Re: [PATCH] Nested VMX: CR emulation fix up

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 10/09/2013 08:31 PM, Zhang, Yang Z wrote:
> Boris Ostrovsky wrote on 2013-10-09:
>> On 10/09/2013 03:28 AM, Zhang, Yang Z wrote:
>>> Boris Ostrovsky wrote on 2013-10-08:
>>>> On 10/08/2013 04:31 AM, Jan Beulich wrote:
>>>>
>>>>> Considering that this touches code common with nested SVM, I'd
>>>>> expect the SVM maintainers to have to approve of the change in any case.
>>>>>
>>>>> In particular I wonder whether this addition isn't obsoleting
>>>>> SVM's ns_cr0.
>>>>>
>>>> I am not sure whether ns_cr0 (replaced with nv_guest_cr[0]) would
>>>> then be updated in paths where it currently is not.
>>>>
>>>> For example in nsvm_vmcb_prepare4vmrun():
>>>>
>>>>        /* CR0 */ svm->ns_cr0 = v->arch.hvm_vcpu.guest_cr[0]; cr0 =
>>>>        nestedsvm_fpu_vmentry(svm->ns_cr0, ns_vmcb, n1vmcb, n2vmcb);
>>>>        v->arch.hvm_vcpu.guest_cr[0] = ns_vmcb->_cr0; rc =
>>>>        hvm_set_cr0(cr0);  <------ nv_guest_cr[0] will get set here.
>>> I am not familiar with SVM code. If you think this change may impact
>>> the
>> nested SVM. Then I will move the change to VMX specific code.
>>
>> No, it doesn't affect SVM code. I was responding to Jan's suggestion
>> to replace SVM's ns_cr0 with the new guest_cr[0].
> So is it ok to change the code according Jan's suggestion?

No. My point was that there may be unintended consequences to the change and
you should leave ns_cr0 alone.


-boris