1perf-stat(1)
2============
3
4NAME
5----
6perf-stat - Run a command and gather performance counter statistics
7
8SYNOPSIS
9--------
10[verse]
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]
15
16DESCRIPTION
17-----------
18This command runs a command and gathers performance counter statistics
19from it.
20
21
22OPTIONS
23-------
24<command>...::
25	Any command you can specify in a shell.
26
27record::
28	See STAT RECORD.
29
30report::
31	See STAT REPORT.
32
33-e::
34--event=::
35	Select the PMU event. Selection can be:
36
37	- a symbolic event name (use 'perf list' to list all events)
38
39	- a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
40	  hexadecimal event descriptor.
41
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/*
45
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/*
51
52	Note that the last two syntaxes support prefix and glob matching in
53	the PMU name to simplify creation of events across multiple instances
54	of the same type of PMU in large systems (e.g. memory controller PMUs).
55	Multiple PMU instances are typical for uncore PMUs, so the prefix
56	'uncore_' is also ignored when performing this match.
57
58
59-i::
60--no-inherit::
61        child tasks do not inherit counters
62-p::
63--pid=<pid>::
64        stat events on existing process id (comma separated list)
65
66-t::
67--tid=<tid>::
68        stat events on existing thread id (comma separated list)
69
70
71-a::
72--all-cpus::
73        system-wide collection from all CPUs (default if no target is specified)
74
75--no-scale::
76	Don't scale/normalize counter values
77
78-d::
79--detailed::
80	print more detailed statistics, can be specified up to 3 times
81
82	   -d:          detailed events, L1 and LLC data cache
83        -d -d:     more detailed events, dTLB and iTLB events
84     -d -d -d:     very detailed events, adding prefetch events
85
86-r::
87--repeat=<n>::
88	repeat command and print average + stddev (max: 100). 0 means forever.
89
90-B::
91--big-num::
92        print large numbers with thousands' separators according to locale
93
94-C::
95--cpu=::
96Count only on the list of CPUs provided. Multiple CPUs can be provided as a
97comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
98In per-thread mode, this option is ignored. The -a option is still necessary
99to activate system-wide monitoring. Default is to count on all CPUs.
100
101-A::
102--no-aggr::
103Do not aggregate counts across all monitored CPUs.
104
105-n::
106--null::
107        null run - don't start any counters
108
109-v::
110--verbose::
111        be more verbose (show counter open errors, etc)
112
113-x SEP::
114--field-separator SEP::
115print counts using a CSV-style output to make it easy to import directly into
116spreadsheets. Columns are separated by the string specified in SEP.
117
118--table:: Display time for each run (-r option), in a table format, e.g.:
119
120  $ perf stat --null -r 5 --table perf bench sched pipe
121
122   Performance counter stats for 'perf bench sched pipe' (5 runs):
123
124             # Table of individual measurements:
125             5.189 (-0.293) #
126             5.189 (-0.294) #
127             5.186 (-0.296) #
128             5.663 (+0.181) ##
129             6.186 (+0.703) ####
130
131             # Final result:
132             5.483 +- 0.198 seconds time elapsed  ( +-  3.62% )
133
134-G name::
135--cgroup name::
136monitor only in the container (cgroup) called "name". This option is available only
137in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
138container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
139can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
140to first event, second cgroup to second event and so on. It is possible to provide
141an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
142corresponding events, i.e., they always refer to events defined earlier on the command
143line. If the user wants to track multiple events for a specific cgroup, the user can
144use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
145
146If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
147command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
148
149-o file::
150--output file::
151Print the output into the designated file.
152
153--append::
154Append to the output file designated with the -o option. Ignored if -o is not specified.
155
156--log-fd::
157
158Log output to fd, instead of stderr.  Complementary to --output, and mutually exclusive
159with it.  --append may be used here.  Examples:
160     3>results  perf stat --log-fd 3          -- $cmd
161     3>>results perf stat --log-fd 3 --append -- $cmd
162
163--pre::
164--post::
165	Pre and post measurement hooks, e.g.:
166
167perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
168
169-I msecs::
170--interval-print msecs::
171Print count deltas every N milliseconds (minimum: 1ms)
172The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals.  Use with caution.
173	example: 'perf stat -I 1000 -e cycles -a sleep 5'
174
175--interval-count times::
176Print count deltas for fixed number of times.
177This option should be used together with "-I" option.
178	example: 'perf stat -I 1000 --interval-count 2 -e cycles -a'
179
180--interval-clear::
181Clear the screen before next interval.
182
183--timeout msecs::
184Stop the 'perf stat' session and print count deltas after N milliseconds (minimum: 10 ms).
185This option is not supported with the "-I" option.
186	example: 'perf stat --time 2000 -e cycles -a'
187
188--metric-only::
189Only print computed metrics. Print them in a single line.
190Don't show any raw values. Not supported with --per-thread.
191
192--per-socket::
193Aggregate counts per processor socket for system-wide mode measurements.  This
194is a useful mode to detect imbalance between sockets.  To enable this mode,
195use --per-socket in addition to -a. (system-wide).  The output includes the
196socket number and the number of online processors on that socket. This is
197useful to gauge the amount of aggregation.
198
199--per-core::
200Aggregate counts per physical processor for system-wide mode measurements.  This
201is a useful mode to detect imbalance between physical cores.  To enable this mode,
202use --per-core in addition to -a. (system-wide).  The output includes the
203core number and the number of online logical processors on that physical processor.
