On Thu, 2023-09-07 at 07:45 -0700, Sean Christopherson wrote: > On Wed, Sep 06, 2023, Kai Huang wrote: > > On Thu, 2023-08-24 at 19:07 -0700, Sean Christopherson wrote: > > > --- > > > arch/x86/kvm/mmu/mmu.c | 3 +++ > > > 1 file changed, 3 insertions(+) > > > > > > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c > > > index 1a5a1e7d1eb7..8e2e07ed1a1b 100644 > > > --- a/arch/x86/kvm/mmu/mmu.c > > > +++ b/arch/x86/kvm/mmu/mmu.c > > > @@ -4334,6 +4334,9 @@ static int kvm_faultin_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault, > > > if (unlikely(!fault->slot)) > > > return kvm_handle_noslot_fault(vcpu, fault, access); > > > > > > + if (mmu_invalidate_retry_hva(vcpu->kvm, fault->mmu_seq, fault->hva)) > > > + return RET_PF_RETRY; > > > + > > > > ... Perhaps a comment saying this is to avoid unnecessary MMU lock contention > > would be nice. Otherwise we have is_page_fault_stale() called later within the > > MMU lock. I suppose people only tend to use git blamer when they cannot find > > answer in the code :-) > > Agreed, will add. > > > > return RET_PF_CONTINUE; > > > } > > > > > > > Btw, currently fault->mmu_seq is set in kvm_faultin_pfn(), which happens after > > fast_page_fault(). Conceptually, should we move this to even before > > fast_page_fault() because I assume the range zapping should also apply to the > > cases that fast_page_fault() handles? > > Nope, fast_page_fault() doesn't need to "manually" detect invalidated SPTEs because > it only modifies shadow-present SPTEs and does so with an atomic CMPXCHG. If a > SPTE is zapped by an mmu_notifier event (or anything else), the CMPXCHG will fail > and fast_page_fault() will see the !PRESENT SPTE on the next retry and bail. Ah yes. Thanks.
