xref: /linux/tools/perf/Documentation/perf-record.txt (revision ab475966455ce285c2c9978a3e3bfe97d75ff8d4)
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] [-a] <command>
12'perf record' [-e <EVENT> | --event=EVENT] [-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 in the form of rN where N is a hexadecimal value
34          that represents the raw register encoding with the layout of the
35          event control registers as described by entries in
36          /sys/bus/event_source/devices/cpu/format/*.
37
38        - a symbolic or raw PMU event followed by an optional colon
39	  and a list of event modifiers, e.g., cpu-cycles:p.  See the
40	  linkperf:perf-list[1] man page for details on event modifiers.
41
42	- a symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
43	  'param1', 'param2', etc 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,config3=K/'
47
48          where M, N, K are numbers (in decimal, hex, octal format). Acceptable
49          values for each of 'config', 'config1' and 'config2' are defined by
50          corresponding entries in /sys/bus/event_source/devices/<pmu>/format/*
51          param1 and param2 are defined as formats for the PMU in:
52          /sys/bus/event_source/devices/<pmu>/format/*
53
54	  There are also some parameters which are not defined in .../<pmu>/format/*.
55	  These params can be used to overload default config values per event.
56	  Here are some common parameters:
57	  - 'period': Set event sampling period
58	  - 'freq': Set event sampling frequency
59	  - 'time': Disable/enable time stamping. Acceptable values are 1 for
60		    enabling time stamping. 0 for disabling time stamping.
61		    The default is 1.
62	  - 'call-graph': Disable/enable callgraph. Acceptable str are "fp" for
63			 FP mode, "dwarf" for DWARF mode, "lbr" for LBR mode and
64			 "no" for disable callgraph.
65	  - 'stack-size': user stack size for dwarf mode
66	  - 'name' : User defined event name. Single quotes (') may be used to
67		    escape symbols in the name from parsing by shell and tool
68		    like this: name=\'CPU_CLK_UNHALTED.THREAD:cmask=0x1\'.
69	  - 'aux-output': Generate AUX records instead of events. This requires
70			  that an AUX area event is also provided.
71	  - 'aux-sample-size': Set sample size for AUX area sampling. If the
72	  '--aux-sample' option has been used, set aux-sample-size=0 to disable
73	  AUX area sampling for the event.
74
75          See the linkperf:perf-list[1] man page for more parameters.
76
77	  Note: If user explicitly sets options which conflict with the params,
78	  the value set by the parameters will be overridden.
79
80	  Also not defined in .../<pmu>/format/* are PMU driver specific
81	  configuration parameters.  Any configuration parameter preceded by
82	  the letter '@' is not interpreted in user space and sent down directly
83	  to the PMU driver.  For example:
84
85	  perf record -e some_event/@cfg1,@cfg2=config/ ...
86
87	  will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
88	  with the event for further processing.  There is no restriction on
89	  what the configuration parameters are, as long as their semantic is
90	  understood and supported by the PMU driver.
91
92        - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
93          where addr is the address in memory you want to break in.
94          Access is the memory access type (read, write, execute) it can
95          be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
96          number of bytes from specified addr, which the breakpoint will cover.
97          If you want to profile read-write accesses in 0x1000, just set
98          'mem:0x1000:rw'.
99          If you want to profile write accesses in [0x1000~1008), just set
100          'mem:0x1000/8:w'.
101
102	- a group of events surrounded by a pair of brace ("{event1,event2,...}").
103	  Each event is separated by commas and the group should be quoted to
104	  prevent the shell interpretation.  You also need to use --group on
105	  "perf report" to view group events together.
106
107--filter=<filter>::
108	Event filter.  This option should follow an event selector (-e).
109	If the event is a tracepoint, the filter string will be parsed by
110	the kernel.  If the event is a hardware trace PMU (e.g. Intel PT
111	or CoreSight), it'll be processed as an address filter.  Otherwise
112	it means a general filter using BPF which can be applied for any
113	kind of event.
114
115	- tracepoint filters
116
117	In the case of tracepoints, multiple '--filter' options are combined
118	using '&&'.
