Linux Developers Talk Again About An Accelerator Subsystem - Or Moving Them Into The GPU/DRM Area

On and off for years has been talk of an accelerator subsystem for the Linux kernel considering that for now most AI training/inference accelerator drivers end up lodged within the "char/misc" area of the kernel. That accelerator subsystem discussion has been restarted with talks of having such a subsystem or moving those drivers within the GPU/DRM subsystem space.

Stemming from the recent controversy around Habana Labs AI driver code changes (though now at least partially addressed with publishing an open-source compiler and user-space library) and how some changes have been skirted by going through "char/misc" and facing less scrutiny than if the same changes were tried to push through the GPU/DRM tree, the discussion is once again happening about where these accelerator drivers should live within the kernel.

Direct Rendering Manager drivers are interested in even moving the accelerator drivers within drivers/gpu given how similar they tend to be with non-display GPU drivers and sharing many interfaces and having common objectives. Living within the hypothetical drivers/accel was also brought up again with DRM maintainers still interested in overseeing then that new area of the kernel.

Greg KH who oversees char/misc welcomes any new reviewers and help managing the code though has questioned the drivers/gpu housing proposal and more firmly figuring out some of the new accelerator driver requirements for upstreaming.

Nothing firmly has been decided yet on an accelerator subsystem or if the GPU/DRM maintainers will oversee such accelerator drivers moving forward. Those interested in the latest discussion can see this kernel thread.