1PERF-REPORT(1) perf Manual PERF-REPORT(1)
2
6 perf-report - Read perf.data (created by perf record) and display the
7 profile
8
10 perf report [-i <file> | --input=file]
11
13 This command displays the performance counter profile information
14 recorded via perf record.
15
17 -i, --input=
18 Input file name. (default: perf.data unless stdin is a fifo)
19 -v, --verbose
21 Be more verbose. (show symbol address, etc)
22 -q, --quiet
24 Do not show any warnings or messages. (Suppress -v)
25 -n, --show-nr-samples
27 Show the number of samples for each symbol
28 --show-cpu-utilization
30 Show sample percentage for different cpu modes.
31 -T, --threads
33 Show per-thread event counters. The input data file should be
34 recorded with -s option.
35 -c, --comms=
37 Only consider symbols in these comms. CSV that understands
38 file://filename entries. This option will affect the percentage of
39 the overhead column. See --percentage for more info.
40 --pid=
42 Only show events for given process ID (comma separated list).
43 --tid=
45 Only show events for given thread ID (comma separated list).
46 -d, --dsos=
48 Only consider symbols in these dsos. CSV that understands
49 file://filename entries. This option will affect the percentage of
50 the overhead column. See --percentage for more info.
51 -S, --symbols=
53 Only consider these symbols. CSV that understands file://filename
54 entries. This option will affect the percentage of the overhead
55 column. See --percentage for more info.
56 --symbol-filter=
58 Only show symbols that match (partially) with this filter.
59 -U, --hide-unresolved
61 Only display entries resolved to a symbol.
62 -s, --sort=
64 Sort histogram entries by given key(s) - multiple keys can be
65 specified in CSV format. Following sort keys are available: pid,
66 comm, dso, symbol, parent, cpu, socket, srcline, weight,
67 local_weight, cgroup_id, addr.
68 Each key has following meaning:
70 • comm: command (name) of the task which can be read via
72 /proc/<pid>/comm
73 • pid: command and tid of the task
75 • dso: name of library or module executed at the time of sample
77 • dso_size: size of library or module executed at the time of
79 sample
80 • symbol: name of function executed at the time of sample
82 • symbol_size: size of function executed at the time of sample
84 • parent: name of function matched to the parent regex filter.
86 Unmatched entries are displayed as "[other]".
87 • cpu: cpu number the task ran at the time of sample
89 • socket: processor socket number the task ran at the time of
91 sample
92 • srcline: filename and line number executed at the time of
94 sample. The DWARF debugging info must be provided.
95 • srcfile: file name of the source file of the samples. Requires
97 dwarf information.
98 • weight: Event specific weight, e.g. memory latency or
100 transaction abort cost. This is the global weight.
101 • local_weight: Local weight version of the weight above.
103 • cgroup_id: ID derived from cgroup namespace device and inode
105 numbers.
106 • cgroup: cgroup pathname in the cgroupfs.
108 • transaction: Transaction abort flags.
110 • overhead: Overhead percentage of sample
112 • overhead_sys: Overhead percentage of sample running in system
114 mode
115 • overhead_us: Overhead percentage of sample running in user mode
117 • overhead_guest_sys: Overhead percentage of sample running in
119 system mode on guest machine
120 • overhead_guest_us: Overhead percentage of sample running in
122 user mode on guest machine
123 • sample: Number of sample
125 • period: Raw number of event count of sample
127 • time: Separate the samples by time stamp with the resolution
129 specified by --time-quantum (default 100ms). Specify with
130 overhead and before it.
131 • code_page_size: the code page size of sampled code address (ip)
133 • ins_lat: Instruction latency in core cycles. This is the global
135 instruction latency
136 • local_ins_lat: Local instruction latency version
138 • p_stage_cyc: On powerpc, this presents the number of cycles
140 spent in a pipeline stage. And currently supported only on
141 powerpc.
