nvidia-driver-390 fails to start GUI

I don't see GDM, Gnome shell, or anything else that should give you a login screen.

If you are using GDM, can you attach the output of this command, please?

systemctl status gdm3.service

LightDM is installed for the login screen. I have tried installing gdm3 but it didn't help.

sudo dpkg-reconfigure gdm3
I selected gdm3 then rebooted.
systemctl status gdm3.service
● gdm.service - GNOME Display Manager
   Loaded: loaded (/lib/systemd/system/gdm.service; static; vendor preset: enabl
  Drop-In: /lib/systemd/system/display-manager.service.d
           └─xdiagnose.conf
   Active: active (running) since Sat 2018-03-17 13:55:15 AEDT; 42min ago
  Process: 1080 ExecStartPre=/usr/share/gdm/generate-config (code=exited, status
 Main PID: 1089 (gdm3)
    Tasks: 3 (limit: 11059)
   CGroup: /system.slice/gdm.service
           └─1089 /usr/sbin/gdm3

Mar 17 14:25:59 dansgpu gdm3[1089]: Child process 28436 was already dead.
Mar 17 14:25:59 dansgpu gdm3[1089]: Unable to kill session worker process
Mar 17 14:25:59 dansgpu gdm-launch-environment][28497]: pam_unix(gdm-launch-envi
Mar 17 14:26:00 dansgpu gdm-launch-environment][28497]: pam_unix(gdm-launch-envi
Mar 17 14:26:00 dansgpu gdm3[1089]: GdmDisplay: display lasted 0.199467 seconds
Mar 17 14:26:00 dansgpu gdm3[1089]: Could not start command '/usr/lib/gdm3/gdm-s
Mar 17 14:26:00 dansgpu gdm3[1089]: GLib: g_child_watch_add_full: assertion 'pid
Mar 17 14:26:00 dansgpu gdm3[1089]: Child process -28522 was already dead.
Mar 17 14:26:00 dansgpu gdm3[1089]: Child process 28497 was already dead.
Mar 17 14:26:00 dansgpu gdm3[1089]: Unable to kill session worker process
daniel@dansgpu:~$

With the terminal wider to catch the full width of lines:
systemctl status gdm3.service
● gdm.service - GNOME Display Manager
   Loaded: loaded (/lib/systemd/system/gdm.service; static; vendor preset: enabled)
  Drop-In: /lib/systemd/system/display-manager.service.d
           └─xdiagnose.conf
   Active: active (running) since Sat 2018-03-17 13:55:15 AEDT; 6h ago
  Process: 1080 ExecStartPre=/usr/share/gdm/generate-config (code=exited, status=0/SUCCESS)
 Main PID: 1089 (gdm3)
    Tasks: 3 (limit: 11059)
   CGroup: /system.slice/gdm.service
           └─1089 /usr/sbin/gdm3

Mar 17 14:25:59 dansgpu gdm3[1089]: Child process 28436 was already dead.
Mar 17 14:25:59 dansgpu gdm3[1089]: Unable to kill session worker process
Mar 17 14:25:59 dansgpu gdm-launch-environment][28497]: pam_unix(gdm-launch-environment:session): session opened for user gdm by (uid=0)
Mar 17 14:26:00 dansgpu gdm-launch-environment][28497]: pam_unix(gdm-launch-environment:session): session closed for user gdm
Mar 17 14:26:00 dansgpu gdm3[1089]: GdmDisplay: display lasted 0.199467 seconds
Mar 17 14:26:00 dansgpu gdm3[1089]: Could not start command '/usr/lib/gdm3/gdm-session-worker': Too many open files
Mar 17 14:26:00 dansgpu gdm3[1089]: GLib: g_child_watch_add_full: assertion 'pid > 0' failed
Mar 17 14:26:00 dansgpu gdm3[1089]: Child process -28522 was already dead.
Mar 17 14:26:00 dansgpu gdm3[1089]: Child process 28497 was already dead.
Mar 17 14:26:00 dansgpu gdm3[1089]: Unable to kill session worker process
daniel@dansgpu:~$

Status changed to 'Confirmed' because the bug affects multiple users.

I have the same issue on Ubuntu MATE 18.04 (daily) which uses lightdm, via slick-greeter, as the display manager on my XPS 15 which has Intel IGP and nvidia GTX 960M.

Ever since the upgrade to nvidia-driver-390 vt7 is a blank screen on boot when the nvidia drivers are installed. The only way to get the display manager to present is to entirely remove the nvidia drivers.

@Daniel: can you attach the output of the following command, please?

sudo cat /sys/module/nvidia_drm/parameters/modeset

Please find attached a dump from `journalctl -k` and logs from Xorg and LightDM.?field.comment=Please find attached a dump from `journalctl -k` and logs from Xorg and LightDM.

@Alberto:
sudo cat /sys/module/nvidia_drm/parameters/modeset
N
The only output is "N".

Exactly the same setup as @dbonner and have exactly the same results. TTY cursor only.

xrandr outputs "Can't open display".

@albertomilone I also get "N."

Update: I removed one GPU and the problem was resolved.

Therefore, the problem appears to be specific to multi-GPU systems.

Maybe X autoconfiguration is not helping. If you have multiple nvidia GPUs, maybe try the following:

sudo nvidia-xconfig --multigpu=on

There are other options that you can try, for example, if you use SLI:
http://us.download.nvidia.com/XFree86/Linux-x86_64/390.42/README/sli.html

Interestingly, I didn't even have an xorg.conf before, apparently.

Running standard nvidia-xconfig autogenerates an xorg.conf with both of the following:

Option "MultiGPU" "on"
Option "SLI" "on".

