Are red-black trees really the best ordered-map mechanism that the kernel hotshots can manage?

Well, that's another advantage to using the big macro in this case. My primary motivation when I started this was to completely eliminate the need for an rbtree user to muck with any of the implementation details. You should treat it as an container. Luckily, I seem to have managed that without incurring a run-time abstraction overhead (I wont absolutely know for certain until I get some good performance tests that compare hand-written search & insert with the generic).

Just remember: abstraction doesn't have to incur a run-time overhead. If done correctly, you can take it out of the compiler. I guess luck doesn't hurt sometimes as well (i.e., having an interface that can be mutated).