#asahi-dev · 2026-01-15 06:35
jannaukitlith: see the can_present in src/broadcom/vulkan/v3dv_wsi.c, the kms and render device fd are in struct renderonly in agx_device#asahi-dev · 2026-01-15 06:33
kitlithbut if it's not supposed to work, given supports_scanout = false...#asahi-dev · 2026-01-15 06:32
kitlithI've already forced can_present to return true w/ gdb and observed it seemingly working correctly (monado + valve index on the fairydust kernel), but was looking for a route that was upstreamable.#asahi-dev · 2026-01-15 06:30
kitlithjannau: can_present is the difference between being able to use vkGetDrmDisplayEXT and vkAcquireDrmDisplayEXT or not. I did not observe any checks of supports_scanout in that path.#asahi-dev · 2026-01-15 06:28
jannaukitlith: does can_present even matter with `wsi_device.supports_scanout = false` in hk_init_wsi()?#asahi-dev · 2026-01-15 06:27
JamesCalligeros[m]With its own drm driver#asahi-dev · 2026-01-15 06:27
JamesCalligeros[m]The display controller is a separate hw block on this platform#asahi-dev · 2026-01-15 06:26
kitlithah. the platform fullname is different. the application i'm testing with passes in a drm fd with platform fullname /soc/display-subsystem, whereas the fd I was comparing against (hk_physical_device.dev.fd) has platform /soc/gpu@<addr>, so they don't compare as equal.#asahi-dev · 2026-01-15 05:21
kitliththink i'm personally leaning towards wsi_common_drm_devices_equal, but am not fussed either way.#asahi-dev · 2026-01-15 05:05
kitlithAre there any preferences for which approach should be taken?#asahi-dev · 2026-01-15 05:04
kitlithI noticed that wsi_device->can_present_on_device in mesa is not overridden by honeykrisp, but it should be since it's a DRM_BUS_PLATFORM device instead of a DRM_BUS_PCI device. There seems to be two approaches taken by the drivers that override this in mesa so far: (1) we're the only DRM_BUS_PLATFORM device on the system, return true if it's a DRM_BUS_PLATFORM device. (2) compare drm device equality using wsi_common_drm_devices_equal.#asahi-dev · 2026-01-14 19:52
chaos_princessand by second copy - 14.latest#asahi-dev · 2026-01-14 19:52
jannausee also https://asahilinux.org/docs/sw/m1n1-hypervisor/#asahi-dev · 2026-01-14 19:52
jannaugatallah: the m3 support isn't yet in a state where support in the curl | sh installer would be useful#asahi-dev · 2026-01-14 19:51
chaos_princessinstall a second copy of macos and overwrite it's kernel with m1n1 using kmutil https://asahilinux.org/docs/sw/m1n1-user-guide/#asahi-dev · 2026-01-14 19:46
gatallahTo clarify, I'd like to have linux on my Macbook M3 Pro, so I'm contributing. I haven't worked in lower level programming, so it's a chance to learn. Interpreting the docs and LLM slop got me to setup a host in Linux but I cannot get my M3 Pro to be the target, since I can't even curl ... | sh. I'd appreciate if someone knowledgable on the topic could point me in the right direction, because I'm blocked. Thank#asahi-dev · 2026-01-14 19:27
gatallahIf I'm on a Macbook M3 Pro, is there a way to develop for it? I'm trying to set it up as a target, but I can't even curl asahi. I've already setup the host in ubuntu VM on a PC#asahi-dev · 2026-01-14 17:27
jannauespecially since it would require vendoring it into the kernel. Producing code seems to be orders of magnitude faster than compiling it#asahi-dev · 2026-01-14 17:01
leiosounds like no point in investigating unsynn then :)#asahi-dev · 2026-01-14 14:54
jannaurewriting the versions macro in syn doesn't seem to affect the compile time much (if at all). within 1% (build time of asahi.o, around 24.3s) and I'm not motivated for statistics#asahi-dev · 2026-01-14 13:29
UncleArthur[m]No#asahi-dev · 2026-01-14 13:29
pitustspam#asahi-dev · 2026-01-14 13:27
UncleArthur[m]Hey 👋... (full message at <https://matrix.org/oftc/media/v1/media/download/AdFzuaNO8Uh8Vflp_vDO0vbjIiJkg1wKUReD034u_R9b3rb-CFk0NHQgP8nVhCU5M0a9fiP7LDKPsftmUFGK8b9CecBMlIdQAG1hdHJpeC5vcmcvdEpUeVZ6elNUZ05mRE54VW55R2tpTXFs>)#asahi-dev · 2026-01-14 12:03
flokliWay more characters to process, lol#asahi-dev · 2026-01-14 08:48
svenliterally the length? So just andding invalid stuff like “aaaaaaaaaaa…aaa” has an effect on the boot time? :D#asahi-dev · 2026-01-14 06:47
LarstiQShiz: the longer the line the faster the progress? ;)#asahi-dev · 2026-01-14 00:59
Shiznot great#asahi-dev · 2026-01-14 00:59
Shizwe got a very interesting state where the kernel command line length seems to have an effect on boot progress :P#asahi-dev · 2026-01-13 20:12
svenmaybe it's time to drop it everywhere and see what breaks :D#asahi-dev · 2026-01-13 20:10
jannaunot sure if we actually need that or are just cargo-culting what macos / the adt does#asahi-dev · 2026-01-13 20:06
svenand i guess we need that because some devices can't use the full iova range? i guess some co-processors that do weird things in the firmware?#asahi-dev · 2026-01-13 20:05
jannaulimits the iova space and allows using "tags" (or whatever macos uses those for) in high bits#asahi-dev · 2026-01-13 19:59
svenand we usually need a good justification for any property like apple,sram-iova-base so i'd rather just hardcode it in the driver for now when it's the same everywhere anyway#asahi-dev · 2026-01-13 19:57
sveni'm writing device tree bindings here and wanted to figure out if this thing has to be variable#asahi-dev · 2026-01-13 19:57
svenhmm?#asahi-dev · 2026-01-13 19:57
Shizsven: can't you just grba that from the templates? :p#asahi-dev · 2026-01-13 19:56
svenwhat does apple,dma-range do again?#asahi-dev · 2026-01-13 19:56
svenjust to verify that the buffer the co-proc requests aren't outside of its sram region#asahi-dev · 2026-01-13 19:55
svennot really#asahi-dev · 2026-01-13 19:50
jannauis it important? We still have to find a upstream solution for "apple,dma-range"#asahi-dev · 2026-01-13 19:44
svenall machines seems to have 0x10000000 as iova base for the sram, just gonna hardcode that instead of adding another apple, device tree property for now i think#asahi-dev · 2026-01-13 19:41
jannaubah, my simple syn based AGX versions macro rewrite breaks down when it comes in contact with other macros#asahi-dev · 2026-01-13 19:37
svengreat, thanks! that should be all i need#asahi-dev · 2026-01-13 19:36
jannausven: m1 max, m1, m3: https://paste.centos.org/view/04bc083a #asahi-dev · 2026-01-13 19:34
sventhanks!#asahi-dev · 2026-01-13 19:33
jannausven: m2, m2 pro, m2 ultra: https://paste.centos.org/view/6574b5f3#asahi-dev · 2026-01-13 19:24
svenbut no need to set that all up, there are usually enough people in here who can grab them from m1n1 and extract the nodes :)#asahi-dev · 2026-01-13 19:24
svenit's about apple device trees