6 perf-record - Run a command and record its profile into perf.data
11 'perf record' [-e <EVENT> | --event=EVENT] [-a] <command>
12 'perf record' [-e <EVENT> | --event=EVENT] [-a] \-- <command> [<options>]
16 This command runs a command and gathers a performance counter profile
17 from it, into perf.data - without displaying anything.
19 This file can then be inspected later on, using 'perf report'.
25 Any command you can specify in a shell.
29 Select the PMU event. Selection can be:
31 - a symbolic event name (use 'perf list' to list all events)
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/*.
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.
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/*.
46 - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
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/*
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.
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.
75 See the linkperf:perf-list[1] man page for more parameters.
77 Note: If user explicitly sets options which conflict with the params,
78 the value set by the parameters will be overridden.
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:
85 perf record -e some_event/@cfg1,@cfg2=config/ ...
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.
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
99 If you want to profile write accesses in [0x1000~1008), just set
102 - a BPF source file (ending in .c) or a precompiled object file (ending
103 in .o) selects one or more BPF events.
104 The BPF program can attach to various perf events based on the ELF section
107 When processing a '.c' file, perf searches an installed LLVM to compile it
108 into an object file first. Optional clang options can be passed via the
109 '--clang-opt' command line option, e.g.:
111 perf record --clang-opt "-DLINUX_VERSION_CODE=0x50000" \
112 -e tests/bpf-script-example.c
114 Note: '--clang-opt' must be placed before '--event/-e'.
116 - a group of events surrounded by a pair of brace ("{event1,event2,...}").
117 Each event is separated by commas and the group should be quoted to
118 prevent the shell interpretation. You also need to use --group on
119 "perf report" to view group events together.
122 Event filter. This option should follow an event selector (-e).
123 If the event is a tracepoint, the filter string will be parsed by
124 the kernel. If the event is a hardware trace PMU (e.g. Intel PT
125 or CoreSight), it'll be processed as an address filter. Otherwise
126 it means a general filter using BPF which can be applied for any
131 In the case of tracepoints, multiple '--filter' options are combined
136 A hardware trace PMU advertises its ability to accept a number of
137 address filters by specifying a non-zero value in
138 /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
140 Address filters have the format:
142 filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
145 - 'filter': defines a region that will be traced.
146 - 'start': defines an address at which tracing will begin.
147 - 'stop': defines an address at which tracing will stop.
148 - 'tracestop': defines a region in which tracing will stop.
150 <file name> is the name of the object file, <start> is the offset to the
151 code to trace in that file, and <size> is the size of the region to
152 trace. 'start' and 'stop' filters need not specify a <size>.
154 If no object file is specified then the kernel is assumed, in which case
155 the start address must be a current kernel memory address.
157 <start> can also be specified by providing the name of a symbol. If the
158 symbol name is not unique, it can be disambiguated by inserting #n where
159 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
160 select only a global symbol. <size> can also be specified by providing
161 the name of a symbol, in which case the size is calculated to the end
162 of that symbol. For 'filter' and 'tracestop' filters, if <size> is
163 omitted and <start> is a symbol, then the size is calculated to the end
166 If <size> is omitted and <start> is '*', then the start and size will
167 be calculated from the first and last symbols, i.e. to trace the whole
170 If symbol names (or '*') are provided, they must be surrounded by white
173 The filter passed to the kernel is not necessarily the same as entered.
174 To see the filter that is passed, use the -v option.
176 The kernel may not be able to configure a trace region if it is not
177 within a single mapping. MMAP events (or /proc/<pid>/maps) can be
178 examined to determine if that is a possibility.
180 Multiple filters can be separated with space or comma.
184 A BPF filter can access the sample data and make a decision based on the
185 data. Users need to set an appropriate sample type to use the BPF
186 filter. BPF filters need root privilege.
