Bounding the execution times of DMA I/O tasks on hard-real-time embedded systems

A cycle-stealing DMA I/O task proceeds by stealing bus cycles from the CPU. The execution time of the DMA I/O task depends on the sequence of CPU instructions executing concurrently with it. This paper presents a method for bounding the worst-case execution time of a cycle-stealing DMA I/O task executing concurrently with a set of CPU tasks on a single-processor system. Our method uses the dynamic-programming technique to minimize the computational cost. We conducted exhaustive simulations on a widely-used embedded controller. The experimental results demonstrate our method safely and tightly bounds the worst-case execution times of cycle-stealing DMA I/O tasks.