sched: remove DEBUGASSERT from nx_waitpid

[pkarashchenko]

Summary

stat_loc of wait and waitpid can be NULL according to POSIX (https://pubs.opengroup.org/onlinepubs/9699919799/functions/wait.html): In this case, if the value of the argument stat_loc is not a null pointer, information shall be stored in the location pointed to by stat_loc.

Remove DEBUGASSERT(stat_loc); from nx_waitpid

Impact

Minimal if this was not reported by anyone till now

Testing

Pass CI
Tested with SAME70-QMTECH board with CONFIG_SCHED_HAVE_PARENT=n

[gustavonihei]

@pkarashchenko Could you please link a reference to the POSIX chapter that states that stat_loc may be NULL?

[pkarashchenko]

@pkarashchenko Could you please link a reference to the POSIX chapter that states that stat_loc may be NULL?

I've updated PR description

[gustavonihei]

@pkarashchenko Could you please link a reference to the POSIX chapter that states that stat_loc may be NULL?

I've updated PR description

Thanks, I had skimmed that page, but I failed to find the information.
The excerpt you pasted does not state that stat_loc may be NULL. It just determines the behavior when stat_loc is non-NULL.

I tried looking at other implementations. In the Linux Kernel, if stat_loc (stat_addr in this case) is NULL, the kernel returns -EFAULT to the userspace:
https://github.com/torvalds/linux/blob/master/kernel/exit.c#L1683

[pkarashchenko]

Yeah, but POSIX does not define an error code if start_loc is NULL, so Linux interpreted that to an error code.
For me it is natural that if start_loc is NULL then the caller "does not care" about exit status.

[gustavonihei]

Yeah, but POSIX does not define an error code if start_loc is NULL, so Linux interpreted that to an error code. For me it is natural that if start_loc is NULL then the caller "does not care" about exit status.

I agree with your point.
In summary:

• Prior to this PR, wait and waitpid had a premise of stat_loc being non-NULL. Passing a NULL pointer resulted in undefined behavior when DEBUG assertions are not enabled.
• This PR defines a proper behavior for a NULL stat_loc.

I am okay with the proposed changes.

[pkarashchenko]

@gustavonihei I think if POSIX want to define start_loc to be non-NULL, then the error code section would state that explicitly. And there would be no sentence that I pasted in the PR description.