Re: [PATCH v12 08/20] KVM: pfncache: allow a cache to be activated with a fixed (userspace) HVA

[Paul Durrant]

On 07/02/2024 04:03, Sean Christopherson wrote:
> +s390 folks (question on kvm_is_error_gpa() for ya)
>
> On Mon, Jan 15, 2024, Paul Durrant wrote:
> > @@ -1398,7 +1414,9 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);
> >   static inline void kvm_gpc_mark_dirty(struct gfn_to_pfn_cache *gpc)
> >   {
> >   	lockdep_assert_held(&gpc->lock);
> > -	mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
> > +
> > +	if (gpc->gpa != KVM_XEN_INVALID_GPA)
>
> KVM_XEN_INVALID_GPA absolutely doesn't belong in common code.  Not to mention
> that it will break when Paolo (rightly) moves it to an x86 header.
>
> https://lore.kernel.org/all/20240131233056.10845-3-pbonzini@xxxxxxxxxx
>
> > +		mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
> >   }
> >   
> >   void kvm_sigset_activate(struct kvm_vcpu *vcpu);
> > diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
> > index 97eec8ee3449..ae822bff812f 100644
> > --- a/virt/kvm/pfncache.c
> > +++ b/virt/kvm/pfncache.c
> > @@ -48,7 +48,10 @@ bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len)
> >   	if (!gpc->active)
> >   		return false;
> >   
> > -	if (gpc->generation != slots->generation || kvm_is_error_hva(gpc->uhva))
> > +	if (gpc->gpa != KVM_XEN_INVALID_GPA && gpc->generation != slots->generation)
>
> This needs a comment.  I know what it's doing, but it wasn't obvious at first
> glance, and it definitely won't be intuitive for readers that aren't intimately
> familiar with memslots.
>
> > +		return false;
> > +
> > +	if (kvm_is_error_hva(gpc->uhva))
> >   		return false;
> >   
> >   	if (offset_in_page(gpc->uhva) + len > PAGE_SIZE)
> > @@ -209,11 +212,13 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
> >   	return -EFAULT;
> >   }
> >   
> > -static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa,
> > +static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long uhva,
> >   			     unsigned long len)
> >   {
> >   	struct kvm_memslots *slots = kvm_memslots(gpc->kvm);
> > -	unsigned long page_offset = offset_in_page(gpa);
> > +	unsigned long page_offset = (gpa != KVM_XEN_INVALID_GPA) ?
> > +		offset_in_page(gpa) :
> > +		offset_in_page(uhva);
>
> This formatting is funky.  I also think it would be worth adding a helper to pair
> with kvm_is_error_hva().
>
> But!  kvm_is_error_gpa() already exists, and it very, very sneakily does a memslot
> lookup and checks for a valid HVA.
>
> s390 people, any objection to renaming kvm_is_error_gpa() to something like
> kvm_gpa_has_memslot() or kvm_gpa_is_in_memslot()?  s390 is the only code that
> uses the existing helper.
>
> That would both to free up the name to pair with kvm_is_error_hva(), and would
> make it obvious what the helper does; I was quite surprised that "error" means
> "is covered by a valid memslot".

Seemingly no response to this; I'll define a local helper rather than 
re-working the open-coded tests to check against INVALID_GPA. This can 
then be trivially replaced if need be.

> Back to this code, then we can have a slightly cleaner:
>
> 	unsigned long page_offset = kvm_is_error_gpa(gpa) ? offset_in_page(gpa) :
> 							    offset_in_page(uhva);
>
>
> And I think it's worth asserting that exactly _one_ of GPA or HVA is valid, e.g.
> to ensure KVM doesn't end up with botched offsets, and to make it a bit more
> clear what's going on.
>
>
> 	if (WARN_ON_ONCE(kvm_is_error_gpa(gpa) == kvm_is_error_hva(uhva))
> 		return -EINVAL;

Sure.