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 Show the number of samples for each symbol
32 --show-cpu-utilization::
33 Show sample percentage for different cpu modes.
37 Show per-thread event counters. The input data file should be recorded
41 Only consider symbols in these comms. CSV that understands
42 file://filename entries. This option will affect the percentage of
43 the overhead column. See --percentage for more info.
45 Only show events for given process ID (comma separated list).
48 Only show events for given thread ID (comma separated list).
51 Only consider symbols in these dsos. CSV that understands
52 file://filename entries. This option will affect the percentage of
53 the overhead column. See --percentage for more info.
56 Only consider these symbols. CSV that understands
57 file://filename entries. This option will affect the percentage of
58 the overhead column. See --percentage for more info.
61 Only show symbols that match (partially) with this filter.
65 Only display entries resolved to a symbol.
69 Sort histogram entries by given key(s) - multiple keys can be specified
70 in CSV format. Following sort keys are available:
71 pid, comm, dso, symbol, parent, cpu, socket, srcline, weight, local_weight.
73 Each key has following meaning:
75 - comm: command (name) of the task which can be read via /proc/<pid>/comm
76 - pid: command and tid of the task
77 - dso: name of library or module executed at the time of sample
78 - symbol: name of function executed at the time of sample
79 - parent: name of function matched to the parent regex filter. Unmatched
80 entries are displayed as "[other]".
81 - cpu: cpu number the task ran at the time of sample
82 - socket: processor socket number the task ran at the time of sample
83 - srcline: filename and line number executed at the time of sample. The
84 DWARF debugging info must be provided.
85 - srcfile: file name of the source file of the same. Requires dwarf
87 - weight: Event specific weight, e.g. memory latency or transaction
88 abort cost. This is the global weight.
89 - local_weight: Local weight version of the weight above.
90 - transaction: Transaction abort flags.
91 - overhead: Overhead percentage of sample
92 - overhead_sys: Overhead percentage of sample running in system mode
93 - overhead_us: Overhead percentage of sample running in user mode
94 - overhead_guest_sys: Overhead percentage of sample running in system mode
96 - overhead_guest_us: Overhead percentage of sample running in user mode on
98 - sample: Number of sample
99 - period: Raw number of event count of sample
101 By default, comm, dso and symbol keys are used.
102 (i.e. --sort comm,dso,symbol)
104 If --branch-stack option is used, following sort keys are also
106 dso_from, dso_to, symbol_from, symbol_to, mispredict.
108 - dso_from: name of library or module branched from
109 - dso_to: name of library or module branched to
110 - symbol_from: name of function branched from
111 - symbol_to: name of function branched to
112 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
113 - in_tx: branch in TSX transaction
114 - abort: TSX transaction abort.
115 - cycles: Cycles in basic block
117 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
118 and symbol_to, see '--branch-stack'.
122 Specify output field - multiple keys can be specified in CSV format.
123 Following fields are available:
124 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
125 Also it can contain any sort key(s).
127 By default, every sort keys not specified in -F will be appended
130 If --mem-mode option is used, following sort keys are also available
131 (incompatible with --branch-stack):
132 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline.
134 - symbol_daddr: name of data symbol being executed on at the time of sample
135 - dso_daddr: name of library or module containing the data being executed
136 on at the time of sample
137 - locked: whether the bus was locked at the time of sample
138 - tlb: type of tlb access for the data at the time of sample
139 - mem: type of memory access for the data at the time of sample
140 - snoop: type of snoop (if any) for the data at the time of sample
141 - dcacheline: the cacheline the data address is on at the time of sample
143 And default sort keys are changed to local_weight, mem, sym, dso,
144 symbol_daddr, dso_daddr, snoop, tlb, locked, see '--mem-mode'.
148 A regex filter to identify parent. The parent is a caller of this
149 function and searched through the callchain, thus it requires callchain
150 information recorded. The pattern is in the exteneded regex format and
151 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
155 Only display entries with parent-match.
158 --column-widths=<width[,width...]>::
159 Force each column width to the provided list, for large terminal
160 readability. 0 means no limit (default behavior).
164 Use a special separator character and don't pad with spaces, replacing
165 all occurrences of this separator in symbol names (and other output)
166 with a '.' character, that thus it's the only non valid separator.
170 Dump raw trace in ASCII.
173 --call-graph=<print_type,threshold[,print_limit],order,sort_key,branch>::
174 Display call chains using type, min percent threshold, print limit,
175 call order, sort key and branch. Note that ordering of parameters is not
176 fixed so any parement can be given in an arbitraty order. One exception
177 is the print_limit which should be preceded by threshold.
179 print_type can be either:
180 - flat: single column, linear exposure of call chains.
181 - graph: use a graph tree, displaying absolute overhead rates. (default)
182 - fractal: like graph, but displays relative rates. Each branch of
183 the tree is considered as a new profiled object.
184 - folded: call chains are displayed in a line, separated by semicolons
185 - none: disable call chain display.
