Re: linux kernel and nvidia - never ending story

[Vincent Lefevre]

On 2022-03-07 18:25:54 -0500, Stefan Monnier wrote:
> Obviously it depends who you ask.  The kernel side doesn't consider
> itself to blame because they do expose a "stable" API which Nvidia
> could use.

But note that the Xorg API is not stable. The 340 branch can no longer
be used at least because of that (it is no longer maintained by Nvidia
and does not support the latest Xorg API). Anyway, it is affected by
security issues, so that even if it still worked, this would not be
secure.

My machines are not concerned by the 340 branch, but by the 390 branch,
whose EOL is at the end of the year. So this will be an issue the next
time Xorg modifies its API.

> I encourage you to file bug reports for that and try to help fix the
> problem as best as you can.  It might be worth trying it with a "fresh"
> install since I've heard several reports that Nouveau installations can
> be impacted by the installation of the proprietary Nvidia driver.

It is useless to report bugs against Nouveau. See for instance:
https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/-/issues/23
https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/-/issues/81
https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/-/issues/202
https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/-/issues/534

In particular, I reported Issue 23 in 2012 (almost 10 years ago)
and provided a script reproducing the bug, which was confirmed.
But there was no attempt to fix it.

-- 
Vincent Lefèvre  - Web: 
100% accessible validated (X)HTML - Blog: 
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)

Re: linux kernel and nvidia - never ending story

[Eric S Fraga]

On Tuesday,  8 Mar 2022 at 22:47, Richmond wrote:
> Now that I have it working I fear to change it.

And this is exactly my modus operandum.  Once I get a system to a stable
productive working state, I leave it alone (except for security issues).

-- 
Eric S Fraga with org 9.5.2 in Emacs 29.0.50 on Debian 11.2

Re: linux kernel and nvidia - never ending story

[Richmond]

didier gaumet  writes:

> Le mardi 08 mars 2022 à 12:12 +, Richmond a écrit :
> [...]
>> I don't know if there are
>> any distributions other than debian which still support kernel 4.
>
> A RHEL 8 clone (Almalinux, Rocky Linux...) should do: kernel is blocked
> to 4.18 until EOL in 2029 (end of full support may 2024, other support
> unspecified right now)

Thanks. It looks like there are plenty of distros I have never heard of
to choose from:

https://distrowatch.com/search.php?pkg=linux=similar=4.18=InLatest#pkgsearch
https://distrowatch.com/search.php?pkg=linux=similar=4.19=InLatest#pkgsearch

But it will be difficult because not all the ones I have tried in the
past have managed to cope with the additional tv-out feature that I want
to use with nvidia-settings. Now that I have it working I fear to change
it.

Re: linux kernel and nvidia - never ending story

[duh]

On 3/8/22 5:05 AM, to...@tuxteam.de wrote:

On Tue, Mar 08, 2022 at 10:46:35AM +0100, Hans wrote:
No, of course not. It's you to blame, for upgrading your system.



Hilarious!!! (Please bear with me for a moment because it is actually
tragic, but ...)

Why might it be hilarious?. Well, speaking in generalities, typically
when someone has a problem, "popular" comments

in the first response(s) is(are) to provide more information, upgrade to
the latest version of the software -- duh, I am a

little (or should I say very) dense. What is a person to do? Guess write
this reply to the list. Unfortunately most people

will probably not think it is funny.


Actually in the opening post, the writer said he was debating posting.
My response: Thank you!!

The problem he describes is just a specific case within the bigger
problem or the more general case. Often times,

as a person starts making changes, new problems proliferate faster than
they can be solved.


I recently came across a list of 10 code words.

    A couple I remember are:

        id10t

        ip8

Don't think that is an accurate assessment of the situation.

FWIW, there is significant merit to:

    DO NOT UPGRADE!!


Just a thought. Am not intending to hijack the thread. Just pointing out
that the current problem is a subset of

a larger problem.

