> >* Bounce buffers.  All of the other bus-mastering drivers need to be
> >updated.  Note also that on EISA machines, we don't need to bounce in
> >many cases, because there's an extra register for the upper 8 bits of
> >the DMA address.  (I haven't had time to look at the PR, so this may
> >have been addressed already.)
> 
> I would put it more strongly: *DON'T* bounce-buffer my EISA machine --
> I don't want to waste the CPU cycles and memory on it since my EISA
> bus accesses that memory nicely by itself.

If your EISA bus has an aha1542 on it, and you try to use that board
with 16M of memory, you'd BETTER have bounce buffer support...


> The point being: please don't break working EISA drivers with bounce
> buffers.

That's not what he was talking about.  Note, however, that EISA
machines still have to have support for mapping bus addresses to
host addresses and vice-versa, and the EISA drivers should be using
those rather than 'vtophys().'  System physical addresses aren't
necesarily the same as bus physical addresses.



cgd