Scripts and ports under development to support secondary Nvidia GPU (Optimus)

[Theron]

Hi All,

I have put together a preliminary release of work I have done to enable 
use of the Nvidia GPU on a dual Intel/Nvidia laptop.  My hope is that 
this may be useful to others with this system configuration, and that, 
at the appropriate time, it can be moved to the ports tree and serve as 
a base for proper Optimus support.

Sources and documentation are here: 
https://github.com/therontarigo/freebsd-gpu-headless

The work I have done is already "ported" and need only be copied into a 
working copy of the ports tree to build, but I will hold off on 
proposing the ports until I can provide some better documentation and 
receive feedback on the approaches I have taken to getting this to work.

The work consists of three parts:

Changes to x11/nvidia-driver to allow it to be used as a slave port with 
hooks for changing some of the file locations
New port x11/nvidia-headless-driver, which is just x11/nvidia-driver but 
installs without breaking an existing OpenGL implementation
New port x11/nvidia-headless-utils, which provides wrapper scripts for 
nvidia-xconfig, for xorg, and for running applications

The work builds off of the information in 
https://wiki.freebsd.org/OptimusVideoSupport

However, x11/nvidia-driver-headless is developed independently from the 
x11/nvidia-driver-optimus which has never made it into ports.  Perhaps I 
have wasted some time in doing this, although if my changes to 
x11/nvidia-driver are considered less invasive, maybe it will be easier 
to get them accepted.  The x11/nvidia-driver-headless is meant to be not 
specific to Optimus, so ideally useful anywhere Nvidia isn't the only 
OpenGL vendor present on the system.

With the new directory $PREFIX/lib/xorg/modules-nvidia-headless I have 
worked around the need to fiddle with extension files to run Xorg 
successfully (see $PREFIX/etc/X11/xorg-nvidia-headless-template.conf).

If you have the time, please take a look into this and let me know if it 
works as intended (I have tested only on 12.0-CURRENT) as well as any 
changes that might be needed before I should propose the new ports.

I am considering splitting x11/nvidia-headless-utils into yet another 
port containing only the VirtualGL bits so it can be a proper 
dependency, but involving the creation of yet another port for such a 
small part may not be the best thing.

Feedback welcomed,
Theron Tarigo