On Mon, Dec 21, 2020 at 04:11:01PM -0800, Linus Torvalds wrote:
> On Mon, Dec 21, 2020 at 4:00 PM Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
> >
> > My first instinct is to be conservative and revert 09854ba94c6a ("mm:
> > do_wp_page() simplification") so people are less likely to come back
> > and complain about performance issues from holding mmap lock for
> > write when clearing pte_write.
>
> Well, the thing is, that simplificaiton was actually part of fixing a
> real regression wrt GUP.
>
> Reverting that would break a308c71bf1e6 ("mm/gup: Remove enfornced COW
> mechanism").
>
> And that one was the (better) fix for commit 17839856fd58 that fixed a
> real security issue, but did it with a big hammer that then caused
> problems for uffd-wp (and some other loads). There's a bit more
> context in the merge message in commit b25d1dc9474e Merge branch
> 'simplify-do_wp_page'.
>
> So while that commit 09854ba94c6a on its own is "just" a
> simplification, it's actually part of a bigger series that fixes
> serious problems.
>
> Linus
Well, that settles it then. I don't any better idea that is as safe
as what Nadav has.
