1PERF-SCRIPT(1) perf Manual PERF-SCRIPT(1)
2
6 perf-script - Read perf.data (created by perf record) and display trace
7 output
8
10 perf script [<options>]
11 perf script [<options>] record <script> [<record-options>] <command>
12 perf script [<options>] report <script> [script-args]
13 perf script [<options>] <script> <required-script-args> [<record-options>] <command>
14 perf script [<options>] <top-script> [script-args]
15
17 This command reads the input file and displays the trace recorded.
18 There are several variants of perf script:
20 'perf script' to see a detailed trace of the workload that was
22 recorded.
23 You can also run a set of pre-canned scripts that aggregate and
25 summarize the raw trace data in various ways (the list of scripts is
26 available via 'perf script -l'). The following variants allow you to
27 record and run those scripts:
28 'perf script record <script> <command>' to record the events required
30 for 'perf script report'. <script> is the name displayed in the
31 output of 'perf script --list' i.e. the actual script name minus any
32 language extension. If <command> is not specified, the events are
33 recorded using the -a (system-wide) 'perf record' option.
34 'perf script report <script> [args]' to run and display the results
36 of <script>. <script> is the name displayed in the output of 'perf
37 script --list' i.e. the actual script name minus any language
38 extension. The perf.data output from a previous run of 'perf script
39 record <script>' is used and should be present for this command to
40 succeed. [args] refers to the (mainly optional) args expected by
41 the script.
42 'perf script <script> <required-script-args> <command>' to both
44 record the events required for <script> and to run the <script>
45 using 'live-mode' i.e. without writing anything to disk. <script>
46 is the name displayed in the output of 'perf script --list' i.e. the
47 actual script name minus any language extension. If <command> is
48 not specified, the events are recorded using the -a (system-wide)
49 'perf record' option. If <script> has any required args, they
50 should be specified before <command>. This mode doesn't allow for
51 optional script args to be specified; if optional script args are
52 desired, they can be specified using separate 'perf script record'
53 and 'perf script report' commands, with the stdout of the record step
54 piped to the stdin of the report script, using the '-o -' and '-i -'
55 options of the corresponding commands.
56 'perf script <top-script>' to both record the events required for
58 <top-script> and to run the <top-script> using 'live-mode'
59 i.e. without writing anything to disk. <top-script> is the name
60 displayed in the output of 'perf script --list' i.e. the actual
61 script name minus any language extension; a <top-script> is defined
62 as any script name ending with the string 'top'.
63 [<record-options>] can be passed to the record steps of 'perf script
65 record' and 'live-mode' variants; this isn't possible however for
66 <top-script> 'live-mode' or 'perf script report' variants.
67 See the 'SEE ALSO' section for links to language-specific
69 information on how to write and run your own trace scripts.
70
72 <command>...
73 Any command you can specify in a shell.
74 -D, --dump-raw-trace=
76 Display verbose dump of the trace data.
77 -L, --Latency=
79 Show latency attributes (irqs/preemption disabled, etc).
80 -l, --list=
82 Display a list of available trace scripts.
83 -s [lang], --script=
85 Process trace data with the given script ([lang]:script[.ext]). If
86 the string lang is specified in place of a script name, a list of
87 supported languages will be displayed instead.
88 -g, --gen-script=
90 Generate perf-script.[ext] starter script for given language, using
91 current perf.data.
92 --dlfilter=<file>
94 Filter sample events using the given shared object file. Refer
95 perf-dlfilter(1)
96 --dlarg=<arg>
98 Pass arg as an argument to the dlfilter. --dlarg may be repeated to
99 add more arguments.
100 --list-dlfilters
102 Display a list of available dlfilters. Use with option -v (must
103 come before option --list-dlfilters) to show long descriptions.
104 -a
106 Force system-wide collection. Scripts run without a <command>
107 normally use -a by default, while scripts run with a <command>
108 normally don’t - this option allows the latter to be run in
109 system-wide mode.
