XNU target config for Raspberry Pi 3 (2016)

XNU target config for Raspberry Pi 3 (2016)(opensource.apple.com)

82 points by beltex 7 years ago | 19 comments

beltex 7 years ago |

Credits for the find: https://twitter.com/s1guza/status/1071951039115464704

beltex 7 years ago | |

About the 2016 tag that was added to the title. While the header does say 2016, I think it's that the file wasn't made open source by Apple till this weekend.

Aside - explicit reference to Raspberry Pi here:

https://opensource.apple.com/source/xnu/xnu-4903.221.2/osfmk...

bpye 7 years ago |

IoT solution based on XNU? - bit outside of Apple's gamut but I could see it. iotOS

EDIT: Eh, maybe not the copyright is quite old - perhaps more likely some internal test? Run ARM64 XNU tests on Raspberry Pis maybe? They are pretty cheap hardware with lots of ports.

mrsteveman1 7 years ago | |

There was a time when it looked like they might switch their Airport and Time Capsule routers, which previously had been running VxWorks and NetBSD, to a variant of iOS. But then they shelved the entire product lines a few years later.

It did turn out to be true that Apple was moving to run iOS (and later on, even APFS) all the way down to watch hardware, and later on the HomePod which is the functional successor to the audio features of the Airport Express.

Had they continued producing routers, it's all but guaranteed that they would be running firmware similar to the HomePod at this point.

The reason they would be building XNU for the bcm2837 however is probably much more "boring", like allowing various teams inside Apple to test some low level change being planned for the kernel, perhaps in the network stack, without the added hassle and potential for leaks that would result from using some other unreleased Apple platform to do it.

emilfihlman 7 years ago | |

iOTOS would be pretty damn interesting!

qalmakka 7 years ago |

It's a pity that Apple let Open Darwin die off, it definitely had a potential as a nice BSD platform.

floatboth 7 years ago | |

What's nice about it?

FreeBSD is a nice BSD platform. Darwin is kind of a weird mess, with Unix awkwardly mashed up with Mach and extra C++ frameworks (IOKit etc.) on top.

tsmarsh 7 years ago | |

I'm still not sure of the history there. I remember one of the selling points of the early OS X was that it was a "Certified UNIX" and I remember that no one I tried to sell one too cared.

Was it just the cost of maintaining the certification? The fact that no other OS went down that route?

4ad 7 years ago | | |

MacOS still is certified UNIX: https://www.opengroup.org/openbrand/register.

naikrovek 7 years ago | |

You can still build it, I believe, they just don't provide ready to go .iso files. I think.

saagarjha 7 years ago |

This file is from 2016, apparently, so Apple has likely been using this for quite some time. It's possible that Apple found it convenient to run certain things (guesses: some IoT thing or prototyping a small embedded processor like the one inside the Lightning Digital AV Adapter) on a Raspberry Pi rather than a full-blown A-series core or other chip that is inconvenient due to {price, availability, etc.}.

bigpicture 7 years ago |

Not being familiar with the term XNU, I did a little clicking around and hopefully I understand a bit better:

This file is evidence that Apple has been running iOS/MacOS on the Raspberry Pi 3?

Wowfunhappy 7 years ago | |

It's evidence they have been running the macOS / iOS kernel, yes.

Whether they had the full stack running—and how much of the stack is necessary for something to be deemed "macOS"—is up for debate.

lukeh 7 years ago | | |

Or maybe it was an intern project.

floatboth 7 years ago |

Why would they even remotely touch that horrendous SoC?

JasonFruit 7 years ago | |

Honest question — why is it horrendous? It's cheap, reasonably capable, can be easily coaxed into doing most of what a person might want, and has enough adoption that advice and code is easy to find. I'm sure you mean horrendous in some more technical sense, but I've found it nothing if not practical. Please teach me.

floatboth 7 years ago | | |

- I/O is comically limited. Everything talking to the outside world is infamously bottlenecked on USB 2.0. (In comparison, any modern Rockchip or Allwinner SoC has Gigabit Ethernet directly connected to the SoC.)

- The memory subsystem is also a joke. 1 GB of slow LPDDR2 is the limit!?

- The boot process is absurd. A proprietary blob is loaded onto the GPU (!) from the SD card before everything else. An independent FOSS replacement boot stack project was started but not finished.

- And the firmware's support for network booting is very, very unreliable.

- There is a custom interrupt controller instead of the standard ARM GIC. (This was in fact one of the frustrations cited in the post about the open boot stack project's abandonment.)

- CPU crypto extensions (AES+PMULL/SHA1/SHA2) are absent. Other similar (4-core A53) SoCs (A64, RK3328) do have it.

- The GPU does not have memory protection. The VC4 driver does its best to try to validate shaders, but… LOL.

And here we're talking about Apple. They make their own SoCs. If they decided to get into IoT, they'd probably make a development platform based on one of their small chips they use in AirPods or whatever