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 --kallsyms=<file>
59 kallsyms pathname
60
61 -m <pages>, --mmap-pages=<pages>
62 Number of mmap data pages (must be a power of two) or size
63 specification with appended unit character - B/K/M/G. The size is
64 rounded up to have nearest pages power of two value.
65
66 -p <pid>, --pid=<pid>
67 Profile events on existing Process ID (comma separated list).
68
69 -t <tid>, --tid=<tid>
70 Profile events on existing thread ID (comma separated list).
71
72 -u, --uid=
73 Record events in threads owned by uid. Name or number.
74
75 -r <priority>, --realtime=<priority>
76 Collect data with this RT SCHED_FIFO priority.
77
78 --sym-annotate=<symbol>
79 Annotate this symbol.
80
81 -K, --hide_kernel_symbols
82 Hide kernel symbols.
83
84 -U, --hide_user_symbols
85 Hide user symbols.
86
87 --demangle-kernel
88 Demangle kernel symbols.
89
90 -D, --dump-symtab
91 Dump the symbol table used for profiling.
92
93 -v, --verbose
94 Be more verbose (show counter open errors, etc).
95
96 -z, --zero
97 Zero history across display updates.
98
99 -s, --sort
100 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
101 local_weight, abort, in_tx, transaction, overhead, sample, period.
102 Please see description of --sort in the perf-report man page.
103
104 --fields=
105 Specify output field - multiple keys can be specified in CSV
106 format. Following fields are available: overhead, overhead_sys,
107 overhead_us, overhead_children, sample and period. Also it can
108 contain any sort key(s).
109
110 By default, every sort keys not specified in --field will be appended
111 automatically.
112
113 -n, --show-nr-samples
114 Show a column with the number of samples.
115
116 --show-total-period
117 Show a column with the sum of periods.
118
119 --dsos
120 Only consider symbols in these dsos. This option will affect the
121 percentage of the overhead column. See --percentage for more info.
122
123 --comms
124 Only consider symbols in these comms. This option will affect the
125 percentage of the overhead column. See --percentage for more info.
126
127 --symbols
128 Only consider these symbols. This option will affect the percentage
129 of the overhead column. See --percentage for more info.
130
131 -M, --disassembler-style=
132 Set disassembler style for objdump.
133
134 --source
135 Interleave source code with assembly code. Enabled by default,
136 disable with --no-source.
137
138 --asm-raw
139 Show raw instruction encoding of assembly instructions.
140
141 -g
142 Enables call-graph (stack chain/backtrace) recording.
143
144 --call-graph [mode,type,min[,limit],order[,key][,branch]]
145 Setup and enable call-graph (stack chain/backtrace) recording,
146 implies -g. See --call-graph section in perf-record and perf-report
147 man pages for details.
148
149 --children
150 Accumulate callchain of children to parent entry so that then can
151 show up in the output. The output will have a new "Children" column
152 and will be sorted on the data. It requires -g/--call-graph option
153 enabled. See the ‘overhead calculation’ section for more details.
154 Enabled by default, disable with --no-children.
155
156 --max-stack
157 Set the stack depth limit when parsing the callchain, anything
158 beyond the specified depth will be ignored. This is a trade-off
159 between information loss and faster processing especially for
160 workloads that can have a very long callchain stack.
161
162 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
163
164 --ignore-callees=<regex>
165 Ignore callees of the function(s) matching the given regex. This
166 has the effect of collecting the callers of each such function into
167 one place in the call-graph tree.
168
169 --percent-limit
170 Do not show entries which have an overhead under that percent.
171 (Default: 0).
172
173 --percentage
174 Determine how to display the overhead percentage of filtered
175 entries. Filters can be applied by --comms, --dsos and/or --symbols
176 options and Zoom operations on the TUI (thread, dso, etc).
177
178 "relative" means it's relative to filtered entries only so that the
179 sum of shown entries will be always 100%. "absolute" means it retains
180 the original value before and after the filter is applied.
181
182 -w, --column-widths=<width[,width...]>
183 Force each column width to the provided list, for large terminal
184 readability. 0 means no limit (default behavior).
185
186 --proc-map-timeout
187 When processing pre-existing threads /proc/XXX/mmap, it may take a
188 long time, because the file may be huge. A time out is needed in
189 such cases. This option sets the time out limit. The default value
190 is 500 ms.
191
192 -b, --branch-any
193 Enable taken branch stack sampling. Any type of taken branch may be
194 sampled. This is a shortcut for --branch-filter any. See
195 --branch-filter for more infos.
196
197 -j, --branch-filter
198 Enable taken branch stack sampling. Each sample captures a series
199 of consecutive taken branches. The number of branches captured with
200 each sample depends on the underlying hardware, the type of
201 branches of interest, and the executed code. It is possible to
202 select the types of branches captured by enabling filters. For a
203 full list of modifiers please see the perf record manpage.
