Bug 20192

Please reproduce the problem and switch to a VT with e.g. Ctrl+Alt+F3, log in, become root and run

    journalctl -ab > journal.txt


and attach journal.txt to this bug report.

Keywords: (none) => NEEDINFO
CC: (none) => marja11
Assignee: bugsquad => kernel

Created attachment 8911 [details]
Journal.txt

result of journalctl -ab on a failing system

Ok, I've posted the journal. 
I want to tell than a friend suffer from the same bug. He run 32bit server as well but his hardware is different.
Only common things are intel cpu and nvidia gpu, both being different models than mines.

I've forgot to say that there it's possible that the bug is from Nvidia itself, since someone, who reported the same bug on the support forum of mageia, tried to use the nvidia installer whithout success.
Here is the link : https://forums.mageia.org/en/viewtopic.php?f=7&t=11586

And sorry for the broken english, i tend to reformulate things and forget some words behind before posting... an edit function would be nice.

Next user reports X does not start after driver update to x11-driver-video-nvidia-current-375.26-1.mga5.nonfree with nVidia GTX 750 Ti.

Errors from journalctl:
(EE) No devices detected.
(EE) Fatal server error:
(EE) no screens found(EE)

Temporary solution:
- to install a driver GeForce 8100 to GeForce 415 (packages of nvidia340, not nvidia-current)

OS:
===
$ cat /etc/issue
Mageia release 5 (Official) for i586
Kernel 4.4.39-desktop-1.mga5 on a Dual-processor i686 / \l

Graphic card:
=============
$ lspci -nn|grep -i vga
01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM107 [GeForce GTX 750 Ti] [10de:1380] (rev a2)

Sources:
========
$ urpmq --list-media active --list-url
Core Release (distrib1) http://mageia.supp.name/distrib/5/i586/media/core/release
Core Updates (distrib3) http://mageia.supp.name/distrib/5/i586/media/core/updates
Nonfree Release (distrib11) http://mageia.supp.name/distrib/5/i586/media/nonfree/release
Nonfree Updates (distrib13) http://mageia.supp.name/distrib/5/i586/media/nonfree/updates
Tainted Release (distrib21) http://mageia.supp.name/distrib/5/i586/media/tainted/release
Tainted Updates (distrib23) http://mageia.supp.name/distrib/5/i586/media/tainted/updates

Installed packages:
===================
$rpm -qa | grep -i -e kernel -e nvidia | sort
dkms-nvidia-current-375.26-1.mga5.nonfree
dkms-nvidia340-340.101-1.mga5.nonfree
kernel-desktop-devel-latest-4.4.39-1.mga5
kernel-desktop-devel-4.4.36-2.mga5-1-1.mga5
kernel-desktop-devel-4.4.39-1.mga5-1-1.mga5
kernel-desktop-latest-4.4.39-1.mga5
kernel-desktop-4.4.36-2.mga5-1-1.mga5
kernel-desktop-4.4.39-1.mga5-1-1.mga5
kernel-firmware-nonfree-20160914-1.mga5.nonfree
kernel-firmware-20160409-1.mga5
kernel-userspace-headers-4.4.39-1.mga5
nvidia-current-doc-html-375.26-1.mga5.nonfree
nvidia340-doc-html-340.101-1.mga5.nonfree
x11-driver-video-nvidia-current-375.26-1.mga5.nonfree
x11-driver-video-nvidia340-340.101-1.mga5.nonfree

CC: (none) => yullaw

Created attachment 8922 [details]
Xorg.0.log

Created attachment 8923 [details]
journalctl -ba

Also valid in cauldron with latest dkms-nvidia-current (375.39-5.mga6.nonfree). The driver builds and loads OK on a 64-bit system, but not on a 32-bit system. The important line from attachment 8911 [details] is:

janv. 31 05:27:02 localhost kernel: nvidia: Unknown symbol phys_to_dma (err 0)

which prevents the kernel module being loaded. This same error is seen in the dkms build log.

CC: (none) => mageia
Keywords: NEEDINFO => (none)
Source RPM: nvidia-current-375.26-1.mga5.nonfree => nvidia-current-375.26-1.mga5.nonfree, nvidia-current-375.39-5.mga6.nonfree.src.rpm

Created attachment 9181 [details]
dkms build log

Crap, the phys_to_dma detection was supposed to be fixed in new nvidia-current drivers, but now it seems only 304 and 340 series were fixed :/

It means I have to build a new kernel set to restore that phys_to_dma export to get this working again...

CC: (none) => tmb

(In reply to Martin Whitaker from comment #10)
> Created attachment 9181 [details]
> dkms build log

Looking on this I see:

 CONFTEST: phys_to_dma

but it apparently fails

Actually I see nvidia is trying to use a kernel internal symbol :/

Ah, I see it's https://bugs.mageia.org/show_bug.cgi?id=17604 that strikes again :/

I wond dupe it as it an advisory bug, but I'll fix the driver the same way as before

And as http://advisories.mageia.org/MGASA-2017-0025.html pushed the broken driver to mga5, I will update that one too..

Ok, I've pushed a fixed nvidia-current-375.39-6.mga6 to cauldron, 

and a nvidia-current-375.39-1.mga5 to mga5 nonfree updates_testing

Please try them out...

The mga6 version builds and loads correctly. I don't have the necessary H/W to check it actually works...

(In reply to Thomas Backlund from comment #15)
> 
> Ok, I've pushed a fixed nvidia-current-375.39-6.mga6 to cauldron, 
> 
> and a nvidia-current-375.39-1.mga5 to mga5 nonfree updates_testing
> 
> Please try them out...

(In reply to Martin Whitaker from comment #16)
> The mga6 version builds and loads correctly. I don't have the necessary H/W
> to check it actually works...

CC'ing QA team and all packagers collectively, because I can't test it either.

CC: (none) => pkg-bugs, qa-bugs

I haven't checked this in a while but thanks for your work !

The 375.39-1 package from mga5 nonfree updates_testing load and work !

I haven't much tested for regressions but it seem fine.

(In reply to Didier Le Gloanic from comment #18)
> I haven't checked this in a while but thanks for your work !
> 
> The 375.39-1 package from mga5 nonfree updates_testing load and work !
> 
> I haven't much tested for regressions but it seem fine.

Thanks for the feedback.

No one commented here that it wasn't fixed for him/her, so closing as FIXED

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