The HW apparently has some issues (or at least a much more complicated VCM
calculation) with non-combined segments, and the closed source driver also
uses combined I/O.  Until I get the last CTS failure resolved (which does
look plausibly like some VPM stomping), let's use combined I/O too.