#asahi-dev · 2026-06-11 00:36
StanfordAppliedCyber[m]btw the issue with implementing them is there's basically two ways for them to talk to each other: some custom protocol managed by sptm (tightbeam) and shared memory buffers#asahi-dev · 2026-06-11 00:36
StanfordAppliedCyber[m]as an extra guest*, two guests total (xnu and secure kernel)#asahi-dev · 2026-06-11 00:34
StanfordAppliedCyber[m]i'd just run secore kernel + exclave kit in el1/0 respectively as two other guests#asahi-dev · 2026-06-11 00:33
StanfordAppliedCyber[m]yes#asahi-dev · 2026-06-11 00:33
StanfordAppliedCyber[m]the biggest risk would be all the apple sysregs ofc#asahi-dev · 2026-06-11 00:32
StanfordAppliedCyber[m]if i had to implement exclaves i think it would take under a week#asahi-dev · 2026-06-11 00:32
StanfordAppliedCyber[m]no doubt, but fwiw i'm still not super worried about them#asahi-dev · 2026-06-11 00:29
StanfordAppliedCyber[m]most of my work has been related to turing them off, which is definitely possible (for now)#asahi-dev · 2026-06-11 00:28
StanfordAppliedCyber[m]if you look at the adt there's a lot of nodes with exclave related stuff in them#asahi-dev · 2026-06-11 00:28
StanfordAppliedCyber[m]i havent looked at it exactly but yes that is my guess#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]they all have some secure exclave mailbox thingy#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]esm#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]isp#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]ane#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]aop#asahi-dev · 2026-06-10 22:59
StanfordAppliedCyber[m]oh god basically everything uses exclaves now#asahi-dev · 2026-06-10 22:57
StanfordAppliedCyber[m]but i figure it can't hurt#asahi-dev · 2026-06-10 22:57
StanfordAppliedCyber[m]idk how much of a difference it'd make#asahi-dev · 2026-06-10 22:57
StanfordAppliedCyber[m]beautiful-- @ sven + yuka i can reach out to my contact with radar numbers if/when yall file the stuff (ik the aic one was already filed)#asahi-dev · 2026-06-10 09:44
StanfordAppliedCyber[m]but given it's almost 3am this is a problem for later today#asahi-dev · 2026-06-10 09:44
StanfordAppliedCyber[m]something's up with the dcp, it's not completing its firmware handshake#asahi-dev · 2026-06-10 09:44
StanfordAppliedCyber[m]well, there's 150 processes, windowserver is running, it feels like i'm very, very close to a working hypervisor#asahi-dev · 2026-06-10 09:38
StanfordAppliedCyber[m]suffice to say, either of these two problems means 26.5 isnt very appealing#asahi-dev · 2026-06-10 09:37
StanfordAppliedCyber[m]sven: i did quite literally nothing but rerun my test and this time i lost every register#asahi-dev · 2026-06-10 09:37
StanfordAppliedCyber[m]well#asahi-dev · 2026-06-10 09:37
StanfordAppliedCyber[m]oh#asahi-dev · 2026-06-10 09:26
StanfordAppliedCyber[m]it's schedule interrupt, write a bunch of registers, wfi, wait for the interrupt to wake us up, dump all registers, return#asahi-dev · 2026-06-10 09:25
StanfordAppliedCyber[m]no this test is artificial#asahi-dev · 2026-06-10 09:25
StanfordAppliedCyber[m]oh ffs#asahi-dev · 2026-06-10 09:25
StanfordAppliedCyber[m]the test is a dumb asm blob#asahi-dev · 2026-06-10 09:25
StanfordAppliedCyber[m]i'm being terse#asahi-dev · 2026-06-10 09:25
StanfordAppliedCyber[m]oh yeah that's what i mean#asahi-dev · 2026-06-10 09:24
StanfordAppliedCyber[m]yeah i ran the test on a secondary#asahi-dev · 2026-06-10 09:23
StanfordAppliedCyber[m]sven: wfi issue is fixed, all register state is saved#asahi-dev · 2026-06-10 09:23
StanfordAppliedCyber[m]my goal is to get it working with the latest macos version on all SoCs including sptm enabled ones#asahi-dev · 2026-06-10 09:23
StanfordAppliedCyber[m]^ at the moment#asahi-dev · 2026-06-10 09:22
StanfordAppliedCyber[m]oh, it's probably worth mentioning making it clear that the hypervisor only supports old macos versions on non sptm enabled socs#asahi-dev · 2026-06-10 08:41
StanfordAppliedCyber[m]sptm itself is not helpful to linux, and my emulator is tainted anyways and i'll never push that code, but we need to find a way to get macos working on sptm enabled systems so we can watch what it does#asahi-dev · 2026-06-10 08:40
StanfordAppliedCyber[m]commpage fixups, timer stuff, register handling, some interrupt stuff (which i'm hoping i can actually delete)#asahi-dev · 2026-06-10 08:40
StanfordAppliedCyber[m]there's a lot of hypervisor stuff that can be upstreamed to m1n1 because it's my code and is definitely needed for the support of newer macos/socs#asahi-dev · 2026-06-10 08:38
StanfordAppliedCyber[m]:(#asahi-dev · 2026-06-10 08:37
StanfordAppliedCyber[m]yeah, msr that register --> Exception: SYNC#asahi-dev · 2026-06-10 08:32
StanfordAppliedCyber[m]sven: reads work fine, writes fault#asahi-dev · 2026-06-10 08:30
StanfordAppliedCyber[m]but this is pure speculation on my part#asahi-dev · 2026-06-10 08:30
StanfordAppliedCyber[m]it's possibly to support machines that don't have exclave support, and the code is basically just vestigal#asahi-dev · 2026-06-10 08:30
StanfordAppliedCyber[m]i wonder why this path even exists? like exclaves are just always turned on#asahi-dev · 2026-06-10 08:25
StanfordAppliedCyber[m]just grep "ex" or "exclave" in the adt#asahi-dev · 2026-06-10 08:25
StanfordAppliedCyber[m]you do have to clean up the adt a bit#asahi-dev · 2026-06-10 08:23
StanfordAppliedCyber[m]how long that'll be true, who knows#asahi-dev · 2026-06-10 08:22
StanfordAppliedCyber[m]good news: with the exception of possibly audio i think they all have graceful non exclave fallback paths