Re: [RFC PATCH 01/11] x86: Add more granularity to the vendors in Kconfig

[Jan Beulich <jbeulich@xxxxxxxx>]

On 27.11.2025 14:44, Alejandro Vallejo wrote:
> On Thu Nov 27, 2025 at 2:19 PM CET, Jan Beulich wrote:
>> On 27.11.2025 13:36, Alejandro Vallejo wrote:
>>> On Thu Nov 27, 2025 at 10:43 AM CET, Jan Beulich wrote:
>>>> On 26.11.2025 17:44, Alejandro Vallejo wrote:
>>>>> --- a/xen/arch/x86/Kconfig.cpu
>>>>> +++ b/xen/arch/x86/Kconfig.cpu
>>>>> @@ -19,4 +19,49 @@ config INTEL
>>>>>     May be turned off in builds targetting other vendors.  Otherwise,
>>>>>     must be enabled for Xen to work suitably on Intel platforms.
>>>>>  
>>>>> +config HYGON
>>>>> + bool "Support Hygon CPUs"
>>>>> + depends on AMD
>>>>> + default y
>>>>> + help
>>>>> +   Detection, tunings and quirks for Hygon platforms.
>>>>> +
>>>>> +   May be turned off in builds targetting other vendors.  Otherwise,
>>>>> +   must be enabled for Xen to work suitably on Hygon platforms.
>>>>> +
>>>>> +
>>>>> +config CENTAUR
>>>>> + bool "Support Centaur CPUs"
>>>>> + depends on INTEL
>>>>> + default y
>>>>> + help
>>>>> +   Detection, tunings and quirks for Centaur platforms.
>>>>> +
>>>>> +   May be turned off in builds targetting other vendors.  Otherwise,
>>>>> +   must be enabled for Xen to work suitably on Centaur platforms.
>>>>> +
>>>>> +config SHANGHAI
>>>>> + bool "Support Shanghai CPUs"
>>>>> + depends on INTEL
>>>>> + default y
>>>>> + help
>>>>> +   Detection, tunings and quirks for Shanghai platforms.
>>>>> +
>>>>> +   May be turned off in builds targetting other vendors.  Otherwise,
>>>>> +   must be enabled for Xen to work suitably on Shanghai platforms.
>>>>> +
>>>>> +config UNKNOWN_CPU
>>>>> + bool "Support unknown CPUs"
>>>>
>>>> "Unknown CPUs" can be of two kinds: Such of vendors we don't explicitly 
>>>> support,
>>>> and such of vendors we do explicitly support, but where we aren't aware of 
>>>> the
>>>> particular model. This needs to be unambiguous here, perhaps by it becoming
>>>> UNKNOWN_CPU_VENDOR (and the prompt changing accordingly).
>>>
>>> Right, what I do in this RFC is have compiled-out vendors fall back onto the
>>> unknown vendor path. Because it really is unknown to the binary.
>>>
>>> I could call it GENERIC_CPU_VENDOR, or anything else, but the main question
>>> is whether a toggle for this seems acceptable upstream. I don't see obvious
>>> drawbacks.
>>
>> I'd recommend against "generic" or anything alike, as it'll rather suggest 
>> any
>> vendor's CPU will work reasonably. I'm fine with "unknown", just that the 
>> nature
>> of the unknown-ness needs making unambiguous.
> 
> Got it, if UNKNOWN_CPU_VENDOR sounds better I'm fine with that.
> 
> What are your thoughts on the panic-on-compiled-out-vendor vs use-unknown?

Both have benefits and downsides, I think. The choice may need to be another
Kconfig or command line option. I'm also curious what other x86 maintainers
may think.

Jan