#asahi-dev · 2026-04-04 20:40
chaos_princessare you sure that your ids are correct#asahi-dev · 2026-04-04 18:53
chaos_princessright, so 0x204 is request, 0x205 is ack, ofc#asahi-dev · 2026-04-04 18:51
chaos_princessi think you are just trying to pattern match scheduler noise#asahi-dev · 2026-04-04 18:51
chaos_princessthe cores that the writes come from?#asahi-dev · 2026-04-04 18:49
chaos_princessi find it unlikely, but if you want to test it - delete everything but cpu0 from adt#asahi-dev · 2026-04-04 18:48
chaos_princessare you talking about cores?#asahi-dev · 2026-04-04 17:51
chaos_princessright, ignore that part, high power is not there on vanilla m1#asahi-dev · 2026-04-04 17:46
chaos_princessdo they have power modes on 13.5?#asahi-dev · 2026-04-04 17:40
chaos_princesstry also loading up the gpu#asahi-dev · 2026-04-04 17:31
chaos_princessso, try all-core#asahi-dev · 2026-04-04 17:30
chaos_princesswhat state did the last one leave it in, 2 or 0?#asahi-dev · 2026-04-04 17:27
chaos_princessi mean what triggers macos setting it to 0 or 2#asahi-dev · 2026-04-04 17:26
chaos_princessi suspect that that is what it is. it is most likely some sort of a separate power state, but i haven't figured out what exactly it is for#asahi-dev · 2026-04-04 17:22
chaos_princessno#asahi-dev · 2026-04-04 17:21
chaos_princess_that_ one is mostly guessing#asahi-dev · 2026-04-04 17:20
chaos_princessi wonder what happens if you fire up an all-core cpu workload#asahi-dev · 2026-04-03 17:06
chaos_princessbut i don't see anything that has an offset of 1048576 in the driver#asahi-dev · 2026-04-03 17:06
chaos_princessno, i don't think so, it is one of those things where they check if addresses are in range at compile time#asahi-dev · 2026-04-03 17:00
chaos_princesscan you post the output?#asahi-dev · 2026-04-03 16:53
chaos_princessthat's vanilla m3 only, right?#asahi-dev · 2026-04-03 15:43
chaos_princessthe file with my adt dump has a creation date in oct 2022#asahi-dev · 2026-04-03 15:42
chaos_princessmine is ancient#asahi-dev · 2026-04-03 15:39
chaos_princess'cause it is a dump using an ancient version of m1n1#asahi-dev · 2026-04-03 15:37
chaos_princesshttps://pastebin.com/gxsvYSUx#asahi-dev · 2026-04-03 15:36
chaos_princesstryhard[m]: again, you are reading it wrong, as the python formatting is confusing#asahi-dev · 2026-04-03 15:34
chaos_princesstryhard[m]: you are reading it wrong#asahi-dev · 2026-04-03 15:34
chaos_princessvanilla m1 is v1, m1 pro/max/ultra and m2 are v2#asahi-dev · 2026-04-03 15:17
chaos_princessmine doesn't and i can't be bothered making it do so#asahi-dev · 2026-04-03 15:16
chaos_princessjfyi, i am on irc, neither of those render as images :P#asahi-dev · 2026-04-03 15:15
chaos_princessyou can look in the pmgr node for things with notify_pmp = True#asahi-dev · 2026-04-03 15:14
chaos_princessmost likely for displays, pmp v2 gets notified about display controllers being turned off and on, so i assume v1 does too#asahi-dev · 2026-04-03 15:12
chaos_princessadt#asahi-dev · 2026-04-03 15:04
chaos_princessthat is most likely it. in pmgr ISP_SYS has id = 100#asahi-dev · 2026-04-03 13:52
chaos_princessno, the firmware runs on a dedicated cpu core in the SoC, we just want to talk to it and let it do it's thing#asahi-dev · 2026-04-03 13:49
chaos_princessif you want something useful - https://patents.google.com/#asahi-dev · 2026-04-03 13:49
chaos_princessi can assure you that llms are almost worthless for it #asahi-dev · 2026-04-02 20:44
chaos_princesssomething called io reporting, it is a common rtkit endpoint, and it does... something. all other drivers just ack whatever the rtkit device sends without doing anything and it seems to work fine#asahi-dev · 2026-04-02 20:23
chaos_princessare you sure about that? https://github.com/AsahiLinux/m1n1/blob/main/proxyclient/hv/trace_pmgr.py#L30#asahi-dev · 2026-04-02 20:23
chaos_princessit already does the right thing (almost)#asahi-dev · 2026-04-02 20:22
chaos_princessmarcan's pmp tracer :P#asahi-dev · 2026-04-02 20:22
chaos_princessfor iomon?#asahi-dev · 2026-04-02 20:21
chaos_princesswell, you still want to figure out where all the junk that macos writes comes from, as your driver will also have to write it there#asahi-dev · 2026-04-02 20:19
chaos_princessyou don't need to, just use iomon.poll() at the right places#asahi-dev · 2026-04-02 20:19
chaos_princessthe iomon stuff is supposed to help you deal with this thing, you register memory addresses you want to watch, and every time you call iomon.poll(), it reads them, and prints a pretty diff#asahi-dev · 2026-04-02 20:18
chaos_princessyep#asahi-dev · 2026-04-02 20:15
chaos_princessin case of 0x10, the payload follows the message, and it is an address in pmp's address space that macos is about to write a bunch of junk to#asahi-dev · 2026-04-02 20:14
chaos_princessso, first, it is not 0x1, it is 0x10, you haven't fully copied over the message structs from where i said, as your tracer still has `TYPE = 56, 48` in PMPMessage, while the other file had `TYPE = 56, 44` (pay attention to the last digit)#asahi-dev · 2026-04-02 20:04
chaos_princesspass the dart to the superclass, it is reading from the wrong address without it#asahi-dev · 2026-04-02 20:03
chaos_princesssuper().start() <- this line does not pass dart to the superclass#asahi-dev · 2026-04-02 20:02
chaos_princessoh, right, that thing, i remember now