Software-Based Online Monitoring of Cache Contents on Platforms without Coherence Fabric

In favor of smaller chip areas and associated fabrication costs, designers of embedded multi-core systems on occasion decide not to include cache coherence logic in the hardware design. However, handling all cache coherence exclusively in software is error-prone, and there are presently no tools supporting developers in this task. Thus, we propose a new software testing method, based on online inspection of the cache contents, to pinpoint programming mistakes related to cache handling. This concept helps localizing the causing data symbol even for complicated cache handling errors, e.g. where the causing and manifesting code-location of an error differ. Our solution is a pure software solution and does not require any specialized hardware. We evaluate our approach by using it in a large application, and show that we can detect typical cache-related errors.