Hi,

How about fixing this in the kernel's PCI bus structures, by adding any
required offsets in the pci_fixup() routines? This would make for clean
inb()/outb() definitions (no offset), and also accomodates multiple I/O
spaces behind more than one bridge. For instance, on the 7x00, both
bandit and chaos have I/O space; the one for the PCI cards, the other
for control (yes, control has an I/O port range! Not sure it serves,
though...).

Any fixed-offset approach would break with multiple I/O ranges.... Not
to mention the 9x00 machines (two separate host bridges, each three
slots), where you can't work around the problem by using hardcoded
addresses for the known motherboard chips.

I guess this approach could be made to work across the PPC
architecture, not only on the PMac, as the PPC in general lacks anything
special about I/O space... after all, those are just small memory
spaces ;-)

   Looks okay to me.  I think it's what Sparc people are already doing
on Ultra/AX.

   If you can, please wait a few days for me to merge my new PCI stuff
with Linus, it should make these fixups easy as I my current code leaves
control of the resource allocation fully to the arch-dependent code.

				Have a nice fortnight
-- 
Martin `MJ' Mares   [off-list ref]   http://atrey.karlin.mff.cuni.cz/~mj/
Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth
"...take my advice, I don't use it anyway."

[[ This message was sent via the linuxppc-dev mailing list.  Replies are ]]
[[ not  forced  back  to the list, so be sure to Cc linuxppc-dev if your ]]
[[ reply is of general interest. Please check http://lists.linuxppc.org/ ]]
[[ and http://www.linuxppc.org/ for useful information before posting.   ]]