This looks like a bug in nouveau (The userspace drivers, not the kernel driver) in that it's trying to load and probably fails due to lack of a kernel module, and has some issue like a double-free in the teardown path. However, I'm not clear why it's getting even that far into init. Do you have the files /usr/lib/gbm/nvidia-drm_gbm.so and /usr/lib32/gbm/nvidia-drm_gbm.so? They should be symlinks to /usr/lib/libnvidia-allocator.so.1 and /usr/lib32/libnvidia-allocator.so.1 respectively.
After adjusting for Fedora pathing (lib64 = 64-bit, lib=32-bit), this is what I get.
matt@gaming-pc:~$ file /usr/lib64/gbm/nvidia-drm_gbm.so
/usr/lib64/gbm/nvidia-drm_gbm.so: symbolic link to ../libnvidia-allocator.so.555.42.02
matt@gaming-pc:~$ file /usr/lib/gbm/nvidia-drm_gbm.so
/usr/lib/gbm/nvidia-drm_gbm.so: cannot open `/usr/lib/gbm/nvidia-drm_gbm.so' (No such file or directory)
matt@gaming-pc:~$ file /usr/lib64/libnvidia-allocator.so.1
/usr/lib64/libnvidia-allocator.so.1: symbolic link to libnvidia-allocator.so.555.42.02
matt@gaming-pc:~$ file /usr/lib/libnvidia-allocator.so.1
/usr/lib/libnvidia-allocator.so.1: symbolic link to libnvidia-allocator.so.555.42.02
it seems like perhaps /usr/lib/gbm/nvidia-drm_gbm.so: cannot open /usr/lib/gbm/nvidia-drm_gbm.so (No such file or directory) is my issue?
I double checked that my PC has no sign of touching nouveau which is especially confusing:
matt@gaming-pc:~$ lsmod | grep nouveau
matt@gaming-pc:~$ sudo dmesg | grep nouveau
[sudo] password for matt:
matt@gaming-pc:~$ sudo journalctl -b | grep nouveau
May 24 09:52:36 gaming-pc xwaylandvideobridge[2911]: DRM kernel driver 'nvidia-drm' in use. NVK requires nouveau.
May 24 09:52:44 gaming-pc plasmashell[3650]: DRM kernel driver 'nvidia-drm' in use. NVK requires nouveau.
I'm testing out Fedora Rawhide on a separate rig that is able to boot embedded gamescope w/ steam, however it crashes when trying to launch games in a similar manner to how my normal Fedora 39 rig crashes before displaying the home screen within embedded gamescope/steam. I will grab logs of that from my NVIDIA laptop shortly and compare them to the logs I've provided so far.
If you manually create that missing 32-bit symlink, pointing to the 32-bit libnvidia-allocator.so.1 file, does it fix things?
That does seem to fix things! I bet I was running into issues with Steam because of the 32-bit portions of it. I think my only issue remaining is now solely within Gamescope so I'm closing this. Thank you so much for the assistance!!
@cubanismo sorry for the additional ping but I've confirmed the official .run file from Nvidia also does not create this symlink properly
i can also confirm this is true
Thanks, yes, I'm aware. This was done on purpose, as I was hoping no one sane would ship a 32-bit app these days, and in prior days, our driver didn't support GBM anyway. Browsers, especially embedded variants like Steam uses, were actually identified as an outlier use case at that point, but they were expected to fall back gracefully, but they can't be expected to work around bugs in NVK failure paths. So much for that. I had already filed a bug to add the symlink in the driver package and of course, the additional test coverage that entails to make sure it remains functional.
Nothing extracted yet.
Your system information
Please describe your issue in as much detail as possible:
While using the new beta NVIDIA 555.42.02 driver, it seems like Steam crashes while trying to use gamescope in embedded mode with the command

gamescope -e -- steamand other variations of the same command. Steam will launch for approximately 5-10 seconds, sometimes even going as far as the boot animation, before it coredumps. This subsequently leads to a gamescope coredump as well. You can see this cycle incoredumpctlhere:Running
coredumpctl infoon one of these coredumps for Steam provides this stack trace:I doubled checked and Nouveau is definitely not being used at all according to my dmesg and journal, along with
lsmod | grep nouveauI tried going one step further of running a backtrace of that in
gdbbut it appears the core is incomplete or corrupted, probably because of Steam's install method (an unofficial RPM package based off the debian version from https://repo.steampowered.com/steam/).Steps for reproducing this issue:
gamescope::ConVar<bool> cv_drm_debug_disable_in_fence_fd( "drm_debug_disable_in_fence_fd", false, "Force disable IN_FENCE_FD being set to avoid over-synchronization on the DRM backend." );into ->gamescope::ConVar<bool> cv_drm_debug_disable_in_fence_fd( "drm_debug_disable_in_fence_fd", true, "Force disable IN_FENCE_FD being set to avoid over-synchronization on the DRM backend." );indrm.cppbefore compilinggamescope -e -- steamThis may very well just be a NVIDIA issue, so I will try making a post on their forums as well. I figured I'd post my logs here anyway considering the steam coredumping in embedded mode/gamescope-session seems to be reproable by someone with a similar setup (2000-series mobile GPU/Arch Linux).