#asahi-dev · 2026-01-06 08:22
integralpilotdoes only MTP use the helper? I assumed other coprocs did but if it’s just MTP might be better to just move that logic to the MTP driver#asahi-dev · 2026-01-06 07:37
svenit’s just that cpu control register which really should be part of mtp itself like we do for all other coprocs #asahi-dev · 2026-01-06 07:37
svenit just calls into rtkit.c like NVMe as well #asahi-dev · 2026-01-06 07:37
svenrtkit-helper doesn’t even implement anything afaict #asahi-dev · 2026-01-06 07:34
svenwhat’s separate and should remain separate is the mailbox hw itself #asahi-dev · 2026-01-06 07:34
svenAnd it’s only used that way for mtp #asahi-dev · 2026-01-06 07:33
svenit feel like abstraction for the sake of building an abstraction #asahi-dev · 2026-01-06 07:33
svenI would merge the helper stuff so that you don’t have another node in the device tree #asahi-dev · 2026-01-06 06:17
integralpilotoh and i realised the actual reason ANS doesn't use helper/asc stuff is because it has its own SART for memory and doesn't use like the asc SRAM API etc. #asahi-dev · 2026-01-06 06:16
integralpilotyeah, i think so. thanks for your help#asahi-dev · 2026-01-06 06:08
JamesCalligeros[m]Right, that probably is the sanest approach then#asahi-dev · 2026-01-06 06:07
integralpilotANS is already upstream because we have rtkit upstream which is all it needs, not the helper/asc stuff#asahi-dev · 2026-01-06 06:07
integralpiloti think ANS (the NVMe co-proc) uses rtkit but not asc/ascwrap because it's already started#asahi-dev · 2026-01-06 06:04
JamesCalligeros[m]Although doing that makes it easier to version asc independently of rtkit which is the point#asahi-dev · 2026-01-06 06:03
JamesCalligeros[m]Upstream might question that logical distinction if we dont expect rtkit to ever use anything else#asahi-dev · 2026-01-06 06:03
JamesCalligeros[m]Do we have any cases of rtkit not using asc/ascwrap#asahi-dev · 2026-01-06 05:59
integralpiloti'm leaning towards b) because i feel like it's a lot cleaner and keeps the protocol and hardware lifecycle separate#asahi-dev · 2026-01-06 05:59
integralpilotwe can either a) merge it all into rtkit.c or b) implement an asc-specific backend (e.g., rtkit-asc.c) within the apple-rtkit library#asahi-dev · 2026-01-06 05:57
integralpiloteither way the current "helper" system is kinda messy and not upstreamable i think#asahi-dev · 2026-01-06 05:57
integralpilotbut actually maybe now thinking about it they ARE different things and can be updated seperate from each other (i.e. M3 - still uses RTKit 12 but upgrades to ASC hardware protocol 6, though the protocol 4 drivers have been working fine so not sure what it changes) so maybe they shouldnt be merged#asahi-dev · 2026-01-06 05:52
integralpiloti think it makes sense to merge them since they basically go hand-in-hand#asahi-dev · 2026-01-06 05:52
integralpilotpreviously the IPC protocol (rtkit.c) and hardware protocol that runs underneath that to map SRAM, control CPU (rtkit-helper.c) were seeprate#asahi-dev · 2026-01-06 05:30
JamesCalligeros[m]i forgor what it looks like but i do seem to remember not quite understaning why we couldnt just integrate whatever it does into the core rtkit api#asahi-dev · 2026-01-06 05:29
integralpiloti think we can integrate it into the main rtkit.c#asahi-dev · 2026-01-06 05:25
integralpiloti'll start prototyping some ideas i had re. that and see if any work#asahi-dev · 2026-01-06 05:25
integralpilotit is a bit messy so maybe that would be ideal#asahi-dev · 2026-01-06 05:22
JamesCalligeros[m]i think s ven would like to get rid of rtkit-helper if we can#asahi-dev · 2026-01-06 05:19
integralpilotmaybe a future 4th series adds trackpad support but that's really messy currently as i think it's partially grafted onto the (already existing upstream) driver for the USB/bluetooth magic mouse and not as important as keyboard#asahi-dev · 2026-01-06 05:17
integralpilot*the previous ones#asahi-dev · 2026-01-06 05:17
integralpiloteach is dependent on each other#asahi-dev · 2026-01-06 05:17
integralpilot3rd series introduces dockchannel hid for keyboard#asahi-dev · 2026-01-06 05:17
integralpilot2nd series introduces the dockchannel protocol#asahi-dev · 2026-01-06 05:17
integralpilot1st series introduces the RTKit helper#asahi-dev · 2026-01-06 05:16
integralpilotokay i've been researching how best to upstream MTP keyboard support i think i've decided it's best split into 3 series #asahi-dev · 2026-01-06 03:12
integralpilotgoodnight!#asahi-dev · 2026-01-06 03:08
fl0_idintegralpilot yeah, I can add adresses... with set_hw_bp I think? will check it tomorrow, gotta get some sleep now#asahi-dev · 2026-01-06 03:02
integralpiloti recall doing something similar but undid it after i got it working and forgot exactly how, sorry#asahi-dev · 2026-01-06 02:58
integralpilotcould you make it so m1n1 adds a BP for the address its about to jump to when it jumps to the kernel, then you can watch everything from the start?#asahi-dev · 2026-01-06 02:54
fl0_idit could very well be there's a better early break point#asahi-dev · 2026-01-06 02:54
integralpilotlooking through it now#asahi-dev · 2026-01-06 02:50
integralpilotactually wait by the time start_kernel is called i think we've already switched to virtual memory because of the assembly that runs before it, sorry#asahi-dev · 2026-01-06 02:48
integralpilotit's so fine, happy to help! i don't know much about linux early boot either, just inferring from my experience making my own (really bad) simple OSes #asahi-dev · 2026-01-06 02:47
fl0_idlike I don't know much about kernel early boot#asahi-dev · 2026-01-06 02:47
fl0_idintegralpilot well that would explain things. sry about my dumb questions, but you know more than me for sure#asahi-dev · 2026-01-06 02:46
integralpilotalso, i think set_sym_bp will use the virtual higher half address but when start_kernel is called we haven't switched to higher half yet and are still running in 0x10... rather than 0xfff... so the breakpoint will never hit (take this with a grain of salt, i'm not 100% sure)#asahi-dev · 2026-01-06 02:45
integralpilotre. breakpoint not working, maybe it never gets there? looking at your traceback it looks to not even be in the part of the kernel with symbols which is super weird#asahi-dev · 2026-01-06 02:44
integralpiloti know nothing about graphics except for when i made a terrible godot 2d game so i might be misunderstanding the results though#asahi-dev · 2026-01-06 02:43
fl0_idI think at some point someone posted about gpu differences in m3, but cannot remember where/who#asahi-dev · 2026-01-06 02:43
fl0_idintegralpilot sounds cool!#asahi-dev · 2026-01-06 02:42
fl0_idI also tried breaking manually and lookign at bt / stepping manually but that didn't yield anything. at the end bt was kimage_limit, before it was always _kernel_size_le_lo32 or maybe my system / symbols map is wrong/not loaded I guess?