TSAN complaint about lock-order inversion

[nathanwbrei]

TSAN complains about a lock-order inversion. In one place we already have the scheduler lock, and we acquire the event source lock. In another place we have the event source lock, and acquire the scheduler lock. Usually a cycle in the lock acquisition graph indicates potential for deadlock. I'm not too concerned about this particular case because it is only triggered when the user presses Ctrl-C, and even if it does deadlock, they can always Ctrl-C a second time to hard-exit. However, we might be redesigning the event source locking for the sake of performance optimization anyhow, and if we do, we should fix this then.

warnings.txt