#asahi-dev · 2026-05-10 12:23
freetolgahow many generations until fuOS is only a normal virtual machine#asahi-dev · 2026-05-10 12:23
nickchanso if sptm is virtualized in EL1, where would exclaves go now? :) #asahi-dev · 2026-05-10 12:23
freetolgaholyyyy#asahi-dev · 2026-05-10 12:23
nickchanand it runs as a guest of sptm#asahi-dev · 2026-05-10 12:23
freetolgai got too rambly#asahi-dev · 2026-05-10 12:23
freetolgaoops sorry#asahi-dev · 2026-05-10 12:23
freetolgathey are cheap for a 16gb machine with good compute#asahi-dev · 2026-05-10 12:22
freetolgabefore they get refreshed#asahi-dev · 2026-05-10 12:22
nickchanruns in GL0 and GL1#asahi-dev · 2026-05-10 12:22
freetolgabut it really is a good time to get those#asahi-dev · 2026-05-10 12:22
nickchanexclave is its own OS with its own kernel (exclaveCore) and rootFS#asahi-dev · 2026-05-10 12:22
freetolgaalright brb grabbing an m4 mac mini jk#asahi-dev · 2026-05-10 12:22
nickchanm5 brought exclaves to macs#asahi-dev · 2026-05-10 12:22
freetolgasven: wait m5 did more?#asahi-dev · 2026-05-10 12:21
flokliand t8140 is the A18Pro, we'd also meed to match T8132 or however the adt calls it#asahi-dev · 2026-05-10 12:21
yukanot t6040#asahi-dev · 2026-05-10 12:21
yukathe J773s (Mac Mini M4 Pro) adt has arm-io,t6041 compatible 🫠#asahi-dev · 2026-05-10 12:11
jannauthe arm-io compatible should be per soc family and not model#asahi-dev · 2026-05-10 12:10
floklijannau: you mean various of these checks for each model?#asahi-dev · 2026-05-10 12:06
jannauand in that case don't check the offset but just the reg directly#asahi-dev · 2026-05-10 12:05
jannauI've discussed with yuka yester to simply add `if (adt_is_compatibe("arm-io,t8140")) { ... }` for every device which needs it#asahi-dev · 2026-05-10 12:01
flokliok, so j773s and j714 both have a different offset to the secondary watchdog (0x3bff8 instead of 0x8008). Tempted to add this as another check to m1n1's wdt.c#asahi-dev · 2026-05-10 11:56
jannaufor each 16k block dockchannel_uart_getbyte() spins 8 times on DATA_RX_COUNT for over 100ms#asahi-dev · 2026-05-10 11:44
jannaulooks like we're missing data#asahi-dev · 2026-05-10 11:44
jannauno, not always and waiting between the reads in the form of debug printf to the fb seems to help#asahi-dev · 2026-05-10 11:28
jannauchainload via debugusb works if I chunk iodev_read() into SZ_16K chunks#asahi-dev · 2026-05-10 11:03
sventrue, that could work#asahi-dev · 2026-05-10 10:59
chaos_princessyou can probably do a "mixed hle", where you hle the gl2 stuff and run everything else in vms#asahi-dev · 2026-05-10 10:55
chadmedM5 doesnt exist and cant hurt us#asahi-dev · 2026-05-10 10:20
svenbut if emulation works for m4 and a18 that’s already great!#asahi-dev · 2026-05-10 10:20
svenM5 is gonna be much more annoying with the secure kernel and exclaves anyway, not sure how feasible emulation is gonna be there #asahi-dev · 2026-05-10 10:19
svenif it’s much more we‘ll see how to deal with it once we get there#asahi-dev · 2026-05-10 10:18
sven1000 ofc, 1000k lines would be a bit too much :D #asahi-dev · 2026-05-10 10:17
svenlet’s see how much SPTM emulation actually requires, if it’s really just ~100-1000k lines the write documentation and then someone else implements it is gonna be fine #asahi-dev · 2026-05-10 10:16
svenWe’re not gonna do linked program crap. That’s just gonna encourage asm2c#asahi-dev · 2026-05-10 08:59
StanfordAppliedCyber[m]that said, i'm not experienced on copyright here other than true clean room == kosher#asahi-dev · 2026-05-10 08:59
StanfordAppliedCyber[m]the latter is mostly a bunch of simple endpoints with a small number of more complicated ones#asahi-dev · 2026-05-10 08:58
StanfordAppliedCyber[m]the former is a bunch of page mappings#asahi-dev · 2026-05-10 08:58
StanfordAppliedCyber[m]like, split it into two things, there's the sptm handoff, and the sptm runtime (the endpoints)#asahi-dev · 2026-05-10 08:58
StanfordAppliedCyber[m]i can say that my current emulation, which makes it past virtual memory init, is not super complicated#asahi-dev · 2026-05-10 08:57
StanfordAppliedCyber[m]I'll defer to the asahi maintainers on this one#asahi-dev · 2026-05-10 08:57
nickchanalongside documentation ofc#asahi-dev · 2026-05-10 08:57
nickchandepending how complex the emulation need to be using LinkedProgram() can make clean implementation easier since it should be legally okay to observe the input->output of the questionable blob the same way it is okay to trace xnu#asahi-dev · 2026-05-10 08:56
StanfordAppliedCyber[m]it does mean m1n1.bin is something like 120mb, but hey you only go through 1tr once#asahi-dev · 2026-05-10 08:55
StanfordAppliedCyber[m]that's how I load the kcache#asahi-dev · 2026-05-10 08:55
StanfordAppliedCyber[m]i'm not kidding, that's how we load the kcache#asahi-dev · 2026-05-10 08:55
StanfordAppliedCyber[m]dumb solution: add macho segment to m1n1, then memcpy()#asahi-dev · 2026-05-10 08:55
StanfordAppliedCyber[m]yeah i hit that#asahi-dev · 2026-05-10 08:54
yukachainload currently doesn't work over debugusb (jannau started looking into that yesterday) and the debug-uart is semi-borked on newer devices#asahi-dev · 2026-05-10 08:53
nickchani thought the way for fast changes is chainload