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[/len][: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]]'. len is the range,
52 number of bytes from specified addr, which the breakpoint will cover.
53 If you want to profile read-write accesses in 0x1000, just set
55 If you want to profile write accesses in [0x1000~1008), just set
58 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
59 Each event is separated by commas and the group should be quoted to
60 prevent the shell interpretation. You also need to use --group on
61 "perf report" to view group events together.
68 System-wide collection from all CPUs.
72 Record events on existing process ID (comma separated list).
76 Record events on existing thread ID (comma separated list).
77 This option also disables inheritance by default. Enable it by adding
82 Record events in threads owned by uid. Name or number.
86 Collect data with this RT SCHED_FIFO priority.
89 Collect data without buffering.
93 Event period to sample.
101 Child tasks do not inherit counters.
104 Profile at this frequency.
108 Number of mmap data pages (must be a power of two) or size
109 specification with appended unit character - B/K/M/G. The
110 size is rounded up to have nearest pages power of two value.
113 Put all events in a single event group. This precedes the --event
114 option and remains only for backward compatibility. See --event.
117 Enables call-graph (stack chain/backtrace) recording.
120 Setup and enable call-graph (stack chain/backtrace) recording,
123 Allows specifying "fp" (frame pointer) or "dwarf"
124 (DWARF's CFI - Call Frame Information) or "lbr"
125 (Hardware Last Branch Record facility) as the method to collect
126 the information used to show the call graphs.
128 In some systems, where binaries are build with gcc
129 --fomit-frame-pointer, using the "fp" method will produce bogus
130 call graphs, using "dwarf", if available (perf tools linked to
131 the libunwind library) should be used instead.
132 Using the "lbr" method doesn't require any compiler options. It
133 will produce call graphs from the hardware LBR registers. The
134 main limition is that it is only available on new Intel
135 platforms, such as Haswell. It can only get user call chain. It
136 doesn't work with branch stack sampling at the same time.
140 Don't print any message, useful for scripting.
144 Be more verbose (show counter open errors, etc).
156 Sample timestamps. Use it with 'perf report -D' to see the timestamps,
165 Collect raw sample records from all opened counters (default for tracepoint counters).
169 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
170 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
171 In per-thread mode with inheritance mode on (default), samples are captured only when
172 the thread executes on the designated CPUs. Default is to monitor all CPUs.
176 Do not update the buildid cache. This saves some overhead in situations
177 where the information in the perf.data file (which includes buildids)
182 monitor only in the container (cgroup) called "name". This option is available only
183 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
184 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
185 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
186 to first event, second cgroup to second event and so on. It is possible to provide
187 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
188 corresponding events, i.e., they always refer to events defined earlier on the command
193 Enable taken branch stack sampling. Any type of taken branch may be sampled.
194 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
198 Enable taken branch stack sampling. Each sample captures a series of consecutive
199 taken branches. The number of branches captured with each sample depends on the
200 underlying hardware, the type of branches of interest, and the executed code.
201 It is possible to select the types of branches captured by enabling filters. The
202 following filters are defined:
204 - any: any type of branches
205 - any_call: any function call or system call
206 - any_ret: any function return or system call return
207 - ind_call: any indirect branch
208 - u: only when the branch target is at the user level
209 - k: only when the branch target is in the kernel
210 - hv: only when the target is at the hypervisor level
211 - in_tx: only when the target is in a hardware transaction
212 - no_tx: only when the target is not in a hardware transaction
213 - abort_tx: only when the target is a hardware transaction abort
214 - cond: conditional branches
217 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
218 The privilege levels may be omitted, in which case, the privilege levels of the associated
219 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
220 levels are subject to permissions. When sampling on multiple events, branch stack sampling
221 is enabled for all the sampling events. The sampled branch type is the same for all events.
222 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
223 Note that this feature may not be available on all processors.
226 Enable weightened sampling. An additional weight is recorded per sample and can be
227 displayed with the weight and local_weight sort keys. This currently works for TSX
228 abort events and some memory events in precise mode on modern Intel CPUs.
231 Record transaction flags for transaction related events.
234 Use per-thread mmaps. By default per-cpu mmaps are created. This option
235 overrides that and uses per-thread mmaps. A side-effect of that is that
236 inheritance is automatically disabled. --per-thread is ignored with a warning
237 if combined with -a or -C options.
241 After starting the program, wait msecs before measuring. This is useful to
242 filter out the startup phase of the program, which is often very different.
246 Capture machine state (registers) at interrupt, i.e., on counter overflows for
247 each sample. List of captured registers depends on the architecture. This option
251 Record running and enabled time for read events (:S)
255 Sets the clock id to use for the various time fields in the perf_event_type
256 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
257 CLOCK_MONOTONIC_RAW are supported, some events might also allow
258 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
262 linkperf:perf-stat[1], linkperf:perf-list[1]