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 symbolic or raw PMU event followed by an optional colon
43 and a list of event modifiers, e.g., cpu-cycles:p. See the
44 linkperf:perf-list[1] man page for details on event modifiers.
46 - a symbolically formed event like 'pmu/param1=0x3,param2/' where
47 param1 and param2 are defined as formats for the PMU in
48 /sys/bus/event_source/devices/<pmu>/format/*
50 'percore' is a event qualifier that sums up the event counts for both
51 hardware threads in a core. For example:
52 perf stat -A -a -e cpu/event,percore=1/,otherevent ...
54 - a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
55 where M, N, K are numbers (in decimal, hex, octal format).
56 Acceptable values for each of 'config', 'config1' and 'config2'
57 parameters are defined by corresponding entries in
58 /sys/bus/event_source/devices/<pmu>/format/*
60 Note that the last two syntaxes support prefix and glob matching in
61 the PMU name to simplify creation of events across multiple instances
62 of the same type of PMU in large systems (e.g. memory controller PMUs).
63 Multiple PMU instances are typical for uncore PMUs, so the prefix
64 'uncore_' is also ignored when performing this match.
69 child tasks do not inherit counters
72 stat events on existing process id (comma separated list)
76 stat events on existing thread id (comma separated list)
80 Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
81 including support for event filters. For example '--pfm-events
82 inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
83 option using the comma separator. Hardware events and generic hardware
84 events cannot be mixed together. The latter must be used with the -e
85 option. The -e option and this one can be mixed and matched. Events
86 can be grouped using the {} notation.
91 system-wide collection from all CPUs (default if no target is specified)
94 Don't scale/normalize counter values
98 print more detailed statistics, can be specified up to 3 times
100 -d: detailed events, L1 and LLC data cache
101 -d -d: more detailed events, dTLB and iTLB events
102 -d -d -d: very detailed events, adding prefetch events
106 repeat command and print average + stddev (max: 100). 0 means forever.
110 print large numbers with thousands' separators according to locale.
111 Enabled by default. Use "--no-big-num" to disable.
112 Default setting can be changed with "perf config stat.big-num=false".
116 Count only on the list of CPUs provided. Multiple CPUs can be provided as a
117 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
118 In per-thread mode, this option is ignored. The -a option is still necessary
119 to activate system-wide monitoring. Default is to count on all CPUs.
123 Do not aggregate counts across all monitored CPUs.
127 null run - don't start any counters
131 be more verbose (show counter open errors, etc)
134 --field-separator SEP::
135 print counts using a CSV-style output to make it easy to import directly into
136 spreadsheets. Columns are separated by the string specified in SEP.
138 --table:: Display time for each run (-r option), in a table format, e.g.:
140 $ perf stat --null -r 5 --table perf bench sched pipe
142 Performance counter stats for 'perf bench sched pipe' (5 runs):
144 # Table of individual measurements:
152 5.483 +- 0.198 seconds time elapsed ( +- 3.62% )
156 monitor only in the container (cgroup) called "name". This option is available only
157 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
158 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
159 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
160 to first event, second cgroup to second event and so on. It is possible to provide
161 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
162 corresponding events, i.e., they always refer to events defined earlier on the command
163 line. If the user wants to track multiple events for a specific cgroup, the user can
164 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
166 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
167 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
171 Print the output into the designated file.
174 Append to the output file designated with the -o option. Ignored if -o is not specified.
178 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
179 with it. --append may be used here. Examples:
180 3>results perf stat --log-fd 3 -- $cmd
181 3>>results perf stat --log-fd 3 --append -- $cmd
183 --control fd:ctl-fd[,ack-fd]
184 Listen on ctl-fd descriptor for command to control measurement ('enable': enable events,
185 'disable': disable events). Measurements can be started with events disabled using
186 --delay=-1 option. Optionally send control command completion ('ack\n') to ack-fd descriptor
187 to synchronize with the controlling process. Example of bash shell script to enable and
188 disable events during measurements:
194 ctl_fifo=${ctl_dir}perf_ctl.fifo
195 test -p ${ctl_fifo} && unlink ${ctl_fifo}
197 exec {ctl_fd}<>${ctl_fifo}
199 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
200 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
201 mkfifo ${ctl_ack_fifo}
202 exec {ctl_fd_ack}<>${ctl_ack_fifo}
204 perf stat -D -1 -e cpu-cycles -a -I 1000 \
205 --control fd:${ctl_fd},${ctl_fd_ack} \
209 sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
210 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
213 unlink ${ctl_ack_fifo}
224 Pre and post measurement hooks, e.g.:
226 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
229 --interval-print msecs::
230 Print count deltas every N milliseconds (minimum: 1ms)
231 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals. Use with caution.
