Ask Your Question

Revision history [back]

Is libglvnd supposed to help a multiseat radeon/nvidia setup ?

Hi,

i just migrated my multi-seat setup From Ubuntu to Fedora. This setup is made of one seat with a radeon card using open source drivers and another with a Nvidia card using 340xx Nvidia binary drivers.

On my previous Ubuntu system i had to workaround a "libGL.so" linking conflict on the radeon seat because nvidia drivers come with their own version (!= mesa) of the library that took precedence on the linking process. This workaround was to add the following on the .profile of the user of the radeon seat:

export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGL.so

On the way to make the same setup on my new Fedora 26, i discovered the libglvnd project which, from my understanding, seems to be designed to solve this library conflict issue. I hoped things to work out of the box once drivers packages installed.

Unfortunatelly, i am not able to make this to work using the xorg-x11-drv-nvidia-340xx drivers from RPM Fusion without having to make the same workaround as described above. It simply brokes the Radeon seat.

xorg-x11-drv-nvidia-340xx package from rpm-fusion comes with its own /usr/lib64/nvidia-340xx/libGL.so.1 which takes precedence to the /usr/lib64/libGL.so.1 provided by libglvnd-glx-1 (verified using ldd on a program that makes use of libGL.so.1 library).

Not to mention a similar issue with a glx Xorg module.

So here are my questions:

  • Is my understanding of the purpose of libglvnd correct ?
  • Does RPMFusion version of Nvidia drivers compatible with and ready for libglvnd ?
  • Is there something broken on my setup ?
  • If libglvnd is not about solving this issue, is there any other more elegant way to solve it ?

Thanks