1PERF-TOP(1) perf Manual PERF-TOP(1)
2
3
4
6 perf-top - System profiling tool.
7
9 perf top [-e <EVENT> | --event=EVENT] [<options>]
10
12 This command generates and displays a performance counter profile in
13 real time.
14
16 -a, --all-cpus
17 System-wide collection. (default)
18
19 -c <count>, --count=<count>
20 Event period to sample.
21
22 -C <cpu-list>, --cpu=<cpu>
23 Monitor only on the list of CPUs provided. Multiple CPUs can be
24 provided as a comma-separated list with no space: 0,1. Ranges of
25 CPUs are specified with -: 0-2. Default is to monitor all CPUS.
26
27 -d <seconds>, --delay=<seconds>
28 Number of seconds to delay between refreshes.
29
30 -e <event>, --event=<event>
31 Select the PMU event. Selection can be a symbolic event name (use
32 perf list to list all events) or a raw PMU event (eventsel+umask)
33 in the form of rNNN where NNN is a hexadecimal event descriptor.
34
35 -E <entries>, --entries=<entries>
36 Display this many functions.
37
38 -f <count>, --count-filter=<count>
39 Only display functions with more events than this.
40
41 --group
42 Put the counters into a counter group.
43
44 -F <freq>, --freq=<freq>
45 Profile at this frequency. Use max to use the currently maximum
46 allowed frequency, i.e. the value in the
47 kernel.perf_event_max_sample_rate sysctl.
48
49 -i, --inherit
50 Child tasks do not inherit counters.
51
52 -k <path>, --vmlinux=<path>
53 Path to vmlinux. Required for annotation functionality.
54
55 --ignore-vmlinux
56 Ignore vmlinux files.
57
58 -m <pages>, --mmap-pages=<pages>
59 Number of mmap data pages (must be a power of two) or size
60 specification with appended unit character - B/K/M/G. The size is
61 rounded up to have nearest pages power of two value.
62
63 -p <pid>, --pid=<pid>
64 Profile events on existing Process ID (comma separated list).
65
66 -t <tid>, --tid=<tid>
67 Profile events on existing thread ID (comma separated list).
68
69 -u, --uid=
70 Record events in threads owned by uid. Name or number.
71
72 -r <priority>, --realtime=<priority>
73 Collect data with this RT SCHED_FIFO priority.
74
75 --sym-annotate=<symbol>
76 Annotate this symbol.
77
78 -K, --hide_kernel_symbols
79 Hide kernel symbols.
80
81 -U, --hide_user_symbols
82 Hide user symbols.
83
84 --demangle-kernel
85 Demangle kernel symbols.
86
87 -D, --dump-symtab
88 Dump the symbol table used for profiling.
89
90 -v, --verbose
91 Be more verbose (show counter open errors, etc).
92
93 -z, --zero
94 Zero history across display updates.
95
96 -s, --sort
97 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
98 local_weight, abort, in_tx, transaction, overhead, sample, period.
99 Please see description of --sort in the perf-report man page.
100
101 --fields=
102 Specify output field - multiple keys can be specified in CSV
103 format. Following fields are available: overhead, overhead_sys,
104 overhead_us, overhead_children, sample and period. Also it can
105 contain any sort key(s).
106
107 By default, every sort keys not specified in --field will be appended
108 automatically.
109
110 -n, --show-nr-samples
111 Show a column with the number of samples.
112
113 --show-total-period
114 Show a column with the sum of periods.
115
116 --dsos
117 Only consider symbols in these dsos. This option will affect the
118 percentage of the overhead column. See --percentage for more info.
119
120 --comms
121 Only consider symbols in these comms. This option will affect the
122 percentage of the overhead column. See --percentage for more info.
123
124 --symbols
125 Only consider these symbols. This option will affect the percentage
126 of the overhead column. See --percentage for more info.
127
128 -M, --disassembler-style=
129 Set disassembler style for objdump.
130
131 --source
132 Interleave source code with assembly code. Enabled by default,
133 disable with --no-source.
134
135 --asm-raw
136 Show raw instruction encoding of assembly instructions.
137
138 -g
139 Enables call-graph (stack chain/backtrace) recording.
140
141 --call-graph [mode,type,min[,limit],order[,key][,branch]]
142 Setup and enable call-graph (stack chain/backtrace) recording,
143 implies -g. See --call-graph section in perf-record and perf-report
144 man pages for details.
145
146 --children
147 Accumulate callchain of children to parent entry so that then can
148 show up in the output. The output will have a new "Children" column
149 and will be sorted on the data. It requires -g/--call-graph option
150 enabled. See the ‘overhead calculation’ section for more details.
151 Enabled by default, disable with --no-children.
152
153 --max-stack
154 Set the stack depth limit when parsing the callchain, anything
155 beyond the specified depth will be ignored. This is a trade-off
156 between information loss and faster processing especially for
157 workloads that can have a very long callchain stack.
158
159 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
160
161 --ignore-callees=<regex>
162 Ignore callees of the function(s) matching the given regex. This
163 has the effect of collecting the callers of each such function into
164 one place in the call-graph tree.
165
166 --percent-limit
167 Do not show entries which have an overhead under that percent.
168 (Default: 0).
169
170 --percentage
171 Determine how to display the overhead percentage of filtered
172 entries. Filters can be applied by --comms, --dsos and/or --symbols
173 options and Zoom operations on the TUI (thread, dso, etc).
