Hacker News Reader

Comment by gorkish | Hacker News Reader

Apple needs to give a little here.

I think it's patently ridiculous that we have to enlist the assistance of console hackers in order to have a chance at porting the most used operating system in the entire world to run on hardware that in all likelihood already happily runs it inside of Apple's labs.

> In particular, we will be reverse engineering the Apple GPU architecture and developing an open source driver for it.

This is all a noble goal, but I can't see this as being anything other than a complete and total non-starter with zero cooperation from Apple. After the street cred of getting a marginally usable OS up and running evaporates, why continue to take the effort of great engineers and waste it developing an ecosystem that doesn't care? It makes no sense to me.

View on HN

> This is all a noble goal, but I can't see this as being anything other than a complete and total non-starter with zero cooperation from Apple.

I think marcan has proven he has the chops for pieces like that. The ps4 linux port had him reverse engineering the ISA for the custom risc cores that are some of the deepest firmware of a pretty modern AMD GPU. Like even the open source drivers just treat this stuff as blobs and moves on with their life, but he had to port a fix over and showed off more publicly than anyone else had.

This Apple stuff is for sure more work, but it all seems within his wheelhouse. Hence the patreon I think, attacking the size of the problem with some funding to make it more viable than a side project.

Also, it sounds like he just loves doing this. Everyone is treating him like a martyr but he’s doing what he’s good at and we’re all benefiting.

Apple sells hardware and now services. Software seems to be a cost center. So I doubt a third party OS like Linux will ever be a priority for them in their consumer products.

Software is an essential supportive pillar which stands between its hardware and lucrative value-added services.

Otherwise you have a shiny brick

My shiny brick is in space gray

They don't necessarily need to _support_ it, but that doesn't preclude being open enough to let others do the work.

They could simply release the specification documents that they obviously already have.

People aren't asking them to go out of their way to provide support or develop anything. Only to be open with their already existing documentation so that those open source devs that want can without doing unneeded work.

They may not want to encourage it. Being able to run linux comfortably would make users less dependent on the ecosystem

This project is obviously a farce, but I would not say Apple is against Linux insofar as supporting it on the backend. Their own backends are designed for raytracing/render farms and will probably always run macOS though.

Why is it a farce? I don't get what you mean. Hector Martin is really good.

Of course it might not actually go anywhere, it doesn't even have a commit yet, literally just a logo and webpage, but let's see.

I don’t think there is any rush at all for Apple to do anything to support Linux on macs that are only a couple of months old.

However I absolutely think they should produce a Bootcamp equivalent designed to support Linux on all Macs and iOS devices that Apple no longer supports with MacOS.

I don't think they have to do anything but publish the specs. That's not 'support' as far as I'm concerned. In fact its easy to argue that Apple will benefit from motivated top tier hacker feedback on their designs.

> This is all a noble goal, but I can't see this as being anything other than a complete and total non-starter with zero cooperation from Apple.

Why? There have been several success cases of reverse engineering in the past, for all sorts of popular and fringe hardware.

I am all about the hacker cred here, but you need only take a look at the most directly comparable project, Nouveau to get an idea of how ridiculous the thought is that greenfield reverse engineering of apple's custom GPU will result in something that is desktop-class usable and reliable. Please remember that Nouveau can't even do better than a framebuffer on anything newer than a geforce 10 series card. GPU development is just too eratic and fast paced to even try to keep up the effort. Anyone willing to devote that kind of time and expense into the project would be better off engineering a GPU from scratch.

The reason why Nouveau got stuck is because Nvidia intentionally changed firmware loading process to break Nouveau and not cooperating on providing firmware. Also part of the problem is that main Nouveau developers are from Red Hat and they are limited in terms of RE of actual Nvidia code. Unfortunately some of Nouveau contributors dropped their efforts once Nvidia began to be actively hostile to the project.

If you look at GPUs like GTX 650 they actually almost have feature-parity and 90% of performance. I personally used GT710 on my PC and while it super-low-end card it's really decent under Nouveau and can drive 3 of 2K displays with 75Hz and lag-free.

As for Apple hardware it's different at least because Marcan isn't as limited in terms of keeping source code purity. After all Apple dont have any Linux drivers so there is no way to RE any code that can be directly applied to Linux graphics stack.

It would be very possible and also related to RE code from the OSX stack though.

Marcan had this topic covered on his stream. While reverse enginering is always gray area there is a lot of device drivers in Linux kernel tree that was created using RE. Also Apple never sued anyone for doing this.

Developer funded by community has more freedom than Red Hat employee. There is simply no profit for hardware vendor in attacking him.

Also OSX graphics stack is completely different from Linux one and Apple only implement their own proprietary Metal API. So even if you gonna decompile Apple drivers chances of copying Apple code are very small since the only common part would be running compilee code on the GPU.

The problem with nouveau is more that it's trying to replicate the Nvidia blob driver (and then some), in that it supports multiple disparate architectures and at different levels (three generations might be a similar uarch, for instance). Replicating a single GPU driver would be significantly easier, albeit still a Herculean task. Especially if your main goal out of the gate is simply framebuffer support and maybe some 2D acceleration.

At least, this is from my experience dealing with the Tegra X1 on a Jetson board for hobby osdev.

Sounds like a fun project, honestly. I wonder how much 2D-only “legacy” functionality even exists on the thing. I expect not much. Going for something like Vulkan right off with a mind for it probably being built to accelerate Metal might be a viable angle.

Total spitballing though. Once they come out with the 16” MBP I may have fun poking at it.

You can read about the architecture in their docs:

https://developer.apple.com/documentation/metal/gpu_features...

It's a fairly generic TBDR architecture based on it's slowly-replaced PowerVR heritage and it still has fixed function hardware in the core of the architecture.

See: the state of nvidia support on Linux today.

Several mobile GPUs have been reverse-engineered so it's not as hard as you think. As for the ecosystem, if users care enough to pay for it then it doesn't matter how little Apple cares.

Small variants/extensions of mobile GPUs have been shoehorned to work with existing drivers or provide very basic framebuffer capability. This isn't exactly vulkan running on Apple-silicon level of functionality. Maybe Apple's implementation isn't so alien after all and these great folks will pull it off. I hope they do, but I'm still tempering my expectations.

Dissecting the Apple M1 GPU, part I: https://rosenzweig.io/blog/asahi-gpu-part-1.html

They could offer an equivalent to Nvidia's PTX ISA for compute, and possibly documentation like Intel and AMD offer for their GPUs. Offcourse, the first thing the latter would do is undermine Metal, so I don't think that's likely.

Nvidia PTX-level openness for GPU-compute would be nice though.

It was not different for PC for many years. Most of drivers were reverse-engineered, including GPU ones.

That said, hardware was simpler back then. Implementing drivers for modern GPU is a hard task, may be an impossible one. It’s like writing a second OS.

Well put, it's very confusing that Apple is willing to reduce the matter of Windows on M1 to a matter of licensing, but Linux on M1 is somehow seen as competition. I was gonna buy an M1 Mac Mini to replace my aging desktop, but Apple can forget my patronage if I can't control the hardware I own.

Reply on news.ycombinator.com