On Mon, Aug 12, 2019 at 05:06:04PM -0700, Sean Christopherson wrote:
> On Thu, Jul 25, 2019 at 11:12:41AM +0800, Yang Weijiang wrote:
> > diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> > index 652b3876ea5c..ce1d6fe21780 100644
> > --- a/arch/x86/kvm/vmx/vmx.c
> > +++ b/arch/x86/kvm/vmx/vmx.c
> > @@ -1637,6 +1637,11 @@ static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
> > return !(val & ~valid_bits);
> > }
> >
> > +static inline u64 vmx_supported_xss(void)
> > +{
> > + return host_xss;
>
> Do you know if the kernel will ever enable CET_USER but not CET_KERNEL,
> and vice versa? I tried hunting down the logic in the main CET enabling
> series but couldn't find the relevant code.
>
> If the kernel does enable USER vs. KERNEL independently, are we sure that
> KVM can correctly virtualize that state and that the guest OS won't die
> due to expecting all CET features or no CET features?
>
> In other words, do we want to return host_xss as is, or do we want to
> make CET_USER and CET_KERNEL a bundle deal and avoid the headache, e.g.:
>
> if (!(host_xss & XFEATURE_MASK_CET_USER) ||
> !(host_xss & XFEATURE_MASK_CET_KERNEL))
> return host_xss & ~(XFEATURE_MASK_CET_USER |
> XFEATURE_MASK_CET_KERNEL);
> return host_xss;
>
Hi, Sean,
Thanks for review! CET_USER and CET_KERNEL are two independent parts of
CET, but CET_KERNEL part has not been fully implemented yet, the final target
is to enable CET_USER + CET_KERNEL in kernel. In the VMM patch, it's supposed
to enable both CET_USER and CET_KERNEL mode at one time, so the patches expose
all the features of CET to guest OS as long as platform and host kernel
support so.
> > +}
> > +
> > static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
> > {
> > switch (msr->index) {
> > @@ -7724,6 +7729,7 @@ static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
> > .get_vmcs12_pages = NULL,
> > .nested_enable_evmcs = NULL,
> > .need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
> > + .supported_xss = vmx_supported_xss,
> > };
> >
> > static void vmx_cleanup_l1d_flush(void)
> > diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
> > index a470ff0868c5..6a1870044752 100644
> > --- a/arch/x86/kvm/x86.h
> > +++ b/arch/x86/kvm/x86.h
> > @@ -288,6 +288,10 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu, unsigned long cr2,
> > | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
> > | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
> > | XFEATURE_MASK_PKRU)
> > +
> > +#define KVM_SUPPORTED_XSS (XFEATURE_MASK_CET_USER \
> > + | XFEATURE_MASK_CET_KERNEL)
> > +
> > extern u64 host_xcr0;
> >
> > extern u64 kvm_supported_xcr0(void);
> > --
> > 2.17.2
> >
