Re: [PATCH v3 1/4] KVM: nVMX: Rework interception of IRQs and NMIs

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 3045 bytes --]

On 2014-03-09 09:03, Paolo Bonzini wrote:
> Il 09/03/2014 08:33, Paolo Bonzini ha scritto:
>> Il 08/03/2014 10:21, Jan Kiszka ha scritto:
>>> On 2014-03-07 20:48, Paolo Bonzini wrote:
>>>> Il 07/03/2014 20:03, Jan Kiszka ha scritto:
>>>>> @@ -4631,22 +4631,8 @@ static void vmx_set_nmi_mask(struct kvm_vcpu
>>>>> *vcpu, bool masked)
>>>>>
>>>>>  static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
>>>>>  {
>>>>> -    if (is_guest_mode(vcpu)) {
>>>>> -        if (to_vmx(vcpu)->nested.nested_run_pending)
>>>>> -            return 0;
>>>>> -        if (nested_exit_on_nmi(vcpu)) {
>>>>> -            nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
>>>>> -                      NMI_VECTOR | INTR_TYPE_NMI_INTR |
>>>>> -                      INTR_INFO_VALID_MASK, 0);
>>>>> -            /*
>>>>> -             * The NMI-triggered VM exit counts as injection:
>>>>> -             * clear this one and block further NMIs.
>>>>> -             */
>>>>> -            vcpu->arch.nmi_pending = 0;
>>>>> -            vmx_set_nmi_mask(vcpu, true);
>>>>> -            return 0;
>>>>> -        }
>>>>> -    }
>>>>> +    if (to_vmx(vcpu)->nested.nested_run_pending)
>>>>> +        return 0;
>>>>>
>>>>>      if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
>>>>>          return 0;
>>>>> @@ -4658,19 +4644,8 @@ static int vmx_nmi_allowed(struct kvm_vcpu
>>>>> *vcpu)
>>>>>
>>>>>  static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
>>>>>  {
>>>>> -    if (is_guest_mode(vcpu)) {
>>>>> -        if (to_vmx(vcpu)->nested.nested_run_pending)
>>>>> -            return 0;
>>>>> -        if (nested_exit_on_intr(vcpu)) {
>>>>> -            nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT,
>>>>> -                      0, 0);
>>>>> -            /*
>>>>> -             * fall through to normal code, but now in L1, not L2
>>>>> -             */
>>>>> -        }
>>>>> -    }
>>>>> -
>>>>> -    return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
>>>>> +    return (!to_vmx(vcpu)->nested.nested_run_pending &&
>>>>> +        vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
>>>>>          !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
>>>>>              (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
>>>>
>>>> The checks on nested_run_pending are not needed anymore and can be
>>>> replaced with a WARN_ON.  Otherwise,
>>>
>>> Nope, that won't be correct: If we have a pending interrupt that L1 does
>>> not intercept, we still trigger this condition legally.
>>
>> Right, this is the case of !nested_exit_on_intr(vcpu) or
>> !nested_exit_on_nmi(vcpu).
>>
>> Why don't we need to request an immediate exit in that case, in order to
>> inject the interrupt into L2?
> 
> Nevermind, this makes no sense.  I was confusing *_allowed with
> enable_*_window.

This code is mind-blowing and probably still not perfect. I wouldn't be
surprised if we are going to find bugs there until we retire. ;)

> 
> Applying v3 to kvm/queue, thanks!
> 

Great, thank you!
Jan



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]