6 perf-trace - strace inspired tool
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.
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.
26 The following options apply to perf trace; options to perf trace record are
27 found in the perf record man page.
34 System-wide collection from all CPUs.
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
46 Event filter. This option should follow an event selector (-e) which
47 selects tracepoint event(s).
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.
61 Record events on existing process ID (comma separated list).
65 Record events on existing thread ID (comma separated list).
69 Record events in threads owned by uid. Name or number.
73 Record events in threads in a cgroup.
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:
78 perf trace -G A -e sched:*switch
80 Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
81 _and_ sched:sched_switch to the 'A' cgroup, while:
83 perf trace -e sched:*switch -G A
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).
91 perf trace -G A -e sched:*switch -G B
93 the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
97 Filter out events for these pids and for 'trace' itself (comma separated list).
104 Child tasks do not inherit counters.
108 Number of mmap data pages (must be a power of two) or size
109 specification with appended unit character - B/K/M/G. The
110 size is rounded up to have nearest pages power of two value.
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.
120 Show only events that had a duration greater than N.M ms.
123 Accrue thread runtime and provide a summary at the end of the session.
126 Show only syscalls that failed, i.e. that returned < 0.
130 Process events from a given perf data file.
134 Print full timestamp rather time relative to first sample.
137 Show process COMM right beside its ID, on by default, disable with --no-comm.
141 Show only a summary of syscalls by thread with min, max, and average times
142 (in msec) and relative stddev.
146 Show all syscalls followed by a summary by thread with min, max, and
147 average times (in msec) and relative stddev.
150 Show tool stats such as number of times fd->pathname was discovered thru
151 hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
155 Don't complain, do it.
159 Trace pagefaults. Optionally, you can specify whether you want minor,
160 major or all pagefaults. Default value is maj.
163 Trace system calls. This options is enabled by default, disable with
166 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
167 Setup and enable call-graph (stack chain/backtrace) recording.
168 See `--call-graph` section in perf-record and perf-report
169 man pages for details. The ones that are most useful in 'perf trace'
170 are 'dwarf' and 'lbr', where available, try: 'perf trace --call-graph dwarf'.
172 Using this will, for the root user, bump the value of --mmap-pages to 4
173 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
174 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
176 --kernel-syscall-graph::
177 Show the kernel callchains on the syscall exit path.
180 Stop after processing N events. Note that strace-like events are considered
181 only at exit time or when a syscall is interrupted, i.e. in those cases this
182 option is equivalent to the number of lines printed.
184 --switch-on EVENT_NAME::
185 Only consider events after this event is found.
187 --switch-off EVENT_NAME::
188 Stop considering events after this event is found.
190 --show-on-off-events::
191 Show the --switch-on/off events too.
194 Set the stack depth limit when parsing the callchain, anything
195 beyond the specified depth will be ignored. Note that at this point
196 this is just about the presentation part, i.e. the kernel is still
197 not limiting, the overhead of callchains needs to be set via the
198 knobs in --call-graph dwarf.
200 Implies '--call-graph dwarf' when --call-graph not present on the
201 command line, on systems where DWARF unwinding was built in.
203 Default: /proc/sys/kernel/perf_event_max_stack when present for
204 live sessions (without --input/-i), 127 otherwise.
207 Set the stack depth limit when parsing the callchain, anything
208 below the specified depth will be ignored. Disabled by default.
210 Implies '--call-graph dwarf' when --call-graph not present on the
211 command line, on systems where DWARF unwinding was built in.
214 Print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info for the
215 raw_syscalls:sys_{enter,exit} tracepoints, for debugging.
218 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
219 because the file may be huge. A time out is needed in such cases.
220 This option sets the time out limit. The default value is 500 ms.
223 Do sorting on batches of events, use when noticing out of order events that
224 may happen, for instance, when a thread gets migrated to a different CPU
225 while processing a syscall.
227 --libtraceevent_print::
228 Use libtraceevent to print tracepoint arguments. By default 'perf trace' uses
229 the same beautifiers used in the strace-like enter+exit lines to augment the
230 tracepoint arguments.
