0001 perf-script(1)
0002 =============
0003
0004 NAME
0005 ----
0006 perf-script - Read perf.data (created by perf record) and display trace output
0007
0008 SYNOPSIS
0009 --------
0010 [verse]
0011 'perf script' [<options>]
0012 'perf script' [<options>] record <script> [<record-options>] <command>
0013 'perf script' [<options>] report <script> [script-args]
0014 'perf script' [<options>] <script> <required-script-args> [<record-options>] <command>
0015 'perf script' [<options>] <top-script> [script-args]
0016
0017 DESCRIPTION
0018 -----------
0019 This command reads the input file and displays the trace recorded.
0020
0021 There are several variants of perf script:
0022
0023 'perf script' to see a detailed trace of the workload that was
0024 recorded.
0025
0026 You can also run a set of pre-canned scripts that aggregate and
0027 summarize the raw trace data in various ways (the list of scripts is
0028 available via 'perf script -l'). The following variants allow you to
0029 record and run those scripts:
0030
0031 'perf script record <script> <command>' to record the events required
0032 for 'perf script report'. <script> is the name displayed in the
0033 output of 'perf script --list' i.e. the actual script name minus any
0034 language extension. If <command> is not specified, the events are
0035 recorded using the -a (system-wide) 'perf record' option.
0036
0037 'perf script report <script> [args]' to run and display the results
0038 of <script>. <script> is the name displayed in the output of 'perf
0039 script --list' i.e. the actual script name minus any language
0040 extension. The perf.data output from a previous run of 'perf script
0041 record <script>' is used and should be present for this command to
0042 succeed. [args] refers to the (mainly optional) args expected by
0043 the script.
0044
0045 'perf script <script> <required-script-args> <command>' to both
0046 record the events required for <script> and to run the <script>
0047 using 'live-mode' i.e. without writing anything to disk. <script>
0048 is the name displayed in the output of 'perf script --list' i.e. the
0049 actual script name minus any language extension. If <command> is
0050 not specified, the events are recorded using the -a (system-wide)
0051 'perf record' option. If <script> has any required args, they
0052 should be specified before <command>. This mode doesn't allow for
0053 optional script args to be specified; if optional script args are
0054 desired, they can be specified using separate 'perf script record'
0055 and 'perf script report' commands, with the stdout of the record step
0056 piped to the stdin of the report script, using the '-o -' and '-i -'
0057 options of the corresponding commands.
0058
0059 'perf script <top-script>' to both record the events required for
0060 <top-script> and to run the <top-script> using 'live-mode'
0061 i.e. without writing anything to disk. <top-script> is the name
0062 displayed in the output of 'perf script --list' i.e. the actual
0063 script name minus any language extension; a <top-script> is defined
0064 as any script name ending with the string 'top'.
0065
0066 [<record-options>] can be passed to the record steps of 'perf script
0067 record' and 'live-mode' variants; this isn't possible however for
0068 <top-script> 'live-mode' or 'perf script report' variants.
0069
0070 See the 'SEE ALSO' section for links to language-specific
0071 information on how to write and run your own trace scripts.
0072
0073 OPTIONS
0074 -------
0075 <command>...::
0076 Any command you can specify in a shell.
0077
0078 -D::
0079 --dump-raw-trace=::
0080 Display verbose dump of the trace data.
0081
0082 --dump-unsorted-raw-trace=::
0083 Same as --dump-raw-trace but not sorted in time order.
0084
0085 -L::
0086 --Latency=::
0087 Show latency attributes (irqs/preemption disabled, etc).
0088
0089 -l::
0090 --list=::
0091 Display a list of available trace scripts.
0092
0093 -s ['lang']::
0094 --script=::
0095 Process trace data with the given script ([lang]:script[.ext]).
0096 If the string 'lang' is specified in place of a script name, a
0097 list of supported languages will be displayed instead.
0098
0099 -g::
0100 --gen-script=::
0101 Generate perf-script.[ext] starter script for given language,
0102 using current perf.data.
