Re: [bug] ata subsystem related crash with latest -git

[Jens Axboe <jens.axboe@oracle.com>]

On Thu, Oct 18 2007, Olof Johansson wrote:
> On Thu, Oct 18, 2007 at 04:38:36PM +0200, Jens Axboe wrote:
> > On Thu, Oct 18 2007, Benny Halevy wrote:
> > > On Oct. 18, 2007, 16:05 +0200, Jens Axboe <jens.axboe@oracle.com> wrote:
> > > > On Thu, Oct 18 2007, Jens Axboe wrote:
> > > >> On Thu, Oct 18 2007, Benny Halevy wrote:
> > > >>>>  	return sg;
> > > >>>>  }
> > > >>>> @@ -83,6 +96,9 @@ static inline struct scatterlist *sg_last(struct scatterlist *sgl,
> > > >>>>  		ret = sg;
> > > >>>>  
> > > >>>>  #endif
> > > >>>> +#ifdef CONFIG_DEBUG_SG
> > > >>>> +	BUG_ON(sgl[0].sg_magic != SG_MAGIC);
> > > >>> can it also do BUG_ON(!sg_is_last(sg))?
> > > >> That would make sense, definitely. I'll add that.
> > > > 
> > > >         BUG_ON(!sg_is_last(ret));
> > > > 
> > > > it should be, not sg. That's what I merged.
> > > > 
> > > right. of course.
> > 
> > OK, that found something interesting - mapping a request may shrink it,
> > so we need to update the end marker to move it earlier in the list.
> > Basically just a
> > 
> >         if (sg)
> >                 __sg_mark_end(sg);
> > 
> > at the bottom of blk_rq_map_sg(). Updated patch below, booted on other
> > machines now as well without incident.
> 
> PPC needs this. I'm guessing other arches might too. Otherwise seems to boot
> and work well (with Jeff's sata patch).

Oh duh, indeed I didn't add that to the archs when converting. Thanks
for the patch, I'll update the rest as well.

And thanks a lot for boot testing it!

-- 
Jens Axboe