#asahi-dev · 2026-08-13 17:14
svenlooks much better#asahi-dev · 2026-08-13 17:11
sveni'm about to grab dinner, so mac os can take its time booting and getting tbt up and running :D#asahi-dev · 2026-08-13 17:10
svenRafay: tbt_data.enter_vdo = (u32)le16_to_cpu(st->intel_vid_status.enter_vdo) << 16; <-- with that enter_vdo 0x3 cannot be possible#asahi-dev · 2026-08-13 17:09
svenhuh#asahi-dev · 2026-08-13 17:09
svenyes#asahi-dev · 2026-08-13 17:09
svenack#asahi-dev · 2026-08-13 17:08
svencan you also grab an atcphy trace from macos?#asahi-dev · 2026-08-13 17:08
svenchaos_princess: latest trace looks like an unbonded link, so it's not lane bonding at fault :(#asahi-dev · 2026-08-13 17:07
sveni don't see how enter_vdo can every contain anything in the lowest 16bits after that one#asahi-dev · 2026-08-13 17:05
svenare you sure you applied that patch?#asahi-dev · 2026-08-13 17:02
svenah, but /proc/interrupts showed they never fired so nvm#asahi-dev · 2026-08-13 17:00
sveni just realized we wouldn't see interrupts happening because the tracer has verbose=1#asahi-dev · 2026-08-13 16:54
svencan you grab a trace to confirm nothing bonded this time? maybe together with that atcphy trace#asahi-dev · 2026-08-13 16:51
svenand while you're at it maybe also trace atcphy both linux and macos to see if i messed up fuses or tunables#asahi-dev · 2026-08-13 16:48
svenmaybe set it to false here already https://elixir.bootlin.com/linux/v7.1.8/source/drivers/thunderbolt/xdomain.c#L1940 though i don't quite see how that should make a difference#asahi-dev · 2026-08-13 16:46
svenoh, weird#asahi-dev · 2026-08-13 16:45
svenwas that run just with slw=1 or also with that forced else path?#asahi-dev · 2026-08-13 16:44
svenyeah, i think the tracer only looks at ring0#asahi-dev · 2026-08-13 16:30
svenyeah, worth a look#asahi-dev · 2026-08-13 16:29
svenmust be somewhere outside of acio then, because that MMIO traces contains those and iirc i compared them and they looked the same#asahi-dev · 2026-08-13 16:27
sveni'm essentially out of ideas at this point :(#asahi-dev · 2026-08-13 16:27
svenwith only slw#asahi-dev · 2026-08-13 16:26
svencan you grab a trace again?#asahi-dev · 2026-08-13 16:19
sventhought you already found it, just set slw and tlw in that LINK_STATE_STATUS_REQUEST to 1#asahi-dev · 2026-08-13 16:10
svenand in the trace it's also trying to initiate that link state change request. so maybe really do both sides to be safe#asahi-dev · 2026-08-13 16:09
svenguess i should also figure out where that actually comes from#asahi-dev · 2026-08-13 16:09
svenhuh, your mac also has a real uuid, the linux driver currently has a fallback one. #asahi-dev · 2026-08-13 16:03
svenor maybe just do both ends to be safe :D#asahi-dev · 2026-08-13 16:03
svenokay, then forcing that else branch should work to skip the states that would eventually initiate bonding#asahi-dev · 2026-08-13 16:02
svener wait.. we're looking at different sides, aren't we?. let me check which side actually initiates the bonding#asahi-dev · 2026-08-13 16:01
svenyeah, that's one state before it reaches your line i think#asahi-dev · 2026-08-13 16:00
svenhttps://elixir.bootlin.com/linux/v7.1.8/source/drivers/thunderbolt/xdomain.c#L1619 or maybe force the else branch there to skip the entire state machine#asahi-dev · 2026-08-13 15:58
sventb_xdomain_lane_bonding_enable is probably too late alreayd#asahi-dev · 2026-08-13 15:58
svenhrm, okay. so maybe it has to be disabled much earlier. let me try to find where it decides to lane bonding#asahi-dev · 2026-08-13 15:56
svenhrm... but didn't something fail with -ENODEV then?#asahi-dev · 2026-08-13 15:53
svenlast attempt was -ENODEV at the beginning of something in xdomain.c, wasnt it?#asahi-dev · 2026-08-13 15:53
svenyeah, let's see how to break that without breaking the entire connection :D#asahi-dev · 2026-08-13 15:53
svenoh right, that didn't work out because my first attempt was 5 minutes before i went to bed :D#asahi-dev · 2026-08-13 15:52
svenah :/#asahi-dev · 2026-08-13 15:51
svenwhat if you try the other way around? so pings from macos and wireshark on linux?#asahi-dev · 2026-08-13 15:49
svenhuh#asahi-dev · 2026-08-13 15:37
svenbecause the traces look very very similar, it's just that one them doesn't work.#asahi-dev · 2026-08-13 15:36
sveni'm wondering if it's some weird "this controller only updates state when enable is set" or something similar#asahi-dev · 2026-08-13 15:35
svenit might just break everything, but given that it's already not working it can't hurt#asahi-dev · 2026-08-13 15:35
svenclears and then sets the enable bit for ring1 #asahi-dev · 2026-08-13 15:33
svenwrite32(0xb01f14010, tx); write32(0xb01f84010, rx)#asahi-dev · 2026-08-13 15:33
svenrx = read32(0xb01f84010)#asahi-dev · 2026-08-13 15:33
svenwrite32(0xb01f14010, tx & ~0x80000000); write32(0xb01f84010, rx & ~0x80000000)