6 perf-script - Read perf.data (created by perf record) and display trace output
11 'perf script' [<options>]
12 'perf script' [<options>] record <script> [<record-options>] <command>
13 'perf script' [<options>] report <script> [script-args]
14 'perf script' [<options>] <script> <required-script-args> [<record-options>] <command>
15 'perf script' [<options>] <top-script> [script-args]
19 This command reads the input file and displays the trace recorded.
21 There are several variants of perf script:
23 'perf script' to see a detailed trace of the workload that was
26 You can also run a set of pre-canned scripts that aggregate and
27 summarize the raw trace data in various ways (the list of scripts is
28 available via 'perf script -l'). The following variants allow you to
29 record and run those scripts:
31 'perf script record <script> <command>' to record the events required
32 for 'perf script report'. <script> is the name displayed in the
33 output of 'perf script --list' i.e. the actual script name minus any
34 language extension. If <command> is not specified, the events are
35 recorded using the -a (system-wide) 'perf record' option.
37 'perf script report <script> [args]' to run and display the results
38 of <script>. <script> is the name displayed in the output of 'perf
39 script --list' i.e. the actual script name minus any language
40 extension. The perf.data output from a previous run of 'perf script
41 record <script>' is used and should be present for this command to
42 succeed. [args] refers to the (mainly optional) args expected by
45 'perf script <script> <required-script-args> <command>' to both
46 record the events required for <script> and to run the <script>
47 using 'live-mode' i.e. without writing anything to disk. <script>
48 is the name displayed in the output of 'perf script --list' i.e. the
49 actual script name minus any language extension. If <command> is
50 not specified, the events are recorded using the -a (system-wide)
51 'perf record' option. If <script> has any required args, they
52 should be specified before <command>. This mode doesn't allow for
53 optional script args to be specified; if optional script args are
54 desired, they can be specified using separate 'perf script record'
55 and 'perf script report' commands, with the stdout of the record step
56 piped to the stdin of the report script, using the '-o -' and '-i -'
57 options of the corresponding commands.
59 'perf script <top-script>' to both record the events required for
60 <top-script> and to run the <top-script> using 'live-mode'
61 i.e. without writing anything to disk. <top-script> is the name
62 displayed in the output of 'perf script --list' i.e. the actual
63 script name minus any language extension; a <top-script> is defined
64 as any script name ending with the string 'top'.
66 [<record-options>] can be passed to the record steps of 'perf script
67 record' and 'live-mode' variants; this isn't possible however for
68 <top-script> 'live-mode' or 'perf script report' variants.
70 See the 'SEE ALSO' section for links to language-specific
71 information on how to write and run your own trace scripts.
76 Any command you can specify in a shell.
80 Display verbose dump of the trace data.
82 --dump-unsorted-raw-trace=::
83 Same as --dump-raw-trace but not sorted in time order.
87 Show latency attributes (irqs/preemption disabled, etc).
91 Display a list of available trace scripts.
95 Process trace data with the given script ([lang]:script[.ext]).
96 If the string 'lang' is specified in place of a script name, a
97 list of supported languages will be displayed instead.
101 Generate perf-script.[ext] starter script for given language,
102 using current perf.data.
105 Filter sample events using the given shared object file.
106 Refer linkperf:perf-dlfilter[1]
109 Pass 'arg' as an argument to the dlfilter. --dlarg may be repeated
110 to add more arguments.
113 Display a list of available dlfilters. Use with option -v (must come
114 before option --list-dlfilters) to show long descriptions.
117 Force system-wide collection. Scripts run without a <command>
118 normally use -a by default, while scripts run with a <command>
119 normally don't - this option allows the latter to be run in
124 Input file name. (default: perf.data unless stdin is a fifo)
128 Do various checks like samples ordering and lost events.
132 Comma separated list of fields to print. Options are:
133 comm, tid, pid, time, cpu, event, trace, ip, sym, dso, dsoff, addr, symoff,
134 srcline, period, iregs, uregs, brstack, brstacksym, flags, bpf-output,
135 brstackinsn, brstackinsnlen, brstackoff, callindent, insn, insnlen, synth,
136 phys_addr, metric, misc, srccode, ipc, data_page_size, code_page_size, ins_lat,
137 machine_pid, vcpu, cgroup, retire_lat.
