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 --group-sort-idx
45 Sort the output by the event at the index n in group. If n is
46 invalid, sort by the first event. It can support multiple groups
47 with different amount of events. WARNING: This should be used on
48 grouped events.
49
50 -F <freq>, --freq=<freq>
51 Profile at this frequency. Use max to use the currently maximum
52 allowed frequency, i.e. the value in the
53 kernel.perf_event_max_sample_rate sysctl.
54
55 -i, --inherit
56 Child tasks do not inherit counters.
57
58 -k <path>, --vmlinux=<path>
59 Path to vmlinux. Required for annotation functionality.
60
61 --ignore-vmlinux
62 Ignore vmlinux files.
63
64 --kallsyms=<file>
65 kallsyms pathname
66
67 -m <pages>, --mmap-pages=<pages>
68 Number of mmap data pages (must be a power of two) or size
69 specification with appended unit character - B/K/M/G. The size is
70 rounded up to have nearest pages power of two value.
71
72 -p <pid>, --pid=<pid>
73 Profile events on existing Process ID (comma separated list).
74
75 -t <tid>, --tid=<tid>
76 Profile events on existing thread ID (comma separated list).
77
78 -u, --uid=
79 Record events in threads owned by uid. Name or number.
80
81 -r <priority>, --realtime=<priority>
82 Collect data with this RT SCHED_FIFO priority.
83
84 --sym-annotate=<symbol>
85 Annotate this symbol.
86
87 -K, --hide_kernel_symbols
88 Hide kernel symbols.
89
90 -U, --hide_user_symbols
91 Hide user symbols.
92
93 --demangle-kernel
94 Demangle kernel symbols.
95
96 -D, --dump-symtab
97 Dump the symbol table used for profiling.
98
99 -v, --verbose
100 Be more verbose (show counter open errors, etc).
101
102 -z, --zero
103 Zero history across display updates.
104
105 -s, --sort
106 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
107 local_weight, abort, in_tx, transaction, overhead, sample, period.
108 Please see description of --sort in the perf-report man page.
109
110 --fields=
111 Specify output field - multiple keys can be specified in CSV
112 format. Following fields are available: overhead, overhead_sys,
113 overhead_us, overhead_children, sample and period. Also it can
114 contain any sort key(s).
115
116 By default, every sort keys not specified in --field will be appended
117 automatically.
118
119 -n, --show-nr-samples
120 Show a column with the number of samples.
121
122 --show-total-period
123 Show a column with the sum of periods.
124
125 --dsos
126 Only consider symbols in these dsos. This option will affect the
127 percentage of the overhead column. See --percentage for more info.
128
129 --comms
130 Only consider symbols in these comms. This option will affect the
131 percentage of the overhead column. See --percentage for more info.
132
133 --symbols
134 Only consider these symbols. This option will affect the percentage
135 of the overhead column. See --percentage for more info.
136
137 -M, --disassembler-style=
138 Set disassembler style for objdump.
139
140 --prefix=PREFIX, --prefix-strip=N
141 Remove first N entries from source file path names in executables
142 and add PREFIX. This allows to display source code compiled on
143 systems with different file system layout.
144
145 --source
146 Interleave source code with assembly code. Enabled by default,
147 disable with --no-source.
148
149 --asm-raw
150 Show raw instruction encoding of assembly instructions.
151
152 -g
153 Enables call-graph (stack chain/backtrace) recording.
154
155 --call-graph [mode,type,min[,limit],order[,key][,branch]]
156 Setup and enable call-graph (stack chain/backtrace) recording,
157 implies -g. See --call-graph section in perf-record and perf-report
158 man pages for details.
159
160 --children
161 Accumulate callchain of children to parent entry so that then can
162 show up in the output. The output will have a new "Children" column
163 and will be sorted on the data. It requires -g/--call-graph option
164 enabled. See the ‘overhead calculation’ section for more details.
165 Enabled by default, disable with --no-children.
166
167 --max-stack
168 Set the stack depth limit when parsing the callchain, anything
169 beyond the specified depth will be ignored. This is a trade-off
170 between information loss and faster processing especially for
171 workloads that can have a very long callchain stack.
