1 perf-trace(1)
2 =============
3
4 NAME
5 ----
6 perf-trace - strace inspired tool
7
8 SYNOPSIS
9 --------
10 [verse]
11 'perf trace'
12 'perf trace record'
13
14 DESCRIPTION
15 -----------
16 This command will show the events associated with the target, initially
17 syscalls, but other system events like pagefaults, task lifetime events,
18 scheduling events, etc.
19
20 This is a live mode tool in addition to working with perf.data files like
21 the other perf tools. Files can be generated using the 'perf record' command
22 but the session needs to include the raw_syscalls events (-e 'raw_syscalls:*').
23 Alternatively, 'perf trace record' can be used as a shortcut to
24 automatically include the raw_syscalls events when writing events to a file.
25
26 The following options apply to perf trace; options to perf trace record are
27 found in the perf record man page.
28
29 OPTIONS
30 -------
31
32 -a::
33 --all-cpus::
34 System-wide collection from all CPUs.
35
36 -e::
37 --expr::
38 --event::
39 List of syscalls and other perf events (tracepoints, HW cache events,
40 etc) to show. Globbing is supported, e.g.: "epoll_*", "*msg*", etc.
41 See 'perf list' for a complete list of events.
42 Prefixing with ! shows all syscalls but the ones specified. You may
43 need to escape it.
44
45 --filter=<filter>::
46 Event filter. This option should follow an event selector (-e) which
47 selects tracepoint event(s).
48
49
50 -D msecs::
51 --delay msecs::
52 After starting the program, wait msecs before measuring. This is useful to
53 filter out the startup phase of the program, which is often very different.
54
55 -o::
56 --output=::
57 Output file name.
58
59 -p::
60 --pid=::
61 Record events on existing process ID (comma separated list).
62
63 -t::
64 --tid=::
65 Record events on existing thread ID (comma separated list).
66
67 -u::
68 --uid=::
69 Record events in threads owned by uid. Name or number.
70
71 -G::
72 --cgroup::
73 Record events in threads in a cgroup.
74
75 Look for cgroups to set at the /sys/fs/cgroup/perf_event directory, then
76 remove the /sys/fs/cgroup/perf_event/ part and try:
77
78 perf trace -G A -e sched:*switch
79
80 Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
81 _and_ sched:sched_switch to the 'A' cgroup, while:
82
83 perf trace -e sched:*switch -G A
84
85 will only set the sched:sched_switch event to the 'A' cgroup, all the
86 other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
87 a cgroup (on the root cgroup, sys wide, etc).
88
89 Multiple cgroups:
90
91 perf trace -G A -e sched:*switch -G B
92
93 the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
94 to the 'B' cgroup.
95
96 --filter-pids=::
97 Filter out events for these pids and for 'trace' itself (comma separated list).
98
99 -v::
100 --verbose::
101 Increase the verbosity level.
102
103 --no-inherit::
104 Child tasks do not inherit counters.
105
106 -m::
107 --mmap-pages=::
108 Number of mmap data pages (must be a power of two) or size
109 specification in bytes with appended unit character - B/K/M/G.
110 The size is rounded up to the nearest power-of-two page value.
111
112 -C::
113 --cpu::
114 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
115 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
116 In per-thread mode with inheritance mode on (default), Events are captured only when
117 the thread executes on the designated CPUs. Default is to monitor all CPUs.
118
119 --duration::
120 Show only events that had a duration greater than N.M ms.
121
122 --sched::
123 Accrue thread runtime and provide a summary at the end of the session.
124
125 --failure::
126 Show only syscalls that failed, i.e. that returned < 0.
127
128 -i::
129 --input::
130 Process events from a given perf data file.
131
132 -T::
133 --time::
134 Print full timestamp rather time relative to first sample.
135
136 --comm::
137 Show process COMM right beside its ID, on by default, disable with --no-comm.
138
139 -s::
140 --summary::
141 Show only a summary of syscalls by thread with min, max, and average times
142 (in msec) and relative stddev.
