NOTES:

Okay, here's what's still broken:

Getting kernel panics in calls to IOMalloc after a while.  I don't think
we're being a bad citizen, AFAICT.  It's just doing the whole 0x300 thing.
Might be bad hardware, I suppose.  Oddly enough, this only started when
I added write support.


WARNINGS (that might be mistakes):
- Need to include headers for kvtophys, delay, flush_dcache, invalidate_dcache, IOMalloc, sleep, wakeup in order to verify usage....
- Verify use of invalidate_dcache in place of invalidate_cache_for_io.
- warning: passing arg 2 of dbdma_start from incompatible pointer type
- crusty warnings in fd.c