nuttx/mutex: do not use non-nx interface in kernel except libs

[pkarashchenko]

Summary

The #6320 redirected nxmutex used in kernel to sem_ API family in case of CONFIG_BUILD_FLAT. That is not correct. This PR is created just for discussion and just highlight the issue.

Impact

Testing

[pkarashchenko]

@xiaoxiang781216 could you please take a look. I noticed during my work that mutexes use sem_ API inside the kernel in case of FLAT build. Seems that we missed that point during the review of #6320 by trying to overcome the libc build error and switching from nxsem_ to _SEM_. This introduced a side effect in case of FLAT build.

[pkarashchenko]

One of the possible ways may be to get back original code for the files that I've listed in this PR and switching nxmutex_ APIs to use nxsem_ APIs only.

[xiaoxiang781216]

One of the possible ways may be to get back original code for the files that I've listed in this PR and switching nxmutex_ APIs to use nxsem_ APIs only.

How about we follow sem_t approach:

• include/mutex.h which call sem_xxx and set errno
• include/nuttx/mutex.h which call nxsem_xxx but not set errno
• add MUTEX_XXXX macro in include/nuttx/mutex.h which call first or second version by check __KERNEL__

[pkarashchenko]

Yes. This can work. Similar approach like _SEM_XXX

[pkarashchenko]

One of the possible ways may be to get back original code for the files that I've listed in this PR and switching nxmutex_ APIs to use nxsem_ APIs only.

How about we follow sem_t approach:

* include/mutex.h which call sem_xxx and set errno

* include/nuttx/mutex.h which call nxsem_xxx but not set errno

* add MUTEX_XXXX macro in include/nuttx/mutex.h which call first or second version by check `__KERNEL__`

Done, but IMO we should merge mutex_t and rmutex_t (as well as APIs) and just add bool recursive; flag to mutex_t and just keep one set of APIs

[pkarashchenko]

@xiaoxiang781216 seems that mutex_ notation conflicts with 3P component naming. Since mutex_ is not a POSIX APIs I'm not sure what is the best way to solve the naming conflict. mtx_ prefix is already occupied (include/threads.h). What do you think, how we can resolve the naming conflict?

[pkarashchenko]

The issue of naming conflict comes from: #include <stdio.h> -> #include <nuttx/fs/fs.h> -> #include <nuttx/mutex.h> because struct file_struct has member rmutex_t fs_lock;

[pkarashchenko]

@xiaoxiang781216 ping

[xiaoxiang781216]

add space after ,

[xiaoxiang781216]

add space after ,

[xiaoxiang781216]

why need save pid in this case?

[pkarashchenko]

In general mutex should have the holder stored and do not allow unlock not from a holder thread. We can even DEBUGASSERT if we try to unlock from not a mutex holder

[xiaoxiang781216]

Since mutex is used very often in the code base. How about we enable this for debugging only?

[pkarashchenko]

Do you mean check for normal mutex? The recursive need to store a holder

[xiaoxiang781216]

The issue of naming conflict comes from: #include <stdio.h> -> #include <nuttx/fs/fs.h> -> #include <nuttx/mutex.h> because struct file_struct has member rmutex_t fs_lock;

It's hard to fix:(

[zyfeier]

@pkarashchenko Hi, pkarashchenko, is this PR still being modified？

[pkarashchenko]

I will review and most probably close this PR as it is outdated

[zyfeier]

@pkarashchenko, i have a similar modify, please help to review. #8645