[Solved] 5.9.* and nvidia-drivers.. again..

Message

389292 · Post by **389292** » Mon Oct 19, 2020 12:13 pm

After upgrading to 5.9 I no longer have vsync with the nvidia-drivers-455.28.
Force Full Composition Pipeline is active in the nvidia-settings, but it doesn't seem to work.
I also have this in dmesg:

Code: Select all

[   22.361079] r8169 0000:05:00.0 enp5s0: renamed from eth0
[   22.377006] r8169 0000:06:00.0 enp6s0: renamed from eth1
[   22.400862] nvidia: loading out-of-tree module taints kernel.
[   22.400870] nvidia: module license 'NVIDIA' taints kernel.
[   22.400870] Disabling lock debugging due to kernel taint
[   22.431806] nvidia-nvlink: Nvlink Core is being initialized, major device number 246
[   22.432186] nvidia 0000:03:00.0: vgaarb: changed VGA decodes: olddecodes=io+mem,decodes=none:owns=io+mem
[   22.601019] BTRFS info (device dm-0): disk space caching is enabled
[   22.603074] BTRFS info (device dm-0): setting nodatacow, compression disabled
[   22.603076] BTRFS info (device dm-0): disk space caching is enabled
[   22.604946] EXT4-fs (dm-1): re-mounted. Opts: (null)
[   22.606694] EXT4-fs (dm-2): re-mounted. Opts: (null)
[   22.671926] Adding 6291452k swap on /swap/swapfile.  Priority:-2 extents:14 across:8153524k 
[   22.921395] EXT4-fs (sdd1): mounted filesystem with ordered data mode. Opts: (null)
[   23.202186] elogind-daemon[2087]: Watching system buttons on /dev/input/event1 (Power Button)
[   23.202226] elogind-daemon[2087]: Watching system buttons on /dev/input/event0 (Power Button)
[   23.394033] elogind-daemon[2087]: Watching system buttons on /dev/input/event18 (USB GAMING MOUSE  Keyboard)
[   23.394098] elogind-daemon[2087]: Watching system buttons on /dev/input/event19 (USB GAMING MOUSE  Consumer Control)
[   23.566010] elogind-daemon[2087]: New seat seat0.
[   23.580863] r8169 0000:06:00.0: Direct firmware load for rtl_nic/rtl8168d-2.fw failed with error -2
[   23.580865] r8169 0000:06:00.0: Unable to load firmware rtl_nic/rtl8168d-2.fw (-2)
[   23.581515] RTL8211B Gigabit Ethernet r8169-600:00: attached PHY driver [RTL8211B Gigabit Ethernet] (mii_bus:phy_addr=r8169-600:00, irq=IGNORE)
[   23.641563] ip (2509) used greatest stack depth: 10656 bytes left
[   23.642994] r8169 0000:06:00.0 enp6s0: Link is Down
[   24.092555] elogind-daemon[2087]: Watching system buttons on /dev/input/event21 (Hexwire Ortho 5x13)
[   25.152974] NVRM: loading NVIDIA UNIX x86_64 Kernel Module  455.28  Wed Sep 30 01:05:16 UTC 2020
[   25.198473] resource sanity check: requesting [mem 0x000e0000-0x000fffff], which spans more than pnp 00:04 [mem 0x000e0000-0x000effff]
[   25.198605] caller _nv032037rm+0x2d/0x60 [nvidia] mapping multiple BARs
[   25.376053] resource sanity check: requesting [mem 0x000c0000-0x000fffff], which spans more than PCI Bus 0000:00 [mem 0x000c0000-0x000dffff window]
[   25.376136] caller _nv000708rm+0x1af/0x200 [nvidia] mapping multiple BARs
[   28.343099] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms  455.28  Wed Sep 30 00:58:12 UTC 2020
[   28.344756] [drm] [nvidia-drm] [GPU ID 0x00000300] Loading driver
[   28.344758] [drm] Initialized nvidia-drm 0.0.0 20160202 for 0000:03:00.0 on minor 0
[   35.070930] r8169 0000:06:00.0 enp6s0: Link is Up - 100Mbps/Full - flow control rx/tx
[   35.070936] IPv6: ADDRCONF(NETDEV_CHANGE): enp6s0: link becomes ready
[   50.914178] elogind-daemon[2087]: New session 1 of user stratex.
[   59.656623] resource sanity check: requesting [mem 0x000c0000-0x000fffff], which spans more than PCI Bus 0000:00 [mem 0x000c0000-0x000dffff window]
[   59.656757] caller _nv000708rm+0x1af/0x200 [nvidia] mapping multiple BARs

Despite enabling this patch for the kernel
https://747028.bugs.gentoo.org/attachment.cgi?id=665762

Code: Select all

lsmod
Module                  Size  Used by
nvidia_drm             49152  2
nvidia_modeset       1167360  4 nvidia_drm
nvidia              27217920  159 nvidia_modeset

Does anyone have it working or the only solution is to downgrade to 5.8 for now?

Post by **Ionen** » Mon Oct 19, 2020 12:29 pm

Can't help with the issue (haven't experimented with 5.9 much yet given I have non-nvidia issues with it), but you can get rid of the patch, you shouldn't need it if you're not loading the uvm module (which has nothing to do with vsync, it's for cuda/opencl).