204
205 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
206 The privilege levels may be omitted, in which case, the privilege levels of the associated
207 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
208 levels are subject to permissions. When sampling on multiple events, branch stack sampling
209 is enabled for all the sampling events. The sampled branch type is the same for all events.
210 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
211 Note that this feature may not be available on all processors.
212
213 --raw-trace
214 When displaying traceevent output, do not use print fmt or plugins.
215
216 --hierarchy
217 Enable hierarchy output.
218
219 --overwrite
220 Enable this to use just the most recent records, which helps in
221 high core count machines such as Knights Landing/Mill, but right
222 now is disabled by default as the pausing used in this technique is
223 leading to loss of metadata events such as PERF_RECORD_MMAP which
224 makes perf top unable to resolve samples, leading to lots of
225 unknown samples appearing on the UI. Enable this if you are in such
226 machines and profiling a workload that doesn’t creates short lived
227 threads and/or doesn’t uses many executable mmap operations. Work
228 is being planed to solve this situation, till then, this will
229 remain disabled by default.
230
231 --force
232 Don’t do ownership validation.
233
234 --num-thread-synthesize
235 The number of threads to run when synthesizing events for existing
236 processes. By default, the number of threads equals to the number
237 of online CPUs.
238
239 --namespaces
240 Record events of type PERF_RECORD_NAMESPACES and display it with
241 the cgroup_id sort key.
242
244 [d]
245 Display refresh delay.
246
247 [e]
248 Number of entries to display.
249
250 [E]
251 Event to display when multiple counters are active.
252
253 [f]
254 Profile display filter (>= hit count).
255
256 [F]
257 Annotation display filter (>= % of total).
258
259 [s]
260 Annotate symbol.
261
262 [S]
263 Stop annotation, return to full profile display.
264
265 [K]
266 Hide kernel symbols.
267
268 [U]
269 Hide user symbols.
270
271 [z]
272 Toggle event count zeroing across display updates.
273
274 [qQ]
275 Quit.
276
277 Pressing any unmapped key displays a menu, and prompts for input.
278
280 The overhead can be shown in two columns as Children and Self when perf
281 collects callchains. The self overhead is simply calculated by adding
282 all period values of the entry - usually a function (symbol). This is
283 the value that perf shows traditionally and sum of all the self
284 overhead values should be 100%.
285
286 The children overhead is calculated by adding all period values of the
287 child functions so that it can show the total overhead of the higher
288 level functions even if they don’t directly execute much. Children here
289 means functions that are called from another (parent) function.
290
291 It might be confusing that the sum of all the children overhead values
292 exceeds 100% since each of them is already an accumulation of self
293 overhead of its child functions. But with this enabled, users can find
294 which function has the most overhead even if samples are spread over
295 the children.
296
297 Consider the following example; there are three functions like below.
298
299
300 .ft C
301 void foo(void) {
302 /* do something */
303 }
304
305 void bar(void) {
306 /* do something */
307 foo();
308 }
309
310 int main(void) {
311 bar()
312 return 0;
313 }
314 .ft
315
316
317 In this case foo is a child of bar, and bar is an immediate child of
318 main so foo also is a child of main. In other words, main is a parent
319 of foo and bar, and bar is a parent of foo.
320
321 Suppose all samples are recorded in foo and bar only. When it’s
322 recorded with callchains the output will show something like below in
323 the usual (self-overhead-only) output of perf report:
324
325
326 .ft C
327 Overhead Symbol
328 ........ .....................
329 60.00% foo
330 |
331 --- foo
332 bar
333 main
334 __libc_start_main
335
336 40.00% bar
337 |
338 --- bar
339 main
340 __libc_start_main
341 .ft
342
343
344 When the --children option is enabled, the self overhead values of
345 child functions (i.e. foo and bar) are added to the parents to
346 calculate the children overhead. In this case the report could be
347 displayed as:
348
349
350 .ft C
351 Children Self Symbol
352 ........ ........ ....................
353 100.00% 0.00% __libc_start_main
354 |
355 --- __libc_start_main
356
357 100.00% 0.00% main
358 |
359 --- main
360 __libc_start_main
361
362 100.00% 40.00% bar
363 |
364 --- bar
365 main
366 __libc_start_main
367
368 60.00% 60.00% foo
369 |
370 --- foo
371 bar
372 main
373 __libc_start_main
374 .ft
375
376
377 In the above output, the self overhead of foo (60%) was add to the
378 children overhead of bar, main and __libc_start_main. Likewise, the
379 self overhead of bar (40%) was added to the children overhead of main
380 and \_\_libc_start_main.
381
382 So \_\_libc_start_main and main are shown first since they have same
383 (100%) children overhead (even though they have zero self overhead) and
384 they are the parents of foo and bar.
385
386 Since v3.16 the children overhead is shown by default and the output is
387 sorted by its values. The children overhead is disabled by specifying
388 --no-children option on the command line or by adding report.children =
389 false or top.children = false in the perf config file.
390
392 perf-stat(1), perf-list(1), perf-report(1)
393
394
395
396perf 11/12/2019 PERF-TOP(1)