Once again, thank you for the first post that started this thread. Some
of the viewpoints expressed in reply have been

both interesting and informative -- perhaps not an area with a lot of
occupants on a VENN diagram

Have a good day. Looking forward to continue following (in silence) the
rest of this discussion

Re: linux kernel and nvidia - never ending story

[Alexander V. Makartsev]

On 08.03.2022 14:40, Hans wrote:

Am Dienstag, 8. März 2022, 00:07:05 CET schrieb Alexander V. Makartsev:
Yes, I am sure, no other version is working. It is GeForce G210 and GeForce
G86m, both NEED 340xx.
In that case, I've never actually tried to do it, but since Debian 
support for 340xx version is officially EoL, I think it's ok to install 
officially supported driver¹ from Nvidia.


Or, if you feel adventurous, you can build a backported packages of 
nvidia legacy 340xx driver from buster-backports.
I've always used parts of this guide² and had successfully build 
packages from source³ package and DKMS module was also build 
successfully during package installation.

You can get source package using "dget":
    $ dget 
http://deb.debian.org/debian/pool/non-free/n/nvidia-graphics-drivers-legacy-340xx/nvidia-graphics-drivers-legacy-340xx_340.108-10~bpo10+1.dsc


No code patches were made by me, except changing "changelog" file via 
'dch' and bumping up version numbers.


    $ uname -a
Linux hostname 5.10.0-11-amd64 #1 SMP Debian 5.10.92-1 (2022-01-18) 
x86_64 GNU/Linux


=8<=8<=8<=8<=
...
Setting up nvidia-legacy-340xx-kernel-dkms (340.108-10~bpo11+1) ...
Loading new nvidia-legacy-340xx-340.108 DKMS files...
Building for 5.10.0-11-amd64
Building initial module for 5.10.0-11-amd64
Done.

nvidia-legacy-340xx.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.10.0-11-amd64/updates/dkms/

nvidia-legacy-340xx-uvm.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.10.0-11-amd64/updates/dkms/

depmod

DKMS: install completed.
...
=8<=8<=8<=8<=

Here is listing of all build packages for amd64 arch:
=8<=8<=8<=8<=
$ ls -1
libegl1-nvidia-legacy-340xx_340.108-10~bpo11+1_amd64.deb
libgl1-nvidia-legacy-340xx-glx_340.108-10~bpo11+1_amd64.deb
libgles1-nvidia-legacy-340xx_340.108-10~bpo11+1_amd64.deb
libgles2-nvidia-legacy-340xx_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-cfg1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-compiler_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-cuda1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-eglcore_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-encode1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-fbc1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-glcore_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-ifr1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-ml1_340.108-10~bpo11+1_amd64.deb
libnvidia-legacy-340xx-nvcuvid1_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-alternative_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-driver_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-driver-bin_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-driver-libs_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-kernel-dkms_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-kernel-source_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-kernel-support_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-opencl-icd_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-smi_340.108-10~bpo11+1_amd64.deb
nvidia-legacy-340xx-vdpau-driver_340.108-10~bpo11+1_amd64.deb
xserver-xorg-video-nvidia-legacy-340xx_340.108-10~bpo11+1_amd64.deb
=8<=8<=8<=8<=

Same listing is for i386 arch too.
To build 32-bit packages the process essentially the same, but you 
(AFAIK still) have to do it under 32-bit environment (chroot, lxc/lxd 
container, VM, etc).


I didn't tested this build on actual hardware yet, but it looks 
promising to me.

Good luck.


¹ https://www.nvidia.co.uk/Download/driverResults.aspx/156193/en-uk
² https://wiki.debian.org/SimpleBackportCreation
³ 
https://packages.debian.org/buster-backports/nvidia-legacy-340xx-kernel-dkms


--
With kindest regards, Alexander.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org
⠈⠳⣄

Re: linux kernel and nvidia - never ending story

[didier gaumet]

Le mardi 08 mars 2022 à 12:12 +, Richmond a écrit :
[...]
> I don't know if there are
> any distributions other than debian which still support kernel 4.

