#asahi-dev · 2022-04-15 14:37
marcanand now everyone is confused.#asahi-dev · 2022-04-15 14:37
marcan*after* this?#asahi-dev · 2022-04-15 14:37
marcan * False, if controller doesn't have any data to be read.#asahi-dev · 2022-04-15 14:37
marcan * some data.#asahi-dev · 2022-04-15 14:37
marcan * Return: True, if controller has, and is going to push after this,#asahi-dev · 2022-04-15 14:37
marcanuhhh okay#asahi-dev · 2022-04-15 14:37
marcan * mbox_chan_received_data()#asahi-dev · 2022-04-15 14:37
marcan * The data is actually passed onto client via the#asahi-dev · 2022-04-15 14:37
marcan * A poke to controller driver for any received data.#asahi-dev · 2022-04-15 14:37
marcansounds vague and reasonable#asahi-dev · 2022-04-15 14:37
marcan * received from remote by the controller.#asahi-dev · 2022-04-15 14:37
marcan * mbox_client_peek_data - A way for client driver to pull data#asahi-dev · 2022-04-15 14:37
marcanthe problem is the doc comment for that function is stupid#asahi-dev · 2022-04-15 14:35
marcanyeah.#asahi-dev · 2022-04-15 14:35
marcansven: remember when I said different drivers interpret the mbox API differently and it's all an insane broken mess and I hate it?#asahi-dev · 2022-04-15 14:32
marcanpoor lina...#asahi-dev · 2022-04-15 14:31
marcanand also its ring buffer base address is a special case#asahi-dev · 2022-04-15 14:31
marcanand also syslog is the only channel with more than one ring buffer on AGX. because of course.#asahi-dev · 2022-04-15 14:31
marcanand doesn't use the RTKit syslog.#asahi-dev · 2022-04-15 14:31
marcansven: in today's stream, AGX has its own syslog ring buffers. 6 of them.#asahi-dev · 2022-04-14 07:32
marcannot yet#asahi-dev · 2022-04-13 06:38
marcanI'm okay with us spending time yak shaving stuff like sleep, but there is no way this much pain is justified for a couple mailbox registers with dedicated IRQs and hardware FIFOs we aren't even using#asahi-dev · 2022-04-13 06:36
marcanpoor abstractions are a code smell#asahi-dev · 2022-04-13 06:36
marcanthat thing is holding us back and provides, as far as I can tell, absolutely zero actual useful benefit/code at this point#asahi-dev · 2022-04-13 06:36
marcanat this point I'd almost consider dropping that mailbox driver, moving the register pokes to something near rtkit, and just having it follow the existing DT binding... especially if it takes more than a day or so to make this mess work with the RTKit refactoring#asahi-dev · 2022-04-13 04:21
marcanhave I mentioned I hate mailbox yet?#asahi-dev · 2022-04-13 04:21
marcanwe need the atomic mailbox stuff to make SMC shutdown/reboot ops work properly :/#asahi-dev · 2022-04-12 10:43
marcanwell, that's 1/3 of what it was yesterday...#asahi-dev · 2022-04-12 10:42
marcaninbox: 666 emails#asahi-dev · 2022-04-11 03:02
marcanbig hypervisor lock#asahi-dev · 2022-04-11 02:45
marcanI forget if I take the BHL to do local FIQ handling... I should probably move at least that to finer-grained locking, so the HV ticks don't fight each other. that'll at least make it scale.#asahi-dev · 2022-04-11 01:34
marcanthat plus the BHL means it doesn't scale#asahi-dev · 2022-04-11 01:34
marcanshould slow down the HV tick on at least the secondaries#asahi-dev · 2022-04-11 01:34
marcantoo many cores probably#asahi-dev · 2022-04-08 17:27
marcanincidentally, those exceptions don't exist on t8103 (or don't get enabled)#asahi-dev · 2022-04-08 17:25
marcanI bet there is one, and if there isn't maybe we can just hook that to disable it at the hardware level#asahi-dev · 2022-04-08 17:25
marcantbh, given we know those cache lines don't actually *break* anything at least the vast majority of the time, I'd first look for a xnu bootarg to turn off those panics#asahi-dev · 2022-04-08 17:23
marcanI'm... not entirely sure what the right fix for that would be. maybe map all guest RAM as noncacheable in the HV stage1 PTEs, since we can't know where framebuffers land? and maybe also all the carveouts in general?#asahi-dev · 2022-04-08 17:22
marcanJamie[m]1, jannau: that panic reason means something like a framebuffer cache line ended up in the CPU cache, which if it happens after initial boot could mean it was randomly caused by speculation in HV code#asahi-dev · 2022-04-08 16:50
marcan(sorry for being a bit slow the past few days, had to take a bit of a rest after a hectic month)#asahi-dev · 2022-04-08 16:49
marcanwas going to maybe do something in the next couple of days, if you want to put that together I'll wait a bit more and clear out my inbox instead#asahi-dev · 2022-04-08 16:49
marcanjannau: go for it :)#asahi-dev · 2022-04-07 07:42
marcan(and if it dies the machine watchdog reboots a few seconds later)#asahi-dev · 2022-04-07 07:42
marcanit does a ton of critical stuff like battery charge management#asahi-dev · 2022-04-07 07:41
marcanSMC is always alive#asahi-dev · 2022-04-07 07:41
marcanyeah#asahi-dev · 2022-04-07 07:41
marcanwe know that, the backdoor SMC channels are not dependent on the main one#asahi-dev · 2022-04-07 07:37
marcanmaybe something is wrong, DCP does something that crashes SMC, then NVMe fails to boot because of that?#asahi-dev · 2022-04-07 07:37
marcanis SMC a common point there?#asahi-dev · 2022-04-06 17:36
marcanso yeah, whole bespoke display controller driver to write there