215 lines · plain
1perf-kwork(1)2=============3 4NAME5----6perf-kwork - Tool to trace/measure kernel work properties (latencies)7 8SYNOPSIS9--------10[verse]11'perf kwork' {record|report|latency|timehist|top}12 13DESCRIPTION14-----------15There are several variants of 'perf kwork':16 17 'perf kwork record <command>' to record the kernel work18 of an arbitrary workload.19 20 'perf kwork report' to report the per kwork runtime.21 22 'perf kwork latency' to report the per kwork latencies.23 24 'perf kwork timehist' provides an analysis of kernel work events.25 26 'perf kwork top' to report the task cpu usage.27 28 Example usage:29 perf kwork record -- sleep 130 perf kwork report31 perf kwork report -b32 perf kwork latency33 perf kwork latency -b34 perf kwork timehist35 perf kwork top36 perf kwork top -b37 38 By default it shows the individual work events such as irq, workqueue,39 including the run time and delay (time between raise and actually entry):40 41 Runtime start Runtime end Cpu Kwork name Runtime Delaytime42 (TYPE)NAME:NUM (msec) (msec)43 ----------------- ----------------- ------ ------------------------- ---------- ----------44 1811186.976062 1811186.976327 [0000] (s)RCU:9 0.266 0.11445 1811186.978452 1811186.978547 [0000] (s)SCHED:7 0.095 0.17146 1811186.980327 1811186.980490 [0000] (s)SCHED:7 0.162 0.08347 1811186.981221 1811186.981271 [0000] (s)SCHED:7 0.050 0.07748 1811186.984267 1811186.984318 [0000] (s)SCHED:7 0.051 0.07549 1811186.987252 1811186.987315 [0000] (s)SCHED:7 0.063 0.08150 1811186.987785 1811186.987843 [0006] (s)RCU:9 0.058 0.64551 1811186.988319 1811186.988383 [0000] (s)SCHED:7 0.064 0.14352 1811186.989404 1811186.989607 [0002] (s)TIMER:1 0.203 0.11153 1811186.989660 1811186.989732 [0002] (s)SCHED:7 0.072 0.31054 1811186.991295 1811186.991407 [0002] eth0:10 0.11255 1811186.991639 1811186.991734 [0002] (s)NET_RX:3 0.095 0.27756 1811186.989860 1811186.991826 [0002] (w)vmstat_shepherd 1.966 0.34557 ...58 59 Times are in msec.usec.60 61OPTIONS62-------63-D::64--dump-raw-trace=::65 Display verbose dump of the sched data.66 67-f::68--force::69 Don't complain, do it.70 71-k::72--kwork::73 List of kwork to profile (irq, softirq, workqueue, sched, etc)74 75-v::76--verbose::77 Be more verbose. (show symbol address, etc)78 79OPTIONS for 'perf kwork report'80----------------------------81 82-b::83--use-bpf::84 Use BPF to measure kwork runtime85 86-C::87--cpu::88 Only show events for the given CPU(s) (comma separated list).89 90-i::91--input::92 Input file name. (default: perf.data unless stdin is a fifo)93 94-n::95--name::96 Only show events for the given name.97 98-s::99--sort::100 Sort by key(s): runtime, max, count101 102-S::103--with-summary::104 Show summary with statistics105 106--time::107 Only analyze samples within given time window: <start>,<stop>. Times108 have the format seconds.microseconds. If start is not given (i.e., time109 string is ',x.y') then analysis starts at the beginning of the file. If110 stop time is not given (i.e, time string is 'x.y,') then analysis goes111 to end of file.112 113OPTIONS for 'perf kwork latency'114----------------------------115 116-b::117--use-bpf::118 Use BPF to measure kwork latency119 120-C::121--cpu::122 Only show events for the given CPU(s) (comma separated list).123 124-i::125--input::126 Input file name. (default: perf.data unless stdin is a fifo)127 128-n::129--name::130 Only show events for the given name.131 132-s::133--sort::134 Sort by key(s): avg, max, count135 136--time::137 Only analyze samples within given time window: <start>,<stop>. Times138 have the format seconds.microseconds. If start is not given (i.e., time139 string is ',x.y') then analysis starts at the beginning of the file. If140 stop time is not given (i.e, time string is 'x.y,') then analysis goes141 to end of file.142 143OPTIONS for 'perf kwork timehist'144---------------------------------145 146-C::147--cpu::148 Only show events for the given CPU(s) (comma separated list).149 150-g::151--call-graph::152 Display call chains if present (default off).153 154-i::155--input::156 Input file name. (default: perf.data unless stdin is a fifo)157 158-k::159--vmlinux=<file>::160 Vmlinux pathname161 162-n::163--name::164 Only show events for the given name.165 166--kallsyms=<file>::167 Kallsyms pathname168 169--max-stack::170 Maximum number of functions to display in backtrace, default 5.171 172--symfs=<directory>::173 Look for files with symbols relative to this directory.174 175--time::176 Only analyze samples within given time window: <start>,<stop>. Times177 have the format seconds.microseconds. If start is not given (i.e., time178 string is ',x.y') then analysis starts at the beginning of the file. If179 stop time is not given (i.e, time string is 'x.y,') then analysis goes180 to end of file.181 182OPTIONS for 'perf kwork top'183---------------------------------184 185-b::186--use-bpf::187 Use BPF to measure task cpu usage.188 189-C::190--cpu::191 Only show events for the given CPU(s) (comma separated list).192 193-i::194--input::195 Input file name. (default: perf.data unless stdin is a fifo)196 197-n::198--name::199 Only show events for the given name.200 201-s::202--sort::203 Sort by key(s): rate, runtime, tid204 205--time::206 Only analyze samples within given time window: <start>,<stop>. Times207 have the format seconds.microseconds. If start is not given (i.e., time208 string is ',x.y') then analysis starts at the beginning of the file. If209 stop time is not given (i.e, time string is 'x.y,') then analysis goes210 to end of file.211 212SEE ALSO213--------214linkperf:perf-record[1]215