[erlang-questions] How does code:purge/1 know to kill processes?

[Roger Lipscombe]

The documentation for code:purge/1
(http://erlang.org/doc/man/code.html#purge-1) says "If some processes
still linger in the old code, these processes are killed before the
code is removed."

How is "lingering" defined?

Does the VM kill the process when it attempts to invoke the old module?
Does the VM walk the stack for all processes and kill those that have
the old code somewhere in their call stack?
Something else?