0103
0104 --dlfilter=<file>::
0105 Filter sample events using the given shared object file.
0106 Refer linkperf:perf-dlfilter[1]
0107
0108 --dlarg=<arg>::
0109 Pass 'arg' as an argument to the dlfilter. --dlarg may be repeated
0110 to add more arguments.
0111
0112 --list-dlfilters::
0113 Display a list of available dlfilters. Use with option -v (must come
0114 before option --list-dlfilters) to show long descriptions.
0115
0116 -a::
0117 Force system-wide collection. Scripts run without a <command>
0118 normally use -a by default, while scripts run with a <command>
0119 normally don't - this option allows the latter to be run in
0120 system-wide mode.
0121
0122 -i::
0123 --input=::
0124 Input file name. (default: perf.data unless stdin is a fifo)
0125
0126 -d::
0127 --debug-mode::
0128 Do various checks like samples ordering and lost events.
0129
0130 -F::
0131 --fields::
0132 Comma separated list of fields to print. Options are:
0133 comm, tid, pid, time, cpu, event, trace, ip, sym, dso, addr, symoff,
0134 srcline, period, iregs, uregs, brstack, brstacksym, flags, bpf-output,
0135 brstackinsn, brstackinsnlen, brstackoff, callindent, insn, insnlen, synth,
0136 phys_addr, metric, misc, srccode, ipc, data_page_size, code_page_size, ins_lat,
0137 machine_pid, vcpu.
0138 Field list can be prepended with the type, trace, sw or hw,
0139 to indicate to which event type the field list applies.
0140 e.g., -F sw:comm,tid,time,ip,sym and -F trace:time,cpu,trace
0141
0142 perf script -F <fields>
0143
0144 is equivalent to:
0145
0146 perf script -F trace:<fields> -F sw:<fields> -F hw:<fields>
0147
0148 i.e., the specified fields apply to all event types if the type string
0149 is not given.
0150
0151 In addition to overriding fields, it is also possible to add or remove
0152 fields from the defaults. For example
0153
0154 -F -cpu,+insn
0155
0156 removes the cpu field and adds the insn field. Adding/removing fields
0157 cannot be mixed with normal overriding.
0158
0159 The arguments are processed in the order received. A later usage can
0160 reset a prior request. e.g.:
0161
0162 -F trace: -F comm,tid,time,ip,sym
0163
0164 The first -F suppresses trace events (field list is ""), but then the
0165 second invocation sets the fields to comm,tid,time,ip,sym. In this case a
0166 warning is given to the user:
0167
0168 "Overriding previous field request for all events."
0169
0170 Alternatively, consider the order:
0171
0172 -F comm,tid,time,ip,sym -F trace:
0173
0174 The first -F sets the fields for all events and the second -F
0175 suppresses trace events. The user is given a warning message about
0176 the override, and the result of the above is that only S/W and H/W
0177 events are displayed with the given fields.
0178
0179 It's possible tp add/remove fields only for specific event type:
0180
0181 -Fsw:-cpu,-period
0182
0183 removes cpu and period from software events.
0184
0185 For the 'wildcard' option if a user selected field is invalid for an
0186 event type, a message is displayed to the user that the option is
0187 ignored for that type. For example:
0188
0189 $ perf script -F comm,tid,trace
0190 'trace' not valid for hardware events. Ignoring.
0191 'trace' not valid for software events. Ignoring.
0192
0193 Alternatively, if the type is given an invalid field is specified it
0194 is an error. For example:
0195
0196 perf script -v -F sw:comm,tid,trace
0197 'trace' not valid for software events.
0198
0199 At this point usage is displayed, and perf-script exits.
0200
0201 The flags field is synthesized and may have a value when Instruction
0202 Trace decoding. The flags are "bcrosyiABExghDt" which stand for branch,
0203 call, return, conditional, system, asynchronous, interrupt,
0204 transaction abort, trace begin, trace end, in transaction, VM-Entry,
0205 VM-Exit, interrupt disabled and interrupt disable toggle respectively.
0206 Known combinations of flags are printed more nicely e.g.
0207 "call" for "bc", "return" for "br", "jcc" for "bo", "jmp" for "b",
0208 "int" for "bci", "iret" for "bri", "syscall" for "bcs", "sysret" for "brs",
0209 "async" for "by", "hw int" for "bcyi", "tx abrt" for "bA", "tr strt" for "bB",
0210 "tr end" for "bE", "vmentry" for "bcg", "vmexit" for "bch".
