Inject the NMI by setting V_NMI in the VMCB interrupt control. processor
will clear V_NMI to acknowledge processing has started and will keep the
V_NMI_MASK set until the processor is done with processing the NMI event.
Signed-off-by: Santosh Shukla <[email protected]>
---
v2:
- Added WARN_ON check for vnmi pending.
- use `get_vnmi_vmcb` to get correct vmcb so to inject vnmi.
arch/x86/kvm/svm/svm.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 44c1f2317b45..c73a1809a7c7 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -3375,12 +3375,20 @@ static void pre_svm_run(struct kvm_vcpu *vcpu)
static void svm_inject_nmi(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
+ struct vmcb *vmcb = NULL;
+
+ ++vcpu->stat.nmi_injections;
+ if (is_vnmi_enabled(svm)) {
+ vmcb = get_vnmi_vmcb(svm);
+ WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
+ vmcb->control.int_ctl |= V_NMI_PENDING;
+ return;
+ }
svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
vcpu->arch.hflags |= HF_NMI_MASK;
if (!sev_es_guest(vcpu->kvm))
svm_set_intercept(svm, INTERCEPT_IRET);
- ++vcpu->stat.nmi_injections;
}
static void svm_inject_irq(struct kvm_vcpu *vcpu)
--
2.25.1
On Sat, Jul 09, 2022, Santosh Shukla wrote:
> Inject the NMI by setting V_NMI in the VMCB interrupt control. processor
> will clear V_NMI to acknowledge processing has started and will keep the
> V_NMI_MASK set until the processor is done with processing the NMI event.
>
> Signed-off-by: Santosh Shukla <[email protected]>
> ---
> v2:
> - Added WARN_ON check for vnmi pending.
> - use `get_vnmi_vmcb` to get correct vmcb so to inject vnmi.
>
> arch/x86/kvm/svm/svm.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index 44c1f2317b45..c73a1809a7c7 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -3375,12 +3375,20 @@ static void pre_svm_run(struct kvm_vcpu *vcpu)
> static void svm_inject_nmi(struct kvm_vcpu *vcpu)
> {
> struct vcpu_svm *svm = to_svm(vcpu);
> + struct vmcb *vmcb = NULL;
> +
> + ++vcpu->stat.nmi_injections;
> + if (is_vnmi_enabled(svm)) {
> + vmcb = get_vnmi_vmcb(svm);
> + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
Haven't read the spec, but based on the changelog I assume the flag doesn't get
cleared until the NMI is fully delivered. That means this WARN will fire if a
VM-Exit occurs during delivery as KVM will re-inject the event, e.g. if KVM is
using shadow paging and a #PF handle by KVM occurs during delivery.
> + vmcb->control.int_ctl |= V_NMI_PENDING;
> + return;
> + }
>
> svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
> vcpu->arch.hflags |= HF_NMI_MASK;
> if (!sev_es_guest(vcpu->kvm))
> svm_set_intercept(svm, INTERCEPT_IRET);
> - ++vcpu->stat.nmi_injections;
> }
>
> static void svm_inject_irq(struct kvm_vcpu *vcpu)
> --
> 2.25.1
>
On Wed, Jul 20, 2022 at 2:41 PM Sean Christopherson <[email protected]> wrote:
>
> On Sat, Jul 09, 2022, Santosh Shukla wrote:
> > Inject the NMI by setting V_NMI in the VMCB interrupt control. processor
> > will clear V_NMI to acknowledge processing has started and will keep the
> > V_NMI_MASK set until the processor is done with processing the NMI event.
> >
> > Signed-off-by: Santosh Shukla <[email protected]>
> > ---
> > v2:
> > - Added WARN_ON check for vnmi pending.
> > - use `get_vnmi_vmcb` to get correct vmcb so to inject vnmi.