188 The sample data field can be specified in lower case letter. Multiple
189 filters can be separated with comma. For example,
191 --filter 'period > 1000, cpu == 1'
193 --filter 'mem_op == load || mem_op == store, mem_lvl > l1'
195 The former filter only accept samples with period greater than 1000 AND
196 CPU number is 1. The latter one accepts either load and store memory
197 operations but it should have memory level above the L1. Since the
198 mem_op and mem_lvl fields come from the (memory) data_source, it'd only
199 work with some events which set the data_source field.
201 Also user should request to collect that information (with -d option in
202 the above case). Otherwise, the following message will be shown.
204 $ sudo perf record -e cycles --filter 'mem_op == load'
205 Error: cycles event does not have PERF_SAMPLE_DATA_SRC
206 Hint: please add -d option to perf record.
207 failed to set filter "BPF" on event cycles with 22 (Invalid argument)
209 Essentially the BPF filter expression is:
211 <term> <operator> <value> (("," | "||") <term> <operator> <value>)*
213 The <term> can be one of:
214 ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
215 code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
216 p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
217 mem_dtlb, mem_blk, mem_hops
219 The <operator> can be one of:
220 ==, !=, >, >=, <, <=, &
222 The <value> can be one of:
223 <number> (for any term)
224 na, load, store, pfetch, exec (for mem_op)
225 l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
226 na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
227 remote (for mem_remote)
228 na, locked (for mem_locked)
229 na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
230 na, by_data, by_addr (for mem_blk)
231 hops0, hops1, hops2, hops3 (for mem_hops)
234 Don't record events issued by perf itself. This option should follow
235 an event selector (-e) which selects tracepoint event(s). It adds a
236 filter expression 'common_pid != $PERFPID' to filters. If other
237 '--filter' exists, the new filter expression will be combined with
242 System-wide collection from all CPUs (default if no target is specified).
246 Record events on existing process ID (comma separated list).
250 Record events on existing thread ID (comma separated list).
251 This option also disables inheritance by default. Enable it by adding
256 Record events in threads owned by uid. Name or number.
260 Collect data with this RT SCHED_FIFO priority.
263 Collect data without buffering.
267 Event period to sample.
275 Child tasks do not inherit counters.
279 Profile at this frequency. Use 'max' to use the currently maximum
280 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
281 sysctl. Will throttle down to the currently maximum allowed frequency.
285 Fail if the specified frequency can't be used.
289 Number of mmap data pages (must be a power of two) or size
290 specification with appended unit character - B/K/M/G. The
291 size is rounded up to have nearest pages power of two value.
292 Also, by adding a comma, the number of mmap pages for AUX
293 area tracing can be specified.
296 Enables call-graph (stack chain/backtrace) recording for both
297 kernel space and user space.
300 Setup and enable call-graph (stack chain/backtrace) recording,
301 implies -g. Default is "fp" (for user space).
303 The unwinding method used for kernel space is dependent on the
304 unwinder used by the active kernel configuration, i.e
305 CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)
307 Any option specified here controls the method used for user space.
309 Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
310 Call Frame Information) or "lbr" (Hardware Last Branch Record
313 In some systems, where binaries are build with gcc
314 --fomit-frame-pointer, using the "fp" method will produce bogus
315 call graphs, using "dwarf", if available (perf tools linked to
316 the libunwind or libdw library) should be used instead.
317 Using the "lbr" method doesn't require any compiler options. It
318 will produce call graphs from the hardware LBR registers. The
319 main limitation is that it is only available on new Intel
320 platforms, such as Haswell. It can only get user call chain. It
321 doesn't work with branch stack sampling at the same time.
323 When "dwarf" recording is used, perf also records (user) stack dump
324 when sampled. Default size of the stack dump is 8192 (bytes).
325 User can change the size by passing the size after comma like
326 "--call-graph dwarf,4096".
328 When "fp" recording is used, perf tries to save stack enties
329 up to the number specified in sysctl.kernel.perf_event_max_stack
330 by default. User can change the number by passing it after comma
331 like "--call-graph fp,32".
