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 message. (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.
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.
118 By default, comm, dso and symbol keys are used.
119 (i.e. --sort comm,dso,symbol)
121 If --branch-stack option is used, following sort keys are also
124 - dso_from: name of library or module branched from
125 - dso_to: name of library or module branched to
126 - symbol_from: name of function branched from
127 - symbol_to: name of function branched to
128 - srcline_from: source file and line branched from
129 - srcline_to: source file and line branched to
130 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
131 - in_tx: branch in TSX transaction
132 - abort: TSX transaction abort.
133 - cycles: Cycles in basic block
135 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
136 and symbol_to, see '--branch-stack'.
138 When the sort key symbol is specified, columns "IPC" and "IPC Coverage"
139 are enabled automatically. Column "IPC" reports the average IPC per function
140 and column "IPC coverage" reports the percentage of instructions with
141 sampled IPC in this function. IPC means Instruction Per Cycle. If it's low,
142 it indicates there may be a performance bottleneck when the function is
143 executed, such as a memory access bottleneck. If a function has high overhead
144 and low IPC, it's worth further analyzing it to optimize its performance.
146 If the --mem-mode option is used, the following sort keys are also available
147 (incompatible with --branch-stack):
148 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline, blocked.
150 - symbol_daddr: name of data symbol being executed on at the time of sample
151 - dso_daddr: name of library or module containing the data being executed
152 on at the time of the sample
153 - locked: whether the bus was locked at the time of the sample
154 - tlb: type of tlb access for the data at the time of the sample
155 - mem: type of memory access for the data at the time of the sample
156 - snoop: type of snoop (if any) for the data at the time of the sample
157 - dcacheline: the cacheline the data address is on at the time of the sample
158 - phys_daddr: physical address of data being executed on at the time of sample
159 - data_page_size: the data page size of data being executed on at the time of sample
160 - blocked: reason of blocked load access for the data at the time of the sample
162 And the default sort keys are changed to local_weight, mem, sym, dso,
163 symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, local_ins_lat,
166 If the data file has tracepoint event(s), following (dynamic) sort keys
168 trace, trace_fields, [<event>.]<field>[/raw]
170 - trace: pretty printed trace output in a single column
171 - trace_fields: fields in tracepoints in separate columns
172 - <field name>: optional event and field name for a specific field
174 The last form consists of event and field names. If event name is
175 omitted, it searches all events for matching field name. The matched
176 field will be shown only for the event has the field. The event name
177 supports substring match so user doesn't need to specify full subsystem
178 and event name everytime. For example, 'sched:sched_switch' event can
179 be shortened to 'switch' as long as it's not ambiguous. Also event can
180 be specified by its index (starting from 1) preceded by the '%'.
181 So '%1' is the first event, '%2' is the second, and so on.
183 The field name can have '/raw' suffix which disables pretty printing
184 and shows raw field value like hex numbers. The --raw-trace option
185 has the same effect for all dynamic sort keys.
187 The default sort keys are changed to 'trace' if all events in the data
192 Specify output field - multiple keys can be specified in CSV format.
193 Following fields are available:
194 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
195 Also it can contain any sort key(s).
197 By default, every sort keys not specified in -F will be appended
200 If the keys starts with a prefix '+', then it will append the specified
201 field(s) to the default field order. For example: perf report -F +period,sample.
205 A regex filter to identify parent. The parent is a caller of this
206 function and searched through the callchain, thus it requires callchain
207 information recorded. The pattern is in the extended regex format and
208 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
212 Only display entries with parent-match.
215 --column-widths=<width[,width...]>::
216 Force each column width to the provided list, for large terminal
217 readability. 0 means no limit (default behavior).
221 Use a special separator character and don't pad with spaces, replacing
222 all occurrences of this separator in symbol names (and other output)
223 with a '.' character, that thus it's the only non valid separator.
227 Dump raw trace in ASCII.
230 Disable raw trace ordering.
233 --call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>::
234 Display call chains using type, min percent threshold, print limit,
235 call order, sort key, optional branch and value. Note that ordering
236 is not fixed so any parameter can be given in an arbitrary order.
237 One exception is the print_limit which should be preceded by threshold.
239 print_type can be either:
240 - flat: single column, linear exposure of call chains.
241 - graph: use a graph tree, displaying absolute overhead rates. (default)
242 - fractal: like graph, but displays relative rates. Each branch of
243 the tree is considered as a new profiled object.
244 - folded: call chains are displayed in a line, separated by semicolons
245 - none: disable call chain display.
247 threshold is a percentage value which specifies a minimum percent to be
248 included in the output call graph. Default is 0.5 (%).
250 print_limit is only applied when stdio interface is used. It's to limit
251 number of call graph entries in a single hist entry. Note that it needs
252 to be given after threshold (but not necessarily consecutive).