142 • addr: (Full) virtual address of the sampled instruction
144 • retire_lat: On X86, this reports pipeline stall of this
146 instruction compared to the previous instruction in cycles. And
147 currently supported only on X86
148 • simd: Flags describing a SIMD operation. "e" for empty Arm SVE
150 predicate. "p" for partial Arm SVE predicate
151 By default, comm, dso and symbol keys are used.
153 (i.e. --sort comm,dso,symbol)
154 If --branch-stack option is used, following sort keys are also
156 available:
157 • dso_from: name of library or module branched from
159 • dso_to: name of library or module branched to
161 • symbol_from: name of function branched from
163 • symbol_to: name of function branched to
165 • srcline_from: source file and line branched from
167 • srcline_to: source file and line branched to
169 • mispredict: "N" for predicted branch, "Y" for mispredicted
171 branch
172 • in_tx: branch in TSX transaction
174 • abort: TSX transaction abort.
176 • cycles: Cycles in basic block
178 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
180 and symbol_to, see '--branch-stack'.
181 When the sort key symbol is specified, columns "IPC" and "IPC Coverage"
183 are enabled automatically. Column "IPC" reports the average IPC per function
184 and column "IPC coverage" reports the percentage of instructions with
185 sampled IPC in this function. IPC means Instruction Per Cycle. If it's low,
186 it indicates there may be a performance bottleneck when the function is
187 executed, such as a memory access bottleneck. If a function has high overhead
188 and low IPC, it's worth further analyzing it to optimize its performance.
189 If the --mem-mode option is used, the following sort keys are also available
191 (incompatible with --branch-stack):
192 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline, blocked.
193 • symbol_daddr: name of data symbol being executed on at the time
195 of sample
196 • dso_daddr: name of library or module containing the data being
198 executed on at the time of the sample
199 • locked: whether the bus was locked at the time of the sample
201 • tlb: type of tlb access for the data at the time of the sample
203 • mem: type of memory access for the data at the time of the
205 sample
206 • snoop: type of snoop (if any) for the data at the time of the
208 sample
209 • dcacheline: the cacheline the data address is on at the time of
211 the sample
212 • phys_daddr: physical address of data being executed on at the
214 time of sample
215 • data_page_size: the data page size of data being executed on at
217 the time of sample
218 • blocked: reason of blocked load access for the data at the time
220 of the sample
221 And the default sort keys are changed to local_weight, mem, sym, dso,
223 symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, local_ins_lat,
224 see '--mem-mode'.
225 If the data file has tracepoint event(s), following (dynamic) sort keys
227 are also available:
228 trace, trace_fields, [<event>.]<field>[/raw]
229 • trace: pretty printed trace output in a single column
231 • trace_fields: fields in tracepoints in separate columns
233 • <field name>: optional event and field name for a specific
235 field
236 The last form consists of event and field names. If event name is
238 omitted, it searches all events for matching field name. The matched
239 field will be shown only for the event has the field. The event name
240 supports substring match so user doesn't need to specify full subsystem
241 and event name everytime. For example, 'sched:sched_switch' event can
242 be shortened to 'switch' as long as it's not ambiguous. Also event can
243 be specified by its index (starting from 1) preceded by the '%'.
244 So '%1' is the first event, '%2' is the second, and so on.
245 The field name can have '/raw' suffix which disables pretty printing
247 and shows raw field value like hex numbers. The --raw-trace option
248 has the same effect for all dynamic sort keys.
249 The default sort keys are changed to 'trace' if all events in the data
251 file are tracepoint.
252 -F, --fields=
254 Specify output field - multiple keys can be specified in CSV
255 format. Following fields are available: overhead, overhead_sys,
256 overhead_us, overhead_children, sample and period. Also it can
257 contain any sort key(s).
258 By default, every sort keys not specified in -F will be appended
260 automatically.
261 If the keys starts with a prefix '+', then it will append the specified
263 field(s) to the default field order. For example: perf report -F +period,sample.
264 -p, --parent=<regex>
266 A regex filter to identify parent. The parent is a caller of this
267 function and searched through the callchain, thus it requires
268 callchain information recorded. The pattern is in the extended
269 regex format and defaults to "^sys_|^do_page_fault", see --sort
270 parent.
271 -x, --exclude-other
273 Only display entries with parent-match.
274 -w, --column-widths=<width[,width...]>
276 Force each column width to the provided list, for large terminal
277 readability. 0 means no limit (default behavior).
278 -t, --field-separator=
280 Use a special separator character and don’t pad with spaces,
281 replacing all occurrences of this separator in symbol names (and
282 other output) with a . character, that thus it’s the only non
283 valid separator.
