Re: [PATCH v5 09/12] x86/kvm/nVMX: allow bare VMXON state migration

From: Jim Mattson
Date: Fri Sep 14 2018 - 12:51:32 EST


On Fri, Sep 14, 2018 at 12:49 AM, Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
> Jim Mattson <jmattson@xxxxxxxxxx> writes:
>
>> On Thu, Sep 13, 2018 at 10:05 AM, Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
>>> It is perfectly valid for a guest to do VMXON and not do VMPTRLD. This
>>> state needs to be preserved on migration.
>>>
>>> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
>>> ---
>>> arch/x86/kvm/vmx.c | 15 ++++++++-------
>>> 1 file changed, 8 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>>> index d3297fadf7ed..25a25fff8dd9 100644
>>> --- a/arch/x86/kvm/vmx.c
>>> +++ b/arch/x86/kvm/vmx.c
>>> @@ -14482,13 +14482,6 @@ static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
>>> if (!page_address_valid(vcpu, kvm_state->vmx.vmxon_pa))
>>> return -EINVAL;
>>>
>>> - if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
>>> - return -EINVAL;
>>> -
>>> - if (kvm_state->vmx.vmcs_pa == kvm_state->vmx.vmxon_pa ||
>>> - !page_address_valid(vcpu, kvm_state->vmx.vmcs_pa))
>>> - return -EINVAL;
>>> -
>>> if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
>>> (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
>>> return -EINVAL;
>>> @@ -14510,6 +14503,14 @@ static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
>>> if (ret)
>>> return ret;
>>>
>>> + /* Empty 'VMXON' state is permitted */
>>> + if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
>>> + return 0;
>>
>> In the original version of this code, the early exit is:
>>
>> if (kvm_state.vmcs_pa == -1ull)
>> return 0;
>
> Time to restore the status quo :-)
>
> In any case this early exit should be done after enter_vmx_operation()
> as even without a valid VMCS we need to preserve VMXON state (e.g. Linux
> with loaded KVM but without any guests running).

Yes. That's also the way it was originally.