Stuck with nvidia graphics only when using the latest NVIDIA 384/387 drivers

[Dylan Borg]

On a brand new laptop with GTX 1050 TI and Kaby Lake graphics I cannot switch back to intel's GPU after installing the nvidia drivers + nvidia-prime. prime-select lists the nvidia driver as well as a prime variant as the only available alternatives. When switching to intel and logging out I get failed logins. If I reboot my screen freezes on login. It is possible to do a hard reboot, switch back to nvidia mode in a TTY, reboot and login after this happens.

ProblemType: Bug
DistroRelease: Ubuntu 17.10
Package: nvidia-prime 0.8.4
Uname: Linux 4.13.5-041305-generic x86_64
NonfreeKernelModules: nvidia_uvm nvidia_drm nvidia_modeset nvidia
ApportVersion: 2.20.7-0ubuntu2
Architecture: amd64
CurrentDesktop: GNOME
Date: Sun Oct 8 09:42:05 2017
InstallationDate: Installed on 2017-10-02 (5 days ago)
InstallationMedia: Ubuntu 17.10 "Artful Aardvark" - Beta amd64 (20171002)
SourcePackage: nvidia-prime
UpgradeStatus: No upgrade log present (probably fresh install)

[dino99 (9d9)]

there is no wayland session for optimus devices in 17.10

https://askubuntu.com/questions/907727/no-way-to-choose-wayland-in-gdm-using-ubuntugnome-17-04
https://devtalk.nvidia.com/default/topic/1010548/linux/bug-gnome-won-t-load-wayland-session-with-latest-nvidia-381-22-drivers-ubuntu-gnome-17-04/

[dino99 (9d9)]

Until nvidia pushed a new driver version supporting that feature, the workaround is to login on a xorg session instaed of the default wayland one.

[Dylan Borg (borgdylan)]

I am using X11 throughout. Wayland has not been used anywhere.

[Dylan Borg (borgdylan)]

Output when running prime-select:

dylan@zenbook-pro:~$ sudo prime-select intel
[sudo] password for dylan:
Info: the current GL alternatives in use are: ['nvidia-387', 'nvidia-387']
Info: the current EGL alternatives in use are: ['nvidia-387', 'nvidia-387']
Info: selecting nvidia-387-prime for the intel profile
update-alternatives: using /usr/lib/nvidia-387-prime/ld.so.conf to provide /etc/ld.so.conf.d/x86_64-linux-gnu_GL.conf (x86_64-linux-gnu_gl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387-prime/ld.so.conf to provide /etc/ld.so.conf.d/x86_64-linux-gnu_EGL.conf (x86_64-linux-gnu_egl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387-prime/alt_ld.so.conf to provide /etc/ld.so.conf.d/i386-linux-gnu_GL.conf (i386-linux-gnu_gl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387-prime/alt_ld.so.conf to provide /etc/ld.so.conf.d/i386-linux-gnu_EGL.conf (i386-linux-gnu_egl_conf) in manual mode
dylan@zenbook-pro:~$ sudo prime-select nvidia
Info: the current GL alternatives in use are: ['nvidia-387-prime', 'nvidia-387-prime']
Info: the current EGL alternatives in use are: ['nvidia-387-prime', 'nvidia-387-prime']
Info: selecting nvidia-387 for the nvidia profile
update-alternatives: using /usr/lib/nvidia-387/ld.so.conf to provide /etc/ld.so.conf.d/x86_64-linux-gnu_GL.conf (x86_64-linux-gnu_gl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387/ld.so.conf to provide /etc/ld.so.conf.d/x86_64-linux-gnu_EGL.conf (x86_64-linux-gnu_egl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387/alt_ld.so.conf to provide /etc/ld.so.conf.d/i386-linux-gnu_GL.conf (i386-linux-gnu_gl_conf) in manual mode
update-alternatives: using /usr/lib/nvidia-387/alt_ld.so.conf to provide /etc/ld.so.conf.d/i386-linux-gnu_EGL.conf (i386-linux-gnu_egl_conf) in manual mode

[Alberto Milone (albertomilone)]

Can you reproduce the problem on the same exact system (kernel, etc.) when using the 375 driver?

