Remember, I'm not a wizard on this area.

I think the bug is that when DMA crosses a page boundry, it sometimes 
does something bad.  It doesn't matter if the DMA is from a 32 or 
64 bit device.

The PCI-PCI bridge breaks transfers at page boundaries so that prevents 
the bug.  I think only the 2 64 bit slots are not behind the bridge. 

I think the firmware knows that some cards don't provoke the problem.  
If it finds anything else in either of the two 64 bit slots, it complains.

I think there is some command to disable that complaint.  I don't 
know if that's wise with your card.

That's all on a first generation Miata.  On the second generation, 
they did some hardware workaround.

-----

I don't know how the slots are numbered.  There is a lot of info 
out on the web at: 
  http://www.digital.com/alphaserver/workstations/auseries/index.html

I think one of the big PDF files has a picture and tells you which 
slot is which.  (Ugh.  The service guide is self unpacking file, 
for an Intel machine, of course.) 

I found an old hardcopy printout.  Slot 4 is the next to the bottom, 
one of the 64 bit slots. 

So try moving your card to another slot.