PC Drivers & Software

[sikhness]

Hi all,

I've had quite a bit of a journey when it comes to using AMD's latest drivers so let me try to summarize it and then quickly describe the latest problem that I'm having.  I'm using the AMD Ryzen 7 7840HS processor which has the Radeon 780M built in, and want to leverage the latest drivers for it of course.  I've been stuck on using drivers 24.3.1 (notice the year), with the reason being that any driver update since has broken 1 of 2 problems that I've seen.  The first issue being that AMF was broken when used within GPU-P via Hyper-V and the second being that many of the latest drivers just wouldn't install without the computer going into a boot loop if Hyper-V was enabled.  Both of those issues are detailed at length in this post if you are interested.

However, I'm happy to say at least those 2 issues seem to have been resolved with AMD's latest 25.3.2 (and 25.3.1) drivers (upon quick testing at least as I didn't get much time to spend on it).  Instead now, I'm seeing a breaking issue with Windows based process-isolated containers.  It's simply broken and somehow it seems related to how drivers may be shared with the container runtime, because when I remove the latest drivers (driver-less) or even go back to 24.3.1, all of the above works as expected.  Let me describe the issue further.

With the latest drivers installed (either 25.3.2 or 25.3.1), I'm able to start any Windows based process-isolated container as normal, however when I stop the container I get the following error in my Event Log:

error unmounting container [container=a695c3c4fd0ea4c4a6a89c7532d6065e652bc284a505d6a4a180bf1a0c858d74 error=hcsshim::UnprepareLayer failed in Win32: Do not detach the filter from the volume at this time. (0x801f0010)]

If I then try to start that same "stopped" container, I then get the following 2 errors every time:

a695c3c4fd0ea4c4a6a89c7532d6065e652bc284a505d6a4a180bf1a0c858d74 cleanup: failed to delete container from containerd: container "a695c3c4fd0ea4c4a6a89c7532d6065e652bc284a505d6a4a180bf1a0c858d74" not found

Handler for POST /v1.48/containers/test/start returned error: hcsshim::ActivateLayer failed in Win32: The process cannot access the file because it is being used by another process. (0x20)

It seems that Windows based docker containers aren't able to cleanly shutdown and remain in a weird "used" state, which effectively breaks any useful docker functionality such as building containers, starting them, and etc. as they all throw those errors above.

Any idea how this can be fixed?  As I had mentioned earlier, it is definitely related to the latest AMD drivers as if I try the same test using no graphics drivers installed or reverting back to 24.3.1, Windows based docker runs as expected without those errors showing up in the Event Log.  I'm hoping this is a known issue and will be fixed as the previous 2 errors were fixed (eventually).

Thanks so much!