#
#asahi-dev · 2026-06-08 09:14
StanfordAppliedCyber[m]LOL
#asahi-dev · 2026-06-08 09:12
chaos_princessok, i have a bad suggestion - buy a m1, and see what happens there
#asahi-dev · 2026-06-08 09:12
StanfordAppliedCyber[m]it's just not responding
#asahi-dev · 2026-06-08 09:12
StanfordAppliedCyber[m]chaos_princess: it's not crashing
#asahi-dev · 2026-06-08 09:12
Sobek[m]chaos_princess: Ahhh, yeah, that rings a bell.
#asahi-dev · 2026-06-08 09:12
chaos_princessStanfordAppliedCyber[m]: look at how exactly it crashes
#asahi-dev · 2026-06-08 09:12
StanfordAppliedCyber[m]yeah, we can find a way to get the hypervisor working just by emulating the registers and lying to userspace that it's in a VM
#asahi-dev · 2026-06-08 09:11
chaos_princessSobek[m]: isn't it some fujitsu supercomputer, so everything that runs on it will be optimized for specific microarch
#asahi-dev · 2026-06-08 09:11
sven*sigh*
#asahi-dev · 2026-06-08 09:11
StanfordAppliedCyber[m]it can't hurt, we're not abusing the system
#asahi-dev · 2026-06-08 09:11
Sobek[m]sven: So, you think there a chance Apple re-enable it eventually, while keeping their bigger more complex security extensions which only macOS uses disabled ?
#asahi-dev · 2026-06-08 09:11
svenI’d just file a radar for tso and see what happens
#asahi-dev · 2026-06-08 09:11
StanfordAppliedCyber[m]chaos_princess: what channels do we have to view sep errors? i know abotu the error and status registers, anything else?
#asahi-dev · 2026-06-08 09:11
nickchanMIDR_EL1 check
#asahi-dev · 2026-06-08 09:10
Sobek[m]sven: That's true ! But do those machine expose to program that they guarantee TSO ?
#asahi-dev · 2026-06-08 09:10
chaos_princesshahaha
#asahi-dev · 2026-06-08 09:10
nickchanthis not recommended is not there previously
#asahi-dev · 2026-06-08 09:09
nickchanfor 0b1111
#asahi-dev · 2026-06-08 09:09
sventhere are arm machine with tso by default
#asahi-dev · 2026-06-08 09:09
nickchanPMUVer now says: IMPLEMENTATION DEFINED form of performance monitors supported, PMUv3 not supported. Arm does not recommend this value for new implementations.
#asahi-dev · 2026-06-08 09:09
nickchanhttps://developer.arm.com/documentation/111107/2026-03/AArch64-Registers/ID-AA64DFR0-EL1--AArch64-Debug-Feature-Register-0
#asahi-dev · 2026-06-08 09:09
Sobek[m]Iirc, the ARM memory model is relaxed, and ARM don't want people to write ARM code that doesn't work on that memory model because they rely on the stricted TSO memory model (whose spec by Intel and AMD is a bit informal too, iirc).
#asahi-dev · 2026-06-08 09:08
sventso disabled might be an accident
#asahi-dev · 2026-06-08 09:08
nickchan(this page is after the lkml arguments)
#asahi-dev · 2026-06-08 09:07
nickchanapple did mention https://developer.apple.com/documentation/virtualization/accelerating-the-performance-of-rosetta TSO here but they did not clearly define TSO either other than "it works with rosetta"
#asahi-dev · 2026-06-08 09:07
StanfordAppliedCyber[m]............................
#asahi-dev · 2026-06-08 09:06
Sobek[m]StanfordAppliedCyber[m]: Is it ?
#asahi-dev · 2026-06-08 09:06
nickchanTSO is a switch in the spec says implementation defined register that switch between two spec compliant memory model
#asahi-dev · 2026-06-08 09:06
chaos_princesswith a secondary of "define TSO"
#asahi-dev · 2026-06-08 09:06
chaos_princesssomeone haven't read lkml, since that is exactly the reasoning they used
#asahi-dev · 2026-06-08 09:06
nickchanTSO is allowed by the arm spec, not in it
#asahi-dev · 2026-06-08 09:06
StanfordAppliedCyber[m](it's 2am in california lol, i need to stop debugging sep and go to bed)
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m]oh i thought tso was in the arm spec, could have hallucinated that
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m]it's not like tso is going to massively fragment the arm ecosystem
#asahi-dev · 2026-06-08 09:05
chaos_princess> "and in the arm spec" - huh????
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m](i think?)
#asahi-dev · 2026-06-08 09:05
nickchannot like that matters for linux though so set that aside
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m]but tso is both useful, and in the arm spec
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m]linux was never going to use all of apple's weird custom stuff
#asahi-dev · 2026-06-08 09:05
StanfordAppliedCyber[m]like the rest of the sysregs, whatever
#asahi-dev · 2026-06-08 09:05
nickchanwell the issue is custom algo register aren't accessible in non entitled mode yet the arm approved id registers say there should be impdef algo
#asahi-dev · 2026-06-08 09:04
StanfordAppliedCyber[m]yeah tso is the most annoying one i'd argue
#asahi-dev · 2026-06-08 09:04
Sobek[m]chadmed: So Apple, is clearly trying to cooperate and not fragment the ISA when it can be avoided. But apparently, TSO mode seems to be a dealbreaker for ARM, even if only ever used for x86 emulation.
#asahi-dev · 2026-06-08 09:03
StanfordAppliedCyber[m]oh, well, trustedaccessory isn't actually gated on bluetooth, as long as the AppleSEPDeviceService is active trustedaccessory will try to talk to it
#asahi-dev · 2026-06-08 09:03
chaos_princesspac allows custom algo
#asahi-dev · 2026-06-08 09:03
nickchanotherwise it would be a violation of saying impdef exists but there's no impdef
#asahi-dev · 2026-06-08 09:02
nickchanunless they managed to somehow change ID_AA64ISAR1_EL1.API
#asahi-dev · 2026-06-08 09:02
nickchanBut by that logic, they can't disable Apple PAC either
#asahi-dev · 2026-06-08 09:01
StanfordAppliedCyber[m]if i can just turn off bluetooth that would indeed be nice
#asahi-dev · 2026-06-08 09:01
StanfordAppliedCyber[m]thanks for the help!

plus ›