Re: [PATCH v3 4/5] KVM: nVMX: Fix conditions for interrupt injection

[Gleb Natapov <gleb@redhat.com>]

On Sun, Mar 24, 2013 at 07:44:47PM +0100, Jan Kiszka wrote:
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> If we are in guest mode, L0 can only inject events into L2 if L1 has
> nothing pending. Otherwise, L0 would overwrite L1's events and they
> would get lost. But even if no injection of L1 is pending, we do not
> want L0 to interrupt unnecessarily an on going vmentry with all its side
> effects on the vmcs. Therefore, injection shall be disallowed during
> L1->L2 transitions. This check is conceptually independent of
> nested_exit_on_intr.
> 
> If L1 traps external interrupts, then we also need to look at L1's
> idt_vectoring_info_field. If it is empty, we can kick the guest from L2
> to L1, just like the previous code worked.
> 
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>  arch/x86/kvm/vmx.c |   28 ++++++++++++++++++++--------
>  1 files changed, 20 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index d1bc834..30aa198 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -4325,16 +4325,28 @@ static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
>  
>  static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
>  {
> -	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
> +	if (is_guest_mode(vcpu)) {
>  		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
> -		if (to_vmx(vcpu)->nested.nested_run_pending ||
> -		    (vmcs12->idt_vectoring_info_field &
> -		     VECTORING_INFO_VALID_MASK))
> +
> +		if (to_vmx(vcpu)->nested.nested_run_pending)
>  			return 0;
> -		nested_vmx_vmexit(vcpu);
> -		vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
> -		vmcs12->vm_exit_intr_info = 0;
> -		/* fall through to normal code, but now in L1, not L2 */
> +		if (nested_exit_on_intr(vcpu)) {
> +			/*
> +			 * Check if the idt_vectoring_info_field is free. We
> +			 * cannot raise EXIT_REASON_EXTERNAL_INTERRUPT if it
> +			 * isn't.
> +			 */
> +			if (vmcs12->idt_vectoring_info_field &
> +			    VECTORING_INFO_VALID_MASK)
> +				return 0;
After patch 2 I do not see how this can be true. Now this case is
handled by the common code: since event queue is not empty the code will not
get here.

> +			nested_vmx_vmexit(vcpu);
> +			vmcs12->vm_exit_reason =
> +				EXIT_REASON_EXTERNAL_INTERRUPT;
> +			vmcs12->vm_exit_intr_info = 0;
> +			/*
> +			 * fall through to normal code, but now in L1, not L2
> +			 */
> +		}
>  	}
>  
>  	return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
> -- 
> 1.7.3.4

--
			Gleb.