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.
46
47 -i, --inherit
48 Child tasks do not inherit counters.
49
50 -k <path>, --vmlinux=<path>
51 Path to vmlinux. Required for annotation functionality.
52
53 -m <pages>, --mmap-pages=<pages>
54 Number of mmap data pages (must be a power of two) or size
55 specification with appended unit character - B/K/M/G. The size is
56 rounded up to have nearest pages power of two value.
57
58 -p <pid>, --pid=<pid>
59 Profile events on existing Process ID (comma separated list).
60
61 -t <tid>, --tid=<tid>
62 Profile events on existing thread ID (comma separated list).
63
64 -u, --uid=
65 Record events in threads owned by uid. Name or number.
66
67 -r <priority>, --realtime=<priority>
68 Collect data with this RT SCHED_FIFO priority.
69
70 --sym-annotate=<symbol>
71 Annotate this symbol.
72
73 -K, --hide_kernel_symbols
74 Hide kernel symbols.
75
76 -U, --hide_user_symbols
77 Hide user symbols.
78
79 --demangle-kernel
80 Demangle kernel symbols.
81
82 -D, --dump-symtab
83 Dump the symbol table used for profiling.
84
85 -v, --verbose
86 Be more verbose (show counter open errors, etc).
87
88 -z, --zero
89 Zero history across display updates.
90
91 -s, --sort
92 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
93 local_weight, abort, in_tx, transaction, overhead, sample, period.
94 Please see description of --sort in the perf-report man page.
95
96 --fields=
97 Specify output field - multiple keys can be specified in CSV
98 format. Following fields are available: overhead, overhead_sys,
99 overhead_us, overhead_children, sample and period. Also it can
100 contain any sort key(s).
101
102 By default, every sort keys not specified in --field will be appended
103 automatically.
104
105 -n, --show-nr-samples
106 Show a column with the number of samples.
107
108 --show-total-period
109 Show a column with the sum of periods.
110
111 --dsos
112 Only consider symbols in these dsos. This option will affect the
113 percentage of the overhead column. See --percentage for more info.
114
115 --comms
116 Only consider symbols in these comms. This option will affect the
117 percentage of the overhead column. See --percentage for more info.
118
119 --symbols
120 Only consider these symbols. This option will affect the percentage
121 of the overhead column. See --percentage for more info.
122
123 -M, --disassembler-style=
124 Set disassembler style for objdump.
125
126 --source
127 Interleave source code with assembly code. Enabled by default,
128 disable with --no-source.
129
130 --asm-raw
131 Show raw instruction encoding of assembly instructions.
132
133 -g
134 Enables call-graph (stack chain/backtrace) recording.
135
136 --call-graph [mode,type,min[,limit],order[,key][,branch]]
137 Setup and enable call-graph (stack chain/backtrace) recording,
138 implies -g. See --call-graph section in perf-record and perf-report
139 man pages for details.
140
141 --children
142 Accumulate callchain of children to parent entry so that then can
143 show up in the output. The output will have a new "Children" column
144 and will be sorted on the data. It requires -g/--call-graph option
145 enabled. See the ‘overhead calculation’ section for more details.
146 Enabled by default, disable with --no-children.
147
148 --max-stack
149 Set the stack depth limit when parsing the callchain, anything
150 beyond the specified depth will be ignored. This is a trade-off
151 between information loss and faster processing especially for
152 workloads that can have a very long callchain stack.
153
154 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
155
156 --ignore-callees=<regex>
157 Ignore callees of the function(s) matching the given regex. This
158 has the effect of collecting the callers of each such function into
159 one place in the call-graph tree.
160
161 --percent-limit
162 Do not show entries which have an overhead under that percent.
163 (Default: 0).
164
165 --percentage
166 Determine how to display the overhead percentage of filtered
167 entries. Filters can be applied by --comms, --dsos and/or --symbols
168 options and Zoom operations on the TUI (thread, dso, etc).
169
170 "relative" means it's relative to filtered entries only so that the
171 sum of shown entries will be always 100%. "absolute" means it retains
172 the original value before and after the filter is applied.
173
174 -w, --column-widths=<width[,width...]>
175 Force each column width to the provided list, for large terminal
176 readability. 0 means no limit (default behavior).
177
178 --proc-map-timeout
179 When processing pre-existing threads /proc/XXX/mmap, it may take a
180 long time, because the file may be huge. A time out is needed in
181 such cases. This option sets the time out limit. The default value
182 is 500 ms.
183
184 -b, --branch-any
185 Enable taken branch stack sampling. Any type of taken branch may be
186 sampled. This is a shortcut for --branch-filter any. See
187 --branch-filter for more infos.
