#asahi-re · 2021-01-21 13:43
never_releasedit's behaving as designed.#asahi-re · 2021-01-21 13:42
never_releasedwell#asahi-re · 2021-01-21 13:42
never_releasedit isn't broken#asahi-re · 2021-01-21 13:41
davidrysk[m](Or the design that leads to this behavior)#asahi-re · 2021-01-21 13:41
davidrysk[m]The big question: is this specific nGnRE breakage intentional?#asahi-re · 2021-01-21 13:40
never_releasedthe workaround is broken, as Apple actually uses nGnRE for some _other_ things#asahi-re · 2021-01-21 13:40
never_releasedwill explain later#asahi-re · 2021-01-21 13:40
never_releaseddavidrysk[m]: even more complex than this#asahi-re · 2021-01-21 13:32
opticronis the Apple device tree on the wiki somewhere?#asahi-re · 2021-01-21 13:31
opticronmaybe they use standard ARM CCN?#asahi-re · 2021-01-21 13:30
opticronthe devices themselves, not the regs#asahi-re · 2021-01-21 13:30
opticronI'm sure it's in the device tree somewhere#asahi-re · 2021-01-21 13:30
davidrysk[m]e.g. what registers are used to configure that#asahi-re · 2021-01-21 13:29
opticronoh...I guess if the interconnects are smart enough they could just do it there#asahi-re · 2021-01-21 13:26
davidrysk[m]Someone did say that they do “fabric config”#asahi-re · 2021-01-21 13:26
davidrysk[m]I’m really curious what’s really up with nGnRE. Does Apple just never use it in their drivers? They do provide a way to map that memory#asahi-re · 2021-01-20 11:57
j`eyhow was that chicken bit sequence found?#asahi-re · 2021-01-20 06:15
maximus64we can install ktext and dump out the kernel page table and verify#asahi-re · 2021-01-20 06:14
svenbut i'd expect the kernel to just map everything as nGnRnE :s#asahi-re · 2021-01-20 06:13
svenhrm, maybe the quirk doesn't apply to all of MMIO but only to some of it. or maybe there's some magic memory poke to fix that issue somewhere in xnu.#asahi-re · 2021-01-20 02:52
hthh_yeah - no idea - it's weird how much they can get away with when it's just their code on their hardware#asahi-re · 2021-01-20 02:50
davidrysk[m]is it possible that nothing in macOS uses that? I guess#asahi-re · 2021-01-20 02:49
davidrysk[m]for IOMemory mapping#asahi-re · 2021-01-20 02:49
davidrysk[m]IOKit exposes a kIOPostedWrite type which maps to nGnRE#asahi-re · 2021-01-20 02:16
davidrysk[m]that's normal memory though#asahi-re · 2021-01-20 02:15
davidrysk[m]configures*#asahi-re · 2021-01-20 02:15
davidrysk[m]but xnu also declares two more memory types (CACHE_ATTRINDX_POSTED_REORDERED and CACHE_ATTRINDX_POSTED_COMBINED_REORDERED) and configured them accordingly#asahi-re · 2021-01-20 02:14
davidrysk[m]xnu maps MT_DEVICE_nGnRE memory as nGnRE#asahi-re · 2021-01-20 02:09
hthh_i assume xnu just maps memory nGnRnE not nGnRE?#asahi-re · 2021-01-20 02:09
davidrysk[m]they do set more things in the MAIR1 register though#asahi-re · 2021-01-20 02:08
davidrysk[m]Apple doesn't have any workaround in their t8101 xnu for this, that I can tell#asahi-re · 2021-01-20 02:06
brentr123[m]Either that or I’m confused#asahi-re · 2021-01-20 02:06
brentr123[m]I believe he renamed it to something else#asahi-re · 2021-01-20 02:03
davidrysk[m]What did marcan end up doing to work around the nGnRE issue?#asahi-re · 2021-01-20 01:37
Shizit's prelinked, you can just jam it into ida afaik#asahi-re · 2021-01-20 01:31
hthh_so i grab from the ipsw (from ipsw.me, which you can unzip), but that kernel loads in ida fine for me too (although it's just xnu, without the drivers)#asahi-re · 2021-01-20 01:30
davidrysk[m]ah, just have to disassemble all the .text sections#asahi-re · 2021-01-20 01:28
davidrysk[m]which isn't an img4#asahi-re · 2021-01-20 01:28
davidrysk[m]I'm looking at /System/Library/Kernels/kernel.release.t8101#asahi-re · 2021-01-20 01:28
davidrysk[m]from where?#asahi-re · 2021-01-20 01:27
hthh_(probably not helpful, but i unpack with xerub/img4lib then just load into ida)#asahi-re · 2021-01-20 01:25
davidrysk[m]https://github.com/vampirecat35/decompkernelcache#asahi-re · 2021-01-20 01:24
davidrysk[m]how does one disassemble the kernel? otool -tV doesn't like it#asahi-re · 2021-01-18 19:42
marcanit's not an x86 CPU#asahi-re · 2021-01-18 19:42
marcanno, how would that work?#asahi-re · 2021-01-18 19:20
suskunOfftop question: i wonder does m1 hypervisor support x86 isa?#asahi-re · 2021-01-18 18:00
modwizcodeI thought you were talking about the IRQ patch and I was like "idk it makes sense to me to treat IPIs similar to normal interrupts"#asahi-re · 2021-01-18 18:00
modwizcodeOh that's nasty#asahi-re · 2021-01-18 17:58
marcanwe're starting fresh (and we will keep up with upstream) so we won't have to deal with that mess#asahi-re · 2021-01-18 17:58
marcanrebasing after being out of date by so much takes forever