Make sure pfn is not changed for a spurious fault by warning in the TDP
MMU. For shadow path, only treat a prefetch fault as spurious when pfn is
not changed, since the rmap removal and add are required when pfn is
changed.

Cc: Sean Christopherson <seanjc@google.com>
Signed-off-by: Yan Zhao <yan.y.zhao@intel.com>
---
 arch/x86/kvm/mmu/mmu.c     | 3 ++-
 arch/x86/kvm/mmu/tdp_mmu.c | 3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 3d74e680006f..47fd3712afe6 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -2846,7 +2846,8 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, struct kvm_memory_slot *slot,
 	}
 
 	if (is_shadow_present_pte(*sptep)) {
-		if (prefetch && is_last_spte(*sptep, level))
+		if (prefetch && is_last_spte(*sptep, level) &&
+		    pfn == spte_to_pfn(*sptep))
 			return RET_PF_SPURIOUS;
 
 		/*
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
index 5f9e7374220e..8b37e4f542f3 100644
--- a/arch/x86/kvm/mmu/tdp_mmu.c
+++ b/arch/x86/kvm/mmu/tdp_mmu.c
@@ -1139,7 +1139,8 @@ static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu,
 
 	if (is_shadow_present_pte(iter->old_spte) &&
 	    is_access_allowed(fault, iter->old_spte) &&
-	    is_last_spte(iter->old_spte, iter->level))
+	    is_last_spte(iter->old_spte, iter->level) &&
+	    !WARN_ON_ONCE(fault->pfn != spte_to_pfn(iter->old_spte)))
 		return RET_PF_SPURIOUS;
 
 	if (unlikely(!fault->slot))
-- 
2.43.2

On Fri, Feb 07, 2025, Yan Zhao wrote:
> Make sure pfn is not changed for a spurious fault by warning in the TDP
> MMU. For shadow path, only treat a prefetch fault as spurious when pfn is
> not changed, since the rmap removal and add are required when pfn is
> changed.

I like sanity checks, but I don't like special casing "prefetch" faults like this.
KVM should _never_ change the PFN of a shadow-present SPTE.  The TDP MMU already
BUG()s on this, and mmu_spte_update() WARNs on the transition.

On Fri, Feb 07, 2025 at 07:07:06AM -0800, Sean Christopherson wrote:
> On Fri, Feb 07, 2025, Yan Zhao wrote:
> > Make sure pfn is not changed for a spurious fault by warning in the TDP
> > MMU. For shadow path, only treat a prefetch fault as spurious when pfn is
> > not changed, since the rmap removal and add are required when pfn is
> > changed.
> 
> I like sanity checks, but I don't like special casing "prefetch" faults like this.
> KVM should _never_ change the PFN of a shadow-present SPTE.  The TDP MMU already
> BUG()s on this, and mmu_spte_update() WARNs on the transition.
However, both TDP MMU and mmu_set_spte() return RET_PF_SPURIOUS directly before
the BUG() in TDP MMU or mmu_spte_update() could be hit.

On Sat, Feb 08, 2025, Yan Zhao wrote:
> On Fri, Feb 07, 2025 at 07:07:06AM -0800, Sean Christopherson wrote:
> > On Fri, Feb 07, 2025, Yan Zhao wrote:
> > > Make sure pfn is not changed for a spurious fault by warning in the TDP
> > > MMU. For shadow path, only treat a prefetch fault as spurious when pfn is
> > > not changed, since the rmap removal and add are required when pfn is
> > > changed.
> > 
> > I like sanity checks, but I don't like special casing "prefetch" faults like this.
> > KVM should _never_ change the PFN of a shadow-present SPTE.  The TDP MMU already
> > BUG()s on this, and mmu_spte_update() WARNs on the transition.
> However, both TDP MMU and mmu_set_spte() return RET_PF_SPURIOUS directly before
> the BUG() in TDP MMU or mmu_spte_update() could be hit.

Ah, that's very different than treating a prefetch fault as !spurious though.  I
would be a-ok with this:

	if (is_shadow_present_pte(iter->old_spte) &&
	    (fault->prefetch || is_access_allowed(fault, iter->old_spte)) &&
	    is_last_spte(iter->old_spte, iter->level)) {
		WARN_ON_ONCE(fault->pfn != spte_to_pfn(iter->old_spte));
		return RET_PF_SPURIOUS;
	}

On Mon, Feb 10, 2025 at 02:23:38PM -0800, Sean Christopherson wrote:
> On Sat, Feb 08, 2025, Yan Zhao wrote:
> > On Fri, Feb 07, 2025 at 07:07:06AM -0800, Sean Christopherson wrote:
> > > On Fri, Feb 07, 2025, Yan Zhao wrote:
> > > > Make sure pfn is not changed for a spurious fault by warning in the TDP
> > > > MMU. For shadow path, only treat a prefetch fault as spurious when pfn is
> > > > not changed, since the rmap removal and add are required when pfn is
> > > > changed.
> > > 
> > > I like sanity checks, but I don't like special casing "prefetch" faults like this.
> > > KVM should _never_ change the PFN of a shadow-present SPTE.  The TDP MMU already
> > > BUG()s on this, and mmu_spte_update() WARNs on the transition.
> > However, both TDP MMU and mmu_set_spte() return RET_PF_SPURIOUS directly before
> > the BUG() in TDP MMU or mmu_spte_update() could be hit.
> 
> Ah, that's very different than treating a prefetch fault as !spurious though.  I
> would be a-ok with this:
> 
> 	if (is_shadow_present_pte(iter->old_spte) &&
> 	    (fault->prefetch || is_access_allowed(fault, iter->old_spte)) &&
> 	    is_last_spte(iter->old_spte, iter->level)) {
> 		WARN_ON_ONCE(fault->pfn != spte_to_pfn(iter->old_spte));
> 		return RET_PF_SPURIOUS;
> 	}

Thanks!
Will also update the shadow MMU part as below.

	if (is_shadow_present_pte(*sptep)) {
		if (prefetch && is_last_spte(*sptep, level)) {
			WARN_ON_ONCE(pfn != spte_to_pfn(*sptep));
			return RET_PF_SPURIOUS;
		}
	}