119
120	- address filters
121
122	A hardware trace PMU advertises its ability to accept a number of
123	address filters	by specifying a non-zero value in
124	/sys/bus/event_source/devices/<pmu>/nr_addr_filters.
125
126	Address filters have the format:
127
128	filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
129
130	Where:
131	- 'filter': defines a region that will be traced.
132	- 'start': defines an address at which tracing will begin.
133	- 'stop': defines an address at which tracing will stop.
134	- 'tracestop': defines a region in which tracing will stop.
135
136	<file name> is the name of the object file, <start> is the offset to the
137	code to trace in that file, and <size> is the size of the region to
138	trace. 'start' and 'stop' filters need not specify a <size>.
139
140	If no object file is specified then the kernel is assumed, in which case
141	the start address must be a current kernel memory address.
142
143	<start> can also be specified by providing the name of a symbol. If the
144	symbol name is not unique, it can be disambiguated by inserting #n where
145	'n' selects the n'th symbol in address order. Alternately #0, #g or #G
146	select only a global symbol. <size> can also be specified by providing
147	the name of a symbol, in which case the size is calculated to the end
148	of that symbol. For 'filter' and 'tracestop' filters, if <size> is
149	omitted and <start> is a symbol, then the size is calculated to the end
150	of that symbol.
151
152	If <size> is omitted and <start> is '*', then the start and size will
153	be calculated from the first and last symbols, i.e. to trace the whole
154	file.
155
156	If symbol names (or '*') are provided, they must be surrounded by white
157	space.
158
159	The filter passed to the kernel is not necessarily the same as entered.
160	To see the filter that is passed, use the -v option.
161
162	The kernel may not be able to configure a trace region if it is not
163	within a single mapping.  MMAP events (or /proc/<pid>/maps) can be
164	examined to determine if that is a possibility.
165
166	Multiple filters can be separated with space or comma.
167
168	- bpf filters
169
170	A BPF filter can access the sample data and make a decision based on the
171	data.  Users need to set an appropriate sample type to use the BPF
172	filter.  BPF filters need root privilege.
173
174	The sample data field can be specified in lower case letter.  Multiple
175	filters can be separated with comma.  For example,
176
177	  --filter 'period > 1000, cpu == 1'
178	or
179	  --filter 'mem_op == load || mem_op == store, mem_lvl > l1'
180
181	The former filter only accept samples with period greater than 1000 AND
182	CPU number is 1.  The latter one accepts either load and store memory
183	operations but it should have memory level above the L1.  Since the
184	mem_op and mem_lvl fields come from the (memory) data_source, it'd only
185	work with some events which set the data_source field.
186
187	Also user should request to collect that information (with -d option in
188	the above case).  Otherwise, the following message will be shown.
189
190	  $ sudo perf record -e cycles --filter 'mem_op == load'
191	  Error: cycles event does not have PERF_SAMPLE_DATA_SRC
192	   Hint: please add -d option to perf record.
193	  failed to set filter "BPF" on event cycles with 22 (Invalid argument)
194
195	Essentially the BPF filter expression is:
196
197	  <term> <operator> <value> (("," | "||") <term> <operator> <value>)*
198
199	The <term> can be one of:
200	  ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
201	  code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
202	  p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
203	  mem_dtlb, mem_blk, mem_hops
204
205	The <operator> can be one of:
206	  ==, !=, >, >=, <, <=, &
207
208	The <value> can be one of:
209	  <number> (for any term)
210	  na, load, store, pfetch, exec (for mem_op)
211	  l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
212	  na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
213	  remote (for mem_remote)
214	  na, locked (for mem_locked)
215	  na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
216	  na, by_data, by_addr (for mem_blk)
217	  hops0, hops1, hops2, hops3 (for mem_hops)
218
219--exclude-perf::
220	Don't record events issued by perf itself. This option should follow
221	an event selector (-e) which selects tracepoint event(s). It adds a
222	filter expression 'common_pid != $PERFPID' to filters. If other
223	'--filter' exists, the new filter expression will be combined with
224	them by '&&'.