138 Field list can be prepended with the type, trace, sw or hw,
139 to indicate to which event type the field list applies.
140 e.g., -F sw:comm,tid,time,ip,sym and -F trace:time,cpu,trace
142 perf script -F <fields>
146 perf script -F trace:<fields> -F sw:<fields> -F hw:<fields>
148 i.e., the specified fields apply to all event types if the type string
151 In addition to overriding fields, it is also possible to add or remove
152 fields from the defaults. For example
156 removes the cpu field and adds the insn field. Adding/removing fields
157 cannot be mixed with normal overriding.
159 The arguments are processed in the order received. A later usage can
160 reset a prior request. e.g.:
162 -F trace: -F comm,tid,time,ip,sym
164 The first -F suppresses trace events (field list is ""), but then the
165 second invocation sets the fields to comm,tid,time,ip,sym. In this case a
166 warning is given to the user:
168 "Overriding previous field request for all events."
170 Alternatively, consider the order:
172 -F comm,tid,time,ip,sym -F trace:
174 The first -F sets the fields for all events and the second -F
175 suppresses trace events. The user is given a warning message about
176 the override, and the result of the above is that only S/W and H/W
177 events are displayed with the given fields.
179 It's possible tp add/remove fields only for specific event type:
183 removes cpu and period from software events.
185 For the 'wildcard' option if a user selected field is invalid for an
186 event type, a message is displayed to the user that the option is
187 ignored for that type. For example:
189 $ perf script -F comm,tid,trace
190 'trace' not valid for hardware events. Ignoring.
191 'trace' not valid for software events. Ignoring.
193 Alternatively, if the type is given an invalid field is specified it
194 is an error. For example:
196 perf script -v -F sw:comm,tid,trace
197 'trace' not valid for software events.
199 At this point usage is displayed, and perf-script exits.
201 The flags field is synthesized and may have a value when Instruction
202 Trace decoding. The flags are "bcrosyiABExghDt" which stand for branch,
203 call, return, conditional, system, asynchronous, interrupt,
204 transaction abort, trace begin, trace end, in transaction, VM-Entry,
205 VM-Exit, interrupt disabled and interrupt disable toggle respectively.
206 Known combinations of flags are printed more nicely e.g.
207 "call" for "bc", "return" for "br", "jcc" for "bo", "jmp" for "b",
208 "int" for "bci", "iret" for "bri", "syscall" for "bcs", "sysret" for "brs",
209 "async" for "by", "hw int" for "bcyi", "tx abrt" for "bA", "tr strt" for "bB",
210 "tr end" for "bE", "vmentry" for "bcg", "vmexit" for "bch".
211 However the "x", "D" and "t" flags will be displayed separately in those
212 cases e.g. "jcc (xD)" for a condition branch within a transaction
213 with interrupts disabled. Note, interrupts becoming disabled is "t",
214 whereas interrupts becoming enabled is "Dt".
216 The callindent field is synthesized and may have a value when
217 Instruction Trace decoding. For calls and returns, it will display the
218 name of the symbol indented with spaces to reflect the stack depth.
220 When doing instruction trace decoding insn and insnlen give the
221 instruction bytes and the instruction length of the current
224 The synth field is used by synthesized events which may be created when
225 Instruction Trace decoding.
227 The ipc (instructions per cycle) field is synthesized and may have a value when
228 Instruction Trace decoding.
230 The machine_pid and vcpu fields are derived from data resulting from using
231 perf inject to insert a perf.data file recorded inside a virtual machine into
232 a perf.data file recorded on the host at the same time.
234 The cgroup fields requires sample having the cgroup id which is saved
235 when "--all-cgroups" option is passed to 'perf record'.
237 Finally, a user may not set fields to none for all event types.
238 i.e., -F "" is not allowed.
