xref: /linux/tools/perf/Documentation/perf-trace.txt (revision bd361f5de2b338218c276d17a510701a16075deb)
1perf-trace(1)
2=============
3
4NAME
5----
6perf-trace - strace inspired tool
7
8SYNOPSIS
9--------
10[verse]
11'perf trace'
12'perf trace record'
13
14DESCRIPTION
15-----------
16This command will show the events associated with the target, initially
17syscalls, but other system events like pagefaults, task lifetime events,
18scheduling events, etc.
19
20This is a live mode tool in addition to working with perf.data files like
21the other perf tools. Files can be generated using the 'perf record' command
22but the session needs to include the raw_syscalls events (-e 'raw_syscalls:*').
23Alternatively, 'perf trace record' can be used as a shortcut to
24automatically include the raw_syscalls events when writing events to a file.
25
26The following options apply to perf trace; options to perf trace record are
27found in the perf record man page.
28
29OPTIONS
30-------
31
32-a::
33--all-cpus::
34        System-wide collection from all CPUs.
35
36-e::
37--expr::
38	List of syscalls to show, currently only syscall names.
39	Prefixing with ! shows all syscalls but the ones specified.  You may
40	need to escape it.
41
42-D msecs::
43--delay msecs::
44After starting the program, wait msecs before measuring. This is useful to
45filter out the startup phase of the program, which is often very different.
46
47-o::
48--output=::
49	Output file name.
50
51-p::
52--pid=::
53	Record events on existing process ID (comma separated list).
54
55-t::
56--tid=::
57        Record events on existing thread ID (comma separated list).
58
59-u::
60--uid=::
61        Record events in threads owned by uid. Name or number.
62
63--filter-pids=::
64	Filter out events for these pids and for 'trace' itself (comma separated list).
65
66-v::
67--verbose=::
68        Verbosity level.
69
70--no-inherit::
71	Child tasks do not inherit counters.
72
73-m::
74--mmap-pages=::
75	Number of mmap data pages (must be a power of two) or size
76	specification with appended unit character - B/K/M/G. The
77	size is rounded up to have nearest pages power of two value.
78
79-C::
80--cpu::
81Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
82comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
83In per-thread mode with inheritance mode on (default), Events are captured only when
84the thread executes on the designated CPUs. Default is to monitor all CPUs.
85
86--duration:
87	Show only events that had a duration greater than N.M ms.
88
89--sched:
90	Accrue thread runtime and provide a summary at the end of the session.
91
92-i
93--input
94	Process events from a given perf data file.
95
96-T
97--time
98	Print full timestamp rather time relative to first sample.
99
100--comm::
101        Show process COMM right beside its ID, on by default, disable with --no-comm.
102
103-s::
104--summary::
105	Show only a summary of syscalls by thread with min, max, and average times
106    (in msec) and relative stddev.
107
108-S::
109--with-summary::
110	Show all syscalls followed by a summary by thread with min, max, and
111    average times (in msec) and relative stddev.
112
113--tool_stats::
114	Show tool stats such as number of times fd->pathname was discovered thru
115	hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
116
117-F=[all|min|maj]::
118--pf=[all|min|maj]::
119	Trace pagefaults. Optionally, you can specify whether you want minor,
120	major or all pagefaults. Default value is maj.
121
122--syscalls::
123	Trace system calls. This options is enabled by default.
124
125--call-graph [mode,type,min[,limit],order[,key][,branch]]::
126        Setup and enable call-graph (stack chain/backtrace) recording.
127        See `--call-graph` section in perf-record and perf-report
128        man pages for details. The ones that are most useful in 'perf trace'
129        are 'dwarf' and 'lbr', where available, try: 'perf trace --call-graph dwarf'.
130
131        Using this will, for the root user, bump the value of --mmap-pages to 4
132        times the maximum for non-root users, based on the kernel.perf_event_mlock_kb
133        sysctl. This is done only if the user doesn't specify a --mmap-pages value.
134
135--kernel-syscall-graph::
136	 Show the kernel callchains on the syscall exit path.
137
138--event::
139	Trace other events, see 'perf list' for a complete list.
140
141--max-stack::
142        Set the stack depth limit when parsing the callchain, anything
143        beyond the specified depth will be ignored. Note that at this point
144        this is just about the presentation part, i.e. the kernel is still
145        not limiting, the overhead of callchains needs to be set via the
146        knobs in --call-graph dwarf.
147
148        Implies '--call-graph dwarf' when --call-graph not present on the
149        command line, on systems where DWARF unwinding was built in.
150
151        Default: /proc/sys/kernel/perf_event_max_stack when present for
152                 live sessions (without --input/-i), 127 otherwise.
153
154--min-stack::
155        Set the stack depth limit when parsing the callchain, anything
156        below the specified depth will be ignored. Disabled by default.
157
158        Implies '--call-graph dwarf' when --call-graph not present on the
159        command line, on systems where DWARF unwinding was built in.
160
161--proc-map-timeout::
162	When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
163	because the file may be huge. A time out is needed in such cases.
164	This option sets the time out limit. The default value is 500 ms.
165
166PAGEFAULTS
167----------
168
169When tracing pagefaults, the format of the trace is as follows:
170
171<min|maj>fault [<ip.symbol>+<ip.offset>] => <addr.dso@addr.offset> (<map type><addr level>).
172
173- min/maj indicates whether fault event is minor or major;
174- ip.symbol shows symbol for instruction pointer (the code that generated the
175  fault); if no debug symbols available, perf trace will print raw IP;
176- addr.dso shows DSO for the faulted address;
177- map type is either 'd' for non-executable maps or 'x' for executable maps;
178- addr level is either 'k' for kernel dso or '.' for user dso.
179
180For symbols resolution you may need to install debugging symbols.
181
182Please be aware that duration is currently always 0 and doesn't reflect actual
183time it took for fault to be handled!
184
185When --verbose specified, perf trace tries to print all available information
186for both IP and fault address in the form of dso@symbol+offset.
187
188EXAMPLES
189--------
190
191Trace only major pagefaults:
192
193 $ perf trace --no-syscalls -F
194
195Trace syscalls, major and minor pagefaults:
196
197 $ perf trace -F all
198
199  1416.547 ( 0.000 ms): python/20235 majfault [CRYPTO_push_info_+0x0] => /lib/x86_64-linux-gnu/libcrypto.so.1.0.0@0x61be0 (x.)
200
201  As you can see, there was major pagefault in python process, from
202  CRYPTO_push_info_ routine which faulted somewhere in libcrypto.so.
203
204SEE ALSO
205--------
206linkperf:perf-record[1], linkperf:perf-script[1]
207