The 2024 State Of Open-Source AMD Firmware

[phoronix]

Phoronix: The 2024 State Of Open-Source AMD Firmware

Last year at FOSDEM 2023 there was a presentation on the state of AMD open-source firmware and since then a lot has changed from the AMD openSIL announcement to new platforms being in the process of being enabled. At FOSDEM 2024 this past weekend in Brussels was a fresh look at the current state of AMD open-source firmware...

https://www.phoronix.com/news/2024-AMD-OSS-Firmware-State

[Kjell]

My wallet is ready for modern motherboards with Coreboot + OpenSIL

[stormcrow]

I realize that AMD likely doesn't want to give away hints about what its future processors will be like 2 generations from now, but developing code essential to the functionality of all future AMD based systems behind closed doors is going to result in mistakes that wouldn't likely have withstood wider scrutiny earlier in the process. It's possible there could even be problems that are unfixable because they implicate hardware design choices in the upcoming physical hardware.

It would be nice, perhaps even prudent, for AMD to solicit feedback through published snapshots of their work on read-only repositories as they reach milestones so more eyes than their own are reading through the work to bring their own supplemental or adversarial perspectives to the code base. It may result in a higher quality foundation for their customers once they reach the point of active production rather than perpetuating the demonstrably broken cycle of broken initial release followed by round after round of fixes, or in some particularly egregious cases, a partially or completely & permanently broken product (ex: Spectre, TERRA, GSM, etc.)

We're long past the point where products developed behind closed doors are considered insecure by default. The Internet in general is littered with news reports of proprietary or even products first developed behind closed doors then publicly released are discovered to be insecure beyond repair. Legal threats won't stop bad actors from successfully attacking buggy hardware and software. It only stops good faith actors from discovering those problems before the bad actors do. The bad actors are already of a criminal mindset and [foreign] governments are beyond traditional legal repercussions to begin with. Removing the low hanging fruit and even part of the mid level stuff will make it considerably harder for attackers.

[skeevy420]

I realize that AMD likely doesn't want to give away hints about what its future processors will be like 2 generations from now, but developing code essential to the functionality of all future AMD based systems behind closed doors is going to result in mistakes that wouldn't likely have withstood wider scrutiny earlier in the process. It's possible there could even be problems that are unfixable because they implicate hardware design choices in the upcoming physical hardware.

I'd rather have open source firmware for existing products. While future products would be nice, I'd be happy if they were able to have open source code available for released products. Yeah, there's Day 1 Implications, but at least the products would be fully open and they'd be able to be fixed by anyone.

I know that should be a why not both situation, but I get why AMD, or anyone else, would want to do some work behind closed doors and not go public with something until they're ready. As is, AMD is really having to play catch-up with their GPUs so they're really doing their damnedest to walk the tightrope between open source and out-innovating NVIDIA like they were able to do to Intel CPUs. NVIDIA is looking for AMD to slip with the tiniest of leak so they can counter it and they probably have some crazy shit up their sleeves. Intel, too, but AMD straight up caught them with their pants down wanking it and have really been taking advantage of it. AMD giving their competitors a 1 to 3 year lead would basically be them committing corporate suicide.

[avph]

I realize that AMD likely doesn't want to give away hints about what its future processors will be like 2 generations from now, but developing code essential to the functionality of all future AMD based systems behind closed doors is going to result in mistakes that wouldn't likely have withstood wider scrutiny earlier in the process. It's possible there could even be problems that are unfixable because they implicate hardware design choices in the upcoming physical hardware.

Usually partners have NDA access before things are publicly released. That's how you get working mainboards on release date. Embracing a full open source model where everything is developed in the open would of course be better (just like for instance how the kernel/mesa is getting in shape for GPUs before release). It takes time for companies to understand and switch to an open source development model: closed firmware has always been the way. Some people at AMD are doing an awesome job on making this easier and better. Convincing marketing, management, lawyers, ... in such big companies that open source code and publishing datasheets is a good for business is not an easy task! So I'd say, cuddos to those making opensil, however imperfect, happen at AMD! It's going in the right direction.

[shmerl]

Am I missing something? How is it open if it's developed in a private repository? Or they periodically publish that?

[skeevy420]

Am I missing something? How is it open if it's developed in a private repository? Or they periodically publish that?

It varies by the product with AMD so some things are developed fully open while others are done behind closed doors with periodic releases. Like, Mesa and the kernel's AMDGPU code is rather open aside from products that aren't released while things like AMDVLK and FSR are done mostly behind closed doors with occasional code dumps.

[jonix]

Am I missing something? How is it open if it's developed in a private repository? Or they periodically publish that?

https://github.com/openSIL the repositories are open, but the development model is closed. From time to time they post code there, and the last update seems to be from last month. Anyway, not the best thing in the world, but it works for this initial bring up at least.

[varikonniemi]

What exactly prevented them from transitioning to this yesterday instead of in 2 years? If they released the specifications for their proprietary firmware the day before yesterday, there would exists open osurce implementation already that they could use.

Intellectual property is not an answer, because it concerns the implementation, not the specification.

[haytham.med]

Can you please provide contrast and saturation control via an app ?