I see __atomic_compare_exchange_n function is being called with different memory order while
doing the same type of atomic operation. In spin lock ACQUIRE AND RELAX being used
but in rw_lock_lock_word_decr function
i see __ATOMIC_SEQ_CST being hard coded in the cmpxchng. I feel we can use consistenly use ACQUIRE AND RELAX memory order for all the mutex implemenation in all the platform.