On Fri, Dec 17, 2021 at 6:17 PM Linus Torvalds
[off-list ref] wrote:

I think the hugepage code should use the exact same logic that the
regular wp fault code does.

IOW, I think that this stupid (AND UNTESTED) patch should likely just
fix David's test-case with the hugepage and splice thing..

Or at least be somewhat close.  But it should be paired with the GUP
side doing the right thing too, of course. Maybe it already does,
maybe it doesn't, I didn't check...

And maybe there's something subtle about the page_count() of a THP
entry. Again, I can't really claim to have tested this all, but I'm
hoping this makes somebody go "Ahh, now I see what Linus means"

            Linus