Re: [PATCH][v2] KVM: Clear pv eoi pending bit only when it is set

[Vitaly Kuznetsov <vkuznets@redhat.com>]

Li RongQing <lirongqing@baidu.com> writes:

> merge pv_eoi_get_pending and pv_eoi_clr_pending into a single
> function pv_eoi_test_and_clear_pending, which returns and clear
> the value of the pending bit.
>
> and clear pv eoi pending bit only when it is set, to avoid calling
> pv_eoi_put_user(), this can speed about 300 nsec on AMD EPYC most
> of the time
>
> and make pv_eoi_set_pending as inline as there is only one user

Compiler is likely smart enough to inline static functions with a single
user anyway.

>
> Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
> Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
> diff with v1:
>  merge as pv_eoi_test_and_clear_pending
>  add inline for pv_eoi_set_pending
>
>  arch/x86/kvm/lapic.c |   47 +++++++++++++++++++++++------------------------
>  1 files changed, 23 insertions(+), 24 deletions(-)
>
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index 76fb009..4da5db8 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -673,18 +673,7 @@ static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
>  	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
>  }
>  
> -static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
> -{
> -	u8 val;
> -	if (pv_eoi_get_user(vcpu, &val) < 0) {
> -		printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
> -			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);
> -		return false;
> -	}
> -	return val & KVM_PV_EOI_ENABLED;
> -}
> -
> -static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
> +static inline void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
>  {
>  	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
>  		printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
> @@ -694,14 +683,31 @@ static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
>  	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
>  }
>  
> -static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
> +static inline bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
>  {
> -	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
> +	u8 val;
> +
> +	if (pv_eoi_get_user(vcpu, &val) < 0) {
> +		printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
> +			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);

pr_warn() would probably be a better choice but looking at this makes me
wonder: isn't it triggerable by the guest? I think it is when the value
written to MSR_KVM_PV_EOI_EN is bogus and this is bad: we don't even
ratelimit these messages! I think this printk() needs to be dropped.

> +		return false;
> +	}
> +
> +	val &= KVM_PV_EOI_ENABLED;
> +
> +	/*
> +	 * Clear pending bit in any case: it will be set again on vmentry.
> +	 * While this might not be ideal from performance point of view,
> +	 * this makes sure pv eoi is only enabled when we know it's safe.
> +	 */
> +	if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
>  		printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
>  			   (unsigned long long)vcpu->arch.pv_eoi.msr_val);

... and this one, probably, too.

> -		return;
> +		return false;
>  	}
>  	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
> +
> +	return !!val;
>  }
>  
>  static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
> @@ -2673,7 +2679,6 @@ void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
>  static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
>  					struct kvm_lapic *apic)
>  {
> -	bool pending;
>  	int vector;
>  	/*
>  	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
> @@ -2687,14 +2692,8 @@ static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
>  	 * 	-> host enabled PV EOI, guest executed EOI.
>  	 */
>  	BUG_ON(!pv_eoi_enabled(vcpu));
> -	pending = pv_eoi_get_pending(vcpu);
> -	/*
> -	 * Clear pending bit in any case: it will be set again on vmentry.
> -	 * While this might not be ideal from performance point of view,
> -	 * this makes sure pv eoi is only enabled when we know it's safe.
> -	 */
> -	pv_eoi_clr_pending(vcpu);
> -	if (pending)
> +
> +	if (pv_eoi_test_and_clr_pending(vcpu))
>  		return;
>  	vector = apic_set_eoi(apic);
>  	trace_kvm_pv_eoi(apic, vector);

-- 
Vitaly