It seems hotspot may occasionally generate a special dynamic code that
is supposed to cause an exception (like SEGV or ILL) first time it's
executed to bring JVM's attention to it.
When used with collector, it may happen that a profiling signal is
delivered right at the moment when another signal (SEGV) is being
handled. AsyncGetCallTrace(), being called from the PROF handler,
fails to unwind the Java stack and reports a fatal error.
See Comments for the entire stack.