Fglrx 8.6 or 8.7 doesn't work with Intel AGP at all

[nbi1]

Shadowfb is enabled by default on 6xx and 7xx cards. There is also an option to force it on with 5xx and RS690 (Option "AccelMethod" "ShadowFB", I think).



No worries, I was just surprised that you hadn't found the existing open source drivers useable since shadowfb runs as fast or faster than traditional 2D acceleration on many systems. The only place you see a real advantage from "2D acceleration" these days is if you are running full EXA Render acceleration (running on the 3D engine) and a very recent X server build to pick up the latest EXA server enhancements. You might be surprised how fast shadowfb really is.



Well, there is the secret conspiracy with Microsoft... darn it, I wasn't supposed to say anything about that

Seriously, your comments about prioritizing 2D over 3D made me think you would have been pretty happy with the open source drivers, since 3D is really where fglrx has a big lead over the open source drivers.



No, not paranoid at all

radeonhd does not install on Debian etch. There are too many dependency issues to attempt building it from source. Don't ask me to change my OS in an attempt to use an unproven driver. There are many alarming statements in the radeonhd documentation:
1. "Future work is happening especially on more advanced features like 2D, 3D and video acceleration." (Say what?)
2. "No XVideo"
3. "there is no DRI support for r6xx based cards yet!"

Any other minor details I overlooked?

I posted my complete X log in my bugzilla report for Catalyst 8.7. I'm waiting to see if anyone will respond.

[israel_miranda]

Take it easy

nbi1, I understand your frustration(I've been there before) but try to see some other aspects.
fglrx is not that bad. I know people who use ATI cards in linux and have no problems. The driver improved a lot last year, believe me, many problems have been solved. What happens in our case is a bug with an intel_agp module. Unfortunately I had found people on the internet asking help to configure fglrx on an intel agp motherboard back to 2005, with the very same behavior blank screen, computer hangs and freezes. But this "bug" happens only in some specific hardware kernel-options/libs combination. I found this discussion which demonstrates what I'm talking about:

http://bbs.archlinux.org/viewtopic.php?id=26016

It "miraculously worked"
They don't know what triggers this bug.

Don't blame Debian's fglrx packages they are in good shape, as you said the best linux distro in the planet. Fglrx is probably not fully tested in AGP platform as it is in PCI-E. All of this problems happens mostly to card with this Rialto chip (a chip which interfaces PCI-E GPUs to AGP slot), and bugs in packages even the Debian stable branch exists once in a while. Maybe the microcode for this chip should be opened to the Xorg community.

I'll attach my logs and experience to the bug report, and bridgman did you check the logs I posted ? My computer just hangs after I try to start kdm and no more message is saved in the log. Could you give me some directions to debug the application ?
I really don't know how to proceed and what to debug: kdm ? X ? fglrx ?
everything ?

[Kano]

Hi! As I dont know anything about your systemconfig, I updated my bench iso to current kernel + fgrlx 8-7 (+ all nvidia drivers). Language is german, but should not matter, it autoinstalls the 3d drivers.

http://debian.tu-bs.de/project/kanotix/preview/kanotix-ng-bench-2.6.26-5-generic.iso

http://debian.tu-bs.de/project/kanotix/preview/kanotix-ng-bench-2.6.26-5-generic.iso.md5

It includes gl2benchmark, something that can only render nvidia correctly (there test 1 to 4 work, fglrx can only render 1+2). You can also check your settings with amdcccle and fgl_glxgears. (if you need to get root: sudo -i)

[mshopf]

radeonhd should build on

radeonhd does not install on Debian etch. There are too many dependency issues to attempt building it from source. Don't ask me to change my OS in an attempt to use an unproven driver. There are many alarming statements in the radeonhd documentation:

We don't have a bug report about radeonhd not compiling on Debian Etch. Which it should, in fact we think that even modern drivers should compile on older X systems, an opinion that isn't shared by all developers in the X community. So please report on bugs.freedesktop.org or on the radeonhd mailing list. Build was broken for BSD for a little while, but that has been fixed lately.

