Thu Aug 16 10:30:49 EDT 2018

Smaller step

SPI works, but writing to buffer, then executing, doesn't work.  What
about making this simpler: add readout first, and see what that gives.

Another thing is to clock the CPU from SPI, and pulling out debug
information that way.

Reading is not expensive: it is just routing.  Writing is expensive:
it requires multiplexing at the input of a register.

So what about adding a tracer for registers?  I have the probe
mechanism, so a state machine could just go through all of this.