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 accross 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-c::
76--scale::
77	scale/normalize counter values
78
79-d::
80--detailed::
81	print more detailed statistics, can be specified up to 3 times
82
83	   -d:          detailed events, L1 and LLC data cache
84        -d -d:     more detailed events, dTLB and iTLB events
85     -d -d -d:     very detailed events, adding prefetch events
86
87-r::
88--repeat=<n>::
89	repeat command and print average + stddev (max: 100). 0 means forever.
90
91-B::
92--big-num::
93        print large numbers with thousands' separators according to locale
94
95-C::
96--cpu=::
97Count only on the list of CPUs provided. Multiple CPUs can be provided as a
98comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
99In per-thread mode, this option is ignored. The -a option is still necessary
100to activate system-wide monitoring. Default is to count on all CPUs.
101
102-A::
103--no-aggr::
104Do not aggregate counts across all monitored CPUs.
105
106-n::
107--null::
108        null run - don't start any counters
109
110-v::
111--verbose::
112        be more verbose (show counter open errors, etc)
113
114-x SEP::
115--field-separator SEP::
116print counts using a CSV-style output to make it easy to import directly into
117spreadsheets. Columns are separated by the string specified in SEP.
118
119--table:: Display time for each run (-r option), in a table format, e.g.:
120
121  $ perf stat --null -r 5 --table perf bench sched pipe
122
123   Performance counter stats for 'perf bench sched pipe' (5 runs):
124
125             # Table of individual measurements:
126             5.189 (-0.293) #
127             5.189 (-0.294) #
128             5.186 (-0.296) #
129             5.663 (+0.181) ##
130             6.186 (+0.703) ####
131
132             # Final result:
133             5.483 +- 0.198 seconds time elapsed  ( +-  3.62% )
134
135-G name::
136--cgroup name::
137monitor only in the container (cgroup) called "name". This option is available only
138in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
139container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
140can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
141to first event, second cgroup to second event and so on. It is possible to provide
142an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
143corresponding events, i.e., they always refer to events defined earlier on the command
144line. If the user wants to track multiple events for a specific cgroup, the user can
145use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
146
147If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
148command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
149
150-o file::
151--output file::
152Print the output into the designated file.
153
154--append::
155Append to the output file designated with the -o option. Ignored if -o is not specified.
156
157--log-fd::
158
159Log output to fd, instead of stderr.  Complementary to --output, and mutually exclusive
160with it.  --append may be used here.  Examples:
161     3>results  perf stat --log-fd 3          -- $cmd
162     3>>results perf stat --log-fd 3 --append -- $cmd
163
164--pre::
165--post::
166	Pre and post measurement hooks, e.g.:
167
168perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
169
170-I msecs::
171--interval-print msecs::
172Print count deltas every N milliseconds (minimum: 1ms)
173The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals.  Use with caution.
174	example: 'perf stat -I 1000 -e cycles -a sleep 5'
175
176--interval-count times::
177Print count deltas for fixed number of times.
178This option should be used together with "-I" option.
179	example: 'perf stat -I 1000 --interval-count 2 -e cycles -a'
180
181--timeout msecs::
182Stop the 'perf stat' session and print count deltas after N milliseconds (minimum: 10 ms).
183This option is not supported with the "-I" option.
184	example: 'perf stat --time 2000 -e cycles -a'
185
186--metric-only::
187Only print computed metrics. Print them in a single line.
188Don't show any raw values. Not supported with --per-thread.
189
190--per-socket::
191Aggregate counts per processor socket for system-wide mode measurements.  This
192is a useful mode to detect imbalance between sockets.  To enable this mode,
193use --per-socket in addition to -a. (system-wide).  The output includes the
194socket number and the number of online processors on that socket. This is
195useful to gauge the amount of aggregation.
196
197--per-core::
198Aggregate counts per physical processor for system-wide mode measurements.  This
199is a useful mode to detect imbalance between physical cores.  To enable this mode,
200use --per-core in addition to -a. (system-wide).  The output includes the
201core number and the number of online logical processors on that physical processor.
202
203--per-thread::
204Aggregate counts per monitored threads, when monitoring threads (-t option)
205or processes (-p option).
206
207-D msecs::
208--delay msecs::
209After starting the program, wait msecs before measuring. This is useful to
210filter out the startup phase of the program, which is often very different.
211
212-T::
213--transaction::
214
215Print statistics of transactional execution if supported.
216
217STAT RECORD
218-----------
219Stores stat data into perf data file.
220
221-o file::
222--output file::
223Output file name.
224
225STAT REPORT
226-----------
227Reads and reports stat data from perf data file.
228
229-i file::
230--input file::
231Input file name.
232
233--per-socket::
234Aggregate counts per processor socket for system-wide mode measurements.
235
236--per-core::
237Aggregate counts per physical processor for system-wide mode measurements.
