The last change in iommureg.h that worked around a problem in
microsparc II CPUs was apparently responsible for occasional
lockups of microsparc I CPUs.

I've changed the patch to insert three `nop's after writing to an
IOMMU register (instead of reading back the register content) which
seems to work Ok on my SS4.

Please get back to me if any of your microsparcs start failing again.

-pk