On Mon, Jun 18, 2018 at 11:25:57PM -0700, Darrick J. Wong wrote:

Is this going to blow up iomap_dax_zero?  It seems to use both bdev and
dax_dev on __dax_zero_page_range, which definitely uses both.

(Or did all that get rearranged when I wasn't looking?)

Ouch, it does.  And that looks pretty broken.

Also, I guess this will break iomap swapfiles since it checks
iomap->bdev which we stop supplying with this patch...
though I have no idea if DAX swapfiles are even supported.

Not sure if we support it.  We didn't use to support it when
swap used ->bmap, so until someone volunteers to test it we
should disable it with the iomap swapfile code as well.  But
even then doing a detour through the block layer and thus
the bdev makes very little sense.

What's the harm in supplying both pointers?

Just blowing up the size of the iomap.  Especially once we add
the inline data as the third option.