Re: [PATCH v3 2/2] x86/vmx: optimize MSR_MISC_FEATURES_ENABLES switch

[Sean Christopherson]

On Mon, Mar 25, 2019 at 04:06:50PM +0800, Xiaoyao Li wrote:
> KVM needs to switch MSR_MISC_FEATURES_ENABLES between host and guest in
> every pcpu/vcpu context switch. Since WRMSR is expensive, this patch tries
> to save cycles by avoiding WRMSR MSR_MISC_FEATURES_ENABLES whenever possible.
> 
> If host's value is zero, nothing needs to be done, since guest can use
> kvm emulated cpuid faulting.
> 
> If host's value is non-zero, it need not clear MSR_MISC_FEATURES_ENABLES
> unconditionally. We can use hardware cpuid faulting if guest's
> value is equal to host'value, thus avoid WRMSR
> MSR_MISC_FEATURES_ENABLES.
> 
> Since hardware cpuid faulting takes higher priority than CPUID vm exit,
> it should be updated to hardware while guest wrmsr and hardware cpuid
> faulting is used for guest.
> 
> Signed-off-by: Xiaoyao Li <xiaoyao.li@xxxxxxxxxxxxxxx>
> ---
>  arch/x86/include/asm/kvm_host.h |  2 ++
>  arch/x86/kvm/vmx/vmx.c          | 13 ++++++++++---
>  arch/x86/kvm/x86.c              | 15 ++++++++++++---
>  3 files changed, 24 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 2c53df4a5a2a..9bcb444b903d 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1343,6 +1343,8 @@ void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw);
>  void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l);
>  int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr);
>  
> +int kvm_supported_msr_misc_features_enables(struct kvm_vcpu *vcpu, u64 data);
> +
>  int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr);
>  int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr);
>  
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 65aa947947ba..73bb11f74b36 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -1035,10 +1035,10 @@ static void vmx_prepare_guest_misc_features_enables(struct vcpu_vmx *vmx)
>  {
>  	u64 msrval = this_cpu_read(msr_misc_features_shadow);
>  
> -	if (!msrval)
> +	if (!msrval || msrval == vmx->vcpu.arch.msr_misc_features_enables)
>  		return;
>  
> -	wrmsrl(MSR_MISC_FEATURES_ENABLES, 0ULL);
> +	wrmsrl(MSR_MISC_FEATURES_ENABLES, vmx->vcpu.arch.msr_misc_features_enables);
>  }
>  
>  void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
> @@ -1136,7 +1136,7 @@ static void vmx_load_host_misc_features_enables(struct vcpu_vmx *vmx)
>  {
>  	u64 msrval = this_cpu_read(msr_misc_features_shadow);
>  
> -	if (!msrval)
> +	if (!msrval || msrval == vmx->vcpu.arch.msr_misc_features_enables)
>  		return;
>  
>  	wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
> @@ -2043,6 +2043,13 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>  		else
>  			vmx->pt_desc.guest.addr_a[index / 2] = data;
>  		break;
> +	case MSR_MISC_FEATURES_ENABLES:
> +		if (!kvm_supported_msr_misc_features_enables(vcpu, data))
> +			return 1;
> +		if (this_cpu_read(msr_misc_features_shadow) && vmx->loaded_cpu_state)
> +			wrmsrl(MSR_MISC_FEATURES_ENABLES, data);
> +		vcpu->arch.msr_misc_features_enables = data;
> +		break;
>  	case MSR_TSC_AUX:
>  		if (!msr_info->host_initiated &&
>  		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index ad1df965574e..d2af90422a51 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2449,6 +2449,17 @@ static void record_steal_time(struct kvm_vcpu *vcpu)
>  		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
>  }
>  
> +int kvm_supported_msr_misc_features_enables(struct kvm_vcpu *vcpu, u64 data)

The name kvm_supported_msr_misc_features_enables() is a bit confusing,
e.g. I expected it to return a bitmask of the supported features.  Maybe
kvm_misc_features_enables_msr_valid() to be consistent with
vmx_feature_control_msr_valid()?

> +{
> +	if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
> +	    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
> +	     !supports_cpuid_fault(vcpu)))
> +		return 0;
> +	else
> +		return 1;

No need for the if-else, you can simply do:

	return (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT) ||
	       (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
		!supports_cpuid_fault(vcpu));

You could even shorten it to:

        return (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT) ||
               (data && !supports_cpuid_fault(vcpu));

although that might be getting a bit too cute.

> +}
> +EXPORT_SYMBOL_GPL(kvm_supported_msr_misc_features_enables);
> +
>  int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>  {
>  	bool pr = false;
> @@ -2669,9 +2680,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>  		vcpu->arch.msr_platform_info = data;
>  		break;
>  	case MSR_MISC_FEATURES_ENABLES:
> -		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
> -		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
> -		     !supports_cpuid_fault(vcpu)))
> +		if (!kvm_supported_msr_misc_features_enables(vcpu, data))
>  			return 1;
>  		vcpu->arch.msr_misc_features_enables = data;
>  		break;
> -- 
> 2.19.1
>