Re: [PATCH] KVM/nSVM: properly map nested VMCB

From: Paolo Bonzini
Date: Wed Aug 14 2019 - 08:43:28 EST


On 10/08/19 20:35, Jiri Palecek wrote:
> Hello,
>
> On 04. 06. 19 19:27, Paolo Bonzini wrote:
>> On 04/06/19 18:09, Vitaly Kuznetsov wrote:
>>> Commit 8c5fbf1a7231 ("KVM/nSVM: Use the new mapping API for mapping
>>> guest
>>> memory") broke nested SVM completely: kvm_vcpu_map()'s second
>>> parameter is
>>> GFN so vmcb_gpa needs to be converted with gpa_to_gfn(), not the
>>> other way
>>> around.
>>>
>>> Fixes: 8c5fbf1a7231 ("KVM/nSVM: Use the new mapping API for mapping
>>> guest memory")
>>> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
>>> ---
>>> Â arch/x86/kvm/svm.c | 4 ++--
>>> Â 1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
>>> index 735b8c01895e..5beca1030c9a 100644
>>> --- a/arch/x86/kvm/svm.c
>>> +++ b/arch/x86/kvm/svm.c
>>> @@ -3293,7 +3293,7 @@ static int nested_svm_vmexit(struct vcpu_svm *svm)
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ vmcb->control.exit_int_info_err,
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ KVM_ISA_SVM);
>>>
>>> -ÂÂÂ rc = kvm_vcpu_map(&svm->vcpu, gfn_to_gpa(svm->nested.vmcb), &map);
>>> +ÂÂÂ rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
>>> ÂÂÂÂÂ if (rc) {
>>> ÂÂÂÂÂÂÂÂÂ if (rc == -EINVAL)
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ kvm_inject_gp(&svm->vcpu, 0);
>>> @@ -3583,7 +3583,7 @@ static bool nested_svm_vmrun(struct vcpu_svm *svm)
>>>
>>> ÂÂÂÂÂ vmcb_gpa = svm->vmcb->save.rax;
>>>
>>> -ÂÂÂ rc = kvm_vcpu_map(&svm->vcpu, gfn_to_gpa(vmcb_gpa), &map);
>>> +ÂÂÂ rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
>>> ÂÂÂÂÂ if (rc) {
>>> ÂÂÂÂÂÂÂÂÂ if (rc == -EINVAL)
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ kvm_inject_gp(&svm->vcpu, 0);
>>>
>> Oops. Queued, thanks.
>
> Given that this fix didn't make it to 5.2, and its straightforwardness,
> could you send it to stable for inclusion?

Done, thanks for the reminder!

Paolo