Re: [PATCH 2/3] KVM: VMX: Short circuit STI; HLT while an interrupt is pending

[Marcelo Tosatti <mtosatti@redhat.com>]

On Mon, Feb 14, 2011 at 04:42:16PM +0200, Avi Kivity wrote:
> Short-circuit an STI; HLT sequence while an interrupt is pending:
> instead of halting, re-entering the guest, and exiting immediately
> on an interrupt window exit, go directly to the last step.
> 
> Saves a vmexit on workloads where interrupts are received synchronously;
> an example is a disk backed by the host page cache where there is no
> latency (from the guest's point of view) between the request and fulfilment.
> 
> Signed-off-by: Avi Kivity <avi@redhat.com>
> ---
>  arch/x86/kvm/vmx.c |    9 +++++++++
>  1 files changed, 9 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index ee1cd1a..541da0e 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -3437,6 +3437,15 @@ static int handle_interrupt_window(struct kvm_vcpu *vcpu)
>  static int handle_halt(struct kvm_vcpu *vcpu)
>  {
>  	skip_emulated_instruction(vcpu);
> +	/*
> +	 * Short-circuit an STI; HLT sequence while an interrupt is pending:
> +	 * instead of halting, re-entering the guest, and exiting immediately
> +	 * on an interrupt window exit, go directly to the last step.
> +	 */
> +	if ((to_vmx(vcpu)->cpu_based_vm_exec_control
> +	     & CPU_BASED_VIRTUAL_INTR_PENDING)
> +	    && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
> +		return handle_interrupt_window(vcpu);
>  	return kvm_emulate_halt(vcpu);
>  }

Why does the normal vcpu entry path fails to inject the interrupt? Because after halt, 
KVM_REQ_EVENT is not set?