Re: [PATCH] device-tree: Move Arm's static-evtchn feature to common

[Jan Beulich <jbeulich@suse.com>]

On 03.06.2025 09:06, Orzel, Michal wrote:
> 
> 
> On 03/06/2025 09:02, Jan Beulich wrote:
>> On 03.06.2025 08:54, Orzel, Michal wrote:
>>>
>>>
>>> On 02/06/2025 10:37, Jan Beulich wrote:
>>>> On 27.05.2025 10:21, Michal Orzel wrote:
>>>>> There's nothing Arm specific about this feature. Move it to common as
>>>>> part of a larger activity to commonalize device tree related features.
>>>>> For now, select it only for ARM until others (e.g. RISC-V) verify it
>>>>> works for them too.
>>>>>
>>>>> Signed-off-by: Michal Orzel <michal.orzel@amd.com>
>>>>
>>>> I realize this was already committed, but ...
>>>>
>>>>> --- a/xen/common/Kconfig
>>>>> +++ b/xen/common/Kconfig
>>>>> @@ -162,6 +162,14 @@ config STATIC_MEMORY
>>>>>  
>>>>>  	  If unsure, say N.
>>>>>  
>>>>> +config STATIC_EVTCHN
>>>>> +	bool "Static event channel support on a dom0less system"
>>>>> +	depends on DOM0LESS_BOOT && ARM
>>>>
>>>> ... I think we should strive to avoid such arch dependencies; they simply
>>>> don't scale very well. Instead (if needed) HAS_* should be introduced, which
>>>> each interested arch can select. In the case here, however, perhaps
>>>> DOM0LESS_BOOT alone would have been sufficient as a dependency?
>>> What if e.g. RISC-V wants to enable dom0less but not static evtchn/memory/shmem
>>> because there are some functions to be implemented and they don't want to do it
>>> now? Protecting with just DOM0LESS_BOOT would not be sufficient here.
>>
>> Imo a transient(!) "depends on !RISCV" would in principle be fine, if ...
> In this case, how can I know that upfront? When moving a feature I need to put
> some dependencies. At that point I don't know RISCV, PPC plans.

You don't need to know this up front, do you? Neither of the two presently selects
DOM0LESS_BOOT afaict, and hence such a transient negative dependency would (if
necessary in the first place) be added when adding such a select.

Jan