Gentoo Forums
Gentoo Forums
Gentoo Forums
Quick Search: in
nvidia-drivers-340.108
View unanswered posts
View posts from last 24 hours

Goto page Previous  1, 2, 3, 4, 5  Next  
Reply to topic    Gentoo Forums Forum Index Kernel & Hardware
View previous topic :: View next topic  
Author Message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Fri Jan 29, 2021 7:59 pm    Post subject: Reply with quote

I'm sorry I haven't been around, I totally forgot about this thread, my bad :roll:

I updated a few things:
  • Patches up to linux 5.10
  • Restored nvidia-uvm module: it compiles, I don't know if it works though cause I don't use CUDA
  • Script to correctly VT switch after resuming from suspend when using elogind
  • Removed the udev rule to call nvidia-smi on boot because it kills TTYs on my laptop; if you need /dev/nvidia* to be initialized before Xorg starts, enable the nvidia-smi service


Poopman wrote:
lspci -k

https://pastebin.com/Hm5rATd1

Seeing nvidia_drm as an available module for your card is really suspicious. Did you mask nvidia-drivers versions other than 340?
Back to top
View user's profile Send private message
dmpogo
Advocate
Advocate


Joined: 02 Sep 2004
Posts: 3446
Location: Canada

PostPosted: Fri Jan 29, 2021 10:06 pm    Post subject: Reply with quote

Shibotto wrote:
I'm sorry I haven't been around, I totally forgot about this thread, my bad :roll:

I updated a few things:
  • Patches up to linux 5.10
  • Restored nvidia-uvm module: it compiles, I don't know if it works though cause I don't use CUDA
  • Script to correctly VT switch after resuming from suspend when using elogind
  • Removed the udev rule to call nvidia-smi on boot because it kills TTYs on my laptop; if you need /dev/nvidia* to be initialized before Xorg starts, enable the nvidia-smi service




Thanks a lot for your effort ! I have one question, which comes from even before nvidia-drivers-340 ebuild was forked

when I run it (successfully) it gives
Code:

>>> Running pre-merge checks for x11-drivers/nvidia-drivers-340.108-r3
 * Determining the location of the kernel source code
 * Found kernel source directory:
 *     /usr/src/linux
 * Found sources for kernel version:
 *     4.14.209-gentoo
 * Checking for suitable kernel configuration options...
 *   CONFIG_DRM:         is not set when it should be.
 *   CONFIG_DRM_KMS_HELPER:      is not set when it should be.
 * Please check to make sure these options are set correctly.
 * Failure to do so may cause unexpected problems.


I never had these options set as far as I was using my GFORCE 8500 card, through generations of nvidia-drivers and kernels since 2007. Moreover, intially it was recommended to disable DRM in kernel if nvidia-drivers are used.
So do you know, is this warning actually applicable ? Should I enable them ? It all runs fine without these kernel options. I am currently on 4.14.209-gentoo kernel.
Back to top
View user's profile Send private message
psycho
Guru
Guru


Joined: 22 Jun 2007
Posts: 544
Location: New Zealand

PostPosted: Fri Jan 29, 2021 11:39 pm    Post subject: Reply with quote

I think I've built a relatively *recent* nvidia kernel module with drm (I'm remember thinking "huh, isn't that only for nouveau?" but then the nvidia-drivers worked nicely with drm in the kernel)...but it also looks like it was around nvidia 364 that the proprietary driver added drm support, so requiring it for 340 does look like it could be a mistake.
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Fri Jan 29, 2021 11:57 pm    Post subject: Reply with quote

Unless you use nvidia-drm.modeset=1 I don't think these options matter no matter the driver version. Originally the kernel config check was added to all ebuilds as a forced requirement but it made me open a bug so it'd (at least) be optional, and so it is and I ignore the message every time.

I don't use 340 anymore (I use nouveau on my old boxes) but I don't even install/load the nvidia_drm module on 360.39 either.

