#asahi-dev · 2022-04-18 15:56
marcanthen again disp0 also claims to have two things, a DART and an SMMU#asahi-dev · 2022-04-18 15:53
marcan(also that's 5?)#asahi-dev · 2022-04-18 15:53
marcanis what that says about those#asahi-dev · 2022-04-18 15:53
marcan> FPADDARTLLT TRADDARTBULK TRADDARTRT UMMSSMMUBULK UMMSSMMURT#asahi-dev · 2022-04-18 15:52
marcanactually the instance stuff says some of them aren't DARTs, they're "SMMU"?#asahi-dev · 2022-04-18 15:51
marcan(maybe because it's used to encode camera output?)#asahi-dev · 2022-04-18 15:51
marcanand ave0 (??)#asahi-dev · 2022-04-18 15:50
marcandisp0 also has that prop#asahi-dev · 2022-04-18 15:50
marcanthere is some "real-time" thing, maybe there's something special about these DARTs since they're used for video capture?#asahi-dev · 2022-04-18 15:49
marcanor maybe it's 3 instances and each one has 2 reg blocks?#asahi-dev · 2022-04-18 15:44
marcan(that's the second reg)#asahi-dev · 2022-04-18 15:44
marcanand also isp0 maps some random pmgr pstate register range and pokes it direclty, because of course it does#asahi-dev · 2022-04-18 15:44
marcanthere's *6*#asahi-dev · 2022-04-18 15:44
marcanyuuup#asahi-dev · 2022-04-18 15:42
marcana lot of the IO traffic is just DART noise#asahi-dev · 2022-04-18 15:41
marcanthere are a ton of DART instances too, which are also part of the wider ISP reg map (and thus show up wrong in mmiotrace)#asahi-dev · 2022-04-18 15:36
marcanbut the funny thing is the firmware does have the usual RTKit strings in it#asahi-dev · 2022-04-18 15:36
marcanall the I/O is 32bit so that's obviously not the usual ASC mailbox#asahi-dev · 2022-04-17 16:13
marcan(if ~nothing else)#asahi-dev · 2022-04-17 16:13
marcanDisk Utility.app works to create a new volume at least#asahi-dev · 2022-04-17 16:12
marcanyou can use the app store or just the install macos option in recovery mode#asahi-dev · 2022-04-17 16:12
marcanyou don't even need any command line tools for that#asahi-dev · 2022-04-17 16:12
marcanjust create a new volume and install to it#asahi-dev · 2022-04-17 16:11
marcanrobinp: you can install a second macOS install on the same APFS container as your primary, which means it can share the slack for updates#asahi-dev · 2022-04-16 09:21
marcanyou don't need the PMGR/iomon stuff, I'm just pointing that one out because it's the most barebones ASCTracer subclass we have lying around I think#asahi-dev · 2022-04-16 09:21
marcanmillenialhacker: look at hv/trace_pmgr.py and PMPTracer (and the commented out instantiation below it) for an example of how to get started#asahi-dev · 2022-04-16 09:17
marcan(there is already a generic ASC tracer but you will need to write the subclass of that to handle ISP and whatever its app-specific endpoints mean and how they work, as you figure it out)#asahi-dev · 2022-04-16 09:16
marcanfor ISP you will need to trace the ASC, which means you will need to sync trace the mailbox registers since the host will have to react to those synchronously to be able to pull payloads out of the buffers at the correct time#asahi-dev · 2022-04-16 09:15
marcanyou can configure the mode individually for everything you trace#asahi-dev · 2022-04-16 09:15
marcansync actually stops the guest and is limited by USB latency#asahi-dev · 2022-04-16 09:15
marcanasync uses a big fat buffer and the host processes that as fast as it can (which is still bottlenecked on python even for simple r/w prints in practice, before the USB bandwidth)#asahi-dev · 2022-04-16 09:14
marcanthere are async and sync tracing modes#asahi-dev · 2022-04-16 09:13
marcanthe HV will happily run with zero I/O if everything is off (as in you can literally yank the cable and the guest keeps running)#asahi-dev · 2022-04-16 09:13
marcanbut that's proportional to how much tracing you decide to enable#asahi-dev · 2022-04-16 09:12
marcantracing is bottlenecked on the host/python most of the time, and a RPi will bottleneck way worse than a proper machine#asahi-dev · 2022-04-16 09:12
marcankeep in mind a RPi will not cut it for high-bandwidth HV usage, but if you're just tracing the camera it's probably OK#asahi-dev · 2022-04-15 15:03
marcanI'm going to let the kernel robot do its thing and fire it off because I hate this so much#asahi-dev · 2022-04-15 15:03
marcanhaven't even compile tested it#asahi-dev · 2022-04-15 15:03
marcansven: see mailbox/i-hate-mailbox#asahi-dev · 2022-04-15 15:02
marcan*sigh*#asahi-dev · 2022-04-15 14:40
marcanso we have a clearly badly defined API and people cargo culting drivers that implement it with no users and no reviewer pointing out how stupid this all is#asahi-dev · 2022-04-15 14:40
marcanyup.#asahi-dev · 2022-04-15 14:39
marcanall the others don't, and clearly implement peek_data for no reason because it makes no sense, i.e. for everyone else it's cargo cult#asahi-dev · 2022-04-15 14:39
marcansven: bcm-flexrm-mailbox.c does what I do#asahi-dev · 2022-04-15 14:38
marcanhave I mentioned I hate mailbox?#asahi-dev · 2022-04-15 14:38
marcanexactly#asahi-dev · 2022-04-15 14:38
marcanyes#asahi-dev · 2022-04-15 14:38
marcanthe name is dumb too#asahi-dev · 2022-04-15 14:37
marcanthe only way it makes any sense is if it polls data on the spot#asahi-dev · 2022-04-15 14:37
marcanit's just broken and wrong.