On Tue, Nov 25 2008, Alexander Beregalov wrote:
> 2008/11/25 <malahal@xxxxxxxxxx>:
> > Jens Axboe [jens.axboe@xxxxxxxxxx] wrote:
> >> On Mon, Nov 24 2008, malahal@xxxxxxxxxx wrote:
> >> > Stephen Rothwell [sfr@xxxxxxxxxxxxxxxx] wrote:
> >> > > > The block timer code calls del_timer(), should it call del_timer_sync()?
> >> > > > It is possible although unlikely that you are hitting del_timer_sync vs
> >> > > > del_timer problem in the block timeout code. Can only be seen on SMP
> >> > > > systems though!
> >> > >
> >> > > Is this still a problem in next-20081121? In that tree, the block commit
> >> > > "block: leave the request timeout timer running even on an empty list"
> >> > > was changed to add this:
> >> > >
> >> > > diff --git a/block/blk-core.c b/block/blk-core.c
> >> > > index 04267d6..44f547c 100644
> >> > > --- a/block/blk-core.c
> >> > > +++ b/block/blk-core.c
> >> > > @@ -391,6 +391,7 @@ EXPORT_SYMBOL(blk_stop_queue);
> >> > > void blk_sync_queue(struct request_queue *q)
> >> > > {
> >> > > del_timer_sync(&q->unplug_timer);
> >> > > + del_timer_sync(&q->timeout);
> >> > > kblockd_flush_work(&q->unplug_work);
> >> > > }
> >> > > EXPORT_SYMBOL(blk_sync_queue);
> >> >
> >> > I was looking at the Linux tree. Clearly same problem doesn't exist with
> >> > the above commit! I wonder why kblockd_flush_work() is called after the
> >> > del_timer_sync(). It makes sense to cancel the work and then shutdown
> >> > the timer(s). I doubt if you are running into this problem though.
> >>
> >> If the kernel tested doesn't include the above fix, it'll surely go
> >> boom. Can someone verify that this is the case?
> >
> > Just looked, next-20081119 doesn't have the above fix. It is included in
> > next-20081120. Also note that the above fix is only partially copied,
> > there is other part that removed deleting the timer when there are no
> > outstanding requests.
> >
> Yes, I can not reproduce it anymore on linux-next 1121 and newer. (I
> did not try 1120) It seems the fix works pretty good. Is it still
> needed and reasonable to investigate the problem on next-20081119?
> Unfortunately I do not have much time for it.
No, you don't have to investigate further. This was a known bug that is
fixed in -next and mainline basically right after next-20081119.
>
> All these problems have gone away on next-1125 except ODEBUG warning
> on HPET.
--
Jens Axboe
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
