Re: [PATCH v4 16/20] KVM:x86: Optimize CET supervisor SSP save/reload

[Yang, Weijiang]

On 7/27/2023 11:27 AM, Chao Gao wrote:
> On Thu, Jul 20, 2023 at 11:03:48PM -0400, Yang Weijiang wrote:
> > /*
> >   * Writes msr value into the appropriate "register".
> >   * Returns 0 on success, non-0 otherwise.
> > @@ -2427,7 +2439,16 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> > #define CET_LEG_BITMAP_BASE(data)	((data) >> 12)
> > #define CET_EXCLUSIVE_BITS		(CET_SUPPRESS | CET_WAIT_ENDBR)
> > 	case MSR_IA32_PL0_SSP ... MSR_IA32_PL3_SSP:
> > -		return kvm_set_msr_common(vcpu, msr_info);
> > +		if (kvm_set_msr_common(vcpu, msr_info))
> > +			return 1;
> > +		/*
> > +		 * Write to the base SSP MSRs should happen ahead of toggling
> > +		 * of IA32_S_CET.SH_STK_EN bit.
> Is this a requirement from SDM? And how is this related to the change below?

No, after a second thought, the usage of the supervisor SSPs doesn't 
necessary mean

supervisor SHSTK is being enabled, e.g., used as some HW registers. I'll 
remove it.

> Note that PLx_SSP MSRs are linear addresses of shadow stacks for different CPLs.
> I may think using the page at 0 (assuming 0 is the reset value of PLx SSP) is
> allowed in architecture although probably no kernel will do so.
>
> I don't understand why this comment is needed. I suggest dropping it.

will drop it, thanks!

> > +		 */
> > +		if (msr_index != MSR_IA32_PL3_SSP && data) {
> > +			vmx_disable_write_intercept_sss_msr(vcpu);
> > +			wrmsrl(msr_index, data);
> > +		}