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 Only consider symbols in these dsos. CSV that understands
31 file://filename entries.
34 Show the number of samples for each symbol
36 --showcpuutilization::
37 Show sample percentage for different cpu modes.
41 Show per-thread event counters
44 Only consider symbols in these comms. CSV that understands
45 file://filename entries.
48 Only consider these symbols. CSV that understands
49 file://filename entries.
52 Only show symbols that match (partially) with this filter.
56 Only display entries resolved to a symbol.
60 Sort histogram entries by given key(s) - multiple keys can be specified
61 in CSV format. Following sort keys are available:
62 pid, comm, dso, symbol, parent, cpu, srcline, weight, local_weight.
64 Each key has following meaning:
66 - comm: command (name) of the task which can be read via /proc/<pid>/comm
67 - pid: command and tid of the task
68 - dso: name of library or module executed at the time of sample
69 - symbol: name of function executed at the time of sample
70 - parent: name of function matched to the parent regex filter. Unmatched
71 entries are displayed as "[other]".
72 - cpu: cpu number the task ran at the time of sample
73 - srcline: filename and line number executed at the time of sample. The
74 DWARF debugging info must be provided.
75 - weight: Event specific weight, e.g. memory latency or transaction
76 abort cost. This is the global weight.
77 - local_weight: Local weight version of the weight above.
78 - transaction: Transaction abort flags.
80 By default, comm, dso and symbol keys are used.
81 (i.e. --sort comm,dso,symbol)
83 If --branch-stack option is used, following sort keys are also
85 dso_from, dso_to, symbol_from, symbol_to, mispredict.
87 - dso_from: name of library or module branched from
88 - dso_to: name of library or module branched to
89 - symbol_from: name of function branched from
90 - symbol_to: name of function branched to
91 - mispredict: "N" for predicted branch, "Y" for mispredicted branch
92 - in_tx: branch in TSX transaction
93 - abort: TSX transaction abort.
95 And default sort keys are changed to comm, dso_from, symbol_from, dso_to
96 and symbol_to, see '--branch-stack'.
100 A regex filter to identify parent. The parent is a caller of this
101 function and searched through the callchain, thus it requires callchain
102 information recorded. The pattern is in the exteneded regex format and
103 defaults to "\^sys_|^do_page_fault", see '--sort parent'.
107 Only display entries with parent-match.
110 --column-widths=<width[,width...]>::
111 Force each column width to the provided list, for large terminal
116 Use a special separator character and don't pad with spaces, replacing
117 all occurrences of this separator in symbol names (and other output)
118 with a '.' character, that thus it's the only non valid separator.
122 Dump raw trace in ASCII.
124 -g [type,min[,limit],order[,key]]::
126 Display call chains using type, min percent threshold, optional print
129 - flat: single column, linear exposure of call chains.
130 - graph: use a graph tree, displaying absolute overhead rates.
131 - fractal: like graph, but displays relative rates. Each branch of
132 the tree is considered as a new profiled object. +
135 - callee: callee based call graph.
136 - caller: inverted caller based call graph.
139 - function: compare on functions
140 - address: compare on individual code addresses
142 Default: fractal,0.5,callee,function.
145 Set the stack depth limit when parsing the callchain, anything
146 beyond the specified depth will be ignored. This is a trade-off
147 between information loss and faster processing especially for
148 workloads that can have a very long callchain stack.
154 alias for inverted caller based call graph.
156 --ignore-callees=<regex>::
157 Ignore callees of the function(s) matching the given regex.
158 This has the effect of collecting the callers of each such
159 function into one place in the call-graph tree.
162 Pretty printing style. key: normal, raw
164 --stdio:: Use the stdio interface.
166 --tui:: Use the TUI interface, that is integrated with annotate and allows
167 zooming into DSOs or threads, among other features. Use of --tui
168 requires a tty, if one is not present, as when piping to other
169 commands, the stdio interface is used.
171 --gtk:: Use the GTK2 interface.
182 Load module symbols. WARNING: This should only be used with -k and
187 Don't complain, do it.
189 --symfs=<directory>::
190 Look for files with symbols relative to this directory.
193 --cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can
194 be provided as a comma-separated list with no space: 0,1. Ranges of
195 CPUs are specified with -: 0-2. Default is to report samples on all
199 --disassembler-style=:: Set disassembler style for objdump.
202 Interleave source code with assembly code. Enabled by default,
203 disable with --no-source.
206 Show raw instruction encoding of assembly instructions.
208 --show-total-period:: Show a column with the sum of periods.
212 Display extended information about the perf.data file. This adds
213 information which may be very large and thus may clutter the display.
214 It currently includes: cpu and numa topology of the host system.
218 Use the addresses of sampled taken branches instead of the instruction
219 address to build the histograms. To generate meaningful output, the
220 perf.data file must have been obtained using perf record -b or
221 perf record --branch-filter xxx where xxx is a branch filter option.
222 perf report is able to auto-detect whether a perf.data file contains
223 branch stacks and it will automatically switch to the branch view mode,
224 unless --no-branch-stack is used.
227 Path to objdump binary.
230 Show event group information together.
233 Demangle symbol names to human readable form. It's enabled by default,
234 disable with --no-demangle.
237 Do not show entries which have an overhead under that percent.
242 linkperf:perf-stat[1], linkperf:perf-annotate[1]