(I'll be posting several more when I remember all subtopics)

Have any CPUs split the MMU in a separate instruction and data MMU? Any immediate pros and cons for doing this?
(I was thinking at 2x the size or number of mmu entries with hopefully no extra slowdown, and possibly tuning mapping sizes to fit the characteristics of either.)

Have any CPUs used MMU status bits to define per-page endianness, priviledge level (ring) or priviledge escalation permission? (or any related ideas)

I believe some versions of the MIPS processors use multiple TLB's to support independent data and code lookups. There's a TLB for code, a TLB for data, then a thrid joint TLB (micro TLB) that is shared for both code and data.

I wonder if there's a processor out there that has more independent MMU's for things like stack, and I/O.

I was just working on code for a 6829 MMU look-a-like teh other day.

Most processors which have an MMU have separate instruction and data TLBs.

Toshi