Re: [PATCH] blk-mq: don't grab rq's refcount in blk_mq_check_expired()

[Keith Busch <kbusch@kernel.org>]

On Wed, Aug 11, 2021 at 10:38:38PM +0800, Ming Lei wrote:
> Inside blk_mq_queue_tag_busy_iter() we already grabbed request's
> refcount before calling ->fn(), so needn't to grab it one more time
> in blk_mq_check_expired().
> 
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>  block/blk-mq.c | 25 +++++++------------------
>  1 file changed, 7 insertions(+), 18 deletions(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index d2725f94491d..4d3457d2957f 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -917,6 +917,10 @@ void blk_mq_put_rq_ref(struct request *rq)
>  		__blk_mq_free_request(rq);
>  }
>  
> +/*
> + * This request won't be re-allocated because its refcount is held when
> + * calling this callback.
> + */
>  static bool blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
>  		struct request *rq, void *priv, bool reserved)
>  {
> @@ -930,27 +934,12 @@ static bool blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
>  		return true;
>  
>  	/*
> -	 * We have reason to believe the request may be expired. Take a
> -	 * reference on the request to lock this request lifetime into its
> -	 * currently allocated context to prevent it from being reallocated in
> -	 * the event the completion by-passes this timeout handler.
> -	 *
> -	 * If the reference was already released, then the driver beat the
> -	 * timeout handler to posting a natural completion.
> -	 */
> -	if (!refcount_inc_not_zero(&rq->ref))
> -		return true;
> -
> -	/*
> -	 * The request is now locked and cannot be reallocated underneath the
> -	 * timeout handler's processing. Re-verify this exact request is truly
> -	 * expired; if it is not expired, then the request was completed and
> -	 * reallocated as a new request.
> +	 * Re-verify this exact request is truly expired; if it is not expired,
> +	 * then the request was completed and reallocated as a new request
> +	 * after returning from blk_mq_check_expired().
>  	 */
>  	if (blk_mq_req_expired(rq, next))
>  		blk_mq_rq_timed_out(rq, reserved);

There's no need to check expired twice if the iterator is already taking a
reference. I had this double check because I didn't want to penalize normal IO
by preventing it from completing while the timeout handler is running, but it
looks like the current timeout iterator is going to do that anyway.

> -
> -	blk_mq_put_rq_ref(rq);
>  	return true;
>  }