#asahi-dev · 2026-01-17 14:36
ShizI wanna say yes, but I'll verify tomorrow :P#asahi-dev · 2026-01-17 14:36
chaos_princess*single p-cluster#asahi-dev · 2026-01-17 14:36
chaos_princessm3pro is single cluster?#asahi-dev · 2026-01-17 14:35
Shizwhich of course also have board IDs :)#asahi-dev · 2026-01-17 14:35
ShizI found myself diving into weird apple hardware for a bit and found out about something called cloudOS which they run on their private compute nodes...#asahi-dev · 2026-01-17 14:35
Shizm3 max/ultra i guess#asahi-dev · 2026-01-17 14:34
Shizahh ;p#asahi-dev · 2026-01-17 14:34
nickchanShiz: was thinking about bigger chips that has 3 clusters or more#asahi-dev · 2026-01-17 14:31
Shiz(one P-core, one E-core)#asahi-dev · 2026-01-17 14:31
Shizunless I misunderstood, that is what we did#asahi-dev · 2026-01-17 14:31
Shiz nickchan : maybe check if it works with 1 core per cluster#asahi-dev · 2026-01-17 14:08
kettenisnote that u-boot maps memory 1:1, so there so mapping disjoint 4k pages isn't a big issue#asahi-dev · 2026-01-17 13:23
sven"With a sufficient number of users of an API, it does not matter what you promise in the contract: all observable behaviours of your system will be depended on by somebody. " ;)#asahi-dev · 2026-01-17 13:21
svenit's just that at least some subsystem (v4l? something like that) abuses the scatter-gatherlist map api which implicitly does that #asahi-dev · 2026-01-17 13:21
sveniirc there's no iommu api that guarantees continous maps#asahi-dev · 2026-01-17 13:12
jannauthe main issue is that you can't map disjoint 4k pages continuously. 4k aligned start/end is supported via the unused sub page protection flags. devices/iommu's would need a flag only use 16k folios for me#asahi-dev · 2026-01-17 13:07
sveni had a very old patchset that fixed that but dropped it because it didn't work for the gpu and we also had stuff like muvm at that point#asahi-dev · 2026-01-17 13:06
svenyeah, the 4k cpu / 16k iommu thing is just hard due to linux abstractions#asahi-dev · 2026-01-17 13:05
svenintegralpilot: i'd put it into the aic2 binding since there's no change except for the compatible#asahi-dev · 2026-01-17 12:48
kettenisit's just the Linux abstraction layers that make it difficult#asahi-dev · 2026-01-17 12:47
kettenisnickchan: there is no fundamental problem with using the DART with 4k pages#asahi-dev · 2026-01-17 12:44
yukabut the proxy protocol does not work over it because it also has the central scrutinizer control stuff mixed in#asahi-dev · 2026-01-17 12:43
yukaI can see the serial output from m1n1#asahi-dev · 2026-01-17 12:43
yukait does do serial passthrough#asahi-dev · 2026-01-17 12:43
Shizoh, I thought that was what the micro-usb on the actual CS board was for :x#asahi-dev · 2026-01-17 12:40
yukathanks for confirming#asahi-dev · 2026-01-17 12:39
jannauyuka: it does not. the interactive control and printing breaks it. someone wanted to rewrite cs firmware so that it offers a transparent serial as well but I don't remember why and haven't heard of it again#asahi-dev · 2026-01-17 12:36
nickchanon m1#asahi-dev · 2026-01-17 12:36
nickchanand uboot has dart translation support so that's working with 4K page on the CPU MMU and 16K IOMMU#asahi-dev · 2026-01-17 12:32
nickchanbut it will not work on hardware that actually does not have 4K paging#asahi-dev · 2026-01-17 12:32
nickchana mircale that works on m1#asahi-dev · 2026-01-17 12:32
nickchanof course uboot assumes 4K everywhere#asahi-dev · 2026-01-17 12:31
nickchanhmm uboot doesn't work on a9-a11#asahi-dev · 2026-01-17 12:27
yukaFor me, I can open the ttyACM device in screen/picocom and see that it's running the proxy, but trying to run the proxyclient on it yields this: https://termbin.com/adby#asahi-dev · 2026-01-17 12:26
yukaHow do you use the m1n1 proxyclient on the central scrutinizer serial?#asahi-dev · 2026-01-17 12:15
integralpilothave to log off now but can look more into it tomorrow! thanks for your help sven working all this out :)#asahi-dev · 2026-01-17 12:11
integralpilotokay doing AIC3 stuff now, are we sure it's okay to bolt onto the apple,aic2 dt binding or is it safer to make a new one?#asahi-dev · 2026-01-17 12:10
integralpilotin good news, M5 also hasn't changed ascwrap or RTBuddy versions#asahi-dev · 2026-01-17 12:07
integralpilotbtw, anyone curious / want me to check anything re. M5 now that i've got the DT open?#asahi-dev · 2026-01-17 12:07
integralpilotgreat uboot work! :)#asahi-dev · 2026-01-17 12:06
nickchanand uboot's usage of the wdt expire now don't work on a8, so I will change it to m1n1's wdt_reboot() behavior which works#asahi-dev · 2026-01-17 12:05
sven-the#asahi-dev · 2026-01-17 12:05
svengreat, so just using the m3 compatible as baseline and hardcoded the values sounds viable then #asahi-dev · 2026-01-17 12:05
integralpilotsame config-offset (they call it extint-baseaddress), cap0-offset and maxnumirq-offset as M3, m3 Pro and m3 max too#asahi-dev · 2026-01-17 12:04
integralpilotso they haven't changed it AGAIN, thankfully#asahi-dev · 2026-01-17 12:04
integralpilotOK yes, just checked M5 and it STILL uses aic,3#asahi-dev · 2026-01-17 12:00
nickchanguess there's no runtime check#asahi-dev · 2026-01-17 12:00
nickchanoh it's actually crc32 instructions not supported#asahi-dev · 2026-01-17 11:59
integralpilotonly differences on ADT AIC node between pro vs max are the strides, capability mask, reg (ofc), and cpu count, as expected#asahi-dev · 2026-01-17 11:58
integralpilotokay so checked, and it's the same on m3 max