172
173 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
174
175 --ignore-callees=<regex>
176 Ignore callees of the function(s) matching the given regex. This
177 has the effect of collecting the callers of each such function into
178 one place in the call-graph tree.
179
180 --percent-limit
181 Do not show entries which have an overhead under that percent.
182 (Default: 0).
183
184 --percentage
185 Determine how to display the overhead percentage of filtered
186 entries. Filters can be applied by --comms, --dsos and/or --symbols
187 options and Zoom operations on the TUI (thread, dso, etc).
188
189 "relative" means it's relative to filtered entries only so that the
190 sum of shown entries will be always 100%. "absolute" means it retains
191 the original value before and after the filter is applied.
192
193 -w, --column-widths=<width[,width...]>
194 Force each column width to the provided list, for large terminal
195 readability. 0 means no limit (default behavior).
196
197 --proc-map-timeout
198 When processing pre-existing threads /proc/XXX/mmap, it may take a
199 long time, because the file may be huge. A time out is needed in
200 such cases. This option sets the time out limit. The default value
201 is 500 ms.
202
203 -b, --branch-any
204 Enable taken branch stack sampling. Any type of taken branch may be
205 sampled. This is a shortcut for --branch-filter any. See
206 --branch-filter for more infos.
207
208 -j, --branch-filter
209 Enable taken branch stack sampling. Each sample captures a series
210 of consecutive taken branches. The number of branches captured with
211 each sample depends on the underlying hardware, the type of
212 branches of interest, and the executed code. It is possible to
213 select the types of branches captured by enabling filters. For a
214 full list of modifiers please see the perf record manpage.
215
216 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
217 The privilege levels may be omitted, in which case, the privilege levels of the associated
218 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
219 levels are subject to permissions. When sampling on multiple events, branch stack sampling
220 is enabled for all the sampling events. The sampled branch type is the same for all events.
221 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
222 Note that this feature may not be available on all processors.
223
224 --raw-trace
225 When displaying traceevent output, do not use print fmt or plugins.
226
227 --hierarchy
228 Enable hierarchy output.
229
230 --overwrite
231 Enable this to use just the most recent records, which helps in
232 high core count machines such as Knights Landing/Mill, but right
233 now is disabled by default as the pausing used in this technique is
234 leading to loss of metadata events such as PERF_RECORD_MMAP which
235 makes perf top unable to resolve samples, leading to lots of
236 unknown samples appearing on the UI. Enable this if you are in such
237 machines and profiling a workload that doesn’t creates short lived
238 threads and/or doesn’t uses many executable mmap operations. Work
239 is being planed to solve this situation, till then, this will
240 remain disabled by default.
241
242 --force
243 Don’t do ownership validation.
244
245 --num-thread-synthesize
246 The number of threads to run when synthesizing events for existing
247 processes. By default, the number of threads equals to the number
248 of online CPUs.
249
250 --namespaces
251 Record events of type PERF_RECORD_NAMESPACES and display it with
252 the cgroup_id sort key.
253
254 --all-cgroups
255 Record events of type PERF_RECORD_CGROUP and display it with the
256 cgroup sort key.
257
258 --switch-on EVENT_NAME
259 Only consider events after this event is found.
260
261 E.g.:
262
263 Find out where broadcast packets are handled
264
265 perf probe -L icmp_rcv
266
267 Insert a probe there:
268
269 perf probe icmp_rcv:59
270
271 Start perf top and ask it to only consider the cycles events when a
272 broadcast packet arrives This will show a menu with two entries and
273 will start counting when a broadcast packet arrives:
274
275 perf top -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv
276
277 Alternatively one can ask for --group and then two overhead columns
278 will appear, the first for cycles and the second for the switch-on event.
279
280 perf top --group -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv
281
282 This may be interesting to measure a workload only after some initialization
283 phase is over, i.e. insert a perf probe at that point and use the above
284 examples replacing probe:icmp_rcv with the just-after-init probe.
285
286 --switch-off EVENT_NAME
287 Stop considering events after this event is found.
288
289 --show-on-off-events
290 Show the --switch-on/off events too. This has no effect in perf top
291 now but probably we’ll make the default not to show the
292 switch-on/off events on the --group mode and if there is only one
293 event besides the off/on ones, go straight to the histogram
294 browser, just like perf top with no events explicitely specified
295 does.
