Abstract EN:

The main characteristic of hard real-time systems is that they must guarantee a correct timing behaviour. Schedulability analysis methods are commonly used in hard real-time systems to check whether or not all tasks deadlines will be met. Most of them rely on the knowledge of an upper bound on the computation time of every task, named WCET (Worst-Case Execution Time). The WCET of a program can be computed by simulation or by performing a static analysis. Dynamic analyses give the actual WCET of a program if we can simulate all possible combinations of input data values and initial system states. Which is clearly impractical due to the exponentially number of simulations required. As a result, we compute an estimate of the actual WCET by performing a static analysis of the program despite of the pessimism generated by the approximations. Most of these analyses are performed at the task so they don't take advantage from the features of the multi-tasking real-time systems such as the tasks' chaining that affects straightforwardly the accuracy of the WCET estimation. We propose an approach that studies the instruction cache behavior of a static tasks scheduling for a single processor multi-tasking real-time application assuming that no preemption is allowed between and inside the tasks. The main goal consists of replacing the conservative approximations that consider an empty or undefined cache state before the task execution, by an abstract cache state. The WCET estimation is thus improved. We also present a free real-time benchmark, PapaBench. This benchmark is designed to be valuable for experimental works in WCET computation and may be also useful for scheduling analysis.