#
✕ nick:svenpeter[m]
#asahi-dev · 2026-10-28 06:16
svenpeter[m]upstream has already rejected the TSO bit, AMX is much more invasive
#asahi-dev · 2026-08-17 09:41
svenpeter[m]maybe just afk_26.c if the changes are bad enough? 😕
#asahi-dev · 2026-07-06 18:16
svenpeter[m]from #asahi-re:matrix.org
#asahi-dev · 2026-07-06 18:16
svenpeter[m]20:13:24 <flokli> The section starts with "we propose", but I don't see if it's only proposed in the docs, or with upstream.
#asahi-dev · 2026-07-06 18:16
svenpeter[m]Neal Gompa (Fedora): 20:12:50 <flokli> Has there been any discussion with upstream about the /lib/firmware/vendor additional path, as per https://asahilinux.org/docs/platform/open-os-interop/#linux-specific ?
#asahi-dev · 2026-01-13 06:16
svenpeter[m]we should either add that to run_guest.py or virtualize that dockchannel serial I think
#asahi-dev · 2026-01-13 06:15
svenpeter[m]https://gist.github.com/dhinakg/3fcd9ad43c82c96964b4f64eb05e6a5e regarding kernel serial output
#asahi-dev · 2026-01-11 21:23
svenpeter[m]but i don't think we ever managed to enter the deepest sleep state and then wake up from it at RVBAR again
#asahi-dev · 2026-01-11 21:22
svenpeter[m]i think that uses the msr that controls at least part of it
#asahi-dev · 2026-01-11 21:22
svenpeter[m]i don't think so
#asahi-dev · 2026-01-11 20:40
svenpeter[m]oh, and you can't observe it from the hypervisor because systemwide sleep breaks that one and results in a reboot
#asahi-dev · 2026-01-11 20:37
svenpeter[m]maybe nickchan knows more from the iphone world
#asahi-dev · 2026-01-11 20:36
svenpeter[m]but i'm not sure if that's enough to properly enter that state with a wfi
#asahi-dev · 2026-01-11 20:36
svenpeter[m]then there's that wfi control register we mess with and iirc it has a lower state you can configure
#asahi-dev · 2026-01-11 20:35
svenpeter[m]there's some information in open source xnu at its entry point
#asahi-dev · 2026-01-11 20:33
svenpeter[m]i know the core is supposed to come up at RVBAR again but i never saw that happen iirc
#asahi-dev · 2026-01-11 20:33
svenpeter[m]we roughly know how to enter it, but i don't think we ever looked into recovering from that deep sleep
#asahi-dev · 2026-01-11 20:32
svenpeter[m]it's that deep sleep state where you can turn off the entire cluster and once it comes back up it lost every state and jumps back to RVBAR (i.e. wherever m1n1 was during boot)
#asahi-dev · 2026-01-11 20:31
svenpeter[m]yeah, but not into the lowest state
#asahi-dev · 2026-01-11 20:30
svenpeter[m]one thing that i'm aware of that's missing is putting the cpu in their lowest power state, we only very roughly know how that works and there's also quite some effort into getting that upstreamable
#asahi-dev · 2026-01-11 20:29
svenpeter[m]i think chaos_princess has looked into power management in general
#asahi-dev · 2026-01-11 18:47
svenpeter[m]right now i take look at it whenever i remember it exists for maybe 30 minutes
#asahi-dev · 2026-01-11 18:47
svenpeter[m]that bug isn't that bad tbh. if it was actually happening regularly and impacting everything we'd probably figure out what's going on in a week or two of actually debugging that thing every day
#asahi-dev · 2026-01-11 18:47
svenpeter[m]😄
#asahi-dev · 2026-01-11 18:33
svenpeter[m]yeah, i'd rather not touch them 🙂
#asahi-dev · 2026-01-11 18:33
svenpeter[m]oh, you meant devices with jtag enabled.
#asahi-dev · 2026-01-11 18:32
svenpeter[m]we know where SWD is, we can even mux it to some pins on the type-c port. it's unfortunately fused off though and doesn't do much
#asahi-dev · 2026-01-11 18:22
svenpeter[m]i'm just showing you more cursed things!
#asahi-dev · 2026-01-11 18:22
svenpeter[m]you told me you like cursed things! 😛
#asahi-dev · 2026-01-11 18:19
svenpeter[m]https://github.com/AsahiLinux/m1n1/issues/463 is actually the first bug in 4 years where i wish i had jtag
#asahi-dev · 2026-01-11 18:01
svenpeter[m]pity that can't explain the weird m3 corruption with the hv though ☹️
#asahi-dev · 2026-01-11 17:38
svenpeter[m]*taken from GL1
#asahi-dev · 2026-01-11 17:37
svenpeter[m]yuuup. it'll update both GL1_ELR and EL1_ELR when the exception is taken from EL1
#asahi-dev · 2026-01-11 17:30
svenpeter[m]https://github.com/AsahiLinux/m1n1/pull/529 🙃
#asahi-dev · 2026-01-11 17:23
svenpeter[m]https://github.com/AsahiLinux/m1n1/blob/fa822be70783bd721589a47c4355e76d6ed9c356/src/exception.c#L357 that's broken as well, isn't it? we always overwrite GL's ELR and the normal ELR there
#asahi-dev · 2026-01-11 16:36
svenpeter[m]that has a few more side effects, doesn't it?
#asahi-dev · 2026-01-11 16:05
svenpeter[m]they look good to me, will wait a bit in case anyone else wants to triple-check before merging
#asahi-dev · 2026-01-11 16:02
svenpeter[m]lol, did we really get stuck in a brk -> EL2 does whatever but doesn't adjust ELR -> back to brk -> EL2 does whatever again loop? 😄
#asahi-dev · 2026-01-11 15:58
svenpeter[m]maybe we should update clang-format inside the CI then
#asahi-dev · 2026-01-11 15:55
svenpeter[m]nickchan: the PR improves "el1_call crashes m1n1" to "el1_call works again", that's good enough for me. feel free to remove el1_call from el1 support if you care
#asahi-dev · 2026-01-11 15:55
svenpeter[m]ugh
#asahi-dev · 2026-01-11 15:53
svenpeter[m]if you run make format and just squash whatever it fixes i'll merge it
#asahi-dev · 2026-01-11 15:53
svenpeter[m]😄
#asahi-dev · 2026-01-11 15:50
svenpeter[m]noopwafel / Shiz: nice! even some drive-by readability fixes
#asahi-dev · 2026-01-11 15:49
svenpeter[m]not looking forward to finding a sane way to do that and then get it upstream 😄
#asahi-dev · 2026-01-11 15:49
svenpeter[m]that's gonna be extra spicy anyway because iirc you have to stuff something that comes from SEP into that weird nvme iommu
#asahi-dev · 2026-01-11 15:47
svenpeter[m]we still need to figure out some PSCI conduit for proper cpu sleep
#asahi-dev · 2026-01-11 15:47
svenpeter[m]oh yeah, there's a bunch of other issues. i was just talking about co-processors there
#asahi-dev · 2026-01-11 15:45
svenpeter[m]so maybe it's just gpu that is going to be fun then. and SEP i guess if we ever support that
#asahi-dev · 2026-01-11 15:37
svenpeter[m]And we also don’t even support the basics like taking CPUs completely offline

plus ›