Re: [RFC] accel: add cpu_reset

[Claudio Fontana]

On 3/22/21 2:45 PM, Paolo Bonzini wrote:
> On 22/03/21 14:35, Claudio Fontana wrote:
>> On 3/22/21 2:31 PM, Paolo Bonzini wrote:
>>> On 22/03/21 14:27, Claudio Fontana wrote:
>>>> This surprisingly works without moving cpu_reset() to a specific_ss
>>>> module, even though accel-common.c is specific_ss, hw/core/cpu.c is
>>>> common_ss. How come the call to accel_reset_cpu works?
>>>
>>> I don't understand the question.  Why wouldn't it work? :)
>>>
>>> Paolo
>>>
>>
>> Heh probably something I forgot or do not understand around the specific_ss 
>> / common_ss distinction.
>>
>> I was under the (wrong?) impression that we build some tools or components 
>> that include common_ss objects, but not specific_ss.
>>
>> And maybe I am just wrong, and things are simpler than I expected.
> 
> No, all emulators include:
> 
> - some parts of common_ss, compiled once per build.  These are files 
> that do not use target-specific definitions.  Other sourcesets also 
> define once-per-build files, and in fact they end up in common_ss via 
> the add_all method of sourcesets; softmmu_ss, for example is added to 
> common_ss under the CONFIG_SOFTMMU condition.
> 
> - some parts of specific_ss, compiled once per target because these 
> files use target-specific definitions.
> 
> - the entirety of the respective hw/ and target/ sourcesets.
> 
> It is possible to include calls from one sourceset to another (including 
> from common to specific) as long as the conditions ensure that the 
> symbol is defined.


I guess this last sentence is the more tricky for me to get: "as long as the 
conditions ensure that the symbol is defined".

> 
> Paolo
> 

Thanks for the explanation, I would assume that "make check" then would be able 
to catch such problems?

Which targets would I need to build to ensure that any problems with this are 
detected? Do we cover all of these cases with our gitlab CI?

Ciao,

Claudio