#asahi-re · 2025-03-05 03:48
nickchan(hopefully soon drivers/perf/apple_m1_cpu_pmu.c can support down to a7 so that kind of source naming is not all that freaky)#asahi-re · 2025-03-05 03:47
nickchanin itself, too#asahi-re · 2025-03-05 03:47
nickchanBut using both "Hx" and "Txxxx" is dodgy #asahi-re · 2025-03-05 03:46
nickchansource file names are just a bad idea because all that means is sharing a codebase#asahi-re · 2025-03-05 03:39
chadmedwe already try not to rely on ADT compatibles#asahi-re · 2025-03-05 03:39
chadmednickchan: ah right yeah its kinda dodgy using apple's naming scheme for things bc they do freaky shit in their drivers that we shouldnt/cant#asahi-re · 2025-03-05 03:34
nickchanchadmed: AppleH7DisplayPipe.cpp#asahi-re · 2025-02-25 12:29
chadmedsmells like leapfrogging design teams#asahi-re · 2025-02-25 12:26
chaos_princessand i know that PMP protocol changed a lot on t600x#asahi-re · 2025-02-25 12:26
chaos_princessthe power delivery is different on t6XXX, vs t8XXX#asahi-re · 2025-02-25 12:25
chadmed"We still don't know exactly what most of this is" oh...#asahi-re · 2025-02-25 12:25
chadmedthat said, why are we only doing it for t8103 and t8112#asahi-re · 2025-02-25 12:24
chadmedi assume marcan wouldnt have just added that for shits/giggles#asahi-re · 2025-02-25 12:24
nickchan"write parts of the definition of cluster->apsc_pstate to the region just after the definitions" #asahi-re · 2025-02-25 12:24
chadmed:p#asahi-re · 2025-02-25 12:23
nickchan>nonsense: #asahi-re · 2025-02-25 12:23
chadmedwe should clear up what that code actually does and comment/document it#asahi-re · 2025-02-25 12:22
nickchanhttps://github.com/AsahiLinux/m1n1/blob/c49bc2c38e4ec9251ee23bc51ca3c46bda582b4a/src/cpufreq.c#L209 m1n1 deals with those registers here but it's nonsense according to my understanding but then again maybe things has changed between a11 and m1#asahi-re · 2025-02-25 12:22
chadmedthe numbers in those fields match the number of reported ANE pstates#asahi-re · 2025-02-25 12:21
chadmedthe DVFMStates for the pcluster reference 'fANE' states for each pcore pstate#asahi-re · 2025-02-25 12:20
chaos_princessshouldn't be? at least on j314 they have a whole separate voltage rail for it#asahi-re · 2025-02-25 12:20
chadmedhuh, are the ANE pstates tied to the pcore's states? #asahi-re · 2025-02-25 12:20
chaos_princessapple does in fact tryhard it, there is a whole separate coprocessor to control fabric bandwidth and power states#asahi-re · 2025-02-25 12:17
chadmedchaos_princess: yeah i figured, but my headcanon is funnier#asahi-re · 2025-02-25 12:17
chaos_princessnickchan: ill poke at those later, my t8103 machine is currently busy :P#asahi-re · 2025-02-25 12:17
chaos_princessapple fabric#asahi-re · 2025-02-25 12:16
chadmedClocks: FAST_AF :p#asahi-re · 2025-02-25 12:15
nickchanhowever, it probably is reading MaxDVMRWeight from those registers #asahi-re · 2025-02-25 12:14
nickchanas configured and filled in by iBoot#asahi-re · 2025-02-25 12:14
nickchan>just from the dt#asahi-re · 2025-02-25 12:13
nickchanadt, because patching the adt changes the frequency displayed by the pmgr kext#asahi-re · 2025-02-25 12:13
nickchanit's also my understanding that xnu do not deal with those register and the frequency information is just from the dt#asahi-re · 2025-02-25 12:12
nickchanchadmed: Not sure I only know a few fields in those register#asahi-re · 2025-02-25 12:12
nickchanxnu calls it "MaxDVMRWeight" https://mrcn.st/p/44xr8mcE#asahi-re · 2025-02-25 12:11
chadmedanything that might help with pcluster idle power? #asahi-re · 2025-02-25 12:11
nickchanthese registers also holds the amount of cores are allowed to be active in the cluster when that state is enter (boost state defs)#asahi-re · 2025-02-25 12:09
chadmedyeah its the same on the gpu aiui#asahi-re · 2025-02-25 12:09
nickchanbut state 1 and above should be fine#asahi-re · 2025-02-25 12:09
nickchanwell state 0 doesn't work like that it's a special state#asahi-re · 2025-02-25 12:09
nickchanyeah that#asahi-re · 2025-02-25 12:08
chadmedi guess we could just overwrite pstate 0 with the definition for the highest turbo pstate and see if it doesnt crash#asahi-re · 2025-02-25 12:08
chadmedcool but we know we can change the frequency. we want to know if the voltage _actually_ changes#asahi-re · 2025-02-25 12:07
nickchanit measures the frequency#asahi-re · 2025-02-25 12:07
nickchanwith the cpu_pstate_latencies experiments#asahi-re · 2025-02-25 12:07
nickchanI cranked up A9 to 2448 MHz and A11 to 2112 MHz Ecore 2376 MHz Pcore like that#asahi-re · 2025-02-25 12:07
chadmedscratch up some tracks? :p#asahi-re · 2025-02-25 12:07
chadmedbut how do we confirm that its _actually_ changing what the regulators are doing#asahi-re · 2025-02-25 12:06
nickchanand see if the experiments/cpu_pstate_latencies.py has anything to say about it#asahi-re · 2025-02-25 12:06
nickchana simple experiment is to try to, say, overwrite definition of state 3 with definition of state 4#asahi-re · 2025-02-25 12:06
nickchanchadmed: it very much does on a7-a11, t2 (different offset and different stride size on some chips)