#asahi-dev · 2026-01-03 12:04
nickchanold alt distro users#asahi-dev · 2026-01-03 12:04
nickchanat this point only alt distro users would be on 12.3 fw#asahi-dev · 2026-01-03 12:04
chadmedbut once we have some 14.x version we can and probably should drop 12.3#asahi-dev · 2026-01-03 12:04
chadmedsome stuff changed for 14.x and i think its a little more than just the method tags#asahi-dev · 2026-01-03 12:03
chadmedeven though there should be no one at all still on 12.x firmware#asahi-dev · 2026-01-03 12:03
chadmedi removed it in one of my test branches but i think we want to keep around at least two versions for safety#asahi-dev · 2026-01-03 12:02
nickchani mean can they be#asahi-dev · 2026-01-03 12:02
nickchan the macos 12.3 support be removed#asahi-dev · 2026-01-03 11:57
chadmedwhos going to tell them not to #asahi-dev · 2026-01-03 11:57
chadmedamd can though #asahi-dev · 2026-01-03 11:57
chadmeddo you think if i dumped millions of lines of machine-generated register definitions into the dcp driver the drm maintainers would take the patches? probably not #asahi-dev · 2026-01-03 11:57
chadmedamd, nvidia, intel#asahi-dev · 2026-01-03 11:56
chadmedred, green blue #asahi-dev · 2026-01-03 11:56
nickchanwhat even is the tricolore#asahi-dev · 2026-01-03 11:55
chadmedhowever fair or unfair it is, we are held to a higher standard, and must do better. and we can.#asahi-dev · 2026-01-03 11:55
chadmedthe tricolore can get away with it because... well what choice do we have but to eat whatever slop they put in front of us#asahi-dev · 2026-01-03 11:54
chadmedbut that said, not every compositor project is going to be so open, and even so its bad manners to half-arsedly implement broken stuff and expect userspace to clean up your mess#asahi-dev · 2026-01-03 11:52
chadmedzamundaaa was working on overlay plane support as it was, but implemented a preliminary "let the cursor go on an OVERLAY instead of CURSOR" thing because we asked nicely about it #asahi-dev · 2026-01-03 11:52
nickchanBut clearly there needs to be an exception as long as it's not @apple.com#asahi-dev · 2026-01-03 11:52
nickchanone isn't expected to have a rpc interface in the kernel either#asahi-dev · 2026-01-03 11:52
chadmedcompositor devs are used to display engines doing stupid things#asahi-dev · 2026-01-03 11:51
lacamar_That sounds like a really tough constraint to work with#asahi-dev · 2026-01-03 11:51
chadmedwe do not have that luxury. we are expected to pass certain tests and uphold the guarantees/invariants of the atomic modesetting api #asahi-dev · 2026-01-03 11:50
chadmedbecause apple control the compositor they can make it work around whatever insane limitations exist in dcp/agx/avd however they want#asahi-dev · 2026-01-03 11:49
chadmedit will put the cursor on the "top" plane and the desktop on the bottom. if theres any media that is not occluded by anything else and can be directly scanned out from the gpu or avd, it will put that on the top plane, unless the cursor is being moved#asahi-dev · 2026-01-03 11:48
chadmedonly two, which is what we do in the asahi kernel right now#asahi-dev · 2026-01-03 11:48
lacamar_What happens in MacOS? Does it not make use of multiple planes for the desktop?#asahi-dev · 2026-01-03 11:45
chadmedother compositors which rely on the legacy cursor plane ioctls (e.g. mutter) will simply not see hw cursor support and ignore it #asahi-dev · 2026-01-03 11:44
chadmedkwin will happily use a "normal" overlay plane for a cursor and is smart enough to immediately fall back to a software composited cursor if an atomic check on the cursor plane fails #asahi-dev · 2026-01-03 11:44
chadmedwell we just dont expose a cursor plane for the legacy cursor ioctls#asahi-dev · 2026-01-03 11:43
lacamar_Just disable mouse input for Asahi#asahi-dev · 2026-01-03 11:43
chadmedbut come on, your desktop display controller has _one_ working plane and crashes if your cursor leaves the screen are you fucking serious #asahi-dev · 2026-01-03 11:42
lacamar_They're hiding all their dirty secrets in this firmware blob#asahi-dev · 2026-01-03 11:42
chadmedand yeah the latter thing is still pretty true, id rather have an apple tv than some random rebadged rockchip thing under my telly #asahi-dev · 2026-01-03 11:42
nickchanor most x86 firmwares for that matter#asahi-dev · 2026-01-03 11:42
chadmedyeah but apple like to present themselves as better than that, and better for media creation/consumption to boot #asahi-dev · 2026-01-03 11:41
nickchanjust take a look at most android downstream kernels#asahi-dev · 2026-01-03 11:41
nickchaneh it's pretty average#asahi-dev · 2026-01-03 11:39
chadmed"yes we have four surfaces but lol dont actually use them" #asahi-dev · 2026-01-03 11:39
chadmedi feel like these are things they should probably fix because it looks rather incompetent #asahi-dev · 2026-01-03 11:38
nickchanthe hardware can output to two screens#asahi-dev · 2026-01-03 11:38
lacamar_nickchan: thanks for that thesis link, I'll try to wrap my head around it#asahi-dev · 2026-01-03 11:38
chadmedlikewise if you try to move a rectangle out of the space clipped by surface 0 you will crash iomfb #asahi-dev · 2026-01-03 11:38
nickchanDcp firmware on m2 still doesn't support to outputting to two external screens#asahi-dev · 2026-01-03 11:38
chadmedim rapidly losing hope that we can actually do more than two planes for example which sucks because kwin will always just use the _one_ overlay for the cursor if it can :/#asahi-dev · 2026-01-03 11:37
chadmedunfortunately some of the limitations we have inherited from its "jack of all trades master of none" firmware are total bs #asahi-dev · 2026-01-03 11:36
chadmedoh yes#asahi-dev · 2026-01-03 11:36
nickchanwhich is another thing#asahi-dev · 2026-01-03 11:36
nickchanRight IOMFB still exist I am referring to the x86 IOFramebuffer#asahi-dev · 2026-01-03 11:36
chadmedits all still called iomfb in the thing too