143
144 -S::
145 --with-summary::
146 Show all syscalls followed by a summary by thread with min, max, and
147 average times (in msec) and relative stddev.
148
149 --errno-summary::
150 To be used with -s or -S, to show stats for the errnos experienced by
151 syscalls, using only this option will trigger --summary.
152
153 --tool_stats::
154 Show tool stats such as number of times fd->pathname was discovered thru
155 hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
156
157 -f::
158 --force::
159 Don't complain, do it.
160
161 -F=[all|min|maj]::
162 --pf=[all|min|maj]::
163 Trace pagefaults. Optionally, you can specify whether you want minor,
164 major or all pagefaults. Default value is maj.
165
166 --syscalls::
167 Trace system calls. This options is enabled by default, disable with
168 --no-syscalls.
169
170 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
171 Setup and enable call-graph (stack chain/backtrace) recording.
172 See `--call-graph` section in perf-record and perf-report
173 man pages for details. The ones that are most useful in 'perf trace'
174 are 'dwarf' and 'lbr', where available, try: 'perf trace --call-graph dwarf'.
175
176 Using this will, for the root user, bump the value of --mmap-pages to 4
177 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
178 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
179
180 --kernel-syscall-graph::
181 Show the kernel callchains on the syscall exit path.
182
183 --max-events=N::
184 Stop after processing N events. Note that strace-like events are considered
185 only at exit time or when a syscall is interrupted, i.e. in those cases this
186 option is equivalent to the number of lines printed.
187
188 --switch-on EVENT_NAME::
189 Only consider events after this event is found.
190
191 --switch-off EVENT_NAME::
192 Stop considering events after this event is found.
193
194 --show-on-off-events::
195 Show the --switch-on/off events too.
196
197 --max-stack::
198 Set the stack depth limit when parsing the callchain, anything
199 beyond the specified depth will be ignored. Note that at this point
200 this is just about the presentation part, i.e. the kernel is still
201 not limiting, the overhead of callchains needs to be set via the
202 knobs in --call-graph dwarf.
203
204 Implies '--call-graph dwarf' when --call-graph not present on the
205 command line, on systems where DWARF unwinding was built in.
206
207 Default: /proc/sys/kernel/perf_event_max_stack when present for
208 live sessions (without --input/-i), 127 otherwise.
209
210 --min-stack::
211 Set the stack depth limit when parsing the callchain, anything
212 below the specified depth will be ignored. Disabled by default.
213
214 Implies '--call-graph dwarf' when --call-graph not present on the
215 command line, on systems where DWARF unwinding was built in.
216
217 --print-sample::
218 Print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info for the
219 raw_syscalls:sys_{enter,exit} tracepoints, for debugging.
220
221 --proc-map-timeout::
222 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
223 because the file may be huge. A time out is needed in such cases.
224 This option sets the time out limit. The default value is 500 ms.
225
226 --sort-events::
227 Do sorting on batches of events, use when noticing out of order events that
228 may happen, for instance, when a thread gets migrated to a different CPU
229 while processing a syscall.
230
231 --libtraceevent_print::
232 Use libtraceevent to print tracepoint arguments. By default 'perf trace' uses
233 the same beautifiers used in the strace-like enter+exit lines to augment the
234 tracepoint arguments.
235
236 --map-dump::
237 Dump BPF maps setup by events passed via -e, for instance the augmented_raw_syscalls
238 living in tools/perf/examples/bpf/augmented_raw_syscalls.c. For now this
239 dumps just boolean map values and integer keys, in time this will print in hex
240 by default and use BTF when available, as well as use functions to do pretty
241 printing using the existing 'perf trace' syscall arg beautifiers to map integer
242 arguments to strings (pid to comm, syscall id to syscall name, etc).
