#asahi · 2022-01-07 05:22
marcanApple (iDevices) and Microsoft (Xbox) are, in my opinion, the only two companies that know how to do consumer product security properly#asahi · 2022-01-07 05:22
marcanisn't pluton the xbox stuff? if so hey, maybe x86 actually has a chance at decent silicon security too, no thanks to intel#asahi · 2022-01-07 05:13
marcanso even if it *did* work for fuOS, we wouldn't be able to use it without some serious craziness#asahi · 2022-01-07 05:12
marcanand all that will only work with real kexts and there's magic authentication/credential passing and it's all a giant confusing thing#asahi · 2022-01-07 05:12
marcanyou can tell because installing kexts ends up in a magic double reboot on macOS#asahi · 2022-01-07 05:12
marcanit doesn't get updated from macOS; stuff gets staged and a reboot through recovery in a special mode happens#asahi · 2022-01-07 05:12
marcanapparently there's a "kcOS" involved#asahi · 2022-01-07 05:11
marcansven: the auxKC stuff is more magic than we thought anyway#asahi · 2022-01-07 05:10
marcanI'm not sure yet, I suspect they will ever since 12.x but maybe we can find a way around that. unclear.#asahi · 2022-01-07 05:10
marcanthe remaining question is whether they will *also* require two reboots, doing step2.sh again from upgraded rOS#asahi · 2022-01-07 05:09
marcanso upgrades will require recovery mode (=1TR in the normal case anyway)#asahi · 2022-01-07 05:09
marcanbut that wouldn't work from linux anyway because we wouldn't want to implement or require the phone home stuff#asahi · 2022-01-07 05:08
marcannormal installs can happen from macOS if Apple still signs that version or from any rOS if they do not (reduced security)#asahi · 2022-01-07 05:08
marcanchanges to the custom boot object require 1TR#asahi · 2022-01-07 05:08
marcanlandscape15[m]: every change to machine software does not require 1TR#asahi · 2022-01-06 13:20
marcanthe round metal button is the ground pin on the adapter side#asahi · 2022-01-06 13:19
marcanmps: assuming europe/schuko plugs, not sure what the easiest store is for you, but this is what you want: https://www.apple.com/de/shop/product/MK122D/A/power-adapter-netzteil-verl%C3%A4ngerungskabel#asahi · 2022-01-06 13:15
marcanthey do exist#asahi · 2022-01-06 13:15
marcanmps: get a grounded plug end for the charger#asahi · 2022-01-06 13:00
marcanjannau: good idea#asahi · 2022-01-06 10:48
marcanseems to work for me with iwd and NetworkManager (you need to say wpa-psk mode in NM, not sae; sae seems to only work/be required for the wpa-supplicant backend while iwd autodetects and just wants to be told psk for both WPA2/3)#asahi · 2022-01-06 10:47
marcanfor wpa3, try this patch: https://github.com/AsahiLinux/linux/commit/4671badd2e5a89e137f054fbdd32bd9c1ef59532#asahi · 2022-01-04 12:41
marcanlet's see if I can fix that one...#asahi · 2022-01-04 12:41
marcanjust upgraded UniFi and created a WPA3 network#asahi · 2022-01-03 13:48
marcanif you see any mentions of the old confused terminology feel free to fix them#asahi · 2022-01-03 13:48
marcantl;dr the last partition on disk is *not* 1TR, it's "System RecoveryOS". 1TR just means "recovery booted by holding the power button down" and it *used* to be system recovery on 11.x but is now the paired OS recovery, and both are called 1TR when booted in this way#asahi · 2022-01-03 13:47
marcanfwiw I just updated some docs wiki pages (boot, storage, glossary) to try to fix the 1TR confusion#asahi · 2022-01-02 13:25
marcanand then if we want to slim it down in the future we certainly can#asahi · 2022-01-02 13:25
marcanso unless it turns out the PSCI code will be basically freestanding once we figure out what needs to go there (i.e. one C file), I'll probably just start by leaving all of m1n1 resident#asahi · 2022-01-02 13:24
marcanbut either way, this is all relatively easy to refactor post facto; m1n1 is small enough it's easy to do major surgery on it#asahi · 2022-01-02 13:24
marcane.g. there's a chance we end up having to talk to SPMI or something for sleep, and if it comes to that I'm not sure I want to start over without our usual scaffolding :p#asahi · 2022-01-02 13:23
marcanyeah, but it's not like there's much of an attack surface here, and m1n1 is the root of trust for our ecosystem anyway#asahi · 2022-01-02 13:23
marcandepends on how much machinery ends up going in there#asahi · 2022-01-02 13:22
marcan(e.g. leaving the UART iodev active, at least if debug was active, is quite legitimate for debugging suspend/resume and will largely play fine with the kernel driver)#asahi · 2022-01-02 13:22
marcanand we certainly could fit the PSCI code in a few kBs if we want, though that would involve forgoing all the m1n1 infra and mostly making it a separate codebase, which may or may not be what we want#asahi · 2022-01-02 13:21
marcanmakes sense, yeah#asahi · 2022-01-02 13:10
marcanthat doesn't mean m1n1 needs to stay entirely resident; I'll probably constrain the PSCI code to not use any dynamic memory, so we can throw away the heap; that would keep the memory wastage to under a couple MBs max, and we could shrink it further if we really want to with some section trickery#asahi · 2022-01-02 13:09
marcanso there has to be *something* there if nothing else#asahi · 2022-01-02 13:09
marcanit is a hard requirement of the platform as it exists today that the computer will wake up at m1n1's base address, the way I have addresses aligned right now#asahi · 2022-01-02 13:08
marcanm1n1 is actually *required* to stay in RAM for us to implement sleep, due to how the CPU RVBAR gets locked down#asahi · 2022-01-02 13:08
marcanamw: subsequent stages do not step over each other if they shouldn't; there are multiple reserved memory mechanisms depending on how you're booting#asahi · 2022-01-02 13:07
marcan*for boot#asahi · 2022-01-02 13:07
marcanbesides, "traditional" linux is at least root, boot, swap; here we'd have stub, EFI, root, maybe swap (no need for root these days) so it's not that different#asahi · 2022-01-02 13:06
marcanlandscape15[m]: GPT can handle up to 128 partitions, so I'm really not worried about typical users running out of partitions by doing multiple linux installs#asahi · 2022-01-02 11:24
marcanpushing low level stuff into m1n1 is always a win from that point of view#asahi · 2022-01-02 11:24
marcanas a bonus, this gives us even *more* chances of things like sleep modes magically working on newer Apple CPUs without any kernel changes#asahi · 2022-01-02 11:24
marcanyes#asahi · 2022-01-02 11:23
marcanbecause every other ARM platform does it via PSCI and we don't want to make a mess of things#asahi · 2022-01-02 11:21
marcanthe problem is this is normally done by a supervisor or hypervisor, but we don't have either of those on M1 (normally), so we need to invent/standardize a new method for doing it without crossing privilege levels#asahi · 2022-01-02 11:21
marcanbut unrelated, we will also have UEFI runtime services via u-boot