225
226-a::
227--all-cpus::
228        System-wide collection from all CPUs (default if no target is specified).
229
230-p::
231--pid=::
232	Record events on existing process ID (comma separated list).
233
234-t::
235--tid=::
236        Record events on existing thread ID (comma separated list).
237        This option also disables inheritance by default.  Enable it by adding
238        --inherit.
239
240-u::
241--uid=::
242        Record events in threads owned by uid. Name or number.
243
244-r::
245--realtime=::
246	Collect data with this RT SCHED_FIFO priority.
247
248--no-buffering::
249	Collect data without buffering.
250
251-c::
252--count=::
253	Event period to sample.
254
255-o::
256--output=::
257	Output file name.
258
259-i::
260--no-inherit::
261	Child tasks do not inherit counters.
262
263-F::
264--freq=::
265	Profile at this frequency. Use 'max' to use the currently maximum
266	allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
267	sysctl. Will throttle down to the currently maximum allowed frequency.
268	See --strict-freq.
269
270--strict-freq::
271	Fail if the specified frequency can't be used.
272
273-m::
274--mmap-pages=::
275	Number of mmap data pages (must be a power of two) or size
276	specification with appended unit character - B/K/M/G. The
277	size is rounded up to have nearest pages power of two value.
278	Also, by adding a comma, the number of mmap pages for AUX
279	area tracing can be specified.
280
281-g::
282	Enables call-graph (stack chain/backtrace) recording for both
283	kernel space and user space.
284
285--call-graph::
286	Setup and enable call-graph (stack chain/backtrace) recording,
287	implies -g.  Default is "fp" (for user space).
288
289	The unwinding method used for kernel space is dependent on the
290	unwinder used by the active kernel configuration, i.e
291	CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)
292
293	Any option specified here controls the method used for user space.
294
295	Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
296	Call Frame Information) or "lbr" (Hardware Last Branch Record
297	facility).
298
299	In some systems, where binaries are build with gcc
300	--fomit-frame-pointer, using the "fp" method will produce bogus
301	call graphs, using "dwarf", if available (perf tools linked to
302	the libunwind or libdw library) should be used instead.
303	Using the "lbr" method doesn't require any compiler options. It
304	will produce call graphs from the hardware LBR registers. The
305	main limitation is that it is only available on new Intel
306	platforms, such as Haswell. It can only get user call chain. It
307	doesn't work with branch stack sampling at the same time.
308
309	When "dwarf" recording is used, perf also records (user) stack dump
310	when sampled.  Default size of the stack dump is 8192 (bytes).
311	User can change the size by passing the size after comma like
312	"--call-graph dwarf,4096".
313
314	When "fp" recording is used, perf tries to save stack enties
315	up to the number specified in sysctl.kernel.perf_event_max_stack
316	by default.  User can change the number by passing it after comma
317	like "--call-graph fp,32".
318
319-q::
320--quiet::
321	Don't print any warnings or messages, useful for scripting.
322
323-v::
324--verbose::
325	Be more verbose (show counter open errors, etc).
326
327-s::
328--stat::
329	Record per-thread event counts.  Use it with 'perf report -T' to see
330	the values.
331
332-d::
333--data::
334	Record the sample virtual addresses.
335
336--phys-data::
337	Record the sample physical addresses.
338
339--data-page-size::
340	Record the sampled data address data page size.
341
342--code-page-size::
343	Record the sampled code address (ip) page size
344
345-T::
346--timestamp::
347	Record the sample timestamps. Use it with 'perf report -D' to see the
348	timestamps, for instance.
349
350-P::
351--period::
352	Record the sample period.
353
354--sample-cpu::
355	Record the sample cpu.
356
357--sample-identifier::
358	Record the sample identifier i.e. PERF_SAMPLE_IDENTIFIER bit set in
359	the sample_type member of the struct perf_event_attr argument to the
360	perf_event_open system call.
361
362-n::
363--no-samples::
364	Don't sample.
365
366-R::
367--raw-samples::
368Collect raw sample records from all opened counters (default for tracepoint counters).
369
370-C::
371--cpu::
372Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
373comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
374In per-thread mode with inheritance mode on (default), samples are captured only when
375the thread executes on the designated CPUs. Default is to monitor all CPUs.
376
377User space tasks can migrate between CPUs, so when tracing selected CPUs,
378a dummy event is created to track sideband for all CPUs.
