#
✕ nick:amarioguy
#asahi-re · 2026-07-18 21:17
amarioguybtw important note, i think a bunch of them being 0 is often symptomatic of those registers normally triggering SErrors
#asahi-re · 2026-07-18 21:10
amarioguyneed to work on some other stuff now, but will respond later to stuff that comes up
#asahi-re · 2026-07-18 21:08
amarioguyplease ping if anyone notices issues with the methodology or with the actual dumps themselves
#asahi-re · 2026-07-18 21:08
amarioguymy experiment python file i used for doing the dumps
#asahi-re · 2026-07-18 21:08
amarioguyhttps://gist.github.com/amarioguy/d16d92227140d098a6af37058c93c0bf
#asahi-re · 2026-07-18 21:05
amarioguyfor 25G5028f
#asahi-re · 2026-07-18 21:05
amarioguyhttps://mega.nz/file/A44mkLDD#i8M1-8cFjg8rkij44wbzoKpPYIEzloUlRba9GifGxfI
#asahi-re · 2026-07-18 21:05
amarioguyfor 25G5043d
#asahi-re · 2026-07-18 21:05
amarioguyhttps://mega.nz/file/UgRjHYbS#mv5FlT4NBxMPsGbH0TXS_-I0y-6APDA1kw2UiDnLwxY
#asahi-re · 2026-07-18 20:51
amarioguy(pastebin itself seems obvious...but unsure if it's that great ngl since iirc it started having some login walls)
#asahi-re · 2026-07-18 20:50
amarioguyany recommended pastebin site?
#asahi-re · 2026-07-18 20:49
amarioguyraw files going up shortly (they don't fit in a github gist)
#asahi-re · 2026-07-18 20:49
amarioguyi hope this is helpful for people wanting to RE the sysreg lock mechanism
#asahi-re · 2026-07-18 20:49
amarioguyaccesses to 0x211ea2000 seemed to hang, i skipped to 0x211f00000 as i was unsure if other regs in that region also hung m1n1 and didn't have the time to test those in between
#asahi-re · 2026-07-18 20:48
amarioguyhere's the diff of that dumped region between the builds
#asahi-re · 2026-07-18 20:48
amarioguyhttps://gist.github.com/amarioguy/8d3400e59e5dc24ee36e9afe642e9e03
#asahi-re · 2026-07-18 20:48
amarioguyalright
#asahi-re · 2026-07-18 20:44
amarioguyah -u makes it more intuitive looking, gotcha, --color as well
#asahi-re · 2026-07-18 20:38
amarioguyi don't imagine the standard diff a.txt b.txt is that intuitive...
#asahi-re · 2026-07-18 20:38
amarioguyok got both dumps, what diff format do you all prefer
#asahi-re · 2026-07-18 20:17
amarioguyi'll post the diff first (in a gist), then i'll send the full raw files separately
#asahi-re · 2026-07-18 20:16
amarioguy34 meg file....
#asahi-re · 2026-07-18 19:39
amarioguyyea, sync exceptions are handled well, those recover just fine
#asahi-re · 2026-07-18 19:38
amarioguydid a trial run and some regs were giving sync exceptions, unsure if i should leave those on or not, whatever's more useful
#asahi-re · 2026-07-18 19:38
amarioguychaos_princess, yuka: going to post the register dumps to gists along with the diff (between the two beta builds, this is related to RE'ing the mechanism for sysreg locks), should i leave sync exceptions on in DAIF or not?
#asahi-re · 2026-07-18 19:26
amarioguyat least i can dump the registers now, so i'll take it i suppose
#asahi-re · 2026-07-18 19:26
amarioguyyuka: an update ^
#asahi-re · 2026-07-18 19:26
amarioguy(when i unmasked the serrors, i started getting that flood of serrors again)
#asahi-re · 2026-07-18 19:25
amarioguytried an experiment where i disabled SErrors in DAIF, read one of the known bad addresses, it read back 0 for it (i doubt it actually read anything however it works at least for suppressing the problem here)
#asahi-re · 2026-07-18 19:24
amarioguyso ummmmmmmmmmm seems like disabling serrors entirely works?
#asahi-re · 2026-07-18 19:07
amarioguyor is there a better way where m1n1 doesn't keep looping the serror
#asahi-re · 2026-07-18 19:07
amarioguynot saying that those registers do exist, more trying to figure out...is doing elr += 4 a good idea for the read8/read16/read32/read64 case specifically?
#asahi-re · 2026-07-18 19:05
amarioguyi did try 32-bit access width on this same addr, lower nibble became 3 but otherwise not much different)
#asahi-re · 2026-07-18 19:04
amarioguythe INF lower bit values seem to be consistent as well
#asahi-re · 2026-07-18 19:04
amarioguyfor example 0x210F80040 (from iBoot disasm) exists and returns a good value but 0x2110400D0 does not
#asahi-re · 2026-07-18 19:03
amarioguythat's an example but most things that aren't IORVBAR or very specific addresses which are known to exist seem to throw these
#asahi-re · 2026-07-18 19:03
amarioguyESR: 0xbe000000 (SError)
#asahi-re · 2026-07-18 19:03
amarioguyL2C_ERR_INF: 0x1400100007
#asahi-re · 2026-07-18 19:03
amarioguyL2C_ERR_ADR: 0x1a80000211010000
#asahi-re · 2026-07-18 19:03
amarioguyL2C_ERR_STS: 0x200
#asahi-re · 2026-07-18 19:01
amarioguybut yes, in the general case you're correct
#asahi-re · 2026-07-18 19:01
amarioguysee it turns out on apple platforms (some previous experience with some m1n1 testing a few years ago gave me similar issues too) accessing some memory addresses wrong basically throws an serror that's synchronous with the instruction
#asahi-re · 2026-07-18 19:00
amarioguyyea..............if only it were that simple
#asahi-re · 2026-07-18 18:57
amarioguyso every time it was doing the ERET, it was returning to the exact same instruction that was SError'ing
#asahi-re · 2026-07-18 18:56
amarioguyi can PR that small addition if upstream wants it though
#asahi-re · 2026-07-18 18:56
amarioguyi say "in quote marks" because i'm sure there's probably a good reason serror handling doesn't have the exception guard code
#asahi-re · 2026-07-18 18:55
amarioguymy "fix" in quote marks is basically just to add the logic from sync exceptions for the purposes of dumping these registers
#asahi-re · 2026-07-18 18:55
amarioguyyuka: tl;dr is that serrors don't have the exception guard code that the synchronous exception path does
#asahi-re · 2026-07-18 18:51
amarioguyah i see why simple local fix should do it
#asahi-re · 2026-07-18 17:47
amarioguy(the problem isn't necessarily the serrors themselves, more that m1n1 seems to be struggling to recover from them...)

plus ›