fix zephyr conflicting types atomic

[hongshui3000]

fix #4049

Signed-off-by: yxh yangxiaohuamail@gmail.com

[sofci]

Can one of the admins verify this patch?

reply test this please to run this test once.

[gkbldcig]

Can one of the admins verify this patch?

[dbaluta]

Please use a prefix for your commit subject.

E.g include: Fix zephyr conflicting types

then please explain in the commit message WHY do we need this change. Do you get a compiler error?

Please give more information so that people can understand the intention of your change.

[lgirdwood]

@zrombel good to merge ? CI pending....

[zrombel]

PR was build and it looks taht it's good to merge.

[lgirdwood]

@lyakh I assume this is good for you ? Any idea why you have not seen this ?

[iuliana-prodan]

@lyakh I assume this is good for you ? Any idea why you have not seen this ?

@lgirdwood @lyakh I'm also working on imx support for SOF with Zephyr and I didn't encounter this issue.
For building and adding support I've followed the steps from documentation , the same as those from this pull request: #4060

[hongshui3000]

@lyakh I assume this is good for you ? Any idea why you have not seen this ?

@lgirdwood @lyakh I'm also working on imx support for SOF with Zephyr and I didn't encounter this issue.
For building and adding support I've followed the steps from documentation , the same as those from this pull request: #4060

This kind of issue won't occur when you just compile a simple sample. When you use src/include/sof/atomic.h (this header file may also be included by others) and zephyr at the same time. The issue will reappear。Please note that my modification is in host arch, and other platforms are not considered.Because my platform is not in the official support list

[lyakh]

@lyakh I assume this is good for you ? Any idea why you have not seen this ?

@lgirdwood because SOF code includes sof/atomic.h whereas Zephyr code includes sys/atomic.h... So, I'm not sure this is a complete fix. A complete one should also remove the definition of atomic_t and include sys/atomic.h instead?

[lyakh]

wouldn't it be better and more correct to move this #ifndef 3 lines up and to #include <sys/atomic.h> on the #else branch? Or better invert the condition to #ifdef to avoid a negation.

[hongshui3000]

Agree

[hongshui3000]

I re-reveiw the sof code, the Cmakefile in the zephyr directory contains two header file paths.

target_include_directories(SOF INTERFACE ../zephyr/include)
target_include_directories(SOF INTERFACE ${SOF_SRC_PATH}/include)

When we compile sof, #include <sof/atomic.h>, then the header files under sof/zephyr will be found first. So we don't see the issue of type conflicts.
But when I compile the app, rtos zephyr includes its own atomic.h by default, and if I use sof with an atomic header, the atomic.h in the sof/src/include directory will be found at the same time. The issue I said happened.

Therefore, there is another modification scheme that does not change the sof code, but it needs to introduce the header files in the sof/zephyr/include directory when the external app is compiled, overwriting the header file path of sof itself
APP BUID:
Originally wrong Cmakefile：

zephyr_library_include_directories(sof_driver PUBLIC
    ${sof_module}/src/arch/${ARCH}/include
    ${sof_module}/src/include
  )

WORK:

zephyr_library_include_directories(sof_driver PUBLIC
    ${sof_module}/src/arch/${ARCH}/include
    ${sof_module}/zephyr/include                        -------------------------->The repetition provided by atomic and zephyr, but can 
                                                                                           overwrite the atomic of sof itself
    ${sof_module}/src/include
  )

But I don’t think this is a good way

[lgirdwood]

@hongshui3000 the plan is to remove all the SOF ARCH header paths as we continue and use more and more native Zephyr APIs (today we still have some in wrapper.c) i.e. SOF include path would only be for audio APIs.
@marc-hb what is the status of your native Zephyr APIs work ? as this should solve all these problems.

[jgunn0262]

I think using Zephyr is a great idea, thanks @hongshui3000.

[lgirdwood]

Jenkins shows unrelated DUT boot failure.