How to install the nvidia driver in F12 using the run method - Page 4

plasmonics · #46 29th May 2010, 11:53 AM

sh NVIDIA-xxxx.run --kernel-source-path=/usr/src/xxxxxx

This is after you have installed the new kernel and rebooted into it.

The source tree should be "prepared". It should be after a successful compilation,
unless you ran "make clean". If the tree is cleaned out, you can prepare it without doing
a full recompile using "make scripts && make prepare".

elelab · #47 26th December 2010, 01:52 AM

Hi plasmonics,

Can we apply the procedure to RHEL 6?

thanks

plasmonics · #48 26th December 2010, 11:17 AM

Quote:

Originally Posted by elelab

Hi plasmonics,

Can we apply the procedure to RHEL 6?

thanks

That is a good question. I have never used RHEL, just Fedora.

If RHEL uses an older kernel that does not integrate the nouveau driver into the kernel, it should be even simpler:

$ sh NVIDIA-xxxxx.run --kernel-source-path=/path-to-kernel-headers

dbonner · #49 22nd April 2011, 03:31 AM

I installed FC15 beta on my desktop and laptop. The rpmfusion drivers just did not work for my desktop (with a nvidia 9800GT) and laptop (nvidia 330M GT). I followed all the advice and married up the exact same driver and kernel that worked for others in this forum .... but not for me.

However .... the .run method (almost exactly word for word as you originally posted) worked perfectly for F15 beta for both my computers. I used the latest driver from the nvidia website: NVIDIA-Linux-x86_64-270.41.06.run

I made one modification to step 13: I shortened the first ModulePath line. The original article quoted:
ModulePath "/usr/lib64/xorg/modules/extensions/nvidia”
However, after I installed the nvidia driver, I found out that the /usr/lib64/xorg/modules/extensions/nvidia directory did not exist. This might be something that’s changed in the 11 months since the article was updated.

plasmonics · #50 22nd April 2011, 11:16 AM

Quote:

Originally Posted by dbonner

I made one modification to step 13: I shortened the first ModulePath line. The original article quoted:
ModulePath "/usr/lib64/xorg/modules/extensions/nvidia”
However, after I installed the nvidia driver, I found out that the /usr/lib64/xorg/modules/extensions/nvidia directory did not exist. This might be something that’s changed in the 11 months since the article was updated.

It should be safe to skip step 13 entirely. I added a note to the original post. Try commenting out the other two module lines and restarting the X server or rebooting. I tried it for F14 and it worked.

JamesRay · #51 6th May 2011, 12:32 AM

I'm a complete Linux novice (or worse), but I thought that I'd give this a try. The graphics computers that I use all run linux so I thought that I would try to get the nVidia driver installed myself this time around.

I tried following this guide primarily because I wanted to be absolutely sure that I was installing nVidia's current driver from the nVidia website.

I skipped step 2 because I thought that I was not running SElinux and now I'm wondering if that was a mistake. I am running FC 14. I also skipped step 5 because this was the first time I had tried to do any sort of driver installation since the fresh install of FC14.

Step 12 gave me errors. It could not find the kernel paths. I used the --kernel-devel-path (or w/e the proper spelling of that option is...) and was able to point it right at specific directories. Doing this allowed me to get past that error, but I still received errors during the nVidia kernel build phase.

At this point I got worried that I was going to cause serious harm and need to format the drive and start over. So I did my best to restore the system back to the nouveau driver and quit for a while.

Any advice or tips would be greatly appreciated.

plasmonics · #52 6th May 2011, 11:00 AM

Quote:

Originally Posted by JamesRay

I skipped step 2 because I thought that I was not running SElinux and now I'm wondering if that was a mistake. I am running FC 14. I also skipped step 5 because this was the first time I had tried to do any sort of driver installation since the fresh install of FC14.

Step 12 gave me errors. It could not find the kernel paths. I used the --kernel-devel-path (or w/e the proper spelling of that option is...) and was able to point it right at specific directories. Doing this allowed me to get past that error, but I still received errors during the nVidia kernel build phase.

You can check if you are using SElinux by examining /etc/selinux/config. If it says "SELINUX=disabled", you are not running it.

