NVIDIA Driver/Kernel Module mismatch

[Ground_Zero]

OK, first a little background.

Recently, I compiled a low-latency kernel from the Mepis kernel sources. I couldn't boot the new kernel into a graphical environment, as the NVIDIA modules were for the other kernel (which I wanted to keep). I attempted to install the NVIDIA driver the NVIDIA way from the wiki, but it did not work. I booted into the live CD and set the driver to nv, and both kernels booted. It was suggested (by kmathern I believe) that I had to install the appropriate driver from the NVIDIA website for both kernels. I intended to follow this course of action, but then my monitor died, so I put off fixing it (I do want 3d acceleration on both kernels). Now to the current problem...

When I got a new(er) monitor, I booted the live CD and switched back to nv. When switching to the nv driver, it says it is removing the NVIDIA drivers, but this may not have gotten rid of the ones installed from the NVIDIA installer. I installed the NVIDIA driver from the Mepis X Assistant for the stock Mepis kernel (haven't messed with the low-latency kernel again yet). Now when I attempt to use accelerated programs, I get the following error:

Error: API mismatch: this NVIDIA driver component has version
96.43.05, but the NVIDIA kernel module's version does not match.
Please make sure that the kernel module and all NVIDIA driver
components have the same version.
NVIDIA: Direct rendering failed; attempting indirect rendering.
Error: API mismatch: this NVIDIA driver component has version
96.43.05, but the NVIDIA kernel module's version does not match.
Please make sure that the kernel module and all NVIDIA driver
components have the same version.
NVIDIA: Direct rendering failed; attempting indirect rendering.

I found some posts with similar issues here, here, and here. I think my situation is closest to the second link, having used the Mepis assistant to install what may have been only modules. Will any of these solutions help me, or can someone point me to some different/better info? Basically it seems I need to get rid of all NVIDIA stuff, and start over from scratch. Can I do this by switching the driver to nv (which should remove the Mepis-installed NVIDIA driver) and running the NVIDIA installer with the uninstall option?

[kmathern]

I'm trying to understand your situation, the error message you posted - is that while running the kernel you compiled or is it while running the original mepis kernel (or both?) ?

Do you have the nvidia driver working while booted into the original mepis kernel?

By the way, back when we discussed this earlier I did succeed in getting the nvidia driver installed using the NVidia installer script to install and run for two different kernels on the same installation (the official mepis kernel [linux-image-2.6.22-1-mepis-smp] and a kernel stevo compiled)

I used the -K option of the nvidia installer (I think I earlier mentioned a -k option which didn't work, the -K option did)

-K, --kernel-module-only
Install a kernel module only, and do not uninstall the existing driver. This is
intended to be used to install kernel modules for additional kernels (in cases where
you might boot between several different kernels). To use this option, you must
already have a driver installed, and the version of the installed driver must match
the version of this kernel module.

[Ground_Zero]

Originally Posted by kmathern

I'm trying to understand your situation, the error message you posted - is that while running the kernel you compiled or is it while running the original mepis kernel (or both?) ?

Sorry if my original post was all over the place. That is running the stock Mepis kernel. Since my monitor died, I have not touched the other kernel.

Originally Posted by kmathern

Do you have the nvidia driver working while booted into the original mepis kernel?

I am running the stock Mepis kernel, and the NVIDIA driver is (apparently) working for 2d, but 3d is not working. I switched to the nv driver from the live CD as soon as I hooked up my new monitor, and it said it was removing the NVIDIA files installed by the X Assistant. I then reinstalled the NVIDIA drivers from X assistant, and that is where I am now (I rebooted where appropriate). It appears my failed attempt at installing drivers for the new kernel has left behind some stuff that is conflicting with the X Assistant driver and needs to be removed.

Originally Posted by kmathern

By the way, back when we discussed this earlier I did succeed in getting the nvidia driver installed using the NVidia installer script to install and run for two different kernels on the same installation (the official mepis kernel [linux-image-2.6.22-1-mepis-smp] and a kernel stevo compiled)

I used the -K option of the nvidia installer (I think I earlier mentioned a -k option which didn't work, the -K option did)

-K, --kernel-module-only
Install a kernel module only, and do not uninstall the existing driver. This is
intended to be used to install kernel modules for additional kernels (in cases where
you might boot between several different kernels). To use this option, you must
already have a driver installed, and the version of the installed driver must match
the version of this kernel module.

I am glad to hear it works, thanks for trying it out! I would like to get rid of everything NVIDIA related, go back to the nv driver, install the NVIDIA driver from the NVIDIA website for the stock kernel, and make the module for the low-latency kernel. Is this what I should be doing, and if so, how?

[kmathern]

Using the NVidia installer script to install the nvidia driver for both kernels is what I'm recommending - basically because I've done it and therefore I am familiar with that method. The one drawback with this method is I don't think you've been able to get the NVidia installer script to successfully work for you previously (is this the case?)