110 -i, --input=
112 Input file name. (default: perf.data unless stdin is a fifo)
113 -d, --debug-mode
115 Do various checks like samples ordering and lost events.
116 -F, --fields
118 Comma separated list of fields to print. Options are: comm, tid,
119 pid, time, cpu, event, trace, ip, sym, dso, addr, symoff, srcline,
120 period, iregs, uregs, brstack, brstacksym, flags, bpf-output,
121 brstackinsn, brstackinsnlen, brstackoff, callindent, insn, insnlen,
122 synth, phys_addr, metric, misc, srccode, ipc, data_page_size,
123 code_page_size, ins_lat. Field list can be prepended with the type,
124 trace, sw or hw, to indicate to which event type the field list
125 applies. e.g., -F sw:comm,tid,time,ip,sym and -F
126 trace:time,cpu,trace
127 perf script -F <fields>
129 is equivalent to:
131 perf script -F trace:<fields> -F sw:<fields> -F hw:<fields>
133 i.e., the specified fields apply to all event types if the type string
135 is not given.
136 In addition to overriding fields, it is also possible to add or remove
138 fields from the defaults. For example
139 -F -cpu,+insn
141 removes the cpu field and adds the insn field. Adding/removing fields
143 cannot be mixed with normal overriding.
144 The arguments are processed in the order received. A later usage can
146 reset a prior request. e.g.:
147 -F trace: -F comm,tid,time,ip,sym
149 The first -F suppresses trace events (field list is ""), but then the
151 second invocation sets the fields to comm,tid,time,ip,sym. In this case a
152 warning is given to the user:
153 "Overriding previous field request for all events."
155 Alternatively, consider the order:
157 -F comm,tid,time,ip,sym -F trace:
159 The first -F sets the fields for all events and the second -F
161 suppresses trace events. The user is given a warning message about
162 the override, and the result of the above is that only S/W and H/W
163 events are displayed with the given fields.
164 It's possible tp add/remove fields only for specific event type:
166 -Fsw:-cpu,-period
168 removes cpu and period from software events.
170 For the 'wildcard' option if a user selected field is invalid for an
172 event type, a message is displayed to the user that the option is
173 ignored for that type. For example:
174 $ perf script -F comm,tid,trace
176 'trace' not valid for hardware events. Ignoring.
177 'trace' not valid for software events. Ignoring.
178 Alternatively, if the type is given an invalid field is specified it
180 is an error. For example:
181 perf script -v -F sw:comm,tid,trace
183 'trace' not valid for software events.
184 At this point usage is displayed, and perf-script exits.
186 The flags field is synthesized and may have a value when Instruction
188 Trace decoding. The flags are "bcrosyiABExghDt" which stand for branch,
189 call, return, conditional, system, asynchronous, interrupt,
190 transaction abort, trace begin, trace end, in transaction, VM-Entry,
191 VM-Exit, interrupt disabled and interrupt disable toggle respectively.
192 Known combinations of flags are printed more nicely e.g.
193 "call" for "bc", "return" for "br", "jcc" for "bo", "jmp" for "b",
194 "int" for "bci", "iret" for "bri", "syscall" for "bcs", "sysret" for "brs",
195 "async" for "by", "hw int" for "bcyi", "tx abrt" for "bA", "tr strt" for "bB",
196 "tr end" for "bE", "vmentry" for "bcg", "vmexit" for "bch".
197 However the "x", "D" and "t" flags will be displayed separately in those
198 cases e.g. "jcc (xD)" for a condition branch within a transaction
199 with interrupts disabled. Note, interrupts becoming disabled is "t",
200 whereas interrupts becoming enabled is "Dt".
201 The callindent field is synthesized and may have a value when
203 Instruction Trace decoding. For calls and returns, it will display the
204 name of the symbol indented with spaces to reflect the stack depth.
205 When doing instruction trace decoding insn and insnlen give the
207 instruction bytes and the instruction length of the current
208 instruction.