The only time you need to specify the path is if you are using a custom kernel. If you using a factory kernel, nVidia's run script finds the path by looking under /lib/modules/2.6.35.xxxxx. There are two symlinks, "build" and "source". These point to where the kernel-devel package is installed. In any case, the path is specified with "--kernel-source-path=xxxxx". This is unnecessary even for a custom kernel if "build" and "source" are set correctly.

I am currently running the latest driver 270.41.06 on F14 with a three year old (GeForce 7000M) graphics chip. No problems so far.

JamesRay · #53 6th May 2011, 08:07 PM

Thank you for the prompt help response!

I checked the SElinux config file and found it enabled so I repeated the process this time adhering to step 2.

I encountered the same problem so I think that it is safe to say that my build and source symlinks are NOT set properly. I assume we are talking about the build and source that live in /lib/modules/2.6.35.xxxxx

Mine currently point to:
source -> build
build -> ../../../usr/src/kernels/2.6.35.6-45.fc14.x86_64/

If I had to guess I would make build w/
ln -s /usr/src/kernels/2.6.35.12-90.fc14.x86_64/ build

and leave source pointing to build.

But I wonder how I ended up with a /lib/modules/ of 2.6.35.6-45 and a /usr/src/kernels/ of 2.6.35.12-90.

Changing the build symlink gets me as far as using the specifc path got me earlier. The nVidia installer gets to 100% in the build kernal phase but then gives me an error that it was unable to load nvidia.ko. It further informs me that the most common causes of this error are:
- the kernel module was built against the wrong or improperly configured kernel sources
- the kernel module was built using the wrong or a different version of gcc than the one used for the target kernel
- a driver (list of examples of drivers including the nouveau driver) is hogging ownership of the NVIDIA graphics hardware

I saved the nvidia log file after both install attempts but I do not think it is necessary to look at that. Instead I believe that the slight version difference I obseved earlier is more than just a minor problem and rather is the source of my frustration. However, I'm not sure what caused it or how I should resolve it.

plasmonics · #54 7th May 2011, 12:00 PM

Quote:

Originally Posted by JamesRay

Mine currently point to:
source -> build
build -> ../../../usr/src/kernels/2.6.35.6-45.fc14.x86_64/

If I had to guess I would make build w/
ln -s /usr/src/kernels/2.6.35.12-90.fc14.x86_64/ build

and leave source pointing to build.

But I wonder how I ended up with a /lib/modules/ of 2.6.35.6-45 and a /usr/src/kernels/ of 2.6.35.12-90.

Changing the build symlink gets me as far as using the specifc path got me earlier. The nVidia installer gets to 100% in the build kernal phase but then gives me an error that it was unable to load nvidia.ko. It further informs me that the most common causes of this error are:
- the kernel module was built against the wrong or improperly configured kernel sources
- the kernel module was built using the wrong or a different version of gcc than the one used for the target kernel
- a driver (list of examples of drivers including the nouveau driver) is hogging ownership of the NVIDIA graphics hardware

I checked mine just to be sure. I have build pointing to -> ../../../usr/src/kernels/2.6.35.12-90.fc14.x86_64 and source pointing to build. This is done automatically by the package handler. It sounds as though you are missing the directory /lib/modules/2.6.35.12-90.fc14.x86_64. I am not sure how this could have happened, as it is part of the kernel package. Manually setting links and source directories is usually done only when you are building your own kernel. For stock kernels, everything is taken care of.

I suggest reinstalling both the kernel and kernel-devel packages. Unfortunately "yum reinstall" command works for kernel-devel but not for kernels. You may have to download the rpm files from the Fedora repository via a web browser and install using "rpm -ivh". rpm has a "--force" option that should be used very sparingly and with caution.

The nouveau driver hogs the graphics card because its automatic loading is embedded in the /boot/initramfs-2.6.xxxxx file. That is why you have to blacklist nouveau and use the dracut command to rebuild the initramfs file without nouveau support. This is part of the kernel mode setting (KMS) mechanism. You can prevent KMS by adding "nomodeset" without the quotes somewhere on the kernel line in /boot/grub/grub.conf. But I think you still have to use dracut to rebuild the initramfs file.

JamesRay · #55 9th May 2011, 09:21 PM

I finally got it installed!