284 -D, --dump-raw-trace
286 Dump raw trace in ASCII.
287 --disable-order
289 Disable raw trace ordering.
290 -g,
292 --call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>
293 Display call chains using type, min percent threshold, print limit,
294 call order, sort key, optional branch and value. Note that ordering
295 is not fixed so any parameter can be given in an arbitrary order.
296 One exception is the print_limit which should be preceded by
297 threshold.
298 print_type can be either:
300 - flat: single column, linear exposure of call chains.
301 - graph: use a graph tree, displaying absolute overhead rates. (default)
302 - fractal: like graph, but displays relative rates. Each branch of
303 the tree is considered as a new profiled object.
304 - folded: call chains are displayed in a line, separated by semicolons
305 - none: disable call chain display.
306 threshold is a percentage value which specifies a minimum percent to be
308 included in the output call graph. Default is 0.5 (%).
309 print_limit is only applied when stdio interface is used. It's to limit
311 number of call graph entries in a single hist entry. Note that it needs
312 to be given after threshold (but not necessarily consecutive).
313 Default is 0 (unlimited).
314 order can be either:
316 - callee: callee based call graph.
317 - caller: inverted caller based call graph.
318 Default is 'caller' when --children is used, otherwise 'callee'.
319 sort_key can be:
321 - function: compare on functions (default)
322 - address: compare on individual code addresses
323 - srcline: compare on source filename and line number
324 branch can be:
326 - branch: include last branch information in callgraph when available.
327 Usually more convenient to use --branch-history for this.
328 value can be:
330 - percent: display overhead percent (default)
331 - period: display event period
332 - count: display event count
333 --children
335 Accumulate callchain of children to parent entry so that then can
336 show up in the output. The output will have a new "Children" column
337 and will be sorted on the data. It requires callchains are
338 recorded. See the ‘overhead calculation’ section for more details.
339 Enabled by default, disable with --no-children.
340 --max-stack
342 Set the stack depth limit when parsing the callchain, anything
343 beyond the specified depth will be ignored. This is a trade-off
344 between information loss and faster processing especially for
345 workloads that can have a very long callchain stack. Note that when
346 using the --itrace option the synthesized callchain size will
347 override this value if the synthesized callchain size is bigger.
348 Default: 127
350 -G, --inverted
352 alias for inverted caller based call graph.
353 --ignore-callees=<regex>
355 Ignore callees of the function(s) matching the given regex. This
356 has the effect of collecting the callers of each such function into
357 one place in the call-graph tree.
358 --pretty=<key>
360 Pretty printing style. key: normal, raw
361 --stdio
363 Use the stdio interface.
364 --stdio-color
366 always, never or auto, allowing configuring color output via the
367 command line, in addition to via "color.ui" .perfconfig. Use
368 --stdio-color always to generate color even when redirecting to a
369 pipe or file. Using just --stdio-color is equivalent to using
370 always.
371 --tui
373 Use the TUI interface, that is integrated with annotate and allows
374 zooming into DSOs or threads, among other features. Use of --tui
375 requires a tty, if one is not present, as when piping to other
376 commands, the stdio interface is used.
377 --gtk
379 Use the GTK2 interface.
380 -k, --vmlinux=<file>
382 vmlinux pathname
383 --ignore-vmlinux
385 Ignore vmlinux files.
386 --kallsyms=<file>
388 kallsyms pathname
389 -m, --modules
391 Load module symbols. WARNING: This should only be used with -k and
392 a LIVE kernel.
393 -f, --force
395 Don’t do ownership validation.
396 --symfs=<directory>
398 Look for files with symbols relative to this directory.
399 -C, --cpu
401 Only report samples for the list of CPUs provided. Multiple CPUs
402 can be provided as a comma-separated list with no space: 0,1.
403 Ranges of CPUs are specified with -: 0-2. Default is to report
404 samples on all CPUs.
405 -M, --disassembler-style=
407 Set disassembler style for objdump.
408 --source
410 Interleave source code with assembly code. Enabled by default,
411 disable with --no-source.
412 --asm-raw
414 Show raw instruction encoding of assembly instructions.
415 --show-total-period
417 Show a column with the sum of periods.
418 -I, --show-info
420 Display extended information about the perf.data file. This adds
421 information which may be very large and thus may clutter the
422 display. It currently includes: cpu and numa topology of the host
423 system.