253 Default is 0 (unlimited).
256 - callee: callee based call graph.
257 - caller: inverted caller based call graph.
258 Default is 'caller' when --children is used, otherwise 'callee'.
261 - function: compare on functions (default)
262 - address: compare on individual code addresses
263 - srcline: compare on source filename and line number
266 - branch: include last branch information in callgraph when available.
267 Usually more convenient to use --branch-history for this.
270 - percent: display overhead percent (default)
271 - period: display event period
272 - count: display event count
275 Accumulate callchain of children to parent entry so that then can
276 show up in the output. The output will have a new "Children" column
277 and will be sorted on the data. It requires callchains are recorded.
278 See the `overhead calculation' section for more details. Enabled by
279 default, disable with --no-children.
282 Set the stack depth limit when parsing the callchain, anything
283 beyond the specified depth will be ignored. This is a trade-off
284 between information loss and faster processing especially for
285 workloads that can have a very long callchain stack.
286 Note that when using the --itrace option the synthesized callchain size
287 will override this value if the synthesized callchain size is bigger.
293 alias for inverted caller based call graph.
295 --ignore-callees=<regex>::
296 Ignore callees of the function(s) matching the given regex.
297 This has the effect of collecting the callers of each such
298 function into one place in the call-graph tree.
301 Pretty printing style. key: normal, raw
303 --stdio:: Use the stdio interface.
306 'always', 'never' or 'auto', allowing configuring color output
307 via the command line, in addition to via "color.ui" .perfconfig.
308 Use '--stdio-color always' to generate color even when redirecting
309 to a pipe or file. Using just '--stdio-color' is equivalent to
312 --tui:: Use the TUI interface, that is integrated with annotate and allows
313 zooming into DSOs or threads, among other features. Use of --tui
314 requires a tty, if one is not present, as when piping to other
315 commands, the stdio interface is used.
317 --gtk:: Use the GTK2 interface.
324 Ignore vmlinux files.
331 Load module symbols. WARNING: This should only be used with -k and
336 Don't do ownership validation.
338 --symfs=<directory>::
339 Look for files with symbols relative to this directory.
342 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
343 be provided as a comma-separated list with no space: 0,1. Ranges of
344 CPUs are specified with -: 0-2. Default is to report samples on all
348 --disassembler-style=:: Set disassembler style for objdump.
351 Interleave source code with assembly code. Enabled by default,
352 disable with --no-source.
355 Show raw instruction encoding of assembly instructions.
357 --show-total-period:: Show a column with the sum of periods.
361 Display extended information about the perf.data file. This adds
362 information which may be very large and thus may clutter the display.
363 It currently includes: cpu and numa topology of the host system.
367 Use the addresses of sampled taken branches instead of the instruction
368 address to build the histograms. To generate meaningful output, the
369 perf.data file must have been obtained using perf record -b or
370 perf record --branch-filter xxx where xxx is a branch filter option.
371 perf report is able to auto-detect whether a perf.data file contains
372 branch stacks and it will automatically switch to the branch view mode,
373 unless --no-branch-stack is used.
376 Add the addresses of sampled taken branches to the callstack.
377 This allows to examine the path the program took to each sample.
378 The data collection must have used -b (or -j) and -g.
381 Path to objdump binary.
385 Remove first N entries from source file path names in executables
386 and add PREFIX. This allows to display source code compiled on systems
387 with different file system layout.
390 Show event group information together. It forces group output also
391 if there are no groups defined in data file.
394 Sort the output by the event at the index n in group. If n is invalid,
395 sort by the first event. It can support multiple groups with different
396 amount of events. WARNING: This should be used on grouped events.
399 Demangle symbol names to human readable form. It's enabled by default,
400 disable with --no-demangle.
403 Demangle kernel symbol names to human readable form (for C++ kernels).
406 Use the data addresses of samples in addition to instruction addresses
407 to build the histograms. To generate meaningful output, the perf.data
408 file must have been obtained using perf record -d -W and using a
409 special event -e cpu/mem-loads/p or -e cpu/mem-stores/p. See
410 'perf mem' for simpler access.
413 Do not show entries which have an overhead under that percent.
414 (Default: 0). Note that this option also sets the percent limit (threshold)
415 of callchains. However the default value of callchain threshold is
416 different than the default value of hist entries. Please see the
417 --call-graph option for details.
420 Determine how to display the overhead percentage of filtered entries.
421 Filters can be applied by --comms, --dsos and/or --symbols options and
422 Zoom operations on the TUI (thread, dso, etc).
424 "relative" means it's relative to filtered entries only so that the
425 sum of shown entries will be always 100%. "absolute" means it retains
426 the original value before and after the filter is applied.
