search

z0rc / debumblebee

DEPRECATED. Don't use it anymore — Optimus graphics support for Debian through VirtualGL
http://suwako.nomanga.net/
GNU General Public License v3.0
26 stars 2 forks source link

nvidia driver not loading - related to issue 22 #23

Closed DEvry closed 13 years ago

DEvry commented 13 years ago

First, thanks for the hack to try and solve the nvidia driver problem with xorg 1.11, per https://github.com/z0rc/debumblebee/issues/22. I tried the manual hack last night, but to no avail. I tried the new debumblebee workaround this morning using the uninstall / install method. Unfortunately, it still doesn't work. I can't seem to force the nvidia driver to load.

I'm not sure if my issue is related to the latest bug in the nvidia driver in sid or not. I altered /etc/default/debumblebee and set ON_DEMAND=no. Rebooted. Still, the Intel i915 MESA video loads by default.

z0rc commented 13 years ago

Provide me the output of update-alternatives --display glx and dpkg -s nvidia-kernel-dkms. Also I need to see the contents of /var/log/Xorg.0.log and /var/log/Xorg.8.log with default installation when trying to run optirun glxgears.

z0rc commented 13 years ago

Also Xlib: extension "NV-GLX" missing on display ":0" is very strange as it should be using display ":8". Provide me your /etc/default/debumblebee too.

DEvry commented 13 years ago

Sorry for the delay. Before I post the details of your request, I uninstalled debumblebee, purged everything again, installed just the nvidia drivers and attempted to log into X with the BIOS set to Discrete. It didn't work. Just a blank screen. Had to reboot in single user mode.

I then purged nvidia drivers, reinstalled debumblebee, and booted. Now glxgears -info shows same Mesa driver, but frames run at 970 - 1100 frames per second. Output of optirun glxgears -info is a blank screen with 250-260 frames per second (completely blank). It gives the message: failed to create drawable. Both commands output the following information:

GL_RENDERER = Software Rasterizer GL_VERSION = 2.1 Mesa 7.11 GL_VENDOR = Mesa Project

i915 video driver is loaded during both attempts.

In theory, I should have a completely fresh installation from the debumblebee install.sh script. Note that I had to change the display to 0 in /etc/default/debumblebee to get any output. Otherwise, it complains there's no display device.

Output of update-alternatives --display glx: http://pastebin.com/m2Z1pJHn

Output of dpkg -s nvidia-kernel-dkms: http://pastebin.com/xLK1GxUH Note that the references to the 2.6.32 kernel are odd. I have two kernels available, but am running 3.0.0-1 and nvidia was built with those kernel headers. The other kernel has to stay around a bit longer for some specific reasons, but it is rarely used. No nvidia support has been built for that kernel.

While running optirun glxgears.... Output of /var/log/Xorg.0.log: http://pastebin.com/wtPz1a3A Output of /var/log/Xorg.8.log: http://pastebin.com/0xcnWU7u

It's clear from both logs that there are problems loading the nvidia module and it complains about not being able to find screens...unfortunately, graphics aren't my area of expertise. I appreciate the help.

Output of /etc/default/debumblebee: http://pastebin.com/d0Yxcv43

As I mentioned, I changed VGL_DISPLAY=8 to VGL_DISPLAY=0. All other values I've checked (between 1 and 9) give the following error: [VGL] ERROR: Could not open display :8. (or whatever display number I give it)

Thanks for the help.

DE

z0rc commented 13 years ago

[ 314.638] (EE) Failed to load /usr/local/lib/debumblebee/libglx.so: /usr/local/lib/debumblebee/libglx.so: cannot open shared object file: No such file or directory OK, what do ls -la /usr/local/lib/debumblebee/libglx.so, ls -la /usr/lib/nvidia/, ls -la /etc/alternatives/nvidia--libglx.so and ls -la /usr/lib/nvidia/current/libglx.so say?

Also two X servers (main one and second for nVidia) cannot share the same screen. So VGL shows the blank screen when we ask it to perform transfer from main X server to main X server.

DEvry commented 13 years ago

Ok. Here's the problem. libglx.so is a non existent. Should the missing locations be symbolic links or copies of the lib?

ls -la /usr/local/lib/debumblebee/libglx.so lrwxrwxrwx 1 root staff 25 Aug 31 14:22 /usr/local/lib/debumblebee/libglx.so -> /usr/lib/nvidia/libglx.so

ls -la /usr/lib/nvidia
total 84 drwxr-xr-x 2 root root 248 Aug 31 14:22 . drwxr-xr-x 232 root root 69816 Aug 31 14:22 .. -rwxr-xr-x 1 root root 1775 Jul 4 12:30 check-for-conflicting-opengl-libraries -rwxr-xr-x 1 root root 1341 Jul 4 07:29 check-for-mismatching-nvidia-module -rwxr-xr-x 1 root root 2566 Jul 12 05:26 create-xorg-nvidia-conf-experimental -rwxr-xr-x 1 root root 441 Apr 27 06:08 pre-install

ls -la /etc/alternatives/nvidia--libglx.so ls: cannot access /etc/alternatives/nvidia--libglx.so: No such file or directory

ls -la /usr/lib/nvidia/current/libglx.so ls: cannot access /usr/lib/nvidia/current/libglx.so: No such file or directory

Thx

z0rc commented 13 years ago

Aha, as I see now you don't have all required nvidia packages installed. Though they aren't correctly installable right now, as they conflict current Xorg 1.11 again. nVidia packages versioned 280.13-3 which were pushed into sid yesterday or today aren't work with current Xorg server.

So there are three possible solutions. First you can perform uninstall and wait for nvidia packages update, then do install again. Second is to downgrade Xorg packages to 1.10.4 from Testing, and perform uninstall-install sequence. Third one is to use nvidia packages versioned 280.13-2, but I have no idea where you can get them as they aren't present in current repositories.

Though in my case I'm stuck with third solution now (haven't upgraded from 280.13-2 to 280.13-3) and also waiting for update from nVidia.

z0rc commented 13 years ago

Hrm, maybe I should try to add nouveau support for such cases...

DEvry commented 13 years ago

I think I'll downgrade Xorg to 1.10 later today just to make sure everything's working, and when nvidia gets the sid update fixed, I'll upgrade everything and see how it goes.

I presume I should boot into single user mode, uninstall, downgrade to xorg 1.10, and to be on the safe side, reboot to single user mode, then install.sh again. Would that be the correct sequence? I want to doublecheck to make sure since clearly the full nvidia package didn't install last time...

Thanks for your help with this.

z0rc commented 13 years ago

I presume I should boot into single user mode, uninstall, downgrade to xorg 1.10, and to be on the safe side, reboot to single user mode, then install.sh again. Would that be the correct sequence? I want to doublecheck to make sure since clearly the full nvidia package didn't install last time...

Well, single user mode isn't required, you can perform Xorg downgrade even with running DE. Though, if you want to make sure, you can just switch to another display (Ctrl+Alt+F1) and stop kdm/gdm service, than start it once you finished with downgrade.

Everything else is fine. Good luck with it.

SnouF commented 13 years ago

Hi

I think I had the same bug, after installing the nvidia drivers from expermimental it's works !

Aptosid on Asus U36SD

z0rc commented 13 years ago

Debian driver packages versioned 280.13.really.275.28-1 work fine. Closing ticket.