188
189 -j, --branch-filter
190 Enable taken branch stack sampling. Each sample captures a series
191 of consecutive taken branches. The number of branches captured with
192 each sample depends on the underlying hardware, the type of
193 branches of interest, and the executed code. It is possible to
194 select the types of branches captured by enabling filters. For a
195 full list of modifiers please see the perf record manpage.
196
197 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
198 The privilege levels may be omitted, in which case, the privilege levels of the associated
199 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
200 levels are subject to permissions. When sampling on multiple events, branch stack sampling
201 is enabled for all the sampling events. The sampled branch type is the same for all events.
202 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
203 Note that this feature may not be available on all processors.
204
205 --raw-trace
206 When displaying traceevent output, do not use print fmt or plugins.
207
208 --hierarchy
209 Enable hierarchy output.
210
211 --force
212 Don’t do ownership validation.
213
214 --num-thread-synthesize
215 The number of threads to run when synthesizing events for existing
216 processes. By default, the number of threads equals to the number
217 of online CPUs.
218
220 [d]
221 Display refresh delay.
222
223 [e]
224 Number of entries to display.
225
226 [E]
227 Event to display when multiple counters are active.
228
229 [f]
230 Profile display filter (>= hit count).
231
232 [F]
233 Annotation display filter (>= % of total).
234
235 [s]
236 Annotate symbol.
237
238 [S]
239 Stop annotation, return to full profile display.
240
241 [z]
242 Toggle event count zeroing across display updates.
243
244 [qQ]
245 Quit.
246
247 Pressing any unmapped key displays a menu, and prompts for input.
248
250 The overhead can be shown in two columns as Children and Self when perf
251 collects callchains. The self overhead is simply calculated by adding
252 all period values of the entry - usually a function (symbol). This is
253 the value that perf shows traditionally and sum of all the self
254 overhead values should be 100%.
255
256 The children overhead is calculated by adding all period values of the
257 child functions so that it can show the total overhead of the higher
258 level functions even if they don’t directly execute much. Children here
259 means functions that are called from another (parent) function.
260
261 It might be confusing that the sum of all the children overhead values
262 exceeds 100% since each of them is already an accumulation of self
263 overhead of its child functions. But with this enabled, users can find
264 which function has the most overhead even if samples are spread over
265 the children.
266
267 Consider the following example; there are three functions like below.
268
269
270 .ft C
271 void foo(void) {
272 /* do something */
273 }
274
275 void bar(void) {
276 /* do something */
277 foo();
278 }
279
280 int main(void) {
281 bar()
282 return 0;
283 }
284 .ft
285
286
287 In this case foo is a child of bar, and bar is an immediate child of
288 main so foo also is a child of main. In other words, main is a parent
289 of foo and bar, and bar is a parent of foo.
290
291 Suppose all samples are recorded in foo and bar only. When it’s
292 recorded with callchains the output will show something like below in
293 the usual (self-overhead-only) output of perf report:
294
295
296 .ft C
297 Overhead Symbol
298 ........ .....................
299 60.00% foo
300 |
301 --- foo
302 bar
303 main
304 __libc_start_main
305
306 40.00% bar
307 |
308 --- bar
309 main
310 __libc_start_main
311 .ft
312
313
314 When the --children option is enabled, the self overhead values of
315 child functions (i.e. foo and bar) are added to the parents to
316 calculate the children overhead. In this case the report could be
317 displayed as:
318
319
320 .ft C
321 Children Self Symbol
322 ........ ........ ....................
323 100.00% 0.00% __libc_start_main
324 |
325 --- __libc_start_main
326
327 100.00% 0.00% main
328 |
329 --- main
330 __libc_start_main
331
332 100.00% 40.00% bar
333 |
334 --- bar
335 main
336 __libc_start_main
337
338 60.00% 60.00% foo
339 |
340 --- foo
341 bar
342 main
343 __libc_start_main
344 .ft
345
346
347 In the above output, the self overhead of foo (60%) was add to the
348 children overhead of bar, main and __libc_start_main. Likewise, the
349 self overhead of bar (40%) was added to the children overhead of main
350 and \_\_libc_start_main.
351
352 So \_\_libc_start_main and main are shown first since they have same
353 (100%) children overhead (even though they have zero self overhead) and
354 they are the parents of foo and bar.
355
356 Since v3.16 the children overhead is shown by default and the output is
357 sorted by its values. The children overhead is disabled by specifying
358 --no-children option on the command line or by adding report.children =
359 false or top.children = false in the perf config file.
360
362 perf-stat(1), perf-list(1), perf-report(1)
363
364
365
366perf 06/18/2019 PERF-TOP(1)