424 -b, --branch-stack
426 Use the addresses of sampled taken branches instead of the
427 instruction address to build the histograms. To generate meaningful
428 output, the perf.data file must have been obtained using perf
429 record -b or perf record --branch-filter xxx where xxx is a branch
430 filter option. perf report is able to auto-detect whether a
431 perf.data file contains branch stacks and it will automatically
432 switch to the branch view mode, unless --no-branch-stack is used.
433 --branch-history
435 Add the addresses of sampled taken branches to the callstack. This
436 allows to examine the path the program took to each sample. The
437 data collection must have used -b (or -j) and -g.
438 --addr2line=<path>
440 Path to addr2line binary.
441 --objdump=<path>
443 Path to objdump binary.
444 --prefix=PREFIX, --prefix-strip=N
446 Remove first N entries from source file path names in executables
447 and add PREFIX. This allows to display source code compiled on
448 systems with different file system layout.
449 --group
451 Show event group information together. It forces group output also
452 if there are no groups defined in data file.
453 --group-sort-idx
455 Sort the output by the event at the index n in group. If n is
456 invalid, sort by the first event. It can support multiple groups
457 with different amount of events. WARNING: This should be used on
458 grouped events.
459 --demangle
461 Demangle symbol names to human readable form. It’s enabled by
462 default, disable with --no-demangle.
463 --demangle-kernel
465 Demangle kernel symbol names to human readable form (for C++
466 kernels).
467 --mem-mode
469 Use the data addresses of samples in addition to instruction
470 addresses to build the histograms. To generate meaningful output,
471 the perf.data file must have been obtained using perf record -d -W
472 and using a special event -e cpu/mem-loads/p or -e
473 cpu/mem-stores/p. See perf mem for simpler access.
474 --percent-limit
476 Do not show entries which have an overhead under that percent.
477 (Default: 0). Note that this option also sets the percent limit
478 (threshold) of callchains. However the default value of callchain
479 threshold is different than the default value of hist entries.
480 Please see the --call-graph option for details.
481 --percentage
483 Determine how to display the overhead percentage of filtered
484 entries. Filters can be applied by --comms, --dsos and/or --symbols
485 options and Zoom operations on the TUI (thread, dso, etc).
486 "relative" means it's relative to filtered entries only so that the
488 sum of shown entries will be always 100%. "absolute" means it retains
489 the original value before and after the filter is applied.
490 --header
492 Show header information in the perf.data file. This includes
493 various information like hostname, OS and perf version, cpu/mem
494 info, perf command line, event list and so on. Currently only
495 --stdio output supports this feature.
496 --header-only
498 Show only perf.data header (forces --stdio).
499 --time
501 Only analyze samples within given time window: <start>,<stop>.
502 Times have the format seconds.nanoseconds. If start is not given
503 (i.e. time string is ,x.y) then analysis starts at the beginning of
504 the file. If stop time is not given (i.e. time string is x.y,) then
505 analysis goes to end of file. Multiple ranges can be separated by
506 spaces, which requires the argument to be quoted e.g. --time
507 "1234.567,1234.789 1235,"
508 Also support time percent with multiple time ranges. Time string is
510 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
511 For example:
513 Select the second 10% time slice:
514 perf report --time 10%/2
516 Select from 0% to 10% time slice:
518 perf report --time 0%-10%
520 Select the first and second 10% time slices:
522 perf report --time 10%/1,10%/2
524 Select from 0% to 10% and 30% to 40% slices:
526 perf report --time 0%-10%,30%-40%
528 --switch-on EVENT_NAME
530 Only consider events after this event is found.
531 This may be interesting to measure a workload only after some initialization
533 phase is over, i.e. insert a perf probe at that point and then using this
534 option with that probe.
535 --switch-off EVENT_NAME
537 Stop considering events after this event is found.
538 --show-on-off-events
540 Show the --switch-on/off events too. This has no effect in perf
541 report now but probably we’ll make the default not to show the
542 switch-on/off events on the --group mode and if there is only one
543 event besides the off/on ones, go straight to the histogram
544 browser, just like perf report with no events explicitly specified
545 does.
