When using ppc64, the screen to screen copies were corrupted.  I tried 6.8, then
6.9, then 7.0 (modularized).  I debugged the nvidia driver, as it appears to be
a problem with endian-ness in the XAA architecture screen to screen copy
operation.  I found in fact that the nv driver 1.0.5 is using DMA operations to
execute the copy.  Is this a hardware bug, or is there some way I can help?

Moving the window vertically one step causes the data to endian-nize.  Moving
the window vertically one step more causes the data to recorrect itself which
would be expected of an endian problem.  Moving a window horizontally causes all
kinds of havok.

I am a kernel developer and can also develop in userland, so if you have ideas
to fix this problem, I'd appreciate them.  I can also test patches.

Is there a mechanism to tell the nvidia driver which endian it is operating
under in the dma operation? (I dont have docs, heh u probably dont either :(

Window copies without acceleration are painfully slow on my dual g5.

The rest of 7.0 via gentoo appears to work on ppc64 though - good job guys.
Regards
-steve