Re: [RFC 7/7] KVM: VMX: Enable PKS for nested VM

[Chenyi Qiang]

On 8/11/2020 8:05 AM, Jim Mattson wrote:
> On Fri, Aug 7, 2020 at 1:47 AM Chenyi Qiang <chenyi.qiang@xxxxxxxxx> wrote:
> > PKS MSR passes through guest directly. Configure the MSR to match the
> > L0/L1 settings so that nested VM runs PKS properly.
> >
> > Signed-off-by: Chenyi Qiang <chenyi.qiang@xxxxxxxxx>
> > ---
> >   arch/x86/kvm/vmx/nested.c | 32 ++++++++++++++++++++++++++++++++
> >   arch/x86/kvm/vmx/vmcs12.c |  2 ++
> >   arch/x86/kvm/vmx/vmcs12.h |  6 +++++-
> >   arch/x86/kvm/vmx/vmx.c    | 10 ++++++++++
> >   arch/x86/kvm/vmx/vmx.h    |  1 +
> >   5 files changed, 50 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> > index df2c2e733549..1f9823d21ecd 100644
> > --- a/arch/x86/kvm/vmx/nested.c
> > +++ b/arch/x86/kvm/vmx/nested.c
> > @@ -647,6 +647,12 @@ static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
> >                                          MSR_IA32_PRED_CMD,
> >                                          MSR_TYPE_W);
> >
> > +       if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PKRS))
> > +               nested_vmx_disable_intercept_for_msr(
> > +                                       msr_bitmap_l1, msr_bitmap_l0,
> > +                                       MSR_IA32_PKRS,
> > +                                       MSR_TYPE_R | MSR_TYPE_W);
>
> What if L1 intercepts only *reads* of MSR_IA32_PKRS?
>
> >          kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
> >
> >          return true;
>
> > @@ -2509,6 +2519,11 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
> >          if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
> >              !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
> >                  vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
> > +
> > +       if (kvm_cpu_cap_has(X86_FEATURE_PKS) &&
>
> Is the above check superfluous? I would assume that the L1 guest can't
> set VM_ENTRY_LOAD_IA32_PKRS unless this is true.

I enforce this check to ensure vmcs_write to the Guest_IA32_PKRS without 
error. if deleted, vmcs_write to GUEST_IA32_PKRS may executed when PKS 
is unsupported.

> > +           (!vmx->nested.nested_run_pending ||
> > +            !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PKRS)))
> > +               vmcs_write64(GUEST_IA32_PKRS, vmx->nested.vmcs01_guest_pkrs);
>
> This doesn't seem right to me. On the target of a live migration, with
> L2 active at the time the snapshot was taken (i.e.,
> vmx->nested.nested_run_pending=0), it looks like we're going to try to
> overwrite the current L2 PKRS value with L1's PKRS value (except that
> in this situation, vmx->nested.vmcs01_guest_pkrs should actually be
> 0). Am I missing something?

We overwrite the L2 PKRS with L1's value when L2 doesn't support PKS. 
Because the L1's VM_ENTRY_LOAD_IA32_PKRS is off, we need to migrate L1's 
PKRS to L2.

> >          vmx_set_rflags(vcpu, vmcs12->guest_rflags);
> >
> >          /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
>
>
> > @@ -3916,6 +3943,8 @@ static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
> >                  vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
> >          if (kvm_mpx_supported())
> >                  vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
> > +       if (kvm_cpu_cap_has(X86_FEATURE_PKS))
>
> Shouldn't we be checking to see if the *virtual* CPU supports PKS
> before writing anything into vmcs12->guest_ia32_pkrs?

Yes, It's reasonable.

> > +               vmcs12->guest_ia32_pkrs = vmcs_read64(GUEST_IA32_PKRS);
> >
> >          vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
> >   }