On 26.09.22 17:23, Jan Beulich wrote:
> On 26.09.2022 16:18, Juergen Gross wrote:
>> --- a/arch/x86/xen/Kconfig
>> +++ b/arch/x86/xen/Kconfig
>> @@ -92,3 +92,12 @@ config XEN_DOM0
>>   	select X86_X2APIC if XEN_PVH && X86_64
>>   	help
>>   	  Support running as a Xen Dom0 guest.
>> +
>> +config XEN_PV_MSR_SAFE
>> +	bool "Always use safe MSR accesses in PV guests"
>> +	default y
> 
> Is there any time line when this default will change, perhaps first to
> DEBUG and later to N?

I'm not sure. I did an initial test with the safe variants disabled in dom0
and it just worked.

I'm not sure we want an intermediate step, as in critical cases the user can
still use the boot parameter.

> 
>> @@ -1010,22 +1020,16 @@ static int xen_write_msr_safe(unsigned int msr, unsigned int low,
>>   
>>   static u64 xen_read_msr(unsigned int msr)
>>   {
>> -	/*
>> -	 * This will silently swallow a #GP from RDMSR.  It may be worth
>> -	 * changing that.
>> -	 */
>>   	int err;
>>   
>> -	return xen_read_msr_safe(msr, &err);
>> +	return xen_do_read_msr(msr, xen_msr_safe ? &err : NULL);
>>   }
> 
> When we were talking at the session, I think I said that at least there
> is no uninitialized value being passed back. But I did look at
> xen_read_msr_safe() only, which indeed is okay. Whereas
> native_read_msr_safe() isn't (nor is native_read_msr() afaict), so I
> think part of this series should be to also eliminate the undefined-
> ness from this code path (possible now only when xen_msr_safe is true,
> but as per above that'll be the default at least for some time), where
> the caller has no way to know that it shouldn't look at the value.

I can add that.


Juergen