#
#asahi-re · 2026-07-01 22:34
StanfordAppliedCyber[m]yes i can verify everything sven + chaos said
#asahi-re · 2026-07-01 17:42
mischa85hmmm. maybe i need to apply the tunables for things to work out
#asahi-re · 2026-07-01 16:39
mischa85now onto ans3 firmware shenanigans
#asahi-re · 2026-07-01 16:39
mischa85i was trying to write to the admin-queue registers directly, the alias +0x40000000 works
#asahi-re · 2026-07-01 11:18
sven+see
#asahi-re · 2026-07-01 11:17
svenif you serrors there’s likely just some bringup missing
#asahi-re · 2026-07-01 11:17
svenyou’ll be able to reach everything from el2
#asahi-re · 2026-07-01 11:17
svenif you run Linux or m1n1 none of that applies
#asahi-re · 2026-07-01 11:16
svenyup, it’s literally only pagetables setup in a way that only allows gl2 access to the nvmmu
#asahi-re · 2026-07-01 10:18
chaos_princessthe protection is only via page tables that are set up from gl2
#asahi-re · 2026-07-01 10:18
chaos_princessfrom what we know, there is no hardware that is capable of discriminating between gl/el
#asahi-re · 2026-07-01 10:18
mischa85StanfordAppliedCyber[m]: so i can't tell if there's a fabric/dart setup that opens the reg for plain el2 or if it's strictly the guarded attribute
#asahi-re · 2026-07-01 10:17
mischa85StanfordAppliedCyber[m]: a plain el2 str to bar+0x24 faults here (L2C_ERR 0x80)
#asahi-re · 2026-07-01 10:17
mischa85StanfordAppliedCyber[m]: when you write the admin-queue regs from el2, is guarded/gxf state actually active during the store, or is it a genuinely plain el2 str?
#asahi-re · 2026-07-01 08:59
chadmedM3 progress, Apple bugs, and more! https://asahilinux.org/2026/06/progress-report-7-1/
#asahi-re · 2026-06-30 23:45
StanfordAppliedCyber[m]they're writable from el2, source: currently doing so
#asahi-re · 2026-06-30 22:34
mischa85exit
#asahi-re · 2026-06-30 22:31
yukaI'm all ears
#asahi-re · 2026-06-30 22:30
mischa85yes i feel close to figuring it out. i know what to try next
#asahi-re · 2026-06-30 22:28
yukait's definitely possible to use the nvme from raw boot mode, since the hypervised mac os of StanfordAppliedCyber already does that
#asahi-re · 2026-06-30 22:27
yukamischa85: wdym?
#asahi-re · 2026-06-30 22:26
mischa85hmmm. the plot thickens
#asahi-re · 2026-06-30 22:23
yukais the most likely explanation
#asahi-re · 2026-06-30 22:23
yukathey moved on m4
#asahi-re · 2026-06-30 22:20
mischa85are those the standard offsets or did they move on ans3?
#asahi-re · 2026-06-30 22:14
chaos_princessel2
#asahi-re · 2026-06-30 22:12
mischa85on m4, are the nvme admin-queue regs writable from el2, or only from gl2?
#asahi-re · 2026-06-30 10:29
Sobek[m]nicolas17: I.e. a Tenured Faculty, not a tenure track one. (The only bad thing that can happen to them now is not being promoted, they have job security already).
#asahi-re · 2026-06-30 08:35
JamesCalligeros[m]We have some group chanop thing but i forgor how it works
#asahi-re · 2026-06-30 08:35
StanfordAppliedCyber[m]lol
#asahi-re · 2026-06-30 08:35
StanfordAppliedCyber[m]i was wondering why your messages weren't showing up
#asahi-re · 2026-06-30 08:34
JamesCalligeros[m]Oh my irc nick is still banned from this room from when i was using iwd lol
#asahi-re · 2026-06-30 08:20
StanfordAppliedCyber[m]that would piss me off also
#asahi-re · 2026-06-30 08:20
StanfordAppliedCyber[m]I do think there are correct ways to use them
#asahi-re · 2026-06-30 08:20
StanfordAppliedCyber[m]oh yeah i'm fully with you that using llms poorly is terrible
#asahi-re · 2026-06-30 08:10
chadmedim sick to death of getting slop-augmented requirements documents that make absolutely no sense because the person we got to gather the requirements did not understand the task at hand and had claude do it for them, and then having to schedule more 4 hour meetings with the customer to actually get the requirements properly
#asahi-re · 2026-06-30 08:10
chaos_princessmaybe kinda useful if you use it as "hey, grok, which part does $whatever", and then go look yourself
#asahi-re · 2026-06-30 08:09
chaos_princessi've found the requirement to re-check everything the parrot says result in it not being an obvious speedup
#asahi-re · 2026-06-30 08:09
chadmedwe have people who use them at $work and all they seem to do is create more annoying rework and fixing for those of us who refuse to use them :P
#asahi-re · 2026-06-30 08:08
StanfordAppliedCyber[m]i've found use for them in other domains (not asahi obviously)
#asahi-re · 2026-06-30 08:08
StanfordAppliedCyber[m]but if it's 90% correct and 10% bs it can be useful
#asahi-re · 2026-06-30 08:07
StanfordAppliedCyber[m]which they will, not denying that
#asahi-re · 2026-06-30 08:07
chaos_princessand if you got the thing to write the "docs", you don't know which part is a mistake
#asahi-re · 2026-06-30 08:07
chaos_princessStanfordAppliedCyber[m]: unless they fuck up in subtle ways.
#asahi-re · 2026-06-30 08:06
StanfordAppliedCyber[m]not to say that the person used them well (i haven't looked at the paper)
#asahi-re · 2026-06-30 08:06
chaos_princessJamesCalligeros[m]: oh, i skipped that section for obvious reasons :P
#asahi-re · 2026-06-30 08:06
StanfordAppliedCyber[m]fwiw llms are actually very good at looking through large blobs of asm and writing docs about it
#asahi-re · 2026-06-30 08:05
JamesCalligeros[m]The section about dart reads like someone got claude to summarise our driver or something like that
#asahi-re · 2026-06-30 08:03
chaos_princessi wouldn't call it fully bs, there is some useful stuff there.
#asahi-re · 2026-06-30 03:21
JamesCalligeros[m]oh god

plus ›