209 The synth field is used by synthesized events which may be created when
211 Instruction Trace decoding.
212 The ipc (instructions per cycle) field is synthesized and may have a value when
214 Instruction Trace decoding.
215 Finally, a user may not set fields to none for all event types.
217 i.e., -F "" is not allowed.
218 The brstack output includes branch related information with raw addresses using the
220 /v/v/v/v/cycles syntax in the following order:
221 FROM: branch source instruction
222 TO : branch target instruction
223 M/P/-: M=branch target mispredicted or branch direction was mispredicted, P=target predicted or direction predicted, -=not supported
224 X/- : X=branch inside a transactional region, -=not in transaction region or not supported
225 A/- : A=TSX abort entry, -=not aborted region or not supported
226 cycles
227 The brstacksym is identical to brstack, except that the FROM and TO addresses are printed in a symbolic form if possible.
229 When brstackinsn is specified the full assembler sequences of branch sequences for each sample
231 is printed. This is the full execution path leading to the sample. This is only supported when the
232 sample was recorded with perf record -b or -j any.
233 Use brstackinsnlen to print the brstackinsn lenght. For example, you
235 can’t know the next sequential instruction after an unconditional branch unless
236 you calculate that based on its length.
237 The brstackoff field will print an offset into a specific dso/binary.
239 With the metric option perf script can compute metrics for
241 sampling periods, similar to perf stat. This requires
242 specifying a group with multiple events defining metrics with the :S option
243 for perf record. perf will sample on the first event, and
244 print computed metrics for all the events in the group. Please note
245 that the metric computed is averaged over the whole sampling
246 period (since the last sample), not just for the sample point.
247 For sample events it's possible to display misc field with -F +misc option,
249 following letters are displayed for each bit:
250 PERF_RECORD_MISC_KERNEL K
252 PERF_RECORD_MISC_USER U
253 PERF_RECORD_MISC_HYPERVISOR H
254 PERF_RECORD_MISC_GUEST_KERNEL G
255 PERF_RECORD_MISC_GUEST_USER g
256 PERF_RECORD_MISC_MMAP_DATA* M
257 PERF_RECORD_MISC_COMM_EXEC E
258 PERF_RECORD_MISC_SWITCH_OUT S
259 PERF_RECORD_MISC_SWITCH_OUT_PREEMPT Sp
260 $ perf script -F +misc ...
262 sched-messaging 1414 K 28690.636582: 4590 cycles ...
263 sched-messaging 1407 U 28690.636600: 325620 cycles ...
264 sched-messaging 1414 K 28690.636608: 19473 cycles ...
265 misc field ___________/
266 -k, --vmlinux=<file>
268 vmlinux pathname
269 --kallsyms=<file>
271 kallsyms pathname
272 --symfs=<directory>
274 Look for files with symbols relative to this directory.
275 -G, --hide-call-graph
277 When printing symbols do not display call chain.
278 --stop-bt
280 Stop display of callgraph at these symbols
281 -C, --cpu
283 Only report samples for the list of CPUs provided. Multiple CPUs
284 can be provided as a comma-separated list with no space: 0,1.
285 Ranges of CPUs are specified with -: 0-2. Default is to report
286 samples on all CPUs.
287 -c, --comms=
289 Only display events for these comms. CSV that understands
290 file://filename entries.
291 --pid=
293 Only show events for given process ID (comma separated list).
294 --tid=
296 Only show events for given thread ID (comma separated list).
297 -I, --show-info
299 Display extended information about the perf.data file. This adds
300 information which may be very large and thus may clutter the
301 display. It currently includes: cpu and numa topology of the host
302 system. It can only be used with the perf script report mode.
303 --show-kernel-path
305 Try to resolve the path of [kernel.kallsyms]
306 --show-task-events Display task related events (e.g. FORK, COMM, EXIT).
308 --show-mmap-events Display mmap related events (e.g. MMAP, MMAP2).