Would reconsider if I wanted to use wayland, but that's not something on the table for me at the moment with things I use.
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Sat Jan 30, 2021 12:15 am    Post subject: Reply with quote

Yes, I just found Ionen report. Prior to this change the options checked where:
Code:
CONFIG_CHECK="!DEBUG_MUTEXES ~!LOCKDEP ~MTRR ~SYSVIPC ~ZONE_DMA"

Since I'm not exactly sure of what is really needed and what is not (and I doubt lazy Nvidia made significant changes since 340 hit legacy) I think I will restore that line.
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Sat Jan 30, 2021 12:27 am    Post subject: Reply with quote

Yeah that makes sense, probably safer to also keep that ZONE_DMA I see (was dropped near those updates).

It's "optional" (I have it disabled with 360.39) but only if you use IOMMU instead (does that even work with 340? odds are old hardware wouldn't have good support either way).
Back to top
View user's profile Send private message
dmpogo
Advocate
Advocate


Joined: 02 Sep 2004
Posts: 3446
Location: Canada

PostPosted: Sat Jan 30, 2021 5:55 am    Post subject: Reply with quote

Shibotto wrote:
I'm sorry I haven't been around, I totally forgot about this thread, my bad :roll:

I updated a few things:
  • Patches up to linux 5.10
  • Restored nvidia-uvm module: it compiles, I don't know if it works though cause I don't use CUDA
  • Script to correctly VT switch after resuming from suspend when using elogind
  • Removed the udev rule to call nvidia-smi on boot because it kills TTYs on my laptop; if you need /dev/nvidia* to be initialized before Xorg starts, enable the nvidia-smi service




Sadly, actually an update r2 to r3 did not compile at the end (and r2 was installed for the same kernel just two days ago with no problems). uvm USE flag is disabled

I got [url] https://pastebin.com/embed_js/pqCmNZuJ [/url]

one difference that happened in last two days since -r2 was installed was an update to mesa-20.2.6

EDIT: reinstall r2, it compiles cleanly even with mesa-20.2.6

(well, cleanly - it produces all that long list of warnings about ./include/linux/cpumask.h:215:9: warning: comparison of integer expressions of different signedness: ‘int’ and ‘unsigned int’ [-Wsign-compare]
that fills my failed pastebin, now I see this is irrelevant, but then finishes all right)

I wonder if there is something in buildfix_kernel_4.11.patch, which would apply to my 4.14 kernel
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Sat Jan 30, 2021 1:20 pm    Post subject: Reply with quote

Ionen wrote:
Yeah that makes sense, probably safer to also keep that ZONE_DMA I see (was dropped near those updates).

It's "optional" (I have it disabled with 360.39) but only if you use IOMMU instead (does that even work with 340? odds are old hardware wouldn't have good support either way).

I guess I'm keeping it then, better safe than sorry. :P

dmpogo wrote:
Sadly, actually an update r2 to r3 did not compile at the end

It turns out that the enormous patch for 5.7 broke a couple of things for older kernels, which I didn't notice because I only tested 5.4 and 5.10. It should now build successfully (with and without uvm) from 4.4 to 5.10, although at runtime I only tested 5.4 and 5.10 (since I don't want to spend half a day building kernels :lol: ).
Back to top
View user's profile Send private message
dmpogo
Advocate
Advocate


Joined: 02 Sep 2004
Posts: 3446
Location: Canada

PostPosted: Sat Jan 30, 2021 10:29 pm    Post subject: Reply with quote

Shibotto wrote:
Ionen wrote:
Yeah that makes sense, probably safer to also keep that ZONE_DMA I see (was dropped near those updates).

It's "optional" (I have it disabled with 360.39) but only if you use IOMMU instead (does that even work with 340? odds are old hardware wouldn't have good support either way).

I guess I'm keeping it then, better safe than sorry. :P

dmpogo wrote:
Sadly, actually an update r2 to r3 did not compile at the end

It turns out that the enormous patch for 5.7 broke a couple of things for older kernels, which I didn't notice because I only tested 5.4 and 5.10. It should now build successfully (with and without uvm) from 4.4 to 5.10, although at runtime I only tested 5.4 and 5.10 (since I don't want to spend half a day building kernels :lol: ).


