Re: [PATCH v3] block: Limit number of items taken from the I/O scheduler in one go

[Christoph Hellwig]

> +	/*
> +	 * A return of -EAGAIN is an indication that hctx->dispatch is not
> +	 * empty and we must run again in order to avoid starving flushes.
> +	 */
> +	if (__blk_mq_sched_dispatch_requests(hctx) == -EAGAIN) {
> +		if (__blk_mq_sched_dispatch_requests(hctx) == -EAGAIN)
> +			blk_mq_run_hw_queue(hctx, true);
> +	}

The comment doesn't explain why we give up after the second attempt and
do a blk_mq_run_hw_queue instead.