1PERF-TRACE(1) perf Manual PERF-TRACE(1)
2
3
4
6 perf-trace - strace inspired tool
7
9 perf trace
10 perf trace record
11
13 This command will show the events associated with the target, initially
14 syscalls, but other system events like pagefaults, task lifetime
15 events, scheduling events, etc.
16
17 This is a live mode tool in addition to working with perf.data files
18 like the other perf tools. Files can be generated using the perf record
19 command but the session needs to include the raw_syscalls events (-e
20 raw_syscalls:*). Alternatively, perf trace record can be used as a
21 shortcut to automatically include the raw_syscalls events when writing
22 events to a file.
23
24 The following options apply to perf trace; options to perf trace record
25 are found in the perf record man page.
26
28 -a, --all-cpus
29 System-wide collection from all CPUs.
30
31 -e, --expr, --event
32 List of syscalls and other perf events (tracepoints, HW cache
33 events, etc) to show. Globbing is supported, e.g.: "epoll_*",
34 "msg", etc. See perf list for a complete list of events. Prefixing
35 with ! shows all syscalls but the ones specified. You may need to
36 escape it.
37
38 --filter=<filter>
39 Event filter. This option should follow an event selector (-e)
40 which selects tracepoint event(s).
41
42 -D msecs, --delay msecs
43 After starting the program, wait msecs before measuring. This is
44 useful to filter out the startup phase of the program, which is
45 often very different.
46
47 -o, --output=
48 Output file name.
49
50 -p, --pid=
51 Record events on existing process ID (comma separated list).
52
53 -t, --tid=
54 Record events on existing thread ID (comma separated list).
55
56 -u, --uid=
57 Record events in threads owned by uid. Name or number.
58
59 -G, --cgroup
60 Record events in threads in a cgroup.
61
62 Look for cgroups to set at the /sys/fs/cgroup/perf_event directory, then
63 remove the /sys/fs/cgroup/perf_event/ part and try:
64
65 perf trace -G A -e sched:*switch
66
67 Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
68 _and_ sched:sched_switch to the 'A' cgroup, while:
69
70 perf trace -e sched:*switch -G A
71
72 will only set the sched:sched_switch event to the 'A' cgroup, all the
73 other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
74 a cgroup (on the root cgroup, sys wide, etc).
75
76 Multiple cgroups:
77
78 perf trace -G A -e sched:*switch -G B
79
80 the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
81 to the 'B' cgroup.
82
83 --filter-pids=
84 Filter out events for these pids and for trace itself (comma
85 separated list).
86
87 -v, --verbose
88 Increase the verbosity level.
89
90 --no-inherit
91 Child tasks do not inherit counters.
92
93 -m, --mmap-pages=
94 Number of mmap data pages (must be a power of two) or size
95 specification with appended unit character - B/K/M/G. The size is
96 rounded up to have nearest pages power of two value.
97
98 -C, --cpu
99 Collect samples only on the list of CPUs provided. Multiple CPUs
100 can be provided as a comma-separated list with no space: 0,1.
101 Ranges of CPUs are specified with -: 0-2. In per-thread mode with
102 inheritance mode on (default), Events are captured only when the
103 thread executes on the designated CPUs. Default is to monitor all
104 CPUs.
105
106 --duration
107 Show only events that had a duration greater than N.M ms.
108
109 --sched
110 Accrue thread runtime and provide a summary at the end of the
111 session.
112
113 --failure
114 Show only syscalls that failed, i.e. that returned < 0.
115
116 -i, --input
117 Process events from a given perf data file.
118
119 -T, --time
120 Print full timestamp rather time relative to first sample.
121
122 --comm
123 Show process COMM right beside its ID, on by default, disable with
124 --no-comm.
125
126 -s, --summary
127 Show only a summary of syscalls by thread with min, max, and
128 average times (in msec) and relative stddev.