[Dylan Borg (borgdylan)]

Is a version of 375 that supports 1050Ti available in the repos (ubuntu or graphics PPA)?

[Dylan Borg (borgdylan)]

The version of nvidia-375 I am seeing points to nvidia-384. So I cannot install nvidia-375 without breaking my system (i.e. using a .run file that wouldn't work with nvidia-prime anyways)

[Dylan Borg (borgdylan)]

Coming to think of it I have not tried nvidia 381. Should I?

[Alberto Milone (albertomilone)]

You can remove nvidia-384 and the transitional packages for nvidia-385, then you can download 375 from the following PPA, and install the deb packages manually:
https://launchpad.net/~graphics-drivers/+archive/ubuntu/ppa

[Dylan Borg (borgdylan)]

nvidia-381 drove both my fans to max :/

[Dylan Borg (borgdylan)]

I have used force version in synaptic to use the PPA version of the packages.

[Dylan Borg (borgdylan)]

With nvidia-375 from the PPA I get the same issue + the fans at max issue as with nvidia-381.

[Alberto Milone (albertomilone)]

It could be a kernel issue. I have seen other systems not playing well with bbswitch disabling the dGPU. I'll see if I can reproduce the problem here.

[Dylan Borg (borgdylan)]

I have seen it happen with both the 4.13.4 kernel currently in artful-proposed as well as the 4.13.5 upstream kernel. I am currently on the upstream kernel since 4.13.4 has an annoying shutdown hang and 4.13.5 does not.

[Alberto Milone (albertomilone)]

what is the newest kernel you can't reproduce the problem with? (you can use the kernel mainline PPA if you want to help testing)

[Dylan Borg (borgdylan)]

None. I cannot even try 4.14 because of some GPL symbol issue which results in the DKMS build of the nvidia driver to fail. As of now, the latest kernel series that support teh binary driver is 4.13.

[Dylan Borg (borgdylan)]

Looking at: https://github.com/Bumblebee-Project/bbswitch/issues/140 indicates that this has been broken since 4.8 when new power management code was enabled for PCIE ports. They suggest using pcie_port_pm=off in teh kernel command line. I will try that whenever I can and see where that leaves me.

[Dylan Borg (borgdylan)]

That kernel boot parameter did not work for me. Also in teh non functional intel mode I am getting fans at maximum even with the recent nvidia drivers. Some GPU switching is really going wrong in intel mode.

[Dylan Borg (borgdylan)]

Could it be that the mesa GL libraries are incorrectly installed or that some library is not in the search path when redirecting calls?

[Dylan Borg (borgdylan)]

For referece, the bbswitch load log:
[ 4.492602] bbswitch: version 0.8
[ 4.492606] bbswitch: Found integrated VGA device 0000:00:02.0: \_SB_.PCI0.GFX0
[ 4.492611] bbswitch: Found discrete VGA device 0000:01:00.0: \_SB_.PCI0.PEG0.PEGP
[ 4.492699] bbswitch: detected an Optimus _DSM function
[ 4.492708] bbswitch: Succesfully loaded. Discrete card 0000:01:00.0 is on

[Dylan Borg (borgdylan)]

list of gl libraries loadable in intel mode

[Dylan Borg (borgdylan)]

list of gl libraries loadable in nvidia mode

[Dylan Borg (borgdylan)]

Looks like libGLX is only lodable in nvidia mode, some library paths may be missing in the intel prime profile

[Dylan Borg (borgdylan)]

I found an alternate implementation of prime-select, and modified it slightly. With it, if I switch to intel mode , then reboot I manage to login with the integrated graphics getting used as expected.

[Launchpad Janitor (janitor)]

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