- * This codepath deserves an explanation: when the thread is
- * quarantined, for is, really, it's already "parked". Though
- * userland doesn't know about, so wake as many threads as
- * userlands would have liked to, and let the wakeup tell
- * userland those should be parked.
+ * This codepath deserves an explanation: waking the thread
+ * "for real" would overcommit, though userspace KNOWS there
+ * is at least one waiting thread. Such threads are threads
+ * that are "quarantined".
+ *
+ * Quarantined threads are woken up one by one, to allow a
+ * slow ramp down, trying to minimize "waiting" <-> "parked"
+ * flip-flops, no matter how many wakes have been asked.
+ *
+ * Since releasing one quarantined thread will wake up a
+ * thread that will (almost) straight go to parked mode, lie
+ * to userland about the fact that we unblocked that thread,
+ * and return 0.