546 --itrace
548 Options for decoding instruction tracing data. The options are:
549 i synthesize instructions events
551 y synthesize cycles events
552 b synthesize branches events (branch misses for Arm SPE)
553 c synthesize branches events (calls only)
554 r synthesize branches events (returns only)
555 x synthesize transactions events
556 w synthesize ptwrite events
557 p synthesize power events (incl. PSB events for Intel PT)
558 o synthesize other events recorded due to the use
559 of aux-output (refer to perf record)
560 I synthesize interrupt or similar (asynchronous) events
561 (e.g. Intel PT Event Trace)
562 e synthesize error events
563 d create a debug log
564 f synthesize first level cache events
565 m synthesize last level cache events
566 M synthesize memory events
567 t synthesize TLB events
568 a synthesize remote access events
569 g synthesize a call chain (use with i or x)
570 G synthesize a call chain on existing event records
571 l synthesize last branch entries (use with i or x)
572 L synthesize last branch entries on existing event records
573 s skip initial number of events
574 q quicker (less detailed) decoding
575 A approximate IPC
576 Z prefer to ignore timestamps (so-called "timeless" decoding)
577 The default is all events i.e. the same as --itrace=iybxwpe,
579 except for perf script where it is --itrace=ce
580 In addition, the period (default 100000, except for perf script where it is 1)
582 for instructions events can be specified in units of:
583 i instructions
585 t ticks
586 ms milliseconds
587 us microseconds
588 ns nanoseconds (default)
589 Also the call chain size (default 16, max. 1024) for instructions or
591 transactions events can be specified.
592 Also the number of last branch entries (default 64, max. 1024) for
594 instructions or transactions events can be specified.
595 Similar to options g and l, size may also be specified for options G and L.
597 On x86, note that G and L work poorly when data has been recorded with
598 large PEBS. Refer linkperf:perf-intel-pt[1] man page for details.
599 It is also possible to skip events generated (instructions, branches, transactions,
601 ptwrite, power) at the beginning. This is useful to ignore initialization code.
602 --itrace=i0nss1000000
604 skips the first million instructions.
606 The 'e' option may be followed by flags which affect what errors will or
608 will not be reported. Each flag must be preceded by either '+' or '-'.
609 The flags are:
610 o overflow
611 l trace data lost
612 If supported, the 'd' option may be followed by flags which affect what
614 debug messages will or will not be logged. Each flag must be preceded
615 by either '+' or '-'. The flags are:
616 a all perf events
617 e output only on errors (size configurable - see linkperf:perf-config[1])
618 o output to stdout
619 If supported, the 'q' option may be repeated to increase the effect.
621 To disable decoding entirely, use --no-itrace.
623 --full-source-path
625 Show the full path for source files for srcline output.
626 --show-ref-call-graph
628 When multiple events are sampled, it may not be needed to collect
629 callgraphs for all of them. The sample sites are usually nearby,
630 and it’s enough to collect the callgraphs on a reference event. So
631 user can use "call-graph=no" event modifier to disable callgraph
632 for other events to reduce the overhead. However, perf report
633 cannot show callgraphs for the event which disable the callgraph.
634 This option extends the perf report to show reference callgraphs,
635 which collected by reference event, in no callgraph event.
636 --stitch-lbr
638 Show callgraph with stitched LBRs, which may have more complete
639 callgraph. The perf.data file must have been obtained using perf
640 record --call-graph lbr. Disabled by default. In common cases with
641 call stack overflows, it can recreate better call stacks than the
642 default lbr call stack output. But this approach is not foolproof.
643 There can be cases where it creates incorrect call stacks from
644 incorrect matches. The known limitations include exception handing
645 such as setjmp/longjmp will have calls/returns not match.
646 --socket-filter
648 Only report the samples on the processor socket that match with
649 this filter
650 --samples=N
652 Save N individual samples for each histogram entry to show context
653 in perf report tui browser.
654 --raw-trace
656 When displaying traceevent output, do not use print fmt or plugins.
657 --hierarchy
659 Enable hierarchical output.
660 --inline
662 If a callgraph address belongs to an inlined function, the inline
663 stack will be printed. Each entry is function name or file/line.
664 Enabled by default, disable with --no-inline.
665 --mmaps
667 Show --tasks output plus mmap information in a format similar to
668 /proc/<PID>/maps.
669 Please note that not all mmaps are stored, options affecting which ones
671 are include 'perf record --data', for instance.
