1PERF-TRACE(1) perf Manual PERF-TRACE(1)
2
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 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 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 -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 -D msecs, --delay msecs
39 After starting the program, wait msecs before measuring. This is
40 useful to filter out the startup phase of the program, which is
41 often very different.
42 -o, --output=
44 Output file name.
45 -p, --pid=
47 Record events on existing process ID (comma separated list).
48 -t, --tid=
50 Record events on existing thread ID (comma separated list).
51 -u, --uid=
53 Record events in threads owned by uid. Name or number.
54 -G, --cgroup
56 Record events in threads in a cgroup.
57 Look for cgroups to set at the /sys/fs/cgroup/perf_event directory, then
59 remove the /sys/fs/cgroup/perf_event/ part and try:
60 perf trace -G A -e sched:*switch
62 Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
64 _and_ sched:sched_switch to the 'A' cgroup, while:
65 perf trace -e sched:*switch -G A
67 will only set the sched:sched_switch event to the 'A' cgroup, all the
69 other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
70 a cgroup (on the root cgroup, sys wide, etc).
71 Multiple cgroups:
73 perf trace -G A -e sched:*switch -G B
75 the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
77 to the 'B' cgroup.
78 --filter-pids=
80 Filter out events for these pids and for trace itself (comma
81 separated list).
82 -v, --verbose=
84 Verbosity level.
85 --no-inherit
87 Child tasks do not inherit counters.
88 -m, --mmap-pages=
90 Number of mmap data pages (must be a power of two) or size
91 specification with appended unit character - B/K/M/G. The size is
92 rounded up to have nearest pages power of two value.
93 -C, --cpu
95 Collect samples only on the list of CPUs provided. Multiple CPUs
96 can be provided as a comma-separated list with no space: 0,1.
97 Ranges of CPUs are specified with -: 0-2. In per-thread mode with
98 inheritance mode on (default), Events are captured only when the
99 thread executes on the designated CPUs. Default is to monitor all
100 CPUs.
101 --duration
103 Show only events that had a duration greater than N.M ms.
104 --sched
106 Accrue thread runtime and provide a summary at the end of the
107 session.
108 --failure
110 Show only syscalls that failed, i.e. that returned < 0.
111 -i, --input
113 Process events from a given perf data file.
114 -T, --time
116 Print full timestamp rather time relative to first sample.
117 --comm
119 Show process COMM right beside its ID, on by default, disable with
120 --no-comm.
121 -s, --summary
123 Show only a summary of syscalls by thread with min, max, and
124 average times (in msec) and relative stddev.
125 -S, --with-summary
127 Show all syscalls followed by a summary by thread with min, max,
128 and average times (in msec) and relative stddev.
129 --tool_stats
131 Show tool stats such as number of times fd→pathname was discovered
132 thru hooking the open syscall return + vfs_getname or via reading
133 /proc/pid/fd, etc.
134 -f, --force
136 Don’t complain, do it.
137 -F=[all|min|maj], --pf=[all|min|maj]
139 Trace pagefaults. Optionally, you can specify whether you want
140 minor, major or all pagefaults. Default value is maj.
141 --syscalls
143 Trace system calls. This options is enabled by default, disable
144 with --no-syscalls.
145 --call-graph [mode,type,min[,limit],order[,key][,branch]]
147 Setup and enable call-graph (stack chain/backtrace) recording. See
148 --call-graph section in perf-record and perf-report man pages for
149 details. The ones that are most useful in perf trace are dwarf and
150 lbr, where available, try: perf trace --call-graph dwarf.
151 Using this will, for the root user, bump the value of --mmap-pages to 4
153 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
154 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
155 --kernel-syscall-graph
157 Show the kernel callchains on the syscall exit path.
158 --max-events=N
160 Stop after processing N events. Note that strace-like events are
161 considered only at exit time or when a syscall is interrupted, i.e.
162 in those cases this option is equivalent to the number of lines
163 printed.
164 --max-stack
166 Set the stack depth limit when parsing the callchain, anything
167 beyond the specified depth will be ignored. Note that at this point
168 this is just about the presentation part, i.e. the kernel is still
169 not limiting, the overhead of callchains needs to be set via the
170 knobs in --call-graph dwarf.
171 Implies '--call-graph dwarf' when --call-graph not present on the
173 command line, on systems where DWARF unwinding was built in.
174 Default: /proc/sys/kernel/perf_event_max_stack when present for
176 live sessions (without --input/-i), 127 otherwise.
177 --min-stack
179 Set the stack depth limit when parsing the callchain, anything
180 below the specified depth will be ignored. Disabled by default.
181 Implies '--call-graph dwarf' when --call-graph not present on the
183 command line, on systems where DWARF unwinding was built in.
184 --print-sample
186 Print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info for the
187 raw_syscalls:sys_{enter,exit} tracepoints, for debugging.
