#
#asahi-gpu · 2022-11-10 21:21
alyssajