Trouble with Radeon KMS on Debian
So with the lack of newer X server support in fglrx, I decided to take the plunge and try out the open source drivers. However, I ran into a minor problem. When I enable KMS and start X, I get a blank black screen. Any ideas?
Thanks in advance.
I'm running Debian testing, have a Radeon HD4670 (ChipID = 0x9490).
X.org - X Server 1.7.5
Kernel - 2.6.32-trunk-686
Libdrm - 2.4.18
Radeon - 6.12.4
I can confirm KMS does work, if I stop X and goto a terminal, I can rmmod radeon, modprobe radeon modeset=1 and clearly see a KMS terminal. dmesg | grep drm shows [drm] radeon: Initializing kernel modesetting. However, when I then try to restart X, either by /etc/init.d/kdm start or startx, the screen goes blank and the machine becomes unresponsive. Booting with modesetting enables causes a similar problem - after grub I get no video output.
I believe this requires radeon ddx driver of at least version 6.13 or a git build. I could be wrong, but I dont think the KMS bits are in 6.12
Originally Posted by Tillin9
Yep, you want 6.13 for KMS on a 7xx GPU. A newer kernel wouldn't hurt either - AFAIK the combination of 6.13 and 2.6.33 is intended to be "the KMS release".
Consider getting a new mesa when you get around to 3D or compositors.
Thanks a bunch - compiled Radeon from git and everything seems to work as advertised with one minor warning. Loving KMS terminal switching, btw.
As far as the minor issue, when running glxinfo or glxgears I get:
do_wait: drmWaitVBlank returned -1, IRQs don't seem to be working correctly.
Try adjusting the vblank_mode configuration parameter.
As far as the heads up - not 100% sure what you mean about needing a new MESA for 3D or compositing (Kwin4 seems to work fine and 3D is fairly fast - at least for what I use). Maybe the above (and other such minor issues) are fixed in Mesa 7.8?
Both kernel 2.6.33 and xf86-video-ati 6.12.191 (the 6.13 pre-release) are available in Debian experimental, so building from git isn't necessary.
IIRC the 2.6.33 kernel enables interrupt support. You'll need an additional microcode binary, ending in ...rlc.bin. I don't remember if you need a newer Mesa or not...
I took a look at Debian experimental. Looks like some dependencies are currently broken so I can't just apt-get install to the newer versions cleanly. Compiling from git doesn't scare me, but I don't like breaking my system if I can avoid it. I think I'll just wait for 2.6.33 - it looks like it will fix my IRQ warning, but its not a big deal.
Thanks again for all the help.
Looking forward to the things just over the horizon in the Open Source driver effort.
The new 2.6.32 kernel that arrived in unstable a day or two ago has the DRM bits from 2.6.33. The Debian version is 2.6.32-10, but it shows up in /boot as 2.6.32-4 for some reason I don't understand. The linux-image-2.6-686 metapackage from sid will give you the correct version (linux-image-2.6-amd64 did for me at least). I don't know if this interrupt support is part of what's been backported though.
The 6.12.192 radeon driver is in experimental now also. It installed for me on Squeeze without pulling in anything else with it.
Just an FYI I guess, I only just installed myself a test partition last night and haven't done much but install those two items. Kwin effects were working nicely enough but I haven't gotten around to booting that install again with KMS enabled yet.
You can try installing the Liquorix kernel at http://liquorix.net. They are built for Debian and use the zen kernel sources along with a configuration for optimized desktop performance.
Right now it's based off of the 2.6.33 kernel, plus all the new stable patches. That's probably what you need too.
I tried your stuff as a quick test and it works well. It actually could become my working kernel (it includes the PHC patch!).
Originally Posted by damentz
As for KMS, I tried it with radeon from experimental (no git for me these busy days) and it's pretty cool. I'm not a big fan of compositing but I tried it all the same. One thing I noticed is that applications that use OpenGL don't have weird artifacts using this while desktop effects are on; i.e. the OpenGL windows are not mangled by what is behind or hovers over them, which is what I get with 2.6.32 and radeon 6.12 (if composite is on).
I was also curious about the BFS scheduler. I wouldn't be able to say for sure, but if anything it seems that scrolling in Firefox with composite on is less laggy, and effects such as expose or window switching come up more snappy.