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_source/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_source/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
121 line. If the user wants to track multiple events for a specific cgroup, the user can
122 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
124 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
125 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
129 Print the output into the designated file.
132 Append to the output file designated with the -o option. Ignored if -o is not specified.
136 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
137 with it. --append may be used here. Examples:
138 3>results perf stat --log-fd 3 -- $cmd
139 3>>results perf stat --log-fd 3 --append -- $cmd
143 Pre and post measurement hooks, e.g.:
145 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
148 --interval-print msecs::
149 Print count deltas every N milliseconds (minimum: 10ms)
150 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals. Use with caution.
151 example: 'perf stat -I 1000 -e cycles -a sleep 5'
153 --interval-count times::
154 Print count deltas for fixed number of times.
155 This option should be used together with "-I" option.
156 example: 'perf stat -I 1000 --interval-count 2 -e cycles -a'
159 Stop the 'perf stat' session and print count deltas after N milliseconds (minimum: 10 ms).
160 This option is not supported with the "-I" option.
161 example: 'perf stat --time 2000 -e cycles -a'
164 Only print computed metrics. Print them in a single line.
165 Don't show any raw values. Not supported with --per-thread.
168 Aggregate counts per processor socket for system-wide mode measurements. This
169 is a useful mode to detect imbalance between sockets. To enable this mode,
170 use --per-socket in addition to -a. (system-wide). The output includes the
171 socket number and the number of online processors on that socket. This is
172 useful to gauge the amount of aggregation.
175 Aggregate counts per physical processor for system-wide mode measurements. This
176 is a useful mode to detect imbalance between physical cores. To enable this mode,
177 use --per-core in addition to -a. (system-wide). The output includes the
178 core number and the number of online logical processors on that physical processor.
181 Aggregate counts per monitored threads, when monitoring threads (-t option)
182 or processes (-p option).
186 After starting the program, wait msecs before measuring. This is useful to
187 filter out the startup phase of the program, which is often very different.
192 Print statistics of transactional execution if supported.
196 Stores stat data into perf data file.
204 Reads and reports stat data from perf data file.
211 Aggregate counts per processor socket for system-wide mode measurements.
214 Aggregate counts per physical processor for system-wide mode measurements.
218 Print metrics or metricgroups specified in a comma separated list.
219 For a group all metrics from the group are added.
220 The events from the metrics are automatically measured.
221 See perf list output for the possble metrics and metricgroups.
225 Do not aggregate counts across all monitored CPUs.
228 Print top down level 1 metrics if supported by the CPU. This allows to
229 determine bottle necks in the CPU pipeline for CPU bound workloads,
230 by breaking the cycles consumed down into frontend bound, backend bound,
231 bad speculation and retiring.
233 Frontend bound means that the CPU cannot fetch and decode instructions fast
234 enough. Backend bound means that computation or memory access is the bottle
235 neck. Bad Speculation means that the CPU wasted cycles due to branch
236 mispredictions and similar issues. Retiring means that the CPU computed without
237 an apparently bottleneck. The bottleneck is only the real bottleneck
238 if the workload is actually bound by the CPU and not by something else.
240 For best results it is usually a good idea to use it with interval
241 mode like -I 1000, as the bottleneck of workloads can change often.
243 The top down metrics are collected per core instead of per
244 CPU thread. Per core mode is automatically enabled
245 and -a (global monitoring) is needed, requiring root rights or
246 perf.perf_event_paranoid=-1.
248 Topdown uses the full Performance Monitoring Unit, and needs
249 disabling of the NMI watchdog (as root):
250 echo 0 > /proc/sys/kernel/nmi_watchdog
251 for best results. Otherwise the bottlenecks may be inconsistent
252 on workload with changing phases.
254 This enables --metric-only, unless overriden with --no-metric-only.
256 To interpret the results it is usually needed to know on which
257 CPUs the workload runs on. If needed the CPUs can be forced using
261 Do not merge results from same PMUs.
264 Measure SMI cost if msr/aperf/ and msr/smi/ events are supported.
266 During the measurement, the /sys/device/cpu/freeze_on_smi will be set to
267 freeze core counters on SMI.
268 The aperf counter will not be effected by the setting.
269 The cost of SMI can be measured by (aperf - unhalted core cycles).
271 In practice, the percentages of SMI cycles is very useful for performance
272 oriented analysis. --metric_only will be applied by default.
273 The output is SMI cycles%, equals to (aperf - unhalted core cycles) / aperf
275 Users who wants to get the actual value can apply --no-metric-only.
280 $ perf stat -- make -j
282 Performance counter stats for 'make -j':
284 8117.370256 task clock ticks # 11.281 CPU utilization factor
285 678 context switches # 0.000 M/sec
286 133 CPU migrations # 0.000 M/sec
287 235724 pagefaults # 0.029 M/sec
288 24821162526 CPU cycles # 3057.784 M/sec
289 18687303457 instructions # 2302.138 M/sec
290 172158895 cache references # 21.209 M/sec
291 27075259 cache misses # 3.335 M/sec
293 Wall-clock time elapsed: 719.554352 msecs
298 With -x, perf stat is able to output a not-quite-CSV format output
299 Commas in the output are not put into "". To make it easy to parse
300 it is recommended to use a different character like -x \;
302 The fields are in this order:
304 - optional usec time stamp in fractions of second (with -I xxx)
305 - optional CPU, core, or socket identifier
306 - optional number of logical CPUs aggregated
308 - unit of the counter value or empty
310 - run time of counter
311 - percentage of measurement time the counter was running
312 - optional variance if multiple values are collected with -r
313 - optional metric value
314 - optional unit of metric
316 Additional metrics may be printed with all earlier fields being empty.
320 linkperf:perf-top[1], linkperf:perf-list[1]