You may also be able to get the nvidia drivers installed by using the Mepis X windows assistant as a starting point. The Mepis X windows assistant would be used to get the nvidia driver working for the original mepis kernel and then I've found that there is a package named nvidia-kernel-legacy-96xx-source which I think would allow you to build the kernel module for your custom kernel (although I'm a little fuzzy on the details about how you do this - stevo or someone else probably knows how)

Which method do want to use?

[Stevo]

I have a theory that the system looks at a symbolic link "linux" in /usr (not the one manually created when compiling a kernel) normally it points to the kernel headers directory. When you boot to a different kernel, you have to point it to the correct headers directory.

Of course I could be completely wrong, but that's what happened when I reinstalled the stock kernel headers in Synaptic.

[kmathern]

Originally Posted by Stevo

I have a theory that the system looks at a symbolic link "linux" in /usr (not the one manually created when compiling a kernel) normally it points to the kernel headers directory. When you boot to a different kernel, you have to point it to the correct headers directory.

Of course I could be completely wrong, but that's what happened when I reinstalled the stock kernel headers in Synaptic.

Stevo,

Would you be refering to the /usr/src/linux symlink ?

By the way I just re-installed your linux-image-2.6.22-rtc-mepcom kernel and headers and then booted that kernel and built the nvidia driver using the -K option from the NVIDIA-Linux-x86-169.12-pkg1.run installer script. My /usr/src/linux symlink was still pointing to the original (2.6.22-1-mepis-smp) headers and the driver built without any problem.

[Ground_Zero]

Originally Posted by kmathern

Using the NVidia installer script to install the nvidia driver for both kernels is what I'm recommending - basically because I've done it and therefore I am familiar with that method. The one drawback with this method is I don't think you've been able to get the NVidia installer script to successfully work for you previously (is this the case?)

This is what I would like to do as well. I think the reason it did not work before is I had the X assistant drivers installed on the Mepis kernel, and I attempted to install the drivers from the NVIDIA website on the realtime kernel, without considering that the root directory and Xconfig are shared between the two kernels. There were references to two driver versions in different places, I think, which is why neither kernel would boot with the NVIDIA driver after my install attempt. So I think if I can get rid of everything NVIDIA that I have now and go back to the nv driver, I can install the NVIDIA driver/modules for both kernels per your method.

I am just not sure how best to be rid of everything NVIDIA. As I said earlier, when switching back to the nv driver in the Mepis X assistant, it says it must uninstall the NVIDIA driver to continue. So I think any references to the Mepis version of the NVIDIA driver will be removed by this. So how do I remove the references to the other driver I installed? Should I run the NVIDIA installer with the --uninstall option? Will I need to remove all the NVIDIA packages in Synaptic per the Ubuntu forum thread above? Should I ruthlessly delete every file and folder in the root directory that refers to NVIDIA? Sorry for all the questions, I just don't want to screw things up worse.

Originally Posted by kmathern

You may also be able to get the nvidia drivers installed by using the Mepis X windows assistant as a starting point. The Mepis X windows assistant would be used to get the nvidia driver working for the original mepis kernel and then I've found that there is a package named nvidia-kernel-legacy-96xx-source which I think would allow you to build the kernel module for your custom kernel (although I'm a little fuzzy on the details about how you do this - stevo or someone else probably knows how)

If the other method fails, I may try this. I think I can make the other method work if I can get rid of all the previous NVIDIA stuff, though.

Originally Posted by Stevo;

I have a theory that the system looks at a symbolic link "linux" in /usr (not the one manually created when compiling a kernel) normally it points to the kernel headers directory. When you boot to a different kernel, you have to point it to the correct headers directory.

Of course I could be completely wrong, but that's what happened when I reinstalled the stock kernel headers in Synaptic.

This may be another reason I had an issue with installing the drivers for the realtime kernel. I did it booted into the realtime kernel from a commandline prompt, and it seemed to build for the stock Mepis kernel, rather than the realtime kernel.

[kmathern]

To start could post the results of:

dpkg -l |grep nvidia

[Stevo]

After I reinstalled the stock kernel headers, there was a linux symlink in /usr, separate from the one in /usr/src. The second one wasn't there before I manually created created it per the kernel compile instructions, but I think you can compile modules with that link, too, since there are headers inside the source tree. I think that's what module-assistant is using as headers.

How is that RTC kernel doing, anyway? It kept resetting the clock for me.

[kmathern]

Originally Posted by Stevo

After I reinstalled the stock kernel headers, there was a linux symlink in /usr, separate from the one in /usr/src. The second one wasn't there before I manually created created it per the kernel compile instructions, but I think you can compile modules with that link, too, since there are headers inside the source tree. I think that's what module-assistant is using as headers.

How is that RTC kernel doing, anyway? It kept resetting the clock for me.

I haven't really used it much, I just installed it to verify that I could have the nvidia driver installed simultaneously for multiple kernels using the NVIDIA-Linux-x86-xxx.yy-pkg1.run script. About a month ago I found I could have multiple nvidia drivers installed but I un-installed your rtc kernel shortly after that. I just loaded it up again tonight.

edit: I have both the stock kernel headers and your rtc kernel headers installed - I don't have any linux symlink in /usr