p37: Introduce atomic spinlocks for multiprocessor concurrency

[2023cs50578]

Objective

This branch initiates the transition from a uniprocessor architecture to Symmetric Multiprocessing (SMP). It introduces the core mutual exclusion primitives (spinlocks) required to protect shared kernel data structures from race conditions once multiple CPU cores are booted.

Core Changes

• x86.h: Ensured the atomic hardware instruction xchg (exchange) is available for read-modify-write operations, leveraging the CPU's hardware-level locking.
• spinlock.h: Defined the struct spinlock to track lock state (locked) and diagnostic data (holding cpu).
• spinlock.c: Implemented the core lock lifecycle functions:
  • initlock: Initializes lock state.
  • acquire: Uses the atomic xchg instruction in a tight loop to spin until the lock is available. Includes __sync_synchronize() barriers to prevent compiler/hardware instruction reordering and wraps the logic in pushcli() to prevent deadlocks from interrupt handlers.
  • release: Uses an atomic movl to clear the lock state and issues memory barriers to ensure critical section stores are visible before the lock is released.
  • holding: Diagnostic utility to verify ownership.
• defs.h: Added the respective prototypes for the spinlock API.

Next Steps

With the atomic primitives in place, subsequent branches can begin retrofitting global kernel structures (e.g., the physical memory allocator kmem, the process table ptable) with these locks in preparation for booting the Application Processors (APs).