#
✕ nick:marcan
#asahi-re · 2023-01-13 03:41
marcandid you look at the bit defines from linux?
#asahi-re · 2023-01-13 03:40
marcanq3k2: I looked at that (mostly blind over the same UART lol), IIRC the closest relative was S3C2400 but some things were moved around
#asahi-re · 2022-12-05 02:25
marcaneiln: given they're two independent blocks I think the answer is to just expose them as separate devices and let userspace figure out how to balance work
#asahi-re · 2022-12-03 14:15
marcanhence my theory about silicon bug or power delivery issues
#asahi-re · 2022-12-03 14:15
marcanso there is only one die, max, with two anes, and one is always disabled
#asahi-re · 2022-12-03 14:15
marcanpro only has one ane
#asahi-re · 2022-12-03 14:15
marcanpro does not have ane1 at all
#asahi-re · 2022-12-03 14:15
marcanane1 is the broken one, which means also ane3 on ultra
#asahi-re · 2022-12-03 14:14
marcanif ane0 works on max then ane2 will work on ultra since it's just two of the same thing
#asahi-re · 2022-12-03 14:14
marcanultra is just two max dies
#asahi-re · 2022-12-03 14:14
marcaneverything has one functional one per die, Ultra has two dies
#asahi-re · 2022-12-03 14:14
marcaneiln: it's 16 cores per neural engine
#asahi-re · 2022-12-03 04:06
marcanand yes, geohot's "research" is, well... yeah what you said tracks
#asahi-re · 2022-12-03 04:04
marcanit was probably intended for M1 Max too, I don't think the reason they ditched the second ANE is because it wasn't ready in time, though it could be
#asahi-re · 2022-12-03 04:03
marcanmacOS has some kind of balancer/dispatcher kext for this
#asahi-re · 2022-12-03 04:03
marcanconcurrency *should* be used on M1 Ultra for the ANE across each die
#asahi-re · 2022-12-03 04:02
marcanmy theory is it's either broken silicon or they decided not to implement it due to power envelope reasons
#asahi-re · 2022-12-03 04:01
marcanthey seem to be hard shut down
#asahi-re · 2022-12-03 04:01
marcanI don't know why, but iBoot filters them out
#asahi-re · 2022-12-03 04:01
marcaneiln: also you may know this already, but ane1/ane3 are fused off on Max/Ultra devices
#asahi-re · 2022-11-29 14:29
marcanBut there's probably a much easier trick: just change the reg to 0 and add 0x4000. DAPF is always one page after DART and instance 0 should match up with the first reg.
#asahi-re · 2022-11-29 14:27
marcaneiln: ah yes, the instancing thing is kind of complicated. I think what's supposed to happen is you count the instances from the length of the `instance` property and then the DAPFs come after the # instances
#asahi-re · 2022-11-28 18:04
marcananyway, I should get some sleep, hope that helps a bit!
#asahi-re · 2022-11-28 18:02
marcanbut as I said the stuff in dapf.c should do that "properly" for you
#asahi-re · 2022-11-28 18:01
marcanif those are wrong then there's no chance the firmware will boot ;)
#asahi-re · 2022-11-28 18:01
marcanthe DAPF tunables I mean
#asahi-re · 2022-11-28 18:01
marcanbecause there's a very high chance those change from version to version (since firmware placement in memory depends on the size of every other firmware and things like that)
#asahi-re · 2022-11-28 18:00
marcanalso you copied the tunable writes from macos? or from the ADT in the 12.3 m1n1?
#asahi-re · 2022-11-28 17:59
marcanyeah :)
#asahi-re · 2022-11-28 17:59
marcanyou'll have to test stuff vs the macos m1n1 first, then see if it works back in 12.3 (if it does, lucky, if not, time to install as 12.3 macos and see what changed)
#asahi-re · 2022-11-28 17:59
marcanyeah, so keep in mind, especially once the ASC comes up, that could mean you have interface differences
#asahi-re · 2022-11-28 17:58
marcanbut for tracing macos you use m1n1 installed on macos, right?
#asahi-re · 2022-11-28 17:58
marcanthen that's 12.3
#asahi-re · 2022-11-28 17:57
marcanare you running against m1n1 installed on top of macos, or the asahi m1n1?
#asahi-re · 2022-11-28 17:57
marcanwhich I guess is conditional on whether we need to use the ASC at all or not, which then brings in the ISP question again
#asahi-re · 2022-11-28 17:56
marcanwe need to support at least certain cherry-picked versions (like we are on 12.3 on t8102/t600x and 12.4 on t8112 right now), and the list will grow over time
#asahi-re · 2022-11-28 17:55
marcanand different firmware versions (from different macos versions) on the same hw
#asahi-re · 2022-11-28 17:55
marcan- and assuming you're using drm/accel... the question is do you want to write the driver in Rust or C :p (if Rust, definitely talk to lina)
#asahi-re · 2022-11-28 17:54
marcan- also would be nice to figure out the firmware and its interface and whether it changes from version to version
#asahi-re · 2022-11-28 17:53
marcan- might want to also lurk in #dri-devel since that's where the rest of the dri crowd is, as far as using the dri/accel subsystems
#asahi-re · 2022-11-28 17:53
marcan- it would be helpful if you can write up some prose on how this all works so others have somewhere to start trying to understand it
#asahi-re · 2022-11-28 17:52
marcanI need to get some sleep, but some pointers
#asahi-re · 2022-11-28 17:50
marcanwell, drm already has that I think
#asahi-re · 2022-11-28 17:50
marcanok, then it's probably a single VM and you don't need to worry about that, you just need some kind of buffer token thing so userspace apps never even see addresses, just some opaque buffer identifiers
#asahi-re · 2022-11-28 17:49
marcanand there's no way for the program to like run a load instruction from an arbitrary address
#asahi-re · 2022-11-28 17:49
marcani.e. there are no dvas in like a compiled program etc
#asahi-re · 2022-11-28 17:49
marcanright, and all that is controlled by the driver
#asahi-re · 2022-11-28 17:47
marcanso the driver controls all dvas?
#asahi-re · 2022-11-28 17:45
marcancan ane programs decide where to do data reads or are the input/output buffers programmed by the driver always?
#asahi-re · 2022-11-28 17:45
marcanwhat is the kernel responsible for as far as deciding where data buffers live

plus ›