> >
> > arch/x86/kvm/svm/svm.c | 10 +++++++++-
> > 1 file changed, 9 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > index 44c1f2317b45..c73a1809a7c7 100644
> > --- a/arch/x86/kvm/svm/svm.c
> > +++ b/arch/x86/kvm/svm/svm.c
> > @@ -3375,12 +3375,20 @@ static void pre_svm_run(struct kvm_vcpu *vcpu)
> > static void svm_inject_nmi(struct kvm_vcpu *vcpu)
> > {
> > struct vcpu_svm *svm = to_svm(vcpu);
> > + struct vmcb *vmcb = NULL;
> > +
> > + ++vcpu->stat.nmi_injections;
> > + if (is_vnmi_enabled(svm)) {
> > + vmcb = get_vnmi_vmcb(svm);
> > + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
>
> Haven't read the spec, but based on the changelog I assume the flag doesn't get
> cleared until the NMI is fully delivered.
Ooh! Is there a spec to read now?!?
On Wed, Jul 20, 2022, Jim Mattson wrote:
> On Wed, Jul 20, 2022 at 2:41 PM Sean Christopherson <[email protected]> wrote:
> >
> > On Sat, Jul 09, 2022, Santosh Shukla wrote:
> > > Inject the NMI by setting V_NMI in the VMCB interrupt control. processor
> > > will clear V_NMI to acknowledge processing has started and will keep the
> > > V_NMI_MASK set until the processor is done with processing the NMI event.
> > >
> > > Signed-off-by: Santosh Shukla <[email protected]>
> > > ---
> > > v2:
> > > - Added WARN_ON check for vnmi pending.
> > > - use `get_vnmi_vmcb` to get correct vmcb so to inject vnmi.
> > >
> > > arch/x86/kvm/svm/svm.c | 10 +++++++++-
> > > 1 file changed, 9 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > > index 44c1f2317b45..c73a1809a7c7 100644
> > > --- a/arch/x86/kvm/svm/svm.c
> > > +++ b/arch/x86/kvm/svm/svm.c
> > > @@ -3375,12 +3375,20 @@ static void pre_svm_run(struct kvm_vcpu *vcpu)
> > > static void svm_inject_nmi(struct kvm_vcpu *vcpu)
> > > {
> > > struct vcpu_svm *svm = to_svm(vcpu);
> > > + struct vmcb *vmcb = NULL;
> > > +
> > > + ++vcpu->stat.nmi_injections;
> > > + if (is_vnmi_enabled(svm)) {
> > > + vmcb = get_vnmi_vmcb(svm);
> > > + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
> >
> > Haven't read the spec, but based on the changelog I assume the flag doesn't get
> > cleared until the NMI is fully delivered.
>
> Ooh! Is there a spec to read now?!?
Apparently not. I just assumed there's a spec, but I haven't found one.
Hi Sean,
On 7/21/2022 3:11 AM, Sean Christopherson wrote:
> On Sat, Jul 09, 2022, Santosh Shukla wrote:
>> Inject the NMI by setting V_NMI in the VMCB interrupt control. processor
>> will clear V_NMI to acknowledge processing has started and will keep the
>> V_NMI_MASK set until the processor is done with processing the NMI event.
>>
>> Signed-off-by: Santosh Shukla <[email protected]>
>> ---
>> v2:
>> - Added WARN_ON check for vnmi pending.
>> - use `get_vnmi_vmcb` to get correct vmcb so to inject vnmi.
>>
>> arch/x86/kvm/svm/svm.c | 10 +++++++++-
>> 1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
>> index 44c1f2317b45..c73a1809a7c7 100644
>> --- a/arch/x86/kvm/svm/svm.c
>> +++ b/arch/x86/kvm/svm/svm.c
>> @@ -3375,12 +3375,20 @@ static void pre_svm_run(struct kvm_vcpu *vcpu)
>> static void svm_inject_nmi(struct kvm_vcpu *vcpu)
>> {
>> struct vcpu_svm *svm = to_svm(vcpu);
>> + struct vmcb *vmcb = NULL;
>> +
>> + ++vcpu->stat.nmi_injections;
>> + if (is_vnmi_enabled(svm)) {
>> + vmcb = get_vnmi_vmcb(svm);
>> + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
>
> Haven't read the spec, but based on the changelog I assume the flag doesn't get
> cleared until the NMI is fully delivered. That means this WARN will fire if a
> VM-Exit occurs during delivery as KVM will re-inject the event, e.g. if KVM is
> using shadow paging and a #PF handle by KVM occurs during delivery.
