6 perf-report - Read perf.data (created by perf record) and display the profile
11 'perf report' [-i <file> | --input=file]
15 This command displays the performance counter profile information recorded
22 Input file name. (default: perf.data unless stdin is a fifo)
26 Be more verbose. (show symbol address, etc)
30 Do not show any warnings or messages. (Suppress -v)
34 Show the number of samples for each symbol
36 --show-cpu-utilization::
37 Show sample percentage for different cpu modes.
41 Show per-thread event counters. The input data file should be recorded
45 Only consider symbols in these comms. CSV that understands
46 file://filename entries. This option will affect the percentage of
47 the overhead column. See --percentage for more info.
49 Only show events for given process ID (comma separated list).
52 Only show events for given thread ID (comma separated list).
55 Only consider symbols in these dsos. CSV that understands
56 file://filename entries. This option will affect the percentage of
57 the overhead column. See --percentage for more info.
60 Only consider these symbols. CSV that understands
61 file://filename entries. This option will affect the percentage of
62 the overhead column. See --percentage for more info.
65 Only show symbols that match (partially) with this filter.
69 Only display entries resolved to a symbol.
73 Sort histogram entries by given key(s) - multiple keys can be specified
74 in CSV format. Following sort keys are available:
75 pid, comm, dso, symbol, parent, cpu, socket, srcline, weight,
76 local_weight, cgroup_id, addr.
78 Each key has following meaning:
80 - comm: command (name) of the task which can be read via /proc/<pid>/comm
81 - pid: command and tid of the task
82 - dso: name of library or module executed at the time of sample
83 - dso_size: size of library or module executed at the time of sample
84 - symbol: name of function executed at the time of sample
85 - symbol_size: size of function executed at the time of sample
86 - parent: name of function matched to the parent regex filter. Unmatched
87 entries are displayed as "[other]".
88 - cpu: cpu number the task ran at the time of sample
89 - socket: processor socket number the task ran at the time of sample
90 - srcline: filename and line number executed at the time of sample. The
91 DWARF debugging info must be provided.
92 - srcfile: file name of the source file of the samples. Requires dwarf
94 - weight: Event specific weight, e.g. memory latency or transaction
95 abort cost. This is the global weight.
96 - local_weight: Local weight version of the weight above.
97 - cgroup_id: ID derived from cgroup namespace device and inode numbers.
98 - cgroup: cgroup pathname in the cgroupfs.
99 - transaction: Transaction abort flags.
100 - overhead: Overhead percentage of sample
101 - overhead_sys: Overhead percentage of sample running in system mode
102 - overhead_us: Overhead percentage of sample running in user mode
103 - overhead_guest_sys: Overhead percentage of sample running in system mode
105 - overhead_guest_us: Overhead percentage of sample running in user mode on
107 - sample: Number of sample
108 - period: Raw number of event count of sample
109 - time: Separate the samples by time stamp with the resolution specified by
110 --time-quantum (default 100ms). Specify with overhead and before it.
111 - code_page_size: the code page size of sampled code address (ip)
112 - ins_lat: Instruction latency in core cycles. This is the global instruction
114 - local_ins_lat: Local instruction latency version
115 - p_stage_cyc: On powerpc, this presents the number of cycles spent in a
116 pipeline stage. And currently supported only on powerpc.
117 - addr: (Full) virtual address of the sampled instruction
118 - retire_lat: On X86, this reports pipeline stall of this instruction compared
119 to the previous instruction in cycles. And currently supported only on X86
120 - simd: Flags describing a SIMD operation. "e" for empty Arm SVE predicate. "p" for partial Arm SVE predicate
121 - type: Data type of sample memory access.
122 - typeoff: Offset in the data type of sample memory access.
123 - symoff: Offset in the symbol.
125 By default, comm, dso and symbol keys are used.
126 (i.e. --sort comm,dso,symbol)
128 If --branch-stack option is used, following sort keys are also
131 - dso_from: name of library or module branched from
132 - dso_to: name of library or module branched to
133 - symbol_from: name of function branched from
134 - symbol_to: name of function branched to
135 - srcline_from: source file and line branched from
136 - srcline_to: source file and line branched to
137 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
138 - in_tx: branch in TSX transaction
139 - abort: TSX transaction abort.
140 - cycles: Cycles in basic block
142 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
143 and symbol_to, see '--branch-stack'.
