amd Adieu AGESA - AMD switching to open-source CPU firmware

[BachChain]

Summary

 

During the OCP Regional Summit, AMD announced a plan to deprecate their current AGESA CPU firmware family, citing lack of flexibility and difficulty with scaling as the primary motivations. Replacing is is what AMD is calling "openSIL", a platform-agnostic open-source set of libraries. openSIL is not yet production-ready, and isn't expected to be until around 2026. Priority is being given to server platforms, but the intention is to eventually migrate AMD's entire processor stack.

 

Quotes

Quote

We had a silicon initialization product called AGESA, and the AGESA firmware was written specifically for interfacing with the UEFI firmware, so it had a lot of UEFI-isms in there and we could not scale it to other host firmwares. So when AMD got some projects to do with Chrome OS, we tried to convert AGESA, change the facets there, and try to change AGESA into Intel FSP and then integrate with CoreBoot. That was a huge Herculean effort, and definitely not scalable. So, we sustained that for a few years until we realized that we need to do something about it, because it's very resource intensive, takes a long time to convert AGESA into Intel FSP and then integrate with code.

Quote

 Today, we are at the POC phase. POR phase for the server is not until 2026. We have proven the openSIL and its feature set [is at] very close parity with AGESA firmware and we've done that with the Genoa platforms, which is epic 4 generation. We do plan to open source the Genoa POC, which is going to be soon released, and same with Gen 5. That is going to be POC as well, and 2026 is when we make it a POR for the gen 6 Epic CPUs.

Quote

We do we do plan to intercept all client and server platforms by the 2026 time-frame.

 

My thoughts

Assuming this pans out as advertised, this could make AMD king of the coreboot community.

 

Sources

https://www.phoronix.com/news/AMD-openSIL-Replace-AGESA

 

[Alex Atkin UK]

18 minutes ago, BachChain said:

Summary

 

During the OCP Regional Summit, AMD announced a plan to deprecate their current AGESA CPU firmware family, citing lack of flexibility and difficulty with scaling as the primary motivations. Replacing is is what AMD is calling "openSIL", a platform-agnostic open-source set of libraries. openSIL is not yet production-ready, and isn't expected to be until around 2026. Priority is being given to server platforms, but the intention is to eventually migrate AMD's entire processor stack.

 

Quotes

 

My thoughts

Assuming this pans out as advertised, this could make AMD king of the coreboot community.

 

Sources

https://www.phoronix.com/news/AMD-openSIL-Replace-AGESA

 

More importantly, it potentially makes it more practical to have a fully open source UEFI, fixing the garbage dump that is current motherboard firmware.

 

Granted we'd need a vendor to be on board as BIOS images are signed, but we just need one vendor to really disrupt this mess.

[williamcll]

This might mean that we might see competitors create AMD-based CPUs in the future. A rather interesting way to diversify against Intel and ARM.

[igormp]

8 minutes ago, Alex Atkin UK said:

More importantly, it potentially makes it more practical to have a fully open source UEFI, fixing the garbage dump that is current motherboard firmware.

 

Granted we'd need a vendor to be on board as BIOS images are signed, but we just need one vendor to really disrupt this mess.

This doesn't change the fact that the firmware is going to be proprietary. openSIL itself is going to be MIT, so vendors can just use it without a care.

 

The only possibility would be if those vendors base their firmware out of coreboot, which is gplv2, but I believe that vendors can work around this. 

[Alex Atkin UK]

4 minutes ago, williamcll said:

This might mean that we might see competitors create AMD-based CPUs in the future. A rather interesting way to diversify against Intel and ARM.

I don't see how opening up the CPUs firmware will allow people to clone the CPU, that will remain legally protected and is still far too cost prohibitive for any third-party to do, even ignoring the legal issues.

 

This is about having more flexibility in developing the motherboards BIOS.  We've had Open Source UEFI for a long time now, AGESA is the closed component that interfaces that to the CPU itself.  So it should open up a lot more flexibility in how motherboards interface with the CPU.  Though it also has its scary side given vendors are managing to still mess it up WITHOUT full control over that part of the BIOS.

[Alex Atkin UK]

1 minute ago, igormp said:

This doesn't change the fact that the firmware is going to be proprietary. openSIL itself is going to be MIT, so vendors can just use it without a care.

 

The only possibility would be if those vendors base their firmware out of coreboot, which is gplv2, but I believe that vendors can work around this. 

Open Source by definition means open to freely use, change and redistribute.  Its not Open Source if it has a restrictive license.

