Add code to allocate extra space for java heap (either 1 page or alignment) and unprotect so that implicit null exceptions at heap_base + 1 page fault.
This fix has been checked in for both hs13 (snapshot for b03) and in hs14.
The ultimate fix should be in runtime, in the short term we may want to have some variant of -UseImplicitNullChecks for narrow oops in C2.
Implict null checks under CompressedOops are not triggering a change on control flow. It appears that the heap_base+1 page is still writeable, so the generated code does not SEGV as expected.