Dekker’s algorithm correctly implements mutual exclusion. However, this is a little bit convoluted. Additionally, each process must know about the other process, which violates modularity for programming. Therefore, we create a new abstraction: locks.
Say we have a lock called l. Then, we can acquire the lock (lock(l)) or release the lock (unlock(l)).
Then, a program consists of the NCS then a lock then CS then unlock.
Note that using a single lock for different resources that need to be protected with a lock can be inefficient. Therefore, if multiple resources can change x, we can assign a lock to x.
Deadlock
It is possible to have nested locks without deadlock, but deadlock can happen, so it is best to avoid. For example, P1 could acquire lock a then b and P2 could acquire lock b then a.
Implementation
How do we implement locks? We could disable interrupts since interrupts are used to switch between processes. However, this does not generalize to multi-core systems.