#
#asahi-dev · 2026-03-06 09:52
oliverYea I'll continue to look into how to tell mesa not to use interchange for window surfaces.
#asahi-dev · 2026-03-06 09:52
olivergot it.
#asahi-dev · 2026-03-06 09:47
jannauexternal buffer probably means allocated on a different device (or dma_heap)
#asahi-dev · 2026-03-06 09:46
jannauinterchange should be avoided for GPU internal purposes
#asahi-dev · 2026-03-06 09:45
oliverIs that what external means?
#asahi-dev · 2026-03-06 09:44
jannauoliver: the scanout buffers are either allocated on the GPU or are linear. the KMS device can only allocate dumb buffers (meaning linear, CPU accessible)
#asahi-dev · 2026-03-06 09:40
chadmedill try
#asahi-dev · 2026-03-06 09:39
oliverDoes it work if you run with stock mesa, but with the kernel patch?
#asahi-dev · 2026-03-06 09:39
oliverWell thanks for nothing @gnome
#asahi-dev · 2026-03-06 09:39
chadmedgnome devs refused to support them for this platform
#asahi-dev · 2026-03-06 09:38
chadmedyes, we have had hardware cursors on kde for a while now
#asahi-dev · 2026-03-06 09:38
oliverAre they normally being created for you? At least gnome was always using the "software" cursor
#asahi-dev · 2026-03-06 09:34
chadmedbut yes the cursor fbs are still not being created
#asahi-dev · 2026-03-06 09:34
chadmedi dont think it really matters as long as the gpu can deal with them
#asahi-dev · 2026-03-06 09:32
oliverI am still looking if i can get mesa to not select the interchange format for window surfaces. currently it creates those with the same modifier set as it created the screen surface. I would prefer gpu tiled in for those and i would love it to only select interchange for the screen
#asahi-dev · 2026-03-06 09:29
oliverI would push some mesa changes though
#asahi-dev · 2026-03-06 09:29
oliverdo you still see issues with the cursor with the latest version? I could not reproduce the problems.
#asahi-dev · 2026-03-06 09:20
chadmedim more concerned about the cursor lol
#asahi-dev · 2026-03-06 09:19
chadmedif it works fine on G13G then i dont see why we cant just allow it to be used everywhere (YCC luma channel thing notwithstanding)
#asahi-dev · 2026-03-06 09:18
chadmedim finishing up my M1 install tonight
#asahi-dev · 2026-03-06 09:16
oliverah nevermind
#asahi-dev · 2026-03-06 09:09
olivermay assumption here being that external referes to something you would want to share for scanout.
#asahi-dev · 2026-03-06 09:08
oliverI think i can configure the interchange format to only be used with external buffers, which - based on the name - might be what we want.
#asahi-dev · 2026-03-06 07:06
chadmedmoving the problem into m1n1 just introduces another thing that has to constantly be updated and tracked by users. its silly
#asahi-dev · 2026-03-06 07:05
chadmedyeah and like i said before its just better for everyone to ship devicetrees with the kernel that are guaranteed to work _with that kernel_
#asahi-dev · 2026-03-06 07:03
chaos_princessgpu is a pretty good example of just how much the binding can change between what we started with and what gets upstream
#asahi-dev · 2026-03-06 07:01
chadmedonce devicetree bindings are upstream, they are guaranteed stable and this stops being a problem
#asahi-dev · 2026-03-06 07:00
chadmedit doesnt matter if m1n1 is filling that or if its compiled into the DTB, the old kernel driver is expecting apple,foo-prop and doesnt find it
#asahi-dev · 2026-03-06 07:00
chadmedlets say that we wrote a driver that looks for apple,foo-prop in the devicetree. for whatever reason, that changes to apple,bar-prop in the next kernel version
#asahi-dev · 2026-03-06 06:58
JamesCalligeros[m]but this stuff isnt the problem, its the bindings themselves
#asahi-dev · 2026-03-06 06:58
JamesCalligeros[m]other stuff like GPU initialisation is computed
#asahi-dev · 2026-03-06 06:58
JamesCalligeros[m]some stuff comes from the ADT, and m1n1 already fills some of this at runtime (see kboot.c)
#asahi-dev · 2026-03-06 06:53
kitlithare the dynamic bits pulled based on information in the dtb?
#asahi-dev · 2026-03-06 05:17
JamesCalligeros[m]it is easier and less annoying to just change the dts and ship the change with the relevant kernel version
#asahi-dev · 2026-03-06 05:17
JamesCalligeros[m]so either m1n1 code needs to change when bindings change, or the dts files do
#asahi-dev · 2026-03-06 05:17
JamesCalligeros[m]and "correct stuff" is a moving target
#asahi-dev · 2026-03-06 05:17
JamesCalligeros[m]m1n1 still needs to overlay the correct stuff
#asahi-dev · 2026-03-06 05:16
JamesCalligeros[m]yeah but having m1n1 overlay stuff on a mostly empty DTB isnt gonna fix that
#asahi-dev · 2026-03-06 05:16
nickchandevice tree is kernel-specific, but the current way they are handled makes it so that there's only one device tree for all of kernels+uboot
#asahi-dev · 2026-03-06 05:14
JamesCalligeros[m]and youd still need to update m1n1
#asahi-dev · 2026-03-06 05:14
JamesCalligeros[m]youre just moving the problem. the issue is that the devicetree binding constitutes an API of sorts between the platform and the drivers. whether we update m1n1 or the devicetree source itself with a break in that API is irrelevant, youd still have broken backwards compat with kernel drivers from older kernels which cant deal with it
#asahi-dev · 2026-03-06 05:10
kitlithI see.
#asahi-dev · 2026-03-06 04:00
tpw_rulesi've proposed that several times and been shot down
#asahi-dev · 2026-03-06 00:26
kitlithsome idle musing about boot flow: could m1n1 theoretically provide the dynamic bits of the device tree as a device tree overlay, to be merged with <kernel specific devicetree>? granted, the eventual goal is for incompatbilities to not happen in the first place...
#asahi-dev · 2026-03-05 00:33
nickchan /arm-io/usb-drd is one and there are others
#asahi-dev · 2026-03-05 00:32
nickchanthe curse of some things being unnumbered in A series SoCs spreads to asahi linux huh
#asahi-dev · 2026-03-04 22:28
TramtristSame weight/size with nearly all the good functions stripped out at the same price
#asahi-dev · 2026-03-04 22:26
TramtristAsahi is the reason Ill keep my M2 and wont look further or at something like the Neo until they can run it
#asahi-dev · 2026-03-04 20:57
jannaut6050 ATC support 4 dp tunnel
#asahi-dev · 2026-03-04 20:51
nicolas17probably a reference to alpha-centauri :D

plus ›