Modular data-race-freedom guarantees in the promising semantics

Abstract

Local data-race-freedom guarantees, ensuring strong semantics for locations accessed by non-racy instructions, provide a fruitful methodology for modular reasoning in relaxed memory concurrency. We observe that standard compiler optimizations are in inherent conflict with such guarantees in general fully-relaxed memory models.Nevertheless, for a certain strengthening of the promising model by Lee et al. that only excludes relaxed RMW-store reorderings, we establish multiple useful local data-racefreedom guarantees that enhance the programmability aspect of the model. We also demonstrate that the performance price of forbidding these reorderings is insignificant. To the best of our knowledge, these results are the first to identify a model that includes the standard concurrency constructs, supports the efficient mapping of relaxed reads and writes to plain hardware loads and stores, and yet validates several local data-race-freedom guarantees. To gain confidence, our results are fully mechanized in Coq.

The leaky semicolon: compositional semantic dependencies for relaxed-memory concurrency

“…In particular, we have only a pen-and-paper proof showing that PwT-mca supports optimal lowering to Arm8. The same is true for local data race-freedom (ldRf-sc)-additionally, our proof sketch for ldRf-sc elides Rmws, which have caused complications in other models [Cho et al 2021].…”

“…Jagadeesan et al [2020] argue that the speculative models allow too many executions, resulting in a failure of temporal reasoning and potentially jeopardizing type safety and other security properties. In a similar vein, Cho et al [2021] argue that local dRf guarantees are violated when read-introduction is followed by if-introduction, branching on the read just introduced. These optimizations are validated by the speculative models-Cho et al manage to avoid the problem by adopting a sub-optimal lowering for Rmws.…”

“…Others have strong dependencies that disallow compiler optimizations and efficient implementation, typically requiring fences for every relaxed read on Arm8 [Boehm and Demsky 2014;Dolan et al 2018;Jeffrey and Riely 2016;Lamport 1979]. Many of the most prominent models are operational models based on speculative execution [Chakraborty and Vafeiadis 2019;Cho et al 2021;Jagadeesan et al 2010;Lee et al 2020;Manson et al 2005].…”

“…This is a non-starter for safe languages like Java and Javascript, and may be an unacceptable cost for low-level languages like C11. • Semantic dependencies [Chakraborty and Vafeiadis 2019;Cho et al 2021;Jagadeesan et al 2010;Lee et al 2020;Manson et al 2005]. These models compute dependencies operationally using alternate worlds, making it impossible to understood a single execution in isolation; they also allow executions that violate temporal reasoning (see §9).…”