Re: [for-4.14 RFC PATCH 0/2] dm rq: eliminate historic blk-mq and .request_fn queue stacking restrictions

From: Christoph Hellwig <hch@lst.de>
To: Mike Snitzer <snitzer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>,
	dm-devel@redhat.com, linux-block@vger.kernel.org,
	linux-scsi@vger.kernel.org
Subject: Re: [for-4.14 RFC PATCH 0/2] dm rq: eliminate historic blk-mq and .request_fn queue stacking restrictions
Date: Sat, 15 Jul 2017 10:44:12 +0200	[thread overview]
Message-ID: <20170715084412.GB23189@lst.de> (raw)
In-Reply-To: <20170714140206.GA18245@redhat.com>

On Fri, Jul 14, 2017 at 10:02:06AM -0400, Mike Snitzer wrote:
> On Fri, Jul 14 2017 at  3:12am -0400,
> Christoph Hellwig <hch@lst.de> wrote:
> 
> > Btw, we might want to expedite this to 4.13, a 4.13 now defaults
> > to blk-mq for scsi, and this patch would make sure that dm keeps
> > on just working with that switch.
> 
> Don't think we need to rush anything in response to that change in
> SCSI's default.  old .request_fn DM multipath works happily ontop of
> blk-mq devices (so long as all paths are blk-mq).

You're right.  In that case I think we should just skip this series
and I'll dust of the patch to just kill the non-mq support for 3.14
if the switch of scsi to default to mq works out for 3.13.

> It is just blk-mq DM multipath ontop of old .request_fn paths that is
> disallowed in current upstream code.
> 
> But again, I really don't see why we should even want/need to support
> that mode... hence my question raised in this RFC.

I think this mode makes sense in the long run - to get rid of the
legacy request code in dm.  But as long as we keep both modes arounds
the use for it seems a big questionable indeed.