splitlock

Splitlock is a term used in computer architecture to describe a condition in which an atomic operation that uses a lock prefix cannot complete atomically because the memory system must service the request with multiple bus transactions. In x86 architectures, the LOCK prefix is intended to serialize access to a memory operand for the duration of the instruction, ensuring exclusive ownership. However, under certain hardware configurations—such as accesses to memory-mapped I/O, non-cache-coherent memory, or complex cache/memory scenarios—the lock may be distributed across more than one bus transaction. When this occurs, the operation is effectively split into segments, potentially compromising the intended atomicity of the instruction and introducing performance penalties.

Detection and handling of split locks vary by processor and platform. Some CPUs can detect split-lock conditions and may stall the issuing core or expose signals to performance monitoring features. In some environments, the operating system or hypervisor may treat repetitive split-lock behavior as a fault to avoid data corruption or severe performance degradation, potentially terminating the offending thread or applying mitigations.

Impact and mitigation center on performance and correctness. Split locks can significantly slow down systems and, in some cases, obscure race conditions by failing to provide the guaranteed atomicity of an operation. Mitigation strategies include keeping critical sections short, avoiding atomic operations that involve I/O or non-cache-coherent memory, using lock-free data structures or higher-level synchronization primitives, and ensuring memory involved in locked operations remains cache-resident and coherent. Profiling tools can help identify excessive split-lock events.

See also: LOCK prefix, atomic operation, memory ordering, x86 architecture, cache coherence.