#asahi-dev · 2022-01-21 13:04
sventhe ADT has a bunch of interrupt numbers for disp0 fwiw#asahi-dev · 2022-01-21 13:03
svenwouldn't surprise me if the IRQs show up somewhere in AIC as well#asahi-dev · 2022-01-21 13:01
svenare they only routed there or do we just not know the AP IRQ number for them?#asahi-dev · 2022-01-21 08:00
sveni'd be very happy that locked DART would just go away#asahi-dev · 2022-01-21 08:00
svenand they still get to keep the lockdown for SEP for all the security stuff that actually matters#asahi-dev · 2022-01-21 07:59
svenyup#asahi-dev · 2022-01-21 07:59
svenyeah#asahi-dev · 2022-01-21 07:58
svenyeah, i haven't had any good reason to take over any IOP so far#asahi-dev · 2022-01-21 07:58
svenyup :D#asahi-dev · 2022-01-21 07:57
svenworst case just build a dumb proxyclient VM using ROP :D#asahi-dev · 2022-01-21 07:57
svenbut other than that i don't think it's very hard to take over DCP with access to .data#asahi-dev · 2022-01-21 07:56
sveni wonder if they also lock down the allowed kernel code phys addrs like they do for the main processor though#asahi-dev · 2022-01-21 07:55
sveni never checked where it stores its pagetables :D#asahi-dev · 2022-01-21 07:55
sventhere are also functions pointers directly at the beginning of the .data section fwiw, so getting control should be easy#asahi-dev · 2022-01-21 07:55
svenyeah, something like that#asahi-dev · 2022-01-21 07:54
svenand I think it doesn't want to reboot from 0x01 where it just doesn't touch anything inside there#asahi-dev · 2022-01-21 07:54
svenyeah, DCP does changes a few values when it's in this 0x10 mode though#asahi-dev · 2022-01-21 07:53
svenwe might even be able to hack that to recover from panics :D#asahi-dev · 2022-01-21 07:53
svenfor DCP you can even re-map .data using DART to any place you want#asahi-dev · 2022-01-21 07:52
sventhis leads to weird situations where apple's special debug commands go through the normal nvme command submission but their buffers must be allowed inside SART because it's the co-processor accessing those and not the NVMe DMA magic#asahi-dev · 2022-01-21 07:51
sven(and the NVMe DMA crap must be whitelisted inside the NVMMU)#asahi-dev · 2022-01-21 07:51
svenand i know for sure that ANS firmware can't access anything that's not allowed by SART#asahi-dev · 2022-01-21 07:50
svenyeah#asahi-dev · 2022-01-21 07:50
svendon't see how that gains us anything#asahi-dev · 2022-01-21 07:50
svenwe could validate the SART ranges inside m1n1 to ensure they are all inside TZ but meh...#asahi-dev · 2022-01-21 07:50
svenand for crashlog/syslog/etc. it just asks us for an address. no need to verify anything there#asahi-dev · 2022-01-21 07:49
svenit's all setup before we get control#asahi-dev · 2022-01-21 07:49
svenand it never sends those addresses to us#asahi-dev · 2022-01-21 07:49
svenat least all those that matter. it has some pre-mapped craped inside SART but that's just the firmware + stuff inside TZ#asahi-dev · 2022-01-21 07:49
svenyes#asahi-dev · 2022-01-21 07:48
svenANS doesn't need that fwiw, but might make sense for DCP/SMC#asahi-dev · 2022-01-21 07:47
sveni remember looking for the crashlog buffer but can't remember if I found it#asahi-dev · 2022-01-21 07:47
svendon't remember how it was called#asahi-dev · 2022-01-21 07:47
svenit certainly contains the firmware sections#asahi-dev · 2022-01-21 07:46
svenerm. s/pre-allocated buffers/pre-mapped crap inside DART or whatever/#asahi-dev · 2022-01-21 07:45
sventhe ADT also contains all the pre-allocated buffers inside last time i checked#asahi-dev · 2022-01-21 07:42
svensounds good to me. would be great if we know that whatever API we end up with works for NVME/SMC and DCP before we submit this upstream#asahi-dev · 2022-01-21 07:41
sven_jannau_: nice, then what I just wrote should be enough to make the crashlog parsing work :-)#asahi-dev · 2022-01-21 07:37
sven(and i just remembered that I wanted to write something about those. feel free to remind me this evening ;))#asahi-dev · 2022-01-21 07:37
svenotherwise you may have to fix those first :D#asahi-dev · 2022-01-21 07:36
svenhrm.. does iova_to_phys work with the DART hacks?#asahi-dev · 2022-01-21 07:34
svenand then just do whatever is required directly in the driver#asahi-dev · 2022-01-21 07:34
svenand just have a single buffer_request and buffer_free callback that takes a pointer to apple_rtkit_shmem #asahi-dev · 2022-01-21 07:34
svenso what i would do is move struct apple_rtkit_shmem to linux/soc/apple/rtkit.h, add a is_syslog bool in there and then drop rtkit_ops->flags completely#asahi-dev · 2022-01-21 07:33
svenlooks reasonable to me. i guess DCP doesn't actually use that oslog endpoint to send any messages?#asahi-dev · 2022-01-21 07:32
svennice!#asahi-dev · 2022-01-21 07:30
sven*changes#asahi-dev · 2022-01-21 07:30
svenmarcan: have you done any changed to rtkit when you wrote SMC and already pushed those somewhere?#asahi-dev · 2022-01-21 05:50
svenor maybe even just have the flag global since so far we haven’t seen a coproc that mixes iomem and ram buffers