kernel - Optimize spinlocks for 48-core contention