1perf-record(1)
2==============
3
4NAME
5----
6perf-record - Run a command and record its profile into perf.data
7
8SYNOPSIS
9--------
10[verse]
11'perf record' [-e <EVENT> | --event=EVENT] [-l] [-a] <command>
12'perf record' [-e <EVENT> | --event=EVENT] [-l] [-a] -- <command> [<options>]
13
14DESCRIPTION
15-----------
16This command runs a command and gathers a performance counter profile
17from it, into perf.data - without displaying anything.
18
19This file can then be inspected later on, using 'perf report'.
20
21
22OPTIONS
23-------
24<command>...::
25	Any command you can specify in a shell.
26
27-e::
28--event=::
29	Select the PMU event. Selection can be:
30
31        - a symbolic event name	(use 'perf list' to list all events)
32
33        - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
34	  hexadecimal event descriptor.
35
36	- a symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
37	  'param1', 'param2', etc are defined as formats for the PMU in
38	  /sys/bus/event_sources/devices/<pmu>/format/*.
39
40	- a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
41
42          where M, N, K are numbers (in decimal, hex, octal format). Acceptable
43          values for each of 'config', 'config1' and 'config2' are defined by
44          corresponding entries in /sys/bus/event_sources/devices/<pmu>/format/*
45          param1 and param2 are defined as formats for the PMU in:
46          /sys/bus/event_sources/devices/<pmu>/format/*
47
48	  There are also some params which are not defined in .../<pmu>/format/*.
49	  These params can be used to set event defaults.
50	  Here is a list of the params.
51	  - 'period': Set event sampling period
52
53	  Note: If user explicitly sets options which conflict with the params,
54	  the value set by the params will be overridden.
55
56        - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
57          where addr is the address in memory you want to break in.
58          Access is the memory access type (read, write, execute) it can
59          be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
60          number of bytes from specified addr, which the breakpoint will cover.
61          If you want to profile read-write accesses in 0x1000, just set
62          'mem:0x1000:rw'.
63          If you want to profile write accesses in [0x1000~1008), just set
64          'mem:0x1000/8:w'.
65
66	- a group of events surrounded by a pair of brace ("{event1,event2,...}").
67	  Each event is separated by commas and the group should be quoted to
68	  prevent the shell interpretation.  You also need to use --group on
69	  "perf report" to view group events together.
70
71--filter=<filter>::
72        Event filter.
73
74-a::
75--all-cpus::
76        System-wide collection from all CPUs.
77
78-p::
79--pid=::
80	Record events on existing process ID (comma separated list).
81
82-t::
83--tid=::
84        Record events on existing thread ID (comma separated list).
85        This option also disables inheritance by default.  Enable it by adding
86        --inherit.
87
88-u::
89--uid=::
90        Record events in threads owned by uid. Name or number.
91
92-r::
93--realtime=::
94	Collect data with this RT SCHED_FIFO priority.
95
96--no-buffering::
97	Collect data without buffering.
98
99-c::
100--count=::
101	Event period to sample.
102
103-o::
104--output=::
105	Output file name.
106
107-i::
108--no-inherit::
109	Child tasks do not inherit counters.
110-F::
111--freq=::
112	Profile at this frequency.
113
114-m::
115--mmap-pages=::
116	Number of mmap data pages (must be a power of two) or size
117	specification with appended unit character - B/K/M/G. The
118	size is rounded up to have nearest pages power of two value.
119	Also, by adding a comma, the number of mmap pages for AUX
120	area tracing can be specified.
121
122--group::
123	Put all events in a single event group.  This precedes the --event
124	option and remains only for backward compatibility.  See --event.
125
126-g::
127	Enables call-graph (stack chain/backtrace) recording.
128
129--call-graph::
130	Setup and enable call-graph (stack chain/backtrace) recording,
131	implies -g.
132
133	Allows specifying "fp" (frame pointer) or "dwarf"
134	(DWARF's CFI - Call Frame Information) or "lbr"
135	(Hardware Last Branch Record facility) as the method to collect
136	the information used to show the call graphs.
137
138	In some systems, where binaries are build with gcc
139	--fomit-frame-pointer, using the "fp" method will produce bogus
140	call graphs, using "dwarf", if available (perf tools linked to
141	the libunwind library) should be used instead.
142	Using the "lbr" method doesn't require any compiler options. It
143	will produce call graphs from the hardware LBR registers. The
144	main limition is that it is only available on new Intel
145	platforms, such as Haswell. It can only get user call chain. It
146	doesn't work with branch stack sampling at the same time.