240 The brstack output includes branch related information with raw addresses using the
241 /v/v/v/v/cycles syntax in the following order:
242 FROM: branch source instruction
243 TO : branch target instruction
244 M/P/-: M=branch target mispredicted or branch direction was mispredicted, P=target predicted or direction predicted, -=not supported
245 X/- : X=branch inside a transactional region, -=not in transaction region or not supported
246 A/- : A=TSX abort entry, -=not aborted region or not supported
249 The brstacksym is identical to brstack, except that the FROM and TO addresses are printed in a symbolic form if possible.
251 When brstackinsn is specified the full assembler sequences of branch sequences for each sample
252 is printed. This is the full execution path leading to the sample. This is only supported when the
253 sample was recorded with perf record -b or -j any.
255 Use brstackinsnlen to print the brstackinsn lenght. For example, you
256 can’t know the next sequential instruction after an unconditional branch unless
257 you calculate that based on its length.
259 The brstackoff field will print an offset into a specific dso/binary.
261 With the metric option perf script can compute metrics for
262 sampling periods, similar to perf stat. This requires
263 specifying a group with multiple events defining metrics with the :S option
264 for perf record. perf will sample on the first event, and
265 print computed metrics for all the events in the group. Please note
266 that the metric computed is averaged over the whole sampling
267 period (since the last sample), not just for the sample point.
269 For sample events it's possible to display misc field with -F +misc option,
270 following letters are displayed for each bit:
272 PERF_RECORD_MISC_KERNEL K
273 PERF_RECORD_MISC_USER U
274 PERF_RECORD_MISC_HYPERVISOR H
275 PERF_RECORD_MISC_GUEST_KERNEL G
276 PERF_RECORD_MISC_GUEST_USER g
277 PERF_RECORD_MISC_MMAP_DATA* M
278 PERF_RECORD_MISC_COMM_EXEC E
279 PERF_RECORD_MISC_SWITCH_OUT S
280 PERF_RECORD_MISC_SWITCH_OUT_PREEMPT Sp
282 $ perf script -F +misc ...
283 sched-messaging 1414 K 28690.636582: 4590 cycles ...
284 sched-messaging 1407 U 28690.636600: 325620 cycles ...
285 sched-messaging 1414 K 28690.636608: 19473 cycles ...
286 misc field ___________/
295 --symfs=<directory>::
296 Look for files with symbols relative to this directory.
300 When printing symbols do not display call chain.
303 Stop display of callgraph at these symbols
306 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
307 be provided as a comma-separated list with no space: 0,1. Ranges of
308 CPUs are specified with -: 0-2. Default is to report samples on all
313 Only display events for these comms. CSV that understands
314 file://filename entries.
317 Only show events for given process ID (comma separated list).
320 Only show events for given thread ID (comma separated list).
324 Display extended information about the perf.data file. This adds
325 information which may be very large and thus may clutter the display.
326 It currently includes: cpu and numa topology of the host system.
327 It can only be used with the perf script report mode.
330 Try to resolve the path of [kernel.kallsyms]
333 Display task related events (e.g. FORK, COMM, EXIT).
336 Display mmap related events (e.g. MMAP, MMAP2).
338 --show-namespace-events
339 Display namespace events i.e. events of type PERF_RECORD_NAMESPACES.
342 Display context switch events i.e. events of type PERF_RECORD_SWITCH or
343 PERF_RECORD_SWITCH_CPU_WIDE.
346 Display lost events i.e. events of type PERF_RECORD_LOST.
349 Display finished round events i.e. events of type PERF_RECORD_FINISHED_ROUND.
352 Display bpf events i.e. events of type PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT.
355 Display cgroup events i.e. events of type PERF_RECORD_CGROUP.
357 --show-text-poke-events
358 Display text poke events i.e. events of type PERF_RECORD_TEXT_POKE and
362 Demangle symbol names to human readable form. It's enabled by default,
363 disable with --no-demangle.