129
130 -S, --with-summary
131 Show all syscalls followed by a summary by thread with min, max,
132 and average times (in msec) and relative stddev.
133
134 --errno-summary
135 To be used with -s or -S, to show stats for the errnos experienced
136 by syscalls, using only this option will trigger --summary.
137
138 --tool_stats
139 Show tool stats such as number of times fd→pathname was discovered
140 thru hooking the open syscall return + vfs_getname or via reading
141 /proc/pid/fd, etc.
142
143 -f, --force
144 Don’t complain, do it.
145
146 -F=[all|min|maj], --pf=[all|min|maj]
147 Trace pagefaults. Optionally, you can specify whether you want
148 minor, major or all pagefaults. Default value is maj.
149
150 --syscalls
151 Trace system calls. This options is enabled by default, disable
152 with --no-syscalls.
153
154 --call-graph [mode,type,min[,limit],order[,key][,branch]]
155 Setup and enable call-graph (stack chain/backtrace) recording. See
156 --call-graph section in perf-record and perf-report man pages for
157 details. The ones that are most useful in perf trace are dwarf and
158 lbr, where available, try: perf trace --call-graph dwarf.
159
160 Using this will, for the root user, bump the value of --mmap-pages to 4
161 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
162 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
163
164 --kernel-syscall-graph
165 Show the kernel callchains on the syscall exit path.
166
167 --max-events=N
168 Stop after processing N events. Note that strace-like events are
169 considered only at exit time or when a syscall is interrupted, i.e.
170 in those cases this option is equivalent to the number of lines
171 printed.
172
173 --switch-on EVENT_NAME
174 Only consider events after this event is found.
175
176 --switch-off EVENT_NAME
177 Stop considering events after this event is found.
178
179 --show-on-off-events
180 Show the --switch-on/off events too.
181
182 --max-stack
183 Set the stack depth limit when parsing the callchain, anything
184 beyond the specified depth will be ignored. Note that at this point
185 this is just about the presentation part, i.e. the kernel is still
186 not limiting, the overhead of callchains needs to be set via the
187 knobs in --call-graph dwarf.
188
189 Implies '--call-graph dwarf' when --call-graph not present on the
190 command line, on systems where DWARF unwinding was built in.
191
192 Default: /proc/sys/kernel/perf_event_max_stack when present for
193 live sessions (without --input/-i), 127 otherwise.
194
195 --min-stack
196 Set the stack depth limit when parsing the callchain, anything
197 below the specified depth will be ignored. Disabled by default.
198
199 Implies '--call-graph dwarf' when --call-graph not present on the
200 command line, on systems where DWARF unwinding was built in.
201
202 --print-sample
203 Print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info for the
204 raw_syscalls:sys_{enter,exit} tracepoints, for debugging.
205
206 --proc-map-timeout
207 When processing pre-existing threads /proc/XXX/mmap, it may take a
208 long time, because the file may be huge. A time out is needed in
209 such cases. This option sets the time out limit. The default value
210 is 500 ms.
211
212 --sort-events
213 Do sorting on batches of events, use when noticing out of order
214 events that may happen, for instance, when a thread gets migrated
215 to a different CPU while processing a syscall.
216
217 --libtraceevent_print
218 Use libtraceevent to print tracepoint arguments. By default perf
219 trace uses the same beautifiers used in the strace-like enter+exit
220 lines to augment the tracepoint arguments.
221
222 --map-dump
223 Dump BPF maps setup by events passed via -e, for instance the
224 augmented_raw_syscalls living in
225 tools/perf/examples/bpf/augmented_raw_syscalls.c. For now this
226 dumps just boolean map values and integer keys, in time this will
227 print in hex by default and use BTF when available, as well as use
228 functions to do pretty printing using the existing perf trace
229 syscall arg beautifiers to map integer arguments to strings (pid to
230 comm, syscall id to syscall name, etc).
