Race condition in port shutdown

[eholk]

Now that ports have a lock, there is a race condition in the shutdown process. One task may call the port destructor while another task is still holding the lock.

[msullivan]

There has been a lot of churn in the comm system. Is this bug still there? The test that it seemed to be tripping, task-comm-15, seems to pass now.

[eholk]

I'm pretty sure the bug is still there, but the window for the race to occur in is much much smaller now.

[brson]

This may not exist, or may not be a problem we're running into. There are still two comm tests that are having trouble: task-comm-15 (windows), and task-comm-5 (periodic failure). Let's make sure this isn't the problem they're having.

[eholk]

I saw task-comm-15 fail on Linux today too.

[brson]

It looks to me like task-comm-15 fails because it's trying to send to a port while it's being destructed.

[brson]

I did some major simplification of the runtime code for ports and channels and I believe that port shutdown is no longer racy.