232 example: 'perf stat -I 1000 -e cycles -a sleep 5'
234 If the metric exists, it is calculated by the counts generated in this interval and the metric is printed after #.
236 --interval-count times::
237 Print count deltas for fixed number of times.
238 This option should be used together with "-I" option.
239 example: 'perf stat -I 1000 --interval-count 2 -e cycles -a'
242 Clear the screen before next interval.
245 Stop the 'perf stat' session and print count deltas after N milliseconds (minimum: 10 ms).
246 This option is not supported with the "-I" option.
247 example: 'perf stat --time 2000 -e cycles -a'
250 Only print computed metrics. Print them in a single line.
251 Don't show any raw values. Not supported with --per-thread.
254 Aggregate counts per processor socket for system-wide mode measurements. This
255 is a useful mode to detect imbalance between sockets. To enable this mode,
256 use --per-socket in addition to -a. (system-wide). The output includes the
257 socket number and the number of online processors on that socket. This is
258 useful to gauge the amount of aggregation.
261 Aggregate counts per processor die for system-wide mode measurements. This
262 is a useful mode to detect imbalance between dies. To enable this mode,
263 use --per-die in addition to -a. (system-wide). The output includes the
264 die number and the number of online processors on that die. This is
265 useful to gauge the amount of aggregation.
268 Aggregate counts per physical processor for system-wide mode measurements. This
269 is a useful mode to detect imbalance between physical cores. To enable this mode,
270 use --per-core in addition to -a. (system-wide). The output includes the
271 core number and the number of online logical processors on that physical processor.
274 Aggregate counts per monitored threads, when monitoring threads (-t option)
275 or processes (-p option).
278 Aggregate counts per NUMA nodes for system-wide mode measurements. This
279 is a useful mode to detect imbalance between NUMA nodes. To enable this
280 mode, use --per-node in addition to -a. (system-wide).
284 After starting the program, wait msecs before measuring (-1: start with events
285 disabled). This is useful to filter out the startup phase of the program,
286 which is often very different.
291 Print statistics of transactional execution if supported.
294 By default, events to compute a metric are placed in weak groups. The
295 group tries to enforce scheduling all or none of the events. The
296 --metric-no-group option places events outside of groups and may
297 increase the chance of the event being scheduled - leading to more
298 accuracy. However, as events may not be scheduled together accuracy
299 for metrics like instructions per cycle can be lower - as both metrics
300 may no longer be being measured at the same time.
303 By default metric events in different weak groups can be shared if one
304 group contains all the events needed by another. In such cases one
305 group will be eliminated reducing event multiplexing and making it so
306 that certain groups of metrics sum to 100%. A downside to sharing a
307 group is that the group may require multiplexing and so accuracy for a
308 small group that need not have multiplexing is lowered. This option
309 forbids the event merging logic from sharing events between groups and
310 may be used to increase accuracy in this case.
314 Stores stat data into perf data file.
322 Reads and reports stat data from perf data file.
329 Aggregate counts per processor socket for system-wide mode measurements.
332 Aggregate counts per processor die for system-wide mode measurements.
335 Aggregate counts per physical processor for system-wide mode measurements.
339 Print metrics or metricgroups specified in a comma separated list.
340 For a group all metrics from the group are added.
341 The events from the metrics are automatically measured.
342 See perf list output for the possble metrics and metricgroups.
346 Do not aggregate counts across all monitored CPUs.
349 Print top down level 1 metrics if supported by the CPU. This allows to
350 determine bottle necks in the CPU pipeline for CPU bound workloads,
351 by breaking the cycles consumed down into frontend bound, backend bound,
352 bad speculation and retiring.
