#asahi-dev · 2026-06-09 06:32
StanfordAppliedCyber[m]well it literally does run in the userspace level of the GLs#asahi-dev · 2026-06-09 06:31
svenputting it into el1 seems painful when you can just put it into gl0 and leave el1 state alone #asahi-dev · 2026-06-09 06:31
svenjust think of TXM as sptm’s userspace #asahi-dev · 2026-06-09 06:31
LarstiQand now it works?#asahi-dev · 2026-06-09 06:30
StanfordAppliedCyber[m]i set a bunch of stuff wrong in txm along with a lot of it being stubbed out#asahi-dev · 2026-06-09 06:30
StanfordAppliedCyber[m]oh for fucks sake i may just be a fucking idiot#asahi-dev · 2026-06-09 06:27
nickchanon the upside though GL0 has not nearly as many state as EL1#asahi-dev · 2026-06-09 06:26
nickchanthat bit is controlling whether you run a VM or run userspace appliaction#asahi-dev · 2026-06-09 06:26
nickchanthat's why I say not really virtualize#asahi-dev · 2026-06-09 06:23
nickchan-> All exceptions that would be routed to EL1 are routed to EL2. #asahi-dev · 2026-06-09 06:23
nickchan-please look HCR_EL2.TGE#asahi-dev · 2026-06-09 06:23
nickchan-https://developer.arm.com/documentation/ddi0601/2026-03/AArch64-Registers/HCR-EL2--Hypervisor-Configuration-Register#asahi-dev · 2026-06-09 06:23
StanfordAppliedCyber[m]idk, tbh i havent' really looked at the txm firmware, i've mostly just looked at the xnu source and gone "how can i make this take the right code path"#asahi-dev · 2026-06-09 06:22
StanfordAppliedCyber[m]i thought there was some apple custom wrinkle to it, not just standard vhe#asahi-dev · 2026-06-09 06:22
NickChan[m]svc turned into hvc is VHE#asahi-dev · 2026-06-09 06:22
StanfordAppliedCyber[m]as a bonus all the shared mappings are really easy because we know exactly what they are by the emulator#asahi-dev · 2026-06-09 06:21
StanfordAppliedCyber[m]i'd probably just run the whole thing in el1 and patch svc --> hvc#asahi-dev · 2026-06-09 06:21
StanfordAppliedCyber[m]but it does some weird stuff involving svc being turned into an hvc effectively by some weird mechanism#asahi-dev · 2026-06-09 06:21
StanfordAppliedCyber[m]yeah i'm familiar#asahi-dev · 2026-06-09 06:19
NickChan[m]well, I say virtualize, but what I mean is run an unprivileged application since txm runs in GL0; still has context switch issue though#asahi-dev · 2026-06-09 06:18
StanfordAppliedCyber[m]that's not a terrible idea, if i run out of ideas i'll look into it#asahi-dev · 2026-06-09 06:18
StanfordAppliedCyber[m]txm is currently the most BS emulator endpoint by far, dart/sart/nvme/uat are all pretty legit, xnu_bootstrap i had to slave over but i'm very confident works#asahi-dev · 2026-06-09 06:18
NickChan[m]txm shouldn't be overly hard to virtualize the main issue is m1n1 hypervisor can only host a single guest#asahi-dev · 2026-06-09 06:15
StanfordAppliedCyber[m]i'll see if i can rule txm out next#asahi-dev · 2026-06-09 06:15
StanfordAppliedCyber[m]wow we had the same thought at the same time#asahi-dev · 2026-06-09 06:15
StanfordAppliedCyber[m]the only "obvious" thing is that a bunch of TXM is currently stubbed out to either return success or not supported#asahi-dev · 2026-06-09 06:15
NickChan[m]Did you run txm?#asahi-dev · 2026-06-09 06:08
StanfordAppliedCyber[m]i can verify that the txm-dart secure channel is both totally fine and not used for this cmd, that the SEP has successfully written to the OOL pages for different cmds in the same endpoint, and that the sep firmware is definitely alive#asahi-dev · 2026-06-09 05:50
StanfordAppliedCyber[m]yeah im really not sure how to debug this-- i just don't get a reply to the stac command#asahi-dev · 2026-06-09 05:35
StanfordAppliedCyber[m]none of the gigalocker endpoints are touched before we panic#asahi-dev · 2026-06-09 05:34
StanfordAppliedCyber[m]okay it's definitely not gigalocker#asahi-dev · 2026-06-09 05:07
StanfordAppliedCyber[m]then again we'd never get a clear Y/N, it'd just be "the sep black box won't work with us and we dont know why"#asahi-dev · 2026-06-09 05:06
StanfordAppliedCyber[m]i suppose that in itself is a finding-- "m1n1 hypervisor will never work on raw boot mode", but that would be a sad discovery#asahi-dev · 2026-06-09 05:06
StanfordAppliedCyber[m]ack, i'm really hoping this is just something i'm doing wrong because if it's not then we have problems#asahi-dev · 2026-06-09 05:03
svenxart is one of the earliest things that gets set up#asahi-dev · 2026-06-09 05:03
StanfordAppliedCyber[m]also, when is it configured in regular macos boot? is it before or after decryption of user data (and thus the publishing of the sks service)?#asahi-dev · 2026-06-09 05:03
StanfordAppliedCyber[m]dart-txm is totally silent, the firmware is definitely loaded#asahi-dev · 2026-06-09 05:02
StanfordAppliedCyber[m]i'm running out of ideas on what we could possibly be breaking that's causing sep xart to be unhappy#asahi-dev · 2026-06-09 05:02
StanfordAppliedCyber[m]chaos_princess: what do we know about gigalocker? like, what sets it up and how does sep talk to it?#asahi-dev · 2026-06-09 04:58
StanfordAppliedCyber[m]this was never an issue with previous hypervisors?#asahi-dev · 2026-06-09 04:44
StanfordAppliedCyber[m]this does make me feel slightly better though, because it likely means it's something I'M doing wrong as opposed to this just being impossible in raw boot mode#asahi-dev · 2026-06-09 04:43
StanfordAppliedCyber[m]okay this is very fun#asahi-dev · 2026-06-09 04:43
StanfordAppliedCyber[m]but xart does work?#asahi-dev · 2026-06-09 04:43
StanfordAppliedCyber[m]well i suppose i get to get xart working then#asahi-dev · 2026-06-09 04:42
StanfordAppliedCyber[m]ah yes this does appear to be a problem, it takes the real aes key, encrypts it with some ephmeral key generated from TRNG, then xnu just takes the encrypted key and passes it to ANS#asahi-dev · 2026-06-09 04:11
StanfordAppliedCyber[m]if it doesn't then it's time to bash my head against a wall hoping SEP works#asahi-dev · 2026-06-09 04:10
StanfordAppliedCyber[m]i have a script that successfully decrypted some files on some other volume on the local machine so i'll see if it works#asahi-dev · 2026-06-09 04:09
StanfordAppliedCyber[m]well, xart (the endpoint) responds, it's stac that's having issues#asahi-dev · 2026-06-09 04:09
svenand sks definitely only starts working once xart is up #asahi-dev · 2026-06-09 04:08
svenit’s even mentioned in the platform security pdf that Apple publishes somewhere