On the Detection of Synchronization Errors

Concurrent systems, either hardware or software, are notoriously difficult to design correctly in large part due to the complexities of nondeterministic execution. A concurrent system can perform many different correct computations for a given input sequence because the absolute order of execution is dependent on factors which cannot be known at design/compile time. Synchronization constructs are used to restrict the set of possible computations to only correct computations, but insertion of synchronization constructs is a manual and error-prone task. The detection of synchronization errors is made difficult because the manifestation of an error can depend on operation timing which can change between executions. We define a class of synchronization errors and define the timing requirements to ensure the detection of these errors. We provide a coverage metric which can be used to determine whether or not a given test execution is sufficient to detect the defined class of synchronization errors.