Re: [PATCHv3 08/17] x86/mm: Implement vma_is_encrypted() and vma_keyid()

From: Kirill A. Shutemov
Date: Fri Jun 15 2018 - 09:14:25 EST


On Wed, Jun 13, 2018 at 06:18:05PM +0000, Dave Hansen wrote:
> On 06/12/2018 07:39 AM, Kirill A. Shutemov wrote:
> > +bool vma_is_encrypted(struct vm_area_struct *vma)
> > +{
> > + return pgprot_val(vma->vm_page_prot) & mktme_keyid_mask;
> > +}
> > +
> > +int vma_keyid(struct vm_area_struct *vma)
> > +{
> > + pgprotval_t prot;
> > +
> > + if (!vma_is_anonymous(vma))
> > + return 0;
> > +
> > + prot = pgprot_val(vma->vm_page_prot);
> > + return (prot & mktme_keyid_mask) >> mktme_keyid_shift;
> > +}
>
> Why do we have a vma_is_anonymous() in one of these but not the other?

It shouldn't be there. It's from earlier approach to the function.
I'll fix this.

And I'll drop vma_is_encrypted(). It is not very useful.

> While this reuse of ->vm_page_prot is cute, is there any downside? It's
> the first place I know of that we can't derive ->vm_page_prot from
> ->vm_flags on non-VM_IO/PFNMAP VMAs. Is that a problem?

I don't think so.

It need to be covered in pte_modify() and such, but it's about it.

That's relatively isolated change and we can move KeyID into a standalone
field, if this approach proves to be problematic.

--
Kirill A. Shutemov