Arm Ethos-U Linux Driver Posted For Machine Learning Processor

Arm and NXP engineers have posted the initial open-source Linux driver patches for an Ethos-U driver for their machine learning processor to enable Linux to dispatch AI inference jobs to the hardware. It's yet another inference/accelerator driver working its way toward the mainline kernel but is off to a rocky start with many code issues being raised.

This morning saw this patch series presented to the Linux Kernel Mailing List for implementing the Ethos-U NPU driver and has been tested on the NXP i.MX93 platform.

"Ethos-U Linux driver is to provide an example of how a rich operating system like Linux can dispatch inferences to an Arm Cortex-M subsystem, consisting of an Arm Cortex-M and an Arm Ethos-U NPU."

The Ethos-U is designed for AI use-cases like sound/speech recognition, language translation, object classification, object detection, noise cancellation, and more. Due to the diversity of Arm deployments, the Ethos-U processor series aims for everything from automobiles and infrastructure to mobile devices and smart home components. Now Arm is working toward getting this Ethos-U NPU driver mainlined into the Linux kernel that comes in at just under four thousand lines.

Greg Kroah-Hartman though was quick to point out contrary to so many engineers signing-off on the driver patches, there were a number of juvenile issues quickly spotted within the code. It was also raised why this driver isn't aiming for the new accelerator "accel" subsystem rather than trying to go under the firmware drivers area of the kernel. In any case, expect more revisions of this driver ahead if Arm and NXP want to get this Ethos-U driver upstreamed.

Over on the user-space side, there is this NXP GitHub repository with the user-space software components to interact with the kernel driver.