Re: [PATCH v2] KVM: x86: use CPUID to locate host page table reserved bits

[Huang, Kai]

On Wed, 2019-12-04 at 09:57 -0600, Tom Lendacky wrote:
> On 12/4/19 9:40 AM, Paolo Bonzini wrote:
> > The comment in kvm_get_shadow_phys_bits refers to MKTME, but the same is
> > actually
> > true of SME and SEV.  Just use CPUID[0x8000_0008].EAX[7:0] unconditionally
> > if
> > available, it is simplest and works even if memory is not encrypted.
> 
> This isn't correct for AMD. The reduction in physical addressing is
> correct. You can't set, e.g. bit 45, in the nested page table, because
> that will be considered a reserved bit and generate an NPF. When memory
> encryption is enabled today, bit 47 is the encryption indicator bit and
> bits 46:43 must be zero or else an NPF is generated. The hardware uses
> these bits internally based on the whether running as the hypervisor or
> based on the ASID of the guest.

Right. Alghouth both MKTME and SME/SEV reduce physical bits, but they treat
those reduced bits differently: MKTME treats those as keyID thus they can be
set, but SME/SEV treats those as reserved bits so you cannot set any of them.

Maybe the naming of shadow_phys_bits is a little bit confusing here. The purpose
of it was to determine first reserved bits, but not actual physical address bits
. Therefore for MKTME it should include the keyID bits, but for SME/SEV it
should not.

Thanks,
-Kai
> 
> Thanks,
> Tom
> 
> > Cc: stable@xxxxxxxxxxxxxxx
> > Reported-by: Tom Lendacky <thomas.lendacky@xxxxxxx>
> > Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> > ---
> >  arch/x86/kvm/mmu/mmu.c | 20 ++++++++++++--------
> >  1 file changed, 12 insertions(+), 8 deletions(-)
> > 
> > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > index 6f92b40d798c..1e4ee4f8de5f 100644
> > --- a/arch/x86/kvm/mmu/mmu.c
> > +++ b/arch/x86/kvm/mmu/mmu.c
> > @@ -538,16 +538,20 @@ void kvm_mmu_set_mask_ptes(u64 user_mask, u64
> > accessed_mask,
> >  static u8 kvm_get_shadow_phys_bits(void)
> >  {
> >  	/*
> > -	 * boot_cpu_data.x86_phys_bits is reduced when MKTME is detected
> > -	 * in CPU detection code, but MKTME treats those reduced bits as
> > -	 * 'keyID' thus they are not reserved bits. Therefore for MKTME
> > -	 * we should still return physical address bits reported by CPUID.
> > +	 * boot_cpu_data.x86_phys_bits is reduced when MKTME or SME are detected
> > +	 * in CPU detection code, but the processor treats those reduced bits as
> > +	 * 'keyID' thus they are not reserved bits. Therefore KVM needs to look
> > at
> > +	 * the physical address bits reported by CPUID.
> >  	 */
> > -	if (!boot_cpu_has(X86_FEATURE_TME) ||
> > -	    WARN_ON_ONCE(boot_cpu_data.extended_cpuid_level < 0x80000008))
> > -		return boot_cpu_data.x86_phys_bits;
> > +	if (likely(boot_cpu_data.extended_cpuid_level >= 0x80000008))
> > +		return cpuid_eax(0x80000008) & 0xff;
> >  
> > -	return cpuid_eax(0x80000008) & 0xff;
> > +	/*
> > +	 * Quite weird to have VMX or SVM but not MAXPHYADDR; probably a VM with
> > +	 * custom CPUID.  Proceed with whatever the kernel found since these
> > features
> > +	 * aren't virtualizable (SME/SEV also require CPUIDs higher than
> > 0x80000008).
> > +	 */
> > +	return boot_cpu_data.x86_phys_bits;
> >  }
> >  
> >  static void kvm_mmu_reset_all_pte_masks(void)
> >