Re: [PATCH V2 3/5] KVM/VMX: Add identical ept table pointer check

[Paolo Bonzini]

On 09/07/2018 11:02, Tianyu Lan wrote:
> +static void check_ept_pointer(struct kvm_vcpu *vcpu, u64 eptp)
> +{
> +	struct kvm *kvm = vcpu->kvm;
> +	u64 tmp_eptp = INVALID_PAGE;
> +	int i;
> +
> +	if (!kvm_x86_ops->tlb_remote_flush)
> +		return;
> +
> +	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
> +	to_vmx(vcpu)->ept_pointer = eptp;
> +
> +	kvm_for_each_vcpu(i, vcpu, kvm) {
> +		if (!VALID_PAGE(tmp_eptp)) {
> +			tmp_eptp = to_vmx(vcpu)->ept_pointer;
> +		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
> +			to_kvm_vmx(kvm)->ept_pointers_match = false;
> +			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
> +			return;
> +		}
> +	}
> +
> +	to_kvm_vmx(kvm)->ept_pointers_match = true;
> +	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
> +}
> +

Is there any reason to do the check here, rather than the first time the
TLB flush is invoked?  You could:

- have a tristate (true, false, check) value for ept_pointers_match

- reset it to "check" in vmx_set_cr3

- set it to either true or false in tlb_remote_flush if it is check, and
do the hypercall if it is true.

Paolo