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 same. 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 - transaction: Transaction abort flags.
99 - overhead: Overhead percentage of sample
100 - overhead_sys: Overhead percentage of sample running in system mode
101 - overhead_us: Overhead percentage of sample running in user mode
102 - overhead_guest_sys: Overhead percentage of sample running in system mode
104 - overhead_guest_us: Overhead percentage of sample running in user mode on
106 - sample: Number of sample
107 - period: Raw number of event count of sample
109 By default, comm, dso and symbol keys are used.
110 (i.e. --sort comm,dso,symbol)
112 If --branch-stack option is used, following sort keys are also
115 - dso_from: name of library or module branched from
116 - dso_to: name of library or module branched to
117 - symbol_from: name of function branched from
118 - symbol_to: name of function branched to
119 - srcline_from: source file and line branched from
120 - srcline_to: source file and line branched to
121 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
122 - in_tx: branch in TSX transaction
123 - abort: TSX transaction abort.
124 - cycles: Cycles in basic block
126 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
127 and symbol_to, see '--branch-stack'.
129 When the sort key symbol is specified, columns "IPC" and "IPC Coverage"
130 are enabled automatically. Column "IPC" reports the average IPC per function
131 and column "IPC coverage" reports the percentage of instructions with
132 sampled IPC in this function. IPC means Instruction Per Cycle. If it's low,
133 it indicates there may be a performance bottleneck when the function is
134 executed, such as a memory access bottleneck. If a function has high overhead
135 and low IPC, it's worth further analyzing it to optimize its performance.
137 If the --mem-mode option is used, the following sort keys are also available
138 (incompatible with --branch-stack):
139 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline.
141 - symbol_daddr: name of data symbol being executed on at the time of sample
142 - dso_daddr: name of library or module containing the data being executed
143 on at the time of the sample
144 - locked: whether the bus was locked at the time of the sample
145 - tlb: type of tlb access for the data at the time of the sample
146 - mem: type of memory access for the data at the time of the sample
147 - snoop: type of snoop (if any) for the data at the time of the sample
148 - dcacheline: the cacheline the data address is on at the time of the sample
149 - phys_daddr: physical address of data being executed on at the time of sample
151 And the default sort keys are changed to local_weight, mem, sym, dso,
152 symbol_daddr, dso_daddr, snoop, tlb, locked, see '--mem-mode'.
154 If the data file has tracepoint event(s), following (dynamic) sort keys
156 trace, trace_fields, [<event>.]<field>[/raw]
158 - trace: pretty printed trace output in a single column
159 - trace_fields: fields in tracepoints in separate columns
160 - <field name>: optional event and field name for a specific field
162 The last form consists of event and field names. If event name is
163 omitted, it searches all events for matching field name. The matched
164 field will be shown only for the event has the field. The event name
165 supports substring match so user doesn't need to specify full subsystem
166 and event name everytime. For example, 'sched:sched_switch' event can
167 be shortened to 'switch' as long as it's not ambiguous. Also event can
168 be specified by its index (starting from 1) preceded by the '%'.
169 So '%1' is the first event, '%2' is the second, and so on.
171 The field name can have '/raw' suffix which disables pretty printing
172 and shows raw field value like hex numbers. The --raw-trace option
173 has the same effect for all dynamic sort keys.
175 The default sort keys are changed to 'trace' if all events in the data
180 Specify output field - multiple keys can be specified in CSV format.
181 Following fields are available:
182 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
183 Also it can contain any sort key(s).
185 By default, every sort keys not specified in -F will be appended
188 If the keys starts with a prefix '+', then it will append the specified
189 field(s) to the default field order. For example: perf report -F +period,sample.
193 A regex filter to identify parent. The parent is a caller of this
194 function and searched through the callchain, thus it requires callchain
195 information recorded. The pattern is in the extended regex format and
196 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
200 Only display entries with parent-match.
203 --column-widths=<width[,width...]>::
204 Force each column width to the provided list, for large terminal
205 readability. 0 means no limit (default behavior).
