#asahi-dev · 2026-01-04 23:57
jannauGPU driver on simpledrm isn't a great experience. glxgears runs with 24 fps#asahi-dev · 2026-01-04 23:52
jannauit was f6c4bdd04b8d ("arm64: dts: apple: t8103: Add missing ps_pmp dependency to ps_gfx")#asahi-dev · 2026-01-04 23:16
jannaubits/210-gpu from asahi-wip on 6.18.3 crashes agx in the same way. we increased the stack and we have a couple of rtkit changes. the latter shouldn't make a difference#asahi-dev · 2026-01-04 22:41
svenand with one of my usb3+dp hubs here at home as well#asahi-dev · 2026-01-04 22:41
svenlol, yeah, at ccc#asahi-dev · 2026-01-04 22:40
jannauadd yourself. you hit that as well#asahi-dev · 2026-01-04 22:37
svenwho all reported hitting the WARN_ON_ONCE in drivers/phy/apple/atc.c:2145 and wants a reported-by in the commit fixing those? i remember at least two different people running into that#asahi-dev · 2026-01-04 22:24
sven(well, disconnects / reconnects actually)#asahi-dev · 2026-01-04 22:24
svenmaybe i still have my slightly broken typec -> a adapter that can trigger *very* quick role switches#asahi-dev · 2026-01-04 22:23
sveni think dwc3_apple doesn't ignore those because the original role switch hack assumed host -> host meant the cable was disconnected inbetween#asahi-dev · 2026-01-04 22:23
jannauit's also what the core dwc3 role switch dowes#asahi-dev · 2026-01-04 22:22
svensounds good to me#asahi-dev · 2026-01-04 22:22
jannauwe also should ignore noop usb role switches in the dwc3_apple. I think that's the better alternative than preventing those in tipd like I did for asahi-wip#asahi-dev · 2026-01-04 22:20
sventbt still hates me though :(#asahi-dev · 2026-01-04 22:20
svenand then we can keep the debounce timeout at 0.5s#asahi-dev · 2026-01-04 22:19
sveni'll add it again and that should get rid of some of the warnings we see#asahi-dev · 2026-01-04 22:19
jannauseems to be something that broke in my v6.19-rc1 branch or a dependency on other bits branches I'm not aware of#asahi-dev · 2026-01-04 22:19
svenit's useful for devices that are very slow at switching from usb3 to usb3+dp#asahi-dev · 2026-01-04 22:19
svenlooks like the "allow mode switches while pipehandler up" code we used to have inside atcphy was actually there for a reason :D#asahi-dev · 2026-01-04 21:32
jannaufl0_id: yes, the acntrdir handling should probably be done via midr_part_features for all m3 devices in chickens.c. I started typing that but haven't pushed anything because I don't yet know what one of the bits do#asahi-dev · 2026-01-04 21:27
svenmy tbt code is back to resetting the SoC after 5 seconds as well :D#asahi-dev · 2026-01-04 21:27
chaos_princessnot recently, the last time was when i was touching that pr. does the initdata match or?#asahi-dev · 2026-01-04 21:26
jannauI guess I have too many moving parts. I'll see that I can tests less changes at once#asahi-dev · 2026-01-04 21:24
jannauchaos_princess: did you try starlight recently? agx crashes for me during initialization with my v6.19-rc1 standalone bits/210-gpu branch. I can't see anything interesting when comparing a trace with asahi-wip. pointers differ but look consistent#asahi-dev · 2026-01-04 21:18
fl0_idhttps://github.com/jannau/m1n1/commit/80381504ed905dadc4eaa5f8369410b0aad73b1a#asahi-dev · 2026-01-04 21:18
fl0_idjannau hey I found this chicken el1/el2 if statement on your github. prob have to do sth like that again. Do you remember if there was any reason why it was not used / not done this way? (just don't want to propose stuff that's already been discarded for some reason) Oh and I assume it should go in chickens_everest I assume#asahi-dev · 2026-01-04 20:54
