#
✕ nick:jannau
#asahi-gpu · 2023-06-07 06:32
jannauchadmed: -EWRONGCHAN, but it's mostly just renaming ebuilds for version bumps
#asahi-gpu · 2023-06-02 05:55
jannau1MB every few seconds sounds a little small for that though. does it show in /sys/kernel/debug/dma_buf/bufinfo?
#asahi-gpu · 2023-06-02 05:55
jannaulina: I didn't manage to look into the buffer leaks I saw with window previews in plasma's taskmanager with the mesa tag from march
#asahi-gpu · 2023-05-02 19:56
jannauthere are two little PRs for the gpu driver on github if you want to take care of those tomorrow
#asahi-gpu · 2023-05-02 19:54
jannaumarcan removed the always-on resulting in dcp crashes on dpms
#asahi-gpu · 2023-05-02 19:53
jannaulina: please check if the gpu on t8112 has an undocumented dependency on ps_pmp like on t8103
#asahi-gpu · 2023-05-02 07:06
jannauI'm aware but I'm not sure apple's design differs that much and it's hard to argue from a point of limited knowledge of the reverse engineered driver
#asahi-gpu · 2023-04-14 06:20
jannaulina: dcp 13.3 are in https://github.com/jannau/linux/tree/t602x/bringup - everything seems to work on m2 (tested briefly). possible that further changes are needed on t602x
#asahi-gpu · 2023-04-13 22:28
jannaubut swapping surfaces works
#asahi-gpu · 2023-04-13 22:28
jannaunot quite, poweroff is broken, probably not a big deal but just another inserted callback
#asahi-gpu · 2023-04-13 22:16
jannauworking now on m2, "4 files changed, 48 insertions(+), 18 deletions(-)"
#asahi-gpu · 2023-04-13 21:36
jannauupdate is not trivial but I'm hopefully down to finding out where the additiona 8 bytes go in the huge swap_submit call
#asahi-gpu · 2023-04-13 20:41
jannauI'd still prefer 13.3 in the hope that it's closer to the version apple releases the next devices with
#asahi-gpu · 2023-04-13 20:40
jannauiboot has a version bump, from 8419.x.y.z (since 13.0 beta 4) to 8422.x.y.z, so maybe 13.2 to 13.3 is a larger update.
#asahi-gpu · 2023-04-13 20:36
jannauturns out dcp firmware 13.2 and 13.3 are not compatible. there is at least one additional callback before D123
#asahi-gpu · 2023-04-13 08:55
jannauprobably make sense to use 13.3 anyway under the assumption that most of changes will remain in later versions
#asahi-gpu · 2023-04-13 07:59
jannauno reason to support a a firmware variant if we don't have to
#asahi-gpu · 2023-04-13 07:59
jannauI'm surprised. no objections from my side. I'll check and fix dcp if it needs changes
#asahi-gpu · 2023-04-13 07:27
jannaulina: I'd be surprised if they changed the older platforms to the way h14j works in 13.3. I don't expect many chnages (if any) for dcp in 13.3
#asahi-gpu · 2023-04-12 23:24
jannauI guess with this design you can start with an empty heap and free available VA spaces, i.e. add multiple disjunct heaps
#asahi-gpu · 2023-04-12 23:18
jannaualyssa: I was expecting an assert in util_vma_heap_free if the address is outside of the range given in util_vma_heap_init()
#asahi-gpu · 2023-04-12 11:24
jannaua little surprising that random VAs can be "returned" to the heap
#asahi-gpu · 2023-04-12 11:24
jannaufixed by https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/45 - free low VA space is now reused instead of exhausting the VA space
#asahi-gpu · 2023-04-12 09:10
jannauchadmed: can you run `ioreg -l -p IODeviceTree | grep perf-states` on the m2 pro and get lina the output
#asahi-gpu · 2023-04-12 08:58
jannaubut returning bad addresses to the heap
#asahi-gpu · 2023-04-12 08:57
jannauso we are neither exhausting the address space nor leaking anything
#asahi-gpu · 2023-04-12 08:56
jannauI'll prepare a MR
#asahi-gpu · 2023-04-12 08:56
jannaulina: found the issue, we don't add dev->shader_base to bo->ptr.gpu in agx_bo_free()
#asahi-gpu · 2023-04-11 21:10
jannauI'll add a comment to the tracking bug
#asahi-gpu · 2023-04-11 21:10
jannaulast working call has data.addr 0x11_0001_0000, the first failing call has 0xfff3_4000
#asahi-gpu · 2023-04-11 21:08
jannaulina: my GEM_BIND failures look like VM_SHADER space exhaustion or failure to reuse freed space. I see GEM_BIND calls getting closer and closer to VM_SHADER_START
#asahi-gpu · 2023-04-11 07:44
jannauI was wondering whether rust-wip or bits/210-gpu is the better target for the PR
#asahi-gpu · 2023-04-11 06:27
jannauI have a fix for argument ordering in a debug statement and a separate debug flag for DRM_IOCTL_ASAHI_SUBMIT
#asahi-gpu · 2023-04-11 06:25
jannaulina: what's the most convenient way to submit patches to the asahi kernel driver? PR against rust-wip?
#asahi-gpu · 2023-04-08 15:29
jannaumy scripted actions were too fast for the memory starved, loaded system. I can push it further manually
#asahi-gpu · 2023-04-08 15:21
jannaumaybe not, now under severe memory pressure it seems to have stopped leaking buffers. I can't get past 529 objects, ~5GB
#asahi-gpu · 2023-04-08 15:16
jannauit looks like this will result in the OOM killer and looks unrelated to the GEM_BIND error
#asahi-gpu · 2023-04-08 15:13
jannauTotal 427 objects, 4065640448 bytes
#asahi-gpu · 2023-04-08 15:08
jannauTotal 276 objects, 2541666304 bytes
#asahi-gpu · 2023-04-08 15:07
jannauthe number of buffers somtimes decreases by 10 to 15 buffers but quicky springs back to the previous value
#asahi-gpu · 2023-04-08 15:05
jannauthanks to uhid the buffer leak reproducer is now automated
#asahi-gpu · 2023-04-08 11:45
jannauunfortunately still no useable debug information in mesa for the following segfault but that's probably just misguided error handling
#asahi-gpu · 2023-04-08 11:43
jannauout of memory appears unlikely on a 32GB M1 Max system
#asahi-gpu · 2023-04-08 11:41
jannaustill falling into the 2-3 days window. not sure if it is related to the buffer leak. last time I looked before the GEM_BIND error it was around 105 buffers with ~800MB
#asahi-gpu · 2023-04-08 11:39
jannauI'm still seeing GEM_BIND errors after flipping the allocations over to the GPU
#asahi-gpu · 2023-04-08 11:37
jannauwould it make sense to have a different debug flag for submit? that is at least 2 orders of magnitude noisier than other ioctls
#asahi-gpu · 2023-04-07 17:54
jannauso possibly a kde bug
#asahi-gpu · 2023-04-07 17:54
jannaulina: reproducer for the buffer leak: 2 konsole windows tiled left and right + kate maximized. Switch between the windows via activation on the panel. I suspect it might be related to the window preview shown after a while
#asahi-gpu · 2023-03-28 19:30
jannauthe gfx cluster on pro/max/ultra have to communicate somehow
#asahi-gpu · 2023-03-28 11:44
jannaulina: did you see https://www.youtube.com/watch?v=dgd88zE3O38 ?

plus ›