RFC:blk-exec-call-completion-if-queue-dead [was Q: blk_execute_rq_nowait() doesn't call completion in case of dead queue

RFC:blk-exec-call-completion-if-queue-dead [was Q: blk_execute_rq_nowait() doesn't call completion in case of dead queue

[Muthu Kumar]

How about this change?

diff --git a/block/blk-exec.c b/block/blk-exec.c
index fb2cbd5..6bf5c0b 100644
--- a/block/blk-exec.c
+++ b/block/blk-exec.c
@@ -56,8 +56,10 @@ void blk_execute_rq_nowait(struct request_queue *q, struct ge
        if (unlikely(blk_queue_dead(q))) {
                spin_unlock_irq(q->queue_lock);
                rq->errors = -ENXIO;
-               if (rq->end_io)
-                       rq->end_io(rq, rq->errors);
+                if (done)
+                    done(rq, rq->errors);
+                else if (rq->end_io)    //XXX Not sure if this check and end_io
+                    rq->end_io(rq, rq->errors);
                return;
        }

Only one driver - sx8.c, doesn't set done() function and every one
else expects done() to be called with error.

Regards,
Muthu


On Wed, Jun 6, 2012 at 11:59 AM, Muthu Kumar <muthu.lkml@gmail.com> wrote:
> Jens, Tejun et al,
> blk_execute_rq() calls blk_execute_rq_nowait() with blk_end_sync_rq()
> as the completion (done) routine. And then waits for the completion().
>
>
>        rq->end_io_data = &wait;
>        blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
>        ...
>        ...
>        wait_for_completion(&wait)
>
> But in blk_execute_rq_nowait() if the queue is dead, completion
> routine is not called. Shouldn't we have to call the above completion
> (done()) to wake up the wait_for_completion()?
>
>
>        if (unlikely(blk_queue_dead(q))) {
>                spin_unlock_irq(q->queue_lock);
>                rq->errors = -ENXIO;
>                if (rq->end_io)
>                        rq->end_io(rq, rq->errors);
>                return;
>       }
>
> Thanks.
>
> Regards,
> Muthu

Re: RFC:blk-exec-call-completion-if-queue-dead [was Q: blk_execute_rq_nowait() doesn't call completion in case of dead queue

[Tejun Heo]

On Wed, Jun 06, 2012 at 02:24:35PM -0700, Muthu Kumar wrote:
> How about this change?
> 
> diff --git a/block/blk-exec.c b/block/blk-exec.c
> index fb2cbd5..6bf5c0b 100644
> --- a/block/blk-exec.c
> +++ b/block/blk-exec.c
> @@ -56,8 +56,10 @@ void blk_execute_rq_nowait(struct request_queue *q, struct ge
>         if (unlikely(blk_queue_dead(q))) {
>                 spin_unlock_irq(q->queue_lock);
>                 rq->errors = -ENXIO;
> -               if (rq->end_io)
> -                       rq->end_io(rq, rq->errors);
> +                if (done)
> +                    done(rq, rq->errors);
> +                else if (rq->end_io)    //XXX Not sure if this check and end_io
> +                    rq->end_io(rq, rq->errors);
>                 return;
>         }
> 
> Only one driver - sx8.c, doesn't set done() function and every one
> else expects done() to be called with error.

Looks like the bug there is rq->rq_disk and rq->end_io assignments
happening after the queue_dead check.  Just move the two lines before
queue_head check?

Thanks.

-- 
tejun