Re: bk 2.4.1pre2 Lombard PCI allocation fails.

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

>Fixed by the kernel, and reassigned to 0x10000000.
>
>> BootX uses the MacOS PCI mappings which are bogus for the Rage LT Pro.
>> Though I'm certain Geert's old PCI resource allocation patch worked fine
>> in spite of this. There was some message about resource conflict but that
>> was fixed by the kernel.
>>
>> What's the PCI resourced for the card after the kernel has finished
>> booting?
>
>I expect it to be 0x10000000.
>
>BTW, this also means we can start using the secondary aperture as well in
>2.4.0. Gives us an additional 4 kB of frame buffer on little-endian boxes.

Well, unfortunately, 0x10000000 is bogus on most cases. (I didn't check
with Grackle, but for example on UniNorth based machines, this address is
not routed to PCI).

Our PCI code still have no knowledge of which physical ranges can
actually be allocated for each bus. That should be implemented via proper
root bus resources, but I've not yet been able to find a good solution
for that (mostly because we would need more resources than the current
pci_bus structure provides in some cases where the bridge can decode
several discontiguous regions). Maybe We could fake this, my knowledge of
the linux resource mecanism is not perfect.

Matt Porter is coming with a better solution based on some platform
callbacks that would take care of such things are re-assigning a given
device. Those callbacks could then take care of doing proper allocation
depending on the host bridge setup. (See our discussion about the enable
device hook)

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/