lkml.org 
[lkml]   [2012]   [Mar]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH] kick ksoftirqd more often to please soft lockup detector
On Wed, Feb 29, 2012 at 10:17:03AM +0100, Peter Zijlstra wrote:
> On Tue, 2012-02-28 at 14:16 -0800, Dan Williams wrote:
> > Looks like everyone is guilty:
> >
> > [ 422.765336] softirq took longer than 1/4 tick: 3 NET_RX ffffffff813f0aa0
> > ...
> > [ 423.971878] softirq took longer than 1/4 tick: 4 BLOCK ffffffff812519c8
> > [ 423.985093] softirq took longer than 1/4 tick: 6 TASKLET ffffffff8103422e
> > [ 423.993157] softirq took longer than 1/4 tick: 7 SCHED ffffffff8105e2e1
> > [ 424.001018] softirq took longer than 1/4 tick: 9 RCU ffffffff810a0fed
> > [ 424.008691] softirq loop took longer than 1/2 tick need_resched:
>
> /me kicks himself for not printing the actual duration.. :-)
>
> > As expected whenever that 1/2 tick message gets emitted the softirq
> > handler is almost running in a need_resched() context.
>
> Yeah.. that's quite expected.
>
> > So is it a good idea to get more aggressive about scheduling ksoftrrqd?
>
> Nah, moving away from softirq more like. I'll put moving the
> load-balancer into a kthread on the todo list. And it looks like
> everybody else should move to kthreads too.

Last year when I tried that, things got a bit ugly. I guess I don't need
the kthreads to be realtime unless CONFIG_RCU_BOOST=y, maybe that will
help. Also IIRC Steven Rostedt made some real-time changes that might
help as well for my case, which would have a per-CPU RT kthread.

Thanx, Paul



\
 
 \ /
  Last update: 2012-03-03 09:41    [W:0.099 / U:0.096 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site