Re: [PATCH] hyperv: root partition faults writing to VP ASSIST MSR PAGE

From: Praveen Kumar
Date: Wed Jul 21 2021 - 03:17:45 EST


On 21-07-2021 09:40, Michael Kelley wrote:
> From: Wei Liu <wei.liu@xxxxxxxxxx> Sent: Tuesday, July 20, 2021 9:29 AM
>>
>> On Tue, Jul 20, 2021 at 04:20:44PM +0000, Michael Kelley wrote:
>>> From: Wei Liu <wei.liu@xxxxxxxxxx> Sent: Tuesday, July 20, 2021 6:35 AM
>>>>
>>>> On Tue, Jul 20, 2021 at 06:55:56PM +0530, Praveen Kumar wrote:
>>>> [...]
>>>>>>
>>>>>>> + if (hv_root_partition &&
>>>>>>> + ms_hyperv.features & HV_MSR_APIC_ACCESS_AVAILABLE) {
>>>>>>
>>>>>> Is HV_MSR_APIC_ACCESS_AVAILABLE a root only flag? Shouldn't non-root
>>>>>> kernel check this too?
>>>>>
>>>>> Yes, you are right. Will update this in v2. thanks.
>>>>
>>>> Please split adding this check to its own patch.
>>>>
>>>> Ideally one patch only does one thing.
>>>>
>>>> Wei.
>>>>
>>>
>>> I was just looking around in the Hyper-V TLFS, and I didn't see
>>> anywhere that the ability to set up a VP Assist page is dependent
>>> on HV_MSR_APIC_ACCESS_AVAILABLE. Or did I just miss it?
>>
>> The feature bit Praveen used is wrong and should be fixed.
>>
>> Per internal discussion this is gated by the AccessIntrCtrlRegs bit.
>>
>> Wei.
>>
>
> The AccessIntrCtrlRegs bit *is* HV_MSR_APIC_ACCESS_AVAILABLE.
> Both are defined as bit 4 of the Partition Privilege flags. :-) I don't
> know why the names don't line up. Even so, it's not clear to me that
> AccessIntrCtrlRegs has any bearing on the VP Assist page. I see this
> description of AccessIntrCtrlRegs:
>

Yup, what I understood as well, this is the one required one for Partition Privilege Flags (4th bit), however, cannot comment on the naming convention.

5 /* Virtual APIC assist and VP assist page registers available */
4 #define HV_MSR_APIC_ACCESS_AVAILABLE BIT(4)

> The partition has access to the synthetic MSRs associated with the
> APIC (HV_X64_MSR_EOI, HV_X64_MSR_ICR and HV_X64_MSR_TPR).
> If this flag is cleared, accesses to these MSRs results in a #GP fault if
> the MSR intercept is not installed.
>

As per what I also understood from the TLFS doc,that we let partition access the MSR and do a fault.
However, the point is, does it make sense to allocate page for vp assist and perform action which is meant to fail when the flag is cleared ?

> But maybe you have additional info that applies to the root
> partition that is not in the TLFS.
>

As per what discussed internally and I understood, the root partition shares the vp assist page provided by hypervisor and its read only for Root kernel.

> Michael
>

Regards,

~Praveen.