Bug 27247

Thank you for this report.
This hardware looks weird, described as "MSI GeForce RTX 2060 SUPER VENTUS OC" it has its own MSI web site:
 https://www.msi.com/graphics-card/geforce-rtx-2060-super-ventus-oc

> I've installed nvidia driver 450.66 via the run file of nvidia site
> and all is ok (With Mageia 7 and 8)
It looks as if for M7 at least, we have just '340' & '390'; I recall seeing 450 somewhere, but cannot find it now.

Assigning to the kernel/drivers team.

Assignee: bugsquad => kernel

Hi,

You should not use Nouveau (open source) driver with Nvidia RTX 2060 as it doesn't support it.

Normally, our installer DrakX for M7 should propose you to install Nonfree Nvidia 430-serie, as support for your card was introduced by Nvidia in 415.27 version of Nvidia drivers. I don't know why this not functioning. Does DrakX ever proposed you to install nonfree driver?

I suggest:
1) With M7 installed and use of VESA driver:
Use 450-serie driver from nonfree backports repository by activate him:
In Mageia Control Center: Software Management --> Select from where software packages are downloaded when updating the system.
Activate backports repository by checking it for x86_64.
In rpmdrake, look for x11-driver-video-nvidia-current-450.57-3 and install it.
Run in Hardware --> Setup the graphical server. Choose first button and select in Vendor -> Nvidia -> Geforce 635 series and later.
Save and reboot.

2) With M8 B1, Nvidia driver series 450 are already proposed by default with online repositories activated when proposed by DrakX. Offline install will fail.
Meanwhile, when M8 will be published, appropriate nvidia nonfree driver will be here.

So, I think we could close this after you test my proposal.

For MGA8 there are drivers 450.66 (glvnd enabled) in nonfree/updates_testing (I suggest also to use ldetect-lst from core/updates_testing, so that it has the newer pcitable).

For MGA7 there are (see bug #27105) drivers 450.57 in nonfree/backports

CC: (none) => ghibomgx

Created attachment 11869 [details]
Mga 7 : install driver 450.57 and launch XFdrake

I've tried to install backported driver on Mga 7.
That doesn't work. The reboot is stopped.
About my first post, VESA driver is finally ok with Mga7.
I've reinstalled manually NVIDIA-Linux-x86_64-450.66.run and that's work

Moreover, i've seen in the boot sequence a message, somewhat like "gpu not detected: i2c timeout"
It is always present even with NVIDIA-Linux-x86_64-450.66.run but that works.

Is it possible there is a link with this thread : https://bugzilla.kernel.org/show_bug.cgi?id=206653

I will try proposal test for Mageia 8 in the next week.

(In reply to Patrice Brunelle from comment #4)
> Created attachment 11869 [details]
> Mga 7 : install driver 450.57 and launch XFdrake
> 
> I've tried to install backported driver on Mga 7.
> That doesn't work. The reboot is stopped.

Could you produce output of:
# dkms status

With backported 450.57 drivers?

> About my first post, VESA driver is finally ok with Mga7.
> I've reinstalled manually NVIDIA-Linux-x86_64-450.66.run and that's work
If so.
But, don't forget that during kernel update, nvidia drivers must be reinstalled.

> Moreover, i've seen in the boot sequence a message, somewhat like "gpu not
> detected: i2c timeout"
> It is always present even with NVIDIA-Linux-x86_64-450.66.run but that works.
> Is it possible there is a link with this thread :
> https://bugzilla.kernel.org/show_bug.cgi?id=206653

Do you see lines like this:
nvidia-gpu 0000:01:00.3: i2c timeout error e0000000
ucsi_ccg 0-0008: i2c_transfer failed -110
ucsi_ccg 0-0008: ucsi_ccg_init failed - -110
ucsi_ccg: probe of 0-0008 failed with error -110

These messages are not harmful.
They are silenced by a patch in our 5.8.6 kernel in M8.

