Re: [PATCH v8 22/26] KVM: VMX: Set up interception for CET MSRs

[Yang, Weijiang]

On 1/17/2024 1:31 PM, Yuan Yao wrote:
> On Wed, Jan 17, 2024 at 09:58:40AM +0800, Yang, Weijiang wrote:
> > On 1/17/2024 9:41 AM, Yang, Weijiang wrote:
> > > On 1/15/2024 5:58 PM, Yuan Yao wrote:
> > > > On Thu, Dec 21, 2023 at 09:02:35AM -0500, Yang Weijiang wrote:
> > [...]
> > > > Looks this leading to MSR_IA32_INT_SSP_TAB not intercepted
> > > > after below steps:
> > > >
> > > > Step 1. User space set cpuid w/  X86_FEATURE_LM, w/  SHSTK.
> > > > Step 2. User space set cpuid w/o X86_FEATURE_LM, w/o SHSTK.
> > > >
> > > > Then MSR_IA32_INT_SSP_TAB won't be intercepted even w/o SHSTK
> > > > on guest cpuid, will this lead to inconsistency when do
> > > > rdmsr(MSR_IA32_INT_SSP_TAB) from guest in this scenario ?
> > > Yes, theoretically it's possible, how about changing it as below?
> > >
> > > vmx_set_intercept_for_msr(vcpu, MSR_IA32_INT_SSP_TAB,
> > > 			  MSR_TYPE_RW,
> > > 			  incpt | !guest_cpuid_has(vcpu, X86_FEATURE_LM));
> > Oops, should be : incpt || !guest_cpuid_has(vcpu, X86_FEATURE_LM)
> It means guest cpuid:
>
> "has X86_FEATURE_SHSTK" + "doesn't have X86_FEATURE_LM"

No, this is invalid within this series. With patch 21 to prevent SHSTK in
32-bit guest, I think the check of LM here can be omitted.

then
vmx_set_intercept_for_msr(vcpu, MSR_IA32_INT_SSP_TAB, MSR_TYPE_RW, incpt); is OK.

> not sure this is valid combination or not.
> If yes it's ok, else just relies on incpt is enough ?