335 Don't print any warnings or messages, useful for scripting.
339 Be more verbose (show counter open errors, etc).
343 Record per-thread event counts. Use it with 'perf report -T' to see
348 Record the sample virtual addresses.
351 Record the sample physical addresses.
354 Record the sampled data address data page size.
357 Record the sampled code address (ip) page size
361 Record the sample timestamps. Use it with 'perf report -D' to see the
362 timestamps, for instance.
366 Record the sample period.
369 Record the sample cpu.
371 --sample-identifier::
372 Record the sample identifier i.e. PERF_SAMPLE_IDENTIFIER bit set in
373 the sample_type member of the struct perf_event_attr argument to the
374 perf_event_open system call.
382 Collect raw sample records from all opened counters (default for tracepoint counters).
386 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
387 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
388 In per-thread mode with inheritance mode on (default), samples are captured only when
389 the thread executes on the designated CPUs. Default is to monitor all CPUs.
393 Do not save the build ids of binaries in the perf.data files. This skips
394 post processing after recording, which sometimes makes the final step in
395 the recording process to take a long time, as it needs to process all
396 events looking for mmap records. The downside is that it can misresolve
397 symbols if the workload binaries used when recording get locally rebuilt
398 or upgraded, because the only key available in this case is the
399 pathname. You can also set the "record.build-id" config variable to
400 'skip to have this behaviour permanently.
404 Do not update the buildid cache. This saves some overhead in situations
405 where the information in the perf.data file (which includes buildids)
406 is sufficient. You can also set the "record.build-id" config variable to
407 'no-cache' to have the same effect.
411 monitor only in the container (cgroup) called "name". This option is available only
412 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
413 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
414 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
415 to first event, second cgroup to second event and so on. It is possible to provide
416 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
417 corresponding events, i.e., they always refer to events defined earlier on the command
418 line. If the user wants to track multiple events for a specific cgroup, the user can
419 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
421 If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
422 command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
426 Enable taken branch stack sampling. Any type of taken branch may be sampled.
427 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
431 Enable taken branch stack sampling. Each sample captures a series of consecutive
432 taken branches. The number of branches captured with each sample depends on the
433 underlying hardware, the type of branches of interest, and the executed code.
434 It is possible to select the types of branches captured by enabling filters. The
435 following filters are defined:
437 - any: any type of branches
438 - any_call: any function call or system call
439 - any_ret: any function return or system call return
440 - ind_call: any indirect branch
441 - ind_jmp: any indirect jump
442 - call: direct calls, including far (to/from kernel) calls
443 - u: only when the branch target is at the user level
444 - k: only when the branch target is in the kernel
445 - hv: only when the target is at the hypervisor level
446 - in_tx: only when the target is in a hardware transaction
447 - no_tx: only when the target is not in a hardware transaction
448 - abort_tx: only when the target is a hardware transaction abort
449 - cond: conditional branches
450 - call_stack: save call stack
451 - no_flags: don't save branch flags e.g prediction, misprediction etc
452 - no_cycles: don't save branch cycles
453 - hw_index: save branch hardware index
454 - save_type: save branch type during sampling in case binary is not available later
455 For the platforms with Intel Arch LBR support (12th-Gen+ client or
456 4th-Gen Xeon+ server), the save branch type is unconditionally enabled
457 when the taken branch stack sampling is enabled.
458 - priv: save privilege state during sampling in case binary is not available later
461 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
462 The privilege levels may be omitted, in which case, the privilege levels of the associated
463 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
464 levels are subject to permissions. When sampling on multiple events, branch stack sampling
465 is enabled for all the sampling events. The sampled branch type is the same for all events.
466 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
467 Note that this feature may not be available on all processors.
471 Enable weightened sampling. An additional weight is recorded per sample and can be
472 displayed with the weight and local_weight sort keys. This currently works for TSX
473 abort events and some memory events in precise mode on modern Intel CPUs.