A RHEL 8 clone (Almalinux, Rocky Linux...) should do: kernel is blocked
to 4.18 until EOL in 2029 (end of full support may 2024, other support
unspecified right now)

Re: linux kernel and nvidia - never ending story

[Richmond]

Hans  writes:

> Dear list,
>
> how find the correct words, without being upset or stepping on
> someones feet.  But I believe, debian hates Nvidia, and debian does
> not want, to use Nvidia.
>
> I am now for a long time using debian and also using nvidia graphic
> cards for almost the same long time.
>
> But whenever debian ships a new kernel version, the proprietrary
> nvidia kernel modules can not be built. If lucky, there is a patch for
> it after months.
>
> Yes, modern Nvidia cards are supported, but using an older notebook
> you can not change the graphics card.
>
> But this is not a problem of Nvidia, not IMO it is a problem with the
> kernel developers. Suddenly, with a kernel the gcc was updated, oh,
> now the kernel module does not want it any more. Wtf? Or, with the new
> kernel, the kernel module crashes during building, but builds
> perfectly at the older kernel.
>
> And suddenly the kernel modul of nvidia disappears completely from the
> repo, problems solved? Get lost, you foolish users with old hardware,
> buy new hardware! What???
>
> Oh, and when someone says: Hey, use the nouveau driver, then tell him,
> nouveau is not working.
>
> I have several older notebooks, that my customers use. They worked
> perfectly with the proprietrary driver from Nvidia. But after update
> to bullseye, it was hardly get them running again. And why? They have
> an old graphics card in their notebooks, and they use Nvidia cards,
> specially the legacy 340xx.
>
> But:
>
> 1. no problem, Install nvidia kernel 340xx, oh no, it is no more in
> the repo, but
>
> 2. no problem, hey, use nouveau, oh no, nouveau crashes and freezes X,
> but
>
> 3. no problem, build just the downloaded 340xx from buster, oh no,
> does not build, wrong gcc installed, gcc to new, but
>
> 4. no problem, just downgrade gcc to the old one, oh no, many other
> packages need to be deinstalled, too, but
>
> 5. no problem, just do it, oh no, does not build with the latest
> kernel, but
>
> 6. no problem, just downgrade the kernel, too, oh no, no kernel from
> bullseye is working, but
>
> 7. no problem, just reactivate buster and install latest kernel from
> buster, and oh yes,
>
> 8. old kernel from buster let build 340xx, but oh no, kernel old...
>
> Well, I and these procedures are now accompanies me since years. New
> kernel, and building fails. Youu feel lost, you feel anger, can you
> believe me?
>
> In earlier times, debian potato and so, there were always prebuild
> kernel modules for graphic cards, Nvidia or AMD or whatever. Today
> these are gone, and people with older cards are lost. IMO here debian
> lost a lots of its quality.
>
> I thought a long time, if I should write this, and maybe I have not
> found the correct words. I do not want to harsh anyone or attack
> anyone, you know what I mean.
>
> But I felt in my heart, I had to say it.
>
> Please apologize, if someone is feeling agry about me now, this was
> not intended. And thanks for reading this.
>

I am in the same position as you. One of my laptops has an old Nvidia
card, so it is stuck on debian 10. But this isn't unique to debian, it
also applies to OpenSUSE which I also have on there dual booting, and
that cannot be upgraded either. It has to stay on the version 4 kernel,
and from what I read at the time this is because Nvidia will not support
later kernels 5+ . So it is ok until long term support ends, but then I
don't know what to do. Nouveau is unstable. I don't know if there are
any distributions other than debian which still support kernel 4.

Re: linux kernel and nvidia - never ending story

[tomas]

On Tue, Mar 08, 2022 at 06:25:54AM -0500, Dan Ritter wrote:
> to...@tuxteam.de wrote: 

[...]

> > (Sorry for the irony, but see, if NVIDIA's sources reference parts of
> > the kernel headers which aren't guaranteed to stay stable, well...)

> There is also the problem that NVidia has no incentive to keep purchasers
> of their hardware happy over the long term [...]

Exactly.

> Now, could NVidia have chosen another path? Certainly [...]

Once I understood that, I decided to not buy NVIDIA.

