Analysis of Execution Behavior for Testing of Multi-Tasking Real-Time Systems

An important issue in software testing is the ability to observe the execution of the software; this is especially true for real-time systems (RTS). RTS are difficult to observe, and the ability to test RTS is inherently low. Embedded RTS have few interfaces for observation and the execution of multi-tasking RTS is usually non-deterministic. As a consequence, testing of RTS cannot be exercised with existing tools for sequential programs. New tools and methods are necessary that enable observation of the system despite few interfaces while at the same time address the non-determinism issue. The contribution in this thesis is three-folded: (1) we present a tool suite that allows deterministic testing of multi-tasking RTS, in which synchronization of tasks is resolved off-line or on-line. (2) We show by building a test bed how to use the tool suite. (3) We present the design and functionality of Asterix the Real-Time Kernel. In (1) we propose an analysis tool that derives all possible system level control-flow paths of multi-tasking RTS in which synchronization between communicating tasks are resolved on-line by using the Priority Ceiling Emulation Protocol (PCEP; also know as the Immediate Inheritance Protocol). The analysis tool is an extension of an existing tool in which synchronization were resolved off-line by using release time offsets or priorities to separate the tasks in time. Based on the number of derived control-flow paths test coverage criteria are defined, and estimations of test effort can be done early in the development of a system. In (2) we show how the defined test coverage criteria relate to the number of traversed control-flow paths during test execution. We also show how the estimation of tasks’ execution times affects the analysis. The analysis tool is applied on multi-tasking RTS in which the tasks are synchronized offline. The real-time applications are then exercised on the test bed using Asterix as the operating system. In (3) we present a small-sized real-time kernel named Asterix that has support for software based instrumentation of kernel events as well as application usage of system calls. The major problem of software instrumentation is the change in execution behavior that occurs when a RTS is executed without or without the probes. In Asterix we avoid this probe-effect by leaving the probes in the kernel during normal operation. Also, we present a literature survey covering the state-of-the-art in the field real-time systems testing. To Andreas and Emelie