The Big Kernel Lock

http://lwn.net/Articles/281437/

BLK    

    ==> semaphore

            ==> The semaphore is handled over to the new thread waiting for it. This introduces performance penalty --- the system's running idle almost 50% of the time.

            ==> lock stealing: A thread acquiring the semaphore could get it once the semaphore becomes available, regardless of whether or not the thread's at the head of the queue. However, this introduces bugs in some situation.

 

    ==> spin lock

               ==> makes the BKL non-preemptible. However, this sub-optimal BKL is currently the best solution for this trade-off problem.