Heterogeneous memory management meets EXPORT_SYMBOL_GPL()

And while in the thread they were talking about the way it interacts with kernel internals as being a guide to deciding whether it should be EXPORT_SYMBOL_GPL or not, I think that doesn't align well with a reasonable copyright-informed interpretation.

I feel a better question to guide that decisionmaking might be "To what degree are the concepts and interfaces linux-specific?". This also aligns somewhat with Linus's stance that driver code written for other OSes and ported to Linux doesn't de facto derive from Linux (https://lwn.net/Articles/13398/).