145 When the sort key symbol is specified, columns "IPC" and "IPC Coverage"
146 are enabled automatically. Column "IPC" reports the average IPC per function
147 and column "IPC coverage" reports the percentage of instructions with
148 sampled IPC in this function. IPC means Instruction Per Cycle. If it's low,
149 it indicates there may be a performance bottleneck when the function is
150 executed, such as a memory access bottleneck. If a function has high overhead
151 and low IPC, it's worth further analyzing it to optimize its performance.
153 If the --mem-mode option is used, the following sort keys are also available
154 (incompatible with --branch-stack):
155 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline, blocked.
157 - symbol_daddr: name of data symbol being executed on at the time of sample
158 - dso_daddr: name of library or module containing the data being executed
159 on at the time of the sample
160 - locked: whether the bus was locked at the time of the sample
161 - tlb: type of tlb access for the data at the time of the sample
162 - mem: type of memory access for the data at the time of the sample
163 - snoop: type of snoop (if any) for the data at the time of the sample
164 - dcacheline: the cacheline the data address is on at the time of the sample
165 - phys_daddr: physical address of data being executed on at the time of sample
166 - data_page_size: the data page size of data being executed on at the time of sample
167 - blocked: reason of blocked load access for the data at the time of the sample
169 And the default sort keys are changed to local_weight, mem, sym, dso,
170 symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, local_ins_lat,
173 If the data file has tracepoint event(s), following (dynamic) sort keys
175 trace, trace_fields, [<event>.]<field>[/raw]
177 - trace: pretty printed trace output in a single column
178 - trace_fields: fields in tracepoints in separate columns
179 - <field name>: optional event and field name for a specific field
181 The last form consists of event and field names. If event name is
182 omitted, it searches all events for matching field name. The matched
183 field will be shown only for the event has the field. The event name
184 supports substring match so user doesn't need to specify full subsystem
185 and event name everytime. For example, 'sched:sched_switch' event can
186 be shortened to 'switch' as long as it's not ambiguous. Also event can
187 be specified by its index (starting from 1) preceded by the '%'.
188 So '%1' is the first event, '%2' is the second, and so on.
190 The field name can have '/raw' suffix which disables pretty printing
191 and shows raw field value like hex numbers. The --raw-trace option
192 has the same effect for all dynamic sort keys.
194 The default sort keys are changed to 'trace' if all events in the data
199 Specify output field - multiple keys can be specified in CSV format.
200 Following fields are available:
201 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
202 Also it can contain any sort key(s).
204 By default, every sort keys not specified in -F will be appended
207 If the keys starts with a prefix '+', then it will append the specified
208 field(s) to the default field order. For example: perf report -F +period,sample.
212 A regex filter to identify parent. The parent is a caller of this
213 function and searched through the callchain, thus it requires callchain
214 information recorded. The pattern is in the extended regex format and
215 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
219 Only display entries with parent-match.
222 --column-widths=<width[,width...]>::
223 Force each column width to the provided list, for large terminal
224 readability. 0 means no limit (default behavior).
228 Use a special separator character and don't pad with spaces, replacing
229 all occurrences of this separator in symbol names (and other output)
230 with a '.' character, that thus it's the only non valid separator.
234 Dump raw trace in ASCII.
237 Disable raw trace ordering.
240 --call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>::
241 Display call chains using type, min percent threshold, print limit,
242 call order, sort key, optional branch and value. Note that ordering
243 is not fixed so any parameter can be given in an arbitrary order.
244 One exception is the print_limit which should be preceded by threshold.
246 print_type can be either:
247 - flat: single column, linear exposure of call chains.
248 - graph: use a graph tree, displaying absolute overhead rates. (default)
249 - fractal: like graph, but displays relative rates. Each branch of
250 the tree is considered as a new profiled object.
251 - folded: call chains are displayed in a line, separated by semicolons
252 - none: disable call chain display.
254 threshold is a percentage value which specifies a minimum percent to be
255 included in the output call graph. Default is 0.5 (%).
257 print_limit is only applied when stdio interface is used. It's to limit
258 number of call graph entries in a single hist entry. Note that it needs
259 to be given after threshold (but not necessarily consecutive).
260 Default is 0 (unlimited).
263 - callee: callee based call graph.
264 - caller: inverted caller based call graph.
265 Default is 'caller' when --children is used, otherwise 'callee'.
268 - function: compare on functions (default)
269 - address: compare on individual code addresses
270 - srcline: compare on source filename and line number
273 - branch: include last branch information in callgraph when available.
