perf/x86: Fix 32-bit perf user callgraph collection
[linux-2.6-block.git] / Documentation / cgroup-v1 / cpuacct.txt
CommitLineData
934352f2
BR
1CPU Accounting Controller
2-------------------------
3
4The CPU accounting controller is used to group tasks using cgroups and
5account the CPU usage of these groups of tasks.
6
7The CPU accounting controller supports multi-hierarchy groups. An accounting
8group accumulates the CPU usage of all of its child groups and the tasks
9directly present in its group.
10
11Accounting groups can be created by first mounting the cgroup filesystem.
12
f6e07d38
JS
13# mount -t cgroup -ocpuacct none /sys/fs/cgroup
14
15With the above step, the initial or the parent accounting group becomes
16visible at /sys/fs/cgroup. At bootup, this group includes all the tasks in
17the system. /sys/fs/cgroup/tasks lists the tasks in this cgroup.
18/sys/fs/cgroup/cpuacct.usage gives the CPU time (in nanoseconds) obtained
19by this group which is essentially the CPU time obtained by all the tasks
934352f2
BR
20in the system.
21
f6e07d38 22New accounting groups can be created under the parent group /sys/fs/cgroup.
934352f2 23
f6e07d38 24# cd /sys/fs/cgroup
934352f2 25# mkdir g1
e47f9d84 26# echo $$ > g1/tasks
934352f2
BR
27
28The above steps create a new group g1 and move the current shell
29process (bash) into it. CPU time consumed by this bash and its children
30can be obtained from g1/cpuacct.usage and the same is accumulated in
f6e07d38 31/sys/fs/cgroup/cpuacct.usage also.
ef12fefa
BR
32
33cpuacct.stat file lists a few statistics which further divide the
34CPU time obtained by the cgroup into user and system times. Currently
35the following statistics are supported:
36
37user: Time spent by tasks of the cgroup in user mode.
38system: Time spent by tasks of the cgroup in kernel mode.
39
40user and system are in USER_HZ unit.
41
42cpuacct controller uses percpu_counter interface to collect user and
43system times. This has two side effects:
44
45- It is theoretically possible to see wrong values for user and system times.
46 This is because percpu_counter_read() on 32bit systems isn't safe
47 against concurrent writes.
48- It is possible to see slightly outdated values for user and system times
49 due to the batch processing nature of percpu_counter.