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>]
13 'perf stat' [-e <EVENT> | --event=EVENT] [-a] record [-o file] -- <command> [<options>]
14 'perf stat' report [-i file]
18 This command runs a command and gathers performance counter statistics
25 Any command you can specify in a shell.
35 Select the PMU event. Selection can be:
37 - a symbolic event name (use 'perf list' to list all events)
39 - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
40 hexadecimal event descriptor.
42 - a symbolically formed event like 'pmu/param1=0x3,param2/' where
43 param1 and param2 are defined as formats for the PMU in
44 /sys/bus/event_sources/devices/<pmu>/format/*
46 - a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
47 where M, N, K are numbers (in decimal, hex, octal format).
48 Acceptable values for each of 'config', 'config1' and 'config2'
49 parameters are defined by corresponding entries in
50 /sys/bus/event_sources/devices/<pmu>/format/*
54 child tasks do not inherit counters
57 stat events on existing process id (comma separated list)
61 stat events on existing thread id (comma separated list)
66 system-wide collection from all CPUs (default if no target is specified)
70 scale/normalize counter values
74 print more detailed statistics, can be specified up to 3 times
76 -d: detailed events, L1 and LLC data cache
77 -d -d: more detailed events, dTLB and iTLB events
78 -d -d -d: very detailed events, adding prefetch events
82 repeat command and print average + stddev (max: 100). 0 means forever.
86 print large numbers with thousands' separators according to locale
90 Count only on the list of CPUs provided. Multiple CPUs can be provided as a
91 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
92 In per-thread mode, this option is ignored. The -a option is still necessary
93 to activate system-wide monitoring. Default is to count on all CPUs.
97 Do not aggregate counts across all monitored CPUs.
101 null run - don't start any counters
105 be more verbose (show counter open errors, etc)
108 --field-separator SEP::
109 print counts using a CSV-style output to make it easy to import directly into
110 spreadsheets. Columns are separated by the string specified in SEP.
114 monitor only in the container (cgroup) called "name". This option is available only
115 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
116 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
117 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
118 to first event, second cgroup to second event and so on. It is possible to provide
119 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
120 corresponding events, i.e., they always refer to events defined earlier on the command
125 Print the output into the designated file.
128 Append to the output file designated with the -o option. Ignored if -o is not specified.
132 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
133 with it. --append may be used here. Examples:
134 3>results perf stat --log-fd 3 -- $cmd
135 3>>results perf stat --log-fd 3 --append -- $cmd
139 Pre and post measurement hooks, e.g.:
141 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
144 --interval-print msecs::
145 Print count deltas every N milliseconds (minimum: 10ms)
146 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals. Use with caution.
147 example: 'perf stat -I 1000 -e cycles -a sleep 5'
150 Only print computed metrics. Print them in a single line.
151 Don't show any raw values. Not supported with --per-thread.
154 Aggregate counts per processor socket for system-wide mode measurements. This
155 is a useful mode to detect imbalance between sockets. To enable this mode,
156 use --per-socket in addition to -a. (system-wide). The output includes the
157 socket number and the number of online processors on that socket. This is
158 useful to gauge the amount of aggregation.
161 Aggregate counts per physical processor for system-wide mode measurements. This
162 is a useful mode to detect imbalance between physical cores. To enable this mode,
163 use --per-core in addition to -a. (system-wide). The output includes the
164 core number and the number of online logical processors on that physical processor.
167 Aggregate counts per monitored threads, when monitoring threads (-t option)
168 or processes (-p option).
172 After starting the program, wait msecs before measuring. This is useful to
173 filter out the startup phase of the program, which is often very different.
178 Print statistics of transactional execution if supported.
182 Stores stat data into perf data file.
190 Reads and reports stat data from perf data file.
197 Aggregate counts per processor socket for system-wide mode measurements.
200 Aggregate counts per physical processor for system-wide mode measurements.
204 Do not aggregate counts across all monitored CPUs.
207 Print top down level 1 metrics if supported by the CPU. This allows to
208 determine bottle necks in the CPU pipeline for CPU bound workloads,
209 by breaking the cycles consumed down into frontend bound, backend bound,
210 bad speculation and retiring.
212 Frontend bound means that the CPU cannot fetch and decode instructions fast
213 enough. Backend bound means that computation or memory access is the bottle
214 neck. Bad Speculation means that the CPU wasted cycles due to branch
215 mispredictions and similar issues. Retiring means that the CPU computed without
216 an apparently bottleneck. The bottleneck is only the real bottleneck
217 if the workload is actually bound by the CPU and not by something else.
219 For best results it is usually a good idea to use it with interval
220 mode like -I 1000, as the bottleneck of workloads can change often.
222 The top down metrics are collected per core instead of per
223 CPU thread. Per core mode is automatically enabled
224 and -a (global monitoring) is needed, requiring root rights or
225 perf.perf_event_paranoid=-1.
227 Topdown uses the full Performance Monitoring Unit, and needs
228 disabling of the NMI watchdog (as root):
229 echo 0 > /proc/sys/kernel/nmi_watchdog
230 for best results. Otherwise the bottlenecks may be inconsistent
231 on workload with changing phases.
233 This enables --metric-only, unless overriden with --no-metric-only.
235 To interpret the results it is usually needed to know on which
236 CPUs the workload runs on. If needed the CPUs can be forced using
240 Do not merge results from same PMUs.
245 $ perf stat -- make -j
247 Performance counter stats for 'make -j':
249 8117.370256 task clock ticks # 11.281 CPU utilization factor
250 678 context switches # 0.000 M/sec
251 133 CPU migrations # 0.000 M/sec
252 235724 pagefaults # 0.029 M/sec
253 24821162526 CPU cycles # 3057.784 M/sec
254 18687303457 instructions # 2302.138 M/sec
255 172158895 cache references # 21.209 M/sec
256 27075259 cache misses # 3.335 M/sec
258 Wall-clock time elapsed: 719.554352 msecs
263 With -x, perf stat is able to output a not-quite-CSV format output
264 Commas in the output are not put into "". To make it easy to parse
265 it is recommended to use a different character like -x \;
267 The fields are in this order:
269 - optional usec time stamp in fractions of second (with -I xxx)
270 - optional CPU, core, or socket identifier
271 - optional number of logical CPUs aggregated
273 - unit of the counter value or empty
275 - run time of counter
276 - percentage of measurement time the counter was running
277 - optional variance if multiple values are collected with -r
278 - optional metric value
279 - optional unit of metric
281 Additional metrics may be printed with all earlier fields being empty.
285 linkperf:perf-top[1], linkperf:perf-list[1]