HeapMon: A helper-thread approach to programmable, automatic, and low-overhead memory bug detection

Abstract

The ability to detect and pinpoint memory-related bugs in production runs is important because in-house testing may miss bugs. This paper presents HeapMon, a heap memory bug-detection scheme that has a very low performance overhead, is automatic, and is easy to deploy. HeapMon relies on two new techniques. First, it decouples application execution from bug monitoring, which executes as a helper thread on a separate core in a chip multiprocessor system. Second, it associates a filter bit with each cached word to safely and significantly reduce bug checking frequency—by 95% on average. We test the effectiveness of these techniques using existing and injected memory bugs in SPEC®2000 applications and show that HeapMon effectively detects and identifies most forms of heap memory bugs. Our results also indicate that the HeapMon performance overhead is only 5%, on average—orders of magnitude less than existing tools. Its overhead is also modest: 3.1% of the cache size and a 32-KB victim cache for on-chip filter bits and 6.2% of the allocated heap memory size for state bits, which are maintained by the helper thread as a software data structure.