Re: [PATCH] xfs: drop experimental warning for FSDAX

["Darrick J. Wong" <djwong@kernel.org>]

On Fri, Sep 29, 2023 at 11:28:02AM -0700, Dan Williams wrote:
> Eric Sandeen wrote:
> > On 9/29/23 9:17 AM, Chandan Babu R wrote:
> > > On Thu, Sep 28, 2023 at 09:20:52 AM -0700, Andrew Morton wrote:
> > >> On Thu, 28 Sep 2023 16:44:00 +0800 Shiyang Ruan <ruansy.fnst@fujitsu.com> wrote:
> > >>
> > >>> But please pick the following patch[1] as well, which fixes failures of 
> > >>> xfs55[0-2] cases.
> > >>>
> > >>> [1] 
> > >>> https://lore.kernel.org/linux-xfs/20230913102942.601271-1-ruansy.fnst@fujitsu.com
> > >>
> > >> I guess I can take that xfs patch, as it fixes a DAX patch.  I hope the xfs team
> > >> are watching.
> > >>
> > >> But
> > >>
> > >> a) I'm not subscribed to linux-xfs and
> > >>
> > >> b) the changelog fails to describe the userspace-visible effects of
> > >>    the bug, so I (and others) are unable to determine which kernel
> > >>    versions should be patched.
> > >>
> > >> Please update that changelog and resend?
> > > 
> > > I will apply "xfs: correct calculation for agend and blockcount" patch to
> > > xfs-linux Git tree and include it for the next v6.6 pull request to Linus.
> > > 
> > > At the outset, It looks like I can pick "mm, pmem, xfs: Introduce
> > > MF_MEM_PRE_REMOVE for unbind"
> > > (i.e. https://lore.kernel.org/linux-xfs/20230928103227.250550-1-ruansy.fnst@fujitsu.com/T/#u)
> > > patch for v6.7 as well. But that will require your Ack. Please let me know
> > > your opinion.
> > > 
> > > Also, I will pick "xfs: drop experimental warning for FSDAX" patch for v6.7.
> > 
> > While I hate to drag it out even longer, it seems slightly optimistic to
> > drop experimental at the same time as the "last" fix, in case it's not
> > really the last fix.
> > 
> > But I don't have super strong feelings about it, and I would be happy to
> > finally see experimental go away. So if those who are more tuned into
> > the details are comfortable with that 6.7 plan, I'll defer to them on
> > the question.
> 
> The main blockage of "experimental" was the inability to specify
> dax+reflink, and the concern that resolving that conflict would end up
> breaking MAP_SYNC semantics or some other regression.
> 
> The dax_notify_failure() work has resolved that conflict without
> regressing semantics.
> 
> Ultimately this is an XFS filesystem maintainer decision, but my
> perspective is that v6.7-rc1 starts the clock on experimental going away
> and if the bug reports stay quiet that state can persist into
> v6.7-final.  If new reports crop up, revert the experimental removal and
> try again for v6.8.

I'm ok with this.  Let's merge the PRE_REMOVE patch (and the arithematic
fix) for 6.7-rc1.  If nobody screams during 6.7, send a patch to Linus
removing EXPERIMENTAL after (say) 6.7-rc8.  DAX will no longer be
experimental for the 2024 LTS.

--D