366 Demangle kernel symbol names to human readable form (for C++ kernels).
369 Show perf.data header.
372 Show only perf.data header.
375 Options for decoding instruction tracing data. The options are:
377 include::itrace.txt[]
379 To disable decoding entirely, use --no-itrace.
382 Show the full path for source files for srcline output.
385 Set the stack depth limit when parsing the callchain, anything
386 beyond the specified depth will be ignored. This is a trade-off
387 between information loss and faster processing especially for
388 workloads that can have a very long callchain stack.
389 Note that when using the --itrace option the synthesized callchain size
390 will override this value if the synthesized callchain size is bigger.
395 Use 9 decimal places when displaying time (i.e. show the nanoseconds)
399 Don't do ownership validation.
402 Only analyze samples within given time window: <start>,<stop>. Times
403 have the format seconds.nanoseconds. If start is not given (i.e. time
404 string is ',x.y') then analysis starts at the beginning of the file. If
405 stop time is not given (i.e. time string is 'x.y,') then analysis goes
406 to end of file. Multiple ranges can be separated by spaces, which
407 requires the argument to be quoted e.g. --time "1234.567,1234.789 1235,"
409 Also support time percent with multiple time ranges. Time string is
410 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
413 Select the second 10% time slice:
414 perf script --time 10%/2
416 Select from 0% to 10% time slice:
417 perf script --time 0%-10%
419 Select the first and second 10% time slices:
420 perf script --time 10%/1,10%/2
422 Select from 0% to 10% and 30% to 40% slices:
423 perf script --time 0%-10%,30%-40%
426 Set the maximum number of program blocks to print with brstackinsn for
430 Print time stamps relative to trace start.
433 Print time stamps relative to previous event.
436 Create per event files with a "perf.data.EVENT.dump" name instead of
437 printing to stdout, useful, for instance, for generating flamegraphs.
440 If a callgraph address belongs to an inlined function, the inline stack
441 will be printed. Each entry has function name and file/line. Enabled by
442 default, disable with --no-inline.
445 Show instruction stream for intel_pt traces. Combine with --xed to
449 Run xed disassembler on output. Requires installing the xed disassembler.
452 --symbols=symbol[,symbol...]::
453 Only consider the listed symbols. Symbols are typically a name
454 but they may also be hexadecimal address.
456 The hexadecimal address may be the start address of a symbol or
457 any other address to filter the trace records
459 For example, to select the symbol noploop or the address 0x4007a0:
460 perf script --symbols=noploop,0x4007a0
462 Support filtering trace records by symbol name, start address of
463 symbol, any hexadecimal address and address range.
465 The comparison order is:
467 1. symbol name comparison
468 2. symbol start address comparison.
469 3. any hexadecimal address comparison.
470 4. address range comparison (see --addr-range).
473 Use with -S or --symbols to list traced records within address range.
475 For example, to list the traced records within the address range
476 [0x4007a0, 0x0x4007a9]:
477 perf script -S 0x4007a0 --addr-range 10
480 Only consider symbols in these DSOs.
483 Show call stream for intel_pt traces. The CPUs are interleaved, but
484 can be filtered with -C.
487 Show call and return stream for intel_pt traces.
490 For itrace only show specified functions and their callees for
491 itrace. Multiple functions can be separated by comma.
493 --switch-on EVENT_NAME::
494 Only consider events after this event is found.
496 --switch-off EVENT_NAME::
497 Stop considering events after this event is found.
499 --show-on-off-events::
500 Show the --switch-on/off events too.
503 Show callgraph with stitched LBRs, which may have more complete
504 callgraph. The perf.data file must have been obtained using
505 perf record --call-graph lbr.
506 Disabled by default. In common cases with call stack overflows,
507 it can recreate better call stacks than the default lbr call stack
508 output. But this approach is not foolproof. There can be cases
509 where it creates incorrect call stacks from incorrect matches.
510 The known limitations include exception handing such as
511 setjmp/longjmp will have calls/returns not match.
513 :GMEXAMPLECMD: script
515 include::guest-files.txt[]
519 linkperf:perf-record[1], linkperf:perf-script-perl[1],
520 linkperf:perf-script-python[1], linkperf:perf-intel-pt[1],
521 linkperf:perf-dlfilter[1]