> I don't blame Hans for picking the wrong vendor. I did that too
> for a while. In that situation, talking to the Nouveau people is
> probably the best bet, other than buying a new graphics card
> from someone else -- and that's expensive this year.

No, nobody's to blame for that. Since a functional computer almost
always implies some "wrong vendor" at some place. I'm a bit miffed
by Hans throwing blame at those who deserve it the least (and he
even anticipates it: that makes it a bit stranger). Perhaps because
it's more difficult yelling at NVIDIA? I don't know.

Cheers
-- 
t


signature.asc
Description: PGP signature

Re: linux kernel and nvidia - never ending story

[Dan Ritter]

to...@tuxteam.de wrote: 
> On Tue, Mar 08, 2022 at 10:46:35AM +0100, Hans wrote:
> > Am Dienstag, 8. März 2022, 00:25:54 CET schrieb Stefan Monnier:
> > No, you are completzely wrong! It is not Nvidia to blam,e when COMPILING 
> > fails! [...]
> 
> No, of course not. It's you to blame, for upgrading your system.
> 
> Just stay with your old kernel, header files and toolchain and all will
> be well.
> 
> (Sorry for the irony, but see, if NVIDIA's sources reference parts of
> the kernel headers which aren't guaranteed to stay stable, well...)
>

There is also the problem that NVidia has no incentive to keep purchasers
of their hardware happy over the long term: the people who buy low-end
cards are not worth the profit margin dip caused by supporting them;
the people who buy high-end cards will continue to buy whatever the
magazines say are fastest this season, and businesses mostly run Windows.

Now, could NVidia have chosen another path? Certainly. There are
at least two more existence proofs: Intel decided that they
would open-source their drivers and not even pretend to be
competitive: for the last decade, nobody has purchased Intel
graphics cards. People buy Intel CPUs and get a just-good-enough
GPU embedded.

And over at AMD, they open-sourced their graphics drivers to
call functions in a big firmware blob, so the drivers need
minimal work each year.

I don't blame Hans for picking the wrong vendor. I did that too
for a while. In that situation, talking to the Nouveau people is
probably the best bet, other than buying a new graphics card
from someone else -- and that's expensive this year.

-dsr-

Re: linux kernel and nvidia - never ending story

[tomas]

On Tue, Mar 08, 2022 at 10:46:35AM +0100, Hans wrote:
> Am Dienstag, 8. März 2022, 00:25:54 CET schrieb Stefan Monnier:
> No, you are completzely wrong! It is not Nvidia to blam,e when COMPILING 
> fails! [...]

No, of course not. It's you to blame, for upgrading your system.

Just stay with your old kernel, header files and toolchain and all will
be well.

(Sorry for the irony, but see, if NVIDIA's sources reference parts of
the kernel headers which aren't guaranteed to stay stable, well...)

Cheers
-- 
t


signature.asc
Description: PGP signature

Re: linux kernel and nvidia - never ending story

[Hans]

Am Dienstag, 8. März 2022, 00:25:54 CET schrieb Stefan Monnier:
No, you are completzely wrong! It is not Nvidia to blam,e when COMPILING 
fails! It is NOT Nvidia to blame, when the packages are REMOVED from the repo, 
because your KERNEL breaks the compiling! And it is NOT Nvidia to blame, when 
the nouveau driver CRASHES! And it is NOT Nvidia to blame, when the new kernel 
BREAKS the compiler!

NO, it is not Nvidia to blame, that is just lame excuse. I know exactly,. whom 
to blame, but I do not tell it here.

