#
#asahi-dev · 2025-05-28 13:44
fl0_id(I use "ported" very losely here)
#asahi-dev · 2025-05-28 13:43
fl0_idchadmed well, yes, it is for a year old mesa. I kind of / maybe (needs more testing) ported the patches to a slightly newer mesa used in ubuntu 24.04 lts. but if they are almost upstream I would rather use that of course
#asahi-dev · 2025-05-28 13:40
chadmedor rather, for a year-old mesa
#asahi-dev · 2025-05-28 13:39
chadmedfl0_id: i dont think that is a thing anymore, especially judging by the fact that the last release was a year ago
#asahi-dev · 2025-05-28 13:39
alyssaand the other 0.1% is shipping in fedora asahi remix since october
#asahi-dev · 2025-05-28 13:39
alyssaI don't know what it is, but the mesa pathces are 99.9% upstream
#asahi-dev · 2025-05-28 13:38
fl0_idor is that also upstream=
#asahi-dev · 2025-05-28 13:38
fl0_idalyssa thx, but I think this is a different one. I was referring to https://copr.fedorainfracloud.org/coprs/slp/mesa-krunkit/package/mesa/ which mainly seems to align some shmem parts along 16k boundary and things like that
#asahi-dev · 2025-05-28 13:35
alyssafl0_id: the mesa patches are 99.9% upstream, it's https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/1527 that needs to go upstream
#asahi-dev · 2025-05-28 13:34
chaos_princesssergio should be on fedora matrix
#asahi-dev · 2025-05-28 13:32
fl0_iddoes anyone know if slopezpa / sinrega is here sometimes or what the status of upstreaming their mesa patches for apple vgpu in containers is?
#asahi-dev · 2025-05-27 22:32
jannau`__asm__ volatile("ldr x11, [sp, #8]" : : : "r11");` after `msr(PAN, 0);` fixes it for me. only difference in the binary is that that meaning instruction is inserted
#asahi-dev · 2025-05-27 21:49
svenso we may still corrupt the stack but just not notice
#asahi-dev · 2025-05-27 21:49
sventhe difference between gcc and clang was that clang had that additional function call inside mmu_setup_secondary while gcc inline that function instead iirc
#asahi-dev · 2025-05-27 21:39
jannaulinux clang-built m1n1 crashes as well
#asahi-dev · 2025-05-27 21:38
jannauthe crash did not reproduce on t6020
#asahi-dev · 2025-05-27 21:03
svenI still have no clue how that happens. I looked for all instrs that follow a register load of 0x20.. and could write it but none made sense
#asahi-dev · 2025-05-27 21:02
jannauright, I wanted to try if that happens with clang on linux
#asahi-dev · 2025-05-27 21:00
svennot great but also not a blocker imho
#asahi-dev · 2025-05-27 21:00
sventhere’s still https://github.com/AsahiLinux/m1n1/issues/463 that hasn’t happened for me in the last release but that one only happens when running the hv
#asahi-dev · 2025-05-27 20:59
svenyeah, good idea
#asahi-dev · 2025-05-27 18:22
jannaushall we plan for a m1n1 release? I think a release before the chickens refactoring and the agx initdata would be good idea
#asahi-dev · 2025-05-27 17:17
svenwould show that there’s interest in that series
#asahi-dev · 2025-05-27 17:17
svengetting some tested-bys,etc. for the pmu support might be more helpful than trying to resend it again and again
#asahi-dev · 2025-05-27 16:02
nickchannothing compared to the amount of patches in asahi linux though
#asahi-dev · 2025-05-27 16:02
nickchandownstream patch load is now 96 (111 for 6.14)
#asahi-dev · 2025-05-27 16:01
nickchanI think I will send i2c, spmi noeds and continue to resend cpu pmu support for the older socs in during the next cycle https://github.com/HoolockLinux/linux b4/apple-cpmu b4/i2c-no-t2 b4/t8015-spmi
#asahi-dev · 2025-05-27 11:14
kettenisI'll see if I can prepare a pull request
#asahi-dev · 2025-05-27 11:13
kettenisso sounds that this way of building is potentially useful for others too
#asahi-dev · 2025-05-27 11:12
chadmedif build-std is stabilised then we should be all clear
#asahi-dev · 2025-05-27 11:12
chadmedso do we, that might be enough then
#asahi-dev · 2025-05-27 11:11
kettenisbut we already package that for OpenBSD
#asahi-dev · 2025-05-27 11:11
kettenisbut of course -Z build-std does need the rust-src component
#asahi-dev · 2025-05-27 11:10
kettenisyou don't really need the full rustup toolchain
#asahi-dev · 2025-05-27 11:10
kettenisso the only missing bit really is just std lib support
#asahi-dev · 2025-05-27 11:10
chadmedthis will also be a bit annoying for gentoo tbh since everyones going to need the softfloat target built/installed
#asahi-dev · 2025-05-27 11:09
kettenisit turns out that the OpenBSD rustc already supports aarch64-unknown-none-softfloat
#asahi-dev · 2025-05-27 11:09
chadmedkettenis: yeah, my question was more assessing whether or not us just not using softfloat at all would help, but if we dont want to touch the fp regs at all on context switches then its not an option
#asahi-dev · 2025-05-27 11:07
kettenisthe -Z build-std is purely to work around the lack of rustup support for OpenBSD
#asahi-dev · 2025-05-27 11:06
kettenischadmed: I'm not changing RUSTARCH; that's still aarch64-unknown-none-softfloat
#asahi-dev · 2025-05-27 10:58
jannauI don't think we want to touch the FP/SIMD registers on HV context switches
#asahi-dev · 2025-05-27 10:56
jannauI'd say using "-Z build-std" is kind of justified for our use case if we can't rely on a prebuilt softfloat toolchain. stabilization is a goal for 2025H1 to support Rust for Linux: https://github.com/rust-lang/rust-project-goals/issues/116
#asahi-dev · 2025-05-27 10:39
chadmedwould it help if we taught the hv how to save/restore fp regs and got rid of the softfloat thing?
#asahi-dev · 2025-05-27 10:38
chadmedkettenis: m1n1 with softfloat rust works fine without extra flags on gentoo so im not certain
#asahi-dev · 2025-05-27 09:52
ketteniswill do so later this week
#asahi-dev · 2025-05-27 09:52
kettenisI'd still like to verify that the binaries work before we commit to having rust in the stage 2 m1n1
#asahi-dev · 2025-05-27 09:51
ketteniswould that be a useful feature for other OSes or distros?
#asahi-dev · 2025-05-27 09:50
kettenisSeems I have a workable way to build m1n1 with rust code enabled using "-Z build-std=alloc,core"
#asahi-dev · 2025-05-26 08:49
svency8aer: nice!
#asahi-dev · 2025-05-26 07:16
chadmedkettenis: yes those cfg warnings are due to an unstable feature that i dont think is required anymore. havent had a look at it though

plus ›