#asahi-dev · 2026-01-08 12:52
svenmaybe that's not true, no idea#asahi-dev · 2026-01-08 12:52
svenessentially what i'm thinking is that MTP, the co-processor, exposes the dockchannel fifo interfaces. so that fifo is really part of the co-proc hw#asahi-dev · 2026-01-08 12:48
svenmaybe it doesn't make sense#asahi-dev · 2026-01-08 12:48
sveni'm mainly trying to avoid apple, phandles ;)#asahi-dev · 2026-01-08 12:47
chaos_princessthat would be harder#asahi-dev · 2026-01-08 12:47
integralpiloton the HID node we could have apple,mtp-asc and apple,mtp-ram maybe?#asahi-dev · 2026-01-08 12:46
sveni just thought about it because the mtp node doesn't seem to do much and because have this apple,mtp thing in the other node#asahi-dev · 2026-01-08 12:46
sventhat's why i'd like to see how the devicetree now looks like#asahi-dev · 2026-01-08 12:46
sveni'm not sure it's a good idea :D#asahi-dev · 2026-01-08 12:43
integralpilotthough it's like from a completely different section on the soc, and does it look bad to mix like widly different addresses in regs?#asahi-dev · 2026-01-08 12:43
svengreat! i'll take a closer look then #asahi-dev · 2026-01-08 12:43
integralpilotthe MTP nodes could be merged, maybe just as more regs on the HID device?#asahi-dev · 2026-01-08 12:42
integralpilotthanks for taking a look, yeah i've run the tools. i can publish the branch tomorrow. #asahi-dev · 2026-01-08 12:42
chaos_princessiirc nothing else#asahi-dev · 2026-01-08 12:40
svendoes mtp do anything that's not dockchannel or could we merge those nodes as well?#asahi-dev · 2026-01-08 12:39
chaos_princessalso i assume you ran all the things that check the dtb and schemas#asahi-dev · 2026-01-08 12:38
svenbut from a quick glance they look quite good#asahi-dev · 2026-01-08 12:34
svenideally with the changes to the dts files#asahi-dev · 2026-01-08 12:34
svenat the dt-bindings i mean#asahi-dev · 2026-01-08 12:34
svencan you push the entire branch somewhere? i can take a look either this evening or tomorrow then#asahi-dev · 2026-01-08 12:33
integralpilotyeah, i'm not sure either#asahi-dev · 2026-01-08 12:33
nickchanit's supposedly a replacement for the S5L uart#asahi-dev · 2026-01-08 12:33
nickchandockchannel is also used as debug console but not sure how it works#asahi-dev · 2026-01-08 12:30
integralpilotthanks for letting me know, i won't in future#asahi-dev · 2026-01-08 12:29
chaos_princessno, do not include anything you do not need in the dt#asahi-dev · 2026-01-08 12:28
integralpiloti'll remove the generic compatibles, thanks#asahi-dev · 2026-01-08 12:28
integralpilotwe don't use the remote registers yet, so i guess they can be removed. i thought it was best to have them documented in case they are needed later but i guess it won't be liked if they are redundant #asahi-dev · 2026-01-08 12:27
chaos_princessremove the "generic" compatibles, use one of the machine-specific compatibles as the base one#asahi-dev · 2026-01-08 12:25
chaos_princessdo we use the remote registers?#asahi-dev · 2026-01-08 12:22
integralpilothttps://gist.github.com/IntegralPilot/5a8b20a7e962d19e973eb00121984e73#asahi-dev · 2026-01-08 12:19
integralpilothere's the link:#asahi-dev · 2026-01-08 12:19
integralpiloti've made DT bindings, and if as suggested someone could take a look and make sure they are right i would really appreicate that.#asahi-dev · 2026-01-08 12:19
integralpiloti think for the first patch series it's best to just add keyboard support and not have firmware and GPIO and everything needed for trackpads so we can merge the core protocol first without overwhelming maintainers#asahi-dev · 2026-01-08 12:18
integralpilotFINALLY finished MTP yay#asahi-dev · 2026-01-08 10:24
integralpiloti guess not, i'll leave it out for now#asahi-dev · 2026-01-08 09:43
integralpilotprobably sending SMC power v2 tonight, does anyone have a preference on emergency_sync() vs no emergency_sync()?#asahi-dev · 2026-01-08 02:09
integralpilotchecked and looks like when orderly_poweroff(true) fails and tries to force power off, it emergency syncs before the kernel_power_off()#asahi-dev · 2026-01-08 02:06
integralpiloti'm just asking because i'm making a v2 for SMC power supply and thought of this while looking through the emergency code#asahi-dev · 2026-01-08 02:05
integralpilotthe 10 seconds power button code has the comment "make sure the NVMe cache is flushed" but doesn't actually do it and just powers off, unless kernel_power_off() also flushes the cache which it doesn't look like#asahi-dev · 2026-01-08 02:02
integralpilotdoesn't look like it, both just kernel_power_off()#asahi-dev · 2026-01-08 02:01
JamesCalligeros[m]arent we already flushing disk caches in emergency scenarios? #asahi-dev · 2026-01-08 01:59
integralpilotmight save a bit more user data and prevent fs corruption but not sure#asahi-dev · 2026-01-08 01:58
integralpiloti guess it depends on if we have time to emergency_sync(), in the power button scenario we have ~4 seconds before SMC kills us, probably less in the case of voltage below design minimum#asahi-dev · 2026-01-08 01:57
integralpiloti've been doing some research on what to do in emergencies, and maybe for critical voltage and power button hold down for 10 seconds, should we emergency_sync() before kernel_power_off()? both currently just kernel_power_off()#asahi-dev · 2026-01-07 21:28
leiootherwise the only option might be stateful v4l; or VA-API if you are into "fun" like trying to make frame-based stuff work with slice-based API that VA-API is.#asahi-dev · 2026-01-07 21:26
leioI believe userspace will pick things up with vulkan video, we shouldn't worry about that for now; main question is - can we do it stateless to even have vulkan video be viable option technically.#asahi-dev · 2026-01-07 21:25
leioseparate vulkan driver can be a bit messy, but Vulkan Core (which clarify having Vulkan driver that have no compute) is coming (or already there, this is months old stuff I'm pulling out)#asahi-dev · 2026-01-07 21:24
leiochats with*#asahi-dev · 2026-01-07 21:24
leioif I were to make time to work on this, I would strongly like if someone else did the kernel component, but as it is today, the work is "free" completely, as I didn't get to digging in beyond reading and some v4l2/gst folks.#asahi-dev · 2026-01-07 21:23
leioif vulkan video, it would presumably go together with a drm component for the command stream handling, and then the bulk of it a new vulkan driver, just for AVD (not the same driver as honeykrisp)