Re: [PATCH v2 01/17] mm/gup: Fixup p*_access_permitted()

[Dave Hansen]

On 12/15/2017 05:10 PM, Linus Torvalds wrote:
> Because *if* we want to check protection keys, I think we should do
> that at the vma layer, partly exactly because the exact implementation
> of protection keys is so architecture-specific, and partly because I
> don't think it makes sense to check them for every page anyway.

So, there are VMA checks against protection keys.  The problem _here_ is
that we are checking against the VMA (and correctly skipping the PKRU
checks) and then _mistakenly_ applying the PTE checks against PKRU.

I think the reason we needed VMA and PTE checks was the
get_user_pages_fast() path not having a VMA.

I need to go re-read the commits, though.