296
297 --stitch-lbr
298 Show callgraph with stitched LBRs, which may have more complete
299 callgraph. The option must be used with --call-graph lbr recording.
300 Disabled by default. In common cases with call stack overflows, it
301 can recreate better call stacks than the default lbr call stack
302 output. But this approach is not full proof. There can be cases
303 where it creates incorrect call stacks from incorrect matches. The
304 known limitations include exception handing such as setjmp/longjmp
305 will have calls/returns not match.
306
308 [d]
309 Display refresh delay.
310
311 [e]
312 Number of entries to display.
313
314 [E]
315 Event to display when multiple counters are active.
316
317 [f]
318 Profile display filter (>= hit count).
319
320 [F]
321 Annotation display filter (>= % of total).
322
323 [s]
324 Annotate symbol.
325
326 [S]
327 Stop annotation, return to full profile display.
328
329 [K]
330 Hide kernel symbols.
331
332 [U]
333 Hide user symbols.
334
335 [z]
336 Toggle event count zeroing across display updates.
337
338 [qQ]
339 Quit.
340
341 Pressing any unmapped key displays a menu, and prompts for input.
342
344 The overhead can be shown in two columns as Children and Self when perf
345 collects callchains. The self overhead is simply calculated by adding
346 all period values of the entry - usually a function (symbol). This is
347 the value that perf shows traditionally and sum of all the self
348 overhead values should be 100%.
349
350 The children overhead is calculated by adding all period values of the
351 child functions so that it can show the total overhead of the higher
352 level functions even if they don’t directly execute much. Children here
353 means functions that are called from another (parent) function.
354
355 It might be confusing that the sum of all the children overhead values
356 exceeds 100% since each of them is already an accumulation of self
357 overhead of its child functions. But with this enabled, users can find
358 which function has the most overhead even if samples are spread over
359 the children.
360
361 Consider the following example; there are three functions like below.
362
363
364 .ft C
365 void foo(void) {
366 /* do something */
367 }
368
369 void bar(void) {
370 /* do something */
371 foo();
372 }
373
374 int main(void) {
375 bar()
376 return 0;
377 }
378 .ft
379
380
381 In this case foo is a child of bar, and bar is an immediate child of
382 main so foo also is a child of main. In other words, main is a parent
383 of foo and bar, and bar is a parent of foo.
384
385 Suppose all samples are recorded in foo and bar only. When it’s
386 recorded with callchains the output will show something like below in
387 the usual (self-overhead-only) output of perf report:
388
389
390 .ft C
391 Overhead Symbol
392 ........ .....................
393 60.00% foo
394 |
395 --- foo
396 bar
397 main
398 __libc_start_main
399
400 40.00% bar
401 |
402 --- bar
403 main
404 __libc_start_main
405 .ft
406
407
408 When the --children option is enabled, the self overhead values of
409 child functions (i.e. foo and bar) are added to the parents to
410 calculate the children overhead. In this case the report could be
411 displayed as:
412
413
414 .ft C
415 Children Self Symbol
416 ........ ........ ....................
417 100.00% 0.00% __libc_start_main
418 |
419 --- __libc_start_main
420
421 100.00% 0.00% main
422 |
423 --- main
424 __libc_start_main
425
426 100.00% 40.00% bar
427 |
428 --- bar
429 main
430 __libc_start_main
431
432 60.00% 60.00% foo
433 |
434 --- foo
435 bar
436 main
437 __libc_start_main
438 .ft
439
440
441 In the above output, the self overhead of foo (60%) was add to the
442 children overhead of bar, main and __libc_start_main. Likewise, the
443 self overhead of bar (40%) was added to the children overhead of main
444 and \_\_libc_start_main.
445
446 So \_\_libc_start_main and main are shown first since they have same
447 (100%) children overhead (even though they have zero self overhead) and
448 they are the parents of foo and bar.
449
450 Since v3.16 the children overhead is shown by default and the output is
451 sorted by its values. The children overhead is disabled by specifying
452 --no-children option on the command line or by adding report.children =
453 false or top.children = false in the perf config file.
454
456 perf-stat(1), perf-list(1), perf-report(1)
457
458
459
460perf 03/30/2021 PERF-TOP(1)