Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 4 Feb 2021 10:18:43 +0800
From:   Ming Lei <ming.lei@redhat.com>
To:     David Jeffery <djeffery@redhat.com>
Cc:     linux-block@vger.kernel.org, Jens Axboe <axboe@kernel.dk>,
        linux-kernel@vger.kernel.org,
        Laurence Oberman <loberman@redhat.com>
Subject: Re: [PATCH] block: recalculate segment count for multi-segment
 discard requests correctly
Message-ID: <20210204021843.GA1108591@T590>
References: <20210201164850.391332-1-djeffery@redhat.com>
 <20210202033343.GA165584@T590>
 <20210202204355.GA31803@redhat>
 <20210203023517.GA948998@T590>
 <20210203162337.GA40163@redhat>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210203162337.GA40163@redhat>
Precedence: bulk

On Wed, Feb 03, 2021 at 11:23:37AM -0500, David Jeffery wrote:
> On Wed, Feb 03, 2021 at 10:35:17AM +0800, Ming Lei wrote:
> > 
> > On Tue, Feb 02, 2021 at 03:43:55PM -0500, David Jeffery wrote:
> > > The return 0 does seem to be an old relic that does not make sense anymore.
> > > Moving REQ_OP_SECURE_ERASE to be with discard and removing the old return 0,
> > > is this what you had in mind?
> > > 
> > >  
> > > diff --git a/block/blk-merge.c b/block/blk-merge.c
> > > index 808768f6b174..68458aa01b05 100644
> > > --- a/block/blk-merge.c
> > > +++ b/block/blk-merge.c
> > > @@ -383,8 +383,14 @@ unsigned int blk_recalc_rq_segments(struct request *rq)
> > >  	switch (bio_op(rq->bio)) {
> > >  	case REQ_OP_DISCARD:
> > >  	case REQ_OP_SECURE_ERASE:
> > > +		if (queue_max_discard_segments(rq->q) > 1) {
> > > +			struct bio *bio = rq->bio;
> > > +			for_each_bio(bio)
> > > +				nr_phys_segs++;
> > > +			return nr_phys_segs;
> > > +		}
> > > +		/* fall through */
> > >  	case REQ_OP_WRITE_ZEROES:
> > > -		return 0;
> > >  	case REQ_OP_WRITE_SAME:
> > >  		return 1;
> > 
> > WRITE_SAME uses same buffer, so the nr_segment is still one; WRITE_ZERO
> > doesn't need extra payload, so nr_segments is zero, see
> > blk_bio_write_zeroes_split(), blk_bio_write_same_split, attempt_merge()
> > and blk_rq_merge_ok().
> > 
> 
> I thought you mentioned virtio-blk because of how some drivers handle
> zeroing and discarding similarly and wanted to align the segment count with
> discard behavior for WRITE_ZEROES too. (Though that would also need an update

virtio-blk is just one example which supports both single discard range
and multiple discard range, meantime virtblk_setup_discard_write_zeroes()
simply maps write zero into discard directly.

Just found blk_rq_nr_discard_segments() returns >=1 segments always, so
looks your patch is enough for avoiding the warning.

> to blk_bio_write_zeroes_split as you pointed out.)  So you want me to leave
> WRITE_ZEROES behavior alone and let blk_rq_nr_discard_segments() keep doing
> the hiding of a 0 rq->nr_phys_segments as 1 segment in the WRITE_ZEROES treated
> as a discard case?
> 
> diff --git a/block/blk-merge.c b/block/blk-merge.c
> index 808768f6b174..756473295f19 100644
> --- a/block/blk-merge.c
> +++ b/block/blk-merge.c
> @@ -383,6 +383,14 @@ unsigned int blk_recalc_rq_segments(struct request *rq)
>  	switch (bio_op(rq->bio)) {
>  	case REQ_OP_DISCARD:
>  	case REQ_OP_SECURE_ERASE:
> +		if (queue_max_discard_segments(rq->q) > 1) {
> +			struct bio *bio = rq->bio;
> +
> +			for_each_bio(bio)
> +				nr_phys_segs++;
> +			return nr_phys_segs;
> +		}
> +		return 1;
>  	case REQ_OP_WRITE_ZEROES:
>  		return 0;
>  	case REQ_OP_WRITE_SAME:

This patch returns 1 for single-range discard explicitly. However, it
isn't necessary because of blk_rq_nr_discard_segments().

Maybe we can align to blk_bio_discard_split() in future, but that can be
done as cleanup.

Thanks,
Ming