Maintainers drop AI-written SM8150 camera patches as unreviewed duplicates
Krzysztof Kozlowski says he may ignore further submissions after an LG G8x series repeated work already on the linux-media list and failed basic style checks.
Linux kernel maintainers have rejected an eight-patch series that would have brought Qualcomm SM8150 camera hardware and three LG G8x ThinQ sensors into mainline, calling the work a duplicate of an existing submission and insufficiently reviewed AI output.
Gaurav Suman posted drivers and device-tree bindings for the SM8150 camera subsystem (Titan 175), a Samsung S5K3M3 13-megapixel wide sensor, a Samsung S5KGD1 32-megapixel front sensor, and the autofocus half of an ON Semiconductor lens controller. He said the sensors streamed raw frames at 30 frames per second on an LG G8x through libcamera, and that he had reviewed every change produced with Anthropic's Claude. The main Sony sensor and the phone's board description were left for later.
Bryan O'Donoghue, who maintains Qualcomm's camera subsystem code, told Suman to check the linux-media list before sending duplicates. An SM8150 camera series from Andrew Gigena was already under review. O'Donoghue also said the usual next step is device-tree patches for the phone itself, so reviewers can see the bindings on real hardware.
Device-tree maintainer Krzysztof Kozlowski dropped the SM8150 binding as a repeat of that existing work. On the new sensor bindings he objected to an AI-assistance paragraph left in the commit message, trivial misindentation, and descriptions that talked about the driver rather than the hardware. "I might be ignoring your contributions based on that," he wrote. He said the sensor driver did not resemble recent kernel drivers and was hard to read, and that sending code whose style is wrong "with naked eye" showed the patches had not been properly reviewed. He dismissed a later binding as more unreviewed AI output and asked whether Suman had compared his own patches with each other.
The exchange is a blunt application of an existing kernel rule: authors may use coding assistants, but they own the result, and maintainers will not spend review time on work that ignores what is already posted or fails checks a human reading of nearby drivers would have caught.