Intel Prepares Updated Linux Graphics Driver Handling For GPU Firmware

[phoronix]

Phoronix: Intel Prepares Updated Linux Graphics Driver Handling For GPU Firmware

Back in July I called attention to the issue how Linux 5.19 was set to break Alder Lake P graphics support unless moving to new graphics micro-controller "GuC" firmware in tandem. That user-space breakage is frowned upon and following that article the upstream DRM kernel maintainers outlined explicit requirements around firmware not breaking driver support. Intel engineers ended up submitting a quick fix for Linux 5.19 to still support the existing firmware while now a more adequate solution has been devised...

https://www.phoronix.com/news/Intel-i915-Versioned-Firmware

[archsway]

Why does the firmware have to be signed, making it impossible for kernel developers to modify? (At least until someone leaks the signing key…)

Is it perhaps because Intel did a bad job at hardware design and don't want you to know that the firmware could take over the rest of the system if compromised?…

On the other hand, the newest Mali GPUs make it very easy to create your own firmware images to run on the 1 GHz Cortex-M7 core that got added. Which is behind an MMU and so unable to be more evil than any other part of the GPU.

[jljusten]

Why does the firmware have to be signed, making it impossible for kernel developers to modify?

Yeah, it could be a cool open source project if not for the signing part.

Another option could be for i915 to continue to support having the CPU submit work to the GPU in the case where the guc firmware fails to load.

[andrebrait]

Why does the firmware have to be signed, making it impossible for kernel developers to modify? (At least until someone leaks the signing key…)

Is it perhaps because Intel did a bad job at hardware design and don't want you to know that the firmware could take over the rest of the system if compromised?…

On the other hand, the newest Mali GPUs make it very easy to create your own firmware images to run on the 1 GHz Cortex-M7 core that got added. Which is behind an MMU and so unable to be more evil than any other part of the GPU.

IP and stuff, locking down features, I guess?
An OSS firmware would allow you to modify it, and without signing it would be runnable.

So they would have to say goodbye to the sweet money they get from locking things down to specific SKUs and whatnot.

[binarybanana]

Why does the firmware have to be signed, making it impossible for kernel developers to modify? (At least until someone leaks the signing key…)

Is it perhaps because Intel did a bad job at hardware design and don't want you to know that the firmware could take over the rest of the system if compromised?…

On the other hand, the newest Mali GPUs make it very easy to create your own firmware images to run on the 1 GHz Cortex-M7 core that got added. Which is behind an MMU and so unable to be more evil than any other part of the GPU.

DRM bullshit is another reason.