Best

Hans 
> > Oh, no, this I see not this way. When there is a new kernel shipped,
> > I expect  that this does not break the system!
> 
> That's indeed what the kernel team aims for.
> 
> > I expect that at least the code, that is already existent, builds.
> 
> Problem is: even the "minor" changes of kernel bug fixes done within
> a stable release ... change the code.  By definition.
> 
> So they do incur the risk of breaking existing code.  To avoid this
> problem, bug fixes make as few changes as possible, but when the third
> party code relies on details considered internal to the kernel, this may
> sometimes break.
> 
> The kernel team would be much happier if they could work *with* Nvidia
> to avoid those problems.  But that would require Nvidia be more open
> about its code.
> 
> > This does NOT work., as I told.  The drivers from the nvidia site do
> > no more  build with a new kernel version.  Who is then to blame?
> 
> Obviously it depends who you ask.  The kernel side doesn't consider
> itself to blame because they do expose a "stable" API which Nvidia
> could use.
> 
> > It is the kernel that changed, who breaks the system, not the driver,
> > who has NOT chenged!
> 
> Not necessarily.  Maybe the part that did not change relied on a bug in
> the kernel, so there might have been no easy way to fix the bug without
> breaking the third party code.
> 
> >> 5. I only now realize that your card is actually too OLD, sorry. What
> >> should I say, the latest release of the 340 branch from nvidia is dating
> >> back to 2019.12.23. If they dont support their older products anymore
> >> themselves, do you expect Debian to hack their closed source driver?
> > 
> > And what? 3 years is too old???
> 
> Tell that to Nvidia.
> I have no such problems with my >10 year old AMD and Intel graphics cards.
> 
> > I do not expect the drivers to be improved,  but I expect, that the
> > drivers can be used further on!
> 
> The kernel changes.  If you want a driver to "survive" it will need to
> adapt.  Commercial interest will inevitably wane after a few years, so
> the only way it can last more than 3-5 years is if the code is Free
> Software so other people can pick up maintenance.
> 
> > This is not to few to be expected, isn't it?
> 
> Nowadays I consider a computer's expected lifetime to be at least 10 years.
> 
> >> 6. If nvidia would only be a little bit more cooperative, nouveau would
> >> be in a much better state. I found it usable for older cards, although
> >> the prorietary driver is of course much better in terms of performance
> >> and power saving.
> > 
> > As I wrote, please read it again: The nouveau driver does not work for
> > this
> > card, X is totally unstable and so often freezing, that it can not be
> > used.
> 
> I encourage you to file bug reports for that and try to help fix the
> problem as best as you can.  It might be worth trying it with a "fresh"
> install since I've heard several reports that Nouveau installations can
> be impacted by the installation of the proprietary Nvidia driver.
> 
> > Did I tell, it is debian/stable?
> 
> FWIW, the "stable" in the name doesn't refer to a claim that the
> resulting systems run reliably, but that they try to avoid behavioral
> changes within a given stable release.
> 
> Obviously, they're having trouble reaching this goal w.r.t compatibility
> with proprietary third party kernel modules.
> 
> >> 7. Intel (what I use these days) and AMD support Linux in a much cleaner
> >> way, avoiding many of the problems with the nvidia blob.
> > 
> > Yes, may be, but who cares.  As I wrote, it is not possible, to change the
> > graphics card in noterbooks.
> 
> No, but you can choose the card in the next laptop you buy (and in
> those you recommend others).
> 
> > Why shhould one? Shall I tell all customers, "Hey, throw away your
> > 5 or 8 year  old stuff!", which yesterday worked well and today, after
> > an upgrade does no  more, because a simple driver can not be compiled?
> > No, that is the wrong way!
> 
> Agreed.  Much better is to work with those who try to keep this hardware
> working with the rest of the software.  In the case of Nvidia cards,
> this is the Nouveau team.
> 
> Noone but Nvidia can do that for the proprietary driver.
> 
> 
> Stefan

Re: linux kernel and nvidia - never ending story

[Hans]

Am Dienstag, 8. März 2022, 00:07:05 CET schrieb Alexander V. Makartsev:
Yes, I am sure, no other version is working. It is GeForce G210 and GeForce 
G86m, both NEED 340xx.

