Memory corruption is one of the most common software failures. For sequential software and multi-tasking software with synchronized data accesses, it has been shown that program faults causing memory corruption can be detected by analyzing the relations between defines and uses of variables (DU-based testing). However, such methods are insufficient in preemptive systems, since they lack the ability to detect inter-task shared variable dependencies. In this paper, we propose the use of a system level shared variable DU analysis of preemptive multi-tasking real-time software. By deriving temporal attributes of each access to shared data using WCET analysis, and combining this information with the real-time schedule information, our method also detects inter-task shared variable dependencies.