#asahi-dev · 2026-06-03 19:22
chaos_princesslook at what v2 is doing, and do same, but over rtkit#asahi-dev · 2026-06-03 19:21
chaos_princessnot how pmp works, the report pm domains should ack power state changes immediately while pmp is booting, and only after it finishes - you start waiting for it to accet the change#asahi-dev · 2026-06-03 19:04
tryhard[m]is there direct way to make linux change devices power state only after pmp is fully configured? or i need to do something manually#asahi-dev · 2026-06-03 18:40
yukaI'm also seeing it on A18 Pro and M4 (even with switching to an e-core), probably because they don't sleep properly#asahi-dev · 2026-06-03 18:40
yukayep#asahi-dev · 2026-06-03 18:39
jannauon the max you could test a core on the second perf cluster. I suspect switching to an efficiency core at the beginning will be sufficient. that probably explains why we did not see it on m1* and m2*#asahi-dev · 2026-06-03 18:27
yukajannau: only if I switch the boot core to a core in another cluster, see the other commits in https://github.com/yuyuyureka/m1n1/commits/pstate-latencies-boot-cpu/#asahi-dev · 2026-06-03 18:22
jannaulooks good now#asahi-dev · 2026-06-03 18:21
jannaudoes the benchmark gives you the expected frequency for higher p-states starting around 3.7 GHz#asahi-dev · 2026-06-03 18:21
yukanow better?#asahi-dev · 2026-06-03 18:20
yukaAh well#asahi-dev · 2026-06-03 18:18
jannauyuka: something went wrong in the cpufreq PR: "yuyuyureka wants to merge 3290 commits into AsahiLinux:bits/001-devicetree-m3 from yuyuyureka:t6031-cpufreq"#asahi-dev · 2026-06-03 18:00
tryhard[m]ok i think i fixed it#asahi-dev · 2026-06-03 17:47
tryhard[m]i suspect something wrong maybe with dt?#asahi-dev · 2026-06-03 17:46
tryhard[m]<tryhard[m]> "https://pastebin.com/UQSTd1ZY" <- still couldn't figure out what the freaky. this is the updated code (https://github.com/dd-dreams/linux/commit/be7695d836dd082c97707782f9eb1fb7d0f3e96b#diff-16054adb2928fc56a1fe64ed8f12c303423a97c43db424c64cdc35efae995863)#asahi-dev · 2026-06-03 16:50
chaos_princessaop init seems to just work(tm), same for als/las, but there seem to be material changes in how the pdm decimator is setup#asahi-dev · 2026-06-03 13:43
svenotherwise that approach sounds good#asahi-dev · 2026-06-03 13:43
svensomeone also got sptm on m2 to run using nested virt fwiw, but that obviously doesn't translate 1:1 to m4+ because that one relies on GXF and SPRR#asahi-dev · 2026-06-03 13:42
svensame for the NVMMU and SART#asahi-dev · 2026-06-03 13:42
svenjust describe that domain N function 12 is dart_tWHATEVER_map page and we can just use the dart code inside m1n1 already#asahi-dev · 2026-06-03 13:41
svenStanfordAppliedCyber[m]: we know how all the iommus work anyway, no need for traces there#asahi-dev · 2026-06-03 12:17
chadmedchaos_princess: if/when you rebase silence on 7.1 make sure to use the new macaudio and i2s bus keeper settings. the amps will ignore the current ones and you will get interesting results.#asahi-dev · 2026-06-03 11:51
chadmedsimilarly the dcp driver has a bunch of insane^H^H cool macros and such to deal with the divergent fw abis#asahi-dev · 2026-06-03 11:50
chadmedyeah, it just has to be reversed and actually expressed#asahi-dev · 2026-06-03 11:49
Sobek[m]The rust macro magic that Lina developped for this sort of versionning can still be used tp help with that issue, right ?#asahi-dev · 2026-06-03 11:22
chadmed"we moved the fb swap submit callback from d408 to d409 because ummm well ahhh you see mmmm"#asahi-dev · 2026-06-03 11:21
chadmedthe gpu and display controller have not stayed the same across a single major os release, it is a bit annoying :P#asahi-dev · 2026-06-03 11:20
