Locks and Synchronization Primitives

A lock is a promise enforced by hardware: atomic compare-and-swap, memory barriers, and a kernel wait queue. This video shows what each primitive actually does on the machine — the atomic instruction, the cache-line contention, and the park/unpark syscall — so you can reason about why a mutex costs 20ns uncontended and a syscall-bound lock costs microseconds.

Topics covered:

• The atomic operations that all locks are built from: CAS, fetch-add, barriers
• Spinlocks: spinning on a cache line and why they're only for short critical sections
• Mutexes: the uncontended fast path vs. the contended futex slow path
• Semaphores and counting: producer/consumer signaling, not just mutual exclusion
• Condition variables: why wait() must atomically release the mutex
• Reader-writer locks and the writer-starvation problem
• Lock-free programming: when CAS-based retry loops beat locks
• The cost model: contention, cache-line bouncing, and false sharing