Thanks, it compiles just fine now for 4.14. I'll be watching the runtime
Back to top
View user's profile Send private message
Poopman
n00b
n00b


Joined: 13 Jan 2021
Posts: 2

PostPosted: Mon Feb 01, 2021 5:38 pm    Post subject: Reply with quote

Shibotto wrote:
I'm sorry I haven't been around, I totally forgot about this thread, my bad :roll:

I updated a few things:
  • Patches up to linux 5.10
  • Restored nvidia-uvm module: it compiles, I don't know if it works though cause I don't use CUDA
  • Script to correctly VT switch after resuming from suspend when using elogind
  • Removed the udev rule to call nvidia-smi on boot because it kills TTYs on my laptop; if you need /dev/nvidia* to be initialized before Xorg starts, enable the nvidia-smi service


Poopman wrote:
lspci -k

https://pastebin.com/Hm5rATd1

Seeing nvidia_drm as an available module for your card is really suspicious. Did you mask nvidia-drivers versions other than 340?


Yep, I masked all of them except 340, but well, i don't have the laptop anymore :cry: (died suddenly). I'll try that and more if it happens in another time.

Thanks for your help.
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Sun Feb 14, 2021 7:46 pm    Post subject: Reply with quote

A couple of updates:
  • Gentoo fixed their libglvnd USE flag situation for x11-base/xorg-server, so I can finally drop it myself. FINALLY.
  • As you may have heard, GTK+2 went EOL with GTK4 release, so Gentoo is proceeding to phase it out. It probably won't be a problem for quite a bit, but if you are eager to jump into the inevitable "future" (LOL) I added nvidia-settings-343.36 (currently masked for testing) which more or less only added initial GTK+3 support compared to 340. Unless someone reports regressions, I'll make it stable in a couple of weeks. nvidia-settings-340.108 won't go away until the removal of GTK+2, so rest assured.

As usual if I managed to break you installation let me know here or on GitLab. :lol:
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Sun Feb 14, 2021 9:29 pm    Post subject: Reply with quote

Yeah, gentoo started cleaning up the situation thanks to new maintainer. So nvidia-drivers lost libglvnd flag plus the FreeBSD ancestral stuff (like your ebuild already did). There's plans to split off nvidia-settings in gentoo as well, but that'll wait for later given versions bumps were getting urgent (security issues too). Those new bumps also all work with kernel 5.10.x without patches (390.xx included), but 340.xx will remain out of the loop :(
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Mon Feb 15, 2021 1:36 am    Post subject: Reply with quote

Ionen wrote:
There's plans to split off nvidia-settings in gentoo as well, but that'll wait for later given versions bumps were getting urgent (security issues too).

This is wonderful news! The second compromise I hate the most is having those "tools" and "static-libs" USE flag in x11-drivers/nvidia-drivers in order not to break ebuilds depending on libXNVCtrl, so I can't wait to be able to get rid of them too. It would also be very welcome if they end up splitting libXNVCtrl from nvidia-settings for obvious reasons, but now I feel like I'm being too greedy :oops:

Ionen wrote:
Those new bumps also all work with kernel 5.10.x without patches (390.xx included), but 340.xx will remain out of the loop :(

That's what we get for having chosen Nvidia :twisted:
Not that I actually regret it in this particular scenario, since back then the alternatives were ATI, also known as more broken driver than Nvidia, and Intel which was barely able to play a 1080p movie. I'm really glad today the situation is incredibly better if you plan to use a GPU long term, and since I don't need CUDA Nvidia is not seeing a penny from me ever again :x
Back to top
View user's profile Send private message
psycho
Guru
Guru


Joined: 22 Jun 2007
Posts: 544
Location: New Zealand

PostPosted: Mon Feb 15, 2021 1:55 am    Post subject: Reply with quote

Shibotto wrote:
since I don't need CUDA Nvidia is not seeing a penny from me ever again :x

If I hadn't needed CUDA they wouldn't have got a penny out of me in the first place, but sadly I invested in Nvidia cards for that reason, and so have stuck with them despite the fact that I could easily live without CUDA now, and that...well, Linus put it best. I continue to hope that they'll open up their drivers and these cards will suddenly sparkle with a whole new aura of reasonableness...but it's not a rational hope.
Back to top
View user's profile Send private message
dmpogo
Advocate
Advocate


Joined: 02 Sep 2004
Posts: 3446
Location: Canada

PostPosted: Mon Feb 15, 2021 8:40 am    Post subject: Reply with quote

Shibotto wrote:
A couple of updates:
  • Gentoo fixed their libglvnd USE flag situation for x11-base/xorg-server, so I can finally drop it myself. FINALLY.
  • As you may have heard, GTK+2 went EOL with GTK4 release, so Gentoo is proceeding to phase it out. It probably won't be a problem for quite a bit, but if you are eager to jump into the inevitable "future" (LOL) I added nvidia-settings-343.36 (currently masked for testing) which more or less only added initial GTK+3 support compared to 340. Unless someone reports regressions, I'll make it stable in a couple of weeks. nvidia-settings-340.108 won't go away until the removal of GTK+2, so rest assured.

