Re: [PATCH v2] x86/vmx: add support for virtualize SPEC_CTRL

[Jan Beulich <jbeulich@xxxxxxxx>]

On 15.02.2024 16:54, Roger Pau Monné wrote:
> On Mon, Feb 12, 2024 at 03:09:01PM +0100, Jan Beulich wrote:
>> On 09.02.2024 12:40, Roger Pau Monne wrote:
>>> --- a/xen/arch/x86/hvm/vmx/vmx.c
>>> +++ b/xen/arch/x86/hvm/vmx/vmx.c
>>> @@ -823,18 +823,29 @@ static void cf_check vmx_cpuid_policy_changed(struct 
>>> vcpu *v)
>>>      {
>>>          vmx_clear_msr_intercept(v, MSR_SPEC_CTRL, VMX_MSR_RW);
>>>  
>>> -        rc = vmx_add_guest_msr(v, MSR_SPEC_CTRL, 0);
>>> -        if ( rc )
>>> -            goto out;
>>> +        if ( !cpu_has_vmx_virt_spec_ctrl )
>>> +        {
>>> +            rc = vmx_add_guest_msr(v, MSR_SPEC_CTRL, 0);
>>> +            if ( rc )
>>> +                goto out;
>>> +        }
>>>      }
>>>      else
>>>      {
>>>          vmx_set_msr_intercept(v, MSR_SPEC_CTRL, VMX_MSR_RW);
>>>  
>>> -        rc = vmx_del_msr(v, MSR_SPEC_CTRL, VMX_MSR_GUEST);
>>> -        if ( rc && rc != -ESRCH )
>>> -            goto out;
>>> -        rc = 0; /* Tolerate -ESRCH */
>>> +        /*
>>> +         * NB: there's no need to clear the virtualize SPEC_CTRL control, 
>>> as
>>> +         * the MSR intercept takes precedence.  The SPEC_CTRL shadow VMCS 
>>> field
>>> +         * is also not loaded on guest entry/exit if the intercept is set.
>>> +         */
>>
>> It wasn't so much the shadow field than the mask one that I was concerned
>> might be used in some way. The shadow one clearly is used only during
>> guest RDMSR/WRMSR processing. To not focus on "shadow", maybe simple say
>> "The SPEC_CTRL shadow VMCS fields are also not ..."?
> 
> What about:
> 
> "The SPEC_CTRL shadow and mask VMCS fields don't take effect if the
> intercept is set."

SGTM.

Jan