#asahi-dev · 2026-01-08 23:27
fl0_id least clarify the help text? Or is my kernel build wrong because it should always create Image.gz#asahi-dev · 2026-01-08 23:27
fl0_idfor run_guest_kernel.sh, is it possible that this detection of gzipped Image is kinda wrong? As I read that script, if a kernel argument is supplied, (but no -k option) it sets kernel to <path>/Image.gz. In my kernel tree, kernel build never gzips the image (or well it does but calls it vmlinuz) The proyclient/tools script only gzips it if the -k option is used iirc. If my understanding is correct, is it ok to improve this handling or to at#asahi-dev · 2026-01-08 23:17
sven:-(#asahi-dev · 2026-01-08 23:07
jannauusb3/dp-altmode hub wasn't fully detected. usb storage in the next test was#asahi-dev · 2026-01-08 23:06
jannausigh, coldplug on m2 mini is still unreliable. I was mostly testing initial hotplug#asahi-dev · 2026-01-08 22:36
kettenisI'll probably skip rebasing the asahi branch on top of v2026.01 though#asahi-dev · 2026-01-08 22:33
kettenisI've found the breaking commit and will send a fix upstream#asahi-dev · 2026-01-08 22:29
svendwc2 must be truly horrible then #asahi-dev · 2026-01-08 22:29
kettenison a different note, u-boot v2026.01 is broken on Apple Silicon#asahi-dev · 2026-01-08 22:28
kettenistruste me, dwc3 is a pleasure to work with compared to dwc2#asahi-dev · 2026-01-08 22:23
svenluckily, we’re dealing with the horrible new DesignWare USB3 controller!#asahi-dev · 2026-01-08 22:23
kettenisthat's the horrible old DesignWare USB2 controller #asahi-dev · 2026-01-08 22:22
svenugh, yes, that should’ve been dwc3#asahi-dev · 2026-01-08 22:22
kettenissven: you probably shouldn't talk about dwc2 in your commit message#asahi-dev · 2026-01-08 20:07
svenand then all the random stuff that will go wrong once this actually hits our release kernel 😨#asahi-dev · 2026-01-08 20:06
svenalright, so that leaves the harmless WARN because pipehandler is already up and those harmless "pipehandler not locked" messages during unplug then #asahi-dev · 2026-01-08 18:04
jannaulocal avd development should be at least safe unlike va-api on intel where you can easily crash the gpu by using va-api wrongly#asahi-dev · 2026-01-08 17:43
leioespecially with the only asahi-capable one being my main workstation too#asahi-dev · 2026-01-08 17:42
leioreally uncomfortable for me to use multiple machines#asahi-dev · 2026-01-08 17:42
leioit would be very motivating for me to poke at it too if I were able to just use this same machine for the dev by not needing it sitting in m1n1#asahi-dev · 2026-01-08 17:36
jannaualso a way to get familiar with the code#asahi-dev · 2026-01-08 17:36
jannauI considered at some point to port the m1n1 + python to vfio-platform + python. mostly because I'm not quite sure where the kernel / userspace break should be and how the interface is going to look#asahi-dev · 2026-01-08 17:15
leiocompared to the status quo#asahi-dev · 2026-01-08 17:15
leioyeah, the biggest win would of course be simply not doing software decoding overall :)#asahi-dev · 2026-01-08 17:14
vupI guess the craziest option would be custom firmware for the cm3#asahi-dev · 2026-01-08 17:14
vupright, but also it just does not actually do that much, my assumption would be the actualy video decoding is the power hungry part#asahi-dev · 2026-01-08 17:12
leiothe latter probably is not relevant for us#asahi-dev · 2026-01-08 17:12
leiothose are the likely reasons apple uses it#asahi-dev · 2026-01-08 17:12
leiothe cm3 probably uses a lot less power, and it can probably be configured to read protected (decrypted) memory (DRM support)#asahi-dev · 2026-01-08 17:08
vupabsolutely. In general I think there is the question of wether to use the "helper" (cortex) m3 and the apple firmware for it, or the lower level stuff that eiln re'd. I think the using the lower level ones and ignoring the cm3 sounds pretty compelling. #asahi-dev · 2026-01-08 16:58
leioit would probably be useful if the command stream and co stuff would end up as some sort of a documentation beyond a few early blog posts#asahi-dev · 2026-01-08 16:57
leioI also find some talks about the stateless stuff with eiln at https://oftc.catirclogs.org/asahi-dev/2023-09-16# but that was early on, things might have changed with new discoveries or something#asahi-dev · 2026-01-08 16:57
leiosome past discussion was in https://oftc.catirclogs.org/asahi-dev/2025-10-07##asahi-dev · 2026-01-08 16:49
vupmakes sense#asahi-dev · 2026-01-08 16:48
leiothe command stream and such things are of interest in there#asahi-dev · 2026-01-08 16:48
leioto take care of and we won't need to worry about that much, but still#asahi-dev · 2026-01-08 16:48
leiovup: I would also be careful about looking at some of the code - I would stay away from looking at the code borrowed from ffmpeg and elsewhere that is with an unsuitable license (mostly LGPL, things about SPS/PPS parsing in there, etc), as if we end up with a vulkan driver, we want it MIT licensed. Most of that code is probably unnecessary too if we can do vulkan video, as that side in my reading and understanding appears to be the job of the vulkan application #asahi-dev · 2026-01-08 16:48
vupyeah that was my short term plan, I have a m1 mpb I can play around with lying around#asahi-dev · 2026-01-08 16:46
leiovup: I would suggest to start with trying to replicate the m1n1 based experiments and get it decoding in m1n1; this will need two machines, the M1 or M2 to sit in m1n1 and do the hardware decode and another that then shows what it decoded. I think the experiment only works on either M1 or M2 (what eiln had) and may need some address tweaks for the other.#asahi-dev · 2026-01-08 16:44
leiovup: I made the mistake of starting from that end, but really should start with the AVD side of it, so you'd even be able to tell if vulkan video would work (just from the basis of stateless or not, to start with). But if it's interesting read, why not :)#asahi-dev · 2026-01-08 16:35
vupjust been reading eilns blogposts and trying to figure out how vulkan video even works :p#asahi-dev · 2026-01-08 16:33
vupno no :)#asahi-dev · 2026-01-08 16:32
leiovup: did we scare you off completely? ;p#asahi-dev · 2026-01-08 16:25
svenjannau: i see the pipehandler lock message during unplug as well fwiw, i think we can just drop the pipehandler dance there. i think it's only required for that weird BIST sequence anyway#asahi-dev · 2026-01-08 14:40
kettenisif there are good technical reasons for changes those are worth considering#asahi-dev · 2026-01-08 14:39
ketteniswell, that is a different matter#asahi-dev · 2026-01-08 14:37
Eighth_Doctorkettenis: I think we'd prefer it if they didn't change too, but they usually get changed due to Linux review requirements#asahi-dev · 2026-01-08 13:37
kettenisand not having a working keyboard in U-Boot will be painful#asahi-dev · 2026-01-08 13:36
kettenisthe current bindings are also in use by U-Boot and OpenBSD#asahi-dev · 2026-01-08 13:36
ketteniscan we please make no arbitrary changes to the bindings?