On Mon, Nov 26, 2018 at 11:23:31AM -0800, Jim Mattson wrote:
> According to the SDM, "interrupt-window exiting" VM-exits wake a
> logical processor from the same inactive states as would an external
> interrupt. Specifically, they wake a logical processor from the states
> entered using the HLT and MWAIT instructions.
>
> Fixes: 6dfacadd5858 ("KVM: nVMX: Add support for activity state HLT")
> Signed-off-by: Jim Mattson <jmattson@xxxxxxxxxx>
> Reviewed-by: Peter Shier <pshier@xxxxxxxxxx>
> ---
> arch/x86/kvm/vmx.c | 17 ++++++++++++++---
> 1 file changed, 14 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 358e96f0c21d..a29921b216f5 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -13498,6 +13498,16 @@ static bool nested_vmx_nmi_window_exit(struct vmcs12 *vmcs12)
> (GUEST_INTR_STATE_NMI | GUEST_INTR_STATE_MOV_SS));
> }
>
> +static bool nested_vmx_intr_window_exit(struct vmcs12 *vmcs12)
> +{
> + return (vmcs12->cpu_based_vm_exec_control &
> + CPU_BASED_VIRTUAL_INTR_PENDING) &&
> + vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI &&
Technically this also needs to weed out GUEST_ACTIVITY_SHUTDOWN as
INTR doesn't break shutdown. That being said, I wonder if it makes
sense to remove the activity state and guest_interruptibility_info
checks altogether from both the NMI and INTR helpers. The current use
case is only for guest_activity_state==HLT, which makes both checks
redundant since guest_activity_state obviously can't be two things at
once and STI/MOVSS blocking are only legal when going to ACTIVE state.
Event blocking and prioritization is so complex that it seems unlikely
that a generic kitchen-sink helper will be useful, e.g.
handle_invalid_guest_state() already has this code and more by way of
vmx_interrupt_allowed().
At that point it's probably easiest to just open code the checks, e.g.:
exec_controls = vmcs12->cpu_based_vm_exec_control;
if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
!(exec_controls & CPU_BASED_VIRTUAL_NMI_PENDING) &&
!((exec_controls & CPU_BASED_VIRTUAL_INTR_PENDING) &&
(vmcs12->guest_rflags & X86_EFLAGS_IF))) {
vmx->nested.nested_run_pending = 0;
return kvm_vcpu_halt(vcpu);
}
return 1;
> + (vmcs12->guest_rflags & X86_EFLAGS_IF) &&
> + !(vmcs12->guest_interruptibility_info &
> + (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
> +}
> +
> /*
> * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
> * for running an L2 nested guest.
> @@ -13590,12 +13600,13 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>
> /*
> * If we're entering a halted L2 vcpu and the L2 vcpu won't be
> - * awakened by event injection or by an NMI-window VM-exit,
> - * halt the vcpu.
> + * awakened by event injection or by an NMI-window VM-exit or
> + * by an interrupt-window VM-exit, halt the vcpu.
> */
> if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
> !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
> - !nested_vmx_nmi_window_exit(vmcs12)) {
> + !nested_vmx_nmi_window_exit(vmcs12) &&
> + !nested_vmx_intr_window_exit(vmcs12)) {
> vmx->nested.nested_run_pending = 0;
> return kvm_vcpu_halt(vcpu);
> }
> --
> 2.20.0.rc0.387.gc7a69e6b6c-goog
>