As usual if I managed to break you installation let me know here or on GitLab. :lol:



Well, they need to get gimp off gtk2 first for sure :)
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Tue Mar 02, 2021 9:18 am    Post subject: Reply with quote

Shibotto wrote:
This is wonderful news! The second compromise I hate the most is having those "tools" and "static-libs" USE flag in x11-drivers/nvidia-drivers in order not to break ebuilds depending on libXNVCtrl, so I can't wait to be able to get rid of them too. It would also be very welcome if they end up splitting libXNVCtrl from nvidia-settings for obvious reasons, but now I feel like I'm being too greedy :oops:
There's chances I may end up maintaining nvidia-drivers myself eventually (I authored the previous 460.56 bump, trivial as it may be), so starting to debate my options now.

Not entirely sure I want to split off nvidia-settings after all (or at least not for now, maybe later), but I do want at least want to make USE=tools optional to get libXNVCtrl.a (haven't check yet but likely easy to build individually). I actually don't use nvidia-settings myself but I have a fan control daemon I wrote that needs the library, and I'd be happy to -tools

Planning to remove virtual/opencl dep as well, as far as I can tell it doesn't make much sense on the drivers until something needs it and they'll depend on opencl themselves. Also debating removing USE=uvm entirely and just installing the module unconditionally (it's always built either way, and would spare things from having to depend on nvidia-drivers[uvm]). USE=kms feels similarly silly (does nothing beside prevent 2 modules install), at most I guess it's convenient not to have the modules get auto-loaded, mostly makes sense on cuda servers.

Will try to make it more viable to install the drivers with -X too (by separating libraries that don't need it, but some hard need GLX), will have to test how this goes on wayland and servers.

Suggestions/oppositions welcome anyway :) Nothing set in stone.

Edit: unfortunately re-adding 340.xx is not the table, I'm not "entirely" against it but given old nvidia-drivers have known vulnerabilities it makes it hard to re-add to official ::gentoo
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Thu Sep 02, 2021 11:19 pm    Post subject: Reply with quote

Long time no see!
I finally took some time to tidy up this overlay a bit, which is now ~99% based on the very neat work Ionen did in the main tree (thank you, it was MUCH needed). It's still in its dedicated branch as a safety measure, but unless some major regression shows up, I plan to go stable and merge it to master by next week. So I won't break you systems for a few days at least :lol:

As I previously said, I really wanted to split nvidia-drivers further, so it now comes in 3 ebuilds: nvidia-kmod, nvidia-drivers and nvidia-settings. If I put the blockers correctly, when the time comes upgrading should be as smooth as adding "x11-drivers/nvidia-kmod NVIDIA-r2" to your package.license .

↓ The following is an explanatory changelog if you want to know why I did things the way I did them ↓

x11-drivers/nvidia-kmod: the Nvidia kernel module.
  • New package
    Since this will get rebuilt very often, just do the bare minimum required. This is what I was aiming for when I split nvidia-drivers and nvidia-settings in the beginning, but I did things in a rush and focused on the wrong aspect of the problem.
  • Drop nvidia-uvm and related patches
    After a night stroll on Wikipedia it became clear that none of the cards supported only by 340.xx is compatible with Unified Memory anyway.
  • Still only support kernels up to 5.10.x
    I only use LTS kernels, so even if I add patches for interim versions, I won't be testing them. As usual, user patches can be added to /etc/portage/patches/ .

x11-drivers/nvidia-drivers: Nvidia libs and tools.
  • Rebased on Gentoo's ebuild
    Some of the changes that happened there also apply here, for example: utilities like nvidia-modprobe and nvidia-persistenced being built from source, obsolete USE flags being dropped, ecc.
  • Drop IUSE=driver
    Unconditionally depend on x11-drivers/nvidia-kmod. If for some reason you need the USE=-driver behavior, add nvidia-kmod to your package.provided.
  • Build libXNVCtrl.a
    Moved here from nvidia-settings-340.108 .

