Re: [PATCH 2/3] KVM: MMU: don not retry #PF for nonpaging guest

[Gleb Natapov]

On Tue, Nov 09, 2010 at 05:52:40PM +0800, Xiao Guangrong wrote:
> >>>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> >>>> index 7f20f2c..606978e 100644
> >>>> --- a/arch/x86/include/asm/kvm_host.h
> >>>> +++ b/arch/x86/include/asm/kvm_host.h
> >>>> @@ -600,6 +600,7 @@ struct kvm_x86_ops {
> >>>>  struct kvm_arch_async_pf {
> >>>>  	u32 token;
> >>>>  	gfn_t gfn;
> >>>> +	bool softmmu;
> >>>>  };
> >>>>  
> >>>>  extern struct kvm_x86_ops *kvm_x86_ops;
> >>>> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> >>>> index f3fad4f..48ca312 100644
> >>>> --- a/arch/x86/kvm/mmu.c
> >>>> +++ b/arch/x86/kvm/mmu.c
> >>>>  static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn)
> >>>> @@ -2602,6 +2607,7 @@ static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn)
> >>>>  	struct kvm_arch_async_pf arch;
> >>>>  	arch.token = (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id;
> >>>>  	arch.gfn = gfn;
> >>>> +	arch.softmmu = mmu_is_softmmu(vcpu);
> >>>>  
> >>> We can do:
> >>>         if (mmu_is_nested(vcpu))
> >>> 		gva = vcpu->mmu.gva_to_gpa(gva);
> >>> And this should fix everything no?
> >>>
> >>
> >> No, since it can't help us to avoid NPF when nested guest run again.
> >>
> > Of course it will not prevent NPF if L2 guest touches it again, but from
> > correctness point of view it is OK. So if L1 will re-use the page for
> > L1 process the page will be already mapped. Not a huge gain I agree, but
> > fix is much more simple.
> > 
> 
> Um, it need hold mmu_lock, and we don't know 'gva''s mapping in PT10 is valid
> or not, also don't know whether it can be accessed later, so the general rule
> is lazily update it. 
> 
We do know that gva's mapping in PT10 is valid since we wouldn't try apf
otherwise. If nested gpa is mapped to a gpa thst is not valid in L0 then
L0 should emulate instruction for L2, no?

> The more important is that we can prefault for softmmu in the later patch,
> it means we can prefault 'gva' in PT20, so don't cook gva here.
> 
So may be just apply second patch then?

--
			Gleb.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/