310 --show-namespace-events Display namespace events i.e. events of type
312 PERF_RECORD_NAMESPACES.
313 --show-switch-events Display context switch events i.e. events of type
315 PERF_RECORD_SWITCH or PERF_RECORD_SWITCH_CPU_WIDE.
316 --show-lost-events Display lost events i.e. events of type
318 PERF_RECORD_LOST.
319 --show-round-events Display finished round events i.e. events of type
321 PERF_RECORD_FINISHED_ROUND.
322 --show-bpf-events Display bpf events i.e. events of type
324 PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT.
325 --show-cgroup-events Display cgroup events i.e. events of type
327 PERF_RECORD_CGROUP.
328 --show-text-poke-events Display text poke events i.e. events of type
330 PERF_RECORD_TEXT_POKE and PERF_RECORD_KSYMBOL.
331 --demangle
333 Demangle symbol names to human readable form. It’s enabled by
334 default, disable with --no-demangle.
335 --demangle-kernel
337 Demangle kernel symbol names to human readable form (for C++
338 kernels).
339 --header Show perf.data header.
341 --header-only Show only perf.data header.
343 --itrace
345 Options for decoding instruction tracing data. The options are:
346 i synthesize instructions events
348 b synthesize branches events (branch misses for Arm SPE)
349 c synthesize branches events (calls only)
350 r synthesize branches events (returns only)
351 x synthesize transactions events
352 w synthesize ptwrite events
353 p synthesize power events (incl. PSB events for Intel PT)
354 o synthesize other events recorded due to the use
355 of aux-output (refer to perf record)
356 I synthesize interrupt or similar (asynchronous) events
357 (e.g. Intel PT Event Trace)
358 e synthesize error events
359 d create a debug log
360 f synthesize first level cache events
361 m synthesize last level cache events
362 M synthesize memory events
363 t synthesize TLB events
364 a synthesize remote access events
365 g synthesize a call chain (use with i or x)
366 G synthesize a call chain on existing event records
367 l synthesize last branch entries (use with i or x)
368 L synthesize last branch entries on existing event records
369 s skip initial number of events
370 q quicker (less detailed) decoding
371 A approximate IPC
372 Z prefer to ignore timestamps (so-called "timeless" decoding)
373 The default is all events i.e. the same as --itrace=ibxwpe,
375 except for perf script where it is --itrace=ce
376 In addition, the period (default 100000, except for perf script where it is 1)
378 for instructions events can be specified in units of:
379 i instructions
381 t ticks
382 ms milliseconds
383 us microseconds
384 ns nanoseconds (default)
385 Also the call chain size (default 16, max. 1024) for instructions or
387 transactions events can be specified.
388 Also the number of last branch entries (default 64, max. 1024) for
390 instructions or transactions events can be specified.
391 Similar to options g and l, size may also be specified for options G and L.
393 On x86, note that G and L work poorly when data has been recorded with
394 large PEBS. Refer linkperf:perf-intel-pt[1] man page for details.
395 It is also possible to skip events generated (instructions, branches, transactions,
397 ptwrite, power) at the beginning. This is useful to ignore initialization code.
398 --itrace=i0nss1000000
400 skips the first million instructions.
402 The 'e' option may be followed by flags which affect what errors will or
404 will not be reported. Each flag must be preceded by either '+' or '-'.
405 The flags are:
406 o overflow
407 l trace data lost
408 If supported, the 'd' option may be followed by flags which affect what
410 debug messages will or will not be logged. Each flag must be preceded
411 by either '+' or '-'. The flags are:
412 a all perf events
413 o output to stdout
414 If supported, the 'q' option may be repeated to increase the effect.
416 To disable decoding entirely, use --no-itrace.
418 --full-source-path
420 Show the full path for source files for srcline output.
421 --max-stack
423 Set the stack depth limit when parsing the callchain, anything
424 beyond the specified depth will be ignored. This is a trade-off
425 between information loss and faster processing especially for
426 workloads that can have a very long callchain stack. Note that when
427 using the --itrace option the synthesized callchain size will
428 override this value if the synthesized callchain size is bigger.
