futher decouple DAX from block devices

[Darrick J. Wong]

On Thu, Nov 04, 2021 at 11:10:19AM -0700, Dan Williams wrote:
> On Thu, Nov 4, 2021 at 10:36 AM Christoph Hellwig <hch at lst.de> wrote:
> >
> > On Thu, Nov 04, 2021 at 10:34:17AM -0700, Darrick J. Wong wrote:
> > > /me wonders, are block devices going away?  Will mkfs.xfs have to learn
> > > how to talk to certain chardevs?  I guess jffs2 and others already do
> > > that kind of thing... but I suppose I can wait for the real draft to
> > > show up to ramble further. ;)
> >
> > Right now I've mostly been looking into the kernel side.  An no, I
> > do not expect /dev/pmem* to go away as you'll still need it for a
> > not DAX aware file system and/or application (such as mkfs initially).
> >
> > But yes, just pointing mkfs to the chardev should be doable with very
> > little work.  We can point it to a regular file after all.
> 
> Note that I've avoided implementing read/write fops for dax devices
> partly out of concern for not wanting to figure out shared-mmap vs
> write coherence issues, but also because of a bet with Dave Hansen
> that device-dax not grow features like what happened to hugetlbfs. So
> it would seem mkfs would need to switch to mmap I/O, or bite the
> bullet and implement read/write fops in the driver.

That ... would require a fair amount of userspace changes, though at
least e2fsprogs has pluggable io drivers, which would make mmapping a
character device not too awful.

xfsprogs would be another story -- porting the buffer cache mignt not be
too bad, but mkfs and repair seem to issue pread/pwrite calls directly.
Note that xfsprogs explicitly screens out chardevs.

--D