Radeon 9000 Pro corruption problem on PPC when DRI is enabled

[Roland Scheidegger]

Ari Entlich wrote:
> Hello,
> 
> I am writing to ask about a video corruption problem I've been trying to
> sort out. The problem arose after I upgraded to X.Org 7.1 from X.Org
> 6.8.2, which didn't have this problem. It only arises when DRI is
> enabled - changing anything to make DRI not start produces correct
> output. I used gentoo's ppc Xorg autoconfig utility (Xorgautoconfig) to
> generate my configuration, which can be found at
> http://home.twcny.rr.com/fandaswebsite/xorg.conf. The only change I made
> to that configuration was to enable UseFBDev, because my monitor acts
> like it's not getting a signal otherwise, for some reason. I do not
> think this is a monitor issue however because, as you might or might not
> be able to see in the pictures, the cursor is unaffected. One thing
> which I noticed was that changing the bit depth changes how the
> corruption looks. Log files with DRI both on and off at bit depth 16
> (there are no significant differences between the 16 bit and 24 bit log
> files) can be found at
> http://home.twcny.rr.com/fandaswebsite/xorg_log_16 and
> http://home.twcny.rr.com/fandaswebsite/xorg_log_no_dri_16. The pictures
> for 16 and 24 bit depths can be found at
> http://home.twcny.rr.com/fandaswebsite/bitdepth_16.jpg and
> http://home.twcny.rr.com/fandaswebsite/bitdepth_24.jpg (sorry about the
> low image quality - they were taken with a digital camcorder...).
Could be some pitch problem. You mentioned you used UseFBDev, you didn't 
use that with Xorg 6.8.2?
I'm not sure where the requirements for pitch are exactly coming from 
(except when you use color tiling, which you can't due to the same 
UseFBDev option), but maybe alignment is wrong somewhere when you're 
using the cp accel path instead of mmio (which is what you get without 
dri). Alex might know more about this.
You might try to get rid of that (imho ugly and utterly useless - well 
the idea is sound the implementation is not because way too limited) 
UseFBDev option, that might fix things. Maybe you need to specify a 
different modeline, I guess there was nothing in the log which indicated 
anything wrong when UseFBDev was not used?

Roland