The driver package now recommends the nvidia-gfxG03-uvm-kmp to be installed as well (it’s needed for CUDA support).
I guess you uninstalled nvidia-gfxG03-uvm-kmp-desktop once already, so it had been added to your /var/lib/zypp/SoftLocks file. In that case, YaST/zypper doesn’t want to install it any more automatically, and chooses the -default flavor instead (and kernel-default to satisfy the dependencies).
So either remove nvidia-gfxG03-uvm-kmp-desktop from /var/lib/zypp/SoftLocks (with text editor f.e.), or select nvidia-gfxG03-uvm-kmp-desktop manually.
The kernel-default should then get deselected automatically.
This is not really related to the driver update (except for the added dependency), but is a (not new) “feature” of libzypp.
This “SoftLocks” behavior has been removed in Factory already though, so such things should not happen any more with 13.2.
I had a look at the packages, and nvidia-uvm-gfxG03-kmp-desktop does indeed require kernel-desktop-3.11.6.
That’s a packaging bug apparently and should be reported at http://bugzilla.novell.com/.
It should not really matter though, since you can have more than one kernel installed (so it’s actually no “downgrade”, the 3.11.6 kernel will be installed additionally unless you disabled that).
When using KMP packages, symlinks are created for all installed kernel versions, so that package should work fine with kernel-desktop 3.11.10-11, which will still be booted afterwards as it is the highest version.
Well, as I said.
kernel-desktop-3.11.6 is installed in addition. Nothing is “downgraded”.
And the rest of the packages that are to be updated in your screenshot are completely unrelated to this.
It’s a too strict dependency in the nvidia-gfxG03-uvm-kmp packages, they require the exact kernel version.
That should be losened, or separate packages have to be provided for all available kernel versions (i.e. 3.11.6, 3.11.10-7.1, 3.11.10-11.1 for 13.1 atm).
And if you are not very careful you could mess up the system
I don’t think so. How?
But as I said, this is clearly a packaging bug, and somebody should report it.
I can’t really, as I don’t even have an nvidia system.
I may do it some time next week (as I should be able then to at least try to update the driver myself on an nvidia system), but would rather prefer somebody else reporting it earlier.
Oh, and btw, you don’t have to install the nvidia-gfxG03-uvm-kmp package at all if you don’t want to. It is only recommended.
But since it is available in different flavors, you might have to taboo all of them.
It is (only) required for CUDA support, i.e. using the GPU for calculations/rendering (in blender f.e.) instead of the CPU.
Oh, and btw, you don’t have to install the nvidia-gfxG03-uvm-kmp package at all if you don’t want to. It is only recommended.
That’s the route I’ve taken… thanks.
Also tried to remove ‘nvidia-computeG03’ as I reasoned that wasn’t needed either … but that wanted to auto-remove
‘x11-video-nvidiaG03’. Now I’m not at all sure what is actually needed, but that sounded like I should keep it…
No.
That bug report is about the kmp packages only available for kernel 3.11.6, i.e. the kernel shipped with openSUSE 13.1.
But that is no bug, that’s how KMP packages are designed to work: you provide/install them for one kernel version, when the KMP package is installed, symlinks are created for all other installed kernels.
That also eliminates the need to re-install a KMP package after a kernel update.
And that’s why that bug report is closed as INVALID.
Here we have a different issue: nvidia-uvm-gfxG03-kmp explicitely requires kernel 3.11.6 as RPM dependency.
Also tried to remove ‘nvidia-computeG03’ as I reasoned that wasn’t needed either … but that wanted to auto-remove
‘x11-video-nvidiaG03’. Now I’m not at all sure what is actually needed, but that sounded like I should keep it…
Well, AFAIK nvidia-computeG03 should also only be needed for CUDA, but this is required by the x11-video-nvidiaG03 package, so cannot be uninstalled.
For accelerated (3D) video you need x11-video-nvidiaG03, nvidia-gfxG03-kmp-xxx and nvidia-glG03.
You can uninstall the last one, but then you have no 3D hardware acceleration (GL). According to the package changelog this might be useful for Optimus systems to still be able to use the intel driver (nvidia-glG03 replaces system libraries, namely libglx and libGL, with nvidia’s own incompatible versions, which breaks all Mesa drivers like intel). In this case you can still use the nvidia GPU for CUDA.
Well, AFAIK nvidia-computeG03 should also only be needed for CUDA, but this is required by the x11-video-nvidiaG03 package, so cannot be uninstalled.
For accelerated (3D) video you need x11-video-nvidiaG03, nvidia-gfxG03-kmp-xxx and nvidia-glG03.
You can uninstall the last one, but then you have no 3D hardware acceleration (GL). According to the package changelog this might be useful for Optimus systems to still be able to use the intel driver (nvidia-glG03 replaces system libraries, namely libglx and libGL, with nvidia’s own incompatible versions, which breaks all Mesa drivers like intel). In this case you can still use the nvidia GPU for CUDA.
Excellent concise description of what’s needed… Many thanks.
Fri Jun 6 13:30:08 UTC 2014 - [EMAIL="sndirsch@suse.com"]sndirsch@suse.com
[/EMAIL] - get rid of the requirement to a specific kernel version
(bnc#881723)
It might still take some time before the fixed packages appear in nvidia’s repo of course…
This fix is still not on the NVIDIA repo.
Yast Software Management still shows a requirement for the older kernel module.
Any idea when it might be on the NVIDIA repo?
For a more precise answer you would have to contact nvidia directly I suppose…
But if you really want to update the driver now, just install kernel 3.11.6 to fulfill the dependency. Grub2 should still boot the latest 3.11.10 kernel afterwards, and the nvidia driver should work fine with that as well.
If you do not want to risk anything, be patient.
Although I had no need to ‘update’ these I just did…
Curiously, on the first reboot the kernel module failed to load.
6.405] (II) LoadModule: "glx"
6.409] (II) Loading /usr/lib64/xorg/modules/updates/extensions/libglx.so
6.617] (II) Module glx: vendor="NVIDIA Corporation"
6.617] compiled for 4.0.2, module version = 1.0.0
6.617] Module class: X.Org Server Extension
6.617] (II) NVIDIA GLX Module 331.79 Sat May 17 23:58:32 PDT 2014
6.618] Loading extension GLX
6.618] (II) LoadModule: "nvidia"
6.618] (II) Loading /usr/lib64/xorg/modules/updates/drivers/nvidia_drv.so
6.630] (II) Module nvidia: vendor="NVIDIA Corporation"
6.630] compiled for 4.0.2, module version = 1.0.0
6.630] Module class: X.Org Video Driver
6.630] (II) NVIDIA dlloader X Driver 331.79 Sat May 17 23:37:42 PDT 2014
6.631] (II) NVIDIA Unified Driver for all Supported NVIDIA GPUs
6.631] (++) using VT number 7
6.632] (II) Loading sub module "fb"
6.632] (II) LoadModule: "fb"
6.633] (II) Loading /usr/lib64/xorg/modules/libfb.so
6.635] (II) Module fb: vendor="X.Org Foundation"
6.635] compiled for 1.14.3.901, module version = 1.0.0
6.635] ABI class: X.Org ANSI C Emulation, version 0.4
6.635] (WW) Unresolved symbol: fbGetGCPrivateKey
6.635] (II) Loading sub module "wfb"
6.635] (II) LoadModule: "wfb"
6.636] (II) Loading /usr/lib64/xorg/modules/libwfb.so
6.638] (II) Module wfb: vendor="X.Org Foundation"
6.638] compiled for 1.14.3.901, module version = 1.0.0
6.638] ABI class: X.Org ANSI C Emulation, version 0.4
6.638] (II) Loading sub module "ramdac"
6.638] (II) LoadModule: "ramdac"
6.638] (II) Module "ramdac" already built-in
6.696] (EE) NVIDIA: Failed to load the NVIDIA kernel module. Please check your
6.696] (EE) NVIDIA: system's kernel log for additional error messages.
6.696] (EE) No devices detected.
6.696] (EE)
Fatal server error:
6.696] (EE) no screens found(EE)
6.696] (EE)
Please consult the The X.Org Foundation support
at http://wiki.x.org
for help.
6.696] (EE) Please also check the log file at "/var/log/Xorg.0.log" for additional information.
6.696] (EE)
So maybe there’s indeed something wrong?
But I have no idea at the moment why it would work then from second reboot on. This doesn’t really make sense to me.
Well, I haven’t tried to install the 331.79 driver packages myself, I suppose I will be able to either tomorrow or on Friday.
No, not in this case. On first boot after install there was no GUI at all. After Grub I found myself at the login prompt of, I believe, tty1. It was from there that I logged in and did a ‘shutdown -r’. The second (and subsequent) reboots were normal. It could, of course just been a weird coincidence.
Well, I haven’t tried to install the 331.79 driver packages myself, I suppose I will be able to either tomorrow or on Friday.
… and your verdict then … WORKSFORME rotfl!
Edit: The ‘update’ (of 331.79) was without error also, no additional RPM output. I didn’t install nvidia-uvm-gfxG03-kmp-desktop, as I have no need for it.
You probably have an xorg.conf in place, because in your log snippet X doesn’t even try to load other drivers.
If you have an xorg.conf that tells X to load the nvidia driver and that fails, you would drop back to text mode because X would fail to start completely.
The other two probably didn’t have an xorg.conf (caf4926 definitely didn’t according to his Xorg log), so X used the fbdev driver instead as fallback when it couldn’t load nvidia.
But as I said, I don’t have any possible explanation for the module failing to load on the first reboot, but being successfully loaded ever thereafter…
… and your verdict then … WORKSFORME rotfl!
We’ll see.
I’m definitely eager to see whether I can reproduce this problem or not.
The question is if I can find out more in case I would see it as well.
If not, I’m not sure if it even makes sense to report this.
Excellent concise description of what’s needed… Many thanks.