media-video/nvidia-settings: ...well, nvidia-settings :P.
  • Default to 343.36 (using GTK+ 3)
    340.108 (GTK+ 2) is still in the overlay and will stay there as long as x11-libs/gtk+:2 is in Gentoo. This situation is the main reason why I'm keeping nvidia-settings split.
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Fri Sep 03, 2021 4:12 am    Post subject: Reply with quote

Shibotto wrote:
  • Drop nvidia-uvm and related patches
    After a night stroll on Wikipedia it became clear that none of the cards supported only by 340.xx is compatible with Unified Memory anyway.
It's also barely usable even with 390.xx, a lot of the libraries/tools don't easily support old cuda (need gcc8 for that too), ffmpeg needs newer nv-codec-headers, despite 390.xx has x86 support uvm doesn't anymore, etc... OpenCL should still work okay with 390.xx though.

Nice to see some updates anyhow :) I do suggest the overlay when I see people want to use 340 still (happens now and then on #gentoo IRC).

I did originally think to reduce amount of things rebuilt as well, but concluded nvidia-settings didn't really take that long.. or at least on non-340-using hardware (and combined with versions typically needing to match, and how the .run normally gives this together, felt it was fine that way). Not that I think splitting it is wrong either, it's just another way to go about it -- plus 340 has its own issues where may need to further drop support for things.
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Wed Sep 08, 2021 1:44 pm    Post subject: Reply with quote

Ionen wrote:
Nice to see some updates anyhow :) I do suggest the overlay when I see people want to use 340 still (happens now and then on #gentoo IRC).

That's good to hear. Knowing someone other than me could be relying on this is what keeps me motivated to not just scatter the files around in my installation and call it a day :lol:

Ionen wrote:
I did originally think to reduce amount of things rebuilt as well, but concluded nvidia-settings didn't really take that long.. or at least on non-340-using hardware (and combined with versions typically needing to match, and how the .run normally gives this together, felt it was fine that way). Not that I think splitting it is wrong either, it's just another way to go about it -- plus 340 has its own issues where may need to further drop support for things.

Yes, even on my 11 years old i3 it takes less than 5 minutes, not exactly an intensive task, especially when things like a webkit implementation join the party... I can't imagine it would make a noticeable difference on a modern system :)

Speaking of this, I was wondering about the packages depending on x11-drivers/nvidia-drivers[static-libs,tools] : do they really need the nvidia-settings executable or is it just a remnant of when "static-libs" needed "tools"? These are the ones I found:
Code:
shiba ~ > find /var/db/repos/gentoo/ -type f -name "*.ebuild" -exec grep "x11-drivers/nvidia-drivers.*\[.*tools" {} +

/var/db/repos/gentoo/xfce-extra/xfce4-sensors-plugin/xfce4-sensors-plugin-1.4.1.ebuild:   video_cards_nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )"
/var/db/repos/gentoo/sys-apps/hwloc/hwloc-2.3.0.ebuild:   gl?      ( x11-drivers/nvidia-drivers[static-libs,tools] )
/var/db/repos/gentoo/sys-apps/hwloc/hwloc-1.11.13.ebuild:   gl? ( x11-drivers/nvidia-drivers[static-libs,tools] )
/var/db/repos/gentoo/sys-apps/hwloc/hwloc-2.5.0.ebuild:   gl?      ( x11-drivers/nvidia-drivers[static-libs,tools] )
/var/db/repos/gentoo/app-admin/conky/conky-1.12.2.ebuild:   nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )
/var/db/repos/gentoo/app-admin/conky/conky-1.12.1-r1.ebuild:   nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )
/var/db/repos/gentoo/media-video/bino/bino-1.6.7.ebuild:   video_cards_nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )"
/var/db/repos/gentoo/sci-libs/vtk/vtk-8.2.0-r4.ebuild:   video_cards_nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )
/var/db/repos/gentoo/sci-libs/vtk/vtk-9.0.3.ebuild:   video_cards_nvidia? ( x11-drivers/nvidia-drivers[tools,static-libs] )
/var/db/repos/gentoo/mate-extra/mate-sensors-applet/mate-sensors-applet-1.24.1.ebuild:   video_cards_nvidia? ( >=x11-drivers/nvidia-drivers-100.14.09:0[static-libs,tools] )

I'm quite sure the MATE/Xfce plugin don't need it, but I don't know about the others.
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Wed Sep 08, 2021 11:25 pm    Post subject: Reply with quote

Shibotto wrote:
Speaking of this, I was wondering about the packages depending on x11-drivers/nvidia-drivers[static-libs,tools] : do they really need the nvidia-settings executable or is it just a remnant of when "static-libs" needed "tools"?
Technically they didn't need to depend tools back then either given the REQUIRED_USE="static-libs? ( tools )" enforcing it. Didn't look but further back in history I imagine this used to not have IUSE=static-libs and tools /was/ the flag to use, then they added static-libs on top.

Since I don't really use nvidia-settings myself but do use the library a bit with my own C code, felt wouldn't hurt to split off the requirement due to less deps.

I doubt anything is using the nvidia-settings binary anyhow, not that I investigated.

Edit: Still waiting to see what nvidia will do with these so it's usable on wayland, may need to refactor a bit eventually (hopefully they give a shared library without having to consider hacking one in like debian did).
Back to top
View user's profile Send private message
Tom_
Guru
Guru


Joined: 20 May 2004
Posts: 448
Location: France

PostPosted: Sat Oct 16, 2021 12:07 pm    Post subject: Reply with quote

I've been using Andrew Church's overlay for quite a while. Even if it works, I'm getting more and more blockers and I have to maintain more and more ebuilds manually in my own overlay. :(

I would like to use a simpler solution. Shibotto's overlay looks promising. I'll give it a try! Thank you for your work :)
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Sat Oct 16, 2021 8:39 pm    Post subject: Reply with quote

Yeah, in my opinion trying to keep old eselect-opengl is a lost cause that deviates far too much from where gentoo is going.

::nvidia-legacy keeps this only specific to nvidia without touching the rest and can live with modern Gentoo just fine (or at least for as long as a new Xorg doesn't a break it).
Back to top
View user's profile Send private message
Shibotto
Apprentice
Apprentice


Joined: 19 Jun 2015
Posts: 157
Location: CET/CEST

PostPosted: Sat Dec 18, 2021 2:12 pm    Post subject: Reply with quote

Some updates:

nvidia-kmod can now be built against anything from 4.4 to 5.15. At runtime though I only tested 4.4.290 (as a safety measure), 5.4.156, 5.10.76 and 5.15.7 but I think there should be no problem with the releases in between. If you decide to upgrade to 5.15 keep in mind that as per the recent trend new LTS kernels are supported for 2 years only (until further notice).

xorg-server 21.1, which was my biggest concern long term, luckily seems to still play nice at least with my hardware (OpenGL, HDMI/VGA output, resume from suspension, vdpau, ecc). I'm now relying on Xorg's "IgnoreABI" though, which essentially means that everything is fine... until it isn't... someday, usually when you are swamped with work :D

I've been running 5.15 + 21.1 for a week and I didn't notice any regression, so this time I'm confident enough to push this right away.
I hope
Back to top
View user's profile Send private message
Ionen
Developer
Developer


Joined: 06 Dec 2018
Posts: 2874

PostPosted: Sat Dec 18, 2021 8:32 pm    Post subject: Reply with quote

Good to know that IgnoreABI works, still hoping for nvidia to update 390.xx for xorg-21 at least (still hope unlike 340) but if it comes down to it I'll have it pass IgnoreABI too :?

xorg-1.20 is still supported for now fwiw (1.20.14 was added not long ago with security fixes).

Update: nvidia updated 390 to xorg-21 and kernel-5.15 with >=390.147.. so no worries for that one for now
Back to top
View user's profile Send private message
Randy Andy
Veteran
Veteran


Joined: 19 Jun 2007
Posts: 1152
Location: /dev/koelsch

PostPosted: Fri Jul 08, 2022 9:14 pm    Post subject: Short story long ;-) Reply with quote

Hi Shibotto and Ionen.

first of all I would like to thank Shibotto for his excellent work in the form of his manual and his repository!
I've only been using it since today, but it looks like it has now saved me and finally brought me back to calmer waters, after a far too long ordeal.

