nvidia-drivers-331.20 was installed as dependency of sci-libs/cholmod-2.1.2 (cuda ? x11-drivers/nvidia-drivers) due to the cuda USE flag. During booting I see almost endless: [ 361.360673] NVRM: No NVIDIA graphics adapter probed! [ 361.360674] [drm] Module unloaded [ 361.360756] NVRM: NVIDIA init module failed! [ 361.409703] NVRM: The NVIDIA probe routine was not called for 1 device(s). [ 361.409707] NVRM: This can occur when a driver such as: NVRM: nouveau, rivafb, nvidiafb or rivatv NVRM: was loaded and obtained ownership of the NVIDIA device(s). [ 361.409709] NVRM: Try unloading the conflicting kernel module (and/or NVRM: reconfigure your kernel without the conflicting NVRM: driver(s)), then try loading the NVIDIA kernel module NVRM: again. After some minutes these messages are finished and I can login, but I can't start X11. In /var/log/Xorg.0.log I find: [ 507.208] (II) NOUVEAU driver for NVIDIA chipset families : [ 507.208] RIVA TNT (NV04) [ 507.208] RIVA TNT2 (NV05) [ 507.208] GeForce 256 (NV10) [ 507.208] GeForce 2 (NV11, NV15) [ 507.208] GeForce 4MX (NV17, NV18) [ 507.208] GeForce 3 (NV20) [ 507.208] GeForce 4Ti (NV25, NV28) [ 507.208] GeForce FX (NV3x) [ 507.208] GeForce 6 (NV4x) [ 507.208] GeForce 7 (G7x) [ 507.208] GeForce 8 (G8x) [ 507.208] GeForce GTX 200 (NVA0) [ 507.209] GeForce GTX 400 (NVC0) [ 507.209] (II) modesetting: Driver for Modesetting Kernel Drivers: kms [ 507.209] (++) using VT number 7 [ 507.218] (II) [drm] nouveau interface version: 1.1.1 [ 507.218] (WW) Falling back to old probe method for modesetting [ 507.218] (II) Loading sub module "dri2" [ 507.218] (II) LoadModule: "dri2" [ 507.218] (II) Module "dri2" already built-in [ 507.218] (--) NOUVEAU(0): Chipset: "NVIDIA NV86" ... and then endless: [ 508.648] removing GPU device /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card1 /dev/dri/card1 [ 508.648] (II) config/udev: Adding drm device (/dev/dri/card1) [ 508.648] (II) xfree86: Adding drm device (/dev/dri/card1) [ 508.685] (II) NOUVEAU(0): EDID vendor "IBM", prod id 10375 [ 508.685] (II) NOUVEAU(0): Printing DDC gathered Modelines: [ 508.685] (II) NOUVEAU(0): Modeline "1680x1050"x0.0 122.00 1680 1712 1776 1904 1050 1051 1054 1066 -hsync -vsync (64.1 kHz eP) [ 508.685] (II) NOUVEAU(0): Modeline "1680x1050"x0.0 101.67 1680 1712 1776 1904 1050 1051 1054 1066 -hsync -vsync (53.4 kHz e) ... [ 858.258] removing GPU device /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card1 /dev/dri/card1 [ 858.258] (II) config/udev: Adding drm device (/dev/dri/card1) [ 858.258] (II) xfree86: Adding drm device (/dev/dri/card1) [ 858.259] removing GPU device /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card1 /dev/dri/card1 [ 858.259] xf86: remove device 1 /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card1 It happens at least on two systems. After removing nvidia-drivers the system boots and starts X11 again as expected.
Yes, you cannot run both nouveau and nvidia-drivers at the same time.
But I do not want to run nvidia-drivers, it is only needed by cholmod and cuda respectively. And it was working for two years without any problems.
(In reply to Juergen Rose from comment #2) > But I do not want to run nvidia-drivers, it is only needed by cholmod and > cuda respectively. And it was working for two years without any problems. You can't have two drivers (nouveau.ko and nvidia.ko) managing the same device, even if they serve different purposes as a display driver or general purpose processor. The upstream README in /usr/share/doc/nvidia-drivers/*/ explains how you can blacklist the nouveau driver in order to use the nvidia driver (at "INTERACTION WITH THE NOUVEAU DRIVER").
(In reply to Jeroen Roovers from comment #3) > (In reply to Juergen Rose from comment #2) > > But I do not want to run nvidia-drivers, it is only needed by cholmod and > > cuda respectively. And it was working for two years without any problems. > > You can't have two drivers (nouveau.ko and nvidia.ko) managing the same > device, even if they serve different purposes as a display driver or general > purpose processor. > > The upstream README in /usr/share/doc/nvidia-drivers/*/ explains how you can > blacklist the nouveau driver in order to use the nvidia driver (at > "INTERACTION WITH THE NOUVEAU DRIVER"). Thanks Jeroen for the hint. But I do not want to use nvidia-drivers as a display driver and disable nouveau. I want only to use cuda. Is there no more any way to use cuda, if nouveau is used as the display driver?
Hi Jeroen, could it be helpfull to blacklist the nvidia.ko module? Does cuda need the kernel module or the libraries from nvidia-drivers?
I removed again nvidia-drivers and addidtional cuda USE flag from /etc/make.conf.
