docker-rdesktop does not use Intel GPU

[ptr33]

docker-rdesktop does not use Intel GPU

---

Expected Behavior

If an Intel GPU is installed in the system, the xrdp server of docker-rdesktop should use it.

Current Behavior

docker-rdesktop always uses software rendering.

Steps to Reproduce

1. Add /dev/dri to the --devices section
2. Start docker-rdesktop
3. Run glxheads to see which render device is used

Environment

OS: Linux / Ubuntu 20.04 CPU architecture: x86_64 How docker service was installed: sudo apt install docker.io

use following Dockerfile command to solve the issue - the problem is access rights for /dev/dri RUN usermod -a -G render abc

[github-actions[bot]]

Thanks for opening your first issue here! Be sure to follow the bug or feature issue templates!

[thelamer]

There is no gl support at all. It's not just a matter of linking in a video device and installing drivers it is xrdp level for opengl support. Some distros have packages like Arch but for Ubuntu/alpine we would need to compile in detect and handle perms etc. In general we are kind of in a holding pattern for this implementation to improve a bit to implement this across our rdesktop base images in a universal fashion.

[ptr33]

Thank you @thelamer for the review. Nevertheless I am wondering about the comment as I have glx running inside the docker container.

I am using a container FROM ghcr.io/linuxserver/rdesktop:mate-focal and added RUN usermod -a -G render abc

In the Dockerfile I used

    devices:
      - /dev/dri:/dev/dri               # GPU support

I have a Intel Iris Plus Graphics 655 and this now works inside container

abc@f4ab7fc03f6f ~ $ glxheads :10.0
Name: :10.0
  Display:     0x55c7e6e78d80
  Window:      0x600002
  Context:     0x55c7e6eac970
  GL_VERSION:  4.6 (Compatibility Profile) Mesa 21.2.6
  GL_VENDOR:   Intel
  GL_RENDERER: Mesa Intel(R) Iris(R) Plus Graphics 655 (CFL GT3)

So I think glx is supported with this minor addition. Thank you

[thelamer]

I'm confused are you seeing the triangle ? Xrdp and vnc has conventionally required virtualgl redirection. CLI stuff will work like gpu video encoding etc, but actual graphical rendering does not due to a software framebuffer being used.

[ptr33]

Hi @thelamer ,

I have a RDP connection from windows to a linux box running the docker server and glxheads renders the triangle with GPU support:

I also see that the GPU is used as intel_gpu_top (I execute this command on the server directly not inside the docker container) shows the activity:

intel-gpu-top -  359/ 364 MHz;    0% RC6;  2.35 Watts;      761 irqs/s

      IMC reads:     1441 MiB/s
     IMC writes:     1179 MiB/s

          ENGINE      BUSY                                                                                MI_SEMA MI_WAIT
     Render/3D/0   16.63% |████████████▊                                                                |      0%      0%
       Blitter/0    0.00% |                                                                             |      0%      0%
         Video/0    0.00% |                                                                             |      0%      0%
         Video/1    0.00% |                                                                             |      0%      0%
  VideoEnhance/0    0.00% |                                                                             |      0%      0%

I also see GPU activity when I simply move the windows inside the docker container xrdp session.

Thank you

[thelamer]

I will look back into this, but it was my understanding this would not work without virtualgl. For testing can you run this on two containers at once ?

[ptr33]

Yes, I can run this in two separate containers simultaneously:

both container are derived rdesktop:mate-focal

FROM ghcr.io/linuxserver/rdesktop:mate-focal
...
RUN usermod -a -G render abc

Thanks

[thelamer]

@ptr33 no joke you just blew my mind.

https://user-images.githubusercontent.com/1852688/156831786-5d40b1e4-a9c9-4f7b-a3da-dcdd4bcce389.mp4

We use these base images for a whole world of desktop app images including webtop. I need to polish the perm automation and get drives into our baseimages for Arch, Alpine, Fedora, and Ubuntu so it may take a bit, but I will ping you when this is implemented.

[ptr33]

@thelamer thank you!

[thelamer]

This is in the latest images simply pass --device=/dev/dri:/dev/dri and it should work.

I need to reformat and retag the images in this repo to bring them in line with https://github.com/linuxserver/docker-webtop along with that will be documentation for this.

[ptr33]

Thank you @thelamer for your fast response, it perfectly works - I tested it with ghcr.io/linuxserver/rdesktop:mate-focal and adding /dev/dri to devices in Dockerfile and my Intel GPU is used inside the docker container.