Thursday, October 18, 2007

A year later, ATI Linux drivers still suck

Folks, just in case you don't know, ATI's display drivers for Linux still suck, one year later:
/2006/09/ati-opengl-20-implementation.html
/2006/09/opengl-acceleration-for-cinelerra-not.html

Here's a bit of the latest pain:
https://init.linpro.no/pipermail/skolelinux.no/cinelerra/2007-October/012026.html

Here's another tale of woe (read starting with post #1014):
http://forums.fedoraforum.org/showthread.php?t=155503&page=102&pp=10

In the interest of having a clean system to start with, I reinstalled Fedora Core 6, 64-bit. I installed all Cinelerra source dependencies and attendent media players and apps as per this post (/2007/09/building-cinelerra-on-fc6-64-bit.html). I then made the very intelligent choice of backing up my system by using a Knoppix Bootable CD (www.knoppix.com) to load partimage and make an image of my system drive. For those who'd like to know how to do this, I will try to post a doc on what to do.
Update 2008/11/16
Here's how you can do this.
end update

This allows me to easily restore from image if anything went drastically wrong with the ATI install process. It is a good thing I did this.

After having the core softwares installed, I compiled Cinelerra and created a project by recording a screen capture with voiceover, some fades and a simple mask. This worked well, so I then decided to tackle building the 8.36.5 version of the ATI installer. I built the ATI rpms with the following command:
./ati-driver-installer-8.36.5-x86.x86_64.run --buildpkg Fedora/FC6
With the recent clean install of FC6 x86-64-bit, I was surprised to see that the build process actually worked! Previously, this build process failed. I speculate that the reason for the previous failure was that I kept my repositories consistent on this OS build. This means I tried not to mix rpm bases from more than two non-Fedora repositories: Dries and Livna, respectively. Refer to http://crazedmuleproductions.blogspot.com/2007/09/building-cinelerra-on-fc6-64-bit.html) for the detail. So having this build process work was unexpected. It gave me hope that I might have some resolution to my ATI woes.

I installed the created ATI rpms and shutdown the box. I popped in the VisionTek card and rebooted to init 3, full multi-user mode, without X enabled. I then ran the ATI configurator:
./aticonfig --initial -f

[sodo@ogre 8.36]# ls -l
total 65244
-rwxr-xr-x 1 root root 53331877 Oct 8 21:18 ati-driver-installer-8.36.5-x86.x86_64.run
-rw-r--r-- 1 root root 6469324 Oct 10 22:49 ATI-fglrx-8.36.5-1.fc6.x86_64.rpm
-rw-r--r-- 1 root root 3367465 Oct 10 22:49 ATI-fglrx-control-center-8.36.5-1.fc6.x86_64.rpm
-rw-r--r-- 1 root root 45017 Oct 10 22:49 ATI-fglrx-devel-8.36.5-1.fc6.x86_64.rpm
-rw-r--r-- 1 root root 3192548 Oct 10 22:49 ATI-fglrx-IA32-libs-8.36.5-1.fc6.x86_64.rpm
-rw-r--r-- 1 root root 309737 Oct 10 22:49 kernel-module-ATI-fglrx-2.6.18-1.2798.fc6-8.36.5-1.fc6.x86_64.rpm



This created a default fglrx /etc/X11/xorg.conf file. I started X and was pleasantly surprised to see both my monitors (Dell FP1901 and FP1907) come alive. I started the ATI Catalyst Display utility and was excited to see that OpenGL was indeed active. I switched from Clone mode (2 separate screens displaying the same thing) to Big Desktop mode, one large desktop. Excitedly, I wanted to see the behavior of xine and mplayer under 8.36.5, as I read some threads that lead me to believe I might get OpenGL without the xine crashes and mplayer sluggishness I had seen using 8.39.4 and 8.40.4. Unfortunately, it was not to be, as I loaded mplayer to see the same sluggishness on HDV content display. I then started xine and true to form, xine displayed the HDV content, but then promptly crashed X.

Ah well. At this point, I decided to restart the system, just to see if the restart improved conditions. When I restarted, I got the following error:
/etc/rc.d/rc.sysinit line 819: Segmentation fault

Oh boy. That does not look good. And the entire system has locked up..I can't even reboot. After a hard shutdown, I restarted, only to find the same error condition reappear. Google is again my best friend:
http://www.linuxforums.org/forum/servers/15670-help-segmentation-fault.html

So, it seems my choices are that:
1) my system memory is corrupt
2) grep is corrupt
3) I've been hacked

This is a new server with ECC memory, so I really doubt the memory has gone bad. But I'd rather be sure. Through the above Google find, I see that memtest86 (www.memtest86.com) is a solid program for testing your system memory that I haven't used before. The doc on the website is very good. I grabbed the ISO online and burned a copy to CD and let my machine churn for about three hours running through memory tests. Thank goodness, no memory errors were found. It seems that Mr. Brady knows his stuff! Thanks Chris!

Secondly, I tried to remark out the offending "for" loop within rc.sysinit. I needed to boot off a Knoppix (www.knoppix.com) CD to be able to assemble my software RAID partition, mount it and edit the file. Doing this and rebooting yielded the same results.

My third and final troubleshooting procedure was to remove and replace "grep" which is where the init loader failed. Again, I booted off the Knoppix CD, assembled and mounted my RAID. This time, however, I changed my root (chroot) partition in order the remove and reinstall grep via rpm and the Fedora install DVD. I did this, rebooted and received the same error.

At this point, I threw my hands up in the air and decided to reinstall my clean partimage image. Knoppix to the rescue again, and I restored the image of my clean install off a second drive I had on the system. A reboot later and I was back up and running. Praise the Lord!

So..what is the moral of this story, you might ask? It must be this, cause I can't think of anything else right now: Don't trust people over 30 and be sure that ATI will always disappoint you if you run Linux.

Harumph!
The Mule

No comments: