Re: [PATCH 1/2] KVM: MMU: Do not treat ZONE_DEVICE pages as being reserved

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Wed, Nov 06, 2019 at 06:14:40PM +0100, Paolo Bonzini wrote:
> On 06/11/19 18:07, Sean Christopherson wrote:
> >  void kvm_get_pfn(kvm_pfn_t pfn)
> >  {
> > -	if (!kvm_is_reserved_pfn(pfn))
> > +	if (!kvm_is_reserved_pfn(pfn) && !WARN_ON(kvm_is_zone_device_pfn(pfn)))
> >  		get_page(pfn_to_page(pfn));
> >  }
> >  EXPORT_SYMBOL_GPL(kvm_get_pfn);
> 
> Can you call remap_pfn_range with a source address that is ZONE_DEVICE?
>  If so, you would get a WARN from the kvm_get_pfn call in
> hva_to_pfn_remapped.

I don't know, at a quick glance I'm guessing it's possible via /dev/mem?
But, get_page() isn't sufficient to properly grab a ZONE_DEVICE page, so
the WARN is a good thing and intentional.

So assuming the answer is "yes", perhaps hva_to_pfn_remapped() should
adds its own check on kvm_is_zone_device_pfn() and return -EINVAL or
something?  That'd likely end up kill the guest, but wouldn't break the
kernel.