Re: [patch 8/9] lguest: the block driver

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

On Thu, May 10 2007, Rene Herman wrote:
> On 05/09/2007 12:22 PM, Pekka Enberg wrote:
> 
> >>+static void end_entire_request(struct request *req, int uptodate)
> >>+{
> >>+       if (end_that_request_first(req, uptodate, req->hard_nr_sectors))
> >>+               BUG();
> >>+       add_disk_randomness(req->rq_disk);
> >>+       blkdev_dequeue_request(req);
> >>+       end_that_request_last(req, uptodate);
> >>+}
> >
> >Again, I would prefer this went straight into block/ll_rw_blk.c. Rene
> >and I am using something similar in the new Mitsumi driver although
> >one of us has a bug already, we're using req->nr_sectors for this...
> 
> The req->hard_nr_sectors is specifically marked block layer internal and 
> since nr_sectors is what we're requesting and reading (we can't deal with 
> partial transfers anyway since the drive won't tell us where it failed as 
> far as I've been able to see and we do want to make read requests for as 
> many sectors as possible at a time for speed) I believe that in our case 
> the nr_sectors is fine.

Using ->nr_sectors is fine, IFF you don't modify ->nr_sectors in the
driver. The generic helper will use ->hard_nr_sectors.

-- 
Jens Axboe