G5 pci probing regression

[Paul Mackerras]

Christoph Hellwig writes:

> sure.  The -unpatched files are plain 2.6.14-rc1, the -patched ones are
> with my little hack.

Thanks.  This would be the problem:

--- proc-pci-patched    2005-09-20 08:08:23.000000000 +1000
+++ proc-pci-unpatched  2005-09-20 08:08:33.000000000 +1000
@@ -28,18 +28,19 @@
     Class 0100: PCI device 1000:0054 (rev 0).
       IRQ 52.
       Master Capable.  Latency=16.  Min Gnt=64.Max Lat=10.
-      I/O at 0x400 [0x4ff].
-      Non-prefetchable 64 bit memory at 0x90004000 [0x90007fff].
-      Non-prefetchable 64 bit memory at 0x90020000 [0x9002ffff].
+      Non-prefetchable 32 bit memory at 0x400 [0x4ff].
+      Non-prefetchable 32 bit memory at 0x90004000 [0x90007fff].
+      Non-prefetchable 32 bit memory at 0x90020000 [0x9002ffff].

The first resource being shown as memory is just a bug.  I'm wondering
how I tell 32-bit BARs from 64-bit BARs in the device tree, though.
Could you tar up your device tree (or at least the directory for this
device) and send it to me?

Thanks,
Paul.