struct orientevt {
	bool status;				/*indicate if event is within range*/
	bool close;				/*indicate if event is destroeyd*/
	unsigned int evt_id;			/*event_id managed by IDR*/
	atomic_t num_proc;			/*number of blocked processes*/
	struct orientation_range *orient;	/*orient range assigned by orientevt_create*/
	wait_queue_head_t blocked_queue;	/*blocked processes attached to this event*/
};

A IDR lock for orientevt and locks inside add_wait_queue, prepare_to_wait and finish_wait.

set_orientation:Update status in every struct orientevt and if it is active wake up wait queue.

orientevt_create:Create struct orientevt and link orientevt with unique ID by IDR.

orientevt_destroy:Remove corresponding struct orientevt from IDR. After making sure no process is blocked on this event(indicated by num_proc) it will free the space.

orientect_wait:If the event exists, create struct wait_queue and synchronize with orientevt. If event is active or event is destroyed it will leave wait queue. Particularly if event is destroyed, it will return -EFAULT.

Test process will fork 4 child processes. Two of them follow facing up event while others follow facing down. After TEST_SEC it will close these two events to let child process exit.

Lock seems work properly because I noticed the syscall order may change sometimes:
1:facing up!
2:facing up!
2:facing up!
1:facing up!
1:facing up!
2:facing up!