[RV730] xorg-edgers: Mist of artifacts in highlight color over menus and buttons

This bug seems fixed with fresh git master drm/mesa/glx and drm-next kernel. but only with KMS/DRI2 enabled. The corruption is still there without mode setting.

Also, xorg-edgers/radeon doesn't work for for me anymore. Everything GL froze in drmCommandWrite() since a few day ago. Perhaps an outdated drm-modules-source 2.4.3+git20090831+r6xx-r7xx-3d.8e297c0a-0ubuntu0tormod4?

Yes, the drm-modules-source might be outdated. I will update them soon, since I see that agd5f's tree has got some updates. But it is probably better to use a kernel with drm-next. I have no time to look at a ppa kernel, but I have asked the kernel-team about it.

I have updated the drm-modules-source in the radeon PPA.

Thank you, Tormod! The app-freezes with the PPA are gone, only for some reason the dri device isn't created automatically. I have to mknod /dev/dri/card0 c 226 0 && chmod 666 /dev/dri/card0. I probably brought this onto myself though. Still investigating.

The dri device should be created by udev (/lib/udev/rules.d/50-udev-default.rules). Try "grep dri/ /var/log/udev" which should list:
DEVNAME=dri/card0
DEVNAME=/dev/dri/card0

This is odd, grep dri/ /var/log lists nothing because the device is created as /dev/card0 instead of /dev/dri/card0

I'll attach the output of
sudo udevadm test /class/drm/card0

Is udev supposed to add /dri to the path or should DEVNAME below be dri/card0?
cat /sys/class/drm/card0/uevent
MAJOR=226
MINOR=0
DEVNAME=card0

Anyway, I've added
SUBSYSTEM=="drm", SYMLINK{unique}+="dri/$kernel" to some rules file
and now I have the device /dev/card0 and a symlink to /dev/dri/card0 and dri works well. Really well actually, except for this little bug.

Great debugging there, marmuta. On my system I see this in the udevadm output:
udev_event_execute_rules: no node name set, will use kernel supplied name 'dri/card0'
while you get:
udev_event_execute_rules: no node name set, will use kernel supplied name 'card0'

I know the old Jaunty udev rules would add "dri/" explicitly.

This is something that has changed in the main kernel (and in the udev rules) but not yet in the drm-modules-source.

> I know the old Jaunty udev rules would add "dri/" explicitly.
>
> This is something that has changed in the main kernel (and in the udev
> rules) but not yet in the drm-modules-source.
That explains it, thanks Tormod.
I've dug up the Jaunty udev rule. It works better than my hack, with only the device being created at the right location.

KERNEL=="card[0-9]*", NAME="dri/%k"

This seems fixed in lucid, didn't find any artifacts there.