On Wed, Feb 07, 2024, Michael Roth wrote:
> On Tue, Jan 30, 2024 at 05:13:00PM -0800, Sean Christopherson wrote:
> > On Mon, Oct 16, 2023, Michael Roth wrote:
> > > For KVM_X86_SNP_VM, only the PFERR_GUEST_ENC_MASK flag is needed to
> > > determine with an #NPF is due to a private/shared access by the guest.
> > > Implement that handling here. Also add handling needed to deal with
> > > SNP guests which in some cases will make MMIO accesses with the
> > > encryption bit.
> >
> > ...
> >
> > > @@ -4356,12 +4357,19 @@ static int __kvm_faultin_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault
> > > return RET_PF_EMULATE;
> > > }
> > >
> > > - if (fault->is_private != kvm_mem_is_private(vcpu->kvm, fault->gfn)) {
> > > + /*
> > > + * In some cases SNP guests will make MMIO accesses with the encryption
> > > + * bit set. Handle these via the normal MMIO fault path.
> > > + */
> > > + if (!slot && private_fault && kvm_is_vm_type(vcpu->kvm, KVM_X86_SNP_VM))
> > > + private_fault = false;
> >
> > Why? This is inarguably a guest bug.
>
> AFAICT this isn't explicitly disallowed by the SNP spec.
There are _lots_ of things that aren't explicitly disallowed by the APM, that
doesn't mean that _KVM_ needs to actively support them.
I am *not* taking on more broken crud in KVM to workaround OVMF's stupidity, the
KVM_X86_QUIRK_CD_NW_CLEARED has taken up literally days of my time at this point.
> So KVM would need to allow for these cases in order to be fully compatible
> with existing SNP guests that do this.
No. KVM does not yet support SNP, so as far as KVM's ABI goes, there are no
existing guests. Yes, I realize that I am burying my head in the sand to some
extent, but it is simply not sustainable for KVM to keep trying to pick up the
pieces of poorly defined hardware specs and broken guest firmware.
> > > +static bool kvm_mmu_fault_is_private(struct kvm *kvm, gpa_t gpa, u64 err)
> > > +{
> > > + bool private_fault = false;
> > > +
> > > + if (kvm_is_vm_type(kvm, KVM_X86_SNP_VM)) {
> > > + private_fault = !!(err & PFERR_GUEST_ENC_MASK);
> > > + } else if (kvm_is_vm_type(kvm, KVM_X86_SW_PROTECTED_VM)) {
> > > + /*
> > > + * This handling is for gmem self-tests and guests that treat
> > > + * userspace as the authority on whether a fault should be
> > > + * private or not.
> > > + */
> > > + private_fault = kvm_mem_is_private(kvm, gpa >> PAGE_SHIFT);
> > > + }
> >
> > This can be more simply:
> >
> > if (kvm_is_vm_type(kvm, KVM_X86_SNP_VM))
> > return !!(err & PFERR_GUEST_ENC_MASK);
> >
> > if (kvm_is_vm_type(kvm, KVM_X86_SW_PROTECTED_VM))
> > return kvm_mem_is_private(kvm, gpa >> PAGE_SHIFT);
> >
>
> Yes, indeed. But TDX has taken a different approach for SW_PROTECTED_VM
> case where they do this check in kvm_mmu_page_fault() and then synthesize
> the PFERR_GUEST_ENC_MASK into error_code before calling
> kvm_mmu_do_page_fault(). It's not in the v18 patchset AFAICT, but it's
> in the tdx-upstream git branch that corresponds to it:
>
> https://github.com/intel/tdx/commit/3717a903ef453aa7b62e7eb65f230566b7f158d4
>
> Would you prefer that SNP adopt the same approach?
Ah, yes, 'twas my suggestion in the first place. FWIW, I was just reviewing the
literal code here and wasn't paying much attention to the content.
https://lore.kernel.org/all/f474282d701aca7af00e4f7171445abb5e734c6f.1689893403.git.isaku.yamahata@xxxxxxxxx
