Problems have been encountered/realized with the practice of
unconditionally setting the multifunction bit for all functions of all
devices. PATCH 2/2 remedies that (details in its own commit
comment). PATCH 1/2 is a one-liner I fixed in the meantime which will
cause a simple conflict if it's not applied to other branches at the
same time as PATCH 2/2, so I'm sending them together.