LXD/GPU Acceleration

   Warning

Use LXD/GPU Acceleration (NVIDIA) to setup LXD with NVIDIA graphics. Soon, I will drop references to NVIDIA in these docs and these will be Open Source Graphics-specific instructions.

Setting up GPU-accelerated containers under Funtoo Linux is quite easy. In this example, we will set up a multilib Gentoo container with the ability to access the host GPU.

   Important

Before starting, make sure you have followed Basic Setup on Funtoo steps on the main LXD page.

Also be sure you are using at least Funtoo Linux 1.4. Funtoo Linux 1.4 and above contain several ebuilds that have been significantly modified to make them container-friendly.

Container Launch

First, let's launch a Gentoo container. By default, LXD configures the remote images: to point to https://images.linuxcontainers.org, and this site hosts Gentoo Linux images. Let's create one as follows:

root # lxc launch images:gentoo gentoo-gpu-test
Creating gentoo-gpu-test
Starting gentoo-gpu-test

Now, let's enter the container and perform an emerge --sync. We will then configure our container to use Intel GPU acceleration, which is what we are using on our host, and we will emerge xorg-x11:

root # lxc exec gentoo-gpu-test -- su --login
gentoo-gpu-test # emerge --sync
gentoo-gpu-test # echo 'VIDEO_CARDS="intel i915 i965"' >> /etc/portage/make.conf
gentoo-gpu-test # emerge -auDN @world xorg-x11 --jobs

Once this completes, we are close to having GPU acceleration but need to perform a few more steps.

Video Card Specifics

If you are using an NVIDIA graphics card, you will want to set VIDEO_CARDS to nvidia above. You will also need to allow all local users to access your GPU, not just the video user, by setting NVreg_DeviceFileMode=0666 in /etc/modprobe.d/nvidia.conf and rebooting. Note that in the future, there will likely be a better solution to this but for now, the host's device nodes are mapped into the container which runs using very high UIDs/GIDs for the root user and this user needs to be able to access your GPU's device nodes.

Host Setup

On your regular system (which we will refer to as your "host", since it "hosts" the containers), you will want to inspect the setting of your DISPLAY environment variable:

user $ echo $DISPLAY
:1

We will want to set the container to use this value. We will also want to add a GPU to our container configuration. Run the following command on the host to edit the container's configuration.

root # lxc config edit gentoo-gpu-test

You will be presented with a YAML file which contains the configuration for the container. You will want to modify the container configuration to add the environment.DISPLAY setting, as well as copy the devices: section in the container configuration snippet below. Note that you should keep all the existing configuration that is in your YAML, and do not wipe your config and paste what is below -- I'm including just the relevant parts to keep things easy-to-read:

   gentoo-gpu-test configuration

architecture: x86_64
config:
  environment.DISPLAY: :1
  image.architecture: amd64
  image.description: Gentoo current amd64 (20190717_17:39)
  image.os: Gentoo
devices:
  X11-unix:
    path: /tmp/.X11-unix
    source: /tmp/.X11-unix
    type: disk
  mygpu:
    type: gpu
ephemeral: false

Now your container is configured to use the host's GPU! All you need to do is run the following command as your regular user on your regular system to allow X connections from the container to your host:

user $ xhost +local:

This xhost command will only be active until your system reboots, and will allow local connections to talk to your X server.

Testing OpenGL

You can now proceed to test use of OpenGL in the container:

root # lxc exec gentoo-gpu-test -- su --login
gentoo-gpu-test # emerge -av --jobs mesa-progs
gentoo-gpu-test # echo $DISPLAY
:1
gentoo-gpu-test # glxgears

You should be presented with a glxgears window, running within your container, but being displayed on your X server. Congratulations -- you now have accelerated OpenGL within a container!