>  > Any ideas on the right way to fix this?  If this is a common
>  > problem, I suppose we could define a quirk for turning on debugging
>  > for some class of cards.
> 
> If turning on debugging makes the card work, this suggests a timing
> problem, yes?

Sure.  

Sticking a delay before the debug printf in wdccommand_intr during
the attach seems to work equivalently.  When I have some more time,
I can dig into it further, but if others have seen similar problems,
maybe there's a common problem.