#asahi-dev · 2026-06-24 02:28
StanfordAppliedCyber[m]in fact it looks like a very normal arm kernel/user space despite running in gl1/0#asahi-dev · 2026-06-24 02:28
StanfordAppliedCyber[m]also i'm looking at the possibility of virtualizing exclaves, they don't contain, from a very stupid python script, anything fun, ie, apple custom instructions, weird sysreg accesses, etc#asahi-dev · 2026-06-24 02:03
StanfordAppliedCyber[m]worst case flash the real kernelcache, boot into real macos, then just look at the disk#asahi-dev · 2026-06-24 02:03
StanfordAppliedCyber[m]this shouldn't be hard to figure out#asahi-dev · 2026-06-24 02:03
StanfordAppliedCyber[m]uhhhhhhhhhhhhhhhhhhhhhhhh check diskutil?#asahi-dev · 2026-06-24 02:03
StanfordAppliedCyber[m]well the solution is just to delete them#asahi-dev · 2026-06-24 01:58
StanfordAppliedCyber[m]i burned through 200gb of space because coredumps + crashlogs used all of it#asahi-dev · 2026-06-24 01:58
StanfordAppliedCyber[m]check if crashlogs are filling it up#asahi-dev · 2026-06-24 01:37
StanfordAppliedCyber[m]these machines, sigh#asahi-dev · 2026-06-24 01:37
StanfordAppliedCyber[m]and for some reason reverting to a known good commit still has the error?#asahi-dev · 2026-06-24 01:33
StanfordAppliedCyber[m]apparently i have a pretty big regression somewhere because i can't even get to windowserver anymore, can't access the encrypted userdata#asahi-dev · 2026-06-24 01:33
StanfordAppliedCyber[m]debugging the sep has to be the worst part of this hypervisor, when it doesn't work we have no way to tell why it doesn't work#asahi-dev · 2026-06-23 07:34
StanfordAppliedCyber[m]if i just allow the accesses we get serrors in el2#asahi-dev · 2026-06-23 07:34
StanfordAppliedCyber[m]i just know i spent a week brute forcing every solution i could think of until i foudn something that could make progress#asahi-dev · 2026-06-23 07:34
StanfordAppliedCyber[m]to be clear i have no real clue of what is happenning at the hardware level#asahi-dev · 2026-06-23 07:33
StanfordAppliedCyber[m]interesting, i'll try that as an experiment sometime#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]and i have more sources of error due to the insane amount of sptm latenccy + cache flushing we do#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]maybe 5% of boots now flake whereas before it was 50%+#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]and that made boot massively more reliable#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]like i added some extra time compensation for time spent in the hypervisor#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]i think current m1n1's timer handling is a bit off#asahi-dev · 2026-06-23 07:32
StanfordAppliedCyber[m]i actually have a mitigation for that one#asahi-dev · 2026-06-23 07:31
StanfordAppliedCyber[m]but the closest xnu symbol for me is "restoreCPUEnergyAccumulator" (i didn't look beyond finding the closest symbol)#asahi-dev · 2026-06-23 07:31
StanfordAppliedCyber[m]it may or may not be related#asahi-dev · 2026-06-23 07:31
StanfordAppliedCyber[m]but i have a horrible file that's 50 lines of god awful hacks that does a combination of intercepting some pmgr writes (based on the program counter), and then either returning 0, performing the access from el2 and returning the restult, or returning a monotonically increasing counter#asahi-dev · 2026-06-23 07:30
StanfordAppliedCyber[m]mine were deterministic tahnkfully#asahi-dev · 2026-06-23 07:30
StanfordAppliedCyber[m]on a18pro#asahi-dev · 2026-06-23 07:30
StanfordAppliedCyber[m]i had some weird pmgr difficulties also#asahi-dev · 2026-06-23 07:30
StanfordAppliedCyber[m]i have an idea of what that nondeterminism may come from#asahi-dev · 2026-06-23 07:28
StanfordAppliedCyber[m]iirc there was some gist about how to boot m3 devices under the hypervisor, where is that guide?#asahi-dev · 2026-06-23 00:18
StanfordAppliedCyber[m]oh, apparently it isn't fully reliable and the run i did after making that change just got lucky, oh well#asahi-dev · 2026-06-23 00:07
StanfordAppliedCyber[m]it's kind of hard to read as it is#asahi-dev · 2026-06-23 00:07
StanfordAppliedCyber[m]does anyone mind if i reformat the apple proprietary instruction page into a regular markdown table?#asahi-dev · 2026-06-22 23:24
StanfordAppliedCyber[m]i'll make a pr to docs with my current understanding of these ops and sysregs to the apple custom instructions page#asahi-dev · 2026-06-22 23:23
StanfordAppliedCyber[m]progress bar is halfway full now#asahi-dev · 2026-06-22 23:23
StanfordAppliedCyber[m]yep that worked#asahi-dev · 2026-06-22 23:08
StanfordAppliedCyber[m]okay i can confirm the sptm sysreg series from m3 doesn't fault on a18 pro, going to wire that up to my emulator and hope it works#asahi-dev · 2026-06-22 23:07
StanfordAppliedCyber[m]yep#asahi-dev · 2026-06-22 22:55
StanfordAppliedCyber[m]okay i need to write a doc about this, this is useful to other people#asahi-dev · 2026-06-22 22:52
StanfordAppliedCyber[m]it appears sptm for m3 has a completely different cache maintenence mechanism that relies on some sysregs#asahi-dev · 2026-06-22 22:51
StanfordAppliedCyber[m]i can confirm that the weird gxf cache instructions are NOT present on m3 socs#asahi-dev · 2026-06-22 22:06
StanfordAppliedCyber[m]it's probably new to m4+ socs#asahi-dev · 2026-06-22 22:06
StanfordAppliedCyber[m]yuka just checked and that instruction doesn't exist on m3 max#asahi-dev · 2026-06-22 21:34
StanfordAppliedCyber[m]sure, i can do that, thanks for the help! will send a pastebin with tests#asahi-dev · 2026-06-22 21:34
StanfordAppliedCyber[m]basically real sptm has a block like dsb 0x00201401 dsb which clearly is doing some page/cache/whatever shenanigans but i'm not sure what and emulating it is proving a pain#asahi-dev · 2026-06-22 21:33
StanfordAppliedCyber[m]UNEXP_RT_HIT_DIR#asahi-dev · 2026-06-22 21:33
StanfordAppliedCyber[m]it's some nondeterministic page fault#asahi-dev · 2026-06-22 21:33
StanfordAppliedCyber[m]these instructions are casuing me problems#asahi-dev · 2026-06-22 21:32
StanfordAppliedCyber[m]oh amazing, can someone with an m3 check if 0x00201401 and 0x00201328 fault?#asahi-dev · 2026-06-22 21:32
StanfordAppliedCyber[m]because it would really help to understand the semantics of some instructions but i can't meaningfully figure them out because they're locked on my machine