On Tue, May 04, 2021, Jim Mattson wrote:
> On Tue, May 4, 2021 at 10:37 AM Jim Mattson <jmattson@xxxxxxxxxx> wrote:
> >
> > On Tue, May 4, 2021 at 10:17 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
> > >
> > > Clear KVM's RDPID capability if the ENABLE_RDTSCP secondary exec control is
> > > unsupported. Despite being enumerated in a separate CPUID flag, RDPID is
> > > bundled under the same VMCS control as RDTSCP and will #UD in VMX non-root
> > > if ENABLE_RDTSCP is not enabled.
> > >
> > > Fixes: 41cd02c6f7f6 ("kvm: x86: Expose RDPID in KVM_GET_SUPPORTED_CPUID")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
> >
> > But KVM will happily emulate RDPID if the instruction causes a #UD
> > VM-exit, won't it? See commit fb6d4d340e05 (KVM: x86: emulate RDPID).
>
> Oh, after reading the second patch, I now see why this is needed.
Yeah. Technically, once common x86 can query MSR_TSC_AUX support directly at
the end of the series, the emulation enumeration could be:
if (kvm_is_supported_user_return_msr(MSR_TSC_AUX))
entry->ecx = F(RDPID);
I think I actually meant to do that, then lost track of that TODO item when
reworking the series for the umpteenth time.
Practically speaking, the only way for kvm_is_supported_user_return_msr() to be
meaningful vs. kvm_cpu_cap_has() is if RDTSCP is supported in hardware but the
VMCS control is not available. And I suppose there's also the case where
X86_FEATURE_RDTSCP was cleared by the kernel, but I feel like KVM should respect
the kernel's avoidance of RDTSCP/MSR_TSC_AUX in that case. Regarding the silly
VMCS case, I have no objection to making the change, but I also don't care if we
sweep it under the rug.
