JMXInterfaceBindingTest filter out loopback addresses: I suspect it's a bug that was added to cover up another bug (which is that the test doesn't work) - and the effect of that might have simply caused the test to be skipped (because it would then find only 1 address which is not the loopback).
I was surprised to see that the -Djava.rmi.server.hostname system property was required, even for 9.
We should fix the test to do something useful and test the loopback interface.