[PATCH] KVM: nVMX: consult PFER_MASK and PFER_MATCH before nested vmexit if inject #PF

[PATCH] KVM: nVMX: consult PFER_MASK and PFER_MATCH before nested vmexit if inject #PF

[Wanpeng Li]

From: Wanpeng Li <wanpeng.li@hotmail.com>

When generating #PF VM-exit, check equality:
(PFEC & PFEC_MASK) == PFEC_MATCH
If there is equality, the 14 bit of exception bitmap is used to take decision
about generating #PF VM-exit. If there is inequality, inverted 14 bit is used.

Reported-by: Jim Mattson <jmattson@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
---
 arch/x86/kvm/vmx.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 29fd8af..8a213f2 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2429,6 +2429,8 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 	vmx_set_interrupt_shadow(vcpu, 0);
 }
 
+static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12, u16 error_code);
+
 /*
  * KVM wants to inject page-faults which it got to the guest. This function
  * checks whether in a nested guest, we need to inject them to L1 or L2.
@@ -2442,6 +2444,10 @@ static int nested_vmx_check_exception(struct kvm_vcpu *vcpu)
 		(nr == PF_VECTOR && vcpu->arch.exception.nested_apf)))
 		return 0;
 
+	if (nr == PF_VECTOR && !vcpu->arch.exception.nested_apf &&
+		!nested_vmx_is_page_fault_vmexit(vmcs12, vcpu->arch.exception.error_code))
+		return 0;
+
 	if (vcpu->arch.exception.nested_apf) {
 		vmcs_write32(VM_EXIT_INTR_ERROR_CODE, vcpu->arch.exception.error_code);
 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
-- 
2.7.4

Re: [PATCH] KVM: nVMX: consult PFER_MASK and PFER_MATCH before nested vmexit if inject #PF

[Paolo Bonzini]

On 23/07/2017 03:05, Wanpeng Li wrote:
> From: Wanpeng Li <wanpeng.li@hotmail.com>
> 
> When generating #PF VM-exit, check equality:
> (PFEC & PFEC_MASK) == PFEC_MATCH
> If there is equality, the 14 bit of exception bitmap is used to take decision
> about generating #PF VM-exit. If there is inequality, inverted 14 bit is used.
> 
> Reported-by: Jim Mattson <jmattson@google.com>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Radim Krčmář <rkrcmar@redhat.com>
> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
> ---
>  arch/x86/kvm/vmx.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 29fd8af..8a213f2 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2429,6 +2429,8 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
>  	vmx_set_interrupt_shadow(vcpu, 0);
>  }
>  
> +static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12, u16 error_code);
> +
>  /*
>   * KVM wants to inject page-faults which it got to the guest. This function
>   * checks whether in a nested guest, we need to inject them to L1 or L2.
> @@ -2442,6 +2444,10 @@ static int nested_vmx_check_exception(struct kvm_vcpu *vcpu)
>  		(nr == PF_VECTOR && vcpu->arch.exception.nested_apf)))
>  		return 0;
>  
> +	if (nr == PF_VECTOR && !vcpu->arch.exception.nested_apf &&
> +		!nested_vmx_is_page_fault_vmexit(vmcs12, vcpu->arch.exception.error_code))
> +		return 0;

This is wrong, the check must replace "(vmcs12->exception_bitmap & (1u
<< nr)" if nr == PF_VECTOR.

Paolo

>  	if (vcpu->arch.exception.nested_apf) {
>  		vmcs_write32(VM_EXIT_INTR_ERROR_CODE, vcpu->arch.exception.error_code);
>  		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
> 

Re: [PATCH] KVM: nVMX: consult PFER_MASK and PFER_MATCH before nested vmexit if inject #PF

[Wanpeng Li]

2017-07-24 21:36 GMT+08:00 Paolo Bonzini <pbonzini@redhat.com>:
> On 23/07/2017 03:05, Wanpeng Li wrote:
>> From: Wanpeng Li <wanpeng.li@hotmail.com>
>>
>> When generating #PF VM-exit, check equality:
>> (PFEC & PFEC_MASK) == PFEC_MATCH
>> If there is equality, the 14 bit of exception bitmap is used to take decision
>> about generating #PF VM-exit. If there is inequality, inverted 14 bit is used.
>>
>> Reported-by: Jim Mattson <jmattson@google.com>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Radim Krčmář <rkrcmar@redhat.com>
>> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
>> ---
>>  arch/x86/kvm/vmx.c | 6 ++++++
>>  1 file changed, 6 insertions(+)
>>
>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> index 29fd8af..8a213f2 100644
>> --- a/arch/x86/kvm/vmx.c
>> +++ b/arch/x86/kvm/vmx.c
>> @@ -2429,6 +2429,8 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
>>       vmx_set_interrupt_shadow(vcpu, 0);
>>  }
>>
>> +static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12, u16 error_code);
>> +
>>  /*
>>   * KVM wants to inject page-faults which it got to the guest. This function
>>   * checks whether in a nested guest, we need to inject them to L1 or L2.
>> @@ -2442,6 +2444,10 @@ static int nested_vmx_check_exception(struct kvm_vcpu *vcpu)
>>               (nr == PF_VECTOR && vcpu->arch.exception.nested_apf)))
>>               return 0;
>>
>> +     if (nr == PF_VECTOR && !vcpu->arch.exception.nested_apf &&
>> +             !nested_vmx_is_page_fault_vmexit(vmcs12, vcpu->arch.exception.error_code))
>> +             return 0;
>
> This is wrong, the check must replace "(vmcs12->exception_bitmap & (1u
> << nr)" if nr == PF_VECTOR.

How about something like this:

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 29fd8af..52afd23 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2439,7 +2439,8 @@ static int nested_vmx_check_exception(struct
kvm_vcpu *vcpu)
     unsigned int nr = vcpu->arch.exception.nr;

     if (!((vmcs12->exception_bitmap & (1u << nr)) ||
-        (nr == PF_VECTOR && vcpu->arch.exception.nested_apf)))
+        (nr == PF_VECTOR && (vcpu->arch.exception.nested_apf ||
+            nested_vmx_is_page_fault_vmexit(vmcs12,
vcpu->arch.exception.error_code)))))
         return 0;

     if (vcpu->arch.exception.nested_apf) {

Regards,
Wanpeng Li

Re: [PATCH] KVM: nVMX: consult PFER_MASK and PFER_MATCH before nested vmexit if inject #PF

[Paolo Bonzini]

On 24/07/2017 15:57, Wanpeng Li wrote:
> 
>      if (!((vmcs12->exception_bitmap & (1u << nr)) ||
> -        (nr == PF_VECTOR && vcpu->arch.exception.nested_apf)))
> +        (nr == PF_VECTOR && (vcpu->arch.exception.nested_apf ||
> +            nested_vmx_is_page_fault_vmexit(vmcs12,
> vcpu->arch.exception.error_code)))))
>          return 0;

No, it would be more like

	if (nr == PF_VECTOR
	    ? (vcpu->arch.exception.nested_apf ||
               nested_vmx_is_page_fault_vmexit(vmcs12, vcpu->arch.exception.error_code))
            : (vmcs12->exception_bitmap & (1u << nr)))

but I've sent an RFC patch that also fixes the exit qualification.

Paolo