Formal Modelling, Testing and Verification of HSA Memory Models using Event-B
This addresses the challenge of ensuring correct implementation of weakly consistent memory models in multi-core systems, though it is incremental as it applies existing formal methods to a specific domain.
The authors tackled the ambiguity in HSA memory model specifications by developing a formal model using Event-B, enabling provably correct refinements and automated validation against litmus tests, with a coverage-driven method to extract tests for hardware and software verification.
The HSA Foundation has produced the HSA Platform System Architecture Specification that goes a long way towards addressing the need for a clear and consistent method for specifying weakly consistent memory. HSA is specified in a natural language which makes it open to multiple ambiguous interpretations and could render bugs in implementations of it in hardware and software. In this paper we present a formal model of HSA which can be used in the development and verification of both concurrent software applications as well as in the development and verification of the HSA-compliant platform itself. We use the Event-B language to build a provably correct hierarchy of models from the most abstract to a detailed refinement of HSA close to implementation level. Our memory models are general in that they represent an arbitrary number of masters, programs and instruction interleavings. We reason about such general models using refinements. Using Rodin tool we are able to model and verify an entire hierarchy of models using proofs to establish that each refinement is correct. We define an automated validation method that allows us to test baseline compliance of the model against a suite of published HSA litmus tests. Once we complete model validation we develop a coverage driven method to extract a richer set of tests from the Event-B model and a user specified coverage model. These tests are used for extensive regression testing of hardware and software systems. Our method of refinement based formal modelling, baseline compliance testing of the model and coverage driven test extraction using the single language of Event-B is a new way to address a key challenge facing the design and verification of multi-core systems.