AMDGPU Patches Under Discussion For Better External GPU Hot Unplug Handling
While Radeon graphics cards can work with various external GPU (eGPU) solutions, currently on Linux if trying to hot unplug such a setup can lead to various problems. An experimental patch series out this weekend is seeking to address that problem.
Andrey Grodzovsky of AMD sent out a patch series in trying to address the issue of when hot unplugging a graphics card (namely through eGPU solutions or also through possible sysfs interfaces) that it would cause "random crashes in user apps."
These random crashes for applications during hot unplugs of Radeon graphics cards stem from the buffer object handling in issues when the backing device disappears.
This RFC patch series aims to address the problem for the AMDGPU kernel driver. There still are some open design questions for more robust handling, but at least so far these patches would be a step forward for Radeon Linux users who rely upon an external GPU configuration and in fact hot unplug their graphics card. We'll see where this work leads but for now is just floating around the dri-devel mailing list for discussion.
Andrey Grodzovsky of AMD sent out a patch series in trying to address the issue of when hot unplugging a graphics card (namely through eGPU solutions or also through possible sysfs interfaces) that it would cause "random crashes in user apps."
These random crashes for applications during hot unplugs of Radeon graphics cards stem from the buffer object handling in issues when the backing device disappears.
This RFC patch series aims to address the problem for the AMDGPU kernel driver. There still are some open design questions for more robust handling, but at least so far these patches would be a step forward for Radeon Linux users who rely upon an external GPU configuration and in fact hot unplug their graphics card. We'll see where this work leads but for now is just floating around the dri-devel mailing list for discussion.
22 Comments