248 - call: direct calls, including far (to/from kernel) calls
249 - u: only when the branch target is at the user level
250 - k: only when the branch target is in the kernel
251 - hv: only when the target is at the hypervisor level
252 - in_tx: only when the target is in a hardware transaction
253 - no_tx: only when the target is not in a hardware transaction
254 - abort_tx: only when the target is a hardware transaction abort
255 - cond: conditional branches
258 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
259 The privilege levels may be omitted, in which case, the privilege levels of the associated
260 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
261 levels are subject to permissions. When sampling on multiple events, branch stack sampling
262 is enabled for all the sampling events. The sampled branch type is the same for all events.
263 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
264 Note that this feature may not be available on all processors.
267 Enable weightened sampling. An additional weight is recorded per sample and can be
268 displayed with the weight and local_weight sort keys. This currently works for TSX
269 abort events and some memory events in precise mode on modern Intel CPUs.
272 Record transaction flags for transaction related events.
275 Use per-thread mmaps. By default per-cpu mmaps are created. This option
276 overrides that and uses per-thread mmaps. A side-effect of that is that
277 inheritance is automatically disabled. --per-thread is ignored with a warning
278 if combined with -a or -C options.
282 After starting the program, wait msecs before measuring. This is useful to
283 filter out the startup phase of the program, which is often very different.
287 Capture machine state (registers) at interrupt, i.e., on counter overflows for
288 each sample. List of captured registers depends on the architecture. This option
289 is off by default. It is possible to select the registers to sample using their
290 symbolic names, e.g. on x86, ax, si. To list the available registers use
291 --intr-regs=\?. To name registers, pass a comma separated list such as
292 --intr-regs=ax,bx. The list of register is architecture dependent.
296 Record running and enabled time for read events (:S)
300 Sets the clock id to use for the various time fields in the perf_event_type
301 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
302 CLOCK_MONOTONIC_RAW are supported, some events might also allow
303 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
307 Select AUX area tracing Snapshot Mode. This option is valid only with an
308 AUX area tracing event. Optionally the number of bytes to capture per
309 snapshot can be specified. In Snapshot Mode, trace data is captured only when
310 signal SIGUSR2 is received.
313 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
314 because the file may be huge. A time out is needed in such cases.
315 This option sets the time out limit. The default value is 500 ms.
318 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
319 PERF_RECORD_SWITCH_CPU_WIDE.
322 Path to clang binary to use for compiling BPF scriptlets.
325 Options passed to clang when compiling BPF scriptlets.
329 linkperf:perf-stat[1], linkperf:perf-list[1]