243
244
245 PAGEFAULTS
246 ----------
247
248 When tracing pagefaults, the format of the trace is as follows:
249
250 <min|maj>fault [<ip.symbol>+<ip.offset>] => <addr.dso@addr.offset> (<map type><addr level>).
251
252 - min/maj indicates whether fault event is minor or major;
253 - ip.symbol shows symbol for instruction pointer (the code that generated the
254 fault); if no debug symbols available, perf trace will print raw IP;
255 - addr.dso shows DSO for the faulted address;
256 - map type is either 'd' for non-executable maps or 'x' for executable maps;
257 - addr level is either 'k' for kernel dso or '.' for user dso.
258
259 For symbols resolution you may need to install debugging symbols.
260
261 Please be aware that duration is currently always 0 and doesn't reflect actual
262 time it took for fault to be handled!
263
264 When --verbose specified, perf trace tries to print all available information
265 for both IP and fault address in the form of dso@symbol+offset.
266
267 EXAMPLES
268 --------
269
270 Trace only major pagefaults:
271
272 $ perf trace --no-syscalls -F
273
274 Trace syscalls, major and minor pagefaults:
275
276 $ perf trace -F all
277
278 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
279
280 As you can see, there was major pagefault in python process, from
281 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
282
283 Trace the first 4 open, openat or open_by_handle_at syscalls (in the future more syscalls may match here):
284
285 $ perf trace -e open* --max-events 4
286 [root@jouet perf]# trace -e open* --max-events 4
287 2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
288 2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
289 3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
290 4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
291 $
292
293 Trace the first minor page fault when running a workload:
294
295 # perf trace -F min --max-stack=7 --max-events 1 sleep 1
296 0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
297 __clear_user ([kernel.kallsyms])
298 load_elf_binary ([kernel.kallsyms])
299 search_binary_handler ([kernel.kallsyms])
300 __do_execve_file.isra.33 ([kernel.kallsyms])
301 __x64_sys_execve ([kernel.kallsyms])
302 do_syscall_64 ([kernel.kallsyms])
303 entry_SYSCALL_64 ([kernel.kallsyms])
304 #
305
306 Trace the next min page page fault to take place on the first CPU:
307
308 # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
309 0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
310 js::gc::FreeSpan::initAsEmpty (inlined)
311 js::gc::Arena::setAsNotAllocated (inlined)
312 js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
313 js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
314 js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
315 js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
316 js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
317 js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
318 js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
319 JSThinInlineString::new_<(js::AllowGC)1> (inlined)
320 AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
321 js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
322 [0x18b26e6bc2bd] (/tmp/perf-17136.map)
323 #
324
325 Trace the next two sched:sched_switch events, four block:*_plug events, the
326 next block:*_unplug and the next three net:*dev_queue events, this last one
327 with a backtrace of at most 16 entries, system wide:
328
329 # perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/
330 0.000 :0/0 sched:sched_switch:swapper/2:0 [120] S ==> rcu_sched:10 [120]
331 0.015 rcu_sched/10 sched:sched_switch:rcu_sched:10 [120] R ==> swapper/2:0 [120]
332 254.198 irq/50-iwlwifi/680 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=66
333 __dev_queue_xmit ([kernel.kallsyms])
334 273.977 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=78
335 __dev_queue_xmit ([kernel.kallsyms])
336 274.007 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051ff00 len=78
337 __dev_queue_xmit ([kernel.kallsyms])
338 2930.140 kworker/u16:58/2722 block:block_plug:[kworker/u16:58]
339 2930.162 kworker/u16:58/2722 block:block_unplug:[kworker/u16:58] 1
340 4466.094 jbd2/dm-2-8/748 block:block_plug:[jbd2/dm-2-8]
341 8050.123 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
342 8050.271 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
343 #
344
345 SEE ALSO
346 --------
347 linkperf:perf-record[1], linkperf:perf-script[1]
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.