#
#asahi-re · 2026-08-22 19:53
svenso there are 12 tx rings and 12 rx rings and 24 dart stream ids. maybe one per id now?
#asahi-re · 2026-08-22 19:52
svenlol, wtf
#asahi-re · 2026-08-22 19:31
chaos_princesswe should probably do something about dart remap
#asahi-re · 2026-08-22 19:31
chaos_princesshttps://github.com/AsahiLinux/linux/pull/575/changes#diff-a764a588064a9b16059358c0e88d318bb4bc74b79692e00d08cd7440a36d05d1R272
#asahi-re · 2026-08-22 19:06
jannauand just to make sure, acio0 on t8122 writes 0x1 to 0x2d0780008
#asahi-re · 2026-08-22 19:04
jannauand 0x350780008 for t6030/acio1, same 0x2 write and poll for 0x0
#asahi-re · 2026-08-22 18:53
sventhat address also looks like t600x style
#asahi-re · 2026-08-22 18:53
svensounds like the t600x style if that was port 2
#asahi-re · 2026-08-22 18:52
jannauit writes 0x2 and polls until it is zero
#asahi-re · 2026-08-22 18:51
chaos_princessm3 should also be using the pro/max style reset so acio0 should be the same
#asahi-re · 2026-08-22 18:50
chaos_princesstyvm
#asahi-re · 2026-08-22 18:50
jannauchaos_princess: t8122/acio1 cio-reset 0x2d0780008
#asahi-re · 2026-08-22 15:35
svenyeah, exactly
#asahi-re · 2026-08-22 15:35
chaos_princessyea, ghidra outright choked on pmgr init, so, meh, not interested in going that way
#asahi-re · 2026-08-22 15:35
svenbut i didn't manage to track that down
#asahi-re · 2026-08-22 15:35
sveni think they somehow initialize that table with the adt entries first and then add a few more 'special' ones
#asahi-re · 2026-08-22 15:34
svenyeah
#asahi-re · 2026-08-22 15:34
chaos_princessSo, my approach is - set a hw breakpoint in kext, read the address
#asahi-re · 2026-08-22 15:34
svenand once i had the right addr it didn't line up with the delta they add to anything in the ADT
#asahi-re · 2026-08-22 15:34
svenin the M2 kext the range for the address was some offset that doesn't exist in the adt
#asahi-re · 2026-08-22 15:33
sveni couldn't figure out where that weird reg range comes from because ugh... c++ vtables
#asahi-re · 2026-08-22 15:33
chaos_princessthe reg range is in the adt on t6031, but i can't figure out how to map kext-internal numbers to adt numbers
#asahi-re · 2026-08-22 15:33
svenyeah, that one also made me get the M2 one wrong at first :(
#asahi-re · 2026-08-22 15:33
chaos_princessyes
#asahi-re · 2026-08-22 15:33
sveni guess it also isn't hardcoded in the kext anymore but goes through some virtual(?) reg range that doesn't exist in the ADT now?
#asahi-re · 2026-08-22 15:31
chaos_princessanyone with a m3 vanilla/pro and a tracing setup on 14.8.3 mind helping out find the tb reset address?
#asahi-re · 2026-08-19 07:02
dj26.5.2, m1n1 is 13.5
#asahi-re · 2026-08-19 06:56
chaos_princesswhat macos version are you using
#asahi-re · 2026-08-19 06:48
djI found something in iboot related to this `enable-sep-rm` which maybe hints that only xnu is allowed to reset sep
#asahi-re · 2026-08-19 06:47
djalso, is it not possible to put sep back into seprom state? I haven't had any success resetting it
#asahi-re · 2026-08-19 06:45
djthanks that's a very helpful clue, is there any existing work I can follow to init xarm?
#asahi-re · 2026-08-19 06:39
djohhh
#asahi-re · 2026-08-19 06:39
chaos_princessdj: initialize xarm and a lot more endpoints will appear
#asahi-re · 2026-08-19 06:38
sveni think sep.rs does that eventually
#asahi-re · 2026-08-19 06:38
svenno
#asahi-re · 2026-08-19 06:38
djhmm, but boot tz0 is not sent?
#asahi-re · 2026-08-19 06:37
svenit uses the rng the rom exposes to pass a random seed to Linux
#asahi-re · 2026-08-19 06:37
NickChan[m]It reads a 128 byte ramdom value
#asahi-re · 2026-08-19 06:37
djm1n1 doesn't do anything with it by default does it?
#asahi-re · 2026-08-19 06:37
svenon the 15.x we use it still runs in the rom when m1n1 gets control
#asahi-re · 2026-08-19 06:36
svenmust’ve been changed then
#asahi-re · 2026-08-19 06:36
djerm I forgot but a very recent release of macOS 26 I believe, so it would be using a relatively new iboot?
#asahi-re · 2026-08-19 06:35
svenwhich macOS version is this?
#asahi-re · 2026-08-19 06:35
djthese endpoints were advertised: stac, cntl, pnon, hdcp, xars, and xarm
#asahi-re · 2026-08-19 06:35
svenhuh, that’s odd
#asahi-re · 2026-08-19 06:35
djmakes sense then, seems like iBoot already boots sepos and you can't talk to the seprom endpoint anymore. it doesn't even seem possible to put sep back into seprom state
#asahi-re · 2026-08-19 06:34
NickChan[m]that boot flow is only correct on ancient iOS
#asahi-re · 2026-08-19 06:32
dj> SEP boot flow: - iBoot preloads SEP firmware into a memory region that's recorded in ADT. - XNU sends a boottz0 message, bringing SEP into second stage boot (a portion of RAM is reserved for this) but from my testing what I find is sepfw is already booted, and if I poke at it like macOS does it advertises its endpoints, but it's only 7 of them and omits SBIO. is this some sort of new behaviour on M2? I'm testing M2 Pro
#asahi-re · 2026-08-19 06:29
djanyone familiar with the sep re work here? I've been poking at it and started by reading https://asahilinux.org/docs/hw/soc/sep/ and sep.rs
#asahi-re · 2026-08-18 21:44
nicolas17"supportsMXUNarrowTileSizes", "supportsPackUnpackSmallInteger", "supportsRGBTextureBuffers", "supportsSIMDGroupParallelForwardProgress", "supportsTextureMultifetch", "supportsTextureViewMinLOD"

plus ›