#asahi-dev · 2026-01-23 20:40
yukaand then you also want to make it an everest not sawtooth (not that it matters right now I think?) and have it use the larger l2 cache#asahi-dev · 2026-01-23 20:39
yukaif you declare only a single cpu#asahi-dev · 2026-01-23 20:39
yukaso you want @10100#asahi-dev · 2026-01-23 20:39
fl0_id__yeah i know#asahi-dev · 2026-01-23 20:39
yukayes#asahi-dev · 2026-01-23 20:39
yukathat's the boot cpu#asahi-dev · 2026-01-23 20:39
fl0_id__is it normal that m1n1 skips cpu 4? (maybe because it's already running on that?#asahi-dev · 2026-01-23 20:38
fl0_id__yuka, m1n1 + linux log https://www.pastery.net/jperwb+srwbmp/#srwbmp#asahi-dev · 2026-01-23 20:36
fl0_id__t6031.dtsi https://www.pastery.net/jperwb/#asahi-dev · 2026-01-23 20:36
fl0_id__yeah, maybe I have to extend the device tree. as my own ones weren't working either I used one by integralpilot to which I then added serial. it might need cpu0@<something> or an actual reg I think...#asahi-dev · 2026-01-23 20:27
yukaand maybe the earlier m1n1 output#asahi-dev · 2026-01-23 20:26
yukacan you share the device tree you're using?#asahi-dev · 2026-01-23 20:26
yuka> this also looks kinda wrong but shouldn't(?) be a problem?#asahi-dev · 2026-01-23 20:26
yuka> Failed to find device node for boot cpu#asahi-dev · 2026-01-23 20:25
yukayay for getting earlycon to work#asahi-dev · 2026-01-23 20:25
sventhat virtual address looks wrong#asahi-dev · 2026-01-23 20:25
fl0_id__and the call trace matches what I traced before with hvc#asahi-dev · 2026-01-23 20:24
fl0_id__https://www.pastery.net/jzqubt/#asahi-dev · 2026-01-23 20:24
fl0_id__I thought I checked reserved memory etc#asahi-dev · 2026-01-23 20:24
fl0_id__with the s5l,address argument I also get some earlycon output. it is trying to read from memory which it cannot.#asahi-dev · 2026-01-23 20:24
svens/anymore/anyway/#asahi-dev · 2026-01-23 20:24
svenwe're not using github's merge feature anymore, as long as the diff looks reasonable everything's good#asahi-dev · 2026-01-23 20:23
sventhe branch is fine#asahi-dev · 2026-01-23 20:23
leioprobably would end up in bits/010-soc in some form#asahi-dev · 2026-01-23 20:20
yuka(lmk if you prefer these PRs to go against any other branch, though I suppose it doesn't matter much if you apply it directly)#asahi-dev · 2026-01-23 20:20
yukahttps://github.com/AsahiLinux/linux/pull/442#asahi-dev · 2026-01-23 20:16
JoshuaPeisach[mds]thanks#asahi-dev · 2026-01-23 20:16
yukaJoshuaPeisach[mds] https://github.com/AsahiLinux/linux/blob/f1879b3d78dcf69a8ff9be3b504c1bd6d5fa2f47/drivers/cpuidle/cpuidle-apple.c#asahi-dev · 2026-01-23 20:15
JoshuaPeisach[mds](which area of code are you guys talking about, so as a newbie I can follow along?)#asahi-dev · 2026-01-23 20:13
yukaThen I can remove the register based check?#asahi-dev · 2026-01-23 20:08
svenyeah#asahi-dev · 2026-01-23 20:05
jannauI'd be surprised if M5 and later would regain ARM64_REG_CYC_OVRD#asahi-dev · 2026-01-23 20:03
svenand i agree with jannau, i'd also list all supported devices#asahi-dev · 2026-01-23 20:03
svenpretty sure it'll be true for everything M4#asahi-dev · 2026-01-23 20:03
jannauthe xnu code was for m4 max#asahi-dev · 2026-01-23 20:03
jannauI'd use a a ositive list with "apple,t8103", ...#asahi-dev · 2026-01-23 20:02
yukaIf I knew that this is true for all M4(/Pro/...) cores, I would check based on that, but I only have the base M4#asahi-dev · 2026-01-23 20:00
yukaor just hard code lists of platforms where it should be skipped?#asahi-dev · 2026-01-23 19:59
yukaso then we also don't need cpuidle-apple there. shall I make a PR to hide it behind a new device tree compatible?#asahi-dev · 2026-01-23 19:58
yukaseems like it#asahi-dev · 2026-01-23 19:57
svenbut that mode may no longer exist on m4#asahi-dev · 2026-01-23 19:57
yukaI see!#asahi-dev · 2026-01-23 19:57
svenwe're not disabling irqs there, we wrote to the register to turn wfi from "regular arm64 wfi" into "apple's wfi that now drops register state"#asahi-dev · 2026-01-23 19:57
yukaOk, the cpuidle-apple code has no comments ;)#asahi-dev · 2026-01-23 19:57
svenkeep in mind we're doing something different in cpuidle#asahi-dev · 2026-01-23 19:57
chaos_princessyea, it seems to just do a wfi now#asahi-dev · 2026-01-23 19:57
yukaand now some irqs are disabled by other means#asahi-dev · 2026-01-23 19:56
yukawhere previously ARM64_REG_CYC_OVRD was used#asahi-dev · 2026-01-23 19:56
yukaok so here is an #else block for NO_CPU_OVRD#asahi-dev · 2026-01-23 19:56
yukahttps://github.com/apple-oss-distributions/xnu/blob/f6217f891ac0bb64f3d375211650a4c1ff8ca1ea/osfmk/arm64/machine_routines_asm.s#L1041