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_sources/devices/<pmu>/format/*
50 - a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
51 where M, N, K are numbers (in decimal, hex, octal format).
52 Acceptable values for each of 'config', 'config1' and 'config2'
53 parameters are defined by corresponding entries in
54 /sys/bus/event_sources/devices/<pmu>/format/*
58 child tasks do not inherit counters
61 stat events on existing process id (comma separated list)
65 stat events on existing thread id (comma separated list)
70 system-wide collection from all CPUs
74 scale/normalize counter values
78 print more detailed statistics, can be specified up to 3 times
80 -d: detailed events, L1 and LLC data cache
81 -d -d: more detailed events, dTLB and iTLB events
82 -d -d -d: very detailed events, adding prefetch events
86 repeat command and print average + stddev (max: 100). 0 means forever.
90 print large numbers with thousands' separators according to locale
94 Count only on the list of CPUs provided. Multiple CPUs can be provided as a
95 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
96 In per-thread mode, this option is ignored. The -a option is still necessary
97 to activate system-wide monitoring. Default is to count on all CPUs.
101 Do not aggregate counts across all monitored CPUs in system-wide mode (-a).
102 This option is only valid in system-wide mode.
106 null run - don't start any counters
110 be more verbose (show counter open errors, etc)
113 --field-separator SEP::
114 print counts using a CSV-style output to make it easy to import directly into
115 spreadsheets. Columns are separated by the string specified in SEP.
119 monitor only in the container (cgroup) called "name". This option is available only
120 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
121 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
122 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
123 to first event, second cgroup to second event and so on. It is possible to provide
124 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
125 corresponding events, i.e., they always refer to events defined earlier on the command
130 Print the output into the designated file.
133 Append to the output file designated with the -o option. Ignored if -o is not specified.
137 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
138 with it. --append may be used here. Examples:
139 3>results perf stat --log-fd 3 -- $cmd
140 3>>results perf stat --log-fd 3 --append -- $cmd
144 Pre and post measurement hooks, e.g.:
146 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
149 --interval-print msecs::
150 Print count deltas every N milliseconds (minimum: 10ms)
151 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals. Use with caution.
152 example: 'perf stat -I 1000 -e cycles -a sleep 5'
155 Only print computed metrics. Print them in a single line.
156 Don't show any raw values. Not supported with --per-thread.
159 Aggregate counts per processor socket for system-wide mode measurements. This
160 is a useful mode to detect imbalance between sockets. To enable this mode,
161 use --per-socket in addition to -a. (system-wide). The output includes the
162 socket number and the number of online processors on that socket. This is
163 useful to gauge the amount of aggregation.
166 Aggregate counts per physical processor for system-wide mode measurements. This
167 is a useful mode to detect imbalance between physical cores. To enable this mode,
168 use --per-core in addition to -a. (system-wide). The output includes the
169 core number and the number of online logical processors on that physical processor.
172 Aggregate counts per monitored threads, when monitoring threads (-t option)
173 or processes (-p option).
177 After starting the program, wait msecs before measuring. This is useful to
178 filter out the startup phase of the program, which is often very different.
183 Print statistics of transactional execution if supported.
187 Stores stat data into perf data file.
195 Reads and reports stat data from perf data file.
202 Aggregate counts per processor socket for system-wide mode measurements.
205 Aggregate counts per physical processor for system-wide mode measurements.
209 Do not aggregate counts across all monitored CPUs.
212 Print top down level 1 metrics if supported by the CPU. This allows to
213 determine bottle necks in the CPU pipeline for CPU bound workloads,
214 by breaking the cycles consumed down into frontend bound, backend bound,
215 bad speculation and retiring.
217 Frontend bound means that the CPU cannot fetch and decode instructions fast
218 enough. Backend bound means that computation or memory access is the bottle
219 neck. Bad Speculation means that the CPU wasted cycles due to branch
220 mispredictions and similar issues. Retiring means that the CPU computed without
221 an apparently bottleneck. The bottleneck is only the real bottleneck
222 if the workload is actually bound by the CPU and not by something else.
224 For best results it is usually a good idea to use it with interval
225 mode like -I 1000, as the bottleneck of workloads can change often.
227 The top down metrics are collected per core instead of per
228 CPU thread. Per core mode is automatically enabled
229 and -a (global monitoring) is needed, requiring root rights or
230 perf.perf_event_paranoid=-1.
232 Topdown uses the full Performance Monitoring Unit, and needs
233 disabling of the NMI watchdog (as root):
234 echo 0 > /proc/sys/kernel/nmi_watchdog
235 for best results. Otherwise the bottlenecks may be inconsistent
236 on workload with changing phases.
238 This enables --metric-only, unless overriden with --no-metric-only.
240 To interpret the results it is usually needed to know on which
241 CPUs the workload runs on. If needed the CPUs can be forced using
247 $ perf stat -- make -j
249 Performance counter stats for 'make -j':
251 8117.370256 task clock ticks # 11.281 CPU utilization factor
252 678 context switches # 0.000 M/sec
253 133 CPU migrations # 0.000 M/sec
254 235724 pagefaults # 0.029 M/sec
255 24821162526 CPU cycles # 3057.784 M/sec
256 18687303457 instructions # 2302.138 M/sec
257 172158895 cache references # 21.209 M/sec
258 27075259 cache misses # 3.335 M/sec
260 Wall-clock time elapsed: 719.554352 msecs
265 With -x, perf stat is able to output a not-quite-CSV format output
266 Commas in the output are not put into "". To make it easy to parse
267 it is recommended to use a different character like -x \;
269 The fields are in this order:
271 - optional usec time stamp in fractions of second (with -I xxx)
272 - optional CPU, core, or socket identifier
273 - optional number of logical CPUs aggregated
275 - unit of the counter value or empty
277 - run time of counter
278 - percentage of measurement time the counter was running
279 - optional variance if multiple values are collected with -r
280 - optional metric value
281 - optional unit of metric
283 Additional metrics may be printed with all earlier fields being empty.
287 linkperf:perf-top[1], linkperf:perf-list[1]