#asahi-dev · 2025-08-19 16:34
nickchansven: I think I will change the comment about enabling linear submission queues and nvmmu#asahi-dev · 2025-08-19 15:50
svenand if that's not acceptable we'll just drop the generic ones, actually use t8103-whatever as fallback and do device tree surgery inside m1n1 for a year or two#asahi-dev · 2025-08-19 15:47
kettenisindeed#asahi-dev · 2025-08-19 15:47
svenessentially just pretend that e.g. "apple,nvme-ans2" is actually "apple,t8103-nvme-ans2"#asahi-dev · 2025-08-19 15:46
svenyeah#asahi-dev · 2025-08-19 15:46
jannaubut I see why not when the existing compatible is still compatible#asahi-dev · 2025-08-19 15:45
jannauI would hope the minimal changes (add t6020 SoC variant compatibles to existing dt-bindings) will be accepted. if not I'd have to make also several driver changes#asahi-dev · 2025-08-19 15:44
sventhat also works :D#asahi-dev · 2025-08-19 15:43
kettenisI suppose we always have the option to do a little bit of devicetree surgery in m1n1 if upstream insists on dropping stuff#asahi-dev · 2025-08-19 15:42
sveni'd keep using the ones that are already upstream and just not add any new ones for e.g. usb or whatever else has no bindings so far#asahi-dev · 2025-08-19 15:39
kettenishopefully we can still add the generic compatibles for the non-upstreamed M2 machines where applicable?#asahi-dev · 2025-08-19 15:37
kettenisbut now that u-boot uses some git trickery to get the official upstream device trees those old trees are no longer useful#asahi-dev · 2025-08-19 15:36
kettenisthey were added to make a maintainer happy who insisted on having a device tree for all boards#asahi-dev · 2025-08-19 15:36
svenso anyway, we can't fix the generic compatibles for most devices without a lot of pain. so imho we just leave them as-is and only use soc-specific ones moving forward#asahi-dev · 2025-08-19 15:36
nickchanthat one comes from corellium i think#asahi-dev · 2025-08-19 15:35
kettenisI should remove those device trees from u-boot#asahi-dev · 2025-08-19 15:35
kettenisprobably#asahi-dev · 2025-08-19 15:35
sventhe upstream binding at least doesn't have "apple,dart", maybe leftovers from the very early days before that was upstream?#asahi-dev · 2025-08-19 15:33
kettenisI found one in the t8103.dtsi that is in the u-boot repo (but that file isn't actually used)#asahi-dev · 2025-08-19 15:33
sventhe current "problem" is apple,nvme-ans2#asahi-dev · 2025-08-19 15:32
sveni think apple,dart doesn't even exist, that was just an example i used#asahi-dev · 2025-08-19 15:31
kettenisI never used "apple,dart" in U-Boot or OpenBSD so quietly dropping that one is ok from my perspective#asahi-dev · 2025-08-19 15:25
svenright, so the current situation sucks but we can't really fix it unless someone here has a time machine ;)#asahi-dev · 2025-08-19 15:24
kettenisthe only "problem" we have is that we started somewhere in the middle now that nickchan is adding support for older hardware#asahi-dev · 2025-08-19 15:24
sventhat generic thing might work for stuff like i2c but anything more complex seems to change every now and then #asahi-dev · 2025-08-19 15:23
svenyeah, i would only stop adding stuff like „apple,dart“ #asahi-dev · 2025-08-19 15:23
kettenisa compatible for the SoC itself and a compatible for the oldest SoC it is compatible with#asahi-dev · 2025-08-19 15:23
kettenisthat is a fairly typical pattern#asahi-dev · 2025-08-19 15:22
jannauwe do not want to tons of compatibles to the drivers. for the t602x series I'll just add t6020 compatibles to the bindings and the device tree gets both. i.e. `compatible = "apple,t6020-dart", "apple,t8110-dart";`#asahi-dev · 2025-08-19 15:19
jannauI wouldn't expect per SoC differences for something low level as DART so using t8110 should be fine. t602x seems to be entirely compatible as well#asahi-dev · 2025-08-19 15:17
nickchansven: there's still some wiggle room for fallback compatibles in the future unless you do mean the driver should match every soc-specific conmpatible#asahi-dev · 2025-08-19 15:14
svenmy point is that it just doesn't matter how and where we messed up in the past. the only thing we can do is to only use tXXX-whatever compatibles with the right XXXX in the future#asahi-dev · 2025-08-19 15:13
nickchani guess so it is probably identical anyways#asahi-dev · 2025-08-19 15:12
svennot like we can fix that at this point#asahi-dev · 2025-08-19 15:11
svenso?#asahi-dev · 2025-08-19 15:11
nickchanalso I think even soc specific compatibles are not completely clean right now, for example the dart t8110 compatible is apple,t8110-dart which is A15 not M2#asahi-dev · 2025-08-19 14:57
svenbut i can reply again on the weekend#asahi-dev · 2025-08-19 14:56
svenkrzysztof didn't comment on that suggestion at all though#asahi-dev · 2025-08-19 14:56
svenyeah, that what i'd do as well#asahi-dev · 2025-08-19 14:56
jannauI'd say leave the existing compatibles as is and use SoC specific ones going forward#asahi-dev · 2025-08-19 14:55
sventhe current status probably sucks but it is what it is, doesn’t help to just complain now but we need to find a way forward #asahi-dev · 2025-08-19 14:55
sveni can try to writer another reply on the weekend, won’t have time before that #asahi-dev · 2025-08-19 14:55
jannauapple's devicetree has up to m4 pro "role = ANS2" whatever role means#asahi-dev · 2025-08-19 14:54
svenI have no idea 🙃#asahi-dev · 2025-08-19 14:54
nickchansven: okay so what happens with the bindings?#asahi-dev · 2025-08-19 14:51
svenhrm, so I guess our original compatible should’ve been ans3 or maybe just t8103-nvme #asahi-dev · 2025-08-19 14:51
nickchanthere are ans3 strings discovered back when m1 is first released#asahi-dev · 2025-08-19 14:50
nickchanso even m1 mentions ans and ans2 #asahi-dev · 2025-08-19 14:50
nickchani believe A9-A10 = ANS (this is a pcie device), T2, A11 - A13 = ANS2 and then A14/M1+ = ANS3 but there are all sorts of remnants from previous generations #asahi-dev · 2025-08-19 14:50
svenkinda weird though