Arch Linux User Repository

Use this forum thread for discussion: https://bbs.archlinux.org/viewtopic.php?pid=1946926

Please don't flag this package out-of-date unless a new version has been released by NVIDIA.

The DKMS package guidelines are explicit that linux-headers should not be a dependency of any DKMS package.

As a concrete example of why including that as a hard dependency is a bad idea, what happens when linux is not an installed kernel?

Thanks Jonathon and collaborators, it still works (2021/08/02) !

My steps

• pacman -Syyu
• ("breaks the dkms driver", with the new kernel version)
• download latest snapshot of this AUR package. Download snapshot
• extract .tar.gz package to new directory.
• go to the new directory and execute with user account makepkg -sri
• Install de builded binary.
• reboot
• And test the new kernel version with the new driver version with primusrun glxinfo

System info

$uname -srv
Linux 5.13.7-arch1-1 #1 SMP PREEMPT Sat, 31 Jul 2021 13:18:52 +0000

$ pacman -Qee | grep -i nvidia
lib32-nvidia-390xx-utils 390.138-2
lib32-opencl-nvidia-390xx 390.138-2
nvidia-390xx-dkms 390.144-1
nvidia-390xx-utils 390.144-1
opencl-nvidia-390xx 390.144-1

(maybe i need to update my 32bit nvidia drivers haha)

Thanks all for your work and contributions!

@jonathon - Confirmed today that nvidia-390xx-dkms 390.143-4 builds correctly and installs without issue against: 5.13.1-arch1-1 as well as against my various 5.12.X kernels.

Cheers!

DKMS no longer builds on 5.13.0. Tried reinstalling DKMS and nvidia-390xx-dkms itself, no luck. Warnings and errors:

[...]

/var/lib/dkms/nvidia/390.143/build/nvidia/nv-mmap.c: In function ���nv_encode_caching���:
/var/lib/dkms/nvidia/390.143/build/nvidia/nv-mmap.c:255:16: warning: this statement may fall through [-Wimplicit-fallthrough=]
  255 |             if (NV_ALLOW_CACHING(memory_type))
      |                ^
/var/lib/dkms/nvidia/390.143/build/nvidia/nv-mmap.c:257:9: note: here
  257 |         default:
      |         ^~~~~~~

[...]

/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:188:23: error: initialization of ���int (*)(struct drm_plane *, struct drm_atomic_state *)��� from incompatible pointer type ���int (*)(struct drm_plane *, struct drm_plane_state *)��� [-Werror=incompatible-pointer-types]
 188 |     .atomic_check   = nv_drm_plane_atomic_check,
     |                       ^~~~~~~~~~~~~~~~~~~~~~~~~
/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:188:23: note: (near initialization for ���nv_plane_helper_funcs.atomic_check���)
/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:189:23: error: initialization of ���void (*)(struct drm_plane *, struct drm_atomic_state *)��� from incompatible pointer type ���void (*)(struct drm_plane *, struct drm_plane_state *)��� [-Werror=incompatible-pointer-types]
  189 |     .atomic_update  = nv_drm_plane_atomic_update,
     |                       ^~~~~~~~~~~~~~~~~~~~~~~~~~
/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:189:23: note: (near initialization for ���nv_plane_helper_funcs.atomic_update���)
/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:190:23: error: initialization of ���void (*)(struct drm_plane *, struct drm_atomic_state *)���   from incompatible pointer type ���void (*)(struct drm_plane *, struct drm_plane_state *)��� [-Werror=incompatible-pointer-types]
  190 |     .atomic_disable = nv_drm_plane_atomic_disable,
      |                       ^~~~~~~~~~~~~~~~~~~~~~~~~~~
/var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.c:190:23: note: (near initialization for ���nv_plane_helper_funcs.atomic_disable���)
  CC [M]  /var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-gem.o
cc1: some warnings being treated as errors
make[2]: *** [scripts/Makefile.build:272: /var/lib/dkms/nvidia/390.143/build/nvidia-drm/nvidia-drm-crtc.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [Makefile:1847: /var/lib/dkms/nvidia/390.143/build] Error 2
make[1]: Leaving directory '/usr/lib/modules/5.13.0-174-tkg-cacule/build'
make: *** [Makefile:81: modules] Error 2

I've been crossing XID errors with screen crashes.

Rolling back to 390.143-1 (+ Linux 5.12.13) is so far stable

Confirmed working on Jun. 9th, 2021

• Linux 5.12.9
• Nvidia 390.143

Linux Kernel 5.12.9 installed packages:

• linux-5.12.9.arch1-1-x86_64
• linux-api-headers-5.12.3-1-any
• linux-firmware-20210511.7685cf4-1-any
• linux-headers-5.12.9.arch1-1-x86_64

Nvidia Driver 390.143 installed packages:

• nvidia-390xx-utils-390.143-3-x86_64
• nvidia-390xx-dkms-390.143-3-x86_64
• nvidia-390xx-settings-390.143-2-x86_64
• opencl-nvidia-390xx-390.143-3-x86_64
• libxnvctrl-390xx-390.143-2-x86_64

Nvidia Driver 390.143 packages not tried:

• nvidia-390xx-390.143-1-x86_64 (It's recommended to always use dkms package!)

Installation Notes:

• 1st. Backup all your old packages from /var/cache/pacman/pkg/
• 2nd. Go into TTY mode and remove all old nvidia packages.
• 3rd. Make sure your linux kernel is not ignored in /etc/pacman.conf
• 4th. Upgrade your linux kernel to latest 5.12 version.
• 5th. Install your new aur compiled nvidia driver packages.

Reboot & Enjoy! Please make sure you also report any bugs!

Thanks everyone, especially Nvidia team, package maintainer jonathon and anyone providing patches!

@stephenboston: That's because the three packages: nvidia-390xx-utils, opencl-nvidia-390xx, nvidia-390xx-dkms are all built from the same source and PKGBUILD file.

This page for nvidia 390 dkms is pointing to the repo for nvida 390 utils.

I don't think that's a packaging issue (e.g. https://github.com/OpenPIV/openpiv-python-gpu/issues/3). Maybe open a thread on the forum?

I have a problem with running an an MPI application (compiled on my computer)

--------------------------------------------------------------------------
An error occurred while trying to map in the address of a function.
  Function Name: cuIpcOpenMemHandle_v2
  Error string:  /usr/lib/libcuda.so.1: undefined symbol: cuIpcOpenMemHandle_v2
CUDA-aware support is disabled.
--------------------------------------------------------------------------

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0  0x146a21675f7f in ???
#1  0x0 in ???

I assume it is related to the recent inclusion of (optional) CUDA support into openMPI (https://bugs.archlinux.org/task/68031).

On other computers, I simply get the message that CUDA support is disabled because libcuda.so is not found.

@hute37: Not a package issue. Report to the paru maintainer.