0001 perf-trace(1)
0002 =============
0003
0004 NAME
0005 ----
0006 perf-trace - strace inspired tool
0007
0008 SYNOPSIS
0009 --------
0010 [verse]
0011 'perf trace'
0012 'perf trace record'
0013
0014 DESCRIPTION
0015 -----------
0016 This command will show the events associated with the target, initially
0017 syscalls, but other system events like pagefaults, task lifetime events,
0018 scheduling events, etc.
0019
0020 This is a live mode tool in addition to working with perf.data files like
0021 the other perf tools. Files can be generated using the 'perf record' command
0022 but the session needs to include the raw_syscalls events (-e 'raw_syscalls:*').
0023 Alternatively, 'perf trace record' can be used as a shortcut to
0024 automatically include the raw_syscalls events when writing events to a file.
0025
0026 The following options apply to perf trace; options to perf trace record are
0027 found in the perf record man page.
0028
0029 OPTIONS
0030 -------
0031
0032 -a::
0033 --all-cpus::
0034 System-wide collection from all CPUs.
0035
0036 -e::
0037 --expr::
0038 --event::
0039 List of syscalls and other perf events (tracepoints, HW cache events,
0040 etc) to show. Globbing is supported, e.g.: "epoll_*", "*msg*", etc.
0041 See 'perf list' for a complete list of events.
0042 Prefixing with ! shows all syscalls but the ones specified. You may
0043 need to escape it.
0044
0045 --filter=<filter>::
0046 Event filter. This option should follow an event selector (-e) which
0047 selects tracepoint event(s).
0048
0049
0050 -D msecs::
0051 --delay msecs::
0052 After starting the program, wait msecs before measuring. This is useful to
0053 filter out the startup phase of the program, which is often very different.
0054
0055 -o::
0056 --output=::
0057 Output file name.
0058
0059 -p::
0060 --pid=::
0061 Record events on existing process ID (comma separated list).
0062
0063 -t::
0064 --tid=::
0065 Record events on existing thread ID (comma separated list).
0066
0067 -u::
0068 --uid=::
0069 Record events in threads owned by uid. Name or number.
0070
0071 -G::
0072 --cgroup::
0073 Record events in threads in a cgroup.
0074
0075 Look for cgroups to set at the /sys/fs/cgroup/perf_event directory, then
0076 remove the /sys/fs/cgroup/perf_event/ part and try:
0077
0078 perf trace -G A -e sched:*switch
0079
0080 Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
0081 _and_ sched:sched_switch to the 'A' cgroup, while:
0082
0083 perf trace -e sched:*switch -G A
0084
0085 will only set the sched:sched_switch event to the 'A' cgroup, all the
0086 other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
0087 a cgroup (on the root cgroup, sys wide, etc).
0088
0089 Multiple cgroups:
0090
0091 perf trace -G A -e sched:*switch -G B
0092
0093 the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
0094 to the 'B' cgroup.
0095
0096 --filter-pids=::
0097 Filter out events for these pids and for 'trace' itself (comma separated list).
0098
0099 -v::
0100 --verbose::
0101 Increase the verbosity level.
0102
0103 --no-inherit::
0104 Child tasks do not inherit counters.
0105
0106 -m::
0107 --mmap-pages=::
0108 Number of mmap data pages (must be a power of two) or size
0109 specification with appended unit character - B/K/M/G. The
0110 size is rounded up to have nearest pages power of two value.
0111
0112 -C::
0113 --cpu::
0114 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
0115 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
0116 In per-thread mode with inheritance mode on (default), Events are captured only when
0117 the thread executes on the designated CPUs. Default is to monitor all CPUs.
0118
0119 --duration::
0120 Show only events that had a duration greater than N.M ms.
0121
0122 --sched::
0123 Accrue thread runtime and provide a summary at the end of the session.
0124
0125 --failure::
0126 Show only syscalls that failed, i.e. that returned < 0.
0127
0128 -i::
0129 --input::
0130 Process events from a given perf data file.
0131
0132 -T::
0133 --time::
0134 Print full timestamp rather time relative to first sample.
0135
0136 --comm::
0137 Show process COMM right beside its ID, on by default, disable with --no-comm.
0138
0139 -s::
0140 --summary::
0141 Show only a summary of syscalls by thread with min, max, and average times
0142 (in msec) and relative stddev.
0143
0144 -S::
0145 --with-summary::
0146 Show all syscalls followed by a summary by thread with min, max, and
0147 average times (in msec) and relative stddev.
0148
0149 --errno-summary::
0150 To be used with -s or -S, to show stats for the errnos experienced by
0151 syscalls, using only this option will trigger --summary.
