#asahi-gpu · 2025-11-01 20:34
jannaupro tip: if a function is only used inside the rust drm bindings it might have been added by lina#asahi-gpu · 2025-11-01 20:32
jannaufinally lockdep complaint free#asahi-gpu · 2025-11-01 14:45
jannausolve by moving the vmap call into add_block. self.cpu_maps is static anyway so we can vmap beforehand and ensure the dma_resv lock is not taken while the mm lock is#asahi-gpu · 2025-11-01 14:12
jannauthe order is reversed. #1 was observed first, should be the self.add_block() on line 923 in the same function#asahi-gpu · 2025-11-01 14:08
jannauI was only using the File debug flag#asahi-gpu · 2025-11-01 14:07
jannauno, I think the vmap is https://github.com/AsahiLinux/linux/blob/asahi/drivers/gpu/drm/asahi/alloc.rs#L954#asahi-gpu · 2025-11-01 14:03
jannauyes, still complains. the complaint looks valid to me#asahi-gpu · 2025-11-01 14:02
jannauas in asahi module param debug flags? yes#asahi-gpu · 2025-11-01 13:46
jannauhttps://paste.debian.net/1403981/#asahi-gpu · 2025-11-01 13:41
jannauit's probably harmless because it happens in .probe() (at least this instance)#asahi-gpu · 2025-11-01 13:39
jannauit's rust/kernel/drm/mm.rs:172 and the dma_resv lock#asahi-gpu · 2025-11-01 13:36
jannauhmm, it now complains about it with my changes backed out. why didn't it do that before#asahi-gpu · 2025-11-01 13:33
jannauwhy does lockdep complains about possible circular locking with a lock I haven't touched? I'm touching the other lock but I don't see how would have anything for the places lockdep complains about#asahi-gpu · 2025-10-31 23:34
jannauproblems for tomorrow#asahi-gpu · 2025-10-31 23:34
jannauworkaround (maybe a fix) found thanks to panthor/tyr. `gst-launch-1.0 v4l2src ! gtk4paintablesink` doesn't hang anymore with vulkan/opengl. Made lockdep unhappy and hitting a warning in drm_gem_mmap_obj()#asahi-gpu · 2025-10-31 19:30
jannauit's the dma_resv lock of external dma bufs. I'm not sure if that was always broken and this is the first case of hitting it or if I broke it#asahi-gpu · 2025-10-31 19:15
jannaulockdep doesn't complain :(#asahi-gpu · 2025-10-31 17:43
jannauat least it only affects the single process#asahi-gpu · 2025-10-31 17:42
jannaubah: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14217#note_3173360#asahi-gpu · 2025-10-31 17:22
jannauany idea what could cause unkillable dead processes (not zombies)? gdb is no help and no kernel messages. reproducible with `gst-launch-1.0 v4l2src ! gtk4paintablesink`#asahi-gpu · 2025-10-31 15:02
jannauthe fedora snapshot package does not use NV12 textures. the flatpak one does but not keen on injecting mesa and pipewire patches into the flatpak runtime#asahi-gpu · 2025-10-31 14:59
jannauthe hk multi-planar issue is hopefully just replacing "1" with "vk_format_get_plane_count(vk_format)" in two places. I've written that but not yet tested#asahi-gpu · 2025-10-31 14:58
jannauMary: yes, I was expecting alyssa to do that. the remaining issue seem to be in pipewire and/or gstreamer#asahi-gpu · 2025-10-18 15:55
jannauella-0: fixed in asahi-wip#asahi-gpu · 2025-10-18 08:59
jannauella-0: I'll fix it for the next kernel tag. thanks#asahi-gpu · 2025-10-11 09:01
jannauhttps://github.com/AsahiLinux/linux/commit/c161a301e1802c8a2757367f79436b67bce8b7d0#asahi-gpu · 2025-10-11 08:39
jannaudoes anyone remember the issues with (compute) preemption? I think it was briefly enabled but lina removed. re: https://github.com/AsahiLinux/linux/issues/417#asahi-gpu · 2025-10-11 08:25
jannauso that leaves the racy allocation as problem#asahi-gpu · 2025-10-11 08:23
jannauaccess to the heap (GPU side) is expected to be serialized so that should be ok (I spotted how that's enforced)#asahi-gpu · 2025-10-11 07:00
jannaudid you run into an issue related to this?#asahi-gpu · 2025-10-11 06:59
jannauolivial: possibly although it looks like the code assumes that concurrent use is not possible#asahi-gpu · 2025-09-07 07:17
jannauerr, wrong channel#asahi-gpu · 2025-09-07 07:01
jannaufinally a use case for aop_las: https://hachyderm.io/@samhenrigold/115159295473019599#asahi-gpu · 2025-08-27 11:22
jannauI have subscribed as well and can try to look as well#asahi-gpu · 2025-08-24 13:03
jannauor at least partially in rust#asahi-gpu · 2025-08-24 13:02
jannauhttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31867#asahi-gpu · 2025-08-24 13:01
jannauleio: nvk's vk video is rust, not merged but there's a PR open#asahi-gpu · 2025-07-23 19:52
jannauhttps://github.com/obsproject/obs-studio/commit/c2b482307599f9bd7ddb7cd96c7fce12f6f3e3e3 adds explicit sync#asahi-gpu · 2025-07-23 19:50
jannauSYNCOBJ ioctls work in general with mutter/kwin#asahi-gpu · 2025-07-23 19:38
jannaumutter tries to use drm syncobj with the kms device but I suppose that failing with EOPNOTSUPP is not causing the crash#asahi-gpu · 2025-07-23 19:14
jannaufails on this assert: https://gitlab.gnome.org/GNOME/mutter/-/blob/main/src/backends/meta-screen-cast-stream-src.c#L1015#asahi-gpu · 2025-07-23 19:11
jannaujust before the crash "meta_drm_timeline_create_syncobj failed: drmSyncobjCreate failed: Operation not supported"#asahi-gpu · 2025-07-23 19:11
jannauscreen capture with obs 31.1 on gnome crashes mutter, looks drm syncobj related#asahi-gpu · 2025-07-23 18:12
jannauasahi-20250723 tagged#asahi-gpu · 2025-07-23 18:11
jannauthere is my dri loader fix for apple (72e64b2713e ("gallium/dril: Add entrypoint for apple (asahi) kms driver"))#asahi-gpu · 2025-07-23 16:11
jannauI guess I should finish the mesa-nightly integration then for testing#asahi-gpu · 2025-07-23 16:10
jannaudo you plan to backport perf features to 25.2? if yes, we could use 25.2.0 as feature release#asahi-gpu · 2025-07-23 16:03
jannauand I think there was something else#asahi-gpu · 2025-07-23 16:02
jannaufrom you, I remember an opencl fix#asahi-gpu · 2025-07-23 16:01
jannauiirc there were 2 two minor fixes for user reported issues