More than one month ago we reported on AMDGPU patches proposed for better hot unplug handling, mainly for the use-case of external GPU solutions if disconnecting them while the system is still running to avoid a range of show-stopping problems. It's been a quiet few weeks but that work has now seen a new revision.

AMD driver developer Andrey Grodzovsky sent out the second version of this hot device unplug handling for AMDGPU but even with the improvements is still considered a proof of concept state. The hope once these patches are fully vetted is to avoid application crashes and other problems that current can happen when unplugging an "eGPU" or otherwise emulating a GPU unplug/remove event via sysfs.

This new version of the hot unplug handling follows feedback on the original patch proposal, better handling around what happens to the buffer objects when the GPU is disconnected, and other changes. But there still are some issues coming up with this code to be addressed before it could be considered for integrating into a future Linux kernel release.

Those with an external GPU setup can find the "v2" patches via this mailing list thread.