Re: [PATCH v2] block: Limit number of items taken from the I/O scheduler in one go

[Christoph Hellwig]

> index 74cedea56034..b69b780351c1 100644
> --- a/block/blk-mq-sched.c
> +++ b/block/blk-mq-sched.c
> @@ -84,12 +84,17 @@ void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx)
>   * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
>   * its queue by itself in its completion handler, so we don't need to
>   * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
> + *
> + * Returns true if hctx->dispatch was found non-empty and
> + * run_work has to be run again.  This is necessary to avoid
> + * starving flushes.
>   */

Please use up all 80 chars for your comments (just like the existing
part of the comment).

>   * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
>   * its queue by itself in its completion handler, so we don't need to
>   * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
> + *
> + * Returns true if hctx->dispatch was found non-empty and
> + * run_work has to be run again.  This is necessary to avoid
> + * starving flushes.

Same here.

> +again:
> +	run_again = false;
> +
>  	/*
>  	 * If we have previous entries on our dispatch list, grab them first for
>  	 * more fair dispatch.
> @@ -208,19 +236,28 @@ void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
>  		blk_mq_sched_mark_restart_hctx(hctx);
>  		if (blk_mq_dispatch_rq_list(q, &rq_list, false)) {
>  			if (has_sched_dispatch)
> -				blk_mq_do_dispatch_sched(hctx);
> +				run_again = blk_mq_do_dispatch_sched(hctx);
>  			else
> -				blk_mq_do_dispatch_ctx(hctx);
> +				run_again = blk_mq_do_dispatch_ctx(hctx);
>  		}
>  	} else if (has_sched_dispatch) {
> -		blk_mq_do_dispatch_sched(hctx);
> +		run_again = blk_mq_do_dispatch_sched(hctx);
>  	} else if (hctx->dispatch_busy) {
>  		/* dequeue request one by one from sw queue if queue is busy */
> -		blk_mq_do_dispatch_ctx(hctx);
> +		run_again = blk_mq_do_dispatch_ctx(hctx);
>  	} else {
>  		blk_mq_flush_busy_ctxs(hctx, &rq_list);
>  		blk_mq_dispatch_rq_list(q, &rq_list, false);
>  	}
> +
> +	if (run_again) {
> +		if (!restarted) {
> +			restarted = true;
> +			goto again;
> +		}
> +
> +		blk_mq_run_hw_queue(hctx, true);
> +	}

This is a weird loop.  I'd split the code betweem the again label and
the run_again check here into a __blk_mq_sched_dispatch_requests
helper, and then you can do:

	if (__blk_mq_sched_dispatch_requests()) {
		if (__blk_mq_sched_dispatch_requests())
			blk_mq_run_hw_queue(hctx, true);
	}

here.  Preferably with ha good comment explaining the logic.