6 perf-stat - Run a command and gather performance counter statistics
11 'perf stat' [-e <EVENT> | --event=EVENT] [-a] <command>
12 'perf stat' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
16 This command runs a command and gathers performance counter statistics
23 Any command you can specify in a shell.
28 Select the PMU event. Selection can be:
30 - a symbolic event name (use 'perf list' to list all events)
32 - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
33 hexadecimal event descriptor.
35 - a symbolic or raw PMU event followed by an optional colon
36 and a list of event modifiers, e.g., cpu-cycles:p. See the
37 linkperf:perf-list[1] man page for details on event modifiers.
39 - a symbolically formed event like 'pmu/param1=0x3,param2/' where
40 param1 and param2 are defined as formats for the PMU in
41 /sys/bus/event_sources/devices/<pmu>/format/*
43 - a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
44 where M, N, K are numbers (in decimal, hex, octal format).
45 Acceptable values for each of 'config', 'config1' and 'config2'
46 parameters are defined by corresponding entries in
47 /sys/bus/event_sources/devices/<pmu>/format/*
51 child tasks do not inherit counters
54 stat events on existing process id (comma separated list)
58 stat events on existing thread id (comma separated list)
63 system-wide collection from all CPUs
67 scale/normalize counter values
71 print more detailed statistics, can be specified up to 3 times
73 -d: detailed events, L1 and LLC data cache
74 -d -d: more detailed events, dTLB and iTLB events
75 -d -d -d: very detailed events, adding prefetch events
79 repeat command and print average + stddev (max: 100). 0 means forever.
83 print large numbers with thousands' separators according to locale
87 Count only on the list of CPUs provided. Multiple CPUs can be provided as a
88 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
89 In per-thread mode, this option is ignored. The -a option is still necessary
90 to activate system-wide monitoring. Default is to count on all CPUs.
94 Do not aggregate counts across all monitored CPUs in system-wide mode (-a).
95 This option is only valid in system-wide mode.
99 null run - don't start any counters
103 be more verbose (show counter open errors, etc)
106 --field-separator SEP::
107 print counts using a CSV-style output to make it easy to import directly into
108 spreadsheets. Columns are separated by the string specified in SEP.
112 monitor only in the container (cgroup) called "name". This option is available only
113 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
114 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
115 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
116 to first event, second cgroup to second event and so on. It is possible to provide
117 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
118 corresponding events, i.e., they always refer to events defined earlier on the command
123 Print the output into the designated file.
126 Append to the output file designated with the -o option. Ignored if -o is not specified.
130 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
131 with it. --append may be used here. Examples:
132 3>results perf stat --log-fd 3 -- $cmd
133 3>>results perf stat --log-fd 3 --append -- $cmd
137 Pre and post measurement hooks, e.g.:
139 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
142 --interval-print msecs::
143 Print count deltas every N milliseconds (minimum: 10ms)
144 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals. Use with caution.
145 example: 'perf stat -I 1000 -e cycles -a sleep 5'
148 Aggregate counts per processor socket for system-wide mode measurements. This
149 is a useful mode to detect imbalance between sockets. To enable this mode,
150 use --per-socket in addition to -a. (system-wide). The output includes the
151 socket number and the number of online processors on that socket. This is
152 useful to gauge the amount of aggregation.
155 Aggregate counts per physical processor for system-wide mode measurements. This
156 is a useful mode to detect imbalance between physical cores. To enable this mode,
157 use --per-core in addition to -a. (system-wide). The output includes the
158 core number and the number of online logical processors on that physical processor.
161 Aggregate counts per monitored threads, when monitoring threads (-t option)
162 or processes (-p option).
166 After starting the program, wait msecs before measuring. This is useful to
167 filter out the startup phase of the program, which is often very different.
172 Print statistics of transactional execution if supported.
177 $ perf stat -- make -j
179 Performance counter stats for 'make -j':
181 8117.370256 task clock ticks # 11.281 CPU utilization factor
182 678 context switches # 0.000 M/sec
183 133 CPU migrations # 0.000 M/sec
184 235724 pagefaults # 0.029 M/sec
185 24821162526 CPU cycles # 3057.784 M/sec
186 18687303457 instructions # 2302.138 M/sec
187 172158895 cache references # 21.209 M/sec
188 27075259 cache misses # 3.335 M/sec
190 Wall-clock time elapsed: 719.554352 msecs
194 linkperf:perf-top[1], linkperf:perf-list[1]