429 Show header information in the perf.data file. This includes
430 various information like hostname, OS and perf version, cpu/mem
431 info, perf command line, event list and so on. Currently only
432 --stdio output supports this feature.
435 Show only perf.data header (forces --stdio).
438 Only analyze samples within given time window: <start>,<stop>. Times
439 have the format seconds.nanoseconds. If start is not given (i.e. time
440 string is ',x.y') then analysis starts at the beginning of the file. If
441 stop time is not given (i.e. time string is 'x.y,') then analysis goes
442 to end of file. Multiple ranges can be separated by spaces, which
443 requires the argument to be quoted e.g. --time "1234.567,1234.789 1235,"
445 Also support time percent with multiple time ranges. Time string is
446 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
449 Select the second 10% time slice:
451 perf report --time 10%/2
453 Select from 0% to 10% time slice:
455 perf report --time 0%-10%
457 Select the first and second 10% time slices:
459 perf report --time 10%/1,10%/2
461 Select from 0% to 10% and 30% to 40% slices:
463 perf report --time 0%-10%,30%-40%
465 --switch-on EVENT_NAME::
466 Only consider events after this event is found.
468 This may be interesting to measure a workload only after some initialization
469 phase is over, i.e. insert a perf probe at that point and then using this
470 option with that probe.
472 --switch-off EVENT_NAME::
473 Stop considering events after this event is found.
475 --show-on-off-events::
476 Show the --switch-on/off events too. This has no effect in 'perf report' now
477 but probably we'll make the default not to show the switch-on/off events
478 on the --group mode and if there is only one event besides the off/on ones,
479 go straight to the histogram browser, just like 'perf report' with no events
480 explicitly specified does.
483 Options for decoding instruction tracing data. The options are:
485 include::itrace.txt[]
487 To disable decoding entirely, use --no-itrace.
490 Show the full path for source files for srcline output.
492 --show-ref-call-graph::
493 When multiple events are sampled, it may not be needed to collect
494 callgraphs for all of them. The sample sites are usually nearby,
495 and it's enough to collect the callgraphs on a reference event.
496 So user can use "call-graph=no" event modifier to disable callgraph
497 for other events to reduce the overhead.
498 However, perf report cannot show callgraphs for the event which
499 disable the callgraph.
500 This option extends the perf report to show reference callgraphs,
501 which collected by reference event, in no callgraph event.
504 Show callgraph with stitched LBRs, which may have more complete
505 callgraph. The perf.data file must have been obtained using
506 perf record --call-graph lbr.
507 Disabled by default. In common cases with call stack overflows,
508 it can recreate better call stacks than the default lbr call stack
509 output. But this approach is not full proof. There can be cases
510 where it creates incorrect call stacks from incorrect matches.
511 The known limitations include exception handing such as
512 setjmp/longjmp will have calls/returns not match.
515 Only report the samples on the processor socket that match with this filter
518 Save N individual samples for each histogram entry to show context in perf
522 When displaying traceevent output, do not use print fmt or plugins.
525 Enable hierarchical output.
528 If a callgraph address belongs to an inlined function, the inline stack
529 will be printed. Each entry is function name or file/line. Enabled by
530 default, disable with --no-inline.
533 Show --tasks output plus mmap information in a format similar to
536 Please note that not all mmaps are stored, options affecting which ones
537 are include 'perf record --data', for instance.
540 Show time stamps in nanoseconds.
543 Display overall events statistics without any further processing.
544 (like the one at the end of the perf report -D command)
547 Display monitored tasks stored in perf data. Displaying pid/tid/ppid
548 plus the command string aligned to distinguish parent and child tasks.
551 Set annotation percent type from following choices:
552 global-period, local-period, global-hits, local-hits
554 The local/global keywords set if the percentage is computed
555 in the scope of the function (local) or the whole data (global).
556 The period/hits keywords set the base the percentage is computed
557 on - the samples period or the number of samples (hits).
560 Configure time quantum for time sort key. Default 100ms.
561 Accepts s, us, ms, ns units.
564 When --total-cycles is specified, it supports sorting for all blocks by
565 'Sampled Cycles%'. This is useful to concentrate on the globally hottest
566 blocks. In output, there are some new columns:
568 'Sampled Cycles%' - block sampled cycles aggregation / total sampled cycles
569 'Sampled Cycles' - block sampled cycles aggregation
570 'Avg Cycles%' - block average sampled cycles / sum of total block average
572 'Avg Cycles' - block average sampled cycles
575 Do not print 0 results in the --stat output.
577 include::callchain-overhead-calculation.txt[]
581 linkperf:perf-stat[1], linkperf:perf-annotate[1], linkperf:perf-record[1],
582 linkperf:perf-intel-pt[1]