RE: [PATCH 1/4] KVM: x86: Add a struct to consolidate host values, e.g. EFER, XCR0, etc...

[Wang, Wei W]

On Thursday, April 25, 2024 10:10 PM, Sean Christopherson wrote:
> On Thu, Apr 25, 2024, Wei W Wang wrote:
> > On Wednesday, April 24, 2024 6:15 AM, Sean Christopherson wrote:
> > > @@ -403,7 +403,7 @@ static void vmx_update_fb_clear_dis(struct
> > > kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
> > >  	 * and VM-Exit.
> > >  	 */
> > >  	vmx->disable_fb_clear
> > > = !cpu_feature_enabled(X86_FEATURE_CLEAR_CPU_BUF) &&
> > > -				(host_arch_capabilities &
> > > ARCH_CAP_FB_CLEAR_CTRL) &&
> > > +				(kvm_host.arch_capabilities &
> > > ARCH_CAP_FB_CLEAR_CTRL) &&
> >
> > The line of code appears to be lengthy. It would be preferable to
> > limit it to under
> > 80 columns per line.
> 
> I agree that staying under 80 is generally preferred, but I find this
> 
> 	vmx->disable_fb_clear = (kvm_host.arch_capabilities &
> ARCH_CAP_FB_CLEAR_CTRL) &&
> 				!boot_cpu_has_bug(X86_BUG_MDS) &&
> 				!boot_cpu_has_bug(X86_BUG_TAA);
> 
> much more readable than this
> 
> 	vmx->disable_fb_clear = (kvm_host.arch_capabilities &
> 			 	 ARCH_CAP_FB_CLEAR_CTRL) &&
> 				!boot_cpu_has_bug(X86_BUG_MDS) &&
> 				!boot_cpu_has_bug(X86_BUG_TAA);
> 
> We should shorten the name to arch_caps, but I don't think that's a net
> positive, e.g. unless we do a bulk rename, it'd diverge from several other
> functions/variables, and IMO it would be less obvious that the field holds
> MSR_IA32_ARCH_CAPABILITIES.

Yeah, the above isn't nice and no need to do bulk rename.
We could just shorten it here, e.g.:

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 4ed8c73f88e4..8d0ab5a6a515 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -393,6 +393,9 @@ static __always_inline void vmx_enable_fb_clear(struct vcpu_vmx *vmx)

 static void vmx_update_fb_clear_dis(struct kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
 {
+       u64 arch_cap = kvm_host.arch_capabilities;
+
        /*
         * Disable VERW's behavior of clearing CPU buffers for the guest if the
         * CPU isn't affected by MDS/TAA, and the host hasn't forcefully enabled
@@ -402,7 +405,7 @@ static void vmx_update_fb_clear_dis(struct kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
         * and VM-Exit.
         */
        vmx->disable_fb_clear = !cpu_feature_enabled(X86_FEATURE_CLEAR_CPU_BUF) &&
-                               (kvm_host.arch_capabilities & ARCH_CAP_FB_CLEAR_CTRL) &&
+                               (arch_cap & ARCH_CAP_FB_CLEAR_CTRL) &&
                                !boot_cpu_has_bug(X86_BUG_MDS) &&
                                !boot_cpu_has_bug(X86_BUG_TAA);


> 
> > >  				!boot_cpu_has_bug(X86_BUG_MDS) &&
> > >  				!boot_cpu_has_bug(X86_BUG_TAA);
> > >
> 
> > > @@ -325,11 +332,8 @@ int x86_emulate_instruction(struct kvm_vcpu
> > > *vcpu, gpa_t cr2_or_gpa,
> > >  			    int emulation_type, void *insn, int insn_len);
> fastpath_t
> > > handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
> > >
> > > -extern u64 host_xcr0;
> > > -extern u64 host_xss;
> > > -extern u64 host_arch_capabilities;
> > > -
> > >  extern struct kvm_caps kvm_caps;
> > > +extern struct kvm_host_values kvm_host;
> >
> > Have you considered merging the kvm_host_values and kvm_caps into one
> > unified structure?
> 
> No really.  I don't see any benefit, only the downside of having to come up
> with a name that is intuitive when reading code related to both.

I thought the two structures perform quite similar jobs and most of the fields in
kvm_cap, e.g. has_tsc_control, supported_perf_cap, could also be interpreted
as host values?