[PATCH v3] KVM: x86: Fix the intel_pt PMI handling wrongly considered from guest

[PATCH v3] KVM: x86: Fix the intel_pt PMI handling wrongly considered from guest

[Yanfei Xu]

When kernel handles the vm-exit caused by external interrupts and NMI,
it always set a type of kvm_intr_type to handling_intr_from_guest to
tell if it's dealing an IRQ or NMI. For the PMI scenario, it could be
IRQ or NMI.
However the intel_pt PMI certainly is a NMI PMI, hence using
kvm_handling_nmi_from_guest() to distinguish if the intel_pt PMI comes
from guest is more appropriate. This modification can avoid the host
wrongly considered the intel_pt PMI comes from a guest once the host
intel_pt PMI breaks the handling of vm-exit of external interrupts.

Fixes: db215756ae59 ("KVM: x86: More precisely identify NMI from guest when handling PMI")
Signed-off-by: Yanfei Xu <yanfei.xu@intel.com>
---
v1->v2:
1.Fix vmx_handle_intel_pt_intr() directly instead of changing the generic function.
2.Tune the commit message.

v2->v3:
Add the NULL pointer check of variable "vcpu".

 arch/x86/kvm/vmx/vmx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 610355b9ccce..982df9c000d3 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -7856,7 +7856,7 @@ static unsigned int vmx_handle_intel_pt_intr(void)
 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
 
 	/* '0' on failure so that the !PT case can use a RET0 static call. */
-	if (!kvm_arch_pmi_in_guest(vcpu))
+	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))
 		return 0;
 
 	kvm_make_request(KVM_REQ_PMI, vcpu);
-- 
2.32.0

Re: [PATCH v3] KVM: x86: Fix the intel_pt PMI handling wrongly considered from guest

[Sean Christopherson]

On Mon, May 23, 2022, Yanfei Xu wrote:
> When kernel handles the vm-exit caused by external interrupts and NMI,
> it always set a type of kvm_intr_type to handling_intr_from_guest to
> tell if it's dealing an IRQ or NMI. For the PMI scenario, it could be
> IRQ or NMI.
> However the intel_pt PMI certainly is a NMI PMI, hence using

It'd be helpful for future readers to explain why it's guaranteed to an NMI.  E.g.

However, intel_pt PMIs are only generated for HARDWARE perf events, and
HARDWARE events are always configured to generate NMIs.  Use
kvm_handling_nmi_from_guest() to precisely identify if the intel_pt PMI
came from the guest to avoid false positives if an intel_pt PMI/NMI
arrives while the host is handling an unrelated IRQ VM-Exit.

> kvm_handling_nmi_from_guest() to distinguish if the intel_pt PMI comes
> from guest is more appropriate. This modification can avoid the host
> wrongly considered the intel_pt PMI comes from a guest once the host
> intel_pt PMI breaks the handling of vm-exit of external interrupts.
> 
> Fixes: db215756ae59 ("KVM: x86: More precisely identify NMI from guest when handling PMI")
> Signed-off-by: Yanfei Xu <yanfei.xu@intel.com>
> ---
> v1->v2:
> 1.Fix vmx_handle_intel_pt_intr() directly instead of changing the generic function.
> 2.Tune the commit message.
> 
> v2->v3:
> Add the NULL pointer check of variable "vcpu".
> 
>  arch/x86/kvm/vmx/vmx.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 610355b9ccce..982df9c000d3 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -7856,7 +7856,7 @@ static unsigned int vmx_handle_intel_pt_intr(void)
>  	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
>  
>  	/* '0' on failure so that the !PT case can use a RET0 static call. */
> -	if (!kvm_arch_pmi_in_guest(vcpu))
> +	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))

Alternatively,

	if (!kvm_arch_pmi_in_guest(vcpu) || !kvm_handling_nmi_from_guest(vcpu))

The generated code is the same since the compiler is smart enough to elide the
handling_intr_from_guest check from kvm_arch_pmi_in_guest.

I'm not actually sure that's better than the !vcpu check though, e.g. it hides the
not-NULL aspect of the check.

Either way, with a tweaked changelog,

Reviewed-by: Sean Christopherson <seanjc@google.com>

Re: [PATCH v3] KVM: x86: Fix the intel_pt PMI handling wrongly considered from guest

[Paolo Bonzini]

Queued, thanks.

Paolo

Re: [PATCH v3] KVM: x86: Fix the intel_pt PMI handling wrongly considered from guest

[Yanfei Xu]

On 2022/5/24 00:43, Sean Christopherson wrote:
> On Mon, May 23, 2022, Yanfei Xu wrote:
>> When kernel handles the vm-exit caused by external interrupts and NMI,
>> it always set a type of kvm_intr_type to handling_intr_from_guest to
>> tell if it's dealing an IRQ or NMI. For the PMI scenario, it could be
>> IRQ or NMI.
>> However the intel_pt PMI certainly is a NMI PMI, hence using
> It'd be helpful for future readers to explain why it's guaranteed to an NMI.  E.g.
>
> However, intel_pt PMIs are only generated for HARDWARE perf events, and
> HARDWARE events are always configured to generate NMIs.  Use
> kvm_handling_nmi_from_guest() to precisely identify if the intel_pt PMI
> came from the guest to avoid false positives if an intel_pt PMI/NMI
> arrives while the host is handling an unrelated IRQ VM-Exit.

It's much better!

>> kvm_handling_nmi_from_guest() to distinguish if the intel_pt PMI comes
>> from guest is more appropriate. This modification can avoid the host
>> wrongly considered the intel_pt PMI comes from a guest once the host
>> intel_pt PMI breaks the handling of vm-exit of external interrupts.
>>
>> Fixes: db215756ae59 ("KVM: x86: More precisely identify NMI from guest when handling PMI")
>> Signed-off-by: Yanfei Xu <yanfei.xu@intel.com>
>> ---
>> v1->v2:
>> 1.Fix vmx_handle_intel_pt_intr() directly instead of changing the generic function.
>> 2.Tune the commit message.
>>
>> v2->v3:
>> Add the NULL pointer check of variable "vcpu".
>>
>>   arch/x86/kvm/vmx/vmx.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>> index 610355b9ccce..982df9c000d3 100644
>> --- a/arch/x86/kvm/vmx/vmx.c
>> +++ b/arch/x86/kvm/vmx/vmx.c
>> @@ -7856,7 +7856,7 @@ static unsigned int vmx_handle_intel_pt_intr(void)
>>   	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
>>   
>>   	/* '0' on failure so that the !PT case can use a RET0 static call. */
>> -	if (!kvm_arch_pmi_in_guest(vcpu))
>> +	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))
> Alternatively,
>
> 	if (!kvm_arch_pmi_in_guest(vcpu) || !kvm_handling_nmi_from_guest(vcpu))
>
> The generated code is the same since the compiler is smart enough to elide the
> handling_intr_from_guest check from kvm_arch_pmi_in_guest.
>
> I'm not actually sure that's better than the !vcpu check though, e.g. it hides the
> not-NULL aspect of the check.
>
> Either way, with a tweaked changelog,
>
> Reviewed-by: Sean Christopherson <seanjc@google.com>

Thanks Sean.

Regards,
Yanfei