Best 

Hans

> On 07.03.2022 23:49, Hans wrote:
> > Dear list,
> > 
> > how find the correct words, without being upset or stepping on someones
> > feet. But I believe, debian hates Nvidia, and debian does not want, to
> > use Nvidia.
> > 
> > I am now for a long time using debian and also using nvidia graphic cards
> > for almost the same long time.
> > 
> > But whenever debian ships a new kernel version, the proprietrary nvidia
> > kernel modules can not be built. If lucky, there is a patch for it after
> > months.
> > 
> > Yes, modern Nvidia cards are supported, but using an older notebook you
> > can
> > not change the graphics card.
> > ...
> > I have several older notebooks, that my customers use. They worked
> > perfectly with the proprietrary driver from Nvidia. But after update to
> > bullseye, it was hardly get them running again. And why? They have an old
> > graphics card in their notebooks, and they use Nvidia cards, specially
> > the legacy 340xx.
> Out of curiosity, can you name a model of nvidia GPU(-s) you have
> trouble with?
> Are you sure you need legacy 340xx driver specifically? That version is
> for hardware that was released in 2009-2010 and older.
> I think there is a chance you can install legacy 390xx version instead.
> You can use 'nvidia-detect' program to check your hardware and get a
> driver version recommendation.

Re: linux kernel and nvidia - never ending story

[Felix Miata]

Stefan Monnier composed on 2022-03-07 18:25 (UTC-0500):

> I've heard several reports that Nouveau installations can
> be impacted by the installation of the proprietary Nvidia driver.

I believe this happens due to incomplete purging of all effects of having
attempted to install, or succeeding to install, then "removing", NVidia's
proprietary drivers. I have a bunch of old NVidia cards, 8 years old and older.
All are running purely on FOSS:

1-the nouveau kernel device driver/module

2-the modesetting DIX display device driver (usually; optionally: nouveau DDX)

#2 is the upstream default for AMD, Intel and NVidia GPUs. Unlike the
reverse-engineered, old, "experimental" and optional nouveau DDX display driver
from package xserver-xorg-video-nouveau, the modesetting DIX is newer 
technology,
and less dependent on device specifications NVidia withholds from FOSS 
developers.

Other than for the issue reported at:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1006907
the FOSS drivers nouveau (module)/modesetting (DIX)/nouveau (DDX) provided by
Debian are fully suited to the needs of all my NVidia GPUs.
-- 
Evolution as taught in public schools is, like religion,
based on faith, not based on science.

 Team OS/2 ** Reg. Linux User #211409 ** a11y rocks!

Felix Miata

Re: linux kernel and nvidia - never ending story

[Alexander V. Makartsev]

On 07.03.2022 23:49, Hans wrote:

Dear list,

how find the correct words, without being upset or stepping on someones feet.
But I believe, debian hates Nvidia, and debian does not want, to use Nvidia.

I am now for a long time using debian and also using nvidia graphic cards for
almost the same long time.

But whenever debian ships a new kernel version, the proprietrary nvidia kernel
modules can not be built. If lucky, there is a patch for it after months.

Yes, modern Nvidia cards are supported, but using an older notebook you can
not change the graphics card.
...
I have several older notebooks, that my customers use. They worked perfectly
with the proprietrary driver from Nvidia. But after update to bullseye, it was
hardly get them running again. And why? They have an old graphics card in
their notebooks, and they use Nvidia cards, specially the legacy 340xx.
Out of curiosity, can you name a model of nvidia GPU(-s) you have 
trouble with?
Are you sure you need legacy 340xx driver specifically? That version is 
for hardware that was released in 2009-2010 and older.

I think there is a chance you can install legacy 390xx version instead.
You can use 'nvidia-detect' program to check your hardware and get a 
driver version recommendation.



--
With kindest regards, Alexander.

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system
⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org
⠈⠳⣄

Re: linux kernel and nvidia - never ending story

[Hans]

Am Montag, 7. März 2022, 20:46:40 CET schrieb Christian Britz:
Hi Christian, 

I feel to answer your post and have to correct some of your essays.
> Hello Hans,
> 
> I understand your frustration, I was also frustrated sometimes back when
> I used nVidia, but IMO Debian is not to blame here, the situation is
> completely the responsibility of nVidia.
> 
> 0. After reading your post again, I realize that sections 1 to 4 of my
> reply do not really apply to your situation. I leave them in my text
> though, because in my opionion they describe the general problem with
> nvidia quite well.
> 

Yes, good idea.

