> > some more thinking led me to come to think of the bus_space_barrier=
()
> > function and that it might be required between the two bus_space_re=
ad_2
> > calls in quick succession in isp_pci_rd_debounced().  So, modifying=
 the
> > diff as shown below moved the problem to somewhere else...
>
> That shouldn't have done anything. The Alpha bus_space operations all=

> still have implicit barriers.

Hm, that's sort of a bummer.  Then I think I'm out of ideas as to
what is the actual source of the problems I'm seeing.

Regards,

- H=E5vard