379
380-B::
381--no-buildid::
382Do not save the build ids of binaries in the perf.data files. This skips
383post processing after recording, which sometimes makes the final step in
384the recording process to take a long time, as it needs to process all
385events looking for mmap records. The downside is that it can misresolve
386symbols if the workload binaries used when recording get locally rebuilt
387or upgraded, because the only key available in this case is the
388pathname. You can also set the "record.build-id" config variable to
389'skip to have this behaviour permanently.
390
391-N::
392--no-buildid-cache::
393Do not update the buildid cache. This saves some overhead in situations
394where the information in the perf.data file (which includes buildids)
395is sufficient.  You can also set the "record.build-id" config variable to
396'no-cache' to have the same effect.
397
398-G name,...::
399--cgroup name,...::
400monitor only in the container (cgroup) called "name". This option is available only
401in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
402container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
403can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
404to first event, second cgroup to second event and so on. It is possible to provide
405an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
406corresponding events, i.e., they always refer to events defined earlier on the command
407line. If the user wants to track multiple events for a specific cgroup, the user can
408use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
409
410If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
411command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
412
413-b::
414--branch-any::
415Enable taken branch stack sampling. Any type of taken branch may be sampled.
416This is a shortcut for --branch-filter any. See --branch-filter for more infos.
417
418-j::
419--branch-filter::
420Enable taken branch stack sampling. Each sample captures a series of consecutive
421taken branches. The number of branches captured with each sample depends on the
422underlying hardware, the type of branches of interest, and the executed code.
423It is possible to select the types of branches captured by enabling filters. The
424following filters are defined:
425
426        - any:  any type of branches
427        - any_call: any function call or system call
428        - any_ret: any function return or system call return
429        - ind_call: any indirect branch
430        - ind_jmp: any indirect jump
431        - call: direct calls, including far (to/from kernel) calls
432        - u:  only when the branch target is at the user level
433        - k: only when the branch target is in the kernel
434        - hv: only when the target is at the hypervisor level
435	- in_tx: only when the target is in a hardware transaction
436	- no_tx: only when the target is not in a hardware transaction
437	- abort_tx: only when the target is a hardware transaction abort
438	- cond: conditional branches
439	- call_stack: save call stack
440	- no_flags: don't save branch flags e.g prediction, misprediction etc
441	- no_cycles: don't save branch cycles
442	- hw_index: save branch hardware index
443	- save_type: save branch type during sampling in case binary is not available later
444		     For the platforms with Intel Arch LBR support (12th-Gen+ client or
445		     4th-Gen Xeon+ server), the save branch type is unconditionally enabled
446		     when the taken branch stack sampling is enabled.
447	- priv: save privilege state during sampling in case binary is not available later
448
449+
450The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
451The privilege levels may be omitted, in which case, the privilege levels of the associated
452event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
453levels are subject to permissions.  When sampling on multiple events, branch stack sampling
454is enabled for all the sampling events. The sampled branch type is the same for all events.
455The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
456Note that this feature may not be available on all processors.
457
458-W::
459--weight::
460Enable weightened sampling. An additional weight is recorded per sample and can be
461displayed with the weight and local_weight sort keys.  This currently works for TSX
462abort events and some memory events in precise mode on modern Intel CPUs.
463
464--namespaces::
465Record events of type PERF_RECORD_NAMESPACES.  This enables 'cgroup_id' sort key.
466
467--all-cgroups::
468Record events of type PERF_RECORD_CGROUP.  This enables 'cgroup' sort key.
469
470--transaction::
471Record transaction flags for transaction related events.
472
473--per-thread::
474Use per-thread mmaps.  By default per-cpu mmaps are created.  This option
475overrides that and uses per-thread mmaps.  A side-effect of that is that
476inheritance is automatically disabled.  --per-thread is ignored with a warning
477if combined with -a or -C options.
478
479-D::
480--delay=::
481After starting the program, wait msecs before measuring (-1: start with events
482disabled), or enable events only for specified ranges of msecs (e.g.
