ATI Radeon 9100: XOrg-drm or fglrx ?

[engelsman]

lspci on my box shows:

[dveatch]

Ok. Lets take this one step at a time.

1. motherboard- intel chipset or via? Which ever it may be you afaik also need intel_agp or via_agp kernel modules. lspci will help you there. Those are located in the AGP section of the kernel config.

2. When running fglrxconfig, please define what "chokes the X server" means. By that I mean, what is the output of Xorg.0.log?

3. Try using "xorgcfg" instead of xorgconfig. I have found the former to be more useful.

4. As for which driver to use, fglrx or XOrg-drm, whichever one you get running first?
_________________
You can tune a piano but you can't tuna fish.

[engelsman]

My full lspci listing shows

[sofar]

about the drivers:

1) Open Source "ati" driver - standard in xorg - does NOT provide 3D
2) Open Source "radeon" driver - comes from the XOrg-drm module, provides 3D for SOME cards
3) closed source "fglrx" driver - comes from the binary "fglrx" module, provides 3D for SOME cards.

when you "pick" one of these "drivers" - you need to tell XOrg that your "video driver" also has changed.

For instance, I use the radeon DRI on my radeon7500 (which has 3D support). My xorg.conf has thus a section like this:

[engelsman]

I've finally done it, based on the clues given above, the contents of /var/log/Xorg.0.log, running 'LIBGL_DEBUG=verbose glxinfo' and searching Google for the error messages, and also the DRI Troubleshooting page!

I've compiled the AGP and ATI_AGP into the 2.6.15.6 kernel (I was trying to use the ati_agp module but maybe it needed the later steps too).

I've compiled DRM and the related RADEON driver as modules (I also tried compiling them into the kernel, but again, maybe it needed the later steps).

[sofar]

you said that you use the following kernel config options:

[engelsman]

That might explaing why it took me so long trying to get the right combination

I had uninstalled both XOrg-drm and XOrg-drm-cvs at one point, and things were still running, but it was about then that I lost libGL.so. Things obviously don't get cleared out properly until reboot.

[engelsman]

I've just run 'lrm XOrg-drm-cvs' so now I have neither Xorg-drm nor XOrg-drm-cvs installed. I've just run 'lin linux-2.6' and rebuilt with the same options given above. I've rebooted, (my libGL.so hasn't disappeared), glxinfo still runs, and still tells me 'direct rendering: Yes'

And I've just found that I do have /dev/agpgart rather than /dev/agpart: a typo or thinko that's cost me a lot of time maybe

I've no idea what would happen if I rebuild XOrg from scratch again though

Seriously though, maybe I've reached the point where I have to find time to read up about the Linux kernel architecture, how compiled drivers work compared to modules, etc. I've been working under the impression that code for common drivers/modules is helpfully supplied as part of the kernel distribution, by the distro's developers, and that other drivers/modules can be added later. Like the madwifi-cvs module that I'm using. I thought that the configuration process allowed you to (a) ignore a feature, or (b) add a stub for a feature so you could hang a kernel module on it later, or (c) compile the code directly into the kernel. It hadn't really crossed my mind that there were kernel supplied drivers/modules *and* also alternative drivers and modules for some features, and that they might conflict. I assumed that everyone was working together on a single unified driver or module.

They say that a little bit of knowledge is a dangerous thing, but on the other hand, they also say that ignorance is bliss

[sofar]

kernel modules are never removed by lunar - whenyou lrm xorg-drm-cvs the 'radeon.ko' module stays intact. this is to insure that you can ue the hardware if something bad happens. the best way to start from scratch is probably:

[engelsman]

With some trepidation, early yesterday evening I started to follow the procedure above. I was running the current 2.6.15.6 kernel, with the apparently strangely configured but working AGP- and DRM-enabled kernel and/or modules. I shut down X as I was going to be removing and rebuilding it anyway, and started work on the console.

I ran the lrm line given in the reply above this one. I ran the rm -rf /lib/modules/`uname -r` line, and then started with lin -c linux-2.6. It didn't run make menuconfig but prompted me for the config one line at a time. I obviously made a mistake or two along the line, like giving a LOCALVERSION suffix that kept reappearing even when I thought I'd got rid of it. Lucky for me, as it turned out. I carefully added AGP and AGP_ATI into the kernel, and said no to DRM and DRM_RADEON in both the kernel and as modules. I let the system build.

It built the kernel. Then it built hostap, pcmcia-cs, madwifi-cvs and iptables, and then it surprised me by rebuilding XOrg too. Total build time was the best part of an hour.

I rebooted the system into the new 2.6.15.6, only to find that my wifi connection had gone. If you've read my User Talk page, you'll know what a saga it was to get the thing to work in the first place

I ran lin madwifi-cvs but it complained about a missing file (/etc/*/ppp/isp or something like that) and refused to build. I ran lin kernel-headers-2.6 and tried again, and this time it worked. On reboot, the wifi was still missing. modprobe ath_pci etc threw up fatal errors about missing modules.

I rebooted into 2.6.14.5, which did have a working wifi, and tried to lin linux-2.6 from there. I went round the loop with kernel-headers-2.6 and madwifi-cvs several times, booting in both 2.6.14.5 and in the new 2.6.15.6 but with the same result. No wifi

I began to get suspicious that I'd done something wrong in the first kernel re-config and that the error was simply being re-used over and over when I kept seeing the LOCALVERSION suffix reappearing, even though I kept deleting it from the config. And by this time I'd run lin -rc linux-2.6 and had specified that I wanted to use menuconfig so it was easier to go back and forth through the menus to check and change the config.

Finally, in desparation, I went back to square one. After booting 2.6.15.6, I ran the lrm again, and then removed /lib/modules/2.6.15.6* and rebooted into 2.6.14.5. I started the lin -rc linux-2.6 again, this time loading a copy of the strange but working 2.6.15.6 config that I had carefully saved elsewhere before I started on this rebuild from scratch.

Once again an hour to build the kernel, hostap, pcmcia-cs, madwifi-cvs, iptables and XOrg. Would it work? On the first reboot, no. But after reloading the kernel-headers-2.6 and madwifi-cvs and rebooting again, we were back in business. The wifi was back

So after two long evenings of rebuilding, the system is back to the strangely configured, but apparently working configuration with AGP and AGP_ATI configured into the kernel, and DRM and DRM_RADEON configured as modules. lvu reports that XOrg-drm, XOrg-drm-cvs and fglrx are not installed. glxinfo still reports that direct rendering is enabled, and glxgears still gives me 720FPS.

lunar nofix has just reported that the only problem is with libperl.so in graphviz, so we're back to normal

Now I know that I can recover to this configuration, given an hour of downtime, I might have another go at reconfiguring over the weekend, but I'm more tempted just to let sleeping dogs lie.

Thanks for the help - and I don't mean that sarcastically.