View Issue Details

IDProjectCategoryView StatusLast Update
0015448CentOS-7mesapublic2018-11-07 14:22
Status newResolutionopen 
PlatformOSCentOSOS Version7
Product Version7.5.1804 
Target VersionFixed in Version 
Summary0015448: Mesa GL 17.2.3-8.20171019.el7 breaks remote GLX for clients with NVidia drivers
DescriptionCentOS remote hosts running mesa-libEGL, mesa-libGL, mesa-libgbm, mesa-libglapi 17.2.3-8.20171019.el7 with libdrm 2.4.83-2.el7 do not work with GLX applications over SSH with X-forwarding if connecting from local hosts with NVidia drivers installed, even when using indirect rendering. The error message received from e.g. glxgears over SSH from a connecting local machine with NVidia driver installed is:

libGL error: No matching fbConfigs or visuals found
libGL error: failed to load driver: swrast
Error: couldn't get an RGB, Double-buffered visual

"Error: couldn't get an RGB, Double-buffered visual" seems to be most relevant as swrast errors can appear on working hosts as well. The problem is fixed by either:
- Downgrading libdrm to 2.4.74-1.el7 and mesa-libEGL, mesa-libGL, mesa-libgbm, mesa-libglapi to 17.0.1-6.20170307.el7
- Replacing NVidia drivers with Nouveau drivers on the local machine (if possible)
Steps To ReproduceEither:
- Install CentOS 7.5.1804 with mesa-ligGL on a new machine (several server and workstation hardware platforms tested with same results)
- Connect via SSH with X-forwarding to the machine from a local machine with NVidia driver 390.87 (various NVidia cards, Linux Distros and NVidia drivers from repos or from NVidia installer script tested with same results)
- Try to launch glxgears

- Install an older release such as 7.2.1511 on a machine without NVidia card (remote GLX still works)
- Update mesa-libGL, mesa-libEGL, mesa-libgbm, mesa-libglapi to 17.0.1-6.20170307.el7 and libdrm to 2.4.74-1.el7 (remote GLX now broken)
- Connect via SSH with X-forwarding to the machine from a local machine with NVidia driver 390.87
- Try to launch glxgears
TagsNo tags attached.




2018-11-07 14:16

manager   ~0033086

If you are installing the proprietary nvidia drivers via their .run file then they silently overwrite various files that are owned by CentOS packages. This will break things. If you rpm -Va then you can get a list of what they broke and reinstall those packages to re-overwrite the files with the correct thing. Or you can use the ELRepo copy of the nvidia drivers which doesn't do this in the first place.


2018-11-07 14:22

reporter   ~0033087

Thanks, but this problem lies with the listed packages on the remote machines where no NVidia hardware or drivers are installed. I.e., a single local NVidia machine can run GLX applications on remote CentOS 7 hosts that don't have NVidia drivers only if the Mesa GL packages on the remote host are older than 17.2.3-8.20171019.el7 (and libdrm is older than 2.4.83-2.el7)

Issue History

Date Modified Username Field Change
2018-11-07 13:43 m_dev34 New Issue
2018-11-07 14:16 TrevorH Note Added: 0033086
2018-11-07 14:22 m_dev34 Note Added: 0033087