476 Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.
479 Record events of type PERF_RECORD_CGROUP. This enables 'cgroup' sort key.
482 Record transaction flags for transaction related events.
485 Use per-thread mmaps. By default per-cpu mmaps are created. This option
486 overrides that and uses per-thread mmaps. A side-effect of that is that
487 inheritance is automatically disabled. --per-thread is ignored with a warning
488 if combined with -a or -C options.
492 After starting the program, wait msecs before measuring (-1: start with events
493 disabled), or enable events only for specified ranges of msecs (e.g.
494 -D 10-20,30-40 means wait 10 msecs, enable for 10 msecs, wait 10 msecs, enable
495 for 10 msecs, then stop). Note, delaying enabling of events is useful to filter
496 out the startup phase of the program, which is often very different.
500 Capture machine state (registers) at interrupt, i.e., on counter overflows for
501 each sample. List of captured registers depends on the architecture. This option
502 is off by default. It is possible to select the registers to sample using their
503 symbolic names, e.g. on x86, ax, si. To list the available registers use
504 --intr-regs=\?. To name registers, pass a comma separated list such as
505 --intr-regs=ax,bx. The list of register is architecture dependent.
508 Similar to -I, but capture user registers at sample time. To list the available
509 user registers use --user-regs=\?.
512 Record running and enabled time for read events (:S)
516 Sets the clock id to use for the various time fields in the perf_event_type
517 records. See clock_gettime(). In particular CLOCK_MONOTONIC and
518 CLOCK_MONOTONIC_RAW are supported, some events might also allow
519 CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
523 Select AUX area tracing Snapshot Mode. This option is valid only with an
524 AUX area tracing event. Optionally, certain snapshot capturing parameters
525 can be specified in a string that follows this option:
526 'e': take one last snapshot on exit; guarantees that there is at least one
527 snapshot in the output file;
528 <size>: if the PMU supports this, specify the desired snapshot size.
530 In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
531 and on exit if the above 'e' option is given.
533 --aux-sample[=OPTIONS]::
534 Select AUX area sampling. At least one of the events selected by the -e option
535 must be an AUX area event. Samples on other events will be created containing
536 data from the AUX area. Optionally sample size may be specified, otherwise it
540 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
541 because the file may be huge. A time out is needed in such cases.
542 This option sets the time out limit. The default value is 500 ms.
545 Record context switch events i.e. events of type PERF_RECORD_SWITCH or
546 PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT, CoreSight or Arm SPE)
547 switch events will be enabled automatically, which can be suppressed by
548 by the option --no-switch-events.
551 Path to clang binary to use for compiling BPF scriptlets.
552 (enabled when BPF support is on)
554 --clang-opt=OPTIONS::
555 Options passed to clang when compiling BPF scriptlets.
556 (enabled when BPF support is on)
559 Specify vmlinux path which has debuginfo.
560 (enabled when BPF prologue is on)
563 Record build-id of all DSOs regardless whether it's actually hit or not.
566 Record build ids in mmap2 events, disables build id cache (implies --no-buildid).
569 Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
570 Asynchronous mode is supported only when linking Perf tool with libc library
571 providing implementation for Posix AIO API.
574 Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
575 node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
576 cpu - thread affinity mask is set to cpu of the processed mmap buffer
578 --mmap-flush=number::
580 Specify minimal number of bytes that is extracted from mmap data pages and
581 processed for output. One can specify the number using B/K/M/G suffixes.
583 The maximal allowed value is a quarter of the size of mmaped data pages.
585 The default option value is 1 byte which means that every time that the output
586 writing thread finds some new data in the mmaped buffer the data is extracted,
587 possibly compressed (-z) and written to the output, perf.data or pipe.
589 Larger data chunks are compressed more effectively in comparison to smaller
590 chunks so extraction of larger chunks from the mmap data pages is preferable
591 from the perspective of output size reduction.
