Inferring Annotations For Device Drivers From Verification Histories
Finding invariants is an important step in automated program analysis. Discovery of precise invariants, however, can be very difficult in practice. The problem can be simplified if one has access to a candidate set of predicates (or annotations) and the search for invariants is limited over the space defined by these annotations. We present an approach that infers program annotations automatically by leveraging the history of verifying related programs. Our algorithm extracts high-quality annotations from previous verification attempts, and then applies them for verifying new programs. We present a case study where we applied our techniques to Microsoft’s Static Driver Verifier (SDV). SDV currently uses manually-tuned heuristics for obtaining a set of annotations. Our techniques can not only replace the need for this manual effort, they even out-perform these heuristics and improve the performance of SDV overall.