On 24/12/2017 17:13, Liran Alon wrote:
> +static bool vmx_cpu_has_nested_posted_interrupt(struct kvm_vcpu *vcpu)
> +{
> + struct vcpu_vmx *vmx = to_vmx(vcpu);
> +
> + return (vcpu->arch.apicv_active &&
> + is_guest_mode(vcpu) &&
> + vmx->nested.pi_pending &&
> + vmx->nested.pi_desc &&
> + pi_test_on(vmx->nested.pi_desc));
> +}
> +
> /*
> * Set up the vmcs's constant host-state fields, i.e., host-state fields that
> * will not change in the lifetime of the guest.
> @@ -12142,6 +12153,8 @@ static int enable_smi_window(struct kvm_vcpu *vcpu)
> .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
> .complete_nested_posted_interrupt =
> vmx_complete_nested_posted_interrupt,
> + .cpu_has_nested_posted_interrupt =
> + vmx_cpu_has_nested_posted_interrupt,
>
> .set_tss_addr = vmx_set_tss_addr,
> .get_tdp_level = get_ept_level,
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index fa088951afc9..a840f2c9bd66 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -8542,7 +8542,8 @@ static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
> return true;
>
> if (kvm_arch_interrupt_allowed(vcpu) &&
> - kvm_cpu_has_interrupt(vcpu))
> + (kvm_cpu_has_interrupt(vcpu) ||
> + kvm_x86_ops->cpu_has_nested_posted_interrupt(vcpu)))
> return true;
kvm_cpu_has_interrupt ultimately calls apic_has_interrupt_for_ppr, which
calls kvm_x86_ops->sync_pir_to_irr.
You already have
+ if (is_guest_mode(vcpu))
+ kvm_x86_ops->complete_nested_posted_interrupt(vcpu);
earlier in the series right after a call to kvm_x86_ops->sync_pir_to_irr.
So I wonder if:
1) kvm_x86_ops->complete_nested_posted_interrupt would do the job here as
well, removing the need for the new kvm_x86_ops member;
2) The call to kvm_x86_ops->complete_nested_posted_interrupt actually
applies to all callers of sync_pir_to_irr, which would remove the need for
that other kvm_x86_ops member.
I'm leaning towards applying patches 1-4, what do you think?
Paolo
