Re: [PATCH] do not call gva_to_gpa on physical address

[Gleb Natapov <gleb@redhat.com>]

On Wed, Jul 14, 2010 at 06:44:15PM +0300, Avi Kivity wrote:
> On 07/14/2010 06:38 PM, Gleb Natapov wrote:
> >When tdp is enabled cr2 passed to emulator is pga not gva, so no need to
> >call gva_to_gpa on it.
> >
> >Signed-off-by: Gleb Natapov<gleb@redhat.com>
> >diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> >index fb08316..721bf0a 100644
> >--- a/arch/x86/kvm/x86.c
> >+++ b/arch/x86/kvm/x86.c
> >@@ -3944,10 +3944,13 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
> >  	if (kvm_mmu_unprotect_page_virt(vcpu, gva))
> >  		return true;
> >
> >-	gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
> >+	if (!tdp_enabled) {
> >+		gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
> >
> >-	if (gpa == UNMAPPED_GVA)
> >-		return true; /* let cpu generate fault */
> >+		if (gpa == UNMAPPED_GVA)
> >+			return true; /* let cpu generate fault */
> >+	} else
> >+		gpa = gva;
> >
> >  	if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa>>  PAGE_SHIFT)))
> >  		return true;
> 
> Why not
> 
>     if (tdp_enabled)
>               return false;
> 
> at the top?  Is there any reason ever to reexecute an instruction
> with tdp enabled?
> 
Can't think of one now. If in the future we will shadow some other data
structures in the guest except page tables (nested vmx/svm?) there may
be the reason.

--
			Gleb.