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.
38 List of syscalls to show, currently only syscall names.
39 Prefixing with ! shows all syscalls but the ones specified. You may
48 Record events on existing process ID (comma separated list).
52 Record events on existing thread ID (comma separated list).
56 Record events in threads owned by uid. Name or number.
59 Filter out events for these pids and for 'trace' itself (comma separated list).
66 Child tasks do not inherit counters.
70 Number of mmap data pages (must be a power of two) or size
71 specification with appended unit character - B/K/M/G. The
72 size is rounded up to have nearest pages power of two value.
76 Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
77 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
78 In per-thread mode with inheritance mode on (default), Events are captured only when
79 the thread executes on the designated CPUs. Default is to monitor all CPUs.
82 Show only events that had a duration greater than N.M ms.
85 Accrue thread runtime and provide a summary at the end of the session.
89 Process events from a given perf data file.
93 Print full timestamp rather time relative to first sample.
96 Show process COMM right beside its ID, on by default, disable with --no-comm.
100 Show only a summary of syscalls by thread with min, max, and average times
101 (in msec) and relative stddev.
105 Show all syscalls followed by a summary by thread with min, max, and
106 average times (in msec) and relative stddev.
109 Show tool stats such as number of times fd->pathname was discovered thru
110 hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
114 Trace pagefaults. Optionally, you can specify whether you want minor,
115 major or all pagefaults. Default value is maj.
118 Trace system calls. This options is enabled by default.
120 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
121 Setup and enable call-graph (stack chain/backtrace) recording.
122 See `--call-graph` section in perf-record and perf-report
123 man pages for details. The ones that are most useful in 'perf trace'
124 are 'dwarf' and 'lbr', where available, try: 'perf trace --call-graph dwarf'.
126 Using this will, for the root user, bump the value of --mmap-pages to 4
127 times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
128 sysctl. This is done only if the user doesn't specify a --mmap-pages value.
130 --kernel-syscall-graph::
131 Show the kernel callchains on the syscall exit path.
134 Trace other events, see 'perf list' for a complete list.
137 Set the stack depth limit when parsing the callchain, anything
138 beyond the specified depth will be ignored. Note that at this point
139 this is just about the presentation part, i.e. the kernel is still
140 not limiting, the overhead of callchains needs to be set via the
141 knobs in --call-graph dwarf.
143 Implies '--call-graph dwarf' when --call-graph not present on the
144 command line, on systems where DWARF unwinding was built in.
146 Default: /proc/sys/kernel/perf_event_max_stack when present for
147 live sessions (without --input/-i), 127 otherwise.
150 Set the stack depth limit when parsing the callchain, anything
151 below the specified depth will be ignored. Disabled by default.
153 Implies '--call-graph dwarf' when --call-graph not present on the
154 command line, on systems where DWARF unwinding was built in.
157 When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
158 because the file may be huge. A time out is needed in such cases.
159 This option sets the time out limit. The default value is 500 ms.
164 When tracing pagefaults, the format of the trace is as follows:
166 <min|maj>fault [<ip.symbol>+<ip.offset>] => <addr.dso@addr.offset> (<map type><addr level>).
168 - min/maj indicates whether fault event is minor or major;
169 - ip.symbol shows symbol for instruction pointer (the code that generated the
170 fault); if no debug symbols available, perf trace will print raw IP;
171 - addr.dso shows DSO for the faulted address;
172 - map type is either 'd' for non-executable maps or 'x' for executable maps;
173 - addr level is either 'k' for kernel dso or '.' for user dso.
175 For symbols resolution you may need to install debugging symbols.
177 Please be aware that duration is currently always 0 and doesn't reflect actual
178 time it took for fault to be handled!
180 When --verbose specified, perf trace tries to print all available information
181 for both IP and fault address in the form of dso@symbol+offset.
186 Trace only major pagefaults:
188 $ perf trace --no-syscalls -F
190 Trace syscalls, major and minor pagefaults:
194 1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
196 As you can see, there was major pagefault in python process, from
197 CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
201 linkperf:perf-record[1], linkperf:perf-script[1]