#asahi-dev · 2022-01-21 05:45
svenyeah, just get rid of flags and have a single buffer_request callback that gets that rtk_buffer structure and just fills it however it sees fit and sets a flag for iomem vs. RAM#asahi-dev · 2022-01-20 20:44
svenor I guess even that could be decided in the callback and left to the driver itself#asahi-dev · 2022-01-20 20:44
sventrue. and then just have a special flag for io memory vs. RAM#asahi-dev · 2022-01-20 20:30
svenhm, I think you’ll need the quirk because you’re not supposed to reply to pre-allocated buffers#asahi-dev · 2022-01-20 20:29
svenOr just add a quirk to the flags #asahi-dev · 2022-01-20 20:29
svenhack the dma_alloc callback? :D#asahi-dev · 2022-01-20 20:28
svenugh#asahi-dev · 2022-01-20 12:49
sveni'm mostly just curious because it's rather weird to have mostly separate structures for the two queues but them have them share that invalidation register#asahi-dev · 2022-01-20 12:49
svenyeah, macos always sets the slot_id to the command_id and just uses that single invalidation register#asahi-dev · 2022-01-20 12:30
sveni don't really understand why that TCB structure is separate for the admin and the io queue while they share that single invalidate register#asahi-dev · 2022-01-20 12:29
svenyes#asahi-dev · 2022-01-20 12:25
sventhe only thing that's still strange is that there only is a single register to invalidate an entry in the NVMMU that seems to index into *both* TCB buffers#asahi-dev · 2022-01-20 12:19
svenyeah#asahi-dev · 2022-01-20 12:05
svenlooks like the NVMMU is actually the starting point for nvme commands. if i write A to that linear sq db the controller will look up TCB[A] and take the command_id from there and then look at SQ[TCB[A].command_id]. not that that matters much.#asahi-dev · 2022-01-19 12:51
svenlinux on the M1 ipad would be incredible. don't see that ever happening though :(#asahi-dev · 2022-01-19 12:50
svenyeah, i think hotplug shouldn't be too bad#asahi-dev · 2022-01-19 12:49
svenand thunderbolt then please :-P#asahi-dev · 2022-01-18 22:35
svenit’s not a bot afaik but something he triggers manually #asahi-dev · 2022-01-18 22:35
svenpovik: sorry https://patchwork.ozlabs.org/project/devicetree-bindings/patch/20220118191839.64086-2-povik+lin@protonmail.com/#asahi-dev · 2022-01-18 21:17
svenalyssa: no need to say anything. Just write a patch to fix it 😇#asahi-dev · 2022-01-18 18:39
sventhere are always more yaks to shave ;)#asahi-dev · 2022-01-18 18:39
svenGood night!#asahi-dev · 2022-01-18 18:37
svenIt’s also rather chatty with lots of strings and function names #asahi-dev · 2022-01-18 18:37
svenI think so #asahi-dev · 2022-01-18 18:35
svenANS has a ton of debug commands to do stuff like fault injection and other crap#asahi-dev · 2022-01-18 18:35
svenand a little bit SIO#asahi-dev · 2022-01-18 18:35
svenANS mostly #asahi-dev · 2022-01-18 18:34
svenI’m still a bit sad we can’t easily recover if those things crash :/#asahi-dev · 2022-01-18 18:33
sven*for#asahi-dev · 2022-01-18 18:33
sveni think one die time critical stuff and another one for stuff that can wait#asahi-dev · 2022-01-18 18:33
svenyeah, the other endpoints usually have two handlers#asahi-dev · 2022-01-18 18:32
svenlol. I assumed something was reading that queue :D#asahi-dev · 2022-01-18 18:32
svenyeah, I’ve seen that code#asahi-dev · 2022-01-18 18:30
svenlol #asahi-dev · 2022-01-18 18:28
svenah, nice#asahi-dev · 2022-01-18 16:39
svenand it crashed the DCP on 11.x when no screen was connected because we maybe didn’t use the backwards-compatible iboot protocol #asahi-dev · 2022-01-18 15:56
svenmarcan: https://github.com/AsahiLinux/m1n1/pull/150 alright, that makes DCP work on 11.x (hopefully without breaking it on 12.x)#asahi-dev · 2022-01-18 13:03
svenok, seems to work now if I accept TYPE_NOTIFY instead of TYPE_REPLY as well#asahi-dev · 2022-01-18 12:55
svenif i wasn't hacking random stuff without understanding it, sure :D#asahi-dev · 2022-01-18 12:55
svenotherwise construct wanted to eat more bytes than it had available somewhere#asahi-dev · 2022-01-18 12:54
sveni just had to remove it from the tracer fwiw#asahi-dev · 2022-01-18 12:54
svenright, but the version is in this sub header while the cookies are in the main header which is... weird#asahi-dev · 2022-01-18 12:53
svenyou should be able to reproduce this anyway by just setting version = 3 on 12.x, shouldn't you?#asahi-dev · 2022-01-18 12:52
svenmarcan: replies are TYPE_NOTIFY instead of TYPE_REPLY and rxcookie/txcookie doesn't seem to exist (which doesn't make a whole lot of sense)#asahi-dev · 2022-01-18 08:50
svenif I add timeout to all those busyloops in afk and use version 3 at least DCP doesn't crash anymore though :D#asahi-dev · 2022-01-18 07:20
svenjust hangs then#asahi-dev · 2022-01-18 07:20
svenEPIC: started interface 0 (disp0-service)#asahi-dev · 2022-01-18 07:20
svenEPIC: got unexpected message type 0 during command#asahi-dev · 2022-01-18 07:20
svenEPIC: received unexpected message during init