nvidia version driver update fails (340.58 -> 340.65) , driver does not work

You did with your last log.

And now fbdev is used because the nvidia kernel module can’t be loaded.
For the likely reason, see my previous post.

Install the correct *-kmp-desktop packages and the driver should hopefully work.

And please decide whether you want to solve this problem in the bug report or here.
You only create confusion by jumping back and forth between the forum and bugzilla.

The next time you probably should ask here in the forum first before filing a bug report, to see whether there actually is a bug.
Problems caused by left-overs from a previous nvidia installation from some home: repo (or the .run installer) are not a bug in the nvidia packages… :wink:

i uninstalled all nvidia packets
then
i uninstalled kernel-default-devel-3.11.10-21.1.x86_64 .
this implied to uninstall kernel-syms-3.11.10-21.1.x86_64 and kernel-syms-3.11.10-17.1.x86_64

kernel-default-devel-3.11.10-17.2.x86_64 is in the list but yast indicates it is not installed

after uninstalling kernel-default-devel-3.11.10-21.1.x86_64 i see that
kernel-default-devel-3.11.10-17.2.x86_64 disappeared in the list


:~ # rpm -qa | egrep "nvidia|kernel"
kernel-desktop-3.11.10-17.2.x86_64
kernel-devel-3.11.10-21.1.noarch
kernel-source-3.11.10-17.2.noarch
kernel-desktop-devel-3.11.10-21.1.x86_64
kernel-desktop-devel-3.11.10-17.2.x86_64
kernel-firmware-20130714git-2.21.1.noarch
kernel-desktop-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-17.2.x86_64
patterns-openSUSE-devel_kernel-13.1-13.6.1.x86_64
kernel-source-3.11.10-21.1.noarch
kernel-devel-3.11.10-17.2.noarch
:~ # 




You did with your last log.

yes some rare time i have the time to do a liitle job before to get an unreadble screen

And please decide whether you want to solve this problem in the bug report or here.
You only create confusion by jumping back and forth between the forum and bugzilla.

yes but how to join a file with the forum ?

SUSE Paste

i try to install the nvidia packets but when i click on x11-video-nvidiaG03
then
yast wants to install
nvidia-uvm-gfxG03-kmp-pae
and
nvidia-gfxG03-kmp-pae

and not the desktop versions

it’s strange it wants to install pae version my system is a 64 bit system .

we are now at the beginning of the story . when booting we have an opensuse graphical panel then a black screen with a text line at the top

here is the xorg lo http://paste.opensuse.org/39605476

:~ # rpm -qa | egrep “nvidia|kernel”
kernel-desktop-3.11.10-17.2.x86_64
kernel-devel-3.11.10-21.1.noarch
kernel-source-3.11.10-17.2.noarch
nvidia-computeG03-340.65-36.1.x86_64
nvidia-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-desktop-devel-3.11.10-21.1.x86_64
nvidia-uvm-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-desktop-devel-3.11.10-17.2.x86_64
kernel-firmware-20130714git-2.21.1.noarch
kernel-desktop-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-17.2.x86_64
x11-video-nvidiaG03-340.65-36.1.x86_64
patterns-openSUSE-devel_kernel-13.1-13.6.1.x86_64
kernel-source-3.11.10-21.1.noarch
nvidia-glG03-340.65-36.1.x86_64
kernel-devel-3.11.10-17.2.noarch
:~ #

where to read the kernel log ?

You should not uninstall kernel-default-devel (and I never told you to do so).
As you see kernel-syms requires kernel-default-devel, and kernel-syms is needed to compile the nvidia kernel module I think.

Yes, because by uninstalling the *-kmp-desktop packages you effectively told YaST/zypper that it should not install it at all any more, so it wants to install *-kmp-pae instead (or *-kmp-default before).
And I already told you a few times that you should select nvidia-uvm-gfxG03-kmp-desktop and nvidia-gfxG03-kmp-desktop manually. This should automatically deselect the *-kmp-pae which only make sense with kernel-pae (which is 32bit).

it’s strange it wants to install pae version my system is a 64 bit system .

No, that’s libzypp’s “SoftLocks” feature that I already explained a few times.
Uninstalling a package soft-locks it and it is never being installed automatically any more.
As a kmp package is required, YaST installs a different one instead which is not locked.

Not quite.
At the beginning the driver did not work correctly because a libglx in a different version was used.

here is the xorg lo SUSE Paste

Well, the kernel module still fails to load.
Have you tried to reboot?

Please post the output of “dmesg|grep -i nvidia”

Maybe try to reinstall the kernel module and post errors that might appear:

sudo zypper in -f nvidia-gfxG03-kmp-desktop

But what I actually find strange in that Xorg log is that in the end nouveau is used. But nouveau should be blacklisted by the nvidia packages and should not be able to be loaded at all.
Apparently the blacklist is missing and the nouveau kernel module is loaded which prevents the nvidia kernel module from being loaded.

As the blacklist is contained in the kmp package, reinstalling that should fix this as well.

:~ # rpm -qa | egrep “nvidia|kernel”
kernel-desktop-3.11.10-17.2.x86_64
kernel-devel-3.11.10-21.1.noarch
kernel-source-3.11.10-17.2.noarch
nvidia-computeG03-340.65-36.1.x86_64
nvidia-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-desktop-devel-3.11.10-21.1.x86_64
nvidia-uvm-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-desktop-devel-3.11.10-17.2.x86_64
kernel-firmware-20130714git-2.21.1.noarch
kernel-desktop-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-21.1.x86_64
kernel-xen-devel-3.11.10-17.2.x86_64
x11-video-nvidiaG03-340.65-36.1.x86_64
patterns-openSUSE-devel_kernel-13.1-13.6.1.x86_64
kernel-source-3.11.10-21.1.noarch
nvidia-glG03-340.65-36.1.x86_64
kernel-devel-3.11.10-17.2.noarch
:~ #

Looks ok now.
But try to re-install nvidia-gfxG03-kmp-desktop as mentioned above.

where to read the kernel log ?

“dmesg”.

Apparently not. It seems that kernel-devel is sufficient.

I have the G02 drivers and kernel-devel. Nothing has brought in kernel-syms.

Correct.
kernel-syms is a so-called “BuildRequires”, i.e. you need it for creating the package (but not for compiling the kernel module itself).
I mixed that up, sorry.

But kernel-devel alone is not sufficient, for installing the kmp-desktop packages (i.e. compiling the kernel modules) these packages are required:
kernel-desktop-devel (which requires kernel-devel in turn), make, and gcc
They will be installed automatically by YaST or zypper if necessary of course when you tell it to install nvidia-gfxG03-kmp-desktop.

You should not uninstall kernel-default-devel (and I never told you to do so).
As you see kernel-syms requires kernel-default-devel, and kernel-syms is needed to compile the nvidia kernel module I think.

i reinstalled these packets

i uninstalled the nvidia packets

i installed the nvidia packets

i check that in /etc/modeprobe.d/nvidia-desktop.conf there is “blacklist nouveau”

i check in kernel there is NO_KMS_IN_INITRD=“yes”

:~ # rpm -qa | egrep “nvidia|kernel”
kernel-desktop-3.11.10-17.2.x86_64
kernel-devel-3.11.10-21.1.noarch
nvidia-glG03-340.65-36.1.x86_64
kernel-source-3.11.10-17.2.noarch
kernel-desktop-devel-3.11.10-21.1.x86_64
nvidia-uvm-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-desktop-devel-3.11.10-17.2.x86_64
kernel-xen-devel-3.11.10-21.1.x86_64
kernel-firmware-20130714git-2.21.1.noarch
kernel-desktop-3.11.10-21.1.x86_64
kernel-syms-3.11.10-21.1.x86_64
x11-video-nvidiaG03-340.65-36.1.x86_64
patterns-openSUSE-devel_kernel-13.1-13.6.1.x86_64
kernel-default-devel-3.11.10-21.1.x86_64
kernel-source-3.11.10-21.1.noarch
nvidia-computeG03-340.65-36.1.x86_64
nvidia-gfxG03-kmp-desktop-340.65_k3.11.6_4-36.1.x86_64
kernel-devel-3.11.10-17.2.noarch
:~ #

soon some results

i started pc 2 times then same bad result

the xorg log http://paste.opensuse.org/73773680

an extract from “messages” file : messages about the second start of the pc http://paste.opensuse.org/78675122

it’s interesting because we have also NVRM messages about nvidia and xorg server which dies and kdm which does not start.

an interesting thing : when switching to nouveau i forget to set to “no” “no_kms_in_initrd” the i get same black screen witha text line at the top . i wonder if is a kms pb for the nvidia driver.

Hm, it still uses nouveau which shouldn’t be possible if nouveau is blacklisted.
Try to recreate the initrd, maybe nouveau is still in there:

sudo /sbin/mkinitrd

If there are errors, please post them.

And/or try to add “nomodeset” to the kernel boot options, although this shouldn’t be necessary either if nouveau is blacklisted and not included in the initrd.

an extract from “messages” file : messages about the second start of the pc SUSE Paste

it’s interesting because we have also NVRM messages about nvidia and xorg server which dies and kdm which does not start.

Right. This doesn’t seem to be from the same boot…

The Xorg log is from Sun Dec 14 16:08:26 2014, whereas according to “messages”, KDM (and Xorg) has been started at 2014-12-14T17:59:16.781603+01:00 .

So X is not starting at all when you are using the nvidia driver?
Your Xorg log is useless then, it is from earlier boots, before nouveau was blacklisted apparently.

Does it work when you try to start Xorg manually in text mode?
I.e. “init 3” followed by “init 5”, or running “startx” (as root).

If X doesn’t even try to start, this does sound related to the latest change to the packages:

Xwrapper: bail out, if an existing module cannot be unloaded;
  this can happen if a second Xsession starts

This is related to 904383 – nvidia version driver update fails (340.46 -> 340.58) it seems. Unloading the nvidia modules when starting X shall prevent the problem that nvidia doesn’t work on the first boot after the installation, AIUI.

Can you please post your /usr/bin/X.x11-video-nvidia-G03 ?
Try to change this:

      # bail out, if an existing module cannot be unloaded
      # this can happen if a second Xsession starts
      /sbin/rmmod -v $m || exit 1

to just

 /sbin/rmmod -v $m

in that file, i.e. remove the " || exit 1".
Maybe this would make X start successfully? I don’t see a point in exiting there anyway if unloading a kernel module fails.

an interesting thing : when switching to nouveau i forget to set to “no” “no_kms_in_initrd” the i get same black screen witha text line at the top .

That’s normal, and actually caused by plymouth. plymouth initializes the graphics display already, and is started from the initrd normally. If nouveau isn’t part of the initrd (i.e. when you have NO_KMS_IN_INITRD=“yes”), it cannot load nouveau and uses a generic driver instead. But this prevents nouveau from correctly initializing the graphics later on (and in particular loading the firmware if necessary) when it is actually loaded and causes problems. The same happens with other KMS drivers like radeon or intel.

i wonder if is a kms pb for the nvidia driver.

No. nvidia doesn’t use nor support KMS. But there should not be a problem if KMS is enabled, unless nouveau is loaded already.

I find the best way to untangle the NVIDA dependencies when they get messed up is to go to Yast-software management search for nvidia assuming G03 be sure that all 5 packages are the same version that all have G03 in them and they all come from the same vendor (version tab bottom right) that the two with a kernel flavor in the name all match the kernel flavor you are using In almost all cases that will be desktop.

1 reboot after /sbin/mkinitrd (i checked no_kms_in_initrd = yes)

os sticks to a first opensuse graphical panel
then
a big througput of text lines without end

i tried to boot 2 times

:~ # /sbin/mkinitrd

Kernel image: /boot/vmlinuz-3.11.10-17-desktop
Initrd image: /boot/initrd-3.11.10-17-desktop
KMS drivers: nouveau nvidia
Root device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part2 (/dev/sdc2) (mounted on / as ext4)
Resume device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part1 (/dev/sdc1)
Microcode: AMD CPU family: 0xf does not support microcode updates
Kernel Modules: thermal_sys thermal processor fan scsi_dh scsi_dh_emc scsi_dh_alua scsi_dh_rdac scsi_dh_hp_sw button wmi video mxm-wmi i2c-algo-bit drm drm_kms_helper ttm nouveau nvidia sata_nv xhci-hcd hid-logitech-dj hid-holtek-kbd hid-lenovo-tpkbd hid-ortek hid-roccat hid-roccat-common hid-roccat-arvo hid-roccat-isku hid-samsung ohci-pci
Features: acpi amd_microcode kms plymouth block usb resume.userspace resume.kernel

Kernel image: /boot/vmlinuz-3.11.10-21-desktop
Initrd image: /boot/initrd-3.11.10-21-desktop
KMS drivers: nouveau nvidia
Root device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part2 (/dev/sdc2) (mounted on / as ext4)
Resume device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part1 (/dev/sdc1)
Microcode: AMD CPU family: 0xf does not support microcode updates
Kernel Modules: thermal_sys thermal processor fan scsi_dh scsi_dh_emc scsi_dh_alua scsi_dh_rdac scsi_dh_hp_sw button wmi video mxm-wmi i2c-algo-bit drm drm_kms_helper ttm nouveau nvidia sata_nv xhci-hcd hid-logitech-dj hid-holtek-kbd hid-lenovo-tpkbd hid-ortek hid-roccat hid-roccat-common hid-roccat-arvo hid-roccat-isku hid-samsung ohci-pci
Features: acpi amd_microcode kms plymouth block usb resume.userspace resume.kernel
:~ #

  1. reboot with nomodeset

success : we get the kdm template to open a kde session

my conclusion : a pb about kms somewhere

I think this shouldn’t be there at all if you really have NO_KMS_IN_INITRD=“yes” in /etc/sysconfig/kernel.
mkinitrd adds both nouveau and nvidia to the initrd, while it should add none of those with NO_KMS_IN_INITRD=“yes”.

So probably something is wrong in your /etc/sysconfig/kernel, can you please post the whole file?

  1. reboot with nomodeset

success : we get the kdm template to open a kde session

my conclusion : a pb about kms somewhere

Yes, looks like it. You shouldn’t have nouveau and nvidia in the initrd, especially not if NO_KMS_IN_INITRD=“yes” is set.

And maybe this triggers the “exit 1” that prevents Xorg from starting afterwards. And it definitely caused your problem mentioned in 904383 – nvidia version driver update fails (340.46 -> 340.58) as well IMHO (because nvidia is part of the initrd, but nvidia-uvm is not).

here is the kernel file http://paste.opensuse.org/50947946

i did an experiment . i execute again /sbin/mkinitrd while nvidia running thus no_kms_in_initrd = yes

this time i can boot without nomedeset and get a kdm template to open a session and nvidia driver is well launch .

i assume this :

when installing nvidia driver this is nouveau which runs thus no_kms_in_initrd = no
then
nvidia installer creates an initrd wrongly for nouveau instead of for nvidia

if you see the log of /sbin/mkinitrd while nvidia runs there is no “KMS drivers: nouveau nvidia” line

:~ # /sbin/mkinitrd

Kernel image: /boot/vmlinuz-3.11.10-17-desktop
Initrd image: /boot/initrd-3.11.10-17-desktop
Root device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part2 (/dev/sdc2) (mounted on / as ext4)
Resume device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part1 (/dev/sdc1)
Microcode: AMD CPU family: 0xf does not support microcode updates
Kernel Modules: thermal_sys thermal processor fan scsi_dh scsi_dh_emc scsi_dh_alua scsi_dh_rdac scsi_dh_hp_sw sata_nv xhci-hcd hid-logitech-dj hid-holtek-kbd hid-lenovo-tpkbd hid-ortek hid-roccat hid-roccat-common hid-roccat-arvo hid-roccat-isku hid-samsung ohci-pci
Features: acpi amd_microcode plymouth block usb resume.userspace resume.kernel

Kernel image: /boot/vmlinuz-3.11.10-21-desktop
Initrd image: /boot/initrd-3.11.10-21-desktop
Root device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part2 (/dev/sdc2) (mounted on / as ext4)
Resume device: /dev/disk/by-id/ata-ST1000DM003-1CH162_S1D7EL8Q-part1 (/dev/sdc1)
Microcode: AMD CPU family: 0xf does not support microcode updates
Kernel Modules: thermal_sys thermal processor fan scsi_dh scsi_dh_emc scsi_dh_alua scsi_dh_rdac scsi_dh_hp_sw sata_nv xhci-hcd hid-logitech-dj hid-holtek-kbd hid-lenovo-tpkbd hid-ortek hid-roccat hid-roccat-common hid-roccat-arvo hid-roccat-isku hid-samsung ohci-pci
Features: acpi amd_microcode plymouth block usb resume.userspace resume.kernel
:~ #

Hm, this actually looks fine.

i did an experiment . i execute again /sbin/mkinitrd while nvidia running thus no_kms_in_initrd = yes

this time i can boot without nomedeset and get a kdm template to open a session and nvidia driver is well launch .

Ok, so everything seems to be ok now.

i assume this :

when installing nvidia driver this is nouveau which runs thus no_kms_in_initrd = no

“NO_KMS_IN_INITRD” is a setting in a configuration file. It tells mkinitrd whether it should add the KMS drivers to the initrd or not.
And this setting is not changed automatically no matter which driver “runs”.

then
nvidia installer creates an initrd wrongly for nouveau instead of for nvidia

Hm? The nvidia installer doesn’t create any initrd.
mkinitrd does.
The nvidia packages do set NO_KMS_IN_INITRD=“yes” on installation and run mkinitrd though. This is to prevent nouveau from being loaded in the initrd.

if you see the log of /sbin/mkinitrd while nvidia runs there is no “KMS drivers: nouveau nvidia” line

Again, it should not matter which driver runs. NO_KMS_IN_INITRD=“yes” in particular means no nouveau driver in the initrd regardless whether nouveau is currently in use or not. Otherwise you couldn’t even install the nvidia driver successfully in the first place.

Did you change NO_KMS_IN_INITRD yourself manually when you “ran” nouveau?
This would of course explain why mkinitrd would add the drivers, because NO_KMS_IN_INITRD was actually set to “no” when you executed mkinitrd, and why /etc/sysconfig/kernel is correct now and everything works.

Did you change NO_KMS_IN_INITRD yourself manually when you “ran” nouveau?

yes i know only this way to get a working conf when nvidia installation fails

there are still some remaining (warning) problems .

in xorg.0.log we get :


26.950] (II) Loading sub module “fb”
26.950] (II) LoadModule: “fb”
26.957] (II) Loading /usr/lib64/xorg/modules/libfb.so
26.958] (II) Module fb: vendor=“X.Org Foundation”
26.958] compiled for 1.14.3.901, module version = 1.0.0
26.958] ABI class: X.Org ANSI C Emulation, version 0.4
26.958] (WW) Unresolved symbol: fbGetGCPrivateKey
26.958] (II) Loading sub module “wfb”
26.958] (II) LoadModule: “wfb”
26.959] (II) Loading /usr/lib64/xorg/modules/libwfb.so
26.959] (II) Module wfb: vendor=“X.Org Foundation”
26.959] compiled for 1.14.3.901, module version = 1.0.0
26.959] ABI class: X.Org ANSI C Emulation, version 0.4
26.959] (II) Loading sub module “ramdac”
26.959] (II) LoadModule: “ramdac”
26.959] (II) Module “ramdac” already built-in
26.960] (WW) Falling back to old probe method for modesetting
26.960] (WW) Falling back to old probe method for fbdev
26.960] (II) Loading sub module “fbdevhw”
26.960] (II) LoadModule: “fbdevhw”
26.961] (II) Loading /usr/lib64/xorg/modules/libfbdevhw.so
26.961] (II) Module fbdevhw: vendor=“X.Org Foundation”
26.961] compiled for 1.14.3.901, module version = 0.0.2
26.961] ABI class: X.Org Video Driver, version 14.1
26.961] (WW) Falling back to old probe method for vesa
26.961] (II) NVIDIA(0): Creating default Display subsection in Screen section
“Default Screen Section” for depth/fbbpp 24/32

also in “messages” file there are some warning problems .

2014-12-15T11:41:01.912317+01:00 linux-b4lz kernel: 21.729451] NVRM: Your system is not currently configured to drive a VGA console
2014-12-15T11:41:01.912330+01:00 linux-b4lz kernel: 21.729457] NVRM: on the primary VGA device. The NVIDIA Linux graphics driver
2014-12-15T11:41:01.912332+01:00 linux-b4lz kernel: 21.729459] NVRM: requires the use of a text-mode VGA console. Use of other console
2014-12-15T11:41:01.912334+01:00 linux-b4lz kernel: 21.729462] NVRM: drivers including, but not limited to, vesafb, may result in
2014-12-15T11:41:01.912336+01:00 linux-b4lz kernel: 21.729463] NVRM: corruption and stability problems, and is not supported.