Kernel module nvidia_uvm fails to load with following error: nvidia_uvm: Unknown symbol radix_tree_preloads (err -2) nVidia announcement: "Due to an incompatibility issue, we advise customers to defer updating to Linux Kernel 5.9+ until mid-November when an NVIDIA Linux GPU driver update with Kernel 5.9+ support is expected to be available. Linux Kernel 5.9+ is incompatible with current and previous NVIDIA Linux GPU drivers." Reproducible: Always Steps to Reproduce: 1. Install kernel 5.9 2. Build nvidia modules 3. Load nvidia modules Actual Results: Module nvidia loads, nvidia_uvm fails to load. As a result CUDA will not execute. Expected Results: All modules load and are functional.
More precisely this bug applies to >=nvidia-drivers-450.80.02[uvm] May be good to keep the bug open for reference until nvidia fix it but don't think much can done here, or at most "maybe" a ebuild warning when USE=uvm + kernel >=5.9. It's possible to workaround by patching the kernel (see bug #747028 comment #8), not that a patch to circumvent license restrictions is something gentoo should add, but users are free to do what they want. Otherwise, if you need USE=uvm / cuda / opencl / nvenc / etc..., do as nvidia advise. For common usage (games, vdpau) then 5.9 is known to be fine.
All in-tree modules now have full 5.10 support and I don't believe this is relevant anymore.