6345154 illustrated a problem where a user's library (actually, NVidia's Cg library) was using signal handlers internally but apparently attempting to be clever and resetting the signal handler state so it appeared from the JVM's crash logs that only HotSpot's internal signal handlers were set up. However, it appears that the Cg library may have cleared the SA_SIGINFO flag on one or more of the signal actions before passing along the signal to HotSpot's internal signal handlers, so that no ucontext was available (which was the root cause of the failures in 6345154). When dumping the hs_err log file, it may be worth testing to see whether the sa_flags are equal to the value set by HotSpot (in particular, the presence of the SA_SIGINFO flag) and printing a warning if any do not match.