Re: xen 4.15.5: msr_relaxed required for MSR 0x1a2

[Jan Beulich <jbeulich@xxxxxxxx>]

On 20.11.2023 09:27, James Dingwall wrote:
> On Fri, Nov 17, 2023 at 10:56:30AM +0100, Jan Beulich wrote:
>> On 17.11.2023 10:18, James Dingwall wrote:
>>> On Thu, Nov 16, 2023 at 04:32:47PM +0000, Andrew Cooper wrote:
>>>> On 16/11/2023 4:15 pm, James Dingwall wrote:
>>>>> Hi,
>>>>>
>>>>> Per the msr_relaxed documentation:
>>>>>
>>>>>    "If using this option is necessary to fix an issue, please report a 
>>>>> bug."
>>>>>
>>>>> After recently upgrading an environment from Xen 4.14.5 to Xen 4.15.5 we
>>>>> started experiencing a BSOD at boot with one of our Windows guests.  We 
>>>>> found
>>>>> that enabling `msr_relaxed = 1` in the guest configuration has resolved 
>>>>> the
>>>>> problem.  With a debug build of Xen and `hvm_debug=2048` on the command 
>>>>> line
>>>>> the following messages were caught as the BSOD happened:
>>>>>
>>>>> (XEN) [HVM:11.0] <vmx_msr_read_intercept> ecx=0x1a2
>>>>> (XEN) vmx.c:3298:d11v0 RDMSR 0x000001a2 unimplemented
>>>>> (XEN) d11v0 VIRIDIAN CRASH: 1e ffffffffc0000096 fffff80b8de81eb5 0 0
>>>>>
>>>>> I found that MSR 0x1a2 is MSR_TEMPERATURE_TARGET and from that this patch
>>>>> series from last month:
>>>>>
>>>>> https://patchwork.kernel.org/project/xen-devel/list/?series=796550
>>>>>
>>>>> Picking out just a small part of that fixes the problem for us. Although 
>>>>> the
>>>>> the patch is against 4.15.5 I think it would be relevant to more recent
>>>>> releases too.
>>>>
>>>> Which version of Windows, and what hardware?
>>>>
>>>> The Viridian Crash isn't about the RDMSR itself - it's presumably
>>>> collateral damage shortly thereafter.
>>>>
>>>> Does filling in 0 for that MSR also resolve the issue?  It's model
>>>> specific and we absolutely cannot pass it through from real hardware
>>>> like that.
>>>>
>>>
>>> Hi Andrew,
>>>
>>> Thanks for your response.  The guest is running Windows 10 and the crash
>>> happens in a proprietary hardware driver.  A little bit of knowledge as
>>> they say was enough to stop the crash but I don't understand the impact
>>> of what I've actually done...
>>>
>>> To rework the patch I'd need a bit of guidance, if I understand your
>>> suggestion I set the MSR to 0 with this change in emul-priv-op.c:
>>
>> For the purpose of the experiment suggested by Andrew ...
>>
>>> diff --git a/xen/arch/x86/pv/emul-priv-op.c b/xen/arch/x86/pv/emul-priv-op.c
>>> index ed97b1d6fcc..66f5e417df6 100644
>>> --- a/xen/arch/x86/pv/emul-priv-op.c
>>> +++ b/xen/arch/x86/pv/emul-priv-op.c
>>> @@ -976,6 +976,10 @@ static int read_msr(unsigned int reg, uint64_t *val,
>>>          *val = 0;
>>>          return X86EMUL_OKAY;
>>>  
>>> +    case MSR_TEMPERATURE_TARGET:
>>> +        *val = 0;
>>> +        return X86EMUL_OKAY;
>>> +
>>>      case MSR_P6_PERFCTR(0) ... MSR_P6_PERFCTR(7):
>>>      case MSR_P6_EVNTSEL(0) ... MSR_P6_EVNTSEL(3):
>>>      case MSR_CORE_PERF_FIXED_CTR0 ... MSR_CORE_PERF_FIXED_CTR2:
>>
>> ... you wouldn't need this (affects PV domains only), and ...
>>
>>> and this in vmx.c:
>>>
>>> diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
>>> index 54023a92587..bbf37b7f272 100644
>>> --- a/xen/arch/x86/hvm/vmx/vmx.c
>>> +++ b/xen/arch/x86/hvm/vmx/vmx.c
>>> @@ -3259,6 +3259,11 @@ static int vmx_msr_read_intercept(unsigned int msr, 
>>> uint64_t *msr_content)
>>>          if ( !nvmx_msr_read_intercept(msr, msr_content) )
>>>              goto gp_fault;
>>>          break;
>>> +
>>> +    case MSR_TEMPERATURE_TARGET:
>>> +        *msr_content = 0;
>>> +        break;
>>> +
>>>      case MSR_IA32_MISC_ENABLE:
>>>          rdmsrl(MSR_IA32_MISC_ENABLE, *msr_content);
>>>          /* Debug Trace Store is not supported. */
>>
>> ... indeed this ought to do. An eventual real patch may want to look
>> different, though.
>>
> 
> Thanks Jan, based on the information I've reduced the patch to what seems the
> minimal necessary to workaround the BSOD.  I assume simply not ending up at
> X86EMUL_EXCEPTION is the resolution regardless of what value is set.

Good. This then confirms that Andrew's expectation of this being enough was
correct. What's not really clear to me though is whether he'd also be okay
to put a (cleaned up) patch along these lines into the tree. Andrew?

Jan