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 --showcpuutilization::
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, 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 - srcline: filename and line number executed at the time of sample. The
83 DWARF debugging info must be provided.
84 - srcfile: file name of the source file of the same. Requires dwarf
86 - weight: Event specific weight, e.g. memory latency or transaction
87 abort cost. This is the global weight.
88 - local_weight: Local weight version of the weight above.
89 - transaction: Transaction abort flags.
90 - overhead: Overhead percentage of sample
91 - overhead_sys: Overhead percentage of sample running in system mode
92 - overhead_us: Overhead percentage of sample running in user mode
93 - overhead_guest_sys: Overhead percentage of sample running in system mode
95 - overhead_guest_us: Overhead percentage of sample running in user mode on
97 - sample: Number of sample
98 - period: Raw number of event count of sample
100 By default, comm, dso and symbol keys are used.
101 (i.e. --sort comm,dso,symbol)
103 If --branch-stack option is used, following sort keys are also
105 dso_from, dso_to, symbol_from, symbol_to, mispredict.
107 - dso_from: name of library or module branched from
108 - dso_to: name of library or module branched to
109 - symbol_from: name of function branched from
110 - symbol_to: name of function branched to
111 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
112 - in_tx: branch in TSX transaction
113 - abort: TSX transaction abort.
114 - cycles: Cycles in basic block
116 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
117 and symbol_to, see '--branch-stack'.
121 Specify output field - multiple keys can be specified in CSV format.
122 Following fields are available:
123 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
124 Also it can contain any sort key(s).
126 By default, every sort keys not specified in -F will be appended
129 If --mem-mode option is used, following sort keys are also available
130 (incompatible with --branch-stack):
131 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline.
133 - symbol_daddr: name of data symbol being executed on at the time of sample
134 - dso_daddr: name of library or module containing the data being executed
135 on at the time of sample
136 - locked: whether the bus was locked at the time of sample
137 - tlb: type of tlb access for the data at the time of sample
138 - mem: type of memory access for the data at the time of sample
139 - snoop: type of snoop (if any) for the data at the time of sample
140 - dcacheline: the cacheline the data address is on at the time of sample
142 And default sort keys are changed to local_weight, mem, sym, dso,
143 symbol_daddr, dso_daddr, snoop, tlb, locked, see '--mem-mode'.
147 A regex filter to identify parent. The parent is a caller of this
148 function and searched through the callchain, thus it requires callchain
149 information recorded. The pattern is in the exteneded regex format and
150 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
154 Only display entries with parent-match.
157 --column-widths=<width[,width...]>::
158 Force each column width to the provided list, for large terminal
159 readability. 0 means no limit (default behavior).
163 Use a special separator character and don't pad with spaces, replacing
164 all occurrences of this separator in symbol names (and other output)
165 with a '.' character, that thus it's the only non valid separator.
169 Dump raw trace in ASCII.
171 -g [type,min[,limit],order[,key][,branch]]::
173 Display call chains using type, min percent threshold, optional print
176 - flat: single column, linear exposure of call chains.
177 - graph: use a graph tree, displaying absolute overhead rates.
178 - fractal: like graph, but displays relative rates. Each branch of
179 the tree is considered as a new profiled object. +
182 - callee: callee based call graph.
183 - caller: inverted caller based call graph.
186 - function: compare on functions
187 - address: compare on individual code addresses
190 - branch: include last branch information in callgraph
191 when available. Usually more convenient to use --branch-history
194 Default: fractal,0.5,callee,function.
197 Accumulate callchain of children to parent entry so that then can
198 show up in the output. The output will have a new "Children" column
199 and will be sorted on the data. It requires callchains are recorded.
200 See the `overhead calculation' section for more details.
203 Set the stack depth limit when parsing the callchain, anything
204 beyond the specified depth will be ignored. This is a trade-off
205 between information loss and faster processing especially for
206 workloads that can have a very long callchain stack.
212 alias for inverted caller based call graph.
214 --ignore-callees=<regex>::
215 Ignore callees of the function(s) matching the given regex.
216 This has the effect of collecting the callers of each such
217 function into one place in the call-graph tree.
220 Pretty printing style. key: normal, raw
222 --stdio:: Use the stdio interface.
224 --tui:: Use the TUI interface, that is integrated with annotate and allows
225 zooming into DSOs or threads, among other features. Use of --tui
226 requires a tty, if one is not present, as when piping to other
227 commands, the stdio interface is used.
229 --gtk:: Use the GTK2 interface.
240 Load module symbols. WARNING: This should only be used with -k and
245 Don't complain, do it.
247 --symfs=<directory>::
248 Look for files with symbols relative to this directory.
251 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
252 be provided as a comma-separated list with no space: 0,1. Ranges of
253 CPUs are specified with -: 0-2. Default is to report samples on all
257 --disassembler-style=:: Set disassembler style for objdump.
260 Interleave source code with assembly code. Enabled by default,
261 disable with --no-source.
264 Show raw instruction encoding of assembly instructions.
266 --show-total-period:: Show a column with the sum of periods.
270 Display extended information about the perf.data file. This adds
271 information which may be very large and thus may clutter the display.
272 It currently includes: cpu and numa topology of the host system.
276 Use the addresses of sampled taken branches instead of the instruction
277 address to build the histograms. To generate meaningful output, the
278 perf.data file must have been obtained using perf record -b or
279 perf record --branch-filter xxx where xxx is a branch filter option.
280 perf report is able to auto-detect whether a perf.data file contains
281 branch stacks and it will automatically switch to the branch view mode,
282 unless --no-branch-stack is used.
285 Add the addresses of sampled taken branches to the callstack.
286 This allows to examine the path the program took to each sample.
287 The data collection must have used -b (or -j) and -g.
290 Path to objdump binary.
293 Show event group information together.
296 Demangle symbol names to human readable form. It's enabled by default,
297 disable with --no-demangle.
300 Demangle kernel symbol names to human readable form (for C++ kernels).
303 Use the data addresses of samples in addition to instruction addresses
304 to build the histograms. To generate meaningful output, the perf.data
305 file must have been obtained using perf record -d -W and using a
306 special event -e cpu/mem-loads/ or -e cpu/mem-stores/. See
307 'perf mem' for simpler access.
310 Do not show entries which have an overhead under that percent.
314 Determine how to display the overhead percentage of filtered entries.
315 Filters can be applied by --comms, --dsos and/or --symbols options and
316 Zoom operations on the TUI (thread, dso, etc).
318 "relative" means it's relative to filtered entries only so that the
319 sum of shown entries will be always 100%. "absolute" means it retains
320 the original value before and after the filter is applied.
323 Show header information in the perf.data file. This includes
324 various information like hostname, OS and perf version, cpu/mem
325 info, perf command line, event list and so on. Currently only
326 --stdio output supports this feature.
329 Show only perf.data header (forces --stdio).
332 Options for decoding instruction tracing data. The options are:
334 include::itrace.txt[]
336 To disable decoding entirely, use --no-itrace.
339 Show the full path for source files for srcline output.
341 --show-ref-call-graph::
342 When multiple events are sampled, it may not be needed to collect
343 callgraphs for all of them. The sample sites are usually nearby,
344 and it's enough to collect the callgraphs on a reference event.
345 So user can use "call-graph=no" event modifier to disable callgraph
346 for other events to reduce the overhead.
347 However, perf report cannot show callgraphs for the event which
348 disable the callgraph.
349 This option extends the perf report to show reference callgraphs,
350 which collected by reference event, in no callgraph event.
352 include::callchain-overhead-calculation.txt[]
356 linkperf:perf-stat[1], linkperf:perf-annotate[1]