>
Right,.
For the above scenario i.e.. if VMEXIT happens during delivery of virtual NMI
then EXITINTINFO will be set accordingly and V_NMI_MASK is saved as 0.
hypervisor will re-inject event in next VMRUN.
I just wanted to track above scenario,. I will replace it with pr_debug().
Thanks,
Santosh
>> + vmcb->control.int_ctl |= V_NMI_PENDING;
>> + return;
>> + }
>>
>> svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
>> vcpu->arch.hflags |= HF_NMI_MASK;
>> if (!sev_es_guest(vcpu->kvm))
>> svm_set_intercept(svm, INTERCEPT_IRET);
>> - ++vcpu->stat.nmi_injections;
>> }
>>
>> static void svm_inject_irq(struct kvm_vcpu *vcpu)
>> --
>> 2.25.1
>>
On Fri, Jul 29, 2022, Shukla, Santosh wrote:
> Hi Sean,
>
> On 7/21/2022 3:11 AM, Sean Christopherson wrote:
> > On Sat, Jul 09, 2022, Santosh Shukla wrote:
> >> + ++vcpu->stat.nmi_injections;
> >> + if (is_vnmi_enabled(svm)) {
> >> + vmcb = get_vnmi_vmcb(svm);
> >> + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
> >
> > Haven't read the spec, but based on the changelog I assume the flag doesn't get
> > cleared until the NMI is fully delivered. That means this WARN will fire if a
> > VM-Exit occurs during delivery as KVM will re-inject the event, e.g. if KVM is
> > using shadow paging and a #PF handle by KVM occurs during delivery.
> >
>
> Right,.
>
>
> For the above scenario i.e.. if VMEXIT happens during delivery of virtual NMI
> then EXITINTINFO will be set accordingly and V_NMI_MASK is saved as 0.
> hypervisor will re-inject event in next VMRUN.
>
> I just wanted to track above scenario,. I will replace it with pr_debug().
No, this is normal (albeit uncommon) behavior, don't print anything. Even if it
weren't normal behavior, pr_debug() is usually the wrong choice in KVM.
On 7/29/2022 7:23 PM, Sean Christopherson wrote:
> On Fri, Jul 29, 2022, Shukla, Santosh wrote:
>> Hi Sean,
>>
>> On 7/21/2022 3:11 AM, Sean Christopherson wrote:
>>> On Sat, Jul 09, 2022, Santosh Shukla wrote:
>>>> + ++vcpu->stat.nmi_injections;
>>>> + if (is_vnmi_enabled(svm)) {
>>>> + vmcb = get_vnmi_vmcb(svm);
>>>> + WARN_ON(vmcb->control.int_ctl & V_NMI_PENDING);
>>>
>>> Haven't read the spec, but based on the changelog I assume the flag doesn't get
>>> cleared until the NMI is fully delivered. That means this WARN will fire if a
>>> VM-Exit occurs during delivery as KVM will re-inject the event, e.g. if KVM is
>>> using shadow paging and a #PF handle by KVM occurs during delivery.
>>>
>>
>> Right,.
>>
>>
>> For the above scenario i.e.. if VMEXIT happens during delivery of virtual NMI
>> then EXITINTINFO will be set accordingly and V_NMI_MASK is saved as 0.
>> hypervisor will re-inject event in next VMRUN.
>>
>> I just wanted to track above scenario,. I will replace it with pr_debug().
>
> No, this is normal (albeit uncommon) behavior, don't print anything. Even if it
Ok.
Thanks,
Santosh
> weren't normal behavior, pr_debug() is usually the wrong choice in KVM.