593 Also at some cases executing less output write syscalls with bigger data size
594 can take less time than executing more output write syscalls with smaller data
595 size thus lowering runtime profiling overhead.
598 --compression-level[=n]::
599 Produce compressed trace using specified level n (default: 1 - fastest compression,
603 Configure all used events to run in kernel space.
606 Configure all used events to run in user space.
608 --kernel-callchains::
609 Collect callchains only from kernel space. I.e. this option sets
610 perf_event_attr.exclude_callchain_user to 1.
613 Collect callchains only from user space. I.e. this option sets
614 perf_event_attr.exclude_callchain_kernel to 1.
616 Don't use both --kernel-callchains and --user-callchains at the same time or no
617 callchains will be collected.
620 Append timestamp to output file name.
622 --timestamp-boundary::
623 Record timestamp boundary (time of first/last samples).
625 --switch-output[=mode]::
626 Generate multiple perf.data files, timestamp prefixed, switching to a new one
627 based on 'mode' value:
628 "signal" - when receiving a SIGUSR2 (default value) or
629 <size> - when reaching the size threshold, size is expected to
630 be a number with appended unit character - B/K/M/G
631 <time> - when reaching the time threshold, size is expected to
632 be a number with appended unit character - s/m/h/d
634 Note: the precision of the size threshold hugely depends
635 on your configuration - the number and size of your ring
636 buffers (-m). It is generally more precise for higher sizes
637 (like >5M), for lower values expect different sizes.
639 A possible use case is to, given an external event, slice the perf.data file
640 that gets then processed, possibly via a perf script, to decide if that
641 particular perf.data snapshot should be kept or not.
643 Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
644 The reason for the latter two is to reduce the data file switching
645 overhead. You can still switch them on with:
647 --switch-output --no-no-buildid --no-no-buildid-cache
649 --switch-output-event::
650 Events that will cause the switch of the perf.data file, auto-selecting
651 --switch-output=signal, the results are similar as internally the side band
652 thread will also send a SIGUSR2 to the main one.
654 Uses the same syntax as --event, it will just not be recorded, serving only to
655 switch the perf.data file as soon as the --switch-output event is processed by
656 a separate sideband thread.
658 This sideband thread is also used to other purposes, like processing the
659 PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
662 --switch-max-files=N::
664 When rotating perf.data with --switch-output, only keep N files.
667 Parse options then exit. --dry-run can be used to detect errors in cmdline
670 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
671 in config file is set to true.
674 Collect and synthesize given type of events (comma separated). Note that
675 this option controls the synthesis from the /proc filesystem which represent
676 task status for pre-existing threads.
678 Kernel (and some other) events are recorded regardless of the
679 choice in this option. For example, --synth=no would have MMAP events for
683 'task' - synthesize FORK and COMM events for each task
684 'mmap' - synthesize MMAP events for each process (implies 'task')
685 'cgroup' - synthesize CGROUP events for each cgroup
686 'all' - synthesize all events (default)
687 'no' - do not synthesize any of the above events
690 Instead of collecting non-sample events (for example, fork, comm, mmap) at
691 the beginning of record, collect them during finalizing an output file.
692 The collected non-sample events reflects the status of the system when
696 Makes all events use an overwritable ring buffer. An overwritable ring
697 buffer works like a flight recorder: when it gets full, the kernel will
698 overwrite the oldest records, that thus will never make it to the
701 When '--overwrite' and '--switch-output' are used perf records and drops
702 events until it receives a signal, meaning that something unusual was
703 detected that warrants taking a snapshot of the most current events,
704 those fitting in the ring buffer at that moment.