0211 However the "x", "D" and "t" flags will be displayed separately in those
0212 cases e.g. "jcc (xD)" for a condition branch within a transaction
0213 with interrupts disabled. Note, interrupts becoming disabled is "t",
0214 whereas interrupts becoming enabled is "Dt".
0215
0216 The callindent field is synthesized and may have a value when
0217 Instruction Trace decoding. For calls and returns, it will display the
0218 name of the symbol indented with spaces to reflect the stack depth.
0219
0220 When doing instruction trace decoding insn and insnlen give the
0221 instruction bytes and the instruction length of the current
0222 instruction.
0223
0224 The synth field is used by synthesized events which may be created when
0225 Instruction Trace decoding.
0226
0227 The ipc (instructions per cycle) field is synthesized and may have a value when
0228 Instruction Trace decoding.
0229
0230 The machine_pid and vcpu fields are derived from data resulting from using
0231 perf inject to insert a perf.data file recorded inside a virtual machine into
0232 a perf.data file recorded on the host at the same time.
0233
0234 Finally, a user may not set fields to none for all event types.
0235 i.e., -F "" is not allowed.
0236
0237 The brstack output includes branch related information with raw addresses using the
0238 /v/v/v/v/cycles syntax in the following order:
0239 FROM: branch source instruction
0240 TO : branch target instruction
0241 M/P/-: M=branch target mispredicted or branch direction was mispredicted, P=target predicted or direction predicted, -=not supported
0242 X/- : X=branch inside a transactional region, -=not in transaction region or not supported
0243 A/- : A=TSX abort entry, -=not aborted region or not supported
0244 cycles
0245
0246 The brstacksym is identical to brstack, except that the FROM and TO addresses are printed in a symbolic form if possible.
0247
0248 When brstackinsn is specified the full assembler sequences of branch sequences for each sample
0249 is printed. This is the full execution path leading to the sample. This is only supported when the
0250 sample was recorded with perf record -b or -j any.
0251
0252 Use brstackinsnlen to print the brstackinsn lenght. For example, you
0253 can’t know the next sequential instruction after an unconditional branch unless
0254 you calculate that based on its length.
0255
0256 The brstackoff field will print an offset into a specific dso/binary.
0257
0258 With the metric option perf script can compute metrics for
0259 sampling periods, similar to perf stat. This requires
0260 specifying a group with multiple events defining metrics with the :S option
0261 for perf record. perf will sample on the first event, and
0262 print computed metrics for all the events in the group. Please note
0263 that the metric computed is averaged over the whole sampling
0264 period (since the last sample), not just for the sample point.
0265
0266 For sample events it's possible to display misc field with -F +misc option,
0267 following letters are displayed for each bit:
0268
0269 PERF_RECORD_MISC_KERNEL K
0270 PERF_RECORD_MISC_USER U
0271 PERF_RECORD_MISC_HYPERVISOR H
0272 PERF_RECORD_MISC_GUEST_KERNEL G
0273 PERF_RECORD_MISC_GUEST_USER g
0274 PERF_RECORD_MISC_MMAP_DATA* M
0275 PERF_RECORD_MISC_COMM_EXEC E
0276 PERF_RECORD_MISC_SWITCH_OUT S
0277 PERF_RECORD_MISC_SWITCH_OUT_PREEMPT Sp
0278
0279 $ perf script -F +misc ...
0280 sched-messaging 1414 K 28690.636582: 4590 cycles ...
0281 sched-messaging 1407 U 28690.636600: 325620 cycles ...
0282 sched-messaging 1414 K 28690.636608: 19473 cycles ...
