#asahi-dev · 2026-08-10 17:37
svendamn, this looked so promising :(#asahi-dev · 2026-08-10 17:26
svenso that 2 and 0x28 is some kind of buffer allocation(?) credit#asahi-dev · 2026-08-10 17:26
svennow this thing from an old xnu uart output have looks rather suspicious: [..] allocateSharedBuffer - Tx 0: 2 Tx 1-5: 40 Tx 6-11: 5#asahi-dev · 2026-08-10 16:50
svenyou also want this on top i think https://f.svpe.de/88f218715c70e7e16b6073bab651d97a0fb95d00dd24719d3f6e620e89bba566_tbt-fix2.patch#asahi-dev · 2026-08-10 16:48
svenactually, wait. it writes to both the original and that copy#asahi-dev · 2026-08-10 16:39
svenack#asahi-dev · 2026-08-10 16:38
svenchaos_princess: https://f.svpe.de/dc0e58423c5e366ec8c45912ac82f080326b44377f5c044b2cd2f9f86ca55b6f_tbt-wip.patch compile tested only since i'm on a $work trip and not near my m1 machine#asahi-dev · 2026-08-10 16:29
svenwhich doesn't matter for ring 0 but would explain why USB4NET packets just disappear because it does matter for ring 1 which carries them#asahi-dev · 2026-08-10 16:29
svenlooks like apple decided to think different again and moved the sof/eof mask to an entirely different mmio region :D#asahi-dev · 2026-08-10 16:04
svenwhy do i always notice these things after complaining there's no difference? just like i always find a typo or mistake just after hitting b4 send *sigh*#asahi-dev · 2026-08-10 16:03
svenyeah... i'm blind i think#asahi-dev · 2026-08-10 15:56
svenor maybe i'm just blind, lol#asahi-dev · 2026-08-10 15:49
svenor maybe that weird dart stream 15 stuff is actually required for that to work#asahi-dev · 2026-08-10 15:49
svenguess i could double-check if the tunables are all applied correctly#asahi-dev · 2026-08-10 15:48
sveni just don't see a difference to what xnu did in that trace from yesterday#asahi-dev · 2026-08-10 15:48
sveni dunno what i'm doing wrong for host2host inside the linux code :/#asahi-dev · 2026-08-10 11:04
svenif it really died it’s probably a software bug somewhere, do you see anything suspicious in dmesg?#asahi-dev · 2026-08-10 11:04
svenwe’re setting every single reset bit we found on disconnect #asahi-dev · 2026-08-10 11:03
svenso usually a disconnect/connect cycle should fix that#asahi-dev · 2026-08-09 19:42
sventhere's CONFIG_USB4_DMA_TEST, still trying to figure out what it does but that could be useful to track down the problem#asahi-dev · 2026-08-09 19:29
svenjust to confirm, this was together with that 0x4000 fix, right?#asahi-dev · 2026-08-09 19:18
svendon't see any difference between the rings other than that#asahi-dev · 2026-08-09 19:12
svenif it's not that i may have to write a real tracer for this :/#asahi-dev · 2026-08-09 19:12
svenack#asahi-dev · 2026-08-09 19:11
svenchaos_princess: can you try ring_iowrite32options(ring, ring->hop ? 0x28 : 2, 4); instead of https://github.com/AsahiLinux/linux/blob/98588460927272c731c6018af25463986fdf9e0e/drivers/thunderbolt/nhi.c#L677 #asahi-dev · 2026-08-09 19:02
svena trace from macos establishing USB4NET and sending some packets would be helpful#asahi-dev · 2026-08-09 18:55
svenand from Rafay's trace i noticed that the offset is wrong which would've nicely explained all that#asahi-dev · 2026-08-09 18:54
svenand the moment it tries to use any other ring something breaks#asahi-dev · 2026-08-09 18:54
svenfrom what i can tell all the setup happens on ring 0 and that works#asahi-dev · 2026-08-09 18:54
svenchaos_princess: so i suspect that something is wrong with anything that's not the control ring (ring 0). i just don't know what is wrong#asahi-dev · 2026-08-09 18:53
svensure, no worries. there's no rush here :)#asahi-dev · 2026-08-09 18:45
svenon top of tbt-reset-wip#asahi-dev · 2026-08-09 18:44
svenRafay: long shot: https://f.svpe.de/6fbf0a90d5ae76360bf95409345957d908d5d28f1617b0dc6dcbba8db944db49_t8112-usb4-acio-port1.patch#asahi-dev · 2026-08-09 18:32
svenchaos_princess: can you change https://github.com/AsahiLinux/linux/blob/tbt-reset-wip/drivers/thunderbolt/apple.c#L89 to 0x4000 and see if that keeps packets out of the shadow realm? :D#asahi-dev · 2026-08-09 18:19
svenbut from looking at the trace it looks like the normal asc mailbox#asahi-dev · 2026-08-09 18:19
svenyeah, that's why i got suspicious#asahi-dev · 2026-08-09 18:15
svennope, i'm just confused. it's the "normal" one#asahi-dev · 2026-08-09 18:14
svenor is that a new variant#asahi-dev · 2026-08-09 18:13
svenwait.. #asahi-dev · 2026-08-09 18:13
svennice, mailbox really looks like the normal ASC mailbox instead of that special m3 one#asahi-dev · 2026-08-09 18:10
svenit was just funny to me that they want to charge money for downloading the thing they already sent to my browser#asahi-dev · 2026-08-09 18:09
svennah, no worries, i just copy/pasted it :)#asahi-dev · 2026-08-09 18:07
svenok, let's see#asahi-dev · 2026-08-09 18:04
sventhat's from a mac mini, right?#asahi-dev · 2026-08-09 18:01
svendon't they know copy/paste works just fine#asahi-dev · 2026-08-09 18:01
svenlol, wtf, pastes.io wants money to allow downloading? :D#asahi-dev · 2026-08-09 18:00
svensounds perfect! i'll just grep for the relevant parts :)#asahi-dev · 2026-08-09 17:16
svenit'll likely be a fairly long trace