1PERF-SCHED(1) perf Manual PERF-SCHED(1)
2
3
4
6 perf-sched - Tool to trace/measure scheduler properties (latencies)
7
9 perf sched {record|latency|map|replay|script|timehist}
10
12 There are several variants of perf sched:
13
14 'perf sched record <command>' to record the scheduling events
15 of an arbitrary workload.
16
17 'perf sched latency' to report the per task scheduling latencies
18 and other scheduling properties of the workload.
19
20 'perf sched script' to see a detailed trace of the workload that
21 was recorded (aliased to 'perf script' for now).
22
23 'perf sched replay' to simulate the workload that was recorded
24 via perf sched record. (this is done by starting up mockup threads
25 that mimic the workload based on the events in the trace. These
26 threads can then replay the timings (CPU runtime and sleep patterns)
27 of the workload as it occurred when it was recorded - and can repeat
28 it a number of times, measuring its performance.)
29
30 'perf sched map' to print a textual context-switching outline of
31 workload captured via perf sched record. Columns stand for
32 individual CPUs, and the two-letter shortcuts stand for tasks that
33 are running on a CPU. A '*' denotes the CPU that had the event, and
34 a dot signals an idle CPU.
35
36 'perf sched timehist' provides an analysis of scheduling events.
37
38 Example usage:
39 perf sched record -- sleep 1
40 perf sched timehist
41
42 By default it shows the individual schedule events, including the wait
43 time (time between sched-out and next sched-in events for the task), the
44 task scheduling delay (time between wakeup and actually running) and run
45 time for the task:
46
47 time cpu task name wait time sch delay run time
48 [tid/pid] (msec) (msec) (msec)
49 -------------- ------ -------------------- --------- --------- ---------
50 79371.874569 [0011] gcc[31949] 0.014 0.000 1.148
51 79371.874591 [0010] gcc[31951] 0.000 0.000 0.024
52 79371.874603 [0010] migration/10[59] 3.350 0.004 0.011
53 79371.874604 [0011] <idle> 1.148 0.000 0.035
54 79371.874723 [0005] <idle> 0.016 0.000 1.383
55 79371.874746 [0005] gcc[31949] 0.153 0.078 0.022
56 ...
57
58 Times are in msec.usec.
59
61 -i, --input=<file>
62 Input file name. (default: perf.data unless stdin is a fifo)
63
64 -v, --verbose
65 Be more verbose. (show symbol address, etc)
66
67 -D, --dump-raw-trace=
68 Display verbose dump of the sched data.
69
70 -f, --force
71 Don’t complain, do it.
72
74 --compact
75 Show only CPUs with activity. Helps visualizing on high core count
76 systems.
77
78 --cpus
79 Show just entries with activities for the given CPUs.
80
81 --color-cpus
82 Highlight the given cpus.
83
84 --color-pids
85 Highlight the given pids.
86
88 -k, --vmlinux=<file>
89 vmlinux pathname
90
91 --kallsyms=<file>
92 kallsyms pathname
93
94 -g, --call-graph
95 Display call chains if present (default on).
96
97 --max-stack
98 Maximum number of functions to display in backtrace, default 5.
99
100 -C=, --cpu=
101 Only show events for the given CPU(s) (comma separated list).
102
103 -p=, --pid=
104 Only show events for given process ID (comma separated list).
105
106 -t=, --tid=
107 Only show events for given thread ID (comma separated list).
108
109 -s, --summary
110 Show only a summary of scheduling by thread with min, max, and
111 average run times (in sec) and relative stddev.
112
113 -S, --with-summary
114 Show all scheduling events followed by a summary by thread with
115 min, max, and average run times (in sec) and relative stddev.
116
117 --symfs=<directory>
118 Look for files with symbols relative to this directory.
119
120 -V, --cpu-visual
121 Show visual aid for sched switches by CPU: i marks idle time, s are
122 scheduler events.
123
124 -w, --wakeups
125 Show wakeup events.
126
127 -M, --migrations
128 Show migration events.
129
130 -n, --next
131 Show next task.
132
133 -I, --idle-hist
134 Show idle-related events only.
135
136 --time
137 Only analyze samples within given time window: <start>,<stop>.
138 Times have the format seconds.microseconds. If start is not given
139 (i.e., time string is ,x.y) then analysis starts at the beginning
140 of the file. If stop time is not given (i.e, time string is x.y,)
141 then analysis goes to end of file.
142
143 --state
144 Show task state when it switched out.
145
147 perf-record(1)
148
149
150
151perf 06/14/2022 PERF-SCHED(1)