cgd writes:

>last i heard, the way to do the right thing there was to provide MD
>read and write register hooks (which may in turn use a bus space tag,
>or whatever) to insert delays as appropriate?

That's what the patch I posted did. (at least it did last,
time it compiled).  So, if I:

	* added a bus-space tag to the attach args and softc,
	* passed that bus-space tag to the register-access macros
        * added an MI declaration for the pdma callback struct

then we'd be OK, or at least no worse than zs and LANCE, right?

As i said to Jason, i have other commitments for the rest of this
week...