Re: [PATCH v2 03/12] block: Send requeued requests to the I/O scheduler

[Christoph Hellwig <hch@lst.de>]

On Fri, Apr 07, 2023 at 04:58:13PM -0700, Bart Van Assche wrote:
> -		/*
> -		 * If RQF_DONTPREP, rq has contained some driver specific
> -		 * data, so insert it to hctx dispatch list to avoid any
> -		 * merge.
> -		 */
> -		if (rq->rq_flags & RQF_DONTPREP)
> -			blk_mq_request_bypass_insert(rq, false, false);
> -		else
> -			blk_mq_sched_insert_request(rq, true, false, false);
> +		blk_mq_sched_insert_request(rq, /*at_head=*/true, false, false);

This effectively undoes commit aef1897cd36d, and instead you add
RQF_DONTPREP to RQF_NOMERGE_FLAGS.  This makes sense to be, but should
probably be more clearly documented in the commit log.

> @@ -2065,9 +2057,14 @@ bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *list,
>  		if (nr_budgets)
>  			blk_mq_release_budgets(q, list);
>  
> -		spin_lock(&hctx->lock);
> -		list_splice_tail_init(list, &hctx->dispatch);
> -		spin_unlock(&hctx->lock);
> +		if (!q->elevator) {
> +			spin_lock(&hctx->lock);
> +			list_splice_tail_init(list, &hctx->dispatch);
> +			spin_unlock(&hctx->lock);
> +		} else {
> +			q->elevator->type->ops.insert_requests(hctx, list,
> +							/*at_head=*/true);
> +		}

But I have no idea how this is related in any way.