Re: [5.4 PATCH] mm/gup: Do not force a COW break on file-backed memory

[Linus Torvalds]

On Thu, Dec 2, 2021 at 11:59 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> Oh absolutely, it's semantically nonsense.  The only reason it fixes the
> problem is that VM_DENYWRITE VMAs are the only ones considered for the
> RO_THP merging, so they're the only ones which we've seen causing a
> problem.

That would be a semantically meaningful argument, but I think the
reverse isn't true: regular pages in VM_DENYWRITE vmas - that aren't
using the RO_THP thing - are open to the same old "COW wrong way"
issue.

So it's not like VM_DENYWRITE is really meaningful for the
conditional, even if it's perhaps a prerequisite for it being a
problem.

> > I suspect a real fix would be bigger and more invasive.
>
> Darn.  I was hoping you were going to say something like "The real
> problem is follow_trans_huge_pmd() is complete garbage and it should
> just do X, Y and Z".  Or "When we force on FOLL_WRITE, we should also
> force on FOLL_SPLIT_PMD".

Well, maybe that "FOLL_SPLIT_PMD" thing would indeed be a valid thing?

But I _think_ that it shouldn't be too painful to do the
should_force_cow_break() call later, when you actually have the exact
page details, so while not exactly a one-liner, I hope that approach
would end up working out.

I only gave it a quick look, though, I might be missing something.

              Linus