Details On The PS4's Radeon GPU With Linux Driver Modifications

[PhilPotter]

This is seriously cool. My hat goes off to him and his colleagues for this amazing bit of work.

[shmerl]

Why sad? The only reason to have that firmware is various DRM stuff (as in bad DRM), which most Linux users tend to avoid anyway. So having an open and DRM-free alternative to firmware blob is always welcome.

[libv]

Yes, he is mostly talking about the CP (and attached cores) firmwares and not about most of the things that atomBIOS covers. But... AtomBIOS covers the less contentious stuff, the things the IP company ATI was less afraid to release. And when RadeonHD was done for, the (sanitised) register level information stopped with it, and so did the proper driver with only very very few magical values. The R600 3d engine had only just been brought up enough to render some triangles when RadeonHD was killed off, so we never even got to the point where we could look further. And to be absolutely honest, it might be that the firmwares for those different engines are mostly about management stuff, and are in reality, not that contentious either. But all of that is academic, thanks to several individuals who these forums here seem to revere.

Honestly though, if we at SuSE/RadeonHD had not been shot down, by ATI, some guys from redhat and a few others, and if we had been able to continue, there would've been:
a) more complete documentation, not just the shader ISA.
b) a clean, sane (set of) driver(s) which isn't trying to mimic the fglrx driver even further (my reason number 2 against atomBIOS: we'd need to be bug compatible with the windows driver, now guess what), so no insane abstractions, little magic numbers, and no dumb working around firmware bugs.
c) native code for coreboot and others.
d) no discussion about holding a vulkan driver back.
e) a much easier time for Marcan and his buddies.

But that is not what the likes of John Bridgman, and his suitors Alex Deucher, Dave Airlie, Daniel Stone, Adam Jackson, Matthew Garrett, and quite a few others, lied, stole and double-crossed for. I find it amazing that Dave Airlie is pulling his stunts these days, he doesn't shy back from claiming victory over situations he created himself, and he's just showing that he is all about powerplay and politics. He's like Nixon and Vietnam, and the whole ATI story is something worthy of "house of cards".

In between changing diapers and sleeping in shifts, i am slowly getting all my data on live storage, i will soon be able to paint a very concise picture of that RadeonHD story, and there will be a set of big multi post blog entries describing what was going on then. It will be quite an eye-opening spectacle.

Despite the above, that was an AWESOME talk. And i was grinning my head off at some of the hacks employed. I just wish that half that talk was not about my failure to have stopped the flood almost a decade ago. But well done Marcan, and keep up the good work and keep kicking ATI and Sony ass.

I wonder who the unhappy ATI employee is though.

[davidbepo]

Yes, he is mostly talking about the CP (and attached cores) firmwares and not about most of the things that atomBIOS covers. But... AtomBIOS covers the less contentious stuff, the things the IP company ATI was less afraid to release. And when RadeonHD was done for, the (sanitised) register level information stopped with it, and so did the proper driver with only very very few magical values. The R600 3d engine had only just been brought up enough to render some triangles when RadeonHD was killed off, so we never even got to the point where we could look further. And to be absolutely honest, it might be that the firmwares for those different engines are mostly about management stuff, and are in reality, not that contentious either. But all of that is academic, thanks to several individuals who these forums here seem to revere.

Honestly though, if we at SuSE/RadeonHD had not been shot down, by ATI, some guys from redhat and a few others, and if we had been able to continue, there would've been:
a) more complete documentation, not just the shader ISA.
b) a clean, sane (set of) driver(s) which isn't trying to mimic the fglrx driver even further (my reason number 2 against atomBIOS: we'd need to be bug compatible with the windows driver, now guess what), so no insane abstractions, little magic numbers, and no dumb working around firmware bugs.
c) native code for coreboot and others.
d) no discussion about holding a vulkan driver back.
e) a much easier time for Marcan and his buddies.

But that is not what the likes of John Bridgman, and his suitors Alex Deucher, Dave Airlie, Daniel Stone, Adam Jackson, Matthew Garrett, and quite a few others, lied, stole and double-crossed for. I find it amazing that Dave Airlie is pulling his stunts these days, he doesn't shy back from claiming victory over situations he created himself, and he's just showing that he is all about powerplay and politics. He's like Nixon and Vietnam, and the whole ATI story is something worthy of "house of cards".

In between changing diapers and sleeping in shifts, i am slowly getting all my data on live storage, i will soon be able to paint a very concise picture of that RadeonHD story, and there will be a set of big multi post blog entries describing what was going on then. It will be quite an eye-opening spectacle.

Despite the above, that was an AWESOME talk. And i was grinning my head off at some of the hacks employed. I just wish that half that talk was not about my failure to have stopped the flood almost a decade ago. But well done Marcan, and keep up the good work and keep kicking ATI and Sony ass.