chaos_princessadded/removed fields, restructured blobs on which field and which struct goes where, things of that nature.#asahi-dev · 2026-06-03 11:18
StanfordAppliedCyber[m]yeah i expected that much, but i'm hoping it's just added/removed some fields, not oh hey we just totally changed the way the whole interface works#asahi-dev · 2026-06-03 11:17
chaos_princessat least the initdata is materially different from 13.5 to 14.8.3#asahi-dev · 2026-06-03 11:17
StanfordAppliedCyber[m]oh it's quite different? unfortunate, i was hoping it'd be similar, but this is apple so i knew better than to expect#asahi-dev · 2026-06-03 11:16
chaos_princessgpu#asahi-dev · 2026-06-03 11:16
StanfordAppliedCyber[m]this is in reply to my 3 part plan? or to yuka's question#asahi-dev · 2026-06-03 11:15
chaos_princess1. yes. 2. reach out to noopwafel/shiz#asahi-dev · 2026-06-03 11:14
StanfordAppliedCyber[m]i'm hoping it's not obviously, but at minimum initdata will need some fixes#asahi-dev · 2026-06-03 11:14
StanfordAppliedCyber[m]but that's what i want to check :)#asahi-dev · 2026-06-03 11:14
StanfordAppliedCyber[m]no clue#asahi-dev · 2026-06-03 11:13
yukaIs the firmware interface even that different from previous ones?#asahi-dev · 2026-06-03 11:09
StanfordAppliedCyber[m]also in the meantime i'm very interested in the gpu and i assume i'm free to use the hypervsior to start REing the firmware interface? i've never disassembled any of the gpu kexts or firmware and am totally untainted#asahi-dev · 2026-06-03 11:03
StanfordAppliedCyber[m]not to mention that sptm is a very ornery piece of software that is pretty much always unhappy in my experience#asahi-dev · 2026-06-03 11:03
StanfordAppliedCyber[m]s/say/think/, s/viable/promising/, s/forward//#asahi-dev · 2026-06-03 11:02
StanfordAppliedCyber[m]by contrast, on our own custom emulator, we can inspect state and then this becomes a lot easier, and i think supporting exclaves is absolutely something we want to do eventually because apple is going to keep putting more stuff in them and it'll likely be required to fully trace the hardware#asahi-dev · 2026-06-03 11:01
StanfordAppliedCyber[m]i say this is the most viable path forward btw because if we take the "just run sptm in a VM, proxy requests" option i think we're going to hit a LOT of issues, none of which are fundamentally unsolvable, but would delay progress by months if not years. it would also make exclaves way harder because sptm is responsible for organizing the shared memory between exclaves and xnu#asahi-dev · 2026-06-03 11:00
StanfordAppliedCyber[m]is this a viable path forward?#asahi-dev · 2026-06-03 10:59
StanfordAppliedCyber[m]yuka has volunteered to do the clean room RE work#asahi-dev · 2026-06-03 10:57
StanfordAppliedCyber[m]literally just:... (full message at <https://matrix.org/oftc/media/v1/media/download/AYYzAOtCaNonclsiuno_tZ5d9TqV7v2ErCcbOhrtr9iOi0rhI_UkYmJhzP-0H8OIwsOZbeskc6GOZbEj6yYtsQJCee1TsWJwAG1hdHJpeC5vcmcvZ1JaZFN3SFZDUUVIamtkcXJFSVBPY3Rn>)#asahi-dev · 2026-06-03 10:56
StanfordAppliedCyber[m]3. a runtime trace from my emulator just describing inputs and outputs#asahi-dev · 2026-06-03 10:56
StanfordAppliedCyber[m]2. clean room docs on how sptm works that i'm very careful to ensure don't include tainted info#asahi-dev · 2026-06-03 10:55
StanfordAppliedCyber[m]1. upstream, myself, all the hypervisor stuff that does NOT involve sptm. this is my own code, where i own the copyright, there's some timer stuff that comes to mind#asahi-dev · 2026-06-03 10:55
tryhard[m]i also have this PM: genpd: Disabling unused power domains but i doubt its related