In JDK 11 only, we see errors in runtime/ErrorHandling/ErrorHandler.java. The test induces an artificial (but real) segfault and checks the VMs reaction to it (that the hs-err file is written correctly and contains expected output).

The test does this by reading from `char* p = NULL;`. This is not guaranteed to work, and does not in this case: the process gets a "BPT/Trap" message instead and dies.

This code (`VMError::controlled_crash()`) has been streamlined and rewritten for later releases, and there we consistently use a known non-null, volatile address to invalid memory instead, which seems to work reliably. We should do this in 11 too (without having to downport the reworkings).