#asahi-dev · 2026-04-21 14:19
svenId really like to figure out wtf is going there #asahi-dev · 2026-04-21 14:19
svenugh, so it’s happening with both compiler now :(#asahi-dev · 2026-04-20 21:35
svenlooks like gcc was the one that worked for me #asahi-dev · 2026-04-20 21:34
svenI mean try the one you’re not using #asahi-dev · 2026-04-20 21:34
svencould just be different instructions that result in different timing #asahi-dev · 2026-04-20 21:34
svenI don’t know #asahi-dev · 2026-04-20 21:33
sveniirc it only happens on gcc or clang, can’t remember which one #asahi-dev · 2026-04-20 21:33
sventry another compiler :(#asahi-dev · 2026-04-20 18:25
svenbut i never investigated in detail#asahi-dev · 2026-04-20 18:25
svenwhat i suspect happens is that it tries to come out of deep sleep and then things are very messed up and m1n1 really doesn't expect that and then it reboots#asahi-dev · 2026-04-20 18:24
svendon't remember the last instruction i could run but i suspect it was the final wfi#asahi-dev · 2026-04-20 18:24
svenand even when single stepping i then eventually lost my shell and the system rebooted iirc#asahi-dev · 2026-04-20 18:23
svenah, on t8103 it went through the whole full cluster sleep sequence back then. pretty sure i confirmed that with a breakpoint#asahi-dev · 2026-04-20 18:22
svenall i remember is that on t8103 if you click on the apple logo and then select sleep the system reboots without any exception visible even on hw serial iirc#asahi-dev · 2026-04-20 18:17
svenbut sleep works if you boot xnu from m1n1 inside el2 without the hv inbetween#asahi-dev · 2026-04-20 18:17
sven<tpw_rules> (does t8103 deep sleep work under macos booted through m1n1 hv?) <-- no, the hv dies and causes a full reset during sleep/wake iirc#asahi-dev · 2026-04-20 16:20
sventhey just didn't turn in on properly, which to be fair is quite annoying for tbt#asahi-dev · 2026-04-20 16:20
svensame story for that issue that claimed tbt doesn't work due to some fabric firewall#asahi-dev · 2026-04-20 16:07
sventhat "fabric firewall" thing seems to be something these models really like to hallucinate #asahi-dev · 2026-04-20 15:57
svenwilfredor: final warning, if you post more LLM slop i will ban you#asahi-dev · 2026-04-20 07:08
svenI guess they moved the chickens to iboot for m4/a18#asahi-dev · 2026-04-19 20:56
sveniirc someone did some reverse engineering in #asahi-re a few weeks (months?) ago#asahi-dev · 2026-04-19 20:55
svenpmgr somehow changed on m4#asahi-dev · 2026-04-19 17:02
sveni vaguely remember testing sleep in that stage because the HV didn't survive that and I wanted to make sure it worked without the HV#asahi-dev · 2026-04-19 17:01
sven*sleep#asahi-dev · 2026-04-19 17:01
svenah true, and deep sleeps works when just using xnu instead of m1n1#asahi-dev · 2026-04-19 17:01
sventhere's https://github.com/Siguza/recfg/tree/master i guess#asahi-dev · 2026-04-19 17:01
svensame#asahi-dev · 2026-04-19 17:00
sven(and if it isn't sane file another bug w/ apple :D)#asahi-dev · 2026-04-19 16:59
svencould maybe take a look at that reconfig engine sequence to see if it looks sane but i'm not sure that's documented anywhere#asahi-dev · 2026-04-19 16:59
svencan't think of anything else#asahi-dev · 2026-04-19 16:52
svenhrm, does xnu do any writes to aic mmio that we don't understand?#asahi-dev · 2026-04-19 16:48
svenand i guess hang here means it goes to sleep and never comes up again?#asahi-dev · 2026-04-19 16:47
svendid you get the msr writes from open source xnu or from disassembled xnu? i guess the latter, just asking because open-source xnu might miss some writes#asahi-dev · 2026-04-17 09:45
sventrue, find_regs should’ve found everything I think #asahi-dev · 2026-04-17 08:34
svenit may not even try to reconfigure that behavior #asahi-dev · 2026-04-17 08:34
sventhough I guess XNU always guards wfi with reg save/restore anyway #asahi-dev · 2026-04-17 08:34
svenand then check if we can write to any of them from raw mode as well #asahi-dev · 2026-04-17 08:33
sventhat’s what I’d do as well #asahi-dev · 2026-04-16 19:09
sven*commits#asahi-dev · 2026-04-16 19:09
sveni think the rest would've taken the commit anyway if they objected#asahi-dev · 2026-04-16 19:09
svenoh, sure#asahi-dev · 2026-04-16 19:07
svenjannau: when you re-send the M3 device trees can you mention in the cover letter that we'd like to take everything through the soc tree unless anyone objects?#asahi-dev · 2026-04-16 18:53
svenhmm… possibly. some regs are going to be GL2-only though #asahi-dev · 2026-04-16 14:40
sveni think we have enough code in there that handles any irq by just acking it#asahi-dev · 2026-04-16 14:35
svensiguza mentioned that the reconfig engine is locked down by iboot#asahi-dev · 2026-04-16 14:35
svenyeah#asahi-dev · 2026-04-16 14:35
svenone problem is that our HV doesn't survive a sleep/wake cycle fwiw