> 1. You are probably not talking about Debian stable and the kernel that
> ships with it. I have not checked, but the binary nVidia packages almost
> certainly have not disappeared from Debian stable non-free within a
> release cycle. Nothing should have broken with the latest dot kernel
> update. If you have problems, please specify further. Testing/unstable
> may break at any time, you have to deal with that, especially when using
> non-free software, which can only be fixed to a certain amount by the
> maintainers.

No, I DO talk about debian/stable! And when I mean stable, I talk about 
bullseye. I expect bullseye to work.
> 
> 2. If you are talking about kernels newer than the one in Debian stable,
> the problem is a problem by design that can and will happen with every
> Linux distribution. The reason is simply that the kernel interfaces
> which nVidia uses are not stable and were never intended to be used by
> non-free software. Eventually, nVidia will adapt their interface module
> and release a new driver package. Eventually, this will be integrated in
> Debian non-free. It is a race between the kernel developers and the
> nvidia developers, which nvidia never can win. If at all, you can blame
> Linus and the developers for not providing a stable interface for
> proprietary non-free software. I fear they will laugh at you, at best.
> Some kernel developers even have the opinion that it is a GPL violation
> what nvidia does.

Oh, no, this I see not this way. When there is a new kernel shipped, I expect 
that this does not break the system! I expect that at least the code, that is 
already existent, builds. This is not too much expected! Why is it called 
stable???


> 
> 3. If you are not satisfied how fast updated nvidia drivers are
> integrated into testing/unstable, you could always install the drivers
> manually from the website of nvidia. It is not very hard to do if you
> are used to Linux.
> 

This does NOT work., as I told. The drivers from the nvidia site do no more 
build with a new kernel version. Who is then to blame? Those, who change the 
kernel or those who do not change the drivers?  It is the kernel that changed, 
who breaks the system, not the driver, who has NOT chenged!

> 4. Given the circumstances, I feel the nvidia integration in Debian
> quite smooth, on stable it works almost out of the box (if your card is
> not too new, but that is a general issue with the stable concept of Debian).
Agreed! If you get the driver compiled, then it is working very well, if you 
get it compiled. 

> 
> 5. I only now realize that your card is actually too OLD, sorry. What
> should I say, the latest release of the 340 branch from nvidia is dating
> back to 2019.12.23. If they dont support their older products anymore
> themselves, do you expect Debian to hack their closed source driver?
> 

And what? 3 years is too old??? I do not expect the drivers to be improved, 
but I expect, that the drivers can be used further on! This is not to few to 
be expected, isn't it?

> 6. If nvidia would only be a little bit more cooperative, nouveau would
> be in a much better state. I found it usable for older cards, although
> the prorietary driver is of course much better in terms of performance
> and power saving.
> 

As I wrote, please read it again: The nouveau driver does not work for this 
card, X is totally unstable and so often freezing, that it can not be used. 
Did I tell, it is debian/stable? 

> 7. Intel (what I use these days) and AMD support Linux in a much cleaner
> way, avoiding many of the problems with the nvidia blob.
> 
Yes, may be, but who cares. As I wrote, it is not possible, to change the 
graphics card in noterbooks.

> 8. If you buy a system with the goal to use it over many years, I can
> only reccomend to choose hardware components wich can be well supported
> by free drivers. Please support companies that support Linux well, like
> Intel and AMD.
> 

Why shhould one? Shall I tell all customers, "Hey, throw away your 5 or 8 year 
old stuff!", which yesterday worked well and today, after an upgrade does no 
more, because a simple driver can not be compiled? No, that is the wrong way!

The whole story with Nvidia-modules is shit, and even when I no come into the 
doghouse, IMO it is NOT the fault of Nvidia, how the developers always try to 
tell! Someone ist just closing the 

Re: linux kernel and nvidia - never ending story

[Christian Britz]

Hello Hans,

I understand your frustration, I was also frustrated sometimes back when
I used nVidia, but IMO Debian is not to blame here, the situation is
completely the responsibility of nVidia.

0. After reading your post again, I realize that sections 1 to 4 of my
reply do not really apply to your situation. I leave them in my text
though, because in my opionion they describe the general problem with
nvidia quite well.

