can't install nvidia driver
 

Debian Sid 2.6.20
Hi ,
I'm trying to install to my GE force MX440 the 1.0-9631 GPU driver . During installation I'm getting :

FATAL: modpost: GPL-incompatible module nvidia.ko uses GPL-only symbol 'para
virt_ops'
make[3]: *** [__modpost] Error 1
make[2]: *** [modules] Error 2
NVIDIA: left KBUILD.
nvidia.ko failed to build!
make[1]: *** [module] Error 1
make: *** [module] Error 2
-> Error.
ERROR: Unable to build the NVIDIA kernel module. "

How can I over come this ?

Thanks.

I had a problem with the 1.096xx driver when installing it for my GeForce 4 Ti 4200 - even though the list at http://us.download.nvidia.com/XFree8...ppendix-a.html says it is supported the installer said it wasn't. I tried the 1.0-71xx driver (for really old cards) and that worked. Maybe that will work for you too, if you think it's an acceptable driver.

As an old Linux user but a fairly new Debian user, getting Nvidia installed was one of the worst experiences I had with Debian, and kept me from adopting the distro until I found this: http://tinyplanet.ca/~lsorense/debia...-dri-howto.txt

This works for me every time. It's bizzare, but does the trick.

Is this the nvidia module which you downloaded from the nvidia web site? The .run package. I have installed it succesfully in Sarge, Etch and Lenny, but kernel version only up to 2.6.18.

jlinkels

No, I've never gotten that stock Nvidia driver from Nvidia.com to work on any Debian flavor. That was my initial frustration with Debian, but once I found that tutorial I posted a link to, Nvidia driver installations work great. The driver is old compared to the latest from Nvidia though, so you might want to see if it is possible to install using maybe unstable or something. Not sure how to pull from multiple levels of apt repositories though - someone else may be able to help with that.

Yes, but that installation procedure is needlessly complicated. I did it once (succesfully) but it is not necessary, and also considered outdated.

If you download the .run package, it runs a script and leaves you with the installed driver, no interference with Debian packages. You don't even need a custom kernel. The relevant kernel module is compiled for you on the fly.

Important to know is that you should have your kernel source installed (e.g. apt-get source linux-source-2.6.18) and that you have the correct gcc version installed which was also used for compiling the kernel.

Beware of the latest driver versions though. There is an error in the GLX libraries for versions > 8x.xx which cause a 10.000 times slowdown. Since the MX440 is considered obsolete anyway you are better off with the latest 7x.xx version.

However, there are also reports that 2.6.20 kernel causes severe slowdown with any nvidia driver. If you think you experience this, you might consider downgrading to 2.6.18.

jlinkels

Interesting. That's exactly the opposite advice I've gotten from Debian forums and other places I've asked - they've always told me to do it "the Debian way", which I also think is convoluted and doesn't make sense. I came off Gentoo, which in comparison to the "Debian way" is easy. I am running a Geforce 6800 OC so have not yet been pushed to use the legacy driver, but will probably be in a couple of years. I might give the manual method a shot again to see if I can get it to work.

Have you tried using module assistant? When I was running Sarge, I could not get nVidia drivers working for me no matter what I did. Then I found this post:

http://www.linuxquestions.org/questi...ight=nvidea+ma

I followed the directions and the drivers installed perfectly. I am not sure if this will work under etch, as I have just installed etch 2 days ago, and I'm still playing with it.

I hope this helps!

:study:

The link I posted earlier in this thread uses module assistant. I have been using testing (previously Etch) and have had no problems. The only problem is with me accepting the convoluted way Debian does the driver install. Since I only have to reaccomplish the procedure when the kernel or Nvidia driver gets updated, it's not a big deal. Because Debian is so darn fast, easy to maintain and tweak, I can overcome my reluctance to accept the "Debian way", no matter how difficult and convoluted I think it is :)

Compile your own kernel/install and reboot into it leaving out the para_virt_ops module however it is named in the .config file there is a conflict between it and the Nvidia driver as you have found.

Sargek,

I did a quick search on what is current now and you are right. Many recommendations in favor of doing it the Debian way.

I think I did that once 2 years ago on a Sarge version. After that I did it the Nvidia way.

Since nvidia advertises their way as a unified driver for all distros, I could not imagine that for two very important distros (Ubuntu and Debian) and a handful more apt based another method should be followed.

I am also aware that usually doing something the 'Debian" way is better in the end. Kernel compiling is really easier doing it the Debian way.

I wonder what the general feeling is about using either the nvidia way or the Debian way.

jlinkels

No problem, I'm a long time Gentoo user so I personally find the Debian way rather difficult because the details are abstracted away from the user. Being a programmer doesn't help either, because I am used to seeing and dealing with details. Building a custom kernel the "standard" way, and installing Nvidia the manual way, to me, is far easier than the Debian way. It IS nice to not think about things sometimes though, so maybe that's why I'm kind of liking Debian. I do feel the Gentoo pull sometimes though...:rolleyes: I just don't like the overnight build times for updates...

Why this happens
 

I have had a similar problem that some of you are reporting, however for me it occurred at the 2.6.20 kernel upgrade, so I'm now back at 2.6.18-4.
And yes i like to try and keep things "the Debian way" as much as possible except for when it comes to the nvidia drivers.

The one thing i found out about this para_virt_ops option is that it is a kernel module compiled for the purposes of helping out programs like vmware and suck. I haven't yet found it in modconf and as happytux reported, the only way that i have seen for removing this option is to build your own kernel with the para_virt_ops module turned off.

The para_virt_ops module, from what i have seen reported, requires that the kernel be "taint free" from any third party modules. Also from what i have seen, it has been logged as a bug and hopefully someone, somewhere is looking into a way to fix it.

Many people have also spoken about asking nvidia the same old question of allowing the source code to be opened. I honestly cant see this happening in a hurry. So for now, until a release for a new kernel without this problem or nvidia go opensource, i am sticking with a 2.6.18-4 kernel.

Just thought i would throw my 2 cents worth in as i spent a weekend trying to solve it and didn't want others to do the same without knowing what they are up for.

DJ :D

I have no problems with the Nvidia 9631 driver on my Etch machines. One has an MX400 and the other has a ti-500. Currently using 2.6.20.3.

You're right, sorry about that; I was in a hurry and wanted to help. I looked at the link you provided, and is very similar to the way I did it. I already had custom kernel, so I didn't update the kernel, but the remainder of the process did work when the other ways wouldn't. I am going to try to get the nvidia drivers installed this weekend on my etch machine.

:study: