CentOS - Mar 2015 - Strange crash after NVidia driver installation from ELRepo

Hi,

I just had the weirdest of system crashed. Here goes.

I wanted to install a desktop step by step, as I usually do. First the 
base system, then the X Window System, then the WindowMaker window 
manager. Everything went OK, then I installed kmod-nvidia from ELRepo. 
Yum informed me that the driver conflicted with x11-glamor (something 
like that), so I removed the corresponding package.

While doing all this, the system still defaulted to boot in console mode 
(meaning multi-user.target and not graphical.target). To test X, I just 
launched WindowMaker using startx.

Anyway, I rebooted, and then... nothing. System refused even to startup, 
and systemd suggested to me that I open journalctl (which taught me 
nothing).

Right now I've started a Slax LiveCD to backup all my data and reinstall 
the whole thing from scratch. I don't have the slightest clue as to what 
could possibly have happened here. The only thing I wonder: could it be 
that installing the kmod-nvidia driver while booting to a default 
console (and not graphical) could lead to this weirdness? A previous 
installation on that machine went OK, but the difference was that the 
first time, I started with a full GNOME installation right away.

Any suggestions?

Niki

On Tue, March 3, 2015 9:41 am, info at microlinux.fr
wrote:
> Hi,
>
> I just had the weirdest of system crashed. Here goes.
>
> I wanted to install a desktop step by step, as I usually do. First the
> base system, then the X Window System, then the WindowMaker window
> manager. Everything went OK, then I installed kmod-nvidia from ELRepo.
> Yum informed me that the driver conflicted with x11-glamor (something
> like that), so I removed the corresponding package.
>
> While doing all this, the system still defaulted to boot in console mode
> (meaning multi-user.target and not graphical.target). To test X, I just
> launched WindowMaker using startx.
>
> Anyway, I rebooted, and then... nothing. System refused even to startup,
> and systemd suggested to me that I open journalctl (which taught me
> nothing).
>
> Right now I've started a Slax LiveCD to backup all my data and
reinstall
> the whole thing from scratch. I don't have the slightest clue as to
what
> could possibly have happened here. The only thing I wonder: could it be
> that installing the kmod-nvidia driver while booting to a default
> console (and not graphical) could lead to this weirdness? A previous
> installation on that machine went OK, but the difference was that the
> first time, I started with a full GNOME installation right away.
>
> Any suggestions?
If I were to troubleshoot this I would do it the following way:
1. wipe everything and set up standard desktop set with Xwindow
2. install nvidia driver
3. remove packages you don't like and replace them with the ones you like
to achieve your final configuration

Why? Because all usually is tested in pretty much standard set of
packages, thus if you follow the route above you will unlikely have your
box bricked after installing nvidia driver. Then in step 3 you will
discover doing which of your customizations breaks things. I call it for
myself lazy person easy path approach ;-)

Just my $0.02.

Valeri

++++++++++++++++++++++++++++++++++++++++
Valeri Galtsev
Sr System Administrator
Department of Astronomy and Astrophysics
Kavli Institute for Cosmological Physics
University of Chicago
Phone: 773-702-4247
++++++++++++++++++++++++++++++++++++++++

On Tue, 3 Mar 2015, info at microlinux.fr wrote:
> Any suggestions?
There are multiple nvidia drivers in Elrepo, depending on the model of 
video card. Install and run nvidia-detect to find out which driver you 
need; just installing the "kmod-nvidia" package is not guaranteed to
give
you a working driver.

Steve

Le 03.03.2015 17:00, Steve Thompson a ?crit?:
> On Tue, 3 Mar 2015, info at microlinux.fr wrote:
> 
> Any suggestions?
> 
> There are multiple nvidia drivers in Elrepo, depending on the model of
> video card. Install and run nvidia-detect to find out which driver you
> need; just installing the "kmod-nvidia" package is not guaranteed
to
> give you a working driver.
I used nvidia-detect, and installed kmod-nvidia. Even then, the previous 
installation ran kmod-nvidia with success on twin monitors.

And even then, installing the wrong NVidia driver should only result in 
X failing to start... but not in a system that won't even *boot*. 
Startup messages stopped short immediately, I'd say right after 
initramfs was loaded.

On a side note: before using CentOS, I ran Slackware for a few years, so 
I know how to configure X and video drivers manually. I've never seen 
this sort of weird crash.

Niki
> 
> Steve
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos

info at microlinux.fr wrote:
>
> I just had the weirdest of system crashed. Here goes.
>
> I wanted to install a desktop step by step, as I usually do. First the
> base system, then the X Window System, then the WindowMaker window
> manager. Everything went OK, then I installed kmod-nvidia from ELRepo.
> Yum informed me that the driver conflicted with x11-glamor (something
> like that), so I removed the corresponding package.
That would be the openGL libraries. kmod-nvidia will install its
own.
>
> While doing all this, the system still defaulted to boot in console mode
> (meaning multi-user.target and not graphical.target). To test X, I just
> launched WindowMaker using startx.
>
> Anyway, I rebooted, and then... nothing. System refused even to startup,
> and systemd suggested to me that I open journalctl (which taught me
> nothing).
I'm a tad confused: you say "nothing", and that system refused to
startup,
but systemd is talking. Where is it in the bootup? What happens if you
edit the grub kernel line one time, to s, so that it boots to single user
mode?
<snip>
       mark

On Tue, Mar 03, 2015 at 04:41:01PM +0100, info at microlinux.fr
wrote:
> Anyway, I rebooted, and then... nothing. System refused even to startup,
and
> systemd suggested to me that I open journalctl (which taught me nothing).
This doesn't make any sense.  You said 'nothing' but it also
suggested
you run journalctl?  So it wasn't nothing.  What exactly did it do?
Did it drop into the rescue shell?  Did it kernel panic?  Did you see
some text then the screen blanked?

Did you try older any older kernels that might have been loaded?

One of the things that the nvidia driver adds is "nouveau.modeset=0
rdblacklist=nouveau" to the kernel arguments.  Do you see them?  I
know that we saw the kernel panic when the nouveau driver was loaded
on a el6.6 system with an NVidia K620, so perhaps when you brought in
X11 you also installed the nouveau drivers?

-- 
Jonathan Billings <billings at negate.org>

Le 03/03/2015 19:25, Jonathan Billings a ?crit :
> One of the things that the nvidia driver adds is "nouveau.modeset=0
> rdblacklist=nouveau" to the kernel arguments.  Do you see them?  I
> know that we saw the kernel panic when the nouveau driver was loaded
> on a el6.6 system with an NVidia K620, so perhaps when you brought in
> X11 you also installed the nouveau drivers?
One thing I did this time was to make sure there weren't any 
incompatible drivers (like 'nouveau') installed *before* actually 
installing and configuring kmod-nvidia. Now everything works as expected.

I can't be sure 100 %, but I think this was the culprit.

Cheers,

Niki


-- 
Microlinux - Solutions informatiques 100% Linux et logiciels libres
7, place de l'?glise - 30730 Montpezat
Web  : http://www.microlinux.fr
Mail : info at microlinux.fr
T?l. : 04 66 63 10 32