#
✕ nick:marcan
#asahi-dev · 2022-05-21 09:02
marcanit looks like the 4xxx range might be ISP-specific and might have some other stuff which you care about
#asahi-dev · 2022-05-21 09:01
marcanthe state report addresses too, etc
#asahi-dev · 2022-05-21 09:01
marcan # 94 GFX id: 99 psreg: 5:31 flags:a C g unk1_0: 0 unk1_1: 0 ctl_reg: 1:0x29 unk3: 8 0 0 1 @ 0x23b7003f8 = 0x0f002300
#asahi-dev · 2022-05-21 09:01
marcanand some of your "noisy" ones are random PMGR devices getting en/disabled, e.g.
#asahi-dev · 2022-05-21 09:00
marcanthose two I see you ran into already
#asahi-dev · 2022-05-21 09:00
marcan #103 ISP_CPU_CORE1 id: 109 psreg: 7: 2 flags: Cd unk1_0: 0 unk1_1: 0 ctl_reg: 1:0x2d unk3: 0 0 0 0 @ 0x23b704010 = 0x0f000300
#asahi-dev · 2022-05-21 09:00
marcan #102 ISP_CPU_CORE0 id: 108 psreg: 7: 1 flags: Cd unk1_0: 0 unk1_1: 0 ctl_reg: 1:0x2c unk3: 0 0 0 0 @ 0x23b704008 = 0x0f000300
#asahi-dev · 2022-05-21 09:00
marcanin particular,
#asahi-dev · 2022-05-21 08:58
marcanApple have this bad habit of having specific device drivers poke PMGR registers directly because they're too lazy to design abstractions properly
#asahi-dev · 2022-05-21 08:58
marcanmillenialhacker: nice! 0x23b7* is PMGR, you can ignore that entire range. xref it with the Linux devicetrees to see if it's something we already know about (clocks, pstates, NCOs, or similar), or this dump: https://mrcn.st/p/nHU2S1Wt
#asahi-dev · 2022-05-17 14:11
marcanbut there are whole C structs in there and such
#asahi-dev · 2022-05-17 14:10
marcanthough it does align sizes to 4 bytes IIRC
#asahi-dev · 2022-05-17 14:10
marcanjn: little endian. ADT doesn't really use cells, it's just "whatever Apple felt like throwing in there"
#asahi-dev · 2022-05-16 07:28
marcanwe should not declare those power domains at all
#asahi-dev · 2022-05-16 07:27
marcanre PMP that sounds like ANE1 on t6001
#asahi-dev · 2022-05-16 07:26
marcanbut I think e.g. cpufreq can be in from the beginning since at this point it's likely this DT will be upstreamed after that binding is in place, so we can just squash all that in
#asahi-dev · 2022-05-16 07:26
marcanideally we keep the not-yet-upstreamed add-ons in separate commits to some extent, especially the more complex ones like audio
#asahi-dev · 2022-05-16 07:25
marcanjannau: the latter, definitely. I was planning on redoing that history anyway.
#asahi-dev · 2022-05-15 09:56
marcanthanks for taking care of this :)
#asahi-dev · 2022-05-15 09:56
marcanI would say /soc/die0, and kboot should be updated to deal with multiple dies properly
#asahi-dev · 2022-05-15 09:43
marcannevermind
#asahi-dev · 2022-05-15 09:43
marcanah wait, that one itself isn't stable yet
#asahi-dev · 2022-05-15 09:42
marcanwe probably want that for m1n1/rust and that might let us use stable?
#asahi-dev · 2022-05-15 09:42
marcanTIL cfg(no_global_oom_handling)
#asahi-dev · 2022-05-11 14:36
marcanoh sorry, misread your sentence
#asahi-dev · 2022-05-11 14:33
marcanthat's the hub, not the controller, right?
#asahi-dev · 2022-05-11 14:30
marcanhuh, really
#asahi-dev · 2022-05-11 14:27
marcanyup
#asahi-dev · 2022-05-11 14:27
marcanI bought one of the standalone cards with the same chip to make sure that whatever firmware loading patch I come up with doesn't affect/break firmwareful devices
#asahi-dev · 2022-05-11 14:26
marcanIIRC it involved writing word offsets to PCI config space and data to a BAR, in a weird swizzled order
#asahi-dev · 2022-05-11 14:26
marcanyeah, I already looked at that, it's kind of insane
#asahi-dev · 2022-05-11 14:23
marcanthe installer already puts the raw kernel image in the EFI partition as a fallback so we can push an extractor to users (that will eventually be done in the installer, it's just an interim hack so that we don't have to tell existing users to reinstall, we can push a re-extractor in a package upgrade)
#asahi-dev · 2022-05-11 14:23
marcanmaz: firmware plan is to pull it from the stuff in the installer as we do for wifi/etc (and avd will be the same)
#asahi-dev · 2022-05-11 13:29
marcannice!
#asahi-dev · 2022-05-05 12:06
marcanlooks like it... even though nobody replied to my email yet :-)
#asahi-dev · 2022-05-05 12:06
marcanhttps://subspace.kernel.org/lists.linux.dev.html
#asahi-dev · 2022-05-04 17:42
marcanneed to see what other drivers are doing there
#asahi-dev · 2022-05-04 17:42
marcanand I might do it differently too...
#asahi-dev · 2022-05-04 17:40
marcanalso added CD debouncing which also makes the insertion more reliable, because apparently these sockets are noncompliant..
#asahi-dev · 2022-05-04 17:39
marcanpushed what I have so far to pcie/wip-sd-pm if anyone wants to look
#asahi-dev · 2022-05-04 16:33
marcanheh :)
#asahi-dev · 2022-05-04 16:31
marcanI can leave the old compatible in for you for a while if you want
#asahi-dev · 2022-05-04 16:31
marcanmostly that it's not a single cluster any more, now that a single instance handles all clusters
#asahi-dev · 2022-05-04 16:30
marcanthat means the only change to rtkit is removing that broken wait function
#asahi-dev · 2022-05-04 16:30
marcansven: ok, it works fine with that bus_lock magic, without requiring atomic context
#asahi-dev · 2022-05-04 15:34
marcanack
#asahi-dev · 2022-05-04 14:38
marcanthanks :)
#asahi-dev · 2022-05-04 14:35
marcanand then I can implement it as d3cold in apcie and support that
#asahi-dev · 2022-05-04 14:35
marcandunno if you saw the last email I sent; I think keeping it the way it is now makes sense, except maybe using a regulator instead of a gpio since it's more generic for power supplies
#asahi-dev · 2022-05-04 14:29
marcanmaz: have you seen the discussion re pcie power? that's what I wanted to get to today, but it ended up all being about SMC GPIO IRQs in the end...

plus ›