#asahi-dev · 2026-05-26 19:29
StanfordAppliedCyber[m]it is power related#asahi-dev · 2026-05-26 19:27
StanfordAppliedCyber[m]is there anyone i can talk to who knows about the piogw and/or soc-dpe fabric?#asahi-dev · 2026-05-25 02:29
StanfordAppliedCyber[m]and i haven't looked at all into the kernelcache to see what this means, so all i have is the raw panic string#asahi-dev · 2026-05-25 02:29
StanfordAppliedCyber[m]the other hypothesis is that the addr values are correct and some routing thing is messed up because the target device is either power or clock gated, which would make some sense, but i'm really not sure#asahi-dev · 2026-05-25 02:05
StanfordAppliedCyber[m]https://patents.google.com/patent/US9195291 i also found this patent on the dpe fabric (i'm sure this is known)#asahi-dev · 2026-05-25 01:35
StanfordAppliedCyber[m]this is very late in iokit init btw, the actual control flow is we're in iokitbsdinit which is basically await all devices to init, after this it's mount root, execve("/sbin/launchd")#asahi-dev · 2026-05-25 01:28
StanfordAppliedCyber[m]and there's two addresses 0x211059000 and 0x211059008 that we blackhole all writes to, but i highly doubt this have to do with the current problem (idk why but if we allow the writes we get serrors)#asahi-dev · 2026-05-25 01:24
StanfordAppliedCyber[m]we intercept the uart writes because we want to use that#asahi-dev · 2026-05-25 01:24
StanfordAppliedCyber[m]we block the cpustart writes and release the core from the spin table as usual#asahi-dev · 2026-05-25 01:24
StanfordAppliedCyber[m]we basically dont touch mmio#asahi-dev · 2026-05-25 01:12
StanfordAppliedCyber[m]i think i may have met my match on this issue because i'm banging my head against a wall and not making much progress#asahi-dev · 2026-05-25 01:11
StanfordAppliedCyber[m]does anyone have an idea of 1. what exactly is going wrong here? and 2. any tips on how to debug this#asahi-dev · 2026-05-25 01:11
StanfordAppliedCyber[m]like the addrhi/addrlo is suspicious because 0x0/0x0 doesn't look valid-- seems like something is trying to read from 0x0 --> obvious fault, but idk why that looks like that#asahi-dev · 2026-05-25 01:10
StanfordAppliedCyber[m]2. ultimately the issue is something our hypervisor introduces that differs from bare metal boot, because i know this exact kernelcache works bare metal. i have no clue what this could be#asahi-dev · 2026-05-25 01:09
StanfordAppliedCyber[m]1. because this is an opaque fabric, i have no clue how to go about debugging it. i looked through irc, and it looks like there's some addresses that look like some sort of config?#asahi-dev · 2026-05-25 01:08
StanfordAppliedCyber[m]the issue is twofold#asahi-dev · 2026-05-25 01:08
StanfordAppliedCyber[m]from the ADT, this corresponds to SOC_DPE MSG trying to access PCPM APCIE_SYS_GP#asahi-dev · 2026-05-25 01:06
StanfordAppliedCyber[m] Ni6 PIOGW 0 ADDR_DECODE error (0x00000002) cmd:0x14 valid/type/tid:0x1/0x1/0x1c... (full message at <https://matrix.org/oftc/media/v1/media/download/ASuuhsuWl5UeCclRlwv4dT5PIaeCVJz54SfOYR1AVLuc8BDvYvI322ZFT-_sauWNtdotagWko5ulc9mu1EAwtI1CeepMQqqgAG1hdHJpeC5vcmcvc1VQWGhYQ3ZRUWpndU9SS2VFZ3N1allJ>)#asahi-dev · 2026-05-25 01:06
StanfordAppliedCyber[m]tl;dr i'm getting the xnu panic #asahi-dev · 2026-05-25 01:06
StanfordAppliedCyber[m]okay guys i need some help here, i've spent almost a week on this panic and haven't really made progress towards fixing it#asahi-dev · 2026-05-21 00:11
StanfordAppliedCyber[m]obviously they're incomplete and mostly contain the initialization stuff, better traces to come after we get to userspace#asahi-dev · 2026-05-21 00:11
StanfordAppliedCyber[m]i gave yuka some pcie traces earlier today#asahi-dev · 2026-05-21 00:11
StanfordAppliedCyber[m]i'm stuck on some weird pio issue right now, but if anyone wants early traces from the neo let me know#asahi-dev · 2026-05-20 06:58
StanfordAppliedCyber[m]that said i am happy rn because we're finally in bsd_init#asahi-dev · 2026-05-20 06:57
StanfordAppliedCyber[m]yeah linux can ignore them 100%, we couldn't even use them because no GL support, but it would make my life harder if i have to get that extra virtualization + emulation working#asahi-dev · 2026-05-20 04:51
