Re: [PATCH v4 08/10] mm/gup: limit number of gup migration failures, honor failures

[Jason Gunthorpe]

On Fri, Jan 15, 2021 at 01:10:27PM -0500, Pavel Tatashin wrote:

> I've been thinking about this some more. Again, I am not sure this is
> a bug. I understand split_huge_pmd() may split the PMD size page into
> PTEs and leave the compound page intact. However, in order for pages[]
> to have non sequential addresses in compound page, those PTEs must
> also be migrated after split_huge_pmd(), 

Why focus on migrated? Anything could happen to those PTEs: they could
be COW'd, they could be mmap/munmap'd, etc.

Jason