On Fedora Linux 13, whether one installs the nvidia driver through the official installer from nvidia or from the rpms provided by the rpmfusion fedora repository, the devices /dev/nvidia0 and /dev/nvidiactl are incorrectly labelled as system_u
bject_r:device_t:s0. Instead, they should be labelled as system_u
bject_r:xserver_misc_device_t:s0, as predefined in the installed selinux policy. This causes applications such as boinc based distributed computing projects to be unable to detect the nvidia graphics card and access its CUDA library.
This issue has already been reported by me and other users on rpmfusion and fedora bugzilla and no one seems to know how to create these two devices with the correct label, in order to allow read and write access to them.
Bug reports with more details: