#
#asahi-dev · 2026-07-03 11:03
chadmedsptm is tightly tightly couple to xnu's idea of memory management, linux will never ever run under it
#asahi-dev · 2026-07-03 11:03
chaos_princesslinux can't run with sptm
#asahi-dev · 2026-07-03 11:03
chaos_princessthis will not happen
#asahi-dev · 2026-07-03 11:02
freetolgaUnless apple introduces some HAL in sptm layer that every os can simply consume and all motivation to RE is lost
#asahi-dev · 2026-07-03 11:02
chadmedit is no more "lokced down" than any other soc, the problem is specifically getting macos to run properly outside of the special "macos boot mode"
#asahi-dev · 2026-07-03 11:02
chaos_princessit will remain a problem for everything from m4 onwards
#asahi-dev · 2026-07-03 11:01
freetolgaIm worried m4 is like maxwell nvidia where its too new and locked down while being too old and locked down
#asahi-dev · 2026-07-03 11:00
chaos_princesssptm is mostly relevant on m4, yes
#asahi-dev · 2026-07-03 11:00
freetolgaFinding how to write drivers and dts
#asahi-dev · 2026-07-03 11:00
freetolgaor general RE progress
#asahi-dev · 2026-07-03 11:00
freetolgaThis is still discussion of "how do we RE m4/a18 pro/m5" right
#asahi-dev · 2026-07-03 09:51
svenweird
#asahi-dev · 2026-07-03 09:51
StanfordAppliedCyber[m]looking at the code next to it there's no way that's an exit instruction
#asahi-dev · 2026-07-03 09:51
StanfordAppliedCyber[m]also it's definitely not just another gexit
#asahi-dev · 2026-07-03 09:51
StanfordAppliedCyber[m]i'm hallucinating but not that bad
#asahi-dev · 2026-07-03 09:51
StanfordAppliedCyber[m]so this is a different instruction
#asahi-dev · 2026-07-03 09:51
StanfordAppliedCyber[m]yeah 0x00201420 is genter
#asahi-dev · 2026-07-03 09:50
svenbut it doesn’t make much sense to have 00201400 for gexit and then 00201401 for something entirely else
#asahi-dev · 2026-07-03 09:50
svenmight be gexit though I don’t think that one takes arguments
#asahi-dev · 2026-07-03 09:49
StanfordAppliedCyber[m]but i can verify it appears in sptm itself, in a codepath adjacent to 0x00201328, i'm literally looking at it in ghidra right now
#asahi-dev · 2026-07-03 09:48
StanfordAppliedCyber[m]wait ofc it takes args, i'm sleepy
#asahi-dev · 2026-07-03 09:48
StanfordAppliedCyber[m]wait no 00201401 is correct, are you sure that's genter with argument 1? i don't think genter takes arguments at all, regardless, that instruction appears directly in the sptm binary, idk why genter would appear in sptm itself
#asahi-dev · 2026-07-03 09:45
StanfordAppliedCyber[m]let me find the right ones
#asahi-dev · 2026-07-03 09:45
StanfordAppliedCyber[m]this is why you don't do work at 3am
#asahi-dev · 2026-07-03 09:45
StanfordAppliedCyber[m]yeah i copy pasted the wrong instructions
#asahi-dev · 2026-07-03 09:45
StanfordAppliedCyber[m]sorry, 0x00201328 was flush by page number
#asahi-dev · 2026-07-03 09:44
sventhink hvc #1 instead of hvc #0
#asahi-dev · 2026-07-03 09:44
sven00201401 is just genter or gexit with argument 1 instead of 0
#asahi-dev · 2026-07-03 09:44
StanfordAppliedCyber[m]no clue, but i don't think it's just tlbs
#asahi-dev · 2026-07-03 09:43
StanfordAppliedCyber[m]in sptm it basically keeps track of dirty pages, if below some threshold it calls 00201320 on all of them, otherwise it calls 00201401 to flush all of them
#asahi-dev · 2026-07-03 09:43
Sobek[m]Flushing from TLBs, or normal caches ?
#asahi-dev · 2026-07-03 09:42
StanfordAppliedCyber[m]tl;dr 00201401 appears to be a flush of all pages, 00201320 appears to be a flush by page number, 00201300 is the same thing by address
#asahi-dev · 2026-07-03 09:42
svenyes, which instruction specifically?
#asahi-dev · 2026-07-03 09:41
StanfordAppliedCyber[m]re: custom cache instructions, sptm has some new instructions present only on m4+, i was discussing them a few days ago
#asahi-dev · 2026-07-03 09:41
StanfordAppliedCyber[m]re: memory mappings, let me think about that, i'm not convinced, though it is currently 2am and my brain isn't functional
#asahi-dev · 2026-07-03 09:39
svenwhat is that „custom cache instruction“ btw?
#asahi-dev · 2026-07-03 09:35
svensecond stage attrs also supersede first stage ones iirc so even without shadow pagetables this would be easy
#asahi-dev · 2026-07-03 09:33
svenuh, no, virtual SPTM without SPRR requires shadow pagetables anyway and we can map things however we want with no patches to the SPTM binary there
#asahi-dev · 2026-07-03 09:31
StanfordAppliedCyber[m]^ it turns out i don't need hardware breakpoints after all
#asahi-dev · 2026-07-03 09:18
StanfordAppliedCyber[m]whereas because i'm emulating sptm i can just always return a NC page for all mappings
#asahi-dev · 2026-07-03 09:17
StanfordAppliedCyber[m]virtualized sptm i think this is a fatal error that we can't get around without some intensive surgery on the sptm binary, which also defeats the point of virtualizing it
#asahi-dev · 2026-07-03 09:17
StanfordAppliedCyber[m]this wouldn't be possible if we were to virtualize sptm btw
#asahi-dev · 2026-07-03 09:15
StanfordAppliedCyber[m]turns out you cant have cache issues if you simply never cache things :)
#asahi-dev · 2026-07-03 09:10
StanfordAppliedCyber[m]got 10 boots in a row without the type 24 fabric access violation returning
#asahi-dev · 2026-07-03 09:10
StanfordAppliedCyber[m]my horrible bump allocator works
#asahi-dev · 2026-07-03 09:10
StanfordAppliedCyber[m]well i managed to sidestep the custom gxf instruction
#asahi-dev · 2026-07-03 07:26
LarstiQStanfordAppliedCyber[m]: ah, so you haven't written an `is_even` function that has cases for the first 100 numbers from 0? :)
#asahi-dev · 2026-07-03 07:04
StanfordAppliedCyber[m]literally just set a hardware breakpoint on the kernelcache location that retypes pages to page table, and then we grab a new page that's marked NC from the start and just give xnu that
#asahi-dev · 2026-07-03 07:02
StanfordAppliedCyber[m]this is some of the dumbest code of my life
#asahi-dev · 2026-07-03 07:02
StanfordAppliedCyber[m]i'm literally writing a bump allocator of noncachable pages right now

plus ›