#asahi-re · 2022-07-20 13:50
rqou_i assume that works on aic2 as well?#asahi-re · 2022-07-20 13:30
rqou_(needed this to work out some IRQ stuff on the m3 side, also i'm cheekily naming this terrible hack "m1cro")#asahi-re · 2022-07-20 13:29
rqou_also, i confirmed that you can run arbitrary code on the m3, no signing etc.#asahi-re · 2022-07-20 13:28
rqou_and then the firmware adds a magic offset to make it work out#asahi-re · 2022-07-20 13:28
rqou_but not the address from the m3's perspective, the address relative to the beginning of the hardware ip's range of addresses#asahi-re · 2022-07-20 13:27
rqou_afaict you queue commands by poking them into the MMIO registers (on the AP) that correspond to the cortex-m data space (at 0x10000000) and then writing the address to a mailbox#asahi-re · 2022-07-20 13:26
rqou_also, unless i'm reading it wrong, apple once again managed to Think Different on this firmware#asahi-re · 2022-07-20 13:25
rqou_i think i have most of the AVD Cortex-M3 mailbox-and-related stuff figured out#asahi-re · 2022-07-20 13:24
rqou_is there an easy way in m1n1 to see whether or not there are IRQs pending (for the AP)?#asahi-re · 2022-07-17 16:42
rqou_i'm guessing probably the latter#asahi-re · 2022-07-17 16:41
rqou_on a different topic, i didn't see anything whatsoever touch btcoex in the "converged ipc" kexts, so if that's the issue it's either a command sent from macos userspace or a problem on the wifi side#asahi-re · 2022-07-17 16:36
rqou_at least this one doesn't involve a hacky layering violation#asahi-re · 2022-07-17 16:36
rqou_maybe#asahi-re · 2022-07-17 16:35
rqou_i mean, the adt outright says so: "function-mcc_dataset = 125:M$DS()"#asahi-re · 2022-07-17 16:34
rqou_the AVD kext definitely also makes calls into the AMCC driver to tune some caching behavior (do we understand any of that yet?)#asahi-re · 2022-07-17 16:32
rqou_it's possible it's only important for performance tuning#asahi-re · 2022-07-17 16:32
rqou_haven't tried at all yet, will need to investigate#asahi-re · 2022-07-17 16:31
rqou_or maybe none of this is actually needed, i haven't tried actually poking the hardware yet#asahi-re · 2022-07-17 16:31
rqou_in the hv trace i see it getting poked multiple times, not just at boot#asahi-re · 2022-07-17 16:31
rqou_yeah#asahi-re · 2022-07-17 16:30
rqou_not sure if this can be handled in m1n1, the tunables might get lost when the device is power/clock gated?#asahi-re · 2022-07-17 16:29
rqou_oh yeah, could do that#asahi-re · 2022-07-17 16:28
rqou_that doesn't give me the masks#asahi-re · 2022-07-17 16:26
rqou_marcan: after some more poking, i think the thing is a "axi2af" block, with a bunch of tunables. unfortunately i can't seem to actually get the tunable data from the PMGR kext, because every tool i've tried chokes on what i assume is the macho chained fixup data (but might be a different problem)#asahi-re · 2022-07-17 09:03
rqou_the first half of avd overlaps pmgr[51] (on m1 max)#asahi-re · 2022-07-17 09:01
rqou_i'm taking a look at avd, half of its regs overlap something mapped in pmgr for some reason, and i cannot for the life of me figure out where any of these registers are being written from#asahi-re · 2022-07-17 09:01
rqou_is there a way to get m1n1 to break into the shell (and disassemble code) when a mmio register is written?#asahi-re · 2022-07-17 00:04
rqou_do you have any notes on avd?#asahi-re · 2022-07-16 23:59
rqou_oof#asahi-re · 2022-07-16 23:44
rqou_https://github.com/AsahiLinux/m1n1/blob/main/proxyclient/experiments/aop.py ?#asahi-re · 2022-07-16 23:40
rqou_are there any blocks of hardware left that nobody's looking at other than ANE?#asahi-re · 2022-07-16 22:43
rqou_i'd also like to know things like "how are filter coeffecients applied, what order are they arranged in, why are there six copies of the scale registers (some of which do not seem to do anything)"#asahi-re · 2022-07-16 22:41
rqou_also if anybody knows what a "DDA" or "ASE" might be#asahi-re · 2022-07-16 22:40
rqou_i now have an experiment that can kind-of scale images, but there's lots of fine details not understood#asahi-re · 2022-07-16 22:40
rqou_is there anybody here who is familiar with actually implementing image/video rescaling algorithms and wants to take a crack at figuring out some of the math involved here? https://github.com/rqou/m1n1/blob/scalers/proxyclient/experiments/scaler.py#asahi-re · 2022-04-17 01:28
rqou_yeah, scalers#asahi-re · 2022-04-17 01:26
rqou_anybody happen to know what a "ASE" might be? or what a "DPE" might be?#asahi-re · 2022-04-17 01:22
rqou_also each tiny group of registers is managed by its own set of classes#asahi-re · 2022-04-17 01:20
rqou_and the driver is implemented as a giant chain of classes inheriting from each other, one for each hardware version#asahi-re · 2022-04-17 01:19
rqou_a bajillion different settings covering basically everything iosurface supports#asahi-re · 2022-04-17 01:19
rqou_ugh, the scalers are such a huge pain to RE#asahi-re · 2022-04-16 23:00
rqou_probably video over dma and everything else configured via the whole RTBuddy/RTKit thing that i'm not familiar with#asahi-re · 2022-04-16 23:00
rqou_probably has some ringbuffers configured somewhere in its register space#asahi-re · 2022-04-16 22:59
rqou_it's just a mmio range#asahi-re · 2022-04-16 22:56
rqou_btw that boot log is similar to the old facetimehd boot log#asahi-re · 2022-04-16 22:56
rqou_millenialhacker: my complete guess is that i2c is used to poke extra settings in the camera sensor#asahi-re · 2022-04-16 11:28
rqou_it just hangs the bus#asahi-re · 2022-04-16 11:27
rqou_hmm, the scaler gets upset if i write to (presumably) read-only registers#asahi-re · 2022-04-16 11:14
rqou_the m1 version obviously has a lot more commands though#asahi-re · 2022-04-16 11:14
rqou_many of the commands are the same if you check the strings in the isp firmware