nVidia drivers and xorg

Jeff Spaleta jspaleta at gmail.com
Wed May 18 16:14:35 UTC 2005


On 5/18/05, Ivan Gyurdiev <ivg2 at cornell.edu> wrote:
> I believe it writes over xorg.conf no matter what, and tries to backup
> the old one. However, SELinux doesn't like that.

There might be an selinux issue here as well...but Paul, the original
poster in this thread, is experiencing other more classic problems. I
think someone more competent with using livna's nvidia rpms should
examine the selinux issue in a seperate thread or bugreport. Paul is
seeing is xorg.conf be overwritten... selinux isnt preventing that
from happening. And I'm not getting any similar avc messages from my
synced rawhide box with regard to the xorg.conf* files at bootup when
using livna's rpms and selinux targetted.
On my rawhide system using recompiled livna nvidia srpms:
ls -Z /etc/X11/xorg.conf*
-rw-r--r--  root  root  system_u:object_r:etc_runtime_t  /etc/X11/xorg.conf
-rw-r--r--  root  root  system_u:object_r:etc_runtime_t 
/etc/X11/xorg.conf.backup-nvidia-glx

Do you perhaps need to run fixfiles relabel on your box?

But you are right.. the script logic from the livna rpms do
always make a backup and does edit the xorg.conf on every boot. 

Please bare with me a moment while I try to explain for the others in
this thread about what Paul is doing wrong.  Anyone interested in
discussing potential selinux issues with livna's rpms can just stop
reading here.

The livna nvidia rpms provide 
 /etc/rc.d/init.d/nvidia 
which calls at boot
/usr/sbin/nvidia-config-display 
which does re-write the xconfig making nv/nvidia specific changes
based whether the nvidia kernel module is available in the expected
place..everytime.

The nvidia installer doesn't provide these utilities.

Paul needs to come back and tell us if he has nvidia rpms installed. 
So I can tell Paul that he should NEVER use both nvidia rpms and
nvidia's installer.  I can tell by his selinux errors that he has used
nvidia's crackrock installer... just by looking at where selinux saw
the libs.

/usr/lib/tls/libnvidia-tls.so.1.0.7174  is from nvidia's installer and
Pauls error message
/usr/lib/nvidia/libnvidia-tls.so.1.0.7174 is from livna's nvidia rpms

I have HIGH confidence that Paul has nvidia rpms installed as well as
using nvidia's crackrock installer... his symptoms are classic of
mixing nvidia rpms with nvidia's installer.
Nvidia's installer places the kernel module in a different directory
than the livna nvidia kernel module rpms. The /etc/rc.d/init.d/nvidia 
script that the livna rpms provide looks specifically in
/lib/modules/"`uname -r`"/kernel/drivers/video/nvidia/  for the kernel
module.
Since the nvidia installer... places the module in a different
location... the livna bootscript does not see the module placed by
nvidia's installer and accordingly reverts to nv mode in the xorg.conf
file.  This is expected behavior.. the livna bootscript is there to
make sure people have a working X config if the nvidia module is not
detected. Never use nvidia's installer as well as livna's rpms and you
won't have this problem.

Users should never...ever... mix nvidia rpms with nvidia's installer.
Personally i think you should never use nvidia's installer and
especially not on a rawhide box because of the way it overwrites
library files from Mesa. The way nvidia's installer works is fragile
and if you aren't aware of whats going on you can easily misinterpret
errors as coming from xorg updates instead of the real
problem...nvidia's installer decision to conflict with files from
Mesa.  The livna rpms delibrately place the nvidia libs in its own
subdirectory and uses a file in /etc/ld.so.conf.d/ to avoid some of
the problems nvidia's installer creates on rpm based systems.




More information about the fedora-test-list mailing list