0152
0153 --tool_stats::
0154 Show tool stats such as number of times fd->pathname was discovered thru
0155 hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
0156
0157 -f::
0158 --force::
0159 Don't complain, do it.
0160
0161 -F=[all|min|maj]::
0162 --pf=[all|min|maj]::
0163 Trace pagefaults. Optionally, you can specify whether you want minor,
0164 major or all pagefaults. Default value is maj.
0165
0166 --syscalls::
0167 Trace system calls. This options is enabled by default, disable with
0168 --no-syscalls.
0169
0170 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
0171 Setup and enable call-graph (stack chain/backtrace) recording.
0172 See `--call-graph` section in perf-record and perf-report
0173 man pages for details. The ones that are most useful in 'perf trace'
0174 are 'dwarf' and 'lbr', where available, try: 'perf trace --call-graph dwarf'.
0175
0176 Using this will, for the root user, bump the value of --mmap-pages to 4
0177 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
0178 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
0179
0180 --kernel-syscall-graph::
0181 Show the kernel callchains on the syscall exit path.
0182
0183 --max-events=N::
0184 Stop after processing N events. Note that strace-like events are considered
0185 only at exit time or when a syscall is interrupted, i.e. in those cases this
0186 option is equivalent to the number of lines printed.
0187
0188 --switch-on EVENT_NAME::
0189 Only consider events after this event is found.
0190
0191 --switch-off EVENT_NAME::
0192 Stop considering events after this event is found.
0193
0194 --show-on-off-events::
0195 Show the --switch-on/off events too.
0196
0197 --max-stack::
0198 Set the stack depth limit when parsing the callchain, anything
0199 beyond the specified depth will be ignored. Note that at this point
0200 this is just about the presentation part, i.e. the kernel is still
0201 not limiting, the overhead of callchains needs to be set via the
0202 knobs in --call-graph dwarf.
0203
0204 Implies '--call-graph dwarf' when --call-graph not present on the
0205 command line, on systems where DWARF unwinding was built in.
0206
0207 Default: /proc/sys/kernel/perf_event_max_stack when present for
0208 live sessions (without --input/-i), 127 otherwise.
0209
0210 --min-stack::
0211 Set the stack depth limit when parsing the callchain, anything
0212 below the specified depth will be ignored. Disabled by default.
0213
0214 Implies '--call-graph dwarf' when --call-graph not present on the
0215 command line, on systems where DWARF unwinding was built in.
0216
0217 --print-sample::
0218 Print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info for the
0219 raw_syscalls:sys_{enter,exit} tracepoints, for debugging.
0220
0221 --proc-map-timeout::
0222 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
0223 because the file may be huge. A time out is needed in such cases.
0224 This option sets the time out limit. The default value is 500 ms.
0225
0226 --sort-events::
0227 Do sorting on batches of events, use when noticing out of order events that
0228 may happen, for instance, when a thread gets migrated to a different CPU
0229 while processing a syscall.
0230
0231 --libtraceevent_print::
0232 Use libtraceevent to print tracepoint arguments. By default 'perf trace' uses
0233 the same beautifiers used in the strace-like enter+exit lines to augment the
0234 tracepoint arguments.
0235
0236 --map-dump::
0237 Dump BPF maps setup by events passed via -e, for instance the augmented_raw_syscalls
0238 living in tools/perf/examples/bpf/augmented_raw_syscalls.c. For now this
0239 dumps just boolean map values and integer keys, in time this will print in hex
0240 by default and use BTF when available, as well as use functions to do pretty
0241 printing using the existing 'perf trace' syscall arg beautifiers to map integer
0242 arguments to strings (pid to comm, syscall id to syscall name, etc).
0243
0244
0245 PAGEFAULTS
0246 ----------
0247
0248 When tracing pagefaults, the format of the trace is as follows:
0249
0250 <min|maj>fault [<ip.symbol>+<ip.offset>] => <addr.dso@addr.offset> (<map type><addr level>).
0251
0252 - min/maj indicates whether fault event is minor or major;
0253 - ip.symbol shows symbol for instruction pointer (the code that generated the
0254 fault); if no debug symbols available, perf trace will print raw IP;
0255 - addr.dso shows DSO for the faulted address;
0256 - map type is either 'd' for non-executable maps or 'x' for executable maps;
0257 - addr level is either 'k' for kernel dso or '.' for user dso.
0258
0259 For symbols resolution you may need to install debugging symbols.
0260
0261 Please be aware that duration is currently always 0 and doesn't reflect actual
0262 time it took for fault to be handled!
0263
0264 When --verbose specified, perf trace tries to print all available information
0265 for both IP and fault address in the form of dso@symbol+offset.