187 threshold is a percentage value which specifies a minimum percent to be
188 included in the output call graph. Default is 0.5 (%).
190 print_limit is only applied when stdio interface is used. It's to limit
191 number of call graph entries in a single hist entry. Note that it needs
192 to be given after threshold (but not necessarily consecutive).
193 Default is 0 (unlimited).
196 - callee: callee based call graph.
197 - caller: inverted caller based call graph.
198 Default is 'caller' when --children is used, otherwise 'callee'.
201 - function: compare on functions (default)
202 - address: compare on individual code addresses
205 - branch: include last branch information in callgraph when available.
206 Usually more convenient to use --branch-history for this.
209 Accumulate callchain of children to parent entry so that then can
210 show up in the output. The output will have a new "Children" column
211 and will be sorted on the data. It requires callchains are recorded.
212 See the `overhead calculation' section for more details.
215 Set the stack depth limit when parsing the callchain, anything
216 beyond the specified depth will be ignored. This is a trade-off
217 between information loss and faster processing especially for
218 workloads that can have a very long callchain stack.
219 Note that when using the --itrace option the synthesized callchain size
220 will override this value if the synthesized callchain size is bigger.
226 alias for inverted caller based call graph.
228 --ignore-callees=<regex>::
229 Ignore callees of the function(s) matching the given regex.
230 This has the effect of collecting the callers of each such
231 function into one place in the call-graph tree.
234 Pretty printing style. key: normal, raw
236 --stdio:: Use the stdio interface.
238 --tui:: Use the TUI interface, that is integrated with annotate and allows
239 zooming into DSOs or threads, among other features. Use of --tui
240 requires a tty, if one is not present, as when piping to other
241 commands, the stdio interface is used.
243 --gtk:: Use the GTK2 interface.
254 Load module symbols. WARNING: This should only be used with -k and
259 Don't complain, do it.
261 --symfs=<directory>::
262 Look for files with symbols relative to this directory.
265 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
266 be provided as a comma-separated list with no space: 0,1. Ranges of
267 CPUs are specified with -: 0-2. Default is to report samples on all
271 --disassembler-style=:: Set disassembler style for objdump.
274 Interleave source code with assembly code. Enabled by default,
275 disable with --no-source.
278 Show raw instruction encoding of assembly instructions.
280 --show-total-period:: Show a column with the sum of periods.
284 Display extended information about the perf.data file. This adds
285 information which may be very large and thus may clutter the display.
286 It currently includes: cpu and numa topology of the host system.
290 Use the addresses of sampled taken branches instead of the instruction
291 address to build the histograms. To generate meaningful output, the
292 perf.data file must have been obtained using perf record -b or
293 perf record --branch-filter xxx where xxx is a branch filter option.
294 perf report is able to auto-detect whether a perf.data file contains
295 branch stacks and it will automatically switch to the branch view mode,
296 unless --no-branch-stack is used.
299 Add the addresses of sampled taken branches to the callstack.
300 This allows to examine the path the program took to each sample.
301 The data collection must have used -b (or -j) and -g.
304 Path to objdump binary.
307 Show event group information together.
310 Demangle symbol names to human readable form. It's enabled by default,
311 disable with --no-demangle.
314 Demangle kernel symbol names to human readable form (for C++ kernels).
317 Use the data addresses of samples in addition to instruction addresses
318 to build the histograms. To generate meaningful output, the perf.data
319 file must have been obtained using perf record -d -W and using a
320 special event -e cpu/mem-loads/ or -e cpu/mem-stores/. See
321 'perf mem' for simpler access.
324 Do not show entries which have an overhead under that percent.
328 Determine how to display the overhead percentage of filtered entries.
329 Filters can be applied by --comms, --dsos and/or --symbols options and
330 Zoom operations on the TUI (thread, dso, etc).
332 "relative" means it's relative to filtered entries only so that the
333 sum of shown entries will be always 100%. "absolute" means it retains
334 the original value before and after the filter is applied.
337 Show header information in the perf.data file. This includes
338 various information like hostname, OS and perf version, cpu/mem
339 info, perf command line, event list and so on. Currently only
340 --stdio output supports this feature.
343 Show only perf.data header (forces --stdio).
346 Options for decoding instruction tracing data. The options are:
348 include::itrace.txt[]
350 To disable decoding entirely, use --no-itrace.
353 Show the full path for source files for srcline output.
355 --show-ref-call-graph::
356 When multiple events are sampled, it may not be needed to collect
357 callgraphs for all of them. The sample sites are usually nearby,
358 and it's enough to collect the callgraphs on a reference event.
359 So user can use "call-graph=no" event modifier to disable callgraph
360 for other events to reduce the overhead.
361 However, perf report cannot show callgraphs for the event which
362 disable the callgraph.
363 This option extends the perf report to show reference callgraphs,
364 which collected by reference event, in no callgraph event.
367 Only report the samples on the processor socket that match with this filter
369 include::callchain-overhead-calculation.txt[]
373 linkperf:perf-stat[1], linkperf:perf-annotate[1]