Re: [PATCH 14/15] KVM: x86: Tie Intel and AMD behavior for MSR_TSC_AUX to guest CPU model

[Maxim Levitsky <mlevitsk@redhat.com>]

On Mon, 2021-05-10 at 10:11 -0700, Jim Mattson wrote:
> On Mon, May 10, 2021 at 9:50 AM Sean Christopherson <seanjc@google.com> wrote:
> > On Mon, May 10, 2021, Maxim Levitsky wrote:
> > > On Tue, 2021-05-04 at 10:17 -0700, Sean Christopherson wrote:
> > > > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > > > index de921935e8de..6c7c6a303cc5 100644
> > > > --- a/arch/x86/kvm/svm/svm.c
> > > > +++ b/arch/x86/kvm/svm/svm.c
> > > > @@ -2663,12 +2663,6 @@ static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> > > >                     msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
> > > >             break;
> > > >     case MSR_TSC_AUX:
> > > > -           if (tsc_aux_uret_slot < 0)
> > > > -                   return 1;
> > > > -           if (!msr_info->host_initiated &&
> > > Not related to this patch, but I do wonder why do we need
> > > to always allow writing this msr if done by the host,
> > > since if neither RDTSPC nor RDPID are supported, the guest
> > > won't be able to read this msr at all.
> > 
> > It's an ordering thing and not specific to MSR_TSC_AUX.  Exempting host userspace
> > from guest CPUID checks allows userspace to set MSR state, e.g. during migration,
> > before setting the guest CPUID model.
> 
> I thought the rule was that if an MSR was enumerated by
> KVM_GET_MSR_INDEX_LIST, then KVM had to accept legal writes from the
> host. The only "ordering thing" is that KVM_GET_MSR_INDEX_LIST is a
> device ioctl, so it can't take guest CPUID information into account.

This makes sense.

Thanks!
Best regards,
	Maxim Levitsky
> 
> > > > -               !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
> > > > -               !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
> > > > -                   return 1;
> > > >             msr_info->data = svm->tsc_aux;
> > > >             break;
> > > >     /*