On Mon, Apr 08, 2019 at 05:54:25PM +0800, Xiaoyao Li wrote:
> On Fri, 2019-04-05 at 14:30 +0200, Thomas Gleixner wrote:
> > On Wed, 3 Apr 2019, Fenghua Yu wrote:
> > > @@ -1663,6 +1663,12 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct
> > > msr_data *msr_info)
> > > u32 index;
> > >
> > > switch (msr_info->index) {
> > > + case MSR_TEST_CTL:
> > > + if (!msr_info->host_initiated &&
> > > + !(vcpu->arch.core_capability & CORE_CAP_SPLIT_LOCK_DETECT))
> >
> > Errm? If the MSR_TEST_CTL is exposed to the guest via CPUID then the
> > rdmsr() in the guest is not supposed to fail just because
> > CORE_CAP_SPLIT_LOCK_DETECT is not set.
>
> you're right.
>
> > vmx->msr_test_ctl holds the proper value, which is either 0 or
> > CORE_CAP_SPLIT_LOCK_DETECT until more bits are supported.
> >
> > So the chek needs to be guest_cpuid_has(X86_FEATURE_CORE_CAPABILITY).
>
> I don't think so. There is no dependecy between
> guest_cpuid_has(X86_FEATURE_CORE_CAPABILITY) and MSR_TEST_CTL.
> guest_cpuid_has(X86_FEATURE_CORE_CAPABILITY) only indicates that guest has MSR
> CORE_CAPABILITY.
>
> Due to the fact that MSR_TEST_CTL is emulated with vmx->msr_test_ctl. I think it
> 's ok to always let userspace or guest to read MSR_TEST_CTL, it just returns the
> emulated value. Like you said, " vmx->msr_test_ctl holds the proper value, which
> is either 0 or CORE_CAP_SPLIT_LOCK_DETECT until more bits are supported."
Assuming the next version implements "vmx->msr_test_ctl_mask", KVM should
inject #GP if the guest attempts RDMSR(MSR_TEST_CTL) and the mask is zero.
It stands to reason that a kernel can only reasonably assume the MSR exists
if the (virtual) CPU supports at least one feature enabled via MSR_TEST_CTL.
