Prev: n2-drv: remove casts from void*
Next: [PATCH 1/3] splice: direct_splice_actor() should not use pos in sd
From: Peter Zijlstra on 29 Jun 2010 06:50
On Tue, 2010-06-29 at 15:10 +0800, shenghui wrote:
> I think some lock on the metadata can fix this issue, but we may
> change plenty of code to add support for lock. I think the easist
> way is just substacting nr_running before dequing tasks.

But all that is fully serialized by the rq->lock.. so I'm really not
seeing how this can happen.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Peter Zijlstra on 29 Jun 2010 07:40
On Tue, 2010-06-29 at 19:24 +0800, shenghui wrote:

> I wonder is there any chance set_next_entity() can get NULL for
> parameter se if so?

Well, if your machine crashes that way, maybe, but I haven't seen that
happen in a long while.

> And will you please give me some instructions on where rq->lock
> is required?

Pretty much everywhere, if you look at sched.c the only sched_class
method not called with rq->lock held is ::task_fork().

The interesting bits are that schedule()->pre_schedule()/idle_balance()
can drop rq->lock as well as ->select_task_rq().



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: n2-drv: remove casts from void*
Next: [PATCH 1/3] splice: direct_splice_actor() should not use pos in sd