Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
From:   "NeilBrown" <neilb@suse.de>
To:     "Christoph Hellwig" <hch@infradead.org>
Cc:     "Trond Myklebust" <trond.myklebust@hammerspace.com>,
        "Anna Schumaker" <anna.schumaker@netapp.com>,
        "Chuck Lever" <chuck.lever@oracle.com>,
        "Andrew Morton" <akpm@linux-foundation.org>,
        "Mel Gorman" <mgorman@suse.de>, linux-nfs@vger.kernel.org,
        linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        linux-block@vger.kernel.org
Subject: Re: [PATCH 12/13] MM: use AIO/DIO for reads from SWP_FS_OPS swap-space
In-reply-to: <YZNsf5yvfb8+SiqB@infradead.org>
References: <163702956672.25805.16457749992977493579.stgit@noble.brown>,
 <163703064458.25805.6777856691611196478.stgit@noble.brown>,
 <YZNsf5yvfb8+SiqB@infradead.org>
Date:   Wed, 17 Nov 2021 08:46:14 +1100
Message-id: <163709917463.13692.6685266362531701682@noble.neil.brown.name>
Precedence: bulk

On Tue, 16 Nov 2021, Christoph Hellwig wrote:
> On Tue, Nov 16, 2021 at 01:44:04PM +1100, NeilBrown wrote:
> > When pages a read from SWP_FS_OPS swap-space, the reads are submitted as
> > separate reads for each page.  This is generally less efficient than
> > larger reads.
> >=20
> > We can use the block-plugging infrastructure to delay submitting the
> > read request until multiple contigious pages have been collected.  This
> > requires using ->direct_IO to submit the read (as ->readpages isn't
> > suitable for swap).
>=20
> Abusing the block code here seems little ugly.  Also this won't
> compile if CONFIG_BLOCK is not set, will it?

There is nothing really block-layer-specific about the plugging
interfaces.  I think it would be quite reasonable to move them to lib/
But you are correct that currently without CONFIG_BLOCK the code will
compile but not work.

>=20
> What is the problem with just batching up manually?

That would require a bigger change to common code, which would only
benefit one user.  The plugging mechanism works well for batching
requests to a block device.  Why not use it for non-block-devices too?

Thanks,
NeilBrown


>=20
> > +	/* nofs needs as ->direct_IO may take the same mutex it takes for write=
 */
>=20
> Overly long line.
>=20
>=20