JavaThread::pd_get_top_frame_for_profiling may fail if the PC is at arbitrary locations

In the review of JDK-8145521, Dean raised the issue whether the profiler would fail if the PC happens to land on the trampoline (which lives in the CDS archive "MC" section)

http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2016-February/018106.html

In general, it's not clear whether the profiler can handle cases when the PC lands in exotic locations, for all CPU/OS combinations.

This bug is filed so we can investigate the current implementation to see if anything needs to be fixed to make the profiler more robust.