0266
0267 EXAMPLES
0268 --------
0269
0270 Trace only major pagefaults:
0271
0272 $ perf trace --no-syscalls -F
0273
0274 Trace syscalls, major and minor pagefaults:
0275
0276 $ perf trace -F all
0277
0278 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
0279
0280 As you can see, there was major pagefault in python process, from
0281 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
0282
0283 Trace the first 4 open, openat or open_by_handle_at syscalls (in the future more syscalls may match here):
0284
0285 $ perf trace -e open* --max-events 4
0286 [root@jouet perf]# trace -e open* --max-events 4
0287 2272.992 ( 0.037 ms): gnome-shell/1370 openat(dfd: CWD, filename: /proc/self/stat) = 31
0288 2277.481 ( 0.139 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
0289 3026.398 ( 0.076 ms): gnome-shell/3039 openat(dfd: CWD, filename: /proc/self/stat) = 65
0290 4294.665 ( 0.015 ms): sed/15879 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
0291 $
0292
0293 Trace the first minor page fault when running a workload:
0294
0295 # perf trace -F min --max-stack=7 --max-events 1 sleep 1
0296 0.000 ( 0.000 ms): sleep/18006 minfault [__clear_user+0x1a] => 0x5626efa56080 (?k)
0297 __clear_user ([kernel.kallsyms])
0298 load_elf_binary ([kernel.kallsyms])
0299 search_binary_handler ([kernel.kallsyms])
0300 __do_execve_file.isra.33 ([kernel.kallsyms])
0301 __x64_sys_execve ([kernel.kallsyms])
0302 do_syscall_64 ([kernel.kallsyms])
0303 entry_SYSCALL_64 ([kernel.kallsyms])
0304 #
0305
0306 Trace the next min page page fault to take place on the first CPU:
0307
0308 # perf trace -F min --call-graph=dwarf --max-events 1 --cpu 0
0309 0.000 ( 0.000 ms): Web Content/17136 minfault [js::gc::Chunk::fetchNextDecommittedArena+0x4b] => 0x7fbe6181b000 (?.)
0310 js::gc::FreeSpan::initAsEmpty (inlined)
0311 js::gc::Arena::setAsNotAllocated (inlined)
0312 js::gc::Chunk::fetchNextDecommittedArena (/usr/lib64/firefox/libxul.so)
0313 js::gc::Chunk::allocateArena (/usr/lib64/firefox/libxul.so)
0314 js::gc::GCRuntime::allocateArena (/usr/lib64/firefox/libxul.so)
0315 js::gc::ArenaLists::allocateFromArena (/usr/lib64/firefox/libxul.so)
0316 js::gc::GCRuntime::tryNewTenuredThing<JSString, (js::AllowGC)1> (inlined)
0317 js::AllocateString<JSString, (js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
0318 js::Allocate<JSThinInlineString, (js::AllowGC)1> (inlined)
0319 JSThinInlineString::new_<(js::AllowGC)1> (inlined)
0320 AllocateInlineString<(js::AllowGC)1, unsigned char> (inlined)
0321 js::ConcatStrings<(js::AllowGC)1> (/usr/lib64/firefox/libxul.so)
0322 [0x18b26e6bc2bd] (/tmp/perf-17136.map)
0323 #
0324
0325 Trace the next two sched:sched_switch events, four block:*_plug events, the
0326 next block:*_unplug and the next three net:*dev_queue events, this last one
0327 with a backtrace of at most 16 entries, system wide:
0328
0329 # perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/
0330 0.000 :0/0 sched:sched_switch:swapper/2:0 [120] S ==> rcu_sched:10 [120]
0331 0.015 rcu_sched/10 sched:sched_switch:rcu_sched:10 [120] R ==> swapper/2:0 [120]
0332 254.198 irq/50-iwlwifi/680 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=66
0333 __dev_queue_xmit ([kernel.kallsyms])
0334 273.977 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051f600 len=78
0335 __dev_queue_xmit ([kernel.kallsyms])
0336 274.007 :0/0 net:net_dev_queue:dev=wlp3s0 skbaddr=0xffff93498051ff00 len=78
0337 __dev_queue_xmit ([kernel.kallsyms])
0338 2930.140 kworker/u16:58/2722 block:block_plug:[kworker/u16:58]
0339 2930.162 kworker/u16:58/2722 block:block_unplug:[kworker/u16:58] 1
0340 4466.094 jbd2/dm-2-8/748 block:block_plug:[jbd2/dm-2-8]
0341 8050.123 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
0342 8050.271 kworker/u16:30/2694 block:block_plug:[kworker/u16:30]
0343 #
0344
0345 SEE ALSO
0346 --------
0347 linkperf:perf-record[1], linkperf:perf-script[1]
