kqueue: drain kqueue taskqueue if syscall tickled it

[kib]

Summary

Otherwise return from the syscall and next syscall, which could be kevent(2) on the kqueue that should be notified, races with the kqueue taskqueue thread, and potentially miss the wakeup. This is reliably visible when kevent(2) only peeks into events using zeroed timeout.

PR: 258310

[arichardson]

I can confirm this reliably fixes the test original test case (converted to ATF in D31862).
Code looks good to me, but since I'm not familiar with that part of the kernel I'd wait for @markj to review too.

[markj]

sys/kern/kern_event.c

1774

Isn't taskqueue_drain_all() sufficient? Or is the idea that we may have a chain of multiple kqueues listening to each other, and we need to wait until all of them have been notified?

[kib]

sys/kern/kern_event.c

1774

Yes, I do not think that it is formally enough to drain only tasks already scheduled on the taskqueue.

This is a rare situation anyway so I do not think there is much overhead to wait for full quiesce.