Re: Very high CPU values in top on idle system (3.0-rc1)

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Mon, 2011-05-30 at 22:45 +0200, Markus Trippelsdorf wrote:
> On 2011.05.30 at 20:23 +0200, Markus Trippelsdorf wrote:
> > On 2011.05.30 at 20:05 +0200, Peter Zijlstra wrote:
> > > On Mon, 2011-05-30 at 19:39 +0200, Markus Trippelsdorf wrote:
> > > > I get very high CPU values when I run 'top' on my (mostly) idle system
> > > > with 3.0-rc1. For example mpd was always in the 1-2% range and is now
> > > > constantly over 50%. 
> > > > 
> > > > This is caused by:
> > > > 
> > > > commit 317f394160e9beb97d19a84c39b7e5eb3d7815a8
> > > > Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > > > Date:   Tue Apr 5 17:23:58 2011 +0200
> > > > 
> > > >     sched: Move the second half of ttwu() to the remote cpu
> > > > 
> > > > When I revert the above I see sane CPU values again.
> > > 
> > > So: echo NO_TTWU_QUEUE > /debug/sched_features, also cures it?
> > 
> > Yes.
> > 
> > > What architecture, what .config, and can you see it with anything other
> > > than mpd (yum search mpd, only seems to result in mpd clients not the
> > > actual server).
> > 
> > Yes, mpd was just an example. _Every_ program that would normally show in
> > the 1-5% CPU range is now in the 30-70% range (X, xterm, etc.).
> 
> IOW:
> 
> % sudo echo TTWU_QUEUE >| /sys/kernel/debug/sched_features
> % top -b -n 1
> top - 22:40:38 up  1:46, 11 users,  load average: 0.00, 0.02, 0.05
> Tasks: 109 total,   1 running, 108 sleeping,   0 stopped,   0 zombie
> Cpu(s):  1.0%us,  0.9%sy,  0.0%ni, 98.0%id,  0.1%wa,  0.0%hi,  0.0%si,  0.0%st
> Mem:   8183132k total,  1365724k used,  6817408k free,     1632k buffers
> Swap:  2097148k total,        0k used,  2097148k free,   655208k cached
> 
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>   881 mpd       20   0  122m  20m 4832 S   74  0.3   4:25.22 mpd
>  1564 root      20   0 99428  40m 4716 S   36  0.5   3:36.15 X
>  1648 markus    20   0 1147m 340m  39m S   16  4.3   1:17.90 firefox-bin
>  1649 markus    20   0  237m  33m  17m S    6  0.4   1:02.31 konsole
> 
> % sudo echo NO_TTWU_QUEUE >| /sys/kernel/debug/sched_features
> % top -b -n 1
> top - 22:42:07 up  1:48, 11 users,  load average: 0.00, 0.01, 0.05
> Tasks: 110 total,   1 running, 109 sleeping,   0 stopped,   0 zombie
> Cpu(s):  1.0%us,  0.9%sy,  0.0%ni, 98.0%id,  0.1%wa,  0.0%hi,  0.0%si,  0.0%st
> Mem:   8183132k total,  1377308k used,  6805824k free,     1632k buffers
> Swap:  2097148k total,        0k used,  2097148k free,   659896k cached
> 
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>  1649 markus    20   0  237m  33m  17m S    2  0.4   1:17.69 konsole
>     1 root      20   0   160   28   12 S    0  0.0   0:16.95 minit
>     2 root      20   0     0    0    0 S    0  0.0   0:00.36 kthreadd

Right, was easy enough to reproduce once you know what to look for
(hadn't noticed as such and simply ascribed it to desktop bloat).

It does look to be an accounting funny because when I see firefox
consume 60% the CPUs line isn't in fact registering much cpu usage at
all.

I poked around with some skip_rq_update bits but couldn't make it go
away, will try more tomorrow.