274 Usually more convenient to use --branch-history for this.
277 - percent: display overhead percent (default)
278 - period: display event period
279 - count: display event count
282 Accumulate callchain of children to parent entry so that then can
283 show up in the output. The output will have a new "Children" column
284 and will be sorted on the data. It requires callchains are recorded.
285 See the `overhead calculation' section for more details. Enabled by
286 default, disable with --no-children.
289 Set the stack depth limit when parsing the callchain, anything
290 beyond the specified depth will be ignored. This is a trade-off
291 between information loss and faster processing especially for
292 workloads that can have a very long callchain stack.
293 Note that when using the --itrace option the synthesized callchain size
294 will override this value if the synthesized callchain size is bigger.
300 alias for inverted caller based call graph.
302 --ignore-callees=<regex>::
303 Ignore callees of the function(s) matching the given regex.
304 This has the effect of collecting the callers of each such
305 function into one place in the call-graph tree.
308 Pretty printing style. key: normal, raw
310 --stdio:: Use the stdio interface.
313 'always', 'never' or 'auto', allowing configuring color output
314 via the command line, in addition to via "color.ui" .perfconfig.
315 Use '--stdio-color always' to generate color even when redirecting
316 to a pipe or file. Using just '--stdio-color' is equivalent to
319 --tui:: Use the TUI interface, that is integrated with annotate and allows
320 zooming into DSOs or threads, among other features. Use of --tui
321 requires a tty, if one is not present, as when piping to other
322 commands, the stdio interface is used.
324 --gtk:: Use the GTK2 interface.
331 Ignore vmlinux files.
338 Load module symbols. WARNING: This should only be used with -k and
343 Don't do ownership validation.
345 --symfs=<directory>::
346 Look for files with symbols relative to this directory.
349 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
350 be provided as a comma-separated list with no space: 0,1. Ranges of
351 CPUs are specified with -: 0-2. Default is to report samples on all
355 --disassembler-style=:: Set disassembler style for objdump.
358 Interleave source code with assembly code. Enabled by default,
359 disable with --no-source.
362 Show raw instruction encoding of assembly instructions.
364 --show-total-period:: Show a column with the sum of periods.
368 Display extended information about the perf.data file. This adds
369 information which may be very large and thus may clutter the display.
370 It currently includes: cpu and numa topology of the host system.
374 Use the addresses of sampled taken branches instead of the instruction
375 address to build the histograms. To generate meaningful output, the
376 perf.data file must have been obtained using perf record -b or
377 perf record --branch-filter xxx where xxx is a branch filter option.
378 perf report is able to auto-detect whether a perf.data file contains
379 branch stacks and it will automatically switch to the branch view mode,
380 unless --no-branch-stack is used.
383 Add the addresses of sampled taken branches to the callstack.
384 This allows to examine the path the program took to each sample.
385 The data collection must have used -b (or -j) and -g.
388 Path to addr2line binary.
391 Path to objdump binary.
395 Remove first N entries from source file path names in executables
396 and add PREFIX. This allows to display source code compiled on systems
397 with different file system layout.
400 Show event group information together. It forces group output also
401 if there are no groups defined in data file.
404 Sort the output by the event at the index n in group. If n is invalid,
405 sort by the first event. It can support multiple groups with different
406 amount of events. WARNING: This should be used on grouped events.
409 Demangle symbol names to human readable form. It's enabled by default,
410 disable with --no-demangle.
413 Demangle kernel symbol names to human readable form (for C++ kernels).
416 Use the data addresses of samples in addition to instruction addresses
417 to build the histograms. To generate meaningful output, the perf.data
418 file must have been obtained using perf record -d -W and using a
419 special event -e cpu/mem-loads/p or -e cpu/mem-stores/p. See
420 'perf mem' for simpler access.
423 Do not show entries which have an overhead under that percent.
424 (Default: 0). Note that this option also sets the percent limit (threshold)
425 of callchains. However the default value of callchain threshold is
426 different than the default value of hist entries. Please see the
427 --call-graph option for details.
430 Determine how to display the overhead percentage of filtered entries.
431 Filters can be applied by --comms, --dsos and/or --symbols options and
432 Zoom operations on the TUI (thread, dso, etc).
434 "relative" means it's relative to filtered entries only so that the
435 sum of shown entries will be always 100%. "absolute" means it retains
436 the original value before and after the filter is applied.
439 Show header information in the perf.data file. This includes
440 various information like hostname, OS and perf version, cpu/mem
441 info, perf command line, event list and so on. Currently only
442 --stdio output supports this feature.
