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