188 --proc-map-timeout
190 When processing pre-existing threads /proc/XXX/mmap, it may take a
191 long time, because the file may be huge. A time out is needed in
192 such cases. This option sets the time out limit. The default value
193 is 500 ms.
194 --sort-events
196 Do sorting on batches of events, use when noticing out of order
197 events that may happen, for instance, when a thread gets migrated
198 to a different CPU while processing a syscall.
199 --map-dump
201 Dump BPF maps setup by events passed via -e, for instance the
202 augmented_raw_syscalls living in
203 tools/perf/examples/bpf/augmented_raw_syscalls.c. For now this
204 dumps just boolean map values and integer keys, in time this will
205 print in hex by default and use BTF when available, as well as use
206 functions to do pretty printing using the existing perf trace
207 syscall arg beautifiers to map integer arguments to strings (pid to
208 comm, syscall id to syscall name, etc).
209
211 When tracing pagefaults, the format of the trace is as follows:
212 <min|maj>fault [<ip.symbol>+<ip.offset>] ⇒ <addr.dso@addr.offset[1]>
214 (<map type><addr level>).
215 · min/maj indicates whether fault event is minor or major;
217 · ip.symbol shows symbol for instruction pointer (the code that
219 generated the fault); if no debug symbols available, perf trace
220 will print raw IP;
221 · addr.dso shows DSO for the faulted address;
223 · map type is either d for non-executable maps or x for executable
225 maps;
226 · addr level is either k for kernel dso or . for user dso.
228 For symbols resolution you may need to install debugging symbols.
230 Please be aware that duration is currently always 0 and doesn’t reflect
232 actual time it took for fault to be handled!
233 When --verbose specified, perf trace tries to print all available
235 information for both IP and fault address in the form of
236 dso@symbol[2]+offset.
237
239 Trace only major pagefaults:
240 $ perf trace --no-syscalls -F
242 Trace syscalls, major and minor pagefaults:
244 $ perf trace -F all
246 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
248 As you can see, there was major pagefault in python process, from
250 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
251 Trace the first 4 open, openat or open_by_handle_at syscalls (in the
253 future more syscalls may match here):
254 $ perf trace -e open* --max-events 4
256 [root@jouet perf]# trace -e open* --max-events 4
257 2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
258 2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
259 3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
260 4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
261 $
262 Trace the first minor page fault when running a workload:
264 # perf trace -F min --max-stack=7 --max-events 1 sleep 1
266 0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
267 __clear_user ([kernel.kallsyms])
268 load_elf_binary ([kernel.kallsyms])
269 search_binary_handler ([kernel.kallsyms])
270 __do_execve_file.isra.33 ([kernel.kallsyms])
271 __x64_sys_execve ([kernel.kallsyms])
272 do_syscall_64 ([kernel.kallsyms])
273 entry_SYSCALL_64 ([kernel.kallsyms])
274 #
275 Trace the next min page page fault to take place on the first CPU:
277 # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
279 0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
280 js::gc::FreeSpan::initAsEmpty (inlined)
281 js::gc::Arena::setAsNotAllocated (inlined)
282 js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
283 js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
284 js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
285 js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
286 js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
287 js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
288 js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
289 JSThinInlineString::new_<(js::AllowGC)1> (inlined)
290 AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
291 js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
292 [0x18b26e6bc2bd] (/tmp/perf-17136.map)
293 #
294 Trace the next two sched:sched_switch events, four block:*_plug events,
296 the next block:*_unplug and the next three net:*dev_queue events, this
297 last one with a backtrace of at most 16 entries, system wide:
298 # perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/
300 0.000 :0/0 sched:sched_switch:swapper/2:0 [120] S ==> rcu_sched:10 [120]
301 0.015 rcu_sched/10 sched:sched_switch:rcu_sched:10 [120] R ==> swapper/2:0 [120]
302 254.198 irq/50-iwlwifi/680 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=66
303 __dev_queue_xmit ([kernel.kallsyms])
304 273.977 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=78
305 __dev_queue_xmit ([kernel.kallsyms])
306 274.007 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051ff00 len=78
307 __dev_queue_xmit ([kernel.kallsyms])
308 2930.140 kworker/u16:58/2722 block:block_plug:[kworker/u16:58]
309 2930.162 kworker/u16:58/2722 block:block_unplug:[kworker/u16:58] 1
310 4466.094 jbd2/dm-2-8/748 block:block_plug:[jbd2/dm-2-8]
311 8050.123 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
312 8050.271 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
313 #
314
316 perf-record(1), perf-script(1)
317
319 1. addr.dso@addr.offset
320 mailto:addr.dso@addr.offset
321 2. dso@symbol
323 mailto:dso@symbol
324perf 11/12/2019 PERF-TRACE(1)