445 Show only perf.data header (forces --stdio).
448 Only analyze samples within given time window: <start>,<stop>. Times
449 have the format seconds.nanoseconds. If start is not given (i.e. time
450 string is ',x.y') then analysis starts at the beginning of the file. If
451 stop time is not given (i.e. time string is 'x.y,') then analysis goes
452 to end of file. Multiple ranges can be separated by spaces, which
453 requires the argument to be quoted e.g. --time "1234.567,1234.789 1235,"
455 Also support time percent with multiple time ranges. Time string is
456 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
459 Select the second 10% time slice:
461 perf report --time 10%/2
463 Select from 0% to 10% time slice:
465 perf report --time 0%-10%
467 Select the first and second 10% time slices:
469 perf report --time 10%/1,10%/2
471 Select from 0% to 10% and 30% to 40% slices:
473 perf report --time 0%-10%,30%-40%
475 --switch-on EVENT_NAME::
476 Only consider events after this event is found.
478 This may be interesting to measure a workload only after some initialization
479 phase is over, i.e. insert a perf probe at that point and then using this
480 option with that probe.
482 --switch-off EVENT_NAME::
483 Stop considering events after this event is found.
485 --show-on-off-events::
486 Show the --switch-on/off events too. This has no effect in 'perf report' now
487 but probably we'll make the default not to show the switch-on/off events
488 on the --group mode and if there is only one event besides the off/on ones,
489 go straight to the histogram browser, just like 'perf report' with no events
490 explicitly specified does.
493 Options for decoding instruction tracing data. The options are:
495 include::itrace.txt[]
497 To disable decoding entirely, use --no-itrace.
500 Show the full path for source files for srcline output.
502 --show-ref-call-graph::
503 When multiple events are sampled, it may not be needed to collect
504 callgraphs for all of them. The sample sites are usually nearby,
505 and it's enough to collect the callgraphs on a reference event.
506 So user can use "call-graph=no" event modifier to disable callgraph
507 for other events to reduce the overhead.
508 However, perf report cannot show callgraphs for the event which
509 disable the callgraph.
510 This option extends the perf report to show reference callgraphs,
511 which collected by reference event, in no callgraph event.
514 Show callgraph with stitched LBRs, which may have more complete
515 callgraph. The perf.data file must have been obtained using
516 perf record --call-graph lbr.
517 Disabled by default. In common cases with call stack overflows,
518 it can recreate better call stacks than the default lbr call stack
519 output. But this approach is not foolproof. There can be cases
520 where it creates incorrect call stacks from incorrect matches.
521 The known limitations include exception handing such as
522 setjmp/longjmp will have calls/returns not match.
525 Only report the samples on the processor socket that match with this filter
528 Save N individual samples for each histogram entry to show context in perf
532 When displaying traceevent output, do not use print fmt or plugins.
535 Enable hierarchical output.
538 If a callgraph address belongs to an inlined function, the inline stack
539 will be printed. Each entry is function name or file/line. Enabled by
540 default, disable with --no-inline.
543 Show --tasks output plus mmap information in a format similar to
546 Please note that not all mmaps are stored, options affecting which ones
547 are include 'perf record --data', for instance.
550 Show time stamps in nanoseconds.
553 Display overall events statistics without any further processing.
554 (like the one at the end of the perf report -D command)
557 Display monitored tasks stored in perf data. Displaying pid/tid/ppid
558 plus the command string aligned to distinguish parent and child tasks.
561 Set annotation percent type from following choices:
562 global-period, local-period, global-hits, local-hits
564 The local/global keywords set if the percentage is computed
565 in the scope of the function (local) or the whole data (global).
566 The period/hits keywords set the base the percentage is computed
567 on - the samples period or the number of samples (hits).
570 Configure time quantum for time sort key. Default 100ms.
571 Accepts s, us, ms, ns units.
574 When --total-cycles is specified, it supports sorting for all blocks by
575 'Sampled Cycles%'. This is useful to concentrate on the globally hottest
576 blocks. In output, there are some new columns:
578 'Sampled Cycles%' - block sampled cycles aggregation / total sampled cycles
579 'Sampled Cycles' - block sampled cycles aggregation
580 'Avg Cycles%' - block average sampled cycles / sum of total block average
582 'Avg Cycles' - block average sampled cycles
585 Do not print 0 results in the --stat output.
587 include::callchain-overhead-calculation.txt[]
591 linkperf:perf-stat[1], linkperf:perf-annotate[1], linkperf:perf-record[1],
592 linkperf:perf-intel-pt[1]