I think I may have discovered a good clue to help track down
the "Everything dumps core" problem.  First, I've have not
seen this problem on the 3/260 in ages.  Second, on the 3/60
the problem disappears with si_obio_options set to zero.

Are these observations consistent with what others see?

My hunch at this point is that the DMA code for OBIO is not
correctly waiting for transfers to finish, leaving partial
reads (or something like that...)

Gordon

P.S.  I've separated si_options into:
  int si_obio_options; int si_vme_options;
where the values of those have the same effect as si_options
but only on the respective DMA code.