ppc32: Weird process scheduling behaviour with 2.6.24-rc

[Michel Dänzer]

On Tue, 2008-01-22 at 15:56 +0100, Michel Dänzer wrote:
> On Fri, 2008-01-18 at 13:34 +0100, Michel Dänzer wrote:
> > This is on a PowerBook5,8.
> > 
> > In a nutshell, things seem more sluggish in general than with 2.6.23.
> > But in particular, processes running at nice levels >0 can get most of
> > the CPU cycles available, slowing down processes running at nice level
> > 0.
> 
> The canonical test case I've come up with is to run an infinite loop
> with
> 
> sudo -u nobody nice -n 19 sh -c 'while true; do true; done'
> 
> This makes my X session (X server running at nice level -1, clients at
> 0) unusably sluggish (it can even take several seconds to process ctrl-c
> to interrupt the infinite loop) with 2.6.24-rc but works as expected
> with 2.6.23.
> 
> Anybody else seeing this?
> 
> 
> > I've seen this since .24-rc5 (the first .24-rc I tried), and it's still
> > there with -rc8. I'd be surprised if this kind of behaviour remained
> > unfixed for that long if it affected x86, so  I presume it's powerpc
> > specific.
> 
> Or maybe not... I've bisected this down to the scheduler changes
> between
> df3d80f5a5c74168be42788364d13cf6c83c7b9c/23fd50450a34f2558070ceabb0bfebc1c9604af5 and b5869ce7f68b233ceb81465a7644be0d9a5f3dbb .

Finished bisecting now. And the winner is...

810e95ccd58d91369191aa4ecc9e6d4a10d8d0c8 is first bad commit
commit 810e95ccd58d91369191aa4ecc9e6d4a10d8d0c8
Author: Peter Zijlstra <a.p.zijlstra at chello.nl>
Date:   Mon Oct 15 17:00:14 2007 +0200

    sched: another wakeup_granularity fix
    
    unit mis-match: wakeup_gran was used against a vruntime
    
    Signed-off-by: Peter Zijlstra <a.p.zijlstra at chello.nl>
    Signed-off-by: Ingo Molnar <mingo at elte.hu>

:040000 040000 61242d589b0082a417657807ed6329321340f7f3 bff39e49275324e15f37d2163157733580b7df1a M      kernel


Unfortunately, I don't understand how that can cause the misbehaviour
described above, and 2.6.24-rc8
(667984d9e481e43a930a478c588dced98cb61fea) with the patch below still
shows the problem. Any ideas Peter or Ingo (or anyone, really :)?


diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index da7c061..a7cc22a 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -843,7 +843,6 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p)
 	struct task_struct *curr = rq->curr;
 	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
 	struct sched_entity *se = &curr->se, *pse = &p->se;
-	unsigned long gran;
 
 	if (unlikely(rt_prio(p->prio))) {
 		update_rq_clock(rq);
@@ -866,11 +865,8 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p)
 		pse = parent_entity(pse);
 	}
 
-	gran = sysctl_sched_wakeup_granularity;
-	if (unlikely(se->load.weight != NICE_0_LOAD))
-		gran = calc_delta_fair(gran, &se->load);
 
-	if (pse->vruntime + gran < se->vruntime)
+	if (pse->vruntime + sysctl_sched_wakeup_granularity < se->vruntime)
 		resched_task(curr);
 }
 


-- 
Earthling Michel Dänzer           |          http://tungstengraphics.com
Libre software enthusiast         |          Debian, X and DRI developer