354 Frontend bound means that the CPU cannot fetch and decode instructions fast
355 enough. Backend bound means that computation or memory access is the bottle
356 neck. Bad Speculation means that the CPU wasted cycles due to branch
357 mispredictions and similar issues. Retiring means that the CPU computed without
358 an apparently bottleneck. The bottleneck is only the real bottleneck
359 if the workload is actually bound by the CPU and not by something else.
361 For best results it is usually a good idea to use it with interval
362 mode like -I 1000, as the bottleneck of workloads can change often.
364 The top down metrics are collected per core instead of per
365 CPU thread. Per core mode is automatically enabled
366 and -a (global monitoring) is needed, requiring root rights or
367 perf.perf_event_paranoid=-1.
369 Topdown uses the full Performance Monitoring Unit, and needs
370 disabling of the NMI watchdog (as root):
371 echo 0 > /proc/sys/kernel/nmi_watchdog
372 for best results. Otherwise the bottlenecks may be inconsistent
373 on workload with changing phases.
375 This enables --metric-only, unless overridden with --no-metric-only.
377 To interpret the results it is usually needed to know on which
378 CPUs the workload runs on. If needed the CPUs can be forced using
382 Do not merge results from same PMUs.
384 When multiple events are created from a single event specification,
385 stat will, by default, aggregate the event counts and show the result
386 in a single row. This option disables that behavior and shows
387 the individual events and counts.
389 Multiple events are created from a single event specification when:
390 1. Prefix or glob matching is used for the PMU name.
391 2. Aliases, which are listed immediately after the Kernel PMU events
392 by perf list, are used.
395 Measure SMI cost if msr/aperf/ and msr/smi/ events are supported.
397 During the measurement, the /sys/device/cpu/freeze_on_smi will be set to
398 freeze core counters on SMI.
399 The aperf counter will not be effected by the setting.
400 The cost of SMI can be measured by (aperf - unhalted core cycles).
402 In practice, the percentages of SMI cycles is very useful for performance
403 oriented analysis. --metric_only will be applied by default.
404 The output is SMI cycles%, equals to (aperf - unhalted core cycles) / aperf
406 Users who wants to get the actual value can apply --no-metric-only.
409 Configure all used events to run in kernel space.
412 Configure all used events to run in user space.
414 --percore-show-thread::
415 The event modifier "percore" has supported to sum up the event counts
416 for all hardware threads in a core and show the counts per core.
418 This option with event modifier "percore" enabled also sums up the event
419 counts for all hardware threads in a core but show the sum counts per
420 hardware thread. This is essentially a replacement for the any bit and
421 convenient for post processing.
424 Print summary for interval mode (-I).
431 Performance counter stats for 'make':
433 83723.452481 task-clock:u (msec) # 1.004 CPUs utilized
434 0 context-switches:u # 0.000 K/sec
435 0 cpu-migrations:u # 0.000 K/sec
436 3,228,188 page-faults:u # 0.039 M/sec
437 229,570,665,834 cycles:u # 2.742 GHz
438 313,163,853,778 instructions:u # 1.36 insn per cycle
439 69,704,684,856 branches:u # 832.559 M/sec
440 2,078,861,393 branch-misses:u # 2.98% of all branches
442 83.409183620 seconds time elapsed
444 74.684747000 seconds user
445 8.739217000 seconds sys
449 As displayed in the example above we can display 3 types of timings.
450 We always display the time the counters were enabled/alive:
452 83.409183620 seconds time elapsed
454 For workload sessions we also display time the workloads spent in
457 74.684747000 seconds user
458 8.739217000 seconds sys
460 Those times are the very same as displayed by the 'time' tool.
465 With -x, perf stat is able to output a not-quite-CSV format output
466 Commas in the output are not put into "". To make it easy to parse
467 it is recommended to use a different character like -x \;
469 The fields are in this order:
471 - optional usec time stamp in fractions of second (with -I xxx)
472 - optional CPU, core, or socket identifier
473 - optional number of logical CPUs aggregated
475 - unit of the counter value or empty
477 - run time of counter
478 - percentage of measurement time the counter was running
479 - optional variance if multiple values are collected with -r
480 - optional metric value
481 - optional unit of metric
483 Additional metrics may be printed with all earlier fields being empty.
487 linkperf:perf-top[1], linkperf:perf-list[1]