#asahi-dev · 2026-05-28 09:00
sofuschaos_princess: no, but i think at some point (when everything works), offloading work to the cm3 is a good idea. And i dont think the instruction stream generation belongs in the kernel#asahi-dev · 2026-05-28 08:55
chaos_princesssofus: yea, that's kinda what i meant by "just proxy the irq", but are you trying to put a bunch of logic into the new cm3 firmware that parses the v4l structs and feeds the hw?#asahi-dev · 2026-05-28 08:53
mazFWIW, I have a patch to allow enabling debugusb from the CS interface. gives you two uarts, just in case! ;-)#asahi-dev · 2026-05-28 08:53
sofusim talking about writing our own firmware for the cm3 if it was not clear#asahi-dev · 2026-05-28 08:52
mazjannau: ah, cool. thanks for letting me know.#asahi-dev · 2026-05-28 08:50
jannaumaz: uart over SBU works. over DATA should work as well since the port controller does the muxing#asahi-dev · 2026-05-28 08:49
sofuschaos_princess: thats what im doing right now#asahi-dev · 2026-05-28 08:49
sofuschaos_princess: im not sure#asahi-dev · 2026-05-28 08:49
chaos_princessor if you can't - just proxy the irq and that's it#asahi-dev · 2026-05-28 08:49
sofusbut yes, everything i have done is based on her work#asahi-dev · 2026-05-28 08:49
chaos_princesscan't you reconfigure the aic to get the interrupts to ap cores?#asahi-dev · 2026-05-28 08:48
sofusKinda, we kinda need it for irq#asahi-dev · 2026-05-28 08:48
JamesCalligeros[m]eiln was going to ignore the m3 interface and talk directly to the actual decoder blocks wasnt she?#asahi-dev · 2026-05-28 08:47
chaos_princesswasn't the plan to ignore the cm3?#asahi-dev · 2026-05-28 08:47
sofuswhich seems possible (at least for h264)#asahi-dev · 2026-05-28 08:47
sofuschaos_princess: i get that. My hope is that the driver esentially copies v4l2's structs to the cm3#asahi-dev · 2026-05-28 08:42
snooze6214[m]jannau: ah I see, thanks#asahi-dev · 2026-05-28 08:38
mazjannau: does CS work on neo? if it doesn't, I'd better capture that in the doc.#asahi-dev · 2026-05-28 08:36
chaos_princessseconding "some driver is better than none", but an advantage of vkvideo/vaapi is that a significant chunk of your driver can live in userspace instead of in the kernel.#asahi-dev · 2026-05-28 08:34
sofusOkay, thanks#asahi-dev · 2026-05-28 08:28
jannauI'd continue for now, a working driver is better than none. From the working driver either support can be added (there are other platforms with v4l2 drivers) or it can be used as base to rewrite the driver#asahi-dev · 2026-05-28 08:05
sofusI have searched, it just seemed like v4l2 was not prefered#asahi-dev · 2026-05-28 08:00
JamesCalligeros[m]Well it’s either v4l2 or vulkan video. Search the logs for previous discussions on this#asahi-dev · 2026-05-28 07:59
sofusIs v4l2 okay? Im not sure how much support it has, firefox does not seem to support it...#asahi-dev · 2026-05-28 07:52
jannausnooze6214[m]: usb device mode doesn't work on the neo since there's a usb2 hub (there ought a way to bypass the hub but we don't know how)#asahi-dev · 2026-05-28 07:49
JamesCalligeros[m]Nice#asahi-dev · 2026-05-28 07:48
sofusIm writing one#asahi-dev · 2026-05-28 07:48
JamesCalligeros[m]Do you have an avd v4l2 driver?#asahi-dev · 2026-05-28 07:43
sofusI decided to try v4l2 for avd. It scores 57/135 on the JVT-AVC_V1 test suite, which i think is okay compared to the 76/135 macos gets#asahi-dev · 2026-05-28 07:22
StanfordAppliedCyber[m]not to mention every other SPTM enabled device#asahi-dev · 2026-05-28 07:19
StanfordAppliedCyber[m]i'm almost to userspace#asahi-dev · 2026-05-28 07:18
StanfordAppliedCyber[m]also i'm doing a bunch of dev work on the macbook neo right now, tl;dr i'm trying to port the m1n1 hypervisor to the macbook neo by emulating SPTM. my emulator is tainted and thus can't be upstreamed (with the exception of the UAT endpoints which are completely clean) but i'm committed to writing very nice clean room docs and finding a way to get it upstreamed so that others can start REing the neo#asahi-dev · 2026-05-28 07:16
snooze6214[m]ykw valid lol, thanks#asahi-dev · 2026-05-28 07:16
StanfordAppliedCyber[m]specifically the macvdmtool readme#asahi-dev · 2026-05-28 07:16
StanfordAppliedCyber[m]..........I don't want to be mean but please RTFM#asahi-dev · 2026-05-28 07:15
snooze6214[m]need another apple silicon mac? did you use macvdmtool? Stanford Applied Cyber #asahi-dev · 2026-05-28 07:14
StanfordAppliedCyber[m]it's a pain, and you need another apple silcion mac and a very nice usb cable#asahi-dev · 2026-05-28 07:14
StanfordAppliedCyber[m]yeah i did#asahi-dev · 2026-05-28 07:11
snooze6214[m]Hey, has anyone gotten m1n1 to run in proxy mode on the macbook neo? Mine says that it's running in proxy mode but it does not appear as a serial device when connecting to my linux host using a usb C to usb A cable.#asahi-dev · 2026-05-27 22:13
chaos_princesssame#asahi-dev · 2026-05-27 22:12
fl0_dchaos_princess, m3 max, with yuka#asahi-dev · 2026-05-27 21:43
StanfordAppliedCyber[m]one consequence of this is that the emulator will only work for one macos version-- not too different from the previous hypervisor, but new devices will require more bringup work than previously because of it assuming this is the method we converge to for m1n1 hypervisor#asahi-dev · 2026-05-27 21:42
StanfordAppliedCyber[m]facepalm#asahi-dev · 2026-05-27 21:41
StanfordAppliedCyber[m]i was using an old firmware revision by accident#asahi-dev · 2026-05-27 21:41
StanfordAppliedCyber[m]....well i found the source of the problem i had earlier, tl;dr 26.5 changed literally every dart endpoint from 26.4.1#asahi-dev · 2026-05-27 17:37
jskah I think I found the linux branch as well. thank you!#asahi-dev · 2026-05-27 17:36
jskthanks! found the m1n1 branch by mildsunrise (or is there a `linux` branch as well?). which experiment in m1n1 were you talking about though?#asahi-dev · 2026-05-27 17:31
yukawith the new spmi features implemented in linux in some capacity#asahi-dev · 2026-05-27 17:30
yukaThere is also a branch by mildsunrise#asahi-dev · 2026-05-27 17:10
chaos_princessnot for usb itself, for the pd controller. protocol is supposed to be same/similar, just over spmi now. there is a m1n1 experiment that extends the existing spmi driver with features used by pd controller.