StanfordAppliedCyber[m]i suppose i probably should get exclaves working eventually because apple is only going to keep adding more stuff to them and m6 or whatever they won't be easily disabled anymore#asahi-dev · 2026-05-20 04:51
StanfordAppliedCyber[m]if some macos functionality is broken, that's fine, but i hope it's not too much of it#asahi-dev · 2026-05-20 04:50
StanfordAppliedCyber[m]from AudioDMAController#asahi-dev · 2026-05-20 04:50
StanfordAppliedCyber[m]oh boy this is somewhat scary: "Exclaves are disabled, and this instance appears to need them to work"#asahi-dev · 2026-05-20 03:33
StanfordAppliedCyber[m]a tiny stage 1 loader that does nothing but read stage2 into some physical address then jump to a hardcoded address. stage2 contains all the actual stuff, including all the hypervisor code and sptm emulation.#asahi-dev · 2026-05-20 03:33
StanfordAppliedCyber[m]actually can i ask now the recommendation for replacing my horrible hacked together iteration setup? tl;dr uart is very slow, and i don't want to stream a whole 120mb kernelcache over a slow and unreliable uart. this means that the kernelcache needs to live in the 1tr image i flash. this also means regular chainload won't work. the way i fixed this locally, is to add a new section for the embedded kernelcache, then have#asahi-dev · 2026-05-20 02:56
StanfordAppliedCyber[m]but i'll push this code (it's untainted, will omit all the emulation stuff) once i make it to the login screen to get some external feedback on the code/overall structure#asahi-dev · 2026-05-20 02:55
StanfordAppliedCyber[m]i think the cleanest way to do it is just separate into sptm and nonsptm hypervisor versions#asahi-dev · 2026-05-20 02:55
StanfordAppliedCyber[m]this is going to need a lot of work to upstream though, most of the old hypervisor stuff doesn't work (especially the sysreg bits) but we can't just delete it because that's still working on m1-m3#asahi-dev · 2026-05-20 02:50
StanfordAppliedCyber[m]we are officially in bsd_init#asahi-dev · 2026-05-19 04:18
StanfordAppliedCyber[m]look at the xnu dump osfmk/kern/exclaves_boot.c line 143#asahi-dev · 2026-05-19 04:12
StanfordAppliedCyber[m]if you read through the xnu source the only thing that reads that flag are the exclave entrypoints#asahi-dev · 2026-05-19 04:12
StanfordAppliedCyber[m]you just set that to false#asahi-dev · 2026-05-19 04:12
StanfordAppliedCyber[m]the sptm init struct has a flag sk_bootstrapped#asahi-dev · 2026-05-19 04:11
StanfordAppliedCyber[m]it'll be somewhat annoying because we basically need to emulate the actual "proxy the request from xnu through sptm to secure kernel" thing but we can find a way to make that work#asahi-dev · 2026-05-19 04:10
StanfordAppliedCyber[m]i'm going to assume the runtime switch is comprehensive for now, which i think it will be, and if it's not, we have a solution for virtualizing exclaves#asahi-dev · 2026-05-19 00:34
StanfordAppliedCyber[m]oh wait i may just be able to turn them off#asahi-dev · 2026-05-18 23:29
StanfordAppliedCyber[m]that'll delay my hypervisor a bit but not too long i think, and will provide more upstream benefits#asahi-dev · 2026-05-18 23:29
StanfordAppliedCyber[m]so i need to do the whole "virtualize gl1 and gl0 as separate guests" thing#asahi-dev · 2026-05-18 23:28
StanfordAppliedCyber[m]apparently the neo does indeed have exclaves enabled#asahi-dev · 2026-05-18 21:37
StanfordAppliedCyber[m]you should probably also ask someone who knows wtf they're talking about#asahi-dev · 2026-05-18 21:36
StanfordAppliedCyber[m]should note: I'm not an experienced contributor, i havent been here two weeks#asahi-dev · 2026-05-18 21:36
StanfordAppliedCyber[m]if you can upstream the initial m1n1 bringup work that'd be very helpful and shouldn't need my hypervisor to be merged#asahi-dev · 2026-05-18 21:36
StanfordAppliedCyber[m]and then see what linux drivers still work and what don't#asahi-dev · 2026-05-18 21:34
StanfordAppliedCyber[m]i think the obvious thing is: get m1n1 booting with serial output