|
From: Bart V. A. <bar...@gm...> - 2008-02-18 20:01:24
|
One of the criticisms on happens-before data race detection tools like Intel Thread Checker and exp-drd is that these tools can miss some violations of a locking discipline. Small examples that illustrate this are well known, and are a.o. published in the original paper about the Eraser algorithm. Since some time I am wondering what the chance is that a happens-before detector misses a locking discipline in a realistic program. Does anyone know whether any papers have been published on this subject ? Some authors argue that data that is shared over threads and protected by mutexes is almost always accessed many times which makes the chance low that locking discipline violations go undetected. Bart. |