238
239-M::
240--metrics::
241Print metrics or metricgroups specified in a comma separated list.
242For a group all metrics from the group are added.
243The events from the metrics are automatically measured.
244See perf list output for the possble metrics and metricgroups.
245
246-A::
247--no-aggr::
248Do not aggregate counts across all monitored CPUs.
249
250--topdown::
251Print top down level 1 metrics if supported by the CPU. This allows to
252determine bottle necks in the CPU pipeline for CPU bound workloads,
253by breaking the cycles consumed down into frontend bound, backend bound,
254bad speculation and retiring.
255
256Frontend bound means that the CPU cannot fetch and decode instructions fast
257enough. Backend bound means that computation or memory access is the bottle
258neck. Bad Speculation means that the CPU wasted cycles due to branch
259mispredictions and similar issues. Retiring means that the CPU computed without
260an apparently bottleneck. The bottleneck is only the real bottleneck
261if the workload is actually bound by the CPU and not by something else.
262
263For best results it is usually a good idea to use it with interval
264mode like -I 1000, as the bottleneck of workloads can change often.
265
266The top down metrics are collected per core instead of per
267CPU thread. Per core mode is automatically enabled
268and -a (global monitoring) is needed, requiring root rights or
269perf.perf_event_paranoid=-1.
270
271Topdown uses the full Performance Monitoring Unit, and needs
272disabling of the NMI watchdog (as root):
273echo 0 > /proc/sys/kernel/nmi_watchdog
274for best results. Otherwise the bottlenecks may be inconsistent
275on workload with changing phases.
276
277This enables --metric-only, unless overriden with --no-metric-only.
278
279To interpret the results it is usually needed to know on which
280CPUs the workload runs on. If needed the CPUs can be forced using
281taskset.
282
283--no-merge::
284Do not merge results from same PMUs.
285
286When multiple events are created from a single event specification,
287stat will, by default, aggregate the event counts and show the result
288in a single row. This option disables that behavior and shows
289the individual events and counts.
290
291Multiple events are created from a single event specification when:
2921. Prefix or glob matching is used for the PMU name.
2932. Aliases, which are listed immediately after the Kernel PMU events
294   by perf list, are used.
295
296--smi-cost::
297Measure SMI cost if msr/aperf/ and msr/smi/ events are supported.
298
299During the measurement, the /sys/device/cpu/freeze_on_smi will be set to
300freeze core counters on SMI.
301The aperf counter will not be effected by the setting.
302The cost of SMI can be measured by (aperf - unhalted core cycles).
303
304In practice, the percentages of SMI cycles is very useful for performance
305oriented analysis. --metric_only will be applied by default.
306The output is SMI cycles%, equals to (aperf - unhalted core cycles) / aperf
307
308Users who wants to get the actual value can apply --no-metric-only.
309
310EXAMPLES
311--------
312
313$ perf stat -- make
314
315   Performance counter stats for 'make':
316
317        83723.452481      task-clock:u (msec)       #    1.004 CPUs utilized
318                   0      context-switches:u        #    0.000 K/sec
319                   0      cpu-migrations:u          #    0.000 K/sec
320           3,228,188      page-faults:u             #    0.039 M/sec
321     229,570,665,834      cycles:u                  #    2.742 GHz
322     313,163,853,778      instructions:u            #    1.36  insn per cycle
323      69,704,684,856      branches:u                #  832.559 M/sec
324       2,078,861,393      branch-misses:u           #    2.98% of all branches
325
326        83.409183620 seconds time elapsed
327
328        74.684747000 seconds user
329         8.739217000 seconds sys
330
331TIMINGS
332-------
333As displayed in the example above we can display 3 types of timings.
334We always display the time the counters were enabled/alive:
335
336        83.409183620 seconds time elapsed
337
338For workload sessions we also display time the workloads spent in
339user/system lands:
340
341        74.684747000 seconds user
342         8.739217000 seconds sys
343
344Those times are the very same as displayed by the 'time' tool.
345
346CSV FORMAT
347----------
348
349With -x, perf stat is able to output a not-quite-CSV format output
350Commas in the output are not put into "". To make it easy to parse
351it is recommended to use a different character like -x \;
352
353The fields are in this order:
354
355	- optional usec time stamp in fractions of second (with -I xxx)
356	- optional CPU, core, or socket identifier
357	- optional number of logical CPUs aggregated
358	- counter value
359	- unit of the counter value or empty
360	- event name
361	- run time of counter
362	- percentage of measurement time the counter was running
363	- optional variance if multiple values are collected with -r
364	- optional metric value
365	- optional unit of metric
366
367Additional metrics may be printed with all earlier fields being empty.
368
369SEE ALSO
370--------
371linkperf:perf-top[1], linkperf:perf-list[1]
372