A licensee using the JDWP stub on an OS with a port of J2SE1.2.2
with HotSpot 1.0.1 encounters a crash during the initialization
of the JDWP stub.

On investigation it turns out that an area malloc'ed by the VM
with malloc cushions is being deallocated by the JDWP without
knowledge of the cushions - the results is a 'free' with a
pointer that differs from the original 'malloc'.

This only happens with the debug version of HotSpot 1.0.1
(jvm_g.dll) and it happens because the JVMDI implementation

See suggested fix section of this bug for fix.