Design of Mutex and Event Synchronization Primitives for a Multithreaded Managed Runtime with Stackful Coroutines

Abstract

Operating system synchronization primitives are unsuitable for managed runtimes with coroutines: acquiring an OS mutex blocks the entire carrier thread, which leads to deadlocks under cooperative multitasking. Existing coroutine-aware solutions are either tightly coupled to the internals of a specific runtime (Go, JVM), or target native environments without a garbage collector (Boost.Fiber), or allow waiting only in an asynchronous context (C#, Kotlin). This paper proposes a self-contained architecture of Mutex and Event synchronization primitives for the multithreaded managed ArkTS/OpenHarmony runtime with stackful coroutines. The primitives are managed-heap objects with built-in queues of waiting coroutines allowing the architecture to be ported to other managed runtimes whose scheduler provides coroutine suspend/resume operations. The one-shot Event replaces the condition variable with simpler semantics. The mutex is adapted to short critical sections with an exponential-backoff spin phase and an adaptive starvation mode guaranteeing acquisition within a bounded number of wakeups. Promise<T> and AsyncLock with wait-for-graph deadlock detection are built on top. On an 8-core arm64 device, the average acquisition time stays within hundreds of nanoseconds for up to 1000 coroutines, with overhead below 1.5x in most configurations versus 10–14x for Go's sync.Mutex. The gain comes at the cost of higher maximum (tail) latencies under long critical sections and high contention.