#asahi-dev · 2026-11-09 21:59
mmediouni[m]KernelSupport alone is such a travesty unfortunately#asahi-dev · 2026-11-09 21:13
mmediouni[m]https://github.com/apple-oss-distributions/xnu/blob/f6217f891ac0bb64f3d375211650a4c1ff8ca1ea/bsd/dev/arm64/sysctl.c#L331#asahi-dev · 2026-11-09 21:13
mmediouni[m]Fun errata with retries (!) in OSS XNU code#asahi-dev · 2026-11-09 21:11
mmediouni[m]vm_compressor_force_sw_wkdm#asahi-dev · 2026-11-09 21:11
mmediouni[m]wksw_force#asahi-dev · 2026-11-09 21:03
mmediouni[m]I think there's a boot arg to disable the hw wkdm path#asahi-dev · 2026-11-09 21:01
mmediouni[m]so apple switched to KTRR#asahi-dev · 2026-11-09 21:01
mmediouni[m]But EL3 just didn't have enough at the time to have a good story to protect the kernel - EL3 FGT came much later#asahi-dev · 2026-11-09 20:59
mmediouni[m]Watchtower back in the pre-KTRR days did a dance where it either trapped neon execution to EL3 or trapped interrupts to be somewhat guaranteed to run constantly without much of a perf impact#asahi-dev · 2026-11-09 20:56
mmediouni[m]Or use MDCR_EL2.TDE and do a breakpoint instr#asahi-dev · 2026-11-09 20:56
mmediouni[m]Yes#asahi-dev · 2026-11-09 20:53
mmediouni[m]sven: a trick to go EL0 to EL2 (if not VHE in-host) is a page mapped at stage 1 but not at stage 2#asahi-dev · 2026-10-26 00:28
mmediouni[m]Think that answers one of the LKML questions on what does Apple TSO mode actually mean#asahi-dev · 2026-10-26 00:23
mmediouni[m]https://www.sigarch.org/compound-memory-models/#asahi-dev · 2026-10-25 22:30
mmediouni[m]Apple and Android-land most likely going to stick with pMTE#asahi-dev · 2026-10-25 22:29
mmediouni[m]MS asked for FEAT_VMTE and x86's memory tagging extension is also virtual memory-based#asahi-dev · 2026-10-25 22:28
mmediouni[m]lower priority item, I expect it to come at some point#asahi-dev · 2026-10-25 22:27
mmediouni[m]sven: virtual MTE (on virtual memory instead of physical, so that you can have different tags on the same underlying physical page depending on the view) is an ARMv10 feature - FEAT_VMTE#asahi-dev · 2026-10-25 22:25
mmediouni[m]tags are stored in a tag memory region that's accessible only to SPTM but with a twist. If the pages that reference that tag page are not using MTE, then the underlying MTE tag page is usable as regular memory#asahi-dev · 2026-10-25 22:24
mmediouni[m]for macs#asahi-dev · 2026-10-25 22:24
mmediouni[m]chaos_princess: no MTE exposed through virtualization right now#asahi-dev · 2026-10-20 22:06
mmediouni[m]ugh, AI everywhere#asahi-dev · 2026-10-20 22:06
mmediouni[m]I thought that USB4 on ACPI plats on arm64 would be fun to work on but... omarchy already has it working on their tree...#asahi-dev · 2026-10-15 11:38
mmediouni[m]Not sure how much of this applies when it's exclusively for HW that has a license anyway#asahi-dev · 2026-10-15 11:36
mmediouni[m]What's Mesa's default?#asahi-dev · 2026-10-08 18:05
mmediouni[m]ESR_ELx being a mix of impdef and architectural bits is a bit intriguing though, I'd have expected Arm to punt the non-architectural bits to another MSR#asahi-dev · 2026-10-08 17:59
mmediouni[m]There'd have been plenty of barriers way before you get to another guest so it'd have been caught by then#asahi-dev · 2026-10-08 17:56
mmediouni[m]tldr is that if ESR bit 22 is there on an SError (along with the impdef bit) - then it's actually a virtual SError to put back to the guest#asahi-dev · 2026-10-08 13:51
mmediouni[m]nickchan: sent over DM, they're using an impdef bit in ESR as far as I can tell#asahi-dev · 2026-10-08 13:25
mmediouni[m]Timers are where things shift outright to crazy though...#asahi-dev · 2026-10-08 13:24
mmediouni[m]They do this thing where if it's vGIC it's always from the guest (tm) - because they only target apple hardware and don't care about anything else#asahi-dev · 2026-10-08 13:16
mmediouni[m]macOS doesn't do this but considers those SErrors as always from the guest#asahi-dev · 2026-10-07 07:14
mmediouni[m]But apple doesn't do that afaik - at least not on any released product#asahi-dev · 2026-10-07 07:13
mmediouni[m]However the vendor can make a SoC where 4KiB pages are only supported at stage1 and stage2 has to use 16KiB pages#asahi-dev · 2026-10-07 07:13
mmediouni[m]*used/not used#asahi-dev · 2026-10-07 07:13
mmediouni[m]If stage1 translation supports 4KB pages, it supports it both when stage2 is supported or not supported#asahi-dev · 2026-10-07 07:12
mmediouni[m]That doesn't change things#asahi-dev · 2026-10-05 11:25
mmediouni[m]jannau: wow, better than the Android pile of hacks from before#asahi-dev · 2026-10-04 14:28
mmediouni[m]and fyi, I did deprecate its= in qemu a while back, the new syntax is msi= (to allow for things like GICv3 + GICv2m)#asahi-dev · 2026-10-04 12:27
mmediouni[m]But filtering by EL would also be wrong. What's conventionally done by other platforms is to filter by PSCI SoC ID but not an option here...#asahi-dev · 2026-10-04 12:26
mmediouni[m]so hvf just zeroes it out (and that's why kernel-irqchip=off,gic-version=3,its=on works fine on a Mac host)#asahi-dev · 2026-10-04 12:25
mmediouni[m]but that's more linked to being heterogenous and macOS not supporting core pinning#asahi-dev · 2026-10-04 12:25
mmediouni[m]Apple deliberately only exposes CPU model 0x0#asahi-dev · 2026-09-26 12:03
mmediouni[m]it could have been even bigger, at least Image Playground got moved to "cloud"/PCC only#asahi-dev · 2026-09-26 12:00
mmediouni[m]the only disable toggle is in screen time...#asahi-dev · 2026-09-26 11:59
mmediouni[m]they want Apple Intelligence to be always-on for everyone yes#asahi-dev · 2026-09-26 11:59
mmediouni[m]jannau: I was told that if you disable all Apple Intelligence toggles through Screen Time and then delete from recovery it doesn't download them again#asahi-dev · 2026-09-26 11:57
mmediouni[m](implementing GXF on top of FGT using non-private isa VMs...)#asahi-dev · 2026-09-26 11:57
mmediouni[m]that's what I did for a cursed project#asahi-dev · 2026-09-26 11:54
mmediouni[m]~22.5GB for the machines with 16GB of RAM or more on M3 and later