#asahi-dev · 2026-10-07 13:00
jannaudrm requires opensource user space#asahi-dev · 2026-10-07 12:58
linkmauveBut they do so already, see Rockchip’s MPP.#asahi-dev · 2026-10-07 12:57
linkmauveMy main fear would be that it would encourage vendors to do proprietary drivers in userspace, where they don’t have the GPL requirement from the kernel.#asahi-dev · 2026-10-07 12:56
jannaundufresne's talk is about replacing v4l2 stateless with minimal drm driver and implement vulkan video in mesa#asahi-dev · 2026-10-07 12:55
linkmauveOh, I recently resumed my work on a Vulkan video layered above V4L2 stateless, fwiw.#asahi-dev · 2026-10-07 12:53
jannausofus: maybe of interest: v4l2 vs. vulkan video talk at lpc: https://lpc.events/event/20/contributions/2354/ https://www.youtube.com/watch?v=gHZd9P5BQWE starting in 7 minutes#asahi-dev · 2026-10-07 11:08
JamesCalligeros[m]Ill go back and test that friday or on the weekend#asahi-dev · 2026-10-07 11:08
MichaelReeves[m]it did so I might have stuffed something up with that#asahi-dev · 2026-10-07 11:07
jannaudid you test the 6.18 tree? I would not be surprised if the rebase requires more than resolving conflicts#asahi-dev · 2026-10-07 11:00
chaos_princessthey do have to deal with app-embedded chromium somehow, as that does jit and runs in 4k#asahi-dev · 2026-10-07 10:59
chadmedor if plasmashell is gonna crash in some qml thing#asahi-dev · 2026-10-07 10:59
chadmedregardless if its going to cause crashes with memfd/shmem/graphics stuff then its useless lol #asahi-dev · 2026-10-07 10:58
j`eyit's funny cos my coworker is trying to upstream the opposite, but that wont work asahi of course :( (4k kernel, 16k user)#asahi-dev · 2026-10-07 10:53
chadmedi just tested this with a win32 game and results were as expected lol#asahi-dev · 2026-10-07 10:52
chadmedyeah pmuch#asahi-dev · 2026-10-07 10:52
chaos_princessso, the only way is to have wine re-exec if it notices that it needs to load x86 code#asahi-dev · 2026-10-07 10:52
chadmedif the elf loader sees a 16k elf it will not do the needful for anything that forks/execs off it later in its life#asahi-dev · 2026-10-07 10:51
jannauat execve() time#asahi-dev · 2026-10-07 10:51
chadmedlinkmauve: doesnt matter, wine itself may not be #asahi-dev · 2026-10-07 10:51
linkmauveAre AArch64 Windows programs also built for 4k pages?#asahi-dev · 2026-10-07 10:51
chaos_princesshow early does it need to be applied?#asahi-dev · 2026-10-07 10:51
jannauwine would probably need to handle it itselff#asahi-dev · 2026-10-07 10:50
jannauor in a binfmt_misc handler for x86#asahi-dev · 2026-10-07 10:50
chadmeddoesnt help with wine #asahi-dev · 2026-10-07 10:50
chaos_princessput it in fex?#asahi-dev · 2026-10-07 10:50
chadmedapple gets away with just having rosetta do this for anything running under it of course, and all the x86 userspace code is obviously compiled for 4k anyway#asahi-dev · 2026-10-07 10:49
chadmedthere probably needs to be a shim similar to loader.c somewhere in the mix if this is going to work on mixed systems#asahi-dev · 2026-10-07 10:46
chadmedyeah i think its still really cool but it needs time in the oven is all#asahi-dev · 2026-10-07 10:46
MichaelReeves[m]was still fun to experiment with a bit & learnt lots abt how memory management works#asahi-dev · 2026-10-07 10:46
MichaelReeves[m]ah yeah, true. makes sense thanks#asahi-dev · 2026-10-07 10:45
chadmedthe parent process will be loaded as 16k because thats what the image is aligned to and then the code inside it will explode#asahi-dev · 2026-10-07 10:45
chadmedand the loader looking for the elf image alignment alone as the indicator of a process being 4k is fine for android apps _compiled_ for 4k but will fall apart for anything like fex or wine that is mixed #asahi-dev · 2026-10-07 10:44
chadmeditll probably have issues with shared memory#asahi-dev · 2026-10-07 10:44
chadmedi assume its going to have major issues with anything that is JITted in any case which is probably why plasmashell is crashing #asahi-dev · 2026-10-07 10:43
chadmedlet them cook and if it ends up being useful we can come back to it#asahi-dev · 2026-10-07 10:43
chadmedwe dont know if upstream would ever accept it#asahi-dev · 2026-10-07 10:43
chadmedwe dont even know if google plans to make this at all upstreamable#asahi-dev · 2026-10-07 10:42
chadmeddont waste too much time on this #asahi-dev · 2026-10-07 10:42
MichaelReeves[m]thank you for testing!!#asahi-dev · 2026-10-07 10:42
MichaelReeves[m]obviously those are much more complex than the programs I tested#asahi-dev · 2026-10-07 10:42
MichaelReeves[m]weird ok, ill look more into it tmrw#asahi-dev · 2026-10-07 10:41
chadmedneeds more time in the oven #asahi-dev · 2026-10-07 10:41
chadmedlinux fex still segfaults on page size issues, wine fex doesnt work, plasmashell constantly crashes under it #asahi-dev · 2026-10-07 10:41
chadmedyeah nah this is actually completely unusable#asahi-dev · 2026-10-07 10:40
MichaelReeves[m]and get rid of the other stuff#asahi-dev · 2026-10-07 10:22
MichaelReeves[m]just replace execv("/proc/self/exe", args); with system(rest of the args...)#asahi-dev · 2026-10-07 10:21
MichaelReeves[m]use this to launch wine https://github.com/IntegralPilot/wts/blob/main/runner/launch.c#asahi-dev · 2026-10-07 10:21
MichaelReeves[m]you should be able to force 4k just for wine#asahi-dev · 2026-10-07 10:21
chadmedyeah probably, that or dcp #asahi-dev · 2026-10-07 10:21
MichaelReeves[m]the main thing I didnt test with that mode is GPU stuff (cause theres none of m3) so maybe its that