Re: [PATCH v5 2/7] KVM: Add KVM_PRE_FAULT_MEMORY vcpu ioctl to pre-populate guest memory

[Paolo Bonzini]

On Wed, Jul 10, 2024 at 11:56 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Wed, Jul 10, 2024, Anish Moorthy wrote:
> > On Wed, Jul 10, 2024 at 10:41 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
> > >
> > > +       if (!PAGE_ALIGNED(range->gpa) ||
> > > +           !PAGE_ALIGNED(range->size) ||
> > > ...
> > > +               return -EINVAL;
> >
> > If 'gpa' and 'size' must be page-aligned anyways, doesn't it make
> > sense to just take a 'gfn' and 'num_pages'  and eliminate this error
> > condition?
>
> The downside is that taking gfn+num_pages prevents supporting sub-page pre-faulting
> in the future.  I highly doubt that sub-page mappings will ever be a thing in KVM,
> but two PAGE_ALIGNED() checks is super cheap, so it's soft of a "why not?" scenario.

With ARM having multiple page sizes, and not necessarily the same size
in host and guest, using a gfn argument is also unnecessarily
confusing.

Paolo