NVIDIA – An Acceptable ‘Taint’ Workaround for Kernel 5.9..
Thanks to mbod, there is a simple kernel patch to restore the Kernel 5.8 ‘inherited taint’ behaviour to Kernel 5.9, and the detail is in this thread: https://forums.developer.nvidia.com/t/opencl-not-working-with-kernel-5-9/156941/6
I have applied this to Kernel 5.9.0, and NVIDIA driver 455.28 now installs and loads/runs OK. Until a proper fix is available, this will – of course – need to be re-applied to every kernel update.
Robert Gadsdon. October 14th 2020.
Nvidia has just released a new driver 455.38, which works on kernel 5.9+.
Although dmesg still shows something like taint, but nvidia_vum module can load and cuda program works:
[ 39.066489] nvidia_uvm: module uses symbols from proprietary module nvidia, inheriting taint.
[ 39.073577] nvidia-uvm: Loaded the UVM driver, major device number 510.
Still not True that NVIDIA 455.38 is Working with CUda on 5.9.6.
I compiled the Kernel and install it, than try to install NVIDIA 455.38 with no use of –no-unified-memory but cant build nvidia-uvm module like before …
So im Sick of getting Blamed from NVIDIA because using Linux.
If they dont fix this in near Future im switching to a AMD card.
(System Faq: Ubuntu 18.04, RTX 2060 Super, using Handbreak 1.3.3 but (NvEnc) still not working.)
For now its the 5 Nov, and Nvidia Released the 455.38 Driver witch in other discusions is still working like they say.
In Fact im On Ubuntu 18.04 compiled Linux Kernel 5.9.6 and try installing Nvidia 455.38 without –no-unified-memory but still not working …
the things i Read in other Posts about this Driver for me is still not True !!!
The Driver cant compile the nvidia-uvm module, so there is no way to use Cuda in Handbreak 1.3.3 (NvEnc).
Hope Nvidia is still fixing this in the next Days cause im sick of getting no Way to use my System with my RTX 2060, i ordert new Parts to Build better System but if this is still working in the near future im switching to AMD card.
Im rid of getting Blamed from NVIDIA because usin Linux.