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 - symbol: name of function executed at the time of sample
84 - symbol_size: size of function executed at the time of sample
85 - parent: name of function matched to the parent regex filter. Unmatched
86 entries are displayed as "[other]".
87 - cpu: cpu number the task ran at the time of sample
88 - socket: processor socket number the task ran at the time of sample
89 - srcline: filename and line number executed at the time of sample. The
90 DWARF debugging info must be provided.
91 - srcfile: file name of the source file of the same. Requires dwarf
93 - weight: Event specific weight, e.g. memory latency or transaction
94 abort cost. This is the global weight.
95 - local_weight: Local weight version of the weight above.
96 - cgroup_id: ID derived from cgroup namespace device and inode numbers.
97 - transaction: Transaction abort flags.
98 - overhead: Overhead percentage of sample
99 - overhead_sys: Overhead percentage of sample running in system mode
100 - overhead_us: Overhead percentage of sample running in user mode
101 - overhead_guest_sys: Overhead percentage of sample running in system mode
103 - overhead_guest_us: Overhead percentage of sample running in user mode on
105 - sample: Number of sample
106 - period: Raw number of event count of sample
108 By default, comm, dso and symbol keys are used.
109 (i.e. --sort comm,dso,symbol)
111 If --branch-stack option is used, following sort keys are also
114 - dso_from: name of library or module branched from
115 - dso_to: name of library or module branched to
116 - symbol_from: name of function branched from
117 - symbol_to: name of function branched to
118 - srcline_from: source file and line branched from
119 - srcline_to: source file and line branched to
120 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
121 - in_tx: branch in TSX transaction
122 - abort: TSX transaction abort.
123 - cycles: Cycles in basic block
125 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
126 and symbol_to, see '--branch-stack'.
128 If the --mem-mode option is used, the following sort keys are also available
129 (incompatible with --branch-stack):
130 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline.
132 - symbol_daddr: name of data symbol being executed on at the time of sample
133 - dso_daddr: name of library or module containing the data being executed
134 on at the time of the sample
135 - locked: whether the bus was locked at the time of the sample
136 - tlb: type of tlb access for the data at the time of the sample
137 - mem: type of memory access for the data at the time of the sample
138 - snoop: type of snoop (if any) for the data at the time of the sample
139 - dcacheline: the cacheline the data address is on at the time of the sample
141 And the default sort keys are changed to local_weight, mem, sym, dso,
142 symbol_daddr, dso_daddr, snoop, tlb, locked, see '--mem-mode'.
144 If the data file has tracepoint event(s), following (dynamic) sort keys
146 trace, trace_fields, [<event>.]<field>[/raw]
148 - trace: pretty printed trace output in a single column
149 - trace_fields: fields in tracepoints in separate columns
150 - <field name>: optional event and field name for a specific field
152 The last form consists of event and field names. If event name is
153 omitted, it searches all events for matching field name. The matched
154 field will be shown only for the event has the field. The event name
155 supports substring match so user doesn't need to specify full subsystem
156 and event name everytime. For example, 'sched:sched_switch' event can
157 be shortened to 'switch' as long as it's not ambiguous. Also event can
158 be specified by its index (starting from 1) preceded by the '%'.
159 So '%1' is the first event, '%2' is the second, and so on.
161 The field name can have '/raw' suffix which disables pretty printing
162 and shows raw field value like hex numbers. The --raw-trace option
163 has the same effect for all dynamic sort keys.
165 The default sort keys are changed to 'trace' if all events in the data
170 Specify output field - multiple keys can be specified in CSV format.
171 Following fields are available:
172 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
173 Also it can contain any sort key(s).
175 By default, every sort keys not specified in -F will be appended
178 If the keys starts with a prefix '+', then it will append the specified
179 field(s) to the default field order. For example: perf report -F +period,sample.
183 A regex filter to identify parent. The parent is a caller of this
184 function and searched through the callchain, thus it requires callchain
185 information recorded. The pattern is in the exteneded regex format and
186 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
190 Only display entries with parent-match.
193 --column-widths=<width[,width...]>::
194 Force each column width to the provided list, for large terminal
195 readability. 0 means no limit (default behavior).
199 Use a special separator character and don't pad with spaces, replacing
200 all occurrences of this separator in symbol names (and other output)
201 with a '.' character, that thus it's the only non valid separator.
205 Dump raw trace in ASCII.
208 --call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>::
209 Display call chains using type, min percent threshold, print limit,
210 call order, sort key, optional branch and value. Note that ordering of
211 parameters is not fixed so any parement can be given in an arbitraty order.
212 One exception is the print_limit which should be preceded by threshold.
214 print_type can be either:
215 - flat: single column, linear exposure of call chains.
216 - graph: use a graph tree, displaying absolute overhead rates. (default)
217 - fractal: like graph, but displays relative rates. Each branch of
218 the tree is considered as a new profiled object.
219 - folded: call chains are displayed in a line, separated by semicolons
220 - none: disable call chain display.
222 threshold is a percentage value which specifies a minimum percent to be
223 included in the output call graph. Default is 0.5 (%).
225 print_limit is only applied when stdio interface is used. It's to limit
226 number of call graph entries in a single hist entry. Note that it needs
227 to be given after threshold (but not necessarily consecutive).
228 Default is 0 (unlimited).
231 - callee: callee based call graph.
232 - caller: inverted caller based call graph.
233 Default is 'caller' when --children is used, otherwise 'callee'.
236 - function: compare on functions (default)
237 - address: compare on individual code addresses
238 - srcline: compare on source filename and line number
241 - branch: include last branch information in callgraph when available.
242 Usually more convenient to use --branch-history for this.
245 - percent: diplay overhead percent (default)
246 - period: display event period
247 - count: display event count
250 Accumulate callchain of children to parent entry so that then can
251 show up in the output. The output will have a new "Children" column
252 and will be sorted on the data. It requires callchains are recorded.
253 See the `overhead calculation' section for more details. Enabled by
254 default, disable with --no-children.
257 Set the stack depth limit when parsing the callchain, anything
258 beyond the specified depth will be ignored. This is a trade-off
259 between information loss and faster processing especially for
260 workloads that can have a very long callchain stack.
261 Note that when using the --itrace option the synthesized callchain size
262 will override this value if the synthesized callchain size is bigger.
268 alias for inverted caller based call graph.
270 --ignore-callees=<regex>::
271 Ignore callees of the function(s) matching the given regex.
272 This has the effect of collecting the callers of each such
273 function into one place in the call-graph tree.
276 Pretty printing style. key: normal, raw
278 --stdio:: Use the stdio interface.
281 'always', 'never' or 'auto', allowing configuring color output
282 via the command line, in addition to via "color.ui" .perfconfig.
283 Use '--stdio-color always' to generate color even when redirecting
284 to a pipe or file. Using just '--stdio-color' is equivalent to
287 --tui:: Use the TUI interface, that is integrated with annotate and allows
288 zooming into DSOs or threads, among other features. Use of --tui
289 requires a tty, if one is not present, as when piping to other
290 commands, the stdio interface is used.
292 --gtk:: Use the GTK2 interface.
303 Load module symbols. WARNING: This should only be used with -k and
308 Don't do ownership validation.
310 --symfs=<directory>::
311 Look for files with symbols relative to this directory.
314 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
315 be provided as a comma-separated list with no space: 0,1. Ranges of
316 CPUs are specified with -: 0-2. Default is to report samples on all
320 --disassembler-style=:: Set disassembler style for objdump.
323 Interleave source code with assembly code. Enabled by default,
324 disable with --no-source.
327 Show raw instruction encoding of assembly instructions.
329 --show-total-period:: Show a column with the sum of periods.
333 Display extended information about the perf.data file. This adds
334 information which may be very large and thus may clutter the display.
335 It currently includes: cpu and numa topology of the host system.
339 Use the addresses of sampled taken branches instead of the instruction
340 address to build the histograms. To generate meaningful output, the
341 perf.data file must have been obtained using perf record -b or
342 perf record --branch-filter xxx where xxx is a branch filter option.
343 perf report is able to auto-detect whether a perf.data file contains
344 branch stacks and it will automatically switch to the branch view mode,
345 unless --no-branch-stack is used.
348 Add the addresses of sampled taken branches to the callstack.
349 This allows to examine the path the program took to each sample.
350 The data collection must have used -b (or -j) and -g.
353 Path to objdump binary.
356 Show event group information together.
359 Demangle symbol names to human readable form. It's enabled by default,
360 disable with --no-demangle.
363 Demangle kernel symbol names to human readable form (for C++ kernels).
366 Use the data addresses of samples in addition to instruction addresses
367 to build the histograms. To generate meaningful output, the perf.data
368 file must have been obtained using perf record -d -W and using a
369 special event -e cpu/mem-loads/ or -e cpu/mem-stores/. See
370 'perf mem' for simpler access.
373 Do not show entries which have an overhead under that percent.
374 (Default: 0). Note that this option also sets the percent limit (threshold)
375 of callchains. However the default value of callchain threshold is
376 different than the default value of hist entries. Please see the
377 --call-graph option for details.
380 Determine how to display the overhead percentage of filtered entries.
381 Filters can be applied by --comms, --dsos and/or --symbols options and
382 Zoom operations on the TUI (thread, dso, etc).
384 "relative" means it's relative to filtered entries only so that the
385 sum of shown entries will be always 100%. "absolute" means it retains
386 the original value before and after the filter is applied.
389 Show header information in the perf.data file. This includes
390 various information like hostname, OS and perf version, cpu/mem
391 info, perf command line, event list and so on. Currently only
392 --stdio output supports this feature.
395 Show only perf.data header (forces --stdio).
398 Only analyze samples within given time window: <start>,<stop>. Times
399 have the format seconds.microseconds. If start is not given (i.e., time
400 string is ',x.y') then analysis starts at the beginning of the file. If
401 stop time is not given (i.e, time string is 'x.y,') then analysis goes
405 Options for decoding instruction tracing data. The options are:
407 include::itrace.txt[]
409 To disable decoding entirely, use --no-itrace.
412 Show the full path for source files for srcline output.
414 --show-ref-call-graph::
415 When multiple events are sampled, it may not be needed to collect
416 callgraphs for all of them. The sample sites are usually nearby,
417 and it's enough to collect the callgraphs on a reference event.
418 So user can use "call-graph=no" event modifier to disable callgraph
419 for other events to reduce the overhead.
420 However, perf report cannot show callgraphs for the event which
421 disable the callgraph.
422 This option extends the perf report to show reference callgraphs,
423 which collected by reference event, in no callgraph event.
426 Only report the samples on the processor socket that match with this filter
429 When displaying traceevent output, do not use print fmt or plugins.
432 Enable hierarchical output.
435 If a callgraph address belongs to an inlined function, the inline stack
436 will be printed. Each entry is function name or file/line.
438 include::callchain-overhead-calculation.txt[]
442 linkperf:perf-stat[1], linkperf:perf-annotate[1]