233 Dump BPF maps setup by events passed via -e, for instance the augmented_raw_syscalls
234 living in tools/perf/examples/bpf/augmented_raw_syscalls.c. For now this
235 dumps just boolean map values and integer keys, in time this will print in hex
236 by default and use BTF when available, as well as use functions to do pretty
237 printing using the existing 'perf trace' syscall arg beautifiers to map integer
238 arguments to strings (pid to comm, syscall id to syscall name, etc).
244 When tracing pagefaults, the format of the trace is as follows:
246 <min|maj>fault [<ip.symbol>+<ip.offset>] => <addr.dso@addr.offset> (<map type><addr level>).
248 - min/maj indicates whether fault event is minor or major;
249 - ip.symbol shows symbol for instruction pointer (the code that generated the
250 fault); if no debug symbols available, perf trace will print raw IP;
251 - addr.dso shows DSO for the faulted address;
252 - map type is either 'd' for non-executable maps or 'x' for executable maps;
253 - addr level is either 'k' for kernel dso or '.' for user dso.
255 For symbols resolution you may need to install debugging symbols.
257 Please be aware that duration is currently always 0 and doesn't reflect actual
258 time it took for fault to be handled!
260 When --verbose specified, perf trace tries to print all available information
261 for both IP and fault address in the form of dso@symbol+offset.
266 Trace only major pagefaults:
268 $ perf trace --no-syscalls -F
270 Trace syscalls, major and minor pagefaults:
274 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
276 As you can see, there was major pagefault in python process, from
277 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
279 Trace the first 4 open, openat or open_by_handle_at syscalls (in the future more syscalls may match here):
281 $ perf trace -e open* --max-events 4
282 [root@jouet perf]# trace -e open* --max-events 4
283 2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
284 2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
285 3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
286 4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
289 Trace the first minor page fault when running a workload:
291 # perf trace -F min --max-stack=7 --max-events 1 sleep 1
292 0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
293 __clear_user ([kernel.kallsyms])
294 load_elf_binary ([kernel.kallsyms])
295 search_binary_handler ([kernel.kallsyms])
296 __do_execve_file.isra.33 ([kernel.kallsyms])
297 __x64_sys_execve ([kernel.kallsyms])
298 do_syscall_64 ([kernel.kallsyms])
299 entry_SYSCALL_64 ([kernel.kallsyms])
302 Trace the next min page page fault to take place on the first CPU:
304 # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
305 0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
306 js::gc::FreeSpan::initAsEmpty (inlined)
307 js::gc::Arena::setAsNotAllocated (inlined)
308 js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
309 js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
310 js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
311 js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
312 js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
313 js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
314 js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
315 JSThinInlineString::new_<(js::AllowGC)1> (inlined)
316 AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
317 js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
318 [0x18b26e6bc2bd] (/tmp/perf-17136.map)
321 Trace the next two sched:sched_switch events, four block:*_plug events, the
322 next block:*_unplug and the next three net:*dev_queue events, this last one
323 with a backtrace of at most 16 entries, system wide:
325 # perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/
326 0.000 :0/0 sched:sched_switch:swapper/2:0 [120] S ==> rcu_sched:10 [120]
327 0.015 rcu_sched/10 sched:sched_switch:rcu_sched:10 [120] R ==> swapper/2:0 [120]
328 254.198 irq/50-iwlwifi/680 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=66
329 __dev_queue_xmit ([kernel.kallsyms])
330 273.977 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=78
331 __dev_queue_xmit ([kernel.kallsyms])
332 274.007 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051ff00 len=78
333 __dev_queue_xmit ([kernel.kallsyms])
334 2930.140 kworker/u16:58/2722 block:block_plug:[kworker/u16:58]
335 2930.162 kworker/u16:58/2722 block:block_unplug:[kworker/u16:58] 1
336 4466.094 jbd2/dm-2-8/748 block:block_plug:[jbd2/dm-2-8]
337 8050.123 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
338 8050.271 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
343 linkperf:perf-record[1], linkperf:perf-script[1]