0283 misc field ___________/
0284
0285 -k::
0286 --vmlinux=<file>::
0287 vmlinux pathname
0288
0289 --kallsyms=<file>::
0290 kallsyms pathname
0291
0292 --symfs=<directory>::
0293 Look for files with symbols relative to this directory.
0294
0295 -G::
0296 --hide-call-graph::
0297 When printing symbols do not display call chain.
0298
0299 --stop-bt::
0300 Stop display of callgraph at these symbols
0301
0302 -C::
0303 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
0304 be provided as a comma-separated list with no space: 0,1. Ranges of
0305 CPUs are specified with -: 0-2. Default is to report samples on all
0306 CPUs.
0307
0308 -c::
0309 --comms=::
0310 Only display events for these comms. CSV that understands
0311 file://filename entries.
0312
0313 --pid=::
0314 Only show events for given process ID (comma separated list).
0315
0316 --tid=::
0317 Only show events for given thread ID (comma separated list).
0318
0319 -I::
0320 --show-info::
0321 Display extended information about the perf.data file. This adds
0322 information which may be very large and thus may clutter the display.
0323 It currently includes: cpu and numa topology of the host system.
0324 It can only be used with the perf script report mode.
0325
0326 --show-kernel-path::
0327 Try to resolve the path of [kernel.kallsyms]
0328
0329 --show-task-events
0330 Display task related events (e.g. FORK, COMM, EXIT).
0331
0332 --show-mmap-events
0333 Display mmap related events (e.g. MMAP, MMAP2).
0334
0335 --show-namespace-events
0336 Display namespace events i.e. events of type PERF_RECORD_NAMESPACES.
0337
0338 --show-switch-events
0339 Display context switch events i.e. events of type PERF_RECORD_SWITCH or
0340 PERF_RECORD_SWITCH_CPU_WIDE.
0341
0342 --show-lost-events
0343 Display lost events i.e. events of type PERF_RECORD_LOST.
0344
0345 --show-round-events
0346 Display finished round events i.e. events of type PERF_RECORD_FINISHED_ROUND.
0347
0348 --show-bpf-events
0349 Display bpf events i.e. events of type PERF_RECORD_KSYMBOL and PERF_RECORD_BPF_EVENT.
0350
0351 --show-cgroup-events
0352 Display cgroup events i.e. events of type PERF_RECORD_CGROUP.
0353
0354 --show-text-poke-events
0355 Display text poke events i.e. events of type PERF_RECORD_TEXT_POKE and
0356 PERF_RECORD_KSYMBOL.
0357
0358 --demangle::
0359 Demangle symbol names to human readable form. It's enabled by default,
0360 disable with --no-demangle.
0361
0362 --demangle-kernel::
0363 Demangle kernel symbol names to human readable form (for C++ kernels).
0364
0365 --header
0366 Show perf.data header.
0367
0368 --header-only
0369 Show only perf.data header.
0370
0371 --itrace::
0372 Options for decoding instruction tracing data. The options are:
0373
0374 include::itrace.txt[]
0375
0376 To disable decoding entirely, use --no-itrace.
0377
0378 --full-source-path::
0379 Show the full path for source files for srcline output.
0380
0381 --max-stack::
0382 Set the stack depth limit when parsing the callchain, anything
0383 beyond the specified depth will be ignored. This is a trade-off
0384 between information loss and faster processing especially for
0385 workloads that can have a very long callchain stack.
0386 Note that when using the --itrace option the synthesized callchain size
0387 will override this value if the synthesized callchain size is bigger.
0388
0389 Default: 127
0390
0391 --ns::
0392 Use 9 decimal places when displaying time (i.e. show the nanoseconds)
0393
0394 -f::
0395 --force::
