Re: [BUG] TASK_DEAD task is able to be woken up in specialcondition

[Mike Galbraith]

On Tue, 2012-01-24 at 13:01 -0500, KOSAKI Motohiro wrote:
> On 1/24/2012 5:23 AM, Peter Zijlstra wrote:
> > On Wed, 2011-12-28 at 16:07 -0500, KOSAKI Motohiro wrote:
> >> I looked at scheduler code today briefly. now I'm afraid following code 
> >> have similar race.
> >>
> >>
> >>          if (task_contributes_to_load(p))
> >>                  rq->nr_uninterruptible--;
> >>
> >>
> >>
> >> Can't following schenario be happen?
> >>
> >>
> >> CPU0                    CPU1
> >> --------------------------------------------------------
> >> deactivate_task()
> >>                         task->state = TASK_UNINTERRUPTIBLE;
> >> activate_task()
> >>    rq->nr_uninterruptible--;
> >>
> >>                         schedule()
> >>                           deactivate_task()
> >>                             rq->nr_uninterruptible++;
> >>
> >> Totally, nr_uninterruptible wasn't incremented.
> >>
> >>
> >> I'm still not sure. I need to read more sched code. 
> > 
> > You shouldn't ever set another tasks ->state. 
> 
> I'm sorry. I haven't catch your point. I think following step is
> valid kernel code. Do you disagree?
> 
> >>                         task->state = TASK_UNINTERRUPTIBLE;
> >>                         schedule()

I think you meant:
	__set_current_state(TASK_UNINTERRUPTIBLE);
	schedule();

The way you wrote it, task doesn't have to be current, so could be doing
the bad thing Peter pointed out, diddling *another* tasks ->state.

	-Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/