lightdm - Job failed to start (xorg problem?!) / bumblebee/cuda related?

Question

Please work through my wall of text, need help!! Thank you!

Booting up my laptop this morning, I encountered the following problem: Boot to black screen with no responsiveness, couldn't even get to tty1 from there.

Then got to a console by pressing F3 on the boot-up splash with the Ubuntu logo and the dots, logged in and tried stopping/starting lightdm manually with the following error messages:

stop: Job has already been stopped: lightdm
start: Job failed to start

So I started digging into log files to try and figure out what is going wrong. /var/log/lightdm/lightdm.log has the following lines:

DEBUG: Process 1754 terminated with signal 6
DEBUG: DisplayServer x-0: X server stopped
DEBUG: Releasing VT 7
DEBUG: DisplayServer x-0: Removing X server authority /var/run/lightdm/root/:0
DEBUG: Seat: Display server stopped
DEBUG: Seat: Stopping; greeter display server failed to start

That ought to be an xorg problem somehow then, right? I checked /var/log/lightdm/x-0.log

Loading extension GLK
(EE)
Fatal server error:
(EE) no screens found(EE)
(EE)

and in /var/log/Xorg.0.log when it tries to (i assume) find the drivers, it is looking for intel, intel, modestting, fbdev, vesa - cannot open any and:

(EE) No drivers available.
(EE)
Fatal server error:
(EE) no screens found(EE)
(EE)

I tried a lot of reinstalling / reconfiguring lightdm and trying out gdm, to no success - which makes sense given it seems to be a driver/display issue?!

So here is background on that: Yd, I installed bumblebee and nvidia-352 and got it working for the first time. I did not boot to black screen and could configure bumblebee, such that glxsphere64 either uses the intel driver or the nvidia driver and card with optirun.

At that point, I installed, after a lot of installing/finding required packages, CUDA 7.5 as well, tested it on the cuda-test deviceQuery and got the expected results. I then also installed matlab all before rebooting / shutting down. It is after this shutdown that the problems occured - so either something happened to the bumblebee config or the cuda install, yes? I cannot for the life of me figure out what is going on, I am not very proficient at these tings..

/etc/bumblebee/xorg.conf.nvidia points to the right places, I did manage to load up the nvidia card/driver with optirun so thats expected.

/etc/bumblebee/bumblebee.conf only has 1 unexpected line, it says:

VirtualDisplay=:8

and I cannot figure out if that is an expected entry here, don't wanna fiddle around with it though.

sudo X -configure tells me I have a missing output driver.

I tried removing the Xorg.conf files, with no results. bumblebee also is not loaded up on the limited boot-up I am getting, as I have no experience I do not know if that is expected. sudo service bumblebee start starts it no problem, but that does not solve any issues as far as I can see. lsmod | grep nvidia and lsmod | grep nouveau do not give any results, I should expect that as nouveau is not installed and the nvidia drivers should only be used for optirun apps, yea? lsmod | grep i9 gives back i915 as well as video *stuff* i915 - that is the kernel driver for my onboard graphics unit, correct? With lspci | grep VGA it returns the Haswell Integrated Graphics Controller.

So, has anyone any clue what might be going wrong here?

Redoing the driver/bumblebee install is of course always an option - as it was such a hassle I am trying to avoid it in fear of not reproducing it or ending up at the same spot.

If I have to resort to it anyway: Do you think I can just keep the CUDA install? Will matlab figure out a newly installed graphics driver?

Here is an update:

I could not figure out anyway to fix this, so I reinstalled, went through the same steps to set up bumblebee and the nvidia-352 drivers - up to here it works! What is different is this: I did not explicitly install virtualgl and libraries so i cannot test with glxspheres64 for example - will that be problematic? What is the best ppa to add to install all that, its does not seem to be in the standard package? I also did not install cuda, I suspect that is where it went wrong! Does installing cuda via package manager and the 1.9GB Deb file also overwrite openGL stuff?

Answer 1

Hmm sounds like an nvidia-driver problem!

Can you uninstall cuda? maybe that is the problem. Did you update before shutting down? That may also be a problem!

The instructions come from here: Nvidia Driver and Heating Problem

In terminal: remove CUDA (I don't know how)

reinstall bumblebee bumblebee-nvidia nvidia-352 nvidia-352-uvm primus nvidia-cuda-toolkit libcuda1-352

With the nvidia-352-uvm package you actually have CUDA.

configure bumblebee like this:

sudo nano /etc/bumblebee/bumblebee.conf

line 22: Driver=nvidia

line 55: KernelDriver=nvidia-352

line 58: LibraryPath=/usr/lib/nvidia-352:/usr/lib32/nvidia-352

line 61: XorgModulePath=/usr/lib/nvidia-352/xorg,/usr/lib/xorg/modules

blacklist it:

sudo nano /etc/modprobe.d/bumblebee.conf

make sure the line

blacklist nvidia-352

is there

Finally add the intel driver and bumblebee switch (bbswitch) to /etc/modules

sudo nano /etc/modules

add those lines:

i915
bbswitch

Then it should work alright.

Samsung NP550P5C S02 intel core i5-3210m+ Intel HD 4000, Geforce GT 650M

EDIT:

1 last thing: You may need to run these commands to get cuda and blender working:

sudo modprobe nvidia-uvm //loads nvidia-uvm
sudo blender             // run blender as superuser, else it doesn't recognize it the first time (you dont need optirun/primusrun for cuda)

then select "CUDA" in the user preferences.. then close blender.

Afterwards for the rest of the day you can use blender cuda without sudo

Only do this if it doesnt work Out-Of-The-Box