#asahi-dev · 2026-05-14 12:33
chadmed"M" doesnt really tell us anything about the core and is confusing because the socs are also M #asahi-dev · 2026-05-14 12:31
chadmedif we've got tahiti are we also gonna get pitcairn as an m5 ghz edition ;)#asahi-dev · 2026-05-14 12:31
nickchanpowermetrics refers to the cluster as M in the shipping macOS of m5 pro/max#asahi-dev · 2026-05-14 12:30
nickchanmaybe call the new performance cores cpu_m0 cpu_m1 cpu_m2 ...#asahi-dev · 2026-05-14 12:29
nickchanatmospheric phenomena -> mountains -> islands#asahi-dev · 2026-05-14 12:28
chadmeddont forget we've got "super" cores on the way too :P #asahi-dev · 2026-05-14 12:28
svenguess we'll just stick to everest/sawtooth and just continue with donan (e-core)/donan (p-core) etc. since those really don't seem to have separate codenames anymore#asahi-dev · 2026-05-14 12:27
svenwhat a mess#asahi-dev · 2026-05-14 12:25
chadmedalso why did we only get two mountains and then move on to islands, which overlap their radio antenna codenames lmfao#asahi-dev · 2026-05-14 12:24
chadmedif they rename the cores for each SKU in the same family theyre going to run out of islands#asahi-dev · 2026-05-14 12:18
nickchanthe cpu names#asahi-dev · 2026-05-14 12:18
nickchanm3 seems not be everest, but really, it's a big mess now#asahi-dev · 2026-05-14 12:18
nickchanhttps://github.com/llvm/llvm-project/blob/5366692aeefd1101cb8dfde7fb65c9565b476d96/llvm/lib/TargetParser/Host.cpp#L1746-L1777#asahi-dev · 2026-05-14 09:33
nickchanthis would make macos run on a single cpu#asahi-dev · 2026-05-14 09:32
nickchanStanfordAppliedCyber[m]: add cpus=1 to boot args #asahi-dev · 2026-05-14 09:28
StanfordAppliedCyber[m]good news: device init works, we're almost all the way to bsd_init#asahi-dev · 2026-05-14 09:28
StanfordAppliedCyber[m]well calling it for today, my hypervisor is essentially blocked on smp bringup, tried pruning a bunch of stuff from the adt to keep going on a single core but adt remains very ornery and i don't feel like monkeypatching it more#asahi-dev · 2026-05-14 08:09
StanfordAppliedCyber[m]which i'd be fine with if i didn't get told "smp just works"#asahi-dev · 2026-05-14 08:07
StanfordAppliedCyber[m]my cpu cores simply will not start#asahi-dev · 2026-05-14 08:06
StanfordAppliedCyber[m]okay, i definitely need some help here-- can someone send me a pointer to a working branch i can just run make on, and a corresponding working binary just to see if something very funky is happening?#asahi-dev · 2026-05-14 07:37
StanfordAppliedCyber[m]yeah ik, i'm familiar, it just surprised me because afaik the secondary cpu start has never involved them, not that mmio reads had side effects generally#asahi-dev · 2026-05-14 07:36
mazinterrupt acknowledgement, for the most common example.#asahi-dev · 2026-05-14 07:35
maza lot of MMIO registers have read side effects, and not only in the Apple world. that's a very common thing to do.#asahi-dev · 2026-05-14 07:08
StanfordAppliedCyber[m]i'm having issues getting a read trace from macos but it's clear reads have side effects#asahi-dev · 2026-05-14 07:08
StanfordAppliedCyber[m]reading from reg-private +0x30 causes an serror storm?#asahi-dev · 2026-05-14 06:18
StanfordAppliedCyber[m]okay, let's replay the macos reads in the exact same order#asahi-dev · 2026-05-14 06:18
StanfordAppliedCyber[m]clean, completely fresh#asahi-dev · 2026-05-14 06:18
StanfordAppliedCyber[m]nothing else#asahi-dev · 2026-05-14 06:18
StanfordAppliedCyber[m]literally p.read32(0x210050020), then cpu0_ps = 0x1f0 and cluster0_ps = 0x21f0#asahi-dev · 2026-05-14 06:16
StanfordAppliedCyber[m]READS are load bearing?#asahi-dev · 2026-05-14 06:16
StanfordAppliedCyber[m]what the fuck#asahi-dev · 2026-05-14 06:16
StanfordAppliedCyber[m]reading from cpu-impl-reg + 0x20 changes the core pstate#asahi-dev · 2026-05-14 06:09
StanfordAppliedCyber[m].......there might be an mmio read here#asahi-dev · 2026-05-14 06:05
StanfordAppliedCyber[m]in a vacuum i'd say this feels like a sysreg-- but afaik sysregs have never been involved in smp init in apple silicon#asahi-dev · 2026-05-14 05:54
StanfordAppliedCyber[m]--> it definitely performed the mmio writes it expected to start the cpu#asahi-dev · 2026-05-14 05:54
StanfordAppliedCyber[m]and this trace is from a run where macos panicked on "waited some ms for cpu core to start"#asahi-dev · 2026-05-14 05:42
StanfordAppliedCyber[m]i'm replaying, quite literally, every single mmio write macos made in the same order#asahi-dev · 2026-05-14 05:42
StanfordAppliedCyber[m]something fishy is going on here#asahi-dev · 2026-05-14 04:46
StanfordAppliedCyber[m]failed to start the secondaries also#asahi-dev · 2026-05-14 04:46
StanfordAppliedCyber[m]the only thing i can think of is that i have a sort of weird setup-- basically chainload.py doesn't work on my machine, so i broke m1n1 into a tiny dumb loader with an embedded kernelcache, which is flashed to 1tr, and the rest of m1n1 which i stream over uart to a specific location in memory then jump to a raw physical address, but i also literally just cloned yuka's m1n1 branch with the commits, flashed that in 1tr, and it#asahi-dev · 2026-05-14 04:44
StanfordAppliedCyber[m]previously i've had other stuff there including _vectors_start, the m1n1 entrypoint, basically anything you can think of#asahi-dev · 2026-05-14 04:41
StanfordAppliedCyber[m]that counter has never been hit#asahi-dev · 2026-05-14 04:40
StanfordAppliedCyber[m]currently i literally just have an instruction that increments a counter then calls WFE just so that if any core drops out of reset i see the side effect#asahi-dev · 2026-05-14 04:40
StanfordAppliedCyber[m]what i've been doing is writing different instructions to that address#asahi-dev · 2026-05-14 04:40
StanfordAppliedCyber[m]sorry, let me clarify, my value in rvbar is 0x10004ef0000 which i can verify i can't change#asahi-dev · 2026-05-14 04:32
JamesCalligeros[m]we can't write to rvbar on M3+, it's locked by iboot#asahi-dev · 2026-05-14 04:30
StanfordAppliedCyber[m]what i don't get is that for everyone else it seemed to "just work" and nothing in my mmio traces stick out as "wow this is super different about my system"#asahi-dev · 2026-05-14 04:26
StanfordAppliedCyber[m]i did previously try m1n1 entrypoint#asahi-dev · 2026-05-14 04:26
StanfordAppliedCyber[m]well i've tried a bunch of things, right now i'm walking through macos mmio via my hypervisor and we deliberately don't write to rvbar#asahi-dev · 2026-05-14 04:25
StanfordAppliedCyber[m]_vectors_start