2.4.18-3 Nvidia driver help needed

[stevnb]

I am running RedHat 7.3 with the 2.4.18-3 kernel. I have attempted to build the NVIDIA_kernel from the src rpm but I get error messages.

install: cannot stat `nvidia.o`: No such file or directory
error: Bad exit status from /var/tmp/rpm-tmp.92699 (%install)

Any help would be appreciated. I am trying to get a Dell Latitude C840 up and running with a dual boot w2k/RH7.3.

Thanks.

[Anthaus]

I think you best option would be to upgrade your kernel. Kernel 2.4.18-3 doesn't get along too well with nvidia's drivers. It's not very difficult with RedHat's rpms.

You will need, AT LEAST, one of these (just one):

Intel (one processor)
ftp://updates.redhat.com/7.3/en/os/i686/kernel-2.4.18-18.7.x.i686.rpm

AMD (one processor)
ftp://updates.redhat.com/7.3/en/os/athlon/kernel-2.4.18-18.7.x.athlon.rpm

If you have two or more processors and/or a LOT of ram, let me know.

It would be convenient to upgrade or install the following as well:
ftp://updates.redhat.com/7.3/en/os/i386/kernel-source-2.4.18-18.7.x.i386.rpm

Type the usual, as root:

rpm -Uvh kernel.blah.blah.blah.rpm

Reboot (one of the few occasions linux has to). After that, you will be tossed to text mode. Login as root and install the nvidia drivers. (I suppose you already know the drill with XF86Config-4).

Since you're comfortable with src rpms, why not give the tar balls a chance? The release notes will take you step by step into the installation procedure.
Don't forget to install the drivers in text mode (X must NOT be running)!!!

[stevnb]

Anthaus, thank you for your reply. Unfortunately, I am unable to upgrade the kernel. The system in question is loaded with a linux load configured and by our development dept. for use with the point of sale software we manufacture and distribute. As such, I must use the system on the kernel level that they are distributing into the field. This means that I need to find an answer within the kernel level we are on.

I have tried both the src rpm and the tar.gz files with similar results.

Is there a nvidia driver known to work with the 2.4.18-3 kernel? If so, which one?

Thanks again for the help.

[Anthaus]

Ok. As you may already know, the nvidia drivers have two components: the kernel driver and the GLX driver (OpenGL).

The GLX is the same for all architectures:
http://download.nvidia.com/XFree86_40/1.0-4191/NVIDIA_GLX-1.0-4191.i386.rpm

There are various flavors for the kernel driver, depending on you architecture:

Intel (i686) one processor:
http://download.nvidia.com/XFree86_40/1.0-4191/NVIDIA_kernel-1.0-4191.rh73up.i686.rpm

AMD one processor:
http://download.nvidia.com/XFree86_40/1.0-4191/NVIDIA_kernel-1.0-4191.rh73up.athlon.rpm

If you have more than one processor and/or more than 4GB RAM, you will need different rpms. Look for them here:
http://www.nvidia.com/view.asp?IO=linux_display_ia32_1.0-4191

In general, you must look for the plain vanilla Red Hat 7.3 driver (without upgrades).

Now, I always recommend people to use the tar files, since they should work on all distros and architectures. But, in order to make them work, you need the kernel-source (the exact same version of your current kernel) and the make libraries.

[bwkaz]

Originally posted by stevnb
The system in question is loaded with a linux load configured and by our development dept. for use with the point of sale software we manufacture and distribute. As such, I must use the system on the kernel level that they are distributing into the field. Does this also mean you don't have root access to the machine? If so, no nVidia drivers will work, at all.

If not, if you do have root access, then take note that the nVidia drivers do in fact change your kernel!

When you load external kernel modules, your kernel changes. It may not affect anything, but it might. You would need to do more testing, which may not mean anything is out of the ordinary -- if this box is getting the drivers for testing purposes. If not, though, then you will have to consider that.

If you really don't care anymore, then just download the NVchooser.sh script that's on the download page of whatever driver you are going to install (3123 or 4191), and run it to tell you which files to download. No need to go checking CPU or whatever, it will do that. Simply sh NVchooser.sh to run it.

[stevnb]

Thank you again for the reply. I do have root access to the system. The loading of an external module for the video is not a concern for us. We simply need to remain on the same rev of the kernel for purposes of our software.

As for the NVchooser.sh, I have run this. I have then downloaded and installed the suggested package with the exact same results.

thanks,

[Anthaus]

As I said, kernel 2.4.18-3 is known as a troublesome rev. Lots of people complained that the nvidia drivers wouldn't work well or not at all with it.

Maybe you should put forward this to your employers.

[Andy Mecham]

Anthaus: Can you give me some links to these complaints? I'm not aware of any specific complaints with 2.4.18-3 that haven't been traced back to issues unrelated to kernel version.

stevnb: Can you verify that the md5sum of the files you've downloaded matches the one posted on the driver download page?

