Re: [PATCH v2 02/12] block: Send flush requests to the I/O scheduler

[Christoph Hellwig <hch@lst.de>]

On Mon, Apr 10, 2023 at 05:15:44PM -0700, Bart Van Assche wrote:
> Subject: [PATCH] block: Send flush requests to the I/O scheduler
>
> Send flush requests to the I/O scheduler such that I/O scheduler policies
> are applied to writes with the FUA flag set. Separate the I/O scheduler
> members from the flush members in struct request since with this patch
> applied a request may pass through both an I/O scheduler and the flush
> machinery.
>
> This change affects the statistics of I/O schedulers that track I/O
> statistics (BFQ and mq-deadline).

This looks reasonably to me, as these special cases are nasty.
But we'll need very careful testing, including performance testing
to ensure this doesn't regress.

> +		blk_mq_sched_insert_request(rq, /*at_head=*/false,
> +					    /*run_queue=*/true, /*async=*/true);

And place drop these silly comments.  If you want to do something about
this rather suboptimal interface convert the three booleans to a flags
argument with properly named flags.

> -	if ((rq->rq_flags & RQF_FLUSH_SEQ) || blk_rq_is_passthrough(rq))
> -		return true;
> -
> -	return false;
> +	return req_op(rq) == REQ_OP_FLUSH || blk_rq_is_passthrough(rq);

This just seem like an arbitrary reformatting.  While I also prefer
your new version, I don't think it belongs into this patch.