Re: [PATCH v1 12/16] xen/domain: introduce domain-emu.h

[Jan Beulich <jbeulich@xxxxxxxx>]

On 01.08.2025 21:30, dmkhn@xxxxxxxxx wrote:
> On Fri, Aug 01, 2025 at 08:02:56AM +0200, Jan Beulich wrote:
>> On 31.07.2025 22:55, dmkhn@xxxxxxxxx wrote:
>>> On Wed, Jul 09, 2025 at 04:57:44PM +0200, Jan Beulich wrote:
>>>> On 24.06.2025 05:56, dmkhn@xxxxxxxxx wrote:
>>>>> @@ -458,16 +459,16 @@ struct arch_domain
>>>>>  } __cacheline_aligned;
>>>>>
>>>>>  #ifdef CONFIG_HVM
>>>>> -#define X86_EMU_LAPIC    XEN_X86_EMU_LAPIC
>>>>> -#define X86_EMU_HPET     XEN_X86_EMU_HPET
>>>>> -#define X86_EMU_PM       XEN_X86_EMU_PM
>>>>> -#define X86_EMU_RTC      XEN_X86_EMU_RTC
>>>>> -#define X86_EMU_IOAPIC   XEN_X86_EMU_IOAPIC
>>>>> -#define X86_EMU_PIC      XEN_X86_EMU_PIC
>>>>> -#define X86_EMU_VGA      XEN_X86_EMU_VGA
>>>>> -#define X86_EMU_IOMMU    XEN_X86_EMU_IOMMU
>>>>> -#define X86_EMU_USE_PIRQ XEN_X86_EMU_USE_PIRQ
>>>>> -#define X86_EMU_VPCI     XEN_X86_EMU_VPCI
>>>>
>>>> The old code deliberately used values from the public interface.
>>>
>>> In next version I am building, I moved all of XEN_X86_EMU_XXX definitions as
>>> is to a new public header under include/public/xen-emu.h:
>>>
>>>   
>>> https://gitlab.com/xen-project/people/dmukhin/xen/-/commit/9b0bc5ffa5710114df8523ae2aa7680b7c6f0942
>>>
>>> That looks less invasive.
>>>
>>> Will that work?
>>>
>>> There should be a common header with emulation flags somewhere, since
>>> there will be SBSA and hwdom vUART definitions there.
>>
>> Yet will there be a strict need for any constants to be identical (i.e.
>> not only have the same name, but also the same value) across architectures?
> 
> I don't think there's strict need for identical values across achitectures.

That's what I was expecting.

> But some of the constants _may_ be reused for non-x86 arches, like VPCI bit
> and, perhaps, IOMMU, PIRQ and future NS16550 (after adding MMIO).

Right, but as you say - they want to use the same name, but they could easily
have a different value there. I hope you understand that what I'm questioning
is the introduction of a single header covering all architectures.

Jan