Please add a Signed-off-by tag to the commit message.

I did

git commit --amend --signoff 

and then

git push --force

Is that what I need to do?

Yes, thanks! Also, can you confirm that you're run the userspace-rcu "make regtest" and "make short_bench" without issues ? Listing the testing done in the commit and adding a Tested-by tag would help.

Added, all tests pass.

OK, I'll merge your patch, but I would recommend implementing futex_async and futex_noasync specifically for NetBSD, as was done for OpenBSD and FreeBSD (see include/urcu/futex.h). For instance, FreeBSD uses _umtx_op(). This will be more efficient than the busy-wait and pthread_condvar fallback implementations.

Considering that this will affect public headers installed on the system, it is a matter of ABI compatibility. It would therefore be simpler to introduce an efficient futex support along with the commit re-introducing NetBSD support before a release is done.

Thanks!

Added the futex implementation and ran the tests again. I am not sure if the tests actually called the futex code though. How do I make sure?

Added the futex implementation and ran the tests again. I am not sure if the tests actually called the futex code though. How do I make sure?

A quick way to make sure that your code is actually compiled is to add

#error "some message"

in the code.

A quick way to make sure that your code is actually run within the testsuite is to add an abort() in your code.

It is compiled and it is called :-) You can commit it.

I notice a few problems with your latest commit:

• it should be rebased on userspace RCU master (which has your first commit merged),
• the patch title should be the first line. Signed-off-by should be at the end of the commit message.
• I notice you used a futex() implementation. Posts like this one: https://mail-index.netbsd.org/tech-kern/2023/12/11/msg029334.html make me wonder if this compatibility layer is well tested. Is there an equivalent to the freebsd umtx as well on netbsd ? If yes, then it might be more robust to implement the wait/wakeup support based on umtx than the futex compatibility layer ?

1. I will rebase it
2. I will amend the commit message
3. We don't have the equivalent of FreeBSD umtx, and I don't think that this system call is well tested (although there are unit tests https://nxr.netbsd.org/xref/src/tests/lib/libc/sys/t_futex_ops.c), since it is not used elsewhere (for example: https://nxr.netbsd.org/xref/src/sys/external/bsd/compiler_rt/dist/lib/sanitizer_common/sanitizer_linux.cc#660 is not using it). The code was added for linux system call compatibility. Perhaps we should not be using it in userspace rcu either. I will ask what other people think

Another issue is that there is currently no sparc (non v9), sh3, and vax support (I will check if there are more missing)