483-D 10-20,30-40 means wait 10 msecs, enable for 10 msecs, wait 10 msecs, enable
484for 10 msecs, then stop). Note, delaying enabling of events is useful to filter
485out the startup phase of the program, which is often very different.
486
487-I::
488--intr-regs::
489Capture machine state (registers) at interrupt, i.e., on counter overflows for
490each sample. List of captured registers depends on the architecture. This option
491is off by default. It is possible to select the registers to sample using their
492symbolic names, e.g. on x86, ax, si. To list the available registers use
493--intr-regs=\?. To name registers, pass a comma separated list such as
494--intr-regs=ax,bx. The list of register is architecture dependent.
495
496--user-regs::
497Similar to -I, but capture user registers at sample time. To list the available
498user registers use --user-regs=\?.
499
500--running-time::
501Record running and enabled time for read events (:S)
502
503-k::
504--clockid::
505Sets the clock id to use for the various time fields in the perf_event_type
506records. See clock_gettime(). In particular CLOCK_MONOTONIC and
507CLOCK_MONOTONIC_RAW are supported, some events might also allow
508CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
509
510-S::
511--snapshot::
512Select AUX area tracing Snapshot Mode. This option is valid only with an
513AUX area tracing event. Optionally, certain snapshot capturing parameters
514can be specified in a string that follows this option:
515
516  - 'e': take one last snapshot on exit; guarantees that there is at least one
517       snapshot in the output file;
518  - <size>: if the PMU supports this, specify the desired snapshot size.
519
520In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
521and on exit if the above 'e' option is given.
522
523--aux-sample[=OPTIONS]::
524Select AUX area sampling. At least one of the events selected by the -e option
525must be an AUX area event. Samples on other events will be created containing
526data from the AUX area. Optionally sample size may be specified, otherwise it
527defaults to 4KiB.
528
529--proc-map-timeout::
530When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
531because the file may be huge. A time out is needed in such cases.
532This option sets the time out limit. The default value is 500 ms.
533
534--switch-events::
535Record context switch events i.e. events of type PERF_RECORD_SWITCH or
536PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT, CoreSight or Arm SPE)
537switch events will be enabled automatically, which can be suppressed by
538by the option --no-switch-events.
539
540--vmlinux=PATH::
541Specify vmlinux path which has debuginfo.
542(enabled when BPF prologue is on)
543
544--buildid-all::
545Record build-id of all DSOs regardless whether it's actually hit or not.
546
547--buildid-mmap::
548Record build ids in mmap2 events, disables build id cache (implies --no-buildid).
549
550--aio[=n]::
551Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
552Asynchronous mode is supported only when linking Perf tool with libc library
553providing implementation for Posix AIO API.
554
555--affinity=mode::
556Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
557
558  - node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
559  - cpu  - thread affinity mask is set to cpu of the processed mmap buffer
560
561--mmap-flush=number::
562
563Specify minimal number of bytes that is extracted from mmap data pages and
564processed for output. One can specify the number using B/K/M/G suffixes.
565
566The maximal allowed value is a quarter of the size of mmaped data pages.
567
568The default option value is 1 byte which means that every time that the output
569writing thread finds some new data in the mmaped buffer the data is extracted,
570possibly compressed (-z) and written to the output, perf.data or pipe.
571
572Larger data chunks are compressed more effectively in comparison to smaller
573chunks so extraction of larger chunks from the mmap data pages is preferable
574from the perspective of output size reduction.
575
576Also at some cases executing less output write syscalls with bigger data size
577can take less time than executing more output write syscalls with smaller data
578size thus lowering runtime profiling overhead.
579
580-z::
581--compression-level[=n]::
582Produce compressed trace using specified level n (default: 1 - fastest compression,
58322 - smallest trace)
584
585--all-kernel::
586Configure all used events to run in kernel space.
587
588--all-user::
589Configure all used events to run in user space.
590
591--kernel-callchains::
592Collect callchains only from kernel space. I.e. this option sets
593perf_event_attr.exclude_callchain_user to 1.
594
595--user-callchains::
596Collect callchains only from user space. I.e. this option sets
597perf_event_attr.exclude_callchain_kernel to 1.
598
599Don't use both --kernel-callchains and --user-callchains at the same time or no
600callchains will be collected.