Edit: actually I'll give it a try by omitting the things I have problems with, I use pipeline compositing as well (set through xorg.conf rather than nvidia-settings but shouldn't matter), might as well check if it's fine. Gimme a bit of time to setup.

jpsollie · Post by **jpsollie** » Mon Oct 19, 2020 12:46 pm

looks like you'll need to wait some time (1 month) before you can get a working nvidia driver on 5.9:
https://www.phoronix.com/scan.php?page= ... .9-Delayed

Post by **Ionen** » Mon Oct 19, 2020 1:11 pm

Ionen wrote:Edit: actually I'll give it a try by omitting the things I have problems with, I use pipeline compositing as well (set through xorg.conf rather than nvidia-settings but shouldn't matter), might as well check if it's fine. Gimme a bit of time to setup.

Well, on nvidia-drivers-455.28 + xfwm4 (which handle compositing, built with -opengl -xpresent since using ForceFullCompositionPipeline to handle tearing) + gentoo-kernel-5.9.1 right now and everything seems fine, don't see any tearing while moving windows and similar.

Tried vkcube+glxgears and they both can use vsync. I still can't help but "looks" fine for me, using a 1070 card. I don't get those dmesg sanity check / BARs messages either.

Code: Select all

$ uname -a; nvidia-smi | grep Version
Linux eos 5.9.1 #1 SMP PREEMPT Mon Oct 19 08:59:48 EDT 2020 x86_64 Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz GenuineIntel GNU/Linux
| NVIDIA-SMI 455.28       Driver Version: 455.28       CUDA Version: 11.1     |

Edit: just tried to boot some game with wine+dxvk and looks normal too. Can't talk for everyone but kernel 5.9 is fine with nvidia for me (I don't use cuda/opencl).

Edit2: might be worth noting that I don't use nvidia-drm.modeset=1 (maybe related if you're using it), I don't load nvidia_drm at all actually... (I have it in INSTALL_MASK, mimic at your own risks). I only have 2 modules loaded:

Code: Select all

$ lsmod
Module                  Size  Used by
nvidia_modeset       1089536  2
nvidia              26042368  84 nvidia_modeset

Hopefully others have further feedback that may give some hint.

389292 · Post by **389292** » Mon Oct 19, 2020 2:28 pm

After more testing, looks like I have working vsync locally, with glxgears, or glxspheres64 (part of virtualgl),
playing this video locally with mpv: https://www.youtube.com/watch?v=Itu4aLp0e-U no tearing.
But in firefox 81.0.2 playing same video on youtube gives me tearing. I recently moved to a new install, and reconfigured firefox from zero, I could messed something up..

Post by **Ionen** » Mon Oct 19, 2020 2:37 pm

Tried the above video in firefox-81.0.2, for what it's worth no tearing on my end

389292 · Post by **389292** » Mon Oct 19, 2020 2:54 pm

Forcefully enabling
layers.acceleration.force-enabled
gfx.direct2d.force-enabled
In about:config seem to fix it. I'm pretty sure it wasn't me who broke it in the first place.
honestly, are we that much different from Windows users in regards of something breaking or being shoved into our throat?
I think the only difference is that we are agreeing to be guenypigs, and they are not, first case is BDSM, second one is RAPE.

Post by **Ionen** » Mon Oct 19, 2020 3:17 pm

Maybe you had USE=hwaccel on firefox before and no longer do? It should add this:

Code: Select all

/* Force hardware accelerated rendering due to USE=hwaccel */
pref("layers.acceleration.force-enabled",  true);
pref("webgl.force-enabled",                true);

Not that I get tearing even with this disabled, but I guess it's because of my compositing setup.

389292 · Post by **389292** » Wed Oct 21, 2020 3:57 pm

I tested my old system, Firefox 82, kernel 5.9.1, no tearing even without these settings set, also no hwaccel use flag was set.
New system, same setup or at least I can't really point out the concrete difference, of course there are fewer packages installed in the new system, it tears without hwaccel. I also don't run a compositor in both systems. I don't have time for this, hwaccel would suffice.

pd1986 · Post by **pd1986** » Thu Oct 22, 2020 11:51 am

News from Nvidia said that nvidia-drivers will not be updated at least within one month from 5.9 release. It is therefore not recommended to upgrade the kernel too aggressive.

Post by **Ionen** » Thu Oct 22, 2020 12:12 pm

pd1986 wrote:News from Nvidia said that nvidia-drivers will not be updated at least within one month from 5.9 release. It is therefore not recommended to upgrade the kernel too aggressive.

Issue from this thread ended up not really being related 5.9.

Also went over this already, but the >=450.80.02 drivers do work with 5.9.x without patches as long as you don't need USE=uvm on nvidia-drivers (it's disabled by default in gentoo, but needed for opencl/cuda). Nvidia just didn't go into details as to what happened and gave a more generalist answer, it's really just due changes to how 5.9 handle kernel taint / licensing which cause it to refuse to load the uvm module (even if it would work).

Not that sticking to 5.8.x meanwhile is a bad thing (but do continue to update your 5.8.x branch, 5.8.16 notably fixed a big bluetooth vulnerability).