(In reply to Aurelien Oudelet from comment #5)
> (In reply to Patrice Brunelle from comment #4)
> > Created attachment 11869 [details]
> > Mga 7 : install driver 450.57 and launch XFdrake
> > 
> > I've tried to install backported driver on Mga 7.
> > That doesn't work. The reboot is stopped.
> 
> Could you produce output of:
> # dkms status
> 
> With backported 450.57 drivers?
> 

I've launched nvidia-uninstall to uninstall driver 450.66 and I've launched XFdrake to install 450.57 from backport repository.
I could not reboot in graphical mode, i've reboot in recovery mode and dkms status command print :
"nvidia-current, 450.57-3.mga7.nonfree, 5.7.19-desktop-1.mga7, x86_64: installed  (WARNING! Diff between built and installed module!) (WARNING! Diff between built and installed module!) (WARNING! Diff between built and installed module!)"

I've line in boot saying nvidia.ko and nvidia_modeset.ko differ (450.66 & 450.57)

After reinstall 450.66 from nvidia website, dkms status print :
"#dkms status
nvidia-current, 450.57-3.mga7.nonfree, 5.7.19-desktop-1.mga7, x86_64: installed  (WARNING! Diff between built and installed module!) (WARNING! Diff between built and installed module!) (WARNING! Diff between built and installed module!)
nvidia, 450.66, 5.7.19-desktop-1.mga7, x86_64: installed"

but X works correctly

> > Moreover, i've seen in the boot sequence a message, somewhat like "gpu not
> > detected: i2c timeout"
> > It is always present even with NVIDIA-Linux-x86_64-450.66.run but that works.
> > Is it possible there is a link with this thread :
> > https://bugzilla.kernel.org/show_bug.cgi?id=206653
> 
> Do you see lines like this:
> nvidia-gpu 0000:01:00.3: i2c timeout error e0000000
> ucsi_ccg 0-0008: i2c_transfer failed -110
> ucsi_ccg 0-0008: ucsi_ccg_init failed - -110
> ucsi_ccg: probe of 0-0008 failed with error -110
> 
> These messages are not harmful.
> They are silenced by a patch in our 5.8.6 kernel in M8.

Yes

Booting with nokmsboot does change something? Using newer ldetect-lst (for mga8 from updates_testing)?

Also check there aren't nvidia kernel modules debris coming from different installations out of rpm/dkms (e.g. find / -type f -name 'nvidia*.ko.*' -print) that maybe aren't properly uninstalled.

I've tested with ldetect-lst (for mga8 from updates_testing) on Mageia 8.
In this case when XFdrake try to install the nvidia driver, there is a windows saying that the driver cannot be installed after 20 or 30 seconds.

Hi

Since a week, the driver nvidia don't install. With XFdrake, I have the same error message that described by Patrice. I try to reinstall mageia 8 with a netinstall install. I reproduce the same problem. Something is broken but I am not determining the cause as much as dkms seems to be doing its job correctly. Is this an Xfdrake bug?

Nb : I use mageia 8 beta with 1060 nvidia graphic card.

CC: (none) => jeanmichel.varvou

Can you try with drivers 455.23.04beta from nonfree/updates_testing?

Hi

At the end, XFdrake wrote : "The proprietary driver could not be installed correctly".So the system reinstall nouveau..

But dkms has build the driver :
[vouf@localhost ~]$ dkms status
nvidia-current, 455.23.04-2.mga8.nonfree, 5.8.10-desktop-1.mga8, x86_64: installed 


I don't know where to look to see which step the installation failed.

See the same message from XFDrake since NVIDIA 450.66.

Workaround is:
1) install from Beta 1 ISO and leave system offline while installing.
2) Let DrakX install NVIDIA non free driver from ISO.
3) Reboot and let first boot run dkms.
4) Go online and add online repos.
5) update with mgaapplet or urpmi --auto-update.
6) Never go back to XFDrake : it bugs and reinstall nouveau drivers.(In reply to 


Jean Michel Varvou from comment #11)
> Hi
> 
> At the end, XFdrake wrote : "The proprietary driver could not be installed
> correctly".So the system reinstall nouveau..
> 
> But dkms has build the driver :
> [vouf@localhost ~]$ dkms status
> nvidia-current, 455.23.04-2.mga8.nonfree, 5.8.10-desktop-1.mga8, x86_64:
> installed 
> 
> 
> I don't know where to look to see which step the installation failed.

CC: (none) => ouaurelien

I wonder whether drakx11 is influenced by the error code as of bug #27080 (which happens nevertheless modules are built successfully).

To complete, a manual procedure, once you already had installed all the required nvidia packages, e.g. these:

dkms-nvidia-current-455.23.04-2.mga8.nonfree
nvidia-current-cuda-opencl-455.23.04-2.mga8.nonfree
nvidia-current-doc-html-455.23.04-2.mga8.nonfree
nvidia-current-lib32-455.23.04-2.mga8.nonfree
nvidia-current-utils-455.23.04-2.mga8.nonfree
x11-driver-video-nvidia-current-455.23.04-2.mga8.nonfree

you can:

1) Edit /etc/X11/xorg.conf

2) Replace the nouveau block device section in /etc/X11/xorg.conf with:

Section "Device"
    Identifier "device1"
    VendorName "NVIDIA Corporation"
    BoardName "NVIDIA GeForce 635 series and later"
    Driver "nvidia"
    Option "DPMS"
    Option "AddARGBGLXVisuals"
    Option "DynamicTwinView" "false"
EndSection

3) check there are no mismatch nvidia modules in kernel: in this case 'dmesg' should tell if there are mismatches, with some warning like:

NVRM: API mismatch: the client has the version XXX.YYY.ZZZ, but this kernel module has the version UUU.VVV.WWW.

3) run:

   update-alternatives --set gl_conf /etc/nvidia-current/ld.so.conf

this will install the proper modprobe commands in /etc/modprobe.d/, which in the end is to add a file with:

install nvidia /sbin/modprobe nvidia-current $CMDLINE_OPTS

Optionally if you get conflicts with preloaded nouveau modules regarding modesetting, you can add a file in /etc/modprobe.d/10-blacklist-nouveau.conf, with an entry:

options nouveau modeset=0

or specify nouveau.modeset=0 in grub booting cmdline.

4) reboot

Hi

Thank you Aurélien. But I use the workaround proposed by Giuseppe. That's works with my graphic card 1060 nvidia with driver 455.23.04 and Mageia 8..

It will therefore remain to correct the XFdrake anomaly.

@Guiseppe, XFDrake works when you install Mageia 7 and Mageia 8 Beta 1 (offline for the later).

But starting with nvidia 450.66 in nonfree repo in Cauldron XFDrake does no longer work.

I found the culprit here in:

http://gitweb.mageia.org/software/drakx-kbd-mouse-x11/tree/lib/Xconfig/proprietary.pm?id=e7e7b854938ae9209a5572ded4ecaa2fb15109c4#n120

line 120: the file /usr/lib64/nvidia-current/xorg/libglx.so was used for validating the proprietary driver proper installation, even if it's not really used anymore (and thus removed). I reintroduced back as a symlink, so to not potentially break older non-glvnd drivers.

Fixed for now in nvidia-current-455.23.04-6.mga8 in nonfree/updates_testing.

Hi

I tested the package nvidia-current-455.23.04-6.mga8.Indeed, the fix is effective and allows to install the proprietary driver with the XFdrake tool without any problem.

Note that this fix should be applied on the 450.66 driver on nonfree_release as long as version 455.23 is not switched to non_free_release. Version 150.66 seems to be affected by the same problem.

Thank you for your intervention

fixed long time ago

Status: NEW => RESOLVED
Resolution: (none) => FIXED