147
148-q::
149--quiet::
150	Don't print any message, useful for scripting.
151
152-v::
153--verbose::
154	Be more verbose (show counter open errors, etc).
155
156-s::
157--stat::
158	Record per-thread event counts.  Use it with 'perf report -T' to see
159	the values.
160
161-d::
162--data::
163	Record the sample addresses.
164
165-T::
166--timestamp::
167	Record the sample timestamps. Use it with 'perf report -D' to see the
168	timestamps, for instance.
169
170-P::
171--period::
172	Record the sample period.
173
174-n::
175--no-samples::
176	Don't sample.
177
178-R::
179--raw-samples::
180Collect raw sample records from all opened counters (default for tracepoint counters).
181
182-C::
183--cpu::
184Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
185comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
186In per-thread mode with inheritance mode on (default), samples are captured only when
187the thread executes on the designated CPUs. Default is to monitor all CPUs.
188
189-N::
190--no-buildid-cache::
191Do not update the buildid cache. This saves some overhead in situations
192where the information in the perf.data file (which includes buildids)
193is sufficient.
194
195-G name,...::
196--cgroup name,...::
197monitor only in the container (cgroup) called "name". This option is available only
198in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
199container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
200can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
201to first event, second cgroup to second event and so on. It is possible to provide
202an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
203corresponding events, i.e., they always refer to events defined earlier on the command
204line.
205
206-b::
207--branch-any::
208Enable taken branch stack sampling. Any type of taken branch may be sampled.
209This is a shortcut for --branch-filter any. See --branch-filter for more infos.
210
211-j::
212--branch-filter::
213Enable taken branch stack sampling. Each sample captures a series of consecutive
214taken branches. The number of branches captured with each sample depends on the
215underlying hardware, the type of branches of interest, and the executed code.
216It is possible to select the types of branches captured by enabling filters. The
217following filters are defined:
218
219        - any:  any type of branches
220        - any_call: any function call or system call
221        - any_ret: any function return or system call return
222        - ind_call: any indirect branch
223        - u:  only when the branch target is at the user level
224        - k: only when the branch target is in the kernel
225        - hv: only when the target is at the hypervisor level
226	- in_tx: only when the target is in a hardware transaction
227	- no_tx: only when the target is not in a hardware transaction
228	- abort_tx: only when the target is a hardware transaction abort
229	- cond: conditional branches
230
231+
232The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
233The privilege levels may be omitted, in which case, the privilege levels of the associated
234event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
235levels are subject to permissions.  When sampling on multiple events, branch stack sampling
236is enabled for all the sampling events. The sampled branch type is the same for all events.
237The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
238Note that this feature may not be available on all processors.
239
240--weight::
241Enable weightened sampling. An additional weight is recorded per sample and can be
242displayed with the weight and local_weight sort keys.  This currently works for TSX
243abort events and some memory events in precise mode on modern Intel CPUs.
244
245--transaction::
246Record transaction flags for transaction related events.
247
248--per-thread::
249Use per-thread mmaps.  By default per-cpu mmaps are created.  This option
250overrides that and uses per-thread mmaps.  A side-effect of that is that
251inheritance is automatically disabled.  --per-thread is ignored with a warning
252if combined with -a or -C options.
253
254-D::
255--delay=::
256After starting the program, wait msecs before measuring. This is useful to
257filter out the startup phase of the program, which is often very different.
258
259-I::
260--intr-regs::
261Capture machine state (registers) at interrupt, i.e., on counter overflows for
262each sample. List of captured registers depends on the architecture. This option
263is off by default.
264
265--running-time::
266Record running and enabled time for read events (:S)
267
268-k::
269--clockid::
270Sets the clock id to use for the various time fields in the perf_event_type
271records. See clock_gettime(). In particular CLOCK_MONOTONIC and
272CLOCK_MONOTONIC_RAW are supported, some events might also allow
273CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
274
275-S::
276--snapshot::
277Select AUX area tracing Snapshot Mode. This option is valid only with an
278AUX area tracing event. Optionally the number of bytes to capture per
279snapshot can be specified. In Snapshot Mode, trace data is captured only when
280signal SIGUSR2 is received.
281
282--proc-map-timeout::
283When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
284because the file may be huge. A time out is needed in such cases.
285This option sets the time out limit. The default value is 500 ms.
286
287SEE ALSO
288--------
289linkperf:perf-stat[1], linkperf:perf-list[1]
290