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 - a hardware breakpoint event in the form of '\mem:addr[:access]'
49 where addr is the address in memory you want to break in.
50 Access is the memory access type (read, write, execute) it can
51 be passed as follows: '\mem:addr[:[r][w][x]]'.
52 If you want to profile read-write accesses in 0x1000, just set
60 System-wide collection from all CPUs.
67 Record events on existing process ID (comma separated list).
71 Record events on existing thread ID (comma separated list).
72 This option also disables inheritance by default. Enable it by adding
77 Record events in threads owned by uid. Name or number.
81 Collect data with this RT SCHED_FIFO priority.
84 Collect data without buffering.
88 Event period to sample.
96 Child tasks do not inherit counters.
99 Profile at this frequency.
103 Number of mmap data pages (must be a power of two) or size
104 specification with appended unit character - B/K/M/G. The
105 size is rounded up to have nearest pages power of two value.
108 Enables call-graph (stack chain/backtrace) recording.
111 Setup and enable call-graph (stack chain/backtrace) recording,
114 Allows specifying "fp" (frame pointer) or "dwarf"
115 (DWARF's CFI - Call Frame Information) as the method to collect
116 the information used to show the call graphs.
118 In some systems, where binaries are build with gcc
119 --fomit-frame-pointer, using the "fp" method will produce bogus
120 call graphs, using "dwarf", if available (perf tools linked to
121 the libunwind library) should be used instead.
125 Don't print any message, useful for scripting.
129 Be more verbose (show counter open errors, etc).
141 Sample timestamps. Use it with 'perf report -D' to see the timestamps,
150 Collect raw sample records from all opened counters (default for tracepoint counters).
154 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
155 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
156 In per-thread mode with inheritance mode on (default), samples are captured only when
157 the thread executes on the designated CPUs. Default is to monitor all CPUs.
161 Do not update the buildid cache. This saves some overhead in situations
162 where the information in the perf.data file (which includes buildids)
167 monitor only in the container (cgroup) called "name". This option is available only
168 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
169 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
170 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
171 to first event, second cgroup to second event and so on. It is possible to provide
172 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
173 corresponding events, i.e., they always refer to events defined earlier on the command
178 Enable taken branch stack sampling. Any type of taken branch may be sampled.
179 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
183 Enable taken branch stack sampling. Each sample captures a series of consecutive
184 taken branches. The number of branches captured with each sample depends on the
185 underlying hardware, the type of branches of interest, and the executed code.
186 It is possible to select the types of branches captured by enabling filters. The
187 following filters are defined:
189 - any: any type of branches
190 - any_call: any function call or system call
191 - any_ret: any function return or system call return
192 - ind_call: any indirect branch
193 - u: only when the branch target is at the user level
194 - k: only when the branch target is in the kernel
195 - hv: only when the target is at the hypervisor level
196 - in_tx: only when the target is in a hardware transaction
197 - no_tx: only when the target is not in a hardware transaction
198 - abort_tx: only when the target is a hardware transaction abort
199 - cond: conditional branches
202 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
203 The privilege levels may be omitted, in which case, the privilege levels of the associated
204 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
205 levels are subject to permissions. When sampling on multiple events, branch stack sampling
206 is enabled for all the sampling events. The sampled branch type is the same for all events.
207 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
208 Note that this feature may not be available on all processors.
211 Enable weightened sampling. An additional weight is recorded per sample and can be
212 displayed with the weight and local_weight sort keys. This currently works for TSX
213 abort events and some memory events in precise mode on modern Intel CPUs.
216 Record transaction flags for transaction related events.
219 Use per-thread mmaps. By default per-cpu mmaps are created. This option
220 overrides that and uses per-thread mmaps. A side-effect of that is that
221 inheritance is automatically disabled. --per-thread is ignored with a warning
222 if combined with -a or -C options.
226 After starting the program, wait msecs before measuring. This is useful to
227 filter out the startup phase of the program, which is often very different.
231 Capture machine state (registers) at interrupt, i.e., on counter overflows for
232 each sample. List of captured registers depends on the architecture. This option
237 linkperf:perf-stat[1], linkperf:perf-list[1]