Re: [Patch mm-hotfixes v4] mm/page_vma_mapped: fix device-private PMD handling

[Lance Yang]

On 2026/6/27 08:38, Wei Yang wrote:
[...]
> > Might be good with this on top:
> >
> > ---8<---
> > diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c
> > index cfa1230c87bb..8b7c062bd81d 100644
> > --- a/mm/page_vma_mapped.c
> > +++ b/mm/page_vma_mapped.c
> > @@ -281,7 +281,7 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
> > 					return not_found(pvmw);
> > 				return true;
> > 			}
> > -			/* THP pmd was split under us: handle on pte level */
> > +			/* THP/device-private pmd was split under us: handle on pte level */
>
> As the comment in commit 65edfda6f3f2 ("mm/rmap: extend rmap and migration
> support device-private entries") says:
>
>    Add device-private THP support...
>
> Per my understanding, we first already setup mapping and "migrate" to device
> memory. This looks a kind of place holder.
>
> Not familiar with this. Just want to clarify, we want to treat device-private
> pmd as some sort of THP or not?

Not a regular THP, obviously. Just the PMD-sized device-private entry case.

It can be split under us too; see commit 146287290023 ("mm/huge_memory:
implement device-private THP splitting").

Nothing deeper meant here. After taking PTL, if that PMD-sized entry
is gone, just drop to the PTE walk.

> > 			spin_unlock(pvmw->ptl);
> > 			pvmw->ptl = NULL;
> > 		} else if (!pmd_present(pmde)) {
> > --