1. "Future work is happening especially on more advanced features like 2D, 3D and video acceleration." (Say what?)
2. "No XVideo"
3. "there is no DRI support for r6xx based cards yet!"

Ok, your point? r6xx is a completely new architecture, and especially the 3D engine has provrn to be notorious difficult (ask Alex Deucher or John Bridgman for confirmation). r5xx acceleration is luckily almost finished thanks to Alex, while on the 3D side unfortunately nobody works on the r5xx mesa driver on a constant basis.

[Kano]

You can install radeonhd with etch. It is just more tricky. Kanotix has it (base still etch) - just without 3d support, therfore I provide the fglrx script.

[nbi1]

You can install radeonhd with etch. It is just more tricky. Kanotix has it (base still etch) - just without 3d support, therfore I provide the fglrx script.

If someone wants to give me the recipe for bludgeoning the "testing" package into usable form for etch I'd be happy to do it as I don't expect to be spoon fed on everything. If it's already available in etch usable form great, but the conversion procedure would be good to know at any rate.

BTW, ich spreche gebrochenes Deutsch. (Ehemaliger Ansbacher)
And I could use a cold Tucher Pilsener right about now!
Cheers.

Peter

[legume]

If someone wants to give me the recipe for bludgeoning the "testing" package into usable form for etch I'd be happy to do it as I don't expect to be spoon fed on everything. If it's already available in etch usable form great, but the conversion procedure would be good to know at any rate.

BTW, ich spreche gebrochenes Deutsch. (Ehemaliger Ansbacher)
And I could use a cold Tucher Pilsener right about now!
Cheers.

Peter

Your logs seem OK - that AGP -1023 is the killer. I used to get that on my X1600 until I found a workaround (setting aperture to 512 in bios) that was with nvidia-agp.

Now I have almost the same card as you and I don't even need the workaround anymore.

I guess intel-agp has it's own different problems :-(

If you happen to have

i82875p_edac
edac_mc

loaded then you could try blacklisting them -

http://peter.peca.dk/art_Linux_Hardware_Compatibility.html

[nbi1]

Your logs seem OK - that AGP -1023 is the killer. I used to get that on my X1600 until I found a workaround (setting aperture to 512 in bios) that was with nvidia-agp.

Now I have almost the same card as you and I don't even need the workaround anymore.

I guess intel-agp has it's own different problems :-(

If you happen to have

i82875p_edac
edac_mc

loaded then you could try blacklisting them -

http://peter.peca.dk/art_Linux_Hardw...atibility.html

I never had the edac code built at all, either as module(s) or into the kernel. Yes, I did find that discussion thread, but thanks for bringing it to my attention. Maybe I should give it a try?

Setting the aperture size in BIOS had no effect on this problem other than to increase/decrease overall system performance.

[legume]

I never had the edac code built at all, either as module(s) or into the kernel. Yes, I did find that discussion thread, but thanks for bringing it to my attention. Maybe I should give it a try?

Setting the aperture size in BIOS had no effect on this problem other than to increase/decrease overall system performance.

He had to disable them so I doubt using them would help - maybe you could try seeing what other modules get loaded that you can live without and blacklist those. Clutching at straws really.

I am surprised changing aperture size in bios (which I thought just sets the upper limit that a driver could use if it wanted too) gives a performance difference.

[nbi1]

You can install radeonhd with etch. It is just more tricky. Kanotix has it (base still etch) - just without 3d support, therfore I provide the fglrx script.

I don't understand what you are saying here. You have already created such a script or are creating one? All I need is some assistance in getting the source built.

I tried building from source, but autogen.sh is missing. I'm guessing the notes simply didn't get updated to reflect the use of 'configure' which bombs as follows:
checking for XORG... configure: error: Package requirements (xorg-server xproto fontsproto ) were not met:

No package 'xorg-server' found
No package 'fontsproto' found

I can't find these packages.