601
602--timestamp-filename
603Append timestamp to output file name.
604
605--timestamp-boundary::
606Record timestamp boundary (time of first/last samples).
607
608--switch-output[=mode]::
609Generate multiple perf.data files, timestamp prefixed, switching to a new one
610based on 'mode' value:
611
612  - "signal" - when receiving a SIGUSR2 (default value) or
613  - <size>   - when reaching the size threshold, size is expected to
614               be a number with appended unit character - B/K/M/G
615  - <time>   - when reaching the time threshold, size is expected to
616               be a number with appended unit character - s/m/h/d
617
618               Note: the precision of  the size  threshold  hugely depends
619               on your configuration  - the number and size of  your  ring
620               buffers (-m). It is generally more precise for higher sizes
621               (like >5M), for lower values expect different sizes.
622
623A possible use case is to, given an external event, slice the perf.data file
624that gets then processed, possibly via a perf script, to decide if that
625particular perf.data snapshot should be kept or not.
626
627Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
628The reason for the latter two is to reduce the data file switching
629overhead. You can still switch them on with:
630
631  --switch-output --no-no-buildid  --no-no-buildid-cache
632
633--switch-output-event::
634Events that will cause the switch of the perf.data file, auto-selecting
635--switch-output=signal, the results are similar as internally the side band
636thread will also send a SIGUSR2 to the main one.
637
638Uses the same syntax as --event, it will just not be recorded, serving only to
639switch the perf.data file as soon as the --switch-output event is processed by
640a separate sideband thread.
641
642This sideband thread is also used to other purposes, like processing the
643PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
644information, etc.
645
646--switch-max-files=N::
647
648When rotating perf.data with --switch-output, only keep N files.
649
650--dry-run::
651Parse options then exit. --dry-run can be used to detect errors in cmdline
652options.
653
654'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
655in config file is set to true.
656
657--synth=TYPE::
658Collect and synthesize given type of events (comma separated).  Note that
659this option controls the synthesis from the /proc filesystem which represent
660task status for pre-existing threads.
661
662Kernel (and some other) events are recorded regardless of the
663choice in this option.  For example, --synth=no would have MMAP events for
664kernel and modules.
665
666Available types are:
667
668  - 'task'    - synthesize FORK and COMM events for each task
669  - 'mmap'    - synthesize MMAP events for each process (implies 'task')
670  - 'cgroup'  - synthesize CGROUP events for each cgroup
671  - 'all'     - synthesize all events (default)
672  - 'no'      - do not synthesize any of the above events
673
674--tail-synthesize::
675Instead of collecting non-sample events (for example, fork, comm, mmap) at
676the beginning of record, collect them during finalizing an output file.
677The collected non-sample events reflects the status of the system when
678record is finished.
679
680--overwrite::
681Makes all events use an overwritable ring buffer. An overwritable ring
682buffer works like a flight recorder: when it gets full, the kernel will
683overwrite the oldest records, that thus will never make it to the
684perf.data file.
685
686When '--overwrite' and '--switch-output' are used perf records and drops
687events until it receives a signal, meaning that something unusual was
688detected that warrants taking a snapshot of the most current events,
689those fitting in the ring buffer at that moment.
690
691'overwrite' attribute can also be set or canceled for an event using
692config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
693
694Implies --tail-synthesize.
695
696--kcore::
697Make a copy of /proc/kcore and place it into a directory with the perf data file.
698
699--max-size=<size>::
700Limit the sample data max size, <size> is expected to be a number with
701appended unit character - B/K/M/G
702
703--num-thread-synthesize::
704	The number of threads to run when synthesizing events for existing processes.
705	By default, the number of threads equals 1.
706
707ifdef::HAVE_LIBPFM[]
708--pfm-events events::
709Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
710including support for event filters. For example '--pfm-events
711inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
712option using the comma separator. Hardware events and generic hardware
713events cannot be mixed together. The latter must be used with the -e
714option. The -e option and this one can be mixed and matched.  Events
715can be grouped using the {} notation.
716endif::HAVE_LIBPFM[]
717
718--control=fifo:ctl-fifo[,ack-fifo]::
719--control=fd:ctl-fd[,ack-fd]::
720ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
721Listen on ctl-fd descriptor for command to control measurement.