174
175 "relative" means it's relative to filtered entries only so that the
176 sum of shown entries will be always 100%. "absolute" means it retains
177 the original value before and after the filter is applied.
178
179 -w, --column-widths=<width[,width...]>
180 Force each column width to the provided list, for large terminal
181 readability. 0 means no limit (default behavior).
182
183 --proc-map-timeout
184 When processing pre-existing threads /proc/XXX/mmap, it may take a
185 long time, because the file may be huge. A time out is needed in
186 such cases. This option sets the time out limit. The default value
187 is 500 ms.
188
189 -b, --branch-any
190 Enable taken branch stack sampling. Any type of taken branch may be
191 sampled. This is a shortcut for --branch-filter any. See
192 --branch-filter for more infos.
193
194 -j, --branch-filter
195 Enable taken branch stack sampling. Each sample captures a series
196 of consecutive taken branches. The number of branches captured with
197 each sample depends on the underlying hardware, the type of
198 branches of interest, and the executed code. It is possible to
199 select the types of branches captured by enabling filters. For a
200 full list of modifiers please see the perf record manpage.
201
202 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
203 The privilege levels may be omitted, in which case, the privilege levels of the associated
204 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
205 levels are subject to permissions. When sampling on multiple events, branch stack sampling
206 is enabled for all the sampling events. The sampled branch type is the same for all events.
207 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
208 Note that this feature may not be available on all processors.
209
210 --raw-trace
211 When displaying traceevent output, do not use print fmt or plugins.
212
213 --hierarchy
214 Enable hierarchy output.
215
216 --force
217 Don’t do ownership validation.
218
219 --num-thread-synthesize
220 The number of threads to run when synthesizing events for existing
221 processes. By default, the number of threads equals to the number
222 of online CPUs.
223
225 [d]
226 Display refresh delay.
227
228 [e]
229 Number of entries to display.
230
231 [E]
232 Event to display when multiple counters are active.
233
234 [f]
235 Profile display filter (>= hit count).
236
237 [F]
238 Annotation display filter (>= % of total).
239
240 [s]
241 Annotate symbol.
242
243 [S]
244 Stop annotation, return to full profile display.
245
246 [K]
247 Hide kernel symbols.
248
249 [U]
250 Hide user symbols.
251
252 [z]
253 Toggle event count zeroing across display updates.
254
255 [qQ]
256 Quit.
257
258 Pressing any unmapped key displays a menu, and prompts for input.
259
261 The overhead can be shown in two columns as Children and Self when perf
262 collects callchains. The self overhead is simply calculated by adding
263 all period values of the entry - usually a function (symbol). This is
264 the value that perf shows traditionally and sum of all the self
265 overhead values should be 100%.
266
267 The children overhead is calculated by adding all period values of the
268 child functions so that it can show the total overhead of the higher
269 level functions even if they don’t directly execute much. Children here
270 means functions that are called from another (parent) function.
271
272 It might be confusing that the sum of all the children overhead values
273 exceeds 100% since each of them is already an accumulation of self
274 overhead of its child functions. But with this enabled, users can find
275 which function has the most overhead even if samples are spread over
276 the children.
277
278 Consider the following example; there are three functions like below.
279
280
281 .ft C
282 void foo(void) {
283 /* do something */
284 }
285
286 void bar(void) {
287 /* do something */
288 foo();
289 }
290
291 int main(void) {
292 bar()
293 return 0;
294 }
295 .ft
296
297
298 In this case foo is a child of bar, and bar is an immediate child of
299 main so foo also is a child of main. In other words, main is a parent
300 of foo and bar, and bar is a parent of foo.
301
302 Suppose all samples are recorded in foo and bar only. When it’s
303 recorded with callchains the output will show something like below in
304 the usual (self-overhead-only) output of perf report:
305
306
307 .ft C
308 Overhead Symbol
309 ........ .....................
310 60.00% foo
311 |
312 --- foo
313 bar
314 main
315 __libc_start_main
316
317 40.00% bar
318 |
319 --- bar
320 main
321 __libc_start_main
322 .ft
323
324
325 When the --children option is enabled, the self overhead values of
326 child functions (i.e. foo and bar) are added to the parents to
327 calculate the children overhead. In this case the report could be
328 displayed as:
329
330
331 .ft C
332 Children Self Symbol
333 ........ ........ ....................
334 100.00% 0.00% __libc_start_main
335 |
336 --- __libc_start_main
337
338 100.00% 0.00% main
339 |
340 --- main
341 __libc_start_main
342
343 100.00% 40.00% bar
344 |
345 --- bar
346 main
347 __libc_start_main
348
349 60.00% 60.00% foo
350 |
351 --- foo
352 bar
353 main
354 __libc_start_main
355 .ft
356
357
358 In the above output, the self overhead of foo (60%) was add to the
359 children overhead of bar, main and __libc_start_main. Likewise, the
360 self overhead of bar (40%) was added to the children overhead of main
361 and \_\_libc_start_main.
362
363 So \_\_libc_start_main and main are shown first since they have same
364 (100%) children overhead (even though they have zero self overhead) and
365 they are the parents of foo and bar.
366
367 Since v3.16 the children overhead is shown by default and the output is
368 sorted by its values. The children overhead is disabled by specifying
369 --no-children option on the command line or by adding report.children =
370 false or top.children = false in the perf config file.
371
373 perf-stat(1), perf-list(1), perf-report(1)
374
375
376
377perf 09/24/2019 PERF-TOP(1)