Bug#461649: Image shakes and shivers when content changes with Radeon RV100 QY (Radeon 7000/VE)

["Stephen Benoit (Linux)" <linux@magnumline.com>]

Brice Goglin wrote:
> Stephen Benoit (Linux) wrote:
>> I tried the git approach, but the autogen.sh script requires
>> xorg-server>=1.2 and libdrm>=2.2 which are more recent than etch.
>>
>> I did not find any xserver-xorg backports for etch, so I will try
>> upgrading xorg to unstable.  I'll report back when it's installed
>> and I can try it out.
>>
>
> Actually, no need to upgrade all the way to unstable.
>
> Just upgrade xserver-xorg-core, and some dependencies such as
> xserver-xorg-dev, libdrm and so to testing/lenny. Building ati from git
> should work then. I will be much easier than upgrading to unstable.

Hi, Brice.

I tried the minimal upgrade to get xserver-xorg-dev to unstable/sid (Maybe
I should have tried testing/lenny).  To satisfy dependencies, I ended up
upgrading xserver-xorg to unstable (7.3+10) with xserver-org-core
(1.4.1~git20080118-1).

1. Tests under xserver-xorg unstable:

At first, I had the shaking/shivering problem under load, but it was less
pronounced (about 1-2 pixel horizontal movement).  Settings that seem to
help:

Option "SWCursor" "on"
Option "AGPMode" "1"
...
Section "Extensions"
    Option "Composite" "disable"
EndSection

With earlier tests I could reproduce the problem instantly.  Now it is not
consistent.  I use an aterm with pseudo-transparency (uses piece of the
root Pixmap as a window background).  I could run my test for several
minutes before the shaking starts.  If I use xrandr to change resolution
while the test is running, it could either stop or restart the shaking.

Overall: this setup is better for my needs.  For some reason my armagetron
dies when starting a match, but I can live without that ;).

2. Trying the git version of xf86-video-ati:

Building from source required the xorg macros:
% git-clone git://anongit.freedesktop.org/git/xorg/util/macros

Now that the source building dependencies are satisfied, The autogen.sh
script works fine, but the CARD64 type used in radeon_crtc.c was
undefined.  Combing through /usr/include/{X11,xorg} this seems to be
conditionally defined in X11/Xmd.h.  My solution?  "typedef unsigned long
CARD64;" in radeon_crtc.c.  Make, manually install *_drv.so, restart xorg,
and, of course, the new driver could not change screen resolution anymore
and was convinced that all the valid modelines were now of the form
"1280x1024"x0.0 and so forth.

I assume I am missing a header file.

3. I will continue breaking in xorg 1.4.1 with xserver-xorg-video-ati
6.7.197-1.

Drat.  I just ran glxgears while my random character generation test was
running.  The screen started shaking all over again, even after I killed
glxgears.  I think the pseudo-transparency of aterm is creating a lot of
traffic over AGP.  After pausing the scrolling characters and moving the
aterm window, the shaking stopped -- even after resuming the scrolling.

Thanks for your assistance, if you have suggestions, I welcome them!

Best regards,
Stephen Benoit
linux@magnumline.com