PhD defence Afshin Amighi

Specification and verification of synchronisation classes in java - a practical approach 

Digital services are becoming an essential part of our daily lives. To provide these services, efficient software plays an important role. Concurrent programming is a technique that developers can exploit to gain more performance. In a concurrent program several threads of execution simultaneously are being executed. Sometimes they have to compete to access shared resources, like memory. This race of accessing shared memories can cause unexpected errors. Programmers use synchronisation constructs to tame the concurrency and control the accesses. In order to develop reliable concurrent software, the correctness of these synchronisation constructs is crucial. 

In this thesis we use a program logic, called permission-based Separation Logic, to statically reason about the correctness of synchronisation constructs. The logic has the power to reason about correct ownership of threads regarding shared memory. A correctly functioning synchroniser is responsible for exchanging a correct permission when a thread requests access to the shared memory. We use our VERCORS verification tool-set to verify the correctness of various synchronisation constructs.

In Chapter 1 we discuss the scope of the thesis. All the required technical background about permission-based Separation Logic and synchronisation classes is explained in Chapter 2. In Chapter 3 we discuss how threads' start and join as minimum synchronisation points can be verified.

To verify correctness of the synchronisation classes we have to first specify expected behaviour of the classes. This is covered in Chapter 4. In this chapter we present a unified approach to abstractly describe the common behaviour of synchronisers. Using our specifications, one is able to reason about the correctness of the client programs that access the shared state through the synchronisers.

The atomic classes of java.util.concurrent are the core element of every synchronisation construct implementation. In Chapter 5 and Chapter 6 we propose a specification for atomic classes. Using this contract, we verified the implementation of synchronisation constructs w.r.t to their specifications from Chapter 4. In our proposed contract the specification of the atomic classes is parameterized with the protocols and resource invariants. Based on the context, the parameters can be defined.

In Chapter 7 we propose a verification stack where each layer of stack verifies one particular aspect of a specified concurrent program in which atomic operations are the main synchronisation constructs. We demonstrate how to verify that a non-blocking data structure is data-race free and well connected. Based on the result of the verification from the lower layers, upper layers can reason about the functional properties of the concurrent data structure.

In Chapter 8 we present a sound specification and verification technique to reason about data race freedom and functional correctness of GPU kernels that use atomic operations as synchronisation mechanism.

Finally, Chapter 9 concludes the thesis with future directions.