672 --ns
674 Show time stamps in nanoseconds.
675 --stats
677 Display overall events statistics without any further processing.
678 (like the one at the end of the perf report -D command)
679 --tasks
681 Display monitored tasks stored in perf data. Displaying
682 pid/tid/ppid plus the command string aligned to distinguish parent
683 and child tasks.
684 --percent-type
686 Set annotation percent type from following choices: global-period,
687 local-period, global-hits, local-hits
688 The local/global keywords set if the percentage is computed
690 in the scope of the function (local) or the whole data (global).
691 The period/hits keywords set the base the percentage is computed
692 on - the samples period or the number of samples (hits).
693 --time-quantum
695 Configure time quantum for time sort key. Default 100ms. Accepts s,
696 us, ms, ns units.
697 --total-cycles
699 When --total-cycles is specified, it supports sorting for all
700 blocks by Sampled Cycles%. This is useful to concentrate on the
701 globally hottest blocks. In output, there are some new columns:
702 'Sampled Cycles%' - block sampled cycles aggregation / total sampled cycles
704 'Sampled Cycles' - block sampled cycles aggregation
705 'Avg Cycles%' - block average sampled cycles / sum of total block average
706 sampled cycles
707 'Avg Cycles' - block average sampled cycles
708 --skip-empty
710 Do not print 0 results in the --stat output.
711
713 The overhead can be shown in two columns as Children and Self when perf
714 collects callchains. The self overhead is simply calculated by adding
715 all period values of the entry - usually a function (symbol). This is
716 the value that perf shows traditionally and sum of all the self
717 overhead values should be 100%.
718 The children overhead is calculated by adding all period values of the
720 child functions so that it can show the total overhead of the higher
721 level functions even if they don’t directly execute much. Children here
722 means functions that are called from another (parent) function.
723 It might be confusing that the sum of all the children overhead values
725 exceeds 100% since each of them is already an accumulation of self
726 overhead of its child functions. But with this enabled, users can find
727 which function has the most overhead even if samples are spread over
728 the children.
729 Consider the following example; there are three functions like below.
731 .ft C
734 void foo(void) {
735 /* do something */
736 }
737 void bar(void) {
739 /* do something */
740 foo();
741 }
742 int main(void) {
744 bar()
745 return 0;
746 }
747 .ft
748 In this case foo is a child of bar, and bar is an immediate child of
751 main so foo also is a child of main. In other words, main is a parent
752 of foo and bar, and bar is a parent of foo.
753 Suppose all samples are recorded in foo and bar only. When it’s
755 recorded with callchains the output will show something like below in
756 the usual (self-overhead-only) output of perf report:
757 .ft C
760 Overhead Symbol
761 ........ .....................
762 60.00% foo
763 |
764 --- foo
765 bar
766 main
767 __libc_start_main
768 40.00% bar
770 |
771 --- bar
772 main
773 __libc_start_main
774 .ft
775 When the --children option is enabled, the self overhead values of
778 child functions (i.e. foo and bar) are added to the parents to
779 calculate the children overhead. In this case the report could be
780 displayed as:
781 .ft C
784 Children Self Symbol
785 ........ ........ ....................
786 100.00% 0.00% __libc_start_main
787 |
788 --- __libc_start_main
789 100.00% 0.00% main
791 |
792 --- main
793 __libc_start_main
794 100.00% 40.00% bar
796 |
797 --- bar
798 main
799 __libc_start_main
800 60.00% 60.00% foo
802 |
803 --- foo
804 bar
805 main
806 __libc_start_main
807 .ft
808 In the above output, the self overhead of foo (60%) was add to the
811 children overhead of bar, main and __libc_start_main. Likewise, the
812 self overhead of bar (40%) was added to the children overhead of main
813 and \_\_libc_start_main.
814 So \_\_libc_start_main and main are shown first since they have same
816 (100%) children overhead (even though they have zero self overhead) and
817 they are the parents of foo and bar.
818 Since v3.16 the children overhead is shown by default and the output is
820 sorted by its values. The children overhead is disabled by specifying
821 --no-children option on the command line or by adding report.children =
822 false or top.children = false in the perf config file.
823
825 perf-stat(1), perf-annotate(1), perf-record(1), perf-intel-pt(1)
826perf 11/28/2023 PERF-REPORT(1)