0396 Don't do ownership validation.
0397
0398 --time::
0399 Only analyze samples within given time window: <start>,<stop>. Times
0400 have the format seconds.nanoseconds. If start is not given (i.e. time
0401 string is ',x.y') then analysis starts at the beginning of the file. If
0402 stop time is not given (i.e. time string is 'x.y,') then analysis goes
0403 to end of file. Multiple ranges can be separated by spaces, which
0404 requires the argument to be quoted e.g. --time "1234.567,1234.789 1235,"
0405
0406 Also support time percent with multiple time ranges. Time string is
0407 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
0408
0409 For example:
0410 Select the second 10% time slice:
0411 perf script --time 10%/2
0412
0413 Select from 0% to 10% time slice:
0414 perf script --time 0%-10%
0415
0416 Select the first and second 10% time slices:
0417 perf script --time 10%/1,10%/2
0418
0419 Select from 0% to 10% and 30% to 40% slices:
0420 perf script --time 0%-10%,30%-40%
0421
0422 --max-blocks::
0423 Set the maximum number of program blocks to print with brstackinsn for
0424 each sample.
0425
0426 --reltime::
0427 Print time stamps relative to trace start.
0428
0429 --deltatime::
0430 Print time stamps relative to previous event.
0431
0432 --per-event-dump::
0433 Create per event files with a "perf.data.EVENT.dump" name instead of
0434 printing to stdout, useful, for instance, for generating flamegraphs.
0435
0436 --inline::
0437 If a callgraph address belongs to an inlined function, the inline stack
0438 will be printed. Each entry has function name and file/line. Enabled by
0439 default, disable with --no-inline.
0440
0441 --insn-trace::
0442 Show instruction stream for intel_pt traces. Combine with --xed to
0443 show disassembly.
0444
0445 --xed::
0446 Run xed disassembler on output. Requires installing the xed disassembler.
0447
0448 -S::
0449 --symbols=symbol[,symbol...]::
0450 Only consider the listed symbols. Symbols are typically a name
0451 but they may also be hexadecimal address.
0452
0453 The hexadecimal address may be the start address of a symbol or
0454 any other address to filter the trace records
0455
0456 For example, to select the symbol noploop or the address 0x4007a0:
0457 perf script --symbols=noploop,0x4007a0
0458
0459 Support filtering trace records by symbol name, start address of
0460 symbol, any hexadecimal address and address range.
0461
0462 The comparison order is:
0463
0464 1. symbol name comparison
0465 2. symbol start address comparison.
0466 3. any hexadecimal address comparison.
0467 4. address range comparison (see --addr-range).
0468
0469 --addr-range::
0470 Use with -S or --symbols to list traced records within address range.
0471
0472 For example, to list the traced records within the address range
0473 [0x4007a0, 0x0x4007a9]:
0474 perf script -S 0x4007a0 --addr-range 10
0475
0476 --dsos=::
0477 Only consider symbols in these DSOs.
0478
0479 --call-trace::
0480 Show call stream for intel_pt traces. The CPUs are interleaved, but
0481 can be filtered with -C.
0482
0483 --call-ret-trace::
0484 Show call and return stream for intel_pt traces.
0485
0486 --graph-function::
0487 For itrace only show specified functions and their callees for
0488 itrace. Multiple functions can be separated by comma.
0489
0490 --switch-on EVENT_NAME::
0491 Only consider events after this event is found.
0492
0493 --switch-off EVENT_NAME::
0494 Stop considering events after this event is found.
0495
0496 --show-on-off-events::
0497 Show the --switch-on/off events too.
0498
0499 --stitch-lbr::
0500 Show callgraph with stitched LBRs, which may have more complete
0501 callgraph. The perf.data file must have been obtained using
0502 perf record --call-graph lbr.
0503 Disabled by default. In common cases with call stack overflows,
0504 it can recreate better call stacks than the default lbr call stack
0505 output. But this approach is not full proof. There can be cases
0506 where it creates incorrect call stacks from incorrect matches.
0507 The known limitations include exception handing such as
0508 setjmp/longjmp will have calls/returns not match.
0509
0510 :GMEXAMPLECMD: script
0511 :GMEXAMPLESUBCMD:
0512 include::guest-files.txt[]
0513
0514 SEE ALSO
0515 --------
0516 linkperf:perf-record[1], linkperf:perf-script-perl[1],
0517 linkperf:perf-script-python[1], linkperf:perf-intel-pt[1],
0518 linkperf:perf-dlfilter[1]
