On Thu, Dec 07, 2017 at 09:33:29AM -0500, Theodore Ts'o wrote: > Let me ask a potentially stupid question, Can someone give an example > of the problem this completion lockdep is actually trying to *catch*. > The lockdep message is sufficiently confusing it's not even clear what > *good* it is in the first place. lock(A) lock(A) wait_for_completion(C) complete(C); > Is there an example of a bug that this lockdep feature hasf actually > caught in real life, as opposed to false positives? Yes, it caught a fair bunch of real deadlocks in cpu hotplug, watchdog, and other places I can't remember. Here's one in the mm: https://lkml.kernel.org/r/20171030152200.ayfnewoqkxbuk4zh@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx I have at least one open issue in perf that I've no real idea what to do about: https://lkml.kernel.org/r/20171027153336.GC3857@worktop Yes lockdep is a pain...