I had already switched to the Nouveau drivers many years ago, when I was still using another graphics card and another PC, and had actually had quite good experiences with them.
Especially because at the time I was still on the ~amd64 arch and at the time the nvidia drivers could not yet be built against the latest kernels without an IgnoreABI patch.

In 2014 I bought a used Dell Precision T3500 with a Nvidia Quadro FX4800 graphics card.
I would still consider it up to date, thanks to 6 physical cores or 12 with hyperthreading and a still quite powerful graphics card.
https://www.techpowerup.com/gpu-specs/quadro-fx-4800.c1320
https://en.wikipedia.org/wiki/Tesla_(microarchitecture)

Finally, it manages a resolution of up to 2880x1620 px at 60 Hz and has one DVI and two display port outputs.
Since I'm not interested in video games, but rather in more constructive 3D and office applications, I absolutely don't see the point in discarding this graphics card. Even if Nvidia tries to discard it as outdated with constantly decreasing support.

Even with the Nouveau drivers I got along for a long time, at least on the DVI output. But after I got myself a 32 inch monitor, I noticed that the picture quality on the DVI output was significantly worse than on the DP output.
From that point on, December 2019, my problems began and they have continued ever since, because the Power Management of the Nouveau driver for this card have been so bad for years that the monitor usually cannot be woken up again after a shutdown.
With kernels up to version 5.4.143 this still worked in about 80% of the cases, but with newer kernels this reversed in about proportion. So here it was necessary to reboot the computer all the time, because all nouveau.<parameters> could not help.
Turning off the screen saver or power management did not help to prevent ghosting. Turning off the monitor didn't help either, because after that the video card didn't wake up either, with errors along the lines of:
nouveau 0000:02:00.0: DRM: core notifier timeout
kernel: nouveau 0000:02:00.0: DRM: base-0: timeout

So at some point within the last two years I tried to switch back to the proprietary nvidia-drivers, which failed thanks to the lack of support from the Gentoo maintainers and became hell for me as a result.
Ionen, it's not just you - in general I find that many packages are removed from the tree far too quickly just because Upstream or someone else discontinues support - it really fucks me off.
GRUB legacy also comes to mind, I still prefer it!

The few additional ebuilds, i.e. text files, won't take up so much space that they can't still be made available to the users. What about the much vaunted and much appreciated "it's all about choice" philosophy.
It's clear that the mantainers want to make their work easier, but from my point of view they are allowed to have a few more instances of the ebuilds. How often I had to experience that a stable version had a bug and I couldn't go back to the previous version, because it was already removed from the tree. Then the huddle with archives, overlays or local overlays starts - not nice from the user's point of view.
The more annoyed I was, to return to the nvidia-driver, when I found out that the version nvidia-drivers-340.x was not even in the Layman overlays or the ones from Zugaina.

Fortunately, thanks to Shibotto's commitment, I found the solution I had hoped for and, thanks to an excellent and simpler description, the solution I had longed for. It's nice to see how all the old dependency problems could be solved in the meantime and how easy the installation can be now.
Finally I can send my PC into standby mode to wake it up again quickly and finally have a picture afterwards.

It's good that the topic of sustainability and planned obsolescence has been brought to people's attention a bit more. Years ago I had to listen to myself all the time - why don't you just buy a new graphics card that is better supported by Linux, but I think that would not have been the right way here in particular.

It's good that the topic of sustainability and planned obsolescence has now been brought to people's attention a bit more. Years ago I had to listen to myself all the time - why don't you just buy a new graphics card that is better supported by Linux, but I think that would not have been the right way here in particular.

Hopefully the situation remains so functional, because as you already mentioned - you never know which of the next changes will break the whole thing again.

Shibotto, in any case, now you see that another user also benefits from your work and appreciates it and is very grateful to you. And so I hope that this will give you a little motivation to continue.

Best regards,
Andy.
_________________
If you want to see a Distro done right, compile it yourself!
Back to top
View user's profile Send private message
Display posts from previous:   
Reply to topic    Gentoo Forums Forum Index Kernel & Hardware All times are GMT
Goto page Previous  1, 2, 3, 4, 5  Next
Page 3 of 5

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum