Re: [PATCH] KVM: SVM: Fix potential wrong physical id in avic_handle_ldr_update

Re: [PATCH] KVM: SVM: Fix potential wrong physical id in avic_handle_ldr_update

[linmiaohe]

Vitaly Kuznetsov <vkuznets@redhat.com> writes:

>> Guest physical APIC ID may not equal to vcpu->vcpu_id in some case.
>> We may set the wrong physical id in avic_handle_ldr_update as we 
>> always use vcpu->vcpu_id.

Hi, Vitaly, thanks for your reply.
Do you think there may be a wrong physical id in avic_handle_ldr_update too ?

>>
>> @@ -4591,6 +4591,8 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
>>  	int ret = 0;
>>  	struct vcpu_svm *svm = to_svm(vcpu);
>>  	u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
>> +	u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
>> +	u32 id = (apic_id_reg >> 24) & 0xff;
>
>If we reach here than we're guaranteed to be in xAPIC mode, right? Could you maybe export and use kvm_xapic_id() here then (and in
>avic_handle_apic_id_update() too)?
>

I think we're guaranteed to be in xAPIC mode when we reach here. I would have a try to export
and use use kvm_xapic_id here and in avic_handle_apic_id_update too.
Thanks for your suggestion.

Have a nice day.
Best wishes.

Re: [PATCH] KVM: SVM: Fix potential wrong physical id in avic_handle_ldr_update

[Vitaly Kuznetsov]

linmiaohe <linmiaohe@huawei.com> writes:

> Vitaly Kuznetsov <vkuznets@redhat.com> writes:
>
>>> Guest physical APIC ID may not equal to vcpu->vcpu_id in some case.
>>> We may set the wrong physical id in avic_handle_ldr_update as we 
>>> always use vcpu->vcpu_id.
>
> Hi, Vitaly, thanks for your reply.
> Do you think there may be a wrong physical id in
> avic_handle_ldr_update too ?

Honestly I'm not sure, however, as we need to put physical id to LDR
we'd rather get it from LAPIC then assume that it's == vcpu_id so I
think your patch makes sense even if it fixes a theoretical issue.

But I may be missing something important about AVIC.

>
>>>
>>> @@ -4591,6 +4591,8 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
>>>  	int ret = 0;
>>>  	struct vcpu_svm *svm = to_svm(vcpu);
>>>  	u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
>>> +	u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
>>> +	u32 id = (apic_id_reg >> 24) & 0xff;
>>
>>If we reach here than we're guaranteed to be in xAPIC mode, right? Could you maybe export and use kvm_xapic_id() here then (and in
>>avic_handle_apic_id_update() too)?
>>
>
> I think we're guaranteed to be in xAPIC mode when we reach here. I would have a try to export
> and use use kvm_xapic_id here and in avic_handle_apic_id_update too.
> Thanks for your suggestion.
>
> Have a nice day.
> Best wishes.

-- 
Vitaly

[PATCH] KVM: SVM: Fix potential wrong physical id in avic_handle_ldr_update

[Miaohe Lin]

Guest physical APIC ID may not equal to vcpu->vcpu_id in some case.
We may set the wrong physical id in avic_handle_ldr_update as we
always use vcpu->vcpu_id.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
 arch/x86/kvm/svm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index f8ecb6d..67cb5ba 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -4591,6 +4591,8 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
 	int ret = 0;
 	struct vcpu_svm *svm = to_svm(vcpu);
 	u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
+	u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
+	u32 id = (apic_id_reg >> 24) & 0xff;
 
 	if (ldr == svm->ldr_reg)
 		return 0;
@@ -4598,7 +4600,7 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
 	avic_invalidate_logical_id_entry(vcpu);
 
 	if (ldr)
-		ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr);
+		ret = avic_ldr_write(vcpu, id, ldr);
 
 	if (!ret)
 		svm->ldr_reg = ldr;
-- 
1.8.3.1

Re: [PATCH] KVM: SVM: Fix potential wrong physical id in avic_handle_ldr_update

[Vitaly Kuznetsov]

Miaohe Lin <linmiaohe@huawei.com> writes:

> Guest physical APIC ID may not equal to vcpu->vcpu_id in some case.
> We may set the wrong physical id in avic_handle_ldr_update as we
> always use vcpu->vcpu_id.
>
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>  arch/x86/kvm/svm.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> index f8ecb6d..67cb5ba 100644
> --- a/arch/x86/kvm/svm.c
> +++ b/arch/x86/kvm/svm.c
> @@ -4591,6 +4591,8 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
>  	int ret = 0;
>  	struct vcpu_svm *svm = to_svm(vcpu);
>  	u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
> +	u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
> +	u32 id = (apic_id_reg >> 24) & 0xff;

If we reach here than we're guaranteed to be in xAPIC mode, right? Could
you maybe export and use kvm_xapic_id() here then (and in
avic_handle_apic_id_update() too)?

>  
>  	if (ldr == svm->ldr_reg)
>  		return 0;
> @@ -4598,7 +4600,7 @@ static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
>  	avic_invalidate_logical_id_entry(vcpu);
>  
>  	if (ldr)
> -		ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr);
> +		ret = avic_ldr_write(vcpu, id, ldr);
>  
>  	if (!ret)
>  		svm->ldr_reg = ldr;

-- 
Vitaly