722
723Available commands:
724
725  - 'enable'           : enable events
726  - 'disable'          : disable events
727  - 'enable name'      : enable event 'name'
728  - 'disable name'     : disable event 'name'
729  - 'snapshot'         : AUX area tracing snapshot).
730  - 'stop'             : stop perf record
731  - 'ping'             : ping
732  - 'evlist [-v|-g|-F] : display all events
733
734                         -F  Show just the sample frequency used for each event.
735                         -v  Show all fields.
736                         -g  Show event group information.
737
738Measurements can be started with events disabled using --delay=-1 option. Optionally
739send control command completion ('ack\n') to ack-fd descriptor to synchronize with the
740controlling process.  Example of bash shell script to enable and disable events during
741measurements:
742
743 #!/bin/bash
744
745 ctl_dir=/tmp/
746
747 ctl_fifo=${ctl_dir}perf_ctl.fifo
748 test -p ${ctl_fifo} && unlink ${ctl_fifo}
749 mkfifo ${ctl_fifo}
750 exec {ctl_fd}<>${ctl_fifo}
751
752 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
753 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
754 mkfifo ${ctl_ack_fifo}
755 exec {ctl_fd_ack}<>${ctl_ack_fifo}
756
757 perf record -D -1 -e cpu-cycles -a               \
758             --control fd:${ctl_fd},${ctl_fd_ack} \
759             -- sleep 30 &
760 perf_pid=$!
761
762 sleep 5  && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
763 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
764
765 exec {ctl_fd_ack}>&-
766 unlink ${ctl_ack_fifo}
767
768 exec {ctl_fd}>&-
769 unlink ${ctl_fifo}
770
771 wait -n ${perf_pid}
772 exit $?
773
774--threads=<spec>::
775Write collected trace data into several data files using parallel threads.
776<spec> value can be user defined list of masks. Masks separated by colon
777define CPUs to be monitored by a thread and affinity mask of that thread
778is separated by slash:
779
780    <cpus mask 1>/<affinity mask 1>:<cpus mask 2>/<affinity mask 2>:...
781
782CPUs or affinity masks must not overlap with other corresponding masks.
783Invalid CPUs are ignored, but masks containing only invalid CPUs are not
784allowed.
785
786For example user specification like the following:
787
788    0,2-4/2-4:1,5-7/5-7
789
790specifies parallel threads layout that consists of two threads,
791the first thread monitors CPUs 0 and 2-4 with the affinity mask 2-4,
792the second monitors CPUs 1 and 5-7 with the affinity mask 5-7.
793
794<spec> value can also be a string meaning predefined parallel threads
795layout:
796
797    - cpu    - create new data streaming thread for every monitored cpu
798    - core   - create new thread to monitor CPUs grouped by a core
799    - package - create new thread to monitor CPUs grouped by a package
800    - numa   - create new threed to monitor CPUs grouped by a NUMA domain
801
802Predefined layouts can be used on systems with large number of CPUs in
803order not to spawn multiple per-cpu streaming threads but still avoid LOST
804events in data directory files. Option specified with no or empty value
805defaults to CPU layout. Masks defined or provided by the option value are
806filtered through the mask provided by -C option.
807
808--debuginfod[=URLs]::
809	Specify debuginfod URL to be used when cacheing perf.data binaries,
810	it follows the same syntax as the DEBUGINFOD_URLS variable, like:
811
812	  http://192.168.122.174:8002
813
814	If the URLs is not specified, the value of DEBUGINFOD_URLS
815	system environment variable is used.
816
817--off-cpu::
818	Enable off-cpu profiling with BPF.  The BPF program will collect
819	task scheduling information with (user) stacktrace and save them
820	as sample data of a software event named "offcpu-time".  The
821	sample period will have the time the task slept in nanoseconds.
822
823	Note that BPF can collect stack traces using frame pointer ("fp")
824	only, as of now.  So the applications built without the frame
825	pointer might see bogus addresses.
826
827include::intel-hybrid.txt[]
828
829SEE ALSO
830--------
831linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]
832