[HOWTO] make a xzm module driver for NVIDIA video card

[S4m4n]

Dear Rava,
I knew that you have that driver prepared. But since I have accomplished nvidia driver installation on other distros I thought that I can handle it on porteus too.
It is a mystery to me that despite I took the same kernel, devel, initrd, vmlinuz from you on same distro and followed the same steps explained on your guide, The final result was disappointing. you are probably a wizard Now your provided driver is working smoothly:
https://imgur.com/a/QKKE60E

Thanks for your help.

[Rava]

It is a mystery to me that despite I took the same kernel, devel, initrd, vmlinuz from you on same distro and followed the same steps explained on your guide, The final result was disappointing. you are probably a wizard Now your provided driver is working smoothly:

I am sure no wizard. If I were I would able to successfully compile NVIDIA-Linux-x86_64-304.137.run using the same kernel.
But I cannot, and the reason being that they stopped supporting 304.x a while ago; donald explained above why:

The 304 driver is out of support.
AFAIK it is not possible getting it to run on a system since xorg-server 1.20
Porteus 4 with xorg 1.19 will be the last one.

nvidia-304.137-kernel 4.16.3-x86_64 (porteus 4)
(built without the 32-bit libraries.)
http://filehorst.de/d/cjsidhnG

_______________________________________________

@donald
Would it make sense to tweak Porteus 5.0rc1 into using kernel 4.16.3 and xorg-server 1.19?
How could that be done? I presume I would have at least alter 002-xorg.xzm.
Maybe I would also have to alter 000-kernel.xzm and 001-core.xzm.
Has anyone ever done this before?

BTW, I already am using the above for Porteus 4.0 as seen by the filename and the md5sum:

Code: Select all

424afaeeacbe727398306ccfca66cae6  /mnt/sda1/Porteus_4.0/porteus/base/010-nvidia-304.137-k.4.16.3-x86_64-don.xzm

CAVE! The discussion about downgrading Xorg to 1.19 was split and moved here:
[HOWTO] downgrade Xorg to 1.19 (Post by donald #76696)
Please reply about that topic in the thread linked above.

[Rava]

Blaze, I tried compiling the NV driver V340.108 for 5.0rc1-pre-rc2 and your kernel 5.4.30. I also used your 05-devel:

Code: Select all

bda976e458a5762b461a22b239ae804a  05-devel_blaze_2020-04-04_gcc-9.3.0.xzm

These are the reasons it failed:

Code: Select all

-> Cleaning kernel module build directory.
   executing: 'cd ./kernel; /usr/bin/make clean'...
   /usr/bin/make: error while loading shared libraries: libguile-3.0.so.1: cannot open shared object file: No such file or directory
-> Building NVIDIA kernel module:
   executing: 'cd ./kernel; /usr/bin/make module SYSSRC=/lib/modules/5.4.30-porteus/source SYSOUT=/lib/modules/5.4.30-porteus/build NV_BUILD_MODULE_INSTANCES='...
   /usr/bin/make: error while loading shared libraries: libguile-3.0.so.1: cannot open shared object file: No such file or directory
-> Error.
ERROR: Unable to build the NVIDIA kernel module.
ERROR: Installation has failed.  Please see the file '/var/log/nvidia-installer.log' for details.  You may find suggestions on fixing installation problems in the README available on the Linux driver download page at www.nvidia.com.

Do you have any idea why that failed?

Your 05-devel and the 000-kernel.xzm should work together, or not?

000-kernel.xzm md5sum is d33c33ba339f0a94844dd73e6749daee. But it should be okay since the kernel tar archive md5sum was okay.

And Porteus usually refuses to activate a corrupted module.

[beny]

see your devel package if the guile version match what make want to build driver.

[Rava]

see your devel package if the guile version match what make want to build driver.

I asked Blaze, who made the kernel tar ( containing crippled_sources-5.4.30-64bit.xzm - vmlinuz - 000-kernel.xzm ), which 05-devel package to use and he told me the one I used is the correct one.

Therefore I addressed the question to him…

[Blaze]

Rava, in my Slackware-current base I have this package

Code: Select all

# ls /var/lib/pkgtools/packages/*guile*
/var/lib/pkgtools/packages/guile-3.0.2-x86_64-1

guile is not a part of devel package

Code: Select all

# ls /mnt/sdb1/porteus/base/001-core/var/lib/pkgtools/packages/*guile*
/mnt/sdb1/porteus/base/001-core/var/lib/pkgtools/packages/guile-3.0.2-x86_64-1

001-core

You can download it manually http://slackware.cs.utah.edu/pub/slackw ... 6_64-1.txz

[beny]

hi blaze /root/slackware/slackware64-current/slackware64/d/guile-3.0.2-x86_64-1.txt better to adds to devel package.....

[Rava]

Yay! Success!
Porteus 5.0rc1 NVidia 340.108 driver for older hardware for kernel 5.4.30

[rych]

6.0.7

Thanks, I'm using this kernel now, and some early-boot-screen minor errors went away. Not a big deal perhaps, still feels like an update that works.

However. I'm now going to boot on a machine with a 1050Ti Nvidia GPU and got myself a few 08-nvidia modules and the latest from this forum wants to have the 5.18.8 kernel, at least it has the following folder inside:

Code: Select all

ls 08-nvidia-515.76-k.5.18.8-porteus-v5.0-x86_64/lib/modules
5.18.8-porteus/

whereas my system now has this folder instead:

Code: Select all

ls /lib/modules
6.0.7-porteus/

It won't work then I guess? What should I do? Revert back to the old kernel? Rename folders inside the nvidia xzm?

[ncmprhnsbl]

It won't work then I guess? What should I do? Revert back to the old kernel? Rename folders inside the nvidia xzm?

either revert to the kernel it's built for or rebuild the nvidia driver for the new kernel are the options.
see here [HOWTO] make a xzm module driver for NVIDIA video card for the method and script to build the driver module..

[rych]

or rebuild the nvidia driver for the new kernel

That looks like a rather involved procedure, not sure I can handle it within a reasonable timeframe. Hm, a difficult choice. Why must GPU driver be kernel-version specific -- doesn't seem to be the case for other Linux distros or Windows -- but I'm a novice. Thanks for a quick answer anyway.

[ncmprhnsbl]

Why must GPU driver be kernel-version specific -- doesn't seem to be the case for other Linux distros

i'm sure it's very much the case with other linux distros..
at some point, now that nvidia have open sourced their driver, it may even happen that it becomes added to the kernel itself, like other opensource drivers, but that might be a pipedream and i'd need a new card for that anyway.
nvidia has a terrible record of not supporting legacy hardware.
more likely i'd go for an amd card in future. just couldn't pass up an old quadro k4000[ for a hundred bucks :p

That looks like a rather involved procedure

Rava probly makes it look harder than it is

[Rava]

hi Rava why do not share the patch link?

I have to look it up first which could take some time since I did not work on that issue for many months.

me thinks donald would be quicker in getting you the patch link, though.

[rych]

it may even happen that it becomes added to the kernel itself, like other opensource drivers

Yes, this feels like a part of the kernel then. At least, it's my first time finding kernel updates posted in this thread "breaking" things due to incompatibility (with nvidia driver that is). I have a question: must the nvidia driver be built on the machine that actually has the target nvidia GPU?! I understand the compiling system has to have the same kernel version it's meant for, but if it doesn't actually have to have the GPU, then in principle it could be a part of the "Porteus Kernel Builder"

[rych]

Must the nvidia driver be built on the machine that actually has the target nvidia GPU?!