#asahi-dev · 2026-01-23 19:55
chaos_princessyes, but that is why you don't copy xnu code.#asahi-dev · 2026-01-23 19:55
sveni mean.. don't paste xnu code in here obviously, but links and discussion are fine#asahi-dev · 2026-01-23 19:55
yukadoes xnu not have a weird license that is potentially not gpl compatible?#asahi-dev · 2026-01-23 19:55
sventhe alternative paths might be redacted from oss xnu anyway#asahi-dev · 2026-01-23 19:54
chaos_princess-re is for disasm and similar#asahi-dev · 2026-01-23 19:54
chaos_princessoss xnu is fine#asahi-dev · 2026-01-23 19:54
yukathat is similar to a lot of the things I discovered in xnu while trying to figure out why chickens were not working, and as sven just said there xnu no longer uses those registers on this platform and one might look at alternative code paths from there but that's more something for #asahi-re, no?#asahi-dev · 2026-01-23 19:53
svennice find! that confirms that register is gone then#asahi-dev · 2026-01-23 19:53
jannaualxdrl: vmlinuz is not gz compressed. `--compression=none` is correct#asahi-dev · 2026-01-23 19:52
chaos_princesshttps://github.com/apple-oss-distributions/xnu/blob/f6217f891ac0bb64f3d375211650a4c1ff8ca1ea/pexpert/pexpert/arm64/board_config.h#L269 #asahi-dev · 2026-01-23 19:52
sven*even when#asahi-dev · 2026-01-23 19:51
svenbut that needs more investigation and some m1n1 code to stay around even linux is running and some PSCI-over-EFI or whatever to hook it all up#asahi-dev · 2026-01-23 19:50
svenand that was using something else anyway iirc#asahi-dev · 2026-01-23 19:50
sven"proper" cpu sleep is much more cursed anyway. it would make the core lose ~all state and have it end up at RVBAR again when coming back up#asahi-dev · 2026-01-23 19:49
svenmaybe the cores can unconditionally enter the higher boost states now ;)#asahi-dev · 2026-01-23 19:49
svenoh right, it was mostly about the higher cpu states and power saving was a nice bonus#asahi-dev · 2026-01-23 19:49
yukabut if xnu no longer uses it there is likely another way to achieve it now#asahi-dev · 2026-01-23 19:49
alxdrljannau: thanks a lot, it gets me further https://gist.github.com/alxdrl/bc71349164e928111e1e711091cfcb93, will have a look tomorrow #asahi-dev · 2026-01-23 19:48
jannaumarcan did add this since it's necessary for entering the higher cpu frequency states and saves power#asahi-dev · 2026-01-23 19:47
yukaindeed#asahi-dev · 2026-01-23 19:45
svenI think when this came up the first time I looked at XNU and also didn't see it access SYS_APL_CYC_OVRD anymore#asahi-dev · 2026-01-23 19:45
svenwe only needed the special one because we wanted the deeper wfi state that drops register state#asahi-dev · 2026-01-23 19:44
sveniirc there's generic code (or a generic driver) that uses wfi#asahi-dev · 2026-01-23 19:43
svenyeah, but no need for a special driver then (assuming wfi really keeps state now)#asahi-dev · 2026-01-23 19:43
yukadoes it still make sense to use wfi without being able to set that?#asahi-dev · 2026-01-23 19:42
svenyes, that changes wfi behavior #asahi-dev · 2026-01-23 19:42
yuka(SYS_APL_CYC_OVRD according to wiki)#asahi-dev · 2026-01-23 19:42
yukas3_5_c15_c5_0 is locked#asahi-dev · 2026-01-23 19:42
yukaso I don't see a way to check this#asahi-dev · 2026-01-23 19:42
yukawell it immediately crashes when attempting to do the apple_cpu_deep_wfi#asahi-dev · 2026-01-23 19:42
sventhough i guess if linux runs for a while and doesn't run into weird crashes that's proof enough that it keeps state :D#asahi-dev · 2026-01-23 19:41
sveni'd still double check if wfi truly never loses state on these machines, that aidr_el1 is impdef iirc. i don't expect them to change it but who knows#asahi-dev · 2026-01-23 19:39
svenwhich is a problem because we can't implement psci on these machines because there's no EL3 or similar#asahi-dev · 2026-01-23 19:39
yukaI see, just re-read the wiki section on that#asahi-dev · 2026-01-23 19:38
svenarm64 doesn't accept custom cpuidle drivers, they want everything to go through PSCI#asahi-dev · 2026-01-23 19:38
svenit can never be upstreamed#asahi-dev · 2026-01-23 19:38
yukasven: you mean cpuidle can never be upstreamed, or not in this state?#asahi-dev · 2026-01-23 19:37
yukaotherwise cpuidle-apple wouldn't register#asahi-dev · 2026-01-23 19:37
yukaaccording to the check of aidr_el1, yes#asahi-dev · 2026-01-23 19:36
svenyuka: cpuidle can't be upstreamed anyway, does wfi retain full register state on your machine?#asahi-dev · 2026-01-23 19:29
jannaualxdrl: https://gist.github.com/jannau/43ce0026a7dd9ba89f2f2c14ff5b04c4 works for me but I dont have a lot space for bootargs. let's see how we can reserve space in u-boots environment#asahi-dev · 2026-01-23 19:24
yukasince currently it just checks for apple,arm-platform#asahi-dev · 2026-01-23 19:23
yukamight have to add some compatible indicator in the device tree for this e.g. cpuidle-t8103 (which is also used in all the others which support this)?#asahi-dev · 2026-01-23 19:22
yukanot a problem on mainline (yet) since there is no cpuidle-apple#asahi-dev · 2026-01-23 19:21
yukaagain some locked register write#asahi-dev · 2026-01-23 19:21
yukaanother m4 oopsie: cpuidle-apple will kill the cpu when trying to apple_enter_idle()#asahi-dev · 2026-01-23 19:20
jannaualxdrl: the u-boot part in linux.py needs to be adjusted. a little bit tricky since there is not a lot of space after we switched to bootflow based booting#asahi-dev · 2026-01-23 18:28
alxdrl(Context is ci-tron)#asahi-dev · 2026-01-23 18:28
alxdrlRequirement : using only pre-build components extracted from official software distribution channels (RPMs) if possible#asahi-dev · 2026-01-23 18:27
alxdrlI think I need to stitch m1n1+uboot+vmlinuz together (linux.py does something similar, but the u-boot use case is kind of obscure to me)