231
233 When tracing pagefaults, the format of the trace is as follows:
234
235 <min|maj>fault [<ip.symbol>+<ip.offset>] ⇒ <addr.dso@addr.offset[1]>
236 (<map type><addr level>).
237
238 • min/maj indicates whether fault event is minor or major;
239
240 • ip.symbol shows symbol for instruction pointer (the code that
241 generated the fault); if no debug symbols available, perf trace
242 will print raw IP;
243
244 • addr.dso shows DSO for the faulted address;
245
246 • map type is either d for non-executable maps or x for executable
247 maps;
248
249 • addr level is either k for kernel dso or . for user dso.
250
251 For symbols resolution you may need to install debugging symbols.
252
253 Please be aware that duration is currently always 0 and doesn’t reflect
254 actual time it took for fault to be handled!
255
256 When --verbose specified, perf trace tries to print all available
257 information for both IP and fault address in the form of
258 dso@symbol[2]+offset.
259
261 Trace only major pagefaults:
262
263 $ perf trace --no-syscalls -F
264
265 Trace syscalls, major and minor pagefaults:
266
267 $ perf trace -F all
268
269 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
270
271 As you can see, there was major pagefault in python process, from
272 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
273
274 Trace the first 4 open, openat or open_by_handle_at syscalls (in the
275 future more syscalls may match here):
276
277 $ perf trace -e open* --max-events 4
278 [root@jouet perf]# trace -e open* --max-events 4
279 2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
280 2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
281 3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
282 4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
283 $
284
285 Trace the first minor page fault when running a workload:
286
287 # perf trace -F min --max-stack=7 --max-events 1 sleep 1
288 0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
289 __clear_user ([kernel.kallsyms])
290 load_elf_binary ([kernel.kallsyms])
291 search_binary_handler ([kernel.kallsyms])
292 __do_execve_file.isra.33 ([kernel.kallsyms])
293 __x64_sys_execve ([kernel.kallsyms])
294 do_syscall_64 ([kernel.kallsyms])
295 entry_SYSCALL_64 ([kernel.kallsyms])
296 #
297
298 Trace the next min page page fault to take place on the first CPU:
299
300 # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
301 0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
302 js::gc::FreeSpan::initAsEmpty (inlined)
303 js::gc::Arena::setAsNotAllocated (inlined)
304 js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
305 js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
306 js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
307 js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
308 js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
309 js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
310 js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
311 JSThinInlineString::new_<(js::AllowGC)1> (inlined)
312 AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
313 js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
314 [0x18b26e6bc2bd] (/tmp/perf-17136.map)
315 #
316
317 Trace the next two sched:sched_switch events, four block:*_plug events,
318 the next block:*_unplug and the next three net:*dev_queue events, this
319 last one with a backtrace of at most 16 entries, system wide:
320
321 # perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/
322 0.000 :0/0 sched:sched_switch:swapper/2:0 [120] S ==> rcu_sched:10 [120]
323 0.015 rcu_sched/10 sched:sched_switch:rcu_sched:10 [120] R ==> swapper/2:0 [120]
324 254.198 irq/50-iwlwifi/680 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=66
325 __dev_queue_xmit ([kernel.kallsyms])
326 273.977 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=78
327 __dev_queue_xmit ([kernel.kallsyms])
328 274.007 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051ff00 len=78
329 __dev_queue_xmit ([kernel.kallsyms])
330 2930.140 kworker/u16:58/2722 block:block_plug:[kworker/u16:58]
331 2930.162 kworker/u16:58/2722 block:block_unplug:[kworker/u16:58] 1
332 4466.094 jbd2/dm-2-8/748 block:block_plug:[jbd2/dm-2-8]
333 8050.123 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
334 8050.271 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
335 #
336
338 perf-record(1), perf-script(1)
339
341 1. addr.dso@addr.offset
342 mailto:addr.dso@addr.offset
343
344 2. dso@symbol
345 mailto:dso@symbol
346
347
348
349perf 06/14/2022 PERF-TRACE(1)