--andy

[bwkaz]

Wait a minute, "cannot stat nvidia.o, no such file or directory", that's not the actual error. That's just a message that the kernel module failed to finish building. There will be errors before that one that will be a pointer to the actual issue.

Verify the md5sum first, then try, instead of rpmbuild --rebuild whatever, rpmbuild --rebuild whatever >log.txt 2>&1, and post the log.txt file here. That will give a better idea of the problem...

[Wolfman [TWP]]

Sounds to me that he is missing the Kernel Source. As it will also be needed when you do a rpm --rebuild..

[stevnb]

Andy:
Yes, the md5sum checks out. The files I downloaded match the ones from the download site. The two files being recommended by NVchooser.sh are the NVIDIA_kernel-1.0-4191.src.rpm and the NVIDIA_GLX-1.0-4191.i386.rpm.

I do an rpm --rebuild NVIDIA_kernel-1.0-4191.src.rpm which results in error's. I have included with this posting a output of these messages.

Wolfman:
I do indeed have the kernel source so I dont believe that is an issue.

bwkaz:

I am doing as you suggested. This post should have the log.txt file from the rpm --rebuild of the kernel.src file.

[Wolfman [TWP]]

It's definately has something to do with the Kernel Source. I just ran the rebuild again, and I didn't get any of the 'can't find xxx.h file' errors that is listed in your *.log file.

You may need to check to see if these files are in you /usr/src/linux-XXX.XXX/include/asm directory.

I found that mine seemed to match, hence no can't find files errors..

[bwkaz]

Yeah, Wolfman is right. The problem is definitely that you don't have kernel-source installed.

The reason I can tell is this bit:

-I/usr/src/linux/include in the compile command. If you have a kernel-source package installed, that will be -I/lib/modules/<whatever your kernel version is>/build/include instead. build is a symlink to the actual kernel source directory (usually /usr/src/linux-<version>, but not always).

Once kernel-source is installed, this path should change. The asm/<whatever>.h files should also show up.

There is still a possibility that it might not work, though. If it doesn't, then you'll have to configure that kernel source directory. If you're using RH's kernel-source package (which it seems you will be, based on the kernel version), then that's pretty easy, just find the file in /usr/src/linux-2.4.18-3/configs that matches the kernel you installed and your hardware (arch, SMP-ness, etc., along with whether you installed e.g. the -secure kernel, or the -enterprise kernel, or whatever). Then cp that file to /usr/src/linux-2.4.18-3/.config so that it can be used in the next steps.

Then do a cd /usr/src/linux-2.4.18-3, then a make mrproper oldconfig dep, then try the nVidia driver again. The oldconfig part does the kernel configuration, based on the defaults found in the .config file that you just put there.

[stevnb]

O.K., I have finally achieved a full kernel source load on the system. I ran the rpm --rebuild and that went through successfully.

I then attempted to install the rpm that was created. This results in:

depmod: *** Unresolved symbols in /lib/modules/2.4.18-3_isi_1/kernel/drivers/video/nvidia.o

Failed to install nvidia.o

error: execution of %post scriptlet from NVIDIA_kernel-1.0-4191 failed, exit status 1

Please advise on course to follow.

Thanks

[stevnb]

further info:

depmod -ae shows:

depmod: *** Unresolved symbols in /lib/modules/2.4.18-3_isi_1/kernel/drivers/video/nvidia.o
depmod: register_chrdev_R02ca188e
depmod: __pollwait_R88d2ac43
depmod: irq_stat_R51b43718
depmod: remove_proc_entry_Rb4cc32b6
depmod: create_proc_entry_Ra60a30fa
depmod: pm_unregister_R94097bd6
depmod: proc_root_Rb08c6a10
depmod: init_mm_R114872b3
depmod: mem_map_R2d57b9df
depmod: pm_register_R027ebe5e
depmod: proc_root_driver_R9f42fa85

[bwkaz]

Originally posted by stevnb
2.4.18-3_isi_1 ???

Why is there an _isi_1 there?

modprobe nvidia looks under /lib/modules/$(uname -r)/ for an nvidia.o, and loads it. The fact that it's looking in 2.4.18-3_isi_1 says to me that the kernel version has changed -- and since RedHat doesn't distribute any _isi_1 kernels that I know of, it seems that someone has changed the top-level Makefile when your "standard" 2.4.18-3 was compiled.

Which brings up the possibility that they changed more than just the version. If they changed the configuration as well, then that explains the unresolved symbols. If someone changed the configuration of the kernel you're running, then the configuration that's stored in RedHat's configs/ directory will (obviously) not match up. You need a .config file that matches your running kernel at compile time, otherwise you'll see unresolved symbols like this.

Soooo...

You need to talk to whoever came up with this _isi_1 kernel, and get the .config file from the source directory, from them. Do what you did with the other config file with this one (that is, copy it to /usr/src/linux-2.4.18-3, then run make mrproper oldconfig dep), and try NVIDIA_kernel again.