lmdk: fix building system-service modules

[lyakh]

System-service modules cannot include Zephyr headers and don't use exported symbols, they link sink_api.c and source_api.c into their images. Disable symbol exporting for them.

will try it first

[softwarecki]

This PR doesn't solve the problem. The __SOF_MODULE_SERVICE_BUILD__ symbol is defined only in one example module. We need more universal solution.

[lyakh]

This PR doesn't solve the problem. The __SOF_MODULE_SERVICE_BUILD__ symbol is defined only in one example module. We need more universal solution.

@softwarecki you don't use it when building LMDK modules? Ok, I thought you did. Is there anything that you do define for all system-service modules?

[softwarecki]

Ok, I thought you did. Is there anything that you do define for all system-service modules?

We don't define it in our examples of lmdk modules. Please check files in lmdk/cmake or use some definition used by llmodules.

[marc-hb]

or use "some" definition used by llmodules.

That sounds very brittle (quotes mine)

Can you confirm MAJOR_IADSP_API_VERSION is here to stay for a while?

[lgirdwood]

@softwarecki btw, why does system services build and link to a base FW file (e.fg the source/sink API) ? Shouldn't these be unresolved in the library and then be resolved when the library is loaded ?

[softwarecki]

@marc-hb MAJOR_IADSP_API_VERSION is currently unused and I planned to remove it in the future as I did in #8546 .

@lgirdwood I moved the common files used for building basefw and lmdk modules to the directory modules. These are simple wrappers that invoke pointers functions provided by the sink/source API.

@lyakh Can we reverse the condition and check for the existence of a definition associated with llext or move these exports to another file?

[softwarecki]

MAJOR_IADSP_API_VERSION is deprecated

[softwarecki]

This is a very surprising solution, but it works so I accept it.

[lyakh]

I think we can safely state, that this PR cannot be a reason for all the CI failures:

• cAVS suspend-resume https://sof-ci.01.org/sofpr/PR8812/build2551/devicetest/index.html
• ACE multiple pipelines and multiple pause-resume with nocodec https://sof-ci.01.org/sofpr/PR8812/build2550/devicetest/index.html
• Broken fuzzing is a recent regression
• Failing zmain builds are a known Zephyr breakage too

[marc-hb]

This is a very surprising solution, but it works so I accept it.

Moreover it's short, well documented and self-contained in only 2 places to change if needed in the future. So very low technical debt if any!