#asahi-dev · 2024-08-24 15:23
nickchanel3 support in m1n1 for spin-table on a9 is done now, but 4096 translation granule sizes are still proving tricky for a7-a8#asahi-dev · 2024-08-24 15:15
nickchanwith some use_local_fast_ipi static branch most likely#asahi-dev · 2024-08-24 15:15
nickchanso the driver will have to be modified with a new apple,t8015-aic compatible#asahi-dev · 2024-08-24 15:14
nickchanafter some testing, it seems that fast IPIs are supported on A11, however local fipi via SYS_IMP_APL_IPI_RR_LOCAL_EL1 is not supported, and all fipi must go through the global SYS_IMP_APL_IPI_RR_GLOBAL_EL1#asahi-dev · 2024-08-20 18:48
nickchanam I right that Documentation/devicetree/bindings in the kernel is only used for documentation and has no effect on how the kernel resolve compatibles#asahi-dev · 2024-08-16 19:46
nickchanokay got it#asahi-dev · 2024-08-16 19:44
nickchani need to debug mmu init on t7000#asahi-dev · 2024-08-16 19:44
nickchananother question how do I call mmu_init with the proxyclient? M1N1Proxy.mmu_init() wants a position argument self#asahi-dev · 2024-08-16 17:24
nickchanah right was trying to find in in cpu specific chickens forgot that every core needs it#asahi-dev · 2024-08-16 17:20
nickchanin m1n1 where is the code that sets the WFI mode in SYS_APL_CYC_OVRD_EL1?#asahi-dev · 2024-08-15 07:29
nickchanthis has happened before in ios 13 where the length changed from 256 to 608#asahi-dev · 2024-08-15 07:28
nickchanand when try to map with mem_size_actual of 0 everything explodes!#asahi-dev · 2024-08-15 07:28
nickchanso the original mem_size_actual is now part of the cmdline#asahi-dev · 2024-08-15 07:28
nickchanjannau: regarding not being able to enable mmu with the macOS 15 stub, in iOS 18/macOS 15 apple changed cmdline length in bootargs struct to 0x400 bytes long#asahi-dev · 2024-08-14 10:10
nickchan*S8001 i meant#asahi-dev · 2024-08-14 10:08
nickchanboth EL1 and EL3 seems to work but running in EL3 is very cursed because it's still modifying the EL1 registers#asahi-dev · 2024-08-14 10:08
nickchanthis time it's about adding a has ecore check in the is ecore check#asahi-dev · 2024-08-14 10:08
nickchanhave working bringup in 8003 too#asahi-dev · 2024-08-14 05:04
nickchanpossibly need to get the tunables right#asahi-dev · 2024-08-14 05:03
nickchanbringup build with no mmu works fine on 8010 though with dwc2 usb#asahi-dev · 2024-08-14 04:58
nickchanfor 8010 though I still need to figure out some issue with enable MMU and for 8003 and older those have EL3 which also need special handling#asahi-dev · 2024-08-14 04:55
nickchan(A7, iPhone 5s, 2013)#asahi-dev · 2024-08-14 04:55
nickchanit seems like the m1 aic is backwards compatible all the way to 8960#asahi-dev · 2024-08-14 04:54
nickchanlike gxf and sprr#asahi-dev · 2024-08-14 04:54
nickchanaside from dumb registers write to make the CPU sane and the USB driver it's mostly about making it skip unsupported parts#asahi-dev · 2024-08-14 04:52
nickchanwhich does not support them#asahi-dev · 2024-08-14 04:52
nickchanThe 8015 does support fast ipi and upmcr0 registers in theory though I disabled those for now since I was doing some testing on 8010 and older#asahi-dev · 2024-08-14 04:52
nickchantuanbles is also different etc#asahi-dev · 2024-08-14 04:51
nickchanand SEPFW/empty doesn't follow so you can see it complain about unknown payload in the log#asahi-dev · 2024-08-14 04:51
nickchanto start with m1n1 is dumped to near the end of DRAM#asahi-dev · 2024-08-14 04:51
nickchanthe previous stage is pongoOS, the memory layout is definitely different#asahi-dev · 2024-08-14 04:45
nickchanfiguring out CPU start without hv will be... interesting#asahi-dev · 2024-08-14 04:38
nickchanbasically about writing dwc2 driver #asahi-dev · 2024-08-14 04:37
nickchanhttps://gist.github.com/asdfugil/e4403b4424620ee792367af6cba6304d brought up m1n1 on 8015 more details soon hopefully#asahi-dev · 2024-07-14 14:52
nickchanthey also said that only macho boot objects should be ran with sptm#asahi-dev · 2024-07-14 14:51
nickchanjannau: someone said running a raw image should be without sptm, so maybe sptm is not running after all#asahi-dev · 2024-07-14 08:53
nickchanand this order of things getting SPTM makes sense when you consider iOS as the most important to apple, followed by iPadOS, then macOS; and that newer devices are more important than older ones.#asahi-dev · 2024-07-14 08:51
nickchanlike moving Core Bringup to it#asahi-dev · 2024-07-14 08:51
nickchanit is all an enginnering time thing that it is not on other devices yet since for SPTM to be a functioning hypervisor there is a lot of stuff beyond renaming PPL to SPTM#asahi-dev · 2024-07-14 08:49
nickchanchadmed: sptm is not a thing yet on j314 since it is the m1 pro and for mac it is on M2 on macOS 15 and nothing else#asahi-dev · 2024-07-14 08:47
nickchanneed SPTM as the hypervisor to run Exclave#asahi-dev · 2024-07-14 08:47
nickchanyeah every exception level is used now GL2: SPTM; EL2: XNU; GL1: ExclaveCore; EL1: Guest Kernel; GL0: ExclaveOS; EL0: macOS userland#asahi-dev · 2024-07-14 08:45
nickchanit is expected that it will eventually be introduced to M1/A14+#asahi-dev · 2024-07-14 08:44
nickchanA15-A17: SPTM introduced in iOS 17; M4: SPTM launched with iPadOS 17.5; M2: SPTM introduced in iOS 18/macOS 15#asahi-dev · 2024-07-14 08:41
nickchanthe fact that it is not in the kernel image is the problem since the code execution primitive we have on apple silicon is being able to replace that#asahi-dev · 2024-07-14 08:40
nickchanit has the highest privilege in macOS, just like PPL#asahi-dev · 2024-07-14 08:39
nickchanSPTM stands for Secure Page Table Monitor, it is a PPL replacement but unlike PPL it is not part of the kernel image#asahi-dev · 2024-07-14 08:18
nickchanAlso under SPTM, XNU cannot ever run with MMU disabled, as it is no longer physically contiguous in memory.#asahi-dev · 2024-07-14 08:15
nickchanhas anyone checked whether m1n1 boots with a macOS 15 beta stub on M2? In macOS 15 on M2, SPTM is introduced to the M2 (and ONLY M2, not any of the Pro/Max/Ultra or M3). The XNU would be loaded in EL2 while GL is enabled, this will break everything in m1n1 unless Apple has special logic when flat arm64 executables are specified in kmutil.