Re: 2.6.27.8 scheduler bug - threads not being scheduled for long periods

[Dimitri Sivanich <sivanich@sgi.com>]

Hi Greg,

On Mon, Jan 05, 2009 at 01:44:41PM -0500, Gregory Haskins wrote:
> Dimitri Sivanich wrote:
> >
> > I'm wondering if anyone else has run into this problem, and whether they've
> > found a resolution for it?
> >
> >   
> 
> Hi Dimitri,
>   I have observed similar problems on my opensuse 11.0 box (2.6.25
> based) when doing kernel builds.  Often times other apps will get
> starved (like firefox, etc).  I was talking to peterz about it and he
> pointed me at a commit that went into 28-rcx that fixes this (the sha
> escapes me at the moment).  I've been meaning to pull the patch into our
> 2.6.25 and 2.6.27 trees to verify if it fixes it, but I have fairly high
> confidence it will.  Perhaps Peter can point you at the same patch and
> you can give it a whirl.  For the time being, can you see if 28 is fixed?
> 

2.6.28 appears to be a little bit worse, actually:
Counts:
0: 50165855
1: 8879
2: 5462
3: 4990
4: 91805284

^C
# uname -r
2.6.28-03114-g3c92ec8

The master thread hangs up after printing only one set of counts.