Re: [PATCH 0/4] workqueue_trace: Add max execution time mesurement for per worklet

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 28 Apr 2009 18:51:33 +0800 Zhaolei <zhaolei@cn.fujitsu.com> wrote:

> Hi,
> 
> This patch add max execution time mesurement for per worklet.
> It is useful for driver-developer and system administrator to
> know which worklet runs how many time.

This seems more useful than the previous feature.

> It need applied on top of patchset of:
> workqueue_tracepoint: Add worklet tracepoints for worklet lifecycle tracing
> 
> Output is like:
>  # CPU INSERTED EXECUTED M_EXECus TASKNAME:PID
>  #  |     |        |        |      `-WORKFUNC
>  #  |     |        |        |        |
>      0      905      973          events/0:5
>      0        1        1       37  |-reg_todo+0x0/0x41f
>      0       13       12     2033  |-rt_worker_func+0x0/0x219
>      0       49       48     3157  |-mce_work_fn+0x0/0x31
>      0       14       13      939  |-check_corruption+0x0/0x2a
>      0      707      706     3826  |-vmstat_update+0x0/0x33
>      0        1        1     1846  |-power_supply_changed_work+0x0/0x3c
>      0       11       11     2879  |-console_callback+0x0/0xe1
>      0       30       29     1579  |-do_cache_clean+0x0/0x37
>      0       74       74     2841  |-test_work+0x0/0x53
>      0        3        2      252  |-rekey_seq_generator+0x0/0x5a
>      0        1        1       21  |-grace_ender+0x0/0x14
>      0        1        1     3701  `-flush_to_ldisc+0x0/0x15d
>      0        0        0          cpuset:6
>      0      658      658          khelper:7
>      0      658      658     5735  `-__call_usermodehelper+0x0/0x77
>      0        0        0          netns:11

What is the meaning of M_EXECus?  Mean execution time in microseconds,
I assume?  I wonder if that is sufficiently high resolution nowadays.

You'll note that all the functions are reported as "foo+0x0/0xN". 
There is a way of suppressing the unneeded and unuseful "+0x0/0xN",
but I forget what it is and the comments over lib/vsprintf.c:pointer()
didn't help.  I think it's %pS.

The patchset adds no user documentation for the feature and should not
(IMO) be merged in such a state.  Please.