]> pilppa.com Git - linux-2.6-omap-h63xx.git/commit
sched: fix update_min_vruntime
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Thu, 15 Jan 2009 13:53:39 +0000 (14:53 +0100)
committerIngo Molnar <mingo@elte.hu>
Thu, 15 Jan 2009 14:12:19 +0000 (15:12 +0100)
commite17036dac189dd034c092a91df56aa740db7146d
treee4dc53a5ad8cccf71e611db3ee001a2e5a1e38f8
parent6bc912b71b6f33b041cfde93ca3f019cbaa852bc
sched: fix update_min_vruntime

Impact: fix SCHED_IDLE latency problems

OK, so we have 1 running task A (which is obviously curr and the tree is
equally obviously empty).

'A' nicely chugs along, doing its thing, carrying min_vruntime along as it
goes.

Then some whacko speed freak SCHED_IDLE task gets inserted due to SMP
balancing, which is very likely far right, in that case

update_curr
  update_min_vruntime
    cfs_rq->rb_leftmost := true (the crazy task sitting in a tree)
      vruntime = se->vruntime

and voila, min_vruntime is waaay right of where it ought to be.

OK, so why did I write it like that to begin with...

Aah, yes.

Say we've just dequeued current

schedule
  deactivate_task(prev)
    dequeue_entity
      update_min_vruntime

Then we'll set

  vruntime = cfs_rq->min_vruntime;

we find !cfs_rq->curr, but do find someone in the tree. Then we _must_
do vruntime = se->vruntime, because

 vruntime = min_vruntime(vruntime := cfs_rq->min_vruntime, se->vruntime)

will not advance vruntime, and cause lags the other way around (which we
fixed with that initial patch: 1af5f730fc1bf7c62ec9fb2d307206e18bf40a69
(sched: more accurate min_vruntime accounting).

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Mike Galbraith <efault@gmx.de>
Acked-by: Mike Galbraith <efault@gmx.de>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c