On Mon, Nov 06 2006, Mike Miller (OS Dev) wrote:
> Patch 12 of 12
>
> This patch replaces complete_buffers with end_that_request_first to fix
> programs like iostat. This has been broken for the last few kernel releases.
> Please consider this for inclusion.
>
> Thanks,
> mikem
>
> Signed-off-by: Mike Miller <mike.miller@xxxxxx>
>
>
> ---
>
>
> ---
>
> drivers/block/cciss.c | 20 ++++----------------
> 1 files changed, 4 insertions(+), 16 deletions(-)
>
> diff -puN drivers/block/cciss.c~cciss_update_diskstats_fix drivers/block/cciss.c
> --- linux-2.6/drivers/block/cciss.c~cciss_update_diskstats_fix 2006-11-06 13:28:53.000000000 -0600
> +++ linux-2.6-root/drivers/block/cciss.c 2006-11-06 13:28:53.000000000 -0600
> @@ -1156,18 +1156,6 @@ static int cciss_ioctl(struct inode *ino
> }
> }
>
> -static inline void complete_buffers(struct bio *bio, int status)
> -{
> - while (bio) {
> - struct bio *xbh = bio->bi_next;
> - int nr_sectors = bio_sectors(bio);
> -
> - bio->bi_next = NULL;
> - bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
> - bio = xbh;
> - }
> -}
> -
> static void cciss_check_queues(ctlr_info_t *h)
> {
> int start_queue = h->next_to_run;
> @@ -1236,15 +1224,15 @@ static void cciss_softirq_done(struct re
> pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
> }
>
> - complete_buffers(rq->bio, rq->errors);
> -
> #ifdef CCISS_DEBUG
> printk("Done with %p\n", rq);
> #endif /* CCISS_DEBUG */
>
> - add_disk_randomness(rq->rq_disk);
> spin_lock_irqsave(&h->lock, flags);
> - end_that_request_last(rq, rq->errors);
> + if (!end_that_request_first(rq, rq->errors, rq->nr_sectors)) {
> + add_disk_randomness(rq->rq_disk);
> + end_that_request_last(rq, rq->errors);
> + }
> cmd_free(h, cmd, 1);
> cciss_check_queues(h);
> spin_unlock_irqrestore(&h->lock, flags);
Ah, so there's where that went. Your code isn't clear, though -
end_that_request_first() _must_ return 0, so the above looks confusing.
It would look cleaner and more informative like:
if (end_that_request_first(rq, rq->errors, rq->nr_sectors))
BUG();
add_disk_randomness(rq->rq_disk);
end_that_request_last(rq, rq->errors);
...
and so on. Additionally you don't need the lock for
end_that_request_first(), so it's a lot more optimal to rearrange it
again.
add_disk_randomness(rq->rq_disk);
if (end_that_request_first(rq, rq->errors, rq->nr_sectors))
BUG();
spin_lock_irqsave(&h->lock, flags);
end_that_request_last(rq, rq->errors);
cmd_free(h, cmd, 1);
...
Not only cleaner to read since it's obvious what will happen, also moves
the heavy path (end_that_request_first()) outside of the controller
lock.
--
Jens Axboe
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