429 Default: 127
431 --ns
433 Use 9 decimal places when displaying time (i.e. show the
434 nanoseconds)
435 -f, --force
437 Don’t do ownership validation.
438 --time
440 Only analyze samples within given time window: <start>,<stop>.
441 Times have the format seconds.nanoseconds. If start is not given
442 (i.e. time string is ,x.y) then analysis starts at the beginning of
443 the file. If stop time is not given (i.e. time string is x.y,) then
444 analysis goes to end of file. Multiple ranges can be separated by
445 spaces, which requires the argument to be quoted e.g. --time
446 "1234.567,1234.789 1235,"
447 Also support time percent with multiple time ranges. Time string is
449 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
450 For example:
452 Select the second 10% time slice:
453 perf script --time 10%/2
454 Select from 0% to 10% time slice:
456 perf script --time 0%-10%
457 Select the first and second 10% time slices:
459 perf script --time 10%/1,10%/2
460 Select from 0% to 10% and 30% to 40% slices:
462 perf script --time 0%-10%,30%-40%
463 --max-blocks
465 Set the maximum number of program blocks to print with brstackinsn
466 for each sample.
467 --reltime
469 Print time stamps relative to trace start.
470 --deltatime
472 Print time stamps relative to previous event.
473 --per-event-dump
475 Create per event files with a "perf.data.EVENT.dump" name instead
476 of printing to stdout, useful, for instance, for generating
477 flamegraphs.
478 --inline
480 If a callgraph address belongs to an inlined function, the inline
481 stack will be printed. Each entry has function name and file/line.
482 Enabled by default, disable with --no-inline.
483 --insn-trace
485 Show instruction stream for intel_pt traces. Combine with --xed to
486 show disassembly.
487 --xed
489 Run xed disassembler on output. Requires installing the xed
490 disassembler.
491 -S, --symbols=symbol[,symbol...]
493 Only consider the listed symbols. Symbols are typically a name but
494 they may also be hexadecimal address.
495 The hexadecimal address may be the start address of a symbol or
497 any other address to filter the trace records
498 For example, to select the symbol noploop or the address 0x4007a0:
500 perf script --symbols=noploop,0x4007a0
501 Support filtering trace records by symbol name, start address of
503 symbol, any hexadecimal address and address range.
504 The comparison order is:
506 1. symbol name comparison
508 2. symbol start address comparison.
510 3. any hexadecimal address comparison.
512 4. address range comparison (see --addr-range).
514 --addr-range
516 Use with -S or --symbols to list traced records within address
517 range.
518 For example, to list the traced records within the address range
520 [0x4007a0, 0x0x4007a9]:
521 perf script -S 0x4007a0 --addr-range 10
522 --dsos=
524 Only consider symbols in these DSOs.
525 --call-trace
527 Show call stream for intel_pt traces. The CPUs are interleaved, but
528 can be filtered with -C.
529 --call-ret-trace
531 Show call and return stream for intel_pt traces.
532 --graph-function
534 For itrace only show specified functions and their callees for
535 itrace. Multiple functions can be separated by comma.
536 --switch-on EVENT_NAME
538 Only consider events after this event is found.
539 --switch-off EVENT_NAME
541 Stop considering events after this event is found.
542 --show-on-off-events
544 Show the --switch-on/off events too.
545 --stitch-lbr
547 Show callgraph with stitched LBRs, which may have more complete
548 callgraph. The perf.data file must have been obtained using perf
549 record --call-graph lbr. Disabled by default. In common cases with
550 call stack overflows, it can recreate better call stacks than the
551 default lbr call stack output. But this approach is not full proof.
552 There can be cases where it creates incorrect call stacks from
553 incorrect matches. The known limitations include exception handing
554 such as setjmp/longjmp will have calls/returns not match.
555
557 perf-record(1), perf-script-perl(1), perf-script-python(1), perf-intel-
558 pt(1), perf-dlfilter(1)
559perf 06/14/2022 PERF-SCRIPT(1)