On Wed, Sep 29, 2021 at 12:57 PM Peter Xu <peterx@xxxxxxxxxx> wrote:
>
> Now we have 3 callers of page_maybe_dma_pinned():
>
> 1. page_needs_cow_for_dma
> 2. pte_is_pinned
> 3. shrink_page_list
>
> The 1st one is good as it takes the seqlock for write properly. The 2nd & 3rd
> are missing, we may need to add them.
Well, the pte_is_pinned() case at least could do the seqlock in
clear_soft_dirty() - it has the vma and mm available.
The page shrinker has always been problematic since it doesn't have
the vm (and by "always" I mean "modern times" - long ago we used to
scan virtually, in the days before rmap)
One option might be for fast-gup to give up on locked pages. I think
the page lock is the only thing that shrink_page_list() serializes
with.
Linus