204
205--per-thread::
206Aggregate counts per monitored threads, when monitoring threads (-t option)
207or processes (-p option).
208
209-D msecs::
210--delay msecs::
211After starting the program, wait msecs before measuring. This is useful to
212filter out the startup phase of the program, which is often very different.
213
214-T::
215--transaction::
216
217Print statistics of transactional execution if supported.
218
219STAT RECORD
220-----------
221Stores stat data into perf data file.
222
223-o file::
224--output file::
225Output file name.
226
227STAT REPORT
228-----------
229Reads and reports stat data from perf data file.
230
231-i file::
232--input file::
233Input file name.
234
235--per-socket::
236Aggregate counts per processor socket for system-wide mode measurements.
237
238--per-core::
239Aggregate counts per physical processor for system-wide mode measurements.
240
241-M::
242--metrics::
243Print metrics or metricgroups specified in a comma separated list.
244For a group all metrics from the group are added.
245The events from the metrics are automatically measured.
246See perf list output for the possble metrics and metricgroups.
247
248-A::
249--no-aggr::
250Do not aggregate counts across all monitored CPUs.
251
252--topdown::
253Print top down level 1 metrics if supported by the CPU. This allows to
254determine bottle necks in the CPU pipeline for CPU bound workloads,
255by breaking the cycles consumed down into frontend bound, backend bound,
256bad speculation and retiring.
257
258Frontend bound means that the CPU cannot fetch and decode instructions fast
259enough. Backend bound means that computation or memory access is the bottle
260neck. Bad Speculation means that the CPU wasted cycles due to branch
261mispredictions and similar issues. Retiring means that the CPU computed without
262an apparently bottleneck. The bottleneck is only the real bottleneck
263if the workload is actually bound by the CPU and not by something else.
264
265For best results it is usually a good idea to use it with interval
266mode like -I 1000, as the bottleneck of workloads can change often.
267
268The top down metrics are collected per core instead of per
269CPU thread. Per core mode is automatically enabled
270and -a (global monitoring) is needed, requiring root rights or
271perf.perf_event_paranoid=-1.
272
273Topdown uses the full Performance Monitoring Unit, and needs
274disabling of the NMI watchdog (as root):
275echo 0 > /proc/sys/kernel/nmi_watchdog
276for best results. Otherwise the bottlenecks may be inconsistent
277on workload with changing phases.
278
279This enables --metric-only, unless overridden with --no-metric-only.
280
281To interpret the results it is usually needed to know on which
282CPUs the workload runs on. If needed the CPUs can be forced using
283taskset.
284
285--no-merge::
286Do not merge results from same PMUs.
287
288When multiple events are created from a single event specification,
289stat will, by default, aggregate the event counts and show the result
290in a single row. This option disables that behavior and shows
291the individual events and counts.
292
293Multiple events are created from a single event specification when:
2941. Prefix or glob matching is used for the PMU name.
2952. Aliases, which are listed immediately after the Kernel PMU events
296   by perf list, are used.
297
298--smi-cost::
299Measure SMI cost if msr/aperf/ and msr/smi/ events are supported.
300
301During the measurement, the /sys/device/cpu/freeze_on_smi will be set to
302freeze core counters on SMI.
303The aperf counter will not be effected by the setting.
304The cost of SMI can be measured by (aperf - unhalted core cycles).
305
306In practice, the percentages of SMI cycles is very useful for performance
307oriented analysis. --metric_only will be applied by default.
308The output is SMI cycles%, equals to (aperf - unhalted core cycles) / aperf
309
310Users who wants to get the actual value can apply --no-metric-only.
311
312EXAMPLES
313--------
314
315$ perf stat -- make
316
317   Performance counter stats for 'make':
318
319        83723.452481      task-clock:u (msec)       #    1.004 CPUs utilized
320                   0      context-switches:u        #    0.000 K/sec
321                   0      cpu-migrations:u          #    0.000 K/sec
322           3,228,188      page-faults:u             #    0.039 M/sec
323     229,570,665,834      cycles:u                  #    2.742 GHz
324     313,163,853,778      instructions:u            #    1.36  insn per cycle
325      69,704,684,856      branches:u                #  832.559 M/sec
326       2,078,861,393      branch-misses:u           #    2.98% of all branches
327
328        83.409183620 seconds time elapsed
329
330        74.684747000 seconds user
331         8.739217000 seconds sys
332
333TIMINGS
334-------
335As displayed in the example above we can display 3 types of timings.
336We always display the time the counters were enabled/alive:
337
338        83.409183620 seconds time elapsed
339
340For workload sessions we also display time the workloads spent in
341user/system lands:
342
343        74.684747000 seconds user
344         8.739217000 seconds sys
345
346Those times are the very same as displayed by the 'time' tool.
347
348CSV FORMAT
349----------
350
351With -x, perf stat is able to output a not-quite-CSV format output
352Commas in the output are not put into "". To make it easy to parse
353it is recommended to use a different character like -x \;
354
355The fields are in this order:
356
357	- optional usec time stamp in fractions of second (with -I xxx)
358	- optional CPU, core, or socket identifier
359	- optional number of logical CPUs aggregated
360	- counter value
361	- unit of the counter value or empty
362	- event name
363	- run time of counter
364	- percentage of measurement time the counter was running
365	- optional variance if multiple values are collected with -r
366	- optional metric value
367	- optional unit of metric
368
369Additional metrics may be printed with all earlier fields being empty.
370
371SEE ALSO
372--------
373linkperf:perf-top[1], linkperf:perf-list[1]
374