I wonder who the unhappy ATI employee is though.

i dont think we should treat AMD as evil only for this. just compare it with nvidia.
also i dont know why ALL companies close the firmware like if it was opened they all die

[libv]

i dont think we should treat AMD as evil only for this. just compare it with nvidia.
also i dont know why ALL companies close the firmware like if it was opened they all die

I did not mention AMD, i mentioned what is left of AMD which i referred to as ATI.

I do treat quite some parts of ATI as evil, due to the fun and games they played (with mr bridgman at the crux of it), to kill a proper driver and free documentation, this against the will of their then new mother company. This company lost a lot of internal power and a lot of the value they had garnered with the AMD hammer, and ended up being worth less than what they paid for ATI a year after having taken over ATI. ATI was a mess that was bankrupt when AMD came along, and they both financially and morally bankrupted AMD.

[Hi-Angel]

This is a disassembler for the Radeon "microcode" processor, which seems to be internally called F32

Would be cool if they made it instead as a part of Radare2 (the FOSS alternative to IDA Pro) instead of drawing it to a separate project.

[Hi-Angel]

libv what do you do now? What happened is a story both interesting and very bad, but not surprising at all. Based on my observations doing bad things is inherent to peoples. It's not necessarily bad — one can pick up an activity they like, and get an opponent to work against for free, which adds up some nice excitement. Depending on what you do for yourself and versus the opponent, the excitement might be very high.

So what do you do? Last Luc Verhaegen's patch I managed to find through the patchwork site is from 2011 on Xorg. Activity for the year neither on github nor in the blog. Did you give up? Do you want to let your opponents — whoever they are — break you?

Stop focusing on the past. It's an interesting story, which you may tell someone over a tea, or may be to even use at some point — but it's not your present. Stop focusing on the opponents. They might be very strong, and often your strikes would not hit — but you should only note how funny it is, that opponents turned out to be wise enough to evade. Focus on your activity, focus on what you like to do, because it's the only thing that matters in the world. And play against the opponents just as a matter of entertainment. Because it's your entertainment, not their — if you lose, you lost some jollity; if they lose, it might be efforts, money, expectations, friends, or clients, or may be even job or life.

Sometimes you need to be subjective, just for you. Just keep in mind the objective picture, for you.

[juno]

I loled at the >90k documentation file for bonaire. Hope that this doesn't harm anyone now, though.

uCode disassembler is focused/limited on f32 (me/mec), so I don't think we'll see free in-place firmwares soon. There are more important things, anyway, imho. While it still would be nice to have, it might not be pleasing for some guys@amd though

[bobwya]

Props to Hector Martin for the amount of work he put in to get the PS4 reverse-engineered to such a degree... Especially when it is apparently a few rusty tin cans duck-traped together... Really well presented as well - just the right amount of detail. I liked the bit where he said he mainly ran Gentoo as well - nice choice :-)

I seriously wonder how much longer Sony can limp on - if they're producing such badly engineered systems...

[Adarion]

I haven't seen the full video yet (but fastforwarded through the slides (not that I'd understand that much of it anyway) but I really respect that work. And I like the Gentoo / openrc (sane) combination.
I also wonder what AMD people think of it. Of course they have a business relationship with Sony and Sony very likely asked to keep everything under NDA. Furthermore they have to obey digital restriction management and implement some of that in their products - otherwise non-nerdy persons would not understand why they can't have netflix 4K and the likes on an AMD system and would blame solely AMD for it.
But now that some independent person found ways in that system... it's not their fault and they might accept patches. (?)
As a user I wouldn't mind some more public microcodes / firmwares.

The bridgman once said that he sees all (here) but I wonder if he wants or is allowed to comment on everything.
But probably AMD people will remain rather silent now to avoid heating any flames.

I seriously wonder how much longer Sony can limp on - if they're producing such badly engineered systems...

Using Marvell chips alone seems to be a pain enough. But if you look at "badly engineered" systems, or if you look around the net, read various forums, where people post on different topics, even read reviews on marketplace websites... you'll see that there is a horrible lot of "badly engineered" products. Or products that were made in a rush, thrown on the market with no intention to support them longer than the initial ROI phase. Sadly bad products don't kill a company. If that were to be true MS should already be bankrupt, right?
And a lot of such products are "shown" at Chaos Communication Congress. (E.g. they had these doorbell/opener systems for large houses and found those would no longer be wired; but work via wireless communications, like a cellphone. That provided totally new attack vectors and silly things you could do with them, even have the doorbell dial "premium" phone numbers and create a huge telephone bill on its account; or remotely open doors and more. *eyeroll*)