[jagdtigger]

3 minutes ago, igormp said:

The only possibility would be if those vendors base their firmware out of coreboot, which is gplv2, but I believe that vendors can work around this. 

I dont think the differing license would cause anything if coreboot loads the CPU FW as a module......

[igormp]

58 minutes ago, Alex Atkin UK said:

Open Source by definition means open to freely use, change and redistribute.  Its not Open Source if it has a restrictive license.

Well, insert the whole permissive vs copyleft thing here, but this is not relevant for this topic.

55 minutes ago, jagdtigger said:

I dont think the differing license would cause anything if coreboot loads the CPU FW as a module......

Yeah, hence why they could easily work around this. If it were GPLv3 or any other more viral license then there could be issues.

[Quackers101]

but xilinx* though, also good luck for beta/alpha testers, lol

[leadeater]

5 hours ago, Alex Atkin UK said:

Open Source by definition means open to freely use, change and redistribute.  Its not Open Source if it has a restrictive license.

Mmmm not really, it currently only means literally what the name says and nothing more. Rights to actually modify and distribute aren't inherent with something being open source, open to see etc.

 

I might make something open source so people can see it, analyze it, report bugs, contribute fixes but I may not allow usage of my thing to be incorporated in to someone else's product even if there's is also open source.

 

Then you also have open source but not free usage vs open source free usage situations, granted honesty system obviously but that is still a thing.

[Alex Atkin UK]

2 hours ago, leadeater said:

Mmmm not really, it currently only means literally what the name says and nothing more. Rights to actually modify and distribute aren't inherent with something being open source, open to see etc.

 

I might make something open source so people can see it, analyze it, report bugs, contribute fixes but I may not allow usage of my thing to be incorporated in to someone else's product even if there's is also open source.

 

Then you also have open source but not free usage vs open source free usage situations, granted honesty system obviously but that is still a thing.

https://opensource.org/osd/

 

[Doobeedoo]

Woah. How would this affect consumer platforms in terms of regular updates though.

[porina]

I'm struggling to understand what this might or might not allow in future. As far as I understand it, AGESA/openSIL acts as an interface between UEFI and silicon. If it does indeed go open, as opposed to the blob that AGESA is, does that mean there is in theory scope for someone to really break they hardware? I mean, tweak it?

 

From a mainstream consumer hardware perspective, I presume AMD will still issue openSIL blobs to the mobo vendors to integrate into their BIOS. It'll still be their choice of UEFI front end. In that sense, I don't feel anything changes at all. Do organisations outside AMD have any ability to adjust that openSIL module for AMD CPUs? For what benefit?

[jagdtigger]

9 hours ago, igormp said:

If it were GPLv3 or any other more viral license then there could be issues.

Off the top of my head you only have to make public and freely available what you changed inside the gplv3 code, it doesnt cover "external" modules thats loaded without changing any of said code.

[leadeater]

4 hours ago, Alex Atkin UK said:

https://opensource.org/osd/

 

https://opensource.org/licenses/

 

Yea no, open source is no more than access to the source code and nothing more. Idealism isn't the same thing as to what actually happens. Some allow basically complete freedom, some do not.

[leadeater]

2 hours ago, porina said:

From a mainstream consumer hardware perspective, I presume AMD will still issue openSIL blobs to the mobo vendors to integrate into their BIOS. It'll still be their choice of UEFI front end. In that sense, I don't feel anything changes at all. Do organisations outside AMD have any ability to adjust that openSIL module for AMD CPUs? For what benefit?

For the most part I doubt much will change with the process now, just the technology used being switched so I agree there.

 

2 hours ago, porina said:

If it does indeed go open, as opposed to the blob that AGESA is, does that mean there is in theory scope for someone to really break they hardware? I mean, tweak it?

Most likely yes, the 3 areas I think that will be utilized the most is high security environments who only what their specific things to work stripping out the rest, hyperscalers to optimize further, probably though not sure XOC (probably useful somehow I guess )

[igormp]

8 hours ago, porina said:

Do organisations outside AMD have any ability to adjust that openSIL module for AMD CPUs? For what benefit?

Hyperscalers that have custom servers/firmware, like Meta and Google, so they can have their own quirks added.

 

[SeriousDad69]

I'm pretty happy about this. AMD and opensource almost always go well together, like how OpenGL performance was completely busted in their crap closed source Radeon drivers for Windows for over a decade, but the opensource community on Linux fixed it for them ages ago. (I say they go well together, but it could just be that AMD just likes having their crap code fixed for them for free lol)