#asahi-dev · 2026-06-15 15:19
nickchani will use my devices then #asahi-dev · 2026-06-15 15:18
sven*arise#asahi-dev · 2026-06-15 15:18
sventhe first step is to get a PoC running that actually ends up at the entry after a deep sleep wfi. we can then just fix any issues that arive#asahi-dev · 2026-06-15 15:17
nickchanthere's also the issue of fb initialized while mmu is off which is very slow#asahi-dev · 2026-06-15 15:05
svensure, that's all fixable once we actually get a PoC with that path working#asahi-dev · 2026-06-15 15:02
nickchanand m1n1 needs to remember where its pagetables are#asahi-dev · 2026-06-15 15:00
nickchanright, it definitely shouldn't init the heapblock again#asahi-dev · 2026-06-15 14:59
nickchanafter exercise that code path two times, would result in `dlmalloc: internal error` at least on a7#asahi-dev · 2026-06-15 14:59
chaos_princessthat is fixable#asahi-dev · 2026-06-15 14:59
nickchanin m1n1#asahi-dev · 2026-06-15 14:59
nickchani think when the rvbar entry on primary cpu code path is exercised, the malloc entries will get corrupted#asahi-dev · 2026-06-15 13:51
jannauwe tried runtime PM in DCP but gave up since marcan and me didn't get it reliable#asahi-dev · 2026-06-15 13:50
chaos_princessmacos does even further, if you load the pmp tracer, there is constant agx on/off spam#asahi-dev · 2026-06-15 13:49
jannaudcp is at most clock-gated via "apple,min-state = 4". my understanding for agx was that it has its own aggressive power management#asahi-dev · 2026-06-15 13:36
chaos_princessfor cpus#asahi-dev · 2026-06-15 13:36
nickchanwasn't pstate 0 usually like, 24 MHz#asahi-dev · 2026-06-15 13:36
chadmednot sure if it goes that far but iirc its supposed to be clock gated at the very least in that state#asahi-dev · 2026-06-15 13:35
chaos_princessit is on via pmgr#asahi-dev · 2026-06-15 13:35
nickchando we keep dcp on? we have dcp_poweroff() and dcp_poweron() in firmware logs when going to s2idle#asahi-dev · 2026-06-15 13:34
chaos_princesscause that is what macos does#asahi-dev · 2026-06-15 13:34
chaos_princessdoes that turn it off via pmgr?#asahi-dev · 2026-06-15 13:34
chadmedi played with making pstate 0 the "base" for agx aaages ago and it helped slightly but caused massive performance issues#asahi-dev · 2026-06-15 13:33
chaos_princessmacos aggressively turns gpu off when it is not being used#asahi-dev · 2026-06-15 13:33
chaos_princesswe also keep dcp and gpu up.#asahi-dev · 2026-06-15 13:32
nickchanspeaking of trampoline m1n1 entry point kind of has a trampoline now to drop to EL1 from EL3#asahi-dev · 2026-06-15 13:30
chadmedthe tcons support it, but are we actually doing it? cant really tell just from iomfb #asahi-dev · 2026-06-15 13:30
chadmedbig one i havent a clue how we'd actually investigate is whether or not PSR is actually being used on the laptops for example #asahi-dev · 2026-06-15 13:29
chadmedi have a feeling a lot of our terrible idle power draw is no longer related to the application cores at this point#asahi-dev · 2026-06-15 13:28
nickchanbut macos does the same thing and power consumption is perfectly fine there#asahi-dev · 2026-06-15 13:28
nickchani mean, wasn't there an issue where the uncore is not powering down on pro/max/ultra on deep wfi#asahi-dev · 2026-06-15 13:27
nickchanpmp :3#asahi-dev · 2026-06-15 13:27
chadmedits just not really been terribly important#asahi-dev · 2026-06-15 13:27
chadmedits been on the pile of things to do for years, and at one point there was even a very dirty poc floating around #asahi-dev · 2026-06-15 13:26
chadmedyeah but if m1n1's entrypoint is a trampoline that can either start m1n1 or go do that other stuff if coming out of sleep on an already booted system then that becomes """"""easy""""""#asahi-dev · 2026-06-15 13:25
nickchannow you need to restore the state of a bunch of system registers#asahi-dev · 2026-06-15 13:24
chaos_princessoverwrite it with a jmp#asahi-dev · 2026-06-15 13:24
chadmedyeah its just that its a lot of work that no one really wants to do rn #asahi-dev · 2026-06-15 13:24
nickchanthe rvbar is still inside m1n1 while in linux#asahi-dev · 2026-06-15 13:23
svensounds like another good argument against psci-via-efi-or-whatever#asahi-dev · 2026-06-15 13:23
nickchaneven the non retention wfi is already annoying enough#asahi-dev · 2026-06-15 13:22
chaos_princesslike 'probably have several gpio drivers become proxies to psci' levels of yes#asahi-dev · 2026-06-15 13:22
chaos_princessyou mean the acc_ovrd deep sleep? - unclear, but signs point to yes#asahi-dev · 2026-06-15 13:21
chaos_princesswell, it is an argument for the list#asahi-dev · 2026-06-15 13:21
svendo we need callbacks or whatever to other drivers for deep sleep? because that would be very annoying with just psci as well#asahi-dev · 2026-06-15 13:20
svenno need to convince me ;)#asahi-dev · 2026-06-15 13:18
chaos_princesscounterargument - we don't have an el3 to hide our sins in#asahi-dev · 2026-06-15 13:17
sveni remember the argument being that they don't want to open the floodgates for N custom cpuidle drivers if they would accept one for apple silicon#asahi-dev · 2026-06-15 13:11
chaos_princesswhich, well, covers acpi, not dt#asahi-dev · 2026-06-15 13:11
chaos_princessthe only thing i can find is "Further, it is essential that the ARMv8 SoC provide a fully functional implementation of PSCI; this will be the only mechanism supported by ACPI to control CPU power state. "#asahi-dev · 2026-06-15 13:06
svenhm, i don't even remember where "must use pci" came from. maybe during early upstreaming on list? or maybe even on irc