Re: [PATCHv2 4/7] KVM: SVM: Report NMI not allowed when Guest busy handling VNMI

From: "Shukla, Santosh" <santosh.shukla@amd.com>
To: Sean Christopherson <seanjc@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
	Maxim Levitsky <mlevitsk@redhat.com>,
	Tom Lendacky <thomas.lendacky@amd.com>,
	kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCHv2 4/7] KVM: SVM: Report NMI not allowed when Guest busy handling VNMI
Date: Thu, 4 Aug 2022 15:21:08 +0530	[thread overview]
Message-ID: <1782cdbb-8274-8c3d-fa98-29147f1e5d1e@amd.com> (raw)
In-Reply-To: <YuPxkMW2aZxrw57n@google.com>

On 7/29/2022 8:11 PM, Sean Christopherson wrote:
> On Fri, Jul 29, 2022, Shukla, Santosh wrote:
>> Hello Sean,
>>
>> On 7/21/2022 3:24 AM, Sean Christopherson wrote:
>>> On Sat, Jul 09, 2022, Santosh Shukla wrote:
> 
> ...
> 
>>>> @@ -3609,6 +3612,9 @@ static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
>>>>  {
>>>>  	struct vcpu_svm *svm = to_svm(vcpu);
>>>>  
>>>> +	if (is_vnmi_enabled(svm))
>>>> +		return;
>>>
>>> Ugh, is there really no way to trigger an exit when NMIs become unmasked?  Because
>>> if there isn't, this is broken for KVM.
>>>
>>
>> Yes. there is.
>>
>> NMI_INTERCEPT will trigger VMEXIT when second NMI arrives while guest is busy handling first NMI.
> 
> But NMI_INTERCEPT only applies to "real" NMIs.  The scenario laid out below is where
> KVM wants to inject a virtual NMI without an associated hardware/real NMI, e.g. if
> multiple vCPUs send NMI IPIs to the target.
> 

Yes, HW supports above case. KVM can inject the pending virtual NMI while guest busy handling
current (virtual) NMI such that after guest finished handling.. HW will take
the pending virtual NMI on the IRET instruction (w/o vmexit).

Thanks,
Santosh

>> And in that scenario, Guest will exit with V_NMI_MASK set to 1, KVM can inject pending(Second)
>> NMI(V_NMI=1). Guest will resume handling the first NMI, then HW will
>> clear the V_NMI_MASK and later HW will take the pending V_NMI in side the guest. 
>>
>> I'll handle above case in v3.
>>
>> Thanks,
>> Santosh
>>
>>> On bare metal, if two NMIs arrive "simultaneously", so long as NMIs aren't blocked,
>>> the first NMI will be delivered and the second will be pended, i.e. software will
>>> see both NMIs.  And if that doesn't hold true, the window for a true collision is
>>> really, really tiny.
>>>
>>> But in KVM, because a vCPU may not be run a long duration, that window becomes
>>> very large.  To not drop NMIs and more faithfully emulate hardware, KVM allows two
>>> NMIs to be _pending_.  And when that happens, KVM needs to trigger an exit when
>>> NMIs become unmasked _after_ the first NMI is injected.
>>>
>>>> +
>>>>  	if ((vcpu->arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) == HF_NMI_MASK)
>>>>  		return; /* IRET will cause a vm exit */
>>>>  
>>>> -- 
>>>> 2.25.1
>>>>