209 Use a special separator character and don't pad with spaces, replacing
210 all occurrences of this separator in symbol names (and other output)
211 with a '.' character, that thus it's the only non valid separator.
215 Dump raw trace in ASCII.
218 --call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>::
219 Display call chains using type, min percent threshold, print limit,
220 call order, sort key, optional branch and value. Note that ordering
221 is not fixed so any parameter can be given in an arbitrary order.
222 One exception is the print_limit which should be preceded by threshold.
224 print_type can be either:
225 - flat: single column, linear exposure of call chains.
226 - graph: use a graph tree, displaying absolute overhead rates. (default)
227 - fractal: like graph, but displays relative rates. Each branch of
228 the tree is considered as a new profiled object.
229 - folded: call chains are displayed in a line, separated by semicolons
230 - none: disable call chain display.
232 threshold is a percentage value which specifies a minimum percent to be
233 included in the output call graph. Default is 0.5 (%).
235 print_limit is only applied when stdio interface is used. It's to limit
236 number of call graph entries in a single hist entry. Note that it needs
237 to be given after threshold (but not necessarily consecutive).
238 Default is 0 (unlimited).
241 - callee: callee based call graph.
242 - caller: inverted caller based call graph.
243 Default is 'caller' when --children is used, otherwise 'callee'.
246 - function: compare on functions (default)
247 - address: compare on individual code addresses
248 - srcline: compare on source filename and line number
251 - branch: include last branch information in callgraph when available.
252 Usually more convenient to use --branch-history for this.
255 - percent: display overhead percent (default)
256 - period: display event period
257 - count: display event count
260 Accumulate callchain of children to parent entry so that then can
261 show up in the output. The output will have a new "Children" column
262 and will be sorted on the data. It requires callchains are recorded.
263 See the `overhead calculation' section for more details. Enabled by
264 default, disable with --no-children.
267 Set the stack depth limit when parsing the callchain, anything
268 beyond the specified depth will be ignored. This is a trade-off
269 between information loss and faster processing especially for
270 workloads that can have a very long callchain stack.
271 Note that when using the --itrace option the synthesized callchain size
272 will override this value if the synthesized callchain size is bigger.
278 alias for inverted caller based call graph.
280 --ignore-callees=<regex>::
281 Ignore callees of the function(s) matching the given regex.
282 This has the effect of collecting the callers of each such
283 function into one place in the call-graph tree.
286 Pretty printing style. key: normal, raw
288 --stdio:: Use the stdio interface.
291 'always', 'never' or 'auto', allowing configuring color output
292 via the command line, in addition to via "color.ui" .perfconfig.
293 Use '--stdio-color always' to generate color even when redirecting
294 to a pipe or file. Using just '--stdio-color' is equivalent to
297 --tui:: Use the TUI interface, that is integrated with annotate and allows
298 zooming into DSOs or threads, among other features. Use of --tui
299 requires a tty, if one is not present, as when piping to other
300 commands, the stdio interface is used.
302 --gtk:: Use the GTK2 interface.
309 Ignore vmlinux files.
316 Load module symbols. WARNING: This should only be used with -k and
321 Don't do ownership validation.
323 --symfs=<directory>::
324 Look for files with symbols relative to this directory.
327 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
328 be provided as a comma-separated list with no space: 0,1. Ranges of
329 CPUs are specified with -: 0-2. Default is to report samples on all
333 --disassembler-style=:: Set disassembler style for objdump.
336 Interleave source code with assembly code. Enabled by default,
337 disable with --no-source.
340 Show raw instruction encoding of assembly instructions.
342 --show-total-period:: Show a column with the sum of periods.
346 Display extended information about the perf.data file. This adds
347 information which may be very large and thus may clutter the display.
348 It currently includes: cpu and numa topology of the host system.
352 Use the addresses of sampled taken branches instead of the instruction
353 address to build the histograms. To generate meaningful output, the
354 perf.data file must have been obtained using perf record -b or
355 perf record --branch-filter xxx where xxx is a branch filter option.
356 perf report is able to auto-detect whether a perf.data file contains
357 branch stacks and it will automatically switch to the branch view mode,
358 unless --no-branch-stack is used.