jannauthe latter mostly because I started in the wrong direction and at the same time wanted to change too much#asahi-dev · 2026-01-04 20:53
jannauI have (trivial) patches which add aux-bridge and aux-hpd-bridge to atc-phy and tipd but are stuck on the dcp side#asahi-dev · 2026-01-04 20:51
jannauseems it wasn't (just) devicetree objections#asahi-dev · 2026-01-04 20:50
jannaurelevant mails: https://lore.kernel.org/dri-devel/20230903214150.2877023-1-dmitry.baryshkov@linaro.org/ https://lore.kernel.org/dri-devel/20231025223027.943563-1-dmitry.baryshkov@linaro.org/#asahi-dev · 2026-01-04 20:32
rosalyntg_(realizing what I said is pretty much the same as what you said. And that's what qualcomm tried and got NAK'd for?)#asahi-dev · 2026-01-04 20:30
rosalyntg_jannu: I'll look into what the qualcomm folks ended up doing...#asahi-dev · 2026-01-04 20:29
rosalyntg_I mean my thought was to have 2 connectors, have DT nodes for them, and then put references to those connector nodes in the displayport = property#asahi-dev · 2026-01-04 20:27
jannauwith the current setup you can't have more than one connector per dcpext. you wouldn't know for which port the hotplug event is. you would have to model 2 things in the device tree which are directly connected two each type-c port and use those as references in the "displayport" property#asahi-dev · 2026-01-04 20:20
jannaurosalyntg_: hej, I was afk. the issue with the current setup with the oob_hotplug event is that the devicetree maintainers don't like it. they afaik nak-ed for qcom x1 elite systems#asahi-dev · 2026-01-04 19:48
Glanzmannsven: Yes, please. :-)#asahi-dev · 2026-01-04 19:34
Eighth_Doctorfrom an old job, maybe one day I can just throw vanilla Fedora KDE on it and everything will work :)#asahi-dev · 2026-01-04 19:33
Eighth_Doctoraditya: while I'm primarily on Matrix as opposed to other places, I'm happy to see T2 Linux working on upstreaming their stuff, I have a T2 Mac sitting around collecting dust as well :P#asahi-dev · 2026-01-04 19:19
Shizmaz: oh useful#asahi-dev · 2026-01-04 18:50
svenwant me to put Reported-by: Thomas Glanzmann <thomas@glanzmann.de> in there?#asahi-dev · 2026-01-04 18:49
svenGlanzmann: yup, that's gonna fix the oops you saw as well#asahi-dev · 2026-01-04 18:36
rosalyntg_sounds good, thanks!#asahi-dev · 2026-01-04 18:33
svenbut jannau really is the right person to talk to, he knows a lot more about drm and dcp than I do#asahi-dev · 2026-01-04 18:32
sventhere's a HPD bit somewhere in the apple variant that is also correct if you have some usb3+dp hub fwiw#asahi-dev · 2026-01-04 18:32
rosalyntg_unfortunately i can't see any device trees that actually use that fwiw--searching for `displayport =` only gives stuff in apple dts....#asahi-dev · 2026-01-04 18:31
rosalyntg_i see, hence direclty calling the oob hp. i wonder why jannau considered it a hack#asahi-dev · 2026-01-04 18:31
svenbut the altmode driver wants to do the negotiation itself iirc#asahi-dev · 2026-01-04 18:31
sveni think we can recover some of the VDOs#asahi-dev · 2026-01-04 18:30
rosalyntg_gotcha. I was kinda working if the status_rx field of tps6598x_dp_sid_status_reg was the vdo considering the vdo is at the same offset for the tbt one, and the dp altmode driver takes that vdo and assings to a `status` field, but that was nothing more than blind conjectore....#asahi-dev · 2026-01-04 18:29
svenI didn't reverse engineering the entire firmware so maybe those are still there but the original TI commands certainly don't exist