Unfair Advantage Aware Spinlock: A Novel and Efficient Way to Counter Spinlock Starvation

Spinlocks, widely used as a synchronization method are implemented assuming that every core, which contends for the same spinlock, has an equal chance of winning, and hence, fairness can come automatically. In reality, due to the inherent hierarchical design of memory using caches, we have seen that the winner core keeps winning the spinlock after every release if it tries to. Fully fair spinlock primitives like ticket or queue-based locks use complicated algorithms and enforce conditions which result in higher. In this paper, we propose a novel solution where the unfair cache advantage of the winner core is nullified, all contending cores gain the same probability of acquiring a released lock and the worst-case wait time in case of spinlock starvation is reduced by a factor of ~100X. The proposed solution has a very low overhead and can be applied to production builds.