#asahi-dev · 2026-08-14 18:19
sveni'm willing to bet it's d6d458d42e1e ("thunderbolt: Handle DisplayPort tunnel activation asynchronously")#asahi-dev · 2026-08-14 18:00
svenyup, it's hanging at wait_for_completion(&anhi->nhi.domain_released) and something still has a reference to that domain#asahi-dev · 2026-08-14 17:50
svenhah nice #asahi-dev · 2026-08-14 17:23
svensomehow a domain leaks#asahi-dev · 2026-08-14 17:23
sventhis is going to be fun, this is something broken in core thunderbolt code i think :(#asahi-dev · 2026-08-14 16:48
svenaaand while focusing on xdomain i somehow broke device so bad that acio doesn't shut down anymore once i disconnect it again :D#asahi-dev · 2026-08-14 16:22
svendoes macos manage to establish that connection?#asahi-dev · 2026-08-14 16:15
svenwtf#asahi-dev · 2026-08-14 16:15
svenlol#asahi-dev · 2026-08-14 15:50
svenodd#asahi-dev · 2026-08-14 15:49
svendoes acio complain with anything in dmesg?#asahi-dev · 2026-08-14 15:44
svendid the tracer spam anything?#asahi-dev · 2026-08-14 15:44
svenbut that's still still surprising because i don't really trace the actual data ring#asahi-dev · 2026-08-14 15:44
svenphew#asahi-dev · 2026-08-14 15:42
svenmy suspicion is something dumb in usb4net that's only triggered on apple platforms but works fine on intel now#asahi-dev · 2026-08-14 15:41
svenuhh.. okay.. that's quite bad#asahi-dev · 2026-08-14 15:36
svenif one of the two linux paths is bottlenecked that makes sense#asahi-dev · 2026-08-14 15:35
svenbut i'm not gonna complain!#asahi-dev · 2026-08-14 15:35
svenhuh, that's unexpected :D#asahi-dev · 2026-08-14 15:31
sven(or vice versa)#asahi-dev · 2026-08-14 15:31
svenjannau: he had 20 Gbps for mac->linux but 7.something for linux->mac on the same connection#asahi-dev · 2026-08-14 15:30
sveni don't think there's anything in the trace that will help speed at this point, except for lane bonding#asahi-dev · 2026-08-14 15:28
svenchaos' had one example where RX was >2x faster than TX so there's obviously something terribly inefficient in one of the two paths#asahi-dev · 2026-08-14 15:27
svenmight also not be all that helpful, once v1 is on the list i'll look into what slows us down#asahi-dev · 2026-08-14 15:26
svenon m2? not yet#asahi-dev · 2026-08-14 15:20
svenalso needs a thunderbolt device or another machine with tbt support and a tbt cable#asahi-dev · 2026-08-14 15:19
sveni'm sure chaos or j_anne will get to it#asahi-dev · 2026-08-14 15:04
svenshould give a line like this for ever plug: VSE WRITE [0x36] = 0x00000407 | hpm1: STATUS=fd820210 DATA_STATUS=03000304 USB4_STATUS=? INTEL_VID_STATUS=030000000002010100 DP_SID_STATUS=? CF_VID_STATUS=?#asahi-dev · 2026-08-14 15:04
svencan someone try that with a pro/max machine?#asahi-dev · 2026-08-14 15:03
svenhttps://f.svpe.de/1e082eb1b38c8d840f4f1daf53f88ca5ecf68fb350381d9fe37f9fc5c1d85e99_trace_tbt_cable.py.txt okay, that should give us the cd3217 regs mac os read + what it wrote to that cable VSE, only tested on t8103 #asahi-dev · 2026-08-14 13:17
svenmaybe it's time to write a tracer for tipd + the VSE write and just collect a bunch of data#asahi-dev · 2026-08-14 13:17
svenffs.. the cable vse thing is still somehow wrong it seems *sigh*#asahi-dev · 2026-08-14 12:01
svenack#asahi-dev · 2026-08-14 12:01
sveni wanted a 20gbps vs. a 40 gbps from the linux side :D#asahi-dev · 2026-08-14 12:01
svenoh lol, that explains it :D#asahi-dev · 2026-08-14 11:59
svengonna take a closer look this evening#asahi-dev · 2026-08-14 11:58
svenjannau: and we're back to "those traces look very similar" :(#asahi-dev · 2026-08-14 09:38
svenso it should've been linux initiating the thing but i also saw macos trying to start that state machine#asahi-dev · 2026-08-14 09:38
svenand macos had the correct uuid#asahi-dev · 2026-08-14 09:38
svenjannau: no, with macos/linux that happened#asahi-dev · 2026-08-14 09:36
sventhe weird thing i saw in chaos's trace is that linux uses "lower uuid initiates lane bonding" but both sides tried to do that which is weird#asahi-dev · 2026-08-14 09:35
svenhttps://f.svenpeter.dev/f9237287e3af548f79fb0d8ceaa85c49b7f8f6ed27a36d07d15d6f86d1292b21_trace_acio.py.txt#asahi-dev · 2026-08-14 09:35
svencan you trace both with trace_acio.py? #asahi-dev · 2026-08-14 09:35
svenhuh, interesting#asahi-dev · 2026-08-14 09:34
sveni'd like to get the uid correct though#asahi-dev · 2026-08-14 09:34
sventhough even single lane would be fine for v1 imho#asahi-dev · 2026-08-14 09:33
svengotta figure out why it doesn't want to bond lanes then#asahi-dev · 2026-08-14 09:32
sventhen chaos' ~20 Gb/s is quite amazing#asahi-dev · 2026-08-14 09:19
svendoes it show if lane bonding works? that's still unclear