5 Tasks encounter delays in execution when they wait
6 for some kernel resource to become available e.g. a
7 runnable task may wait for a free CPU to run on.
9 The per-task delay accounting functionality measures
10 the delays experienced by a task while
12 a) waiting for a CPU (while being runnable)
13 b) completion of synchronous block I/O initiated by the task
21 and makes these statistics available to userspace through
22 the taskstats interface.
24 Such delays provide feedback for setting a task's cpu priority,
25 io priority and rss limit values appropriately. Long delays for
26 important tasks could be a trigger for raising its corresponding priority.
28 The functionality, through its use of the taskstats interface, also provides
29 delay statistics aggregated for all tasks (or threads) belonging to a
30 thread group (corresponding to a traditional Unix process). This is a commonly
31 needed aggregation that is more efficiently done by the kernel.
33 Userspace utilities, particularly resource management applications, can also
34 aggregate delay statistics into arbitrary groups. To enable this, delay
35 statistics of a task are available both during its lifetime as well as on its
36 exit, ensuring continuous and complete monitoring can be done.
42 Delay accounting uses the taskstats interface which is described
43 in detail in a separate document in this directory. Taskstats returns a
44 generic data structure to userspace corresponding to per-pid and per-tgid
45 statistics. The delay accounting functionality populates specific fields of
48 include/uapi/linux/taskstats.h
50 for a description of the fields pertaining to delay accounting.
51 It will generally be in the form of counters returning the cumulative
52 delay seen for cpu, sync block I/O, swapin, memory reclaim, thrash page
53 cache, direct compact, write-protect copy, IRQ/SOFTIRQ etc.
55 Taking the difference of two successive readings of a given
56 counter (say cpu_delay_total) for a task will give the delay
57 experienced by the task waiting for the corresponding resource
60 When a task exits, records containing the per-task statistics
61 are sent to userspace without requiring a command. If it is the last exiting
62 task of a thread group, the per-tgid statistics are also sent. More details
63 are given in the taskstats interface description.
65 The getdelays.c userspace utility in tools/accounting directory allows simple
66 commands to be run and the corresponding delay statistics to be displayed. It
67 also serves as an example of using the taskstats interface.
72 Compile the kernel with::
74 CONFIG_TASK_DELAY_ACCT=y
77 Delay accounting is disabled by default at boot up.
82 to the kernel boot options. The rest of the instructions below assume this has
83 been done. Alternatively, use sysctl kernel.task_delayacct to switch the state
84 at runtime. Note however that only tasks started after enabling it will have
85 delayacct information.
87 After the system has booted up, use a utility
88 similar to getdelays.c to access the delays
89 seen by a given task or a task group (tgid).
90 The utility also allows a given command to be
91 executed and the corresponding delays to be
94 General format of the getdelays command::
96 getdelays [-dilv] [-t tgid] [-p pid]
98 Get delays, since system boot, for pid 10::
100 # ./getdelays -d -p 10
101 (output similar to next case)
103 Get sum of delays, since system boot, for all pids with tgid 5::
105 # ./getdelays -d -t 5
106 print delayacct stats ON
110 CPU count real total virtual total delay total delay average
111 8 7000000 6872122 3382277 0.423ms
112 IO count delay total delay average
114 SWAP count delay total delay average
116 RECLAIM count delay total delay average
118 THRASHING count delay total delay average
120 COMPACT count delay total delay average
122 WPCOPY count delay total delay average
124 IRQ count delay total delay average
127 Get IO accounting for pid 1, it works only with -p::
129 # ./getdelays -i -p 1
130 printing IO accounting
131 linuxrc: read=65536, write=0, cancelled_write=0
133 The above command can be used with -v to get more debug information.