The problem was only having the 2.6.35.6-45.fc14.x86_64 version of the kernel installed. But then when I executed step 9 the first time through it installed the 2.6.35.12-90.fc14.x86_64 version of the kernel-devel package. So then it took me a while to hunt down and finally have the same version installed and configured at all levels.

I finally decided on using the 12-90 version and redid all steps making sure to get step 8 done with the new version (12-90) rather than the old one. I also had to do " yum install kernel-2.6.35.12-90.fc14.x86_64 " in order to actually get the desired kernel installed before I did any of the steps listed.

I am pretty sure that my system was originally set up with the .6-45 version and not the .12-90 version. But for whatever reason step 9 installed the 12-90 development package and therefore the sym links in my /lib/modules/2.6.35.6-45.fc14.x86_64 directory were broken links (b/c they were looking for a /usr/src/ for .6-45 and only the .12-90 one existed).

I hope that my comments on this matter might help someone else who runs into installation troubles. Make sure that the version numbers used in every step (especially 8 and 9) agree with the actual kernel version numbers that your system is using ( look in the /lib/modules directory).

Well thank you for the guide and the help responses Plasmonic! I am now happily using the latest nVidia driver and have learned a bit about Linux software configuration to boot~

plasmonics · #56 9th May 2011, 11:00 PM

Quote:

Originally Posted by JamesRay

I finally got it installed!

The problem was only having the 2.6.35.6-45.fc14.x86_64 version of the kernel installed. But then when I executed step 9 the first time through it installed the 2.6.35.12-90.fc14.x86_64 version of the kernel-devel package. So then it took me a while to hunt down and finally have the same version installed and configured at all levels.

I made a change to step 9 which guarantees against a version mismatch.

vfulco · #57 24th May 2011, 06:30 PM

[QUOTE=plasmonics;1294394](Note added on 5/31/2010: Surprisingly, these steps also work for Fedora 13. For x86_64 F13, the i686 libraries are no longer required in step 9. The x86_64 libs are sufficient. I have tried it with and without i686 and could not detect any difference.)

Dear Plasmonics- Thx for the excellent tips. Upon final reboot however on F15, the process stops right after 'Starting LSB: Daemon to access a smart card using PC/SC." Any hints, tips, tricks, to proceed from here would be welcome. TIA, V.

plasmonics · #58 25th May 2011, 06:50 PM

[QUOTE=vfulco;1474507]

Quote:

Originally Posted by plasmonics

(Note added on 5/31/2010: Surprisingly, these steps also work for Fedora 13. For x86_64 F13, the i686 libraries are no longer required in step 9. The x86_64 libs are sufficient. I have tried it with and without i686 and could not detect any difference.)

Dear Plasmonics- Thx for the excellent tips. Upon final reboot however on F15, the process stops right after 'Starting LSB: Daemon to access a smart card using PC/SC." Any hints, tips, tricks, to proceed from here would be welcome. TIA, V.

I have not yet installed F15. It sounds like it is more to do with the smart card daemon (pcscd) than with the video.

This and other services can be turned of using chkconfig pcscd off.

---------- Post added at 01:50 PM ---------- Previous post was at 06:16 AM ----------

i just installed F15 on a laptop. The steps still work.

vfulco · #59 25th May 2011, 10:03 PM

[QUOTE=plasmonics;1474830]

Quote:

Originally Posted by vfulco

I have not yet installed F15. It sounds like it is more to do with the smart card daemon (pcscd) than with the video.

This and other services can be turned of using chkconfig pcscd off.

---------- Post added at 01:50 PM ---------- Previous post was at 06:16 AM ----------

i just installed F15 on a laptop. The steps still work.

It is actually the service right after the smart card daemon but I am unable to find an ordered list so I can check the boot process against it?!?

plasmonics · #60 25th May 2011, 11:04 PM

[QUOTE=vfulco;1475166]

Quote:

Originally Posted by plasmonics

It is actually the service right after the smart card daemon but I am unable to find an ordered list so I can check the boot process against it?!?

I did not have to turn off any daemons. Did you try removing the nouveau driver (one of the steps) and re-running dracut? The kernel and kernel-devel packages have to match in version numbers before building the video driver.

In any case, "chkconfig --list" lists the services.