706 'overwrite' attribute can also be set or canceled for an event using
707 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
709 Implies --tail-synthesize.
712 Make a copy of /proc/kcore and place it into a directory with the perf data file.
715 Limit the sample data max size, <size> is expected to be a number with
716 appended unit character - B/K/M/G
718 --num-thread-synthesize::
719 The number of threads to run when synthesizing events for existing processes.
720 By default, the number of threads equals 1.
723 --pfm-events events::
724 Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
725 including support for event filters. For example '--pfm-events
726 inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
727 option using the comma separator. Hardware events and generic hardware
728 events cannot be mixed together. The latter must be used with the -e
729 option. The -e option and this one can be mixed and matched. Events
730 can be grouped using the {} notation.
733 --control=fifo:ctl-fifo[,ack-fifo]::
734 --control=fd:ctl-fd[,ack-fd]::
735 ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
736 Listen on ctl-fd descriptor for command to control measurement.
739 'enable' : enable events
740 'disable' : disable events
741 'enable name' : enable event 'name'
742 'disable name' : disable event 'name'
743 'snapshot' : AUX area tracing snapshot).
744 'stop' : stop perf record
747 'evlist [-v|-g|-F] : display all events
748 -F Show just the sample frequency used for each event.
750 -g Show event group information.
752 Measurements can be started with events disabled using --delay=-1 option. Optionally
753 send control command completion ('ack\n') to ack-fd descriptor to synchronize with the
754 controlling process. Example of bash shell script to enable and disable events during
761 ctl_fifo=${ctl_dir}perf_ctl.fifo
762 test -p ${ctl_fifo} && unlink ${ctl_fifo}
764 exec {ctl_fd}<>${ctl_fifo}
766 ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
767 test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
768 mkfifo ${ctl_ack_fifo}
769 exec {ctl_fd_ack}<>${ctl_ack_fifo}
771 perf record -D -1 -e cpu-cycles -a \
772 --control fd:${ctl_fd},${ctl_fd_ack} \
776 sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
777 sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
780 unlink ${ctl_ack_fifo}
789 Write collected trace data into several data files using parallel threads.
790 <spec> value can be user defined list of masks. Masks separated by colon
791 define CPUs to be monitored by a thread and affinity mask of that thread
792 is separated by slash:
794 <cpus mask 1>/<affinity mask 1>:<cpus mask 2>/<affinity mask 2>:...
796 CPUs or affinity masks must not overlap with other corresponding masks.
797 Invalid CPUs are ignored, but masks containing only invalid CPUs are not
800 For example user specification like the following:
804 specifies parallel threads layout that consists of two threads,
805 the first thread monitors CPUs 0 and 2-4 with the affinity mask 2-4,
806 the second monitors CPUs 1 and 5-7 with the affinity mask 5-7.
808 <spec> value can also be a string meaning predefined parallel threads
811 cpu - create new data streaming thread for every monitored cpu
812 core - create new thread to monitor CPUs grouped by a core
813 package - create new thread to monitor CPUs grouped by a package
814 numa - create new threed to monitor CPUs grouped by a NUMA domain
816 Predefined layouts can be used on systems with large number of CPUs in
817 order not to spawn multiple per-cpu streaming threads but still avoid LOST
818 events in data directory files. Option specified with no or empty value
819 defaults to CPU layout. Masks defined or provided by the option value are
820 filtered through the mask provided by -C option.
822 --debuginfod[=URLs]::
823 Specify debuginfod URL to be used when cacheing perf.data binaries,
824 it follows the same syntax as the DEBUGINFOD_URLS variable, like:
826 http://192.168.122.174:8002
828 If the URLs is not specified, the value of DEBUGINFOD_URLS
829 system environment variable is used.
832 Enable off-cpu profiling with BPF. The BPF program will collect
833 task scheduling information with (user) stacktrace and save them
834 as sample data of a software event named "offcpu-time". The
835 sample period will have the time the task slept in nanoseconds.
837 Note that BPF can collect stack traces using frame pointer ("fp")
838 only, as of now. So the applications built without the frame
839 pointer might see bogus addresses.
841 include::intel-hybrid.txt[]
845 linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]