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