1. You are probably not talking about Debian stable and the kernel that
ships with it. I have not checked, but the binary nVidia packages almost
certainly have not disappeared from Debian stable non-free within a
release cycle. Nothing should have broken with the latest dot kernel
update. If you have problems, please specify further. Testing/unstable
may break at any time, you have to deal with that, especially when using
non-free software, which can only be fixed to a certain amount by the
maintainers.

2. If you are talking about kernels newer than the one in Debian stable,
the problem is a problem by design that can and will happen with every
Linux distribution. The reason is simply that the kernel interfaces
which nVidia uses are not stable and were never intended to be used by
non-free software. Eventually, nVidia will adapt their interface module
and release a new driver package. Eventually, this will be integrated in
Debian non-free. It is a race between the kernel developers and the
nvidia developers, which nvidia never can win. If at all, you can blame
Linus and the developers for not providing a stable interface for
proprietary non-free software. I fear they will laugh at you, at best.
Some kernel developers even have the opinion that it is a GPL violation
what nvidia does.

3. If you are not satisfied how fast updated nvidia drivers are
integrated into testing/unstable, you could always install the drivers
manually from the website of nvidia. It is not very hard to do if you
are used to Linux.

4. Given the circumstances, I feel the nvidia integration in Debian
quite smooth, on stable it works almost out of the box (if your card is
not too new, but that is a general issue with the stable concept of Debian).

5. I only now realize that your card is actually too OLD, sorry. What
should I say, the latest release of the 340 branch from nvidia is dating
back to 2019.12.23. If they dont support their older products anymore
themselves, do you expect Debian to hack their closed source driver?

6. If nvidia would only be a little bit more cooperative, nouveau would
be in a much better state. I found it usable for older cards, although
the prorietary driver is of course much better in terms of performance
and power saving.

7. Intel (what I use these days) and AMD support Linux in a much cleaner
way, avoiding many of the problems with the nvidia blob.

8. If you buy a system with the goal to use it over many years, I can
only reccomend to choose hardware components wich can be well supported
by free drivers. Please support companies that support Linux well, like
Intel and AMD.

Best Regards,
Christian



On 2022-03-07 19:49 UTC+0100, Hans wrote:
> Dear list,
> 
> how find the correct words, without being upset or stepping on someones feet.
> But I believe, debian hates Nvidia, and debian does not want, to use Nvidia.
> 
> I am now for a long time using debian and also using nvidia graphic cards for 
> almost the same long time. 
> 
> But whenever debian ships a new kernel version, the proprietrary nvidia 
> kernel 
> modules can not be built. If lucky, there is a patch for it after months.
> 
> Yes, modern Nvidia cards are supported, but using an older notebook you can 
> not change the graphics card. 
> 
> But this is not a problem of Nvidia, not IMO it is a problem with the kernel 
> developers. Suddenly, with a  kernel the gcc was updated, oh, now the kernel 
> module does not want it any more. Wtf? Or, with the new kernel, the kernel 
> module crashes during building, but builds perfectly at the older kernel.
> 
> And suddenly the kernel modul of nvidia disappears completely from the repo, 
> problems solved? Get lost, you foolish users with old hardware, buy new 
> hardware! What???
> 
> Oh, and when someone says: Hey, use the nouveau driver, then tell him, 
> nouveau 
> is not working.
> 
> I have several older notebooks, that my customers use. They worked perfectly 
> with the proprietrary driver from Nvidia. But after update to bullseye, it 
> was 
> hardly get them running again. And why? They have an old graphics card in 
> their notebooks, and they use Nvidia cards, specially the legacy 340xx.
> 
> But: 
> 
> 1. no problem, Install nvidia kernel 340xx, oh no, it is no more in the repo, 
> but 
> 
> 2. no problem, hey, use nouveau, oh no, nouveau crashes and freezes X, but
> 
> 3. no problem, build just the downloaded 340xx from buster, oh no, does not 
> build, wrong gcc installed, gcc to new, but
> 
> 4. no problem, just downgrade gcc to the old one, oh no, many other