361 Add the addresses of sampled taken branches to the callstack.
362 This allows to examine the path the program took to each sample.
363 The data collection must have used -b (or -j) and -g.
366 Path to objdump binary.
369 Show event group information together. It forces group output also
370 if there are no groups defined in data file.
373 Demangle symbol names to human readable form. It's enabled by default,
374 disable with --no-demangle.
377 Demangle kernel symbol names to human readable form (for C++ kernels).
380 Use the data addresses of samples in addition to instruction addresses
381 to build the histograms. To generate meaningful output, the perf.data
382 file must have been obtained using perf record -d -W and using a
383 special event -e cpu/mem-loads/p or -e cpu/mem-stores/p. See
384 'perf mem' for simpler access.
387 Do not show entries which have an overhead under that percent.
388 (Default: 0). Note that this option also sets the percent limit (threshold)
389 of callchains. However the default value of callchain threshold is
390 different than the default value of hist entries. Please see the
391 --call-graph option for details.
394 Determine how to display the overhead percentage of filtered entries.
395 Filters can be applied by --comms, --dsos and/or --symbols options and
396 Zoom operations on the TUI (thread, dso, etc).
398 "relative" means it's relative to filtered entries only so that the
399 sum of shown entries will be always 100%. "absolute" means it retains
400 the original value before and after the filter is applied.
403 Show header information in the perf.data file. This includes
404 various information like hostname, OS and perf version, cpu/mem
405 info, perf command line, event list and so on. Currently only
406 --stdio output supports this feature.
409 Show only perf.data header (forces --stdio).
412 Only analyze samples within given time window: <start>,<stop>. Times
413 have the format seconds.microseconds. If start is not given (i.e., time
414 string is ',x.y') then analysis starts at the beginning of the file. If
415 stop time is not given (i.e, time string is 'x.y,') then analysis goes
418 Also support time percent with multiple time range. Time string is
419 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'.
422 Select the second 10% time slice:
424 perf report --time 10%/2
426 Select from 0% to 10% time slice:
428 perf report --time 0%-10%
430 Select the first and second 10% time slices:
432 perf report --time 10%/1,10%/2
434 Select from 0% to 10% and 30% to 40% slices:
436 perf report --time 0%-10%,30%-40%
439 Options for decoding instruction tracing data. The options are:
441 include::itrace.txt[]
443 To disable decoding entirely, use --no-itrace.
446 Show the full path for source files for srcline output.
448 --show-ref-call-graph::
449 When multiple events are sampled, it may not be needed to collect
450 callgraphs for all of them. The sample sites are usually nearby,
451 and it's enough to collect the callgraphs on a reference event.
452 So user can use "call-graph=no" event modifier to disable callgraph
453 for other events to reduce the overhead.
454 However, perf report cannot show callgraphs for the event which
455 disable the callgraph.
456 This option extends the perf report to show reference callgraphs,
457 which collected by reference event, in no callgraph event.
460 Only report the samples on the processor socket that match with this filter
463 When displaying traceevent output, do not use print fmt or plugins.
466 Enable hierarchical output.
469 If a callgraph address belongs to an inlined function, the inline stack
470 will be printed. Each entry is function name or file/line. Enabled by
471 default, disable with --no-inline.
474 Show --tasks output plus mmap information in a format similar to
477 Please note that not all mmaps are stored, options affecting which ones
478 are include 'perf record --data', for instance.
481 Display overall events statistics without any further processing.
482 (like the one at the end of the perf report -D command)
485 Display monitored tasks stored in perf data. Displaying pid/tid/ppid
486 plus the command string aligned to distinguish parent and child tasks.
489 Set annotation percent type from following choices:
490 global-period, local-period, global-hits, local-hits
492 The local/global keywords set if the percentage is computed
493 in the scope of the function (local) or the whole data (global).
494 The period/hits keywords set the base the percentage is computed
495 on - the samples period or the number of samples (hits).
497 include::callchain-overhead-calculation.txt[]
501 linkperf:perf-stat[1], linkperf:perf-annotate[1], linkperf:perf-record[1]