I tried removing the SLI bridge from my computer and having just "MultiGPU" on but that didn't work either.

So far nothing has resulted in video output. Will keep trying.

Interestingly, my Motherboard's GPU0 is detected as GPU1 in nvidia-xconfig, and vice versa.

Relevant lines from Xorg log:

[ 22.370] (EE) NVIDIA(GPU-0): The NVIDIA graphics device PCI:8:0:0 bound to this SLI X
[ 22.370] (EE) NVIDIA(GPU-0): screen is not the SLI parent device. This configuration
[ 22.370] (EE) NVIDIA(GPU-0): is not currently supported. Please add 'BusID
[ 22.370] (EE) NVIDIA(GPU-0): "PCI:67:0:0"' to the SLI "Device" section in the X
[ 22.370] (EE) NVIDIA(GPU-0): configuration file.
[ 23.555] (EE) NVIDIA(GPU-0): Only one GPU will be used for this X screen.

So I added that line to Xorg.conf, and now my screen keeps turning on and off.

After doing that I can no longer switch to tty1 and get video, so I'll probably have to start over before trying some more.

Continuing to grasp at straws:

sudo nvidia-xconfig
WARNING: Unable to locate/open X configuration file.

Package xorg-server was not found in the pkg-config search path.
Perhaps you should add the directory containing `xorg-server.pc' to the PKG_CONFIG_PATH environment variable
No package 'xorg-server' found'

Update:

I am able to get X video if I boot with the monitor plugged into GPU1 and then switch the cable to GPU0 (according to nvidia settings)

TTY/UEFI video only works with the cable plugged into GPU1 (since this is my motherboard's first PCI slot)
X video only works with the cable plugged into GPU0.

@Brendan: does moving away /usr/share/X11/xorg.conf.d/nvidia-drm-outputclass-ubuntu.conf (e.g. to your home directory) and rebooting help? (also, remove your xorg.conf, just in case)

actually, simply comment out the line that says Option "PrimaryGPU" "yes" in /usr/share/X11/xorg.conf.d/nvidia-drm-outputclass-ubuntu.conf, and leave the file where it is.

@Alberto
Commenting out the PrimaryGPU line worked for me in a 4 GPU system I have.

Commenting out PrimaryGPU also worked on my XPS 15 with GTX 960M.

Commenting out PrimaryGPU also worked on my 2 GPU system. It might be time to patch the driver so we don't have to do this manually. I'm so happy. It took 5 weeks before we found a fix.

The main problem here is that the PrimaryGPU option, that we use for system Intel+NVIDIA, causes problems on systems with multiple NVIDIA GPUs.

I am working on a solution.

Just to clarify my XPS 15 is an Intel+nvidia computer and requires PrimaryGPU be commented out when using the 390 drivers.

@Martin: can you attach your /var/log/Xorg.0.log with PrimaryGPU commented out, please?

Here is my Xorg.0.log:

https://paste.ubuntu.com/p/mfkBK8hQ3F/

Here is my Xorg.0.log with nvidia enabled by PrimaryGPU commented out:

https://paste.ubuntu.com/p/XptX6tTdRr/

This bug was fixed in the package nvidia-graphics-drivers-390 - 390.48-0ubuntu2

---------------
nvidia-graphics-drivers-390 (390.48-0ubuntu2) bionic; urgency=medium

  [ Alberto Milone ]
  * debian/templates/10-nvidia.conf.in,
    debian/templates/xserver-xorg-video-nvidia-flavour.install.in:
    - Rename xorg.conf.d file.
    - Drop the PrimaryGpu option from the xorg.conf.d file, as it
      has a bad interaction with systems with multiple NVIDIA GPUs
      (LP: #1756226).

  [ Graham Inggs ]
  * debian/templates/control.in:
    - Make sure that libnvidia-compute does not conflict with or
      replace opencl-icd, so that other OpenCL ICDs can be installed
      at the same time.

 -- Alberto Milone <email address hidden> Thu, 12 Apr 2018 16:45:37 +0200

This bug was fixed in the package ubuntu-drivers-common - 1:0.5.1

---------------
ubuntu-drivers-common (1:0.5.1) bionic; urgency=medium

  * debian/control:
    - Build depend on dh-python.
    - Drop dh-systemd, and use debhelper (>= 9.20160709).
  * tests/ubuntu_drivers.py:
    - Raise time to 40.0 for i386 in
      test_system_driver_packages_performance().
  * NvidiaDetector/nvidiadetector.py:
    - Fix crash with new name scheme of the NVIDIA
      packages (LP: #1054458).
  * gpu-manager.(c|py):
    - Create a xorg.conf.d file for systems with PRIME.
      This way we can drop the PrimaryGpu option from
      the main NVIDIA configuration xorg.conf.d snippet,
      and generate an additional custom snippet on demand
      with that option, to avoid causing problems on systems
      with multiple NVIDIA GPUs (LP: #1756226).

 -- Alberto Milone <email address hidden> Thu, 12 Apr 2018 16:35:07 +0200

...........:~$ apt-cache policy nvidia-390
nvidia-390:
  Installeret: 390.87-0ubuntu0~gpu18.04.2
  Kandidat: 390.87-0ubuntu0~gpu18.04.2

....
.............:~$ apt-cache policy ubuntu-drivers-common
ubuntu-drivers-common:
  Installeret: 1:0.5.2.3
  Kandidat: 1:0.5.2.3

I seem be having this issue ? how can i verify? if i nvidia active i get black screen with a cursor indicated in top left, and i hear the ubunti logon sound. and then nothing.

i have to do:
 sudo prime-select intel

to get back in to GUI

please help