Re: [PATCH 1/2] mm: Allow architectures to request 'old' entries when prefaulting

[Linus Torvalds]

On Mon, Dec 28, 2020 at 2:05 PM Kirill A. Shutemov <kirill@xxxxxxxxxxxxx> wrote:
>
>
> But I *think* we should be fine here: do_fault_around() limits start_pgoff
> and end_pgoff to stay within the page table.

Yeah, but I was thinking it would then update ->pte to just past the edge.

But looking at that logic, you're right - it will update ->pte and
->address only just before installing the pte, so it will never go
_to_ the edge, it will always stay inside.

So scratch my suspicion. It looked promising mainly because that ->pte
pointer update was one of the things that changed when you instead
compared against the address.

         Linus