Dynamic synthesis for relaxed memory models

Modern architectures implement relaxed memory models which may reorder memory operations or execute them non-atomically. Special instructions called memory fences are provided, allowing control of this behavior. To implement a concurrent algorithm for a modern architecture, the programmer is forced to manually reason about subtle relaxed behaviors and figure out ways to control these behaviors by adding fences to the program. Not only is this process time consuming and error-prone, but it has to be repeated every time the implementation is ported to a different architecture. In this paper, we present the first scalable framework for han-dling real-world concurrent algorithms running on relaxed archi-tectures. Given a concurrent C program, a safety specification, and a description of the memory model, our framework tests the pro-gram on the memory model to expose violations of the specifica-tion, and synthesizes a set of necessary ordering constraints that prevent these violations. The ordering constraints are then realized as additional fences in the program. We implemented our approach in a tool called DFENCE based on LLVM and used it to infer fences in a number of concurrent al-gorithms. Using DFENCE, we perform the first in-depth study of the interaction between fences in real-world concurrent C programs, correctness criteria such as sequential consistency and linearizabil-ity, and memory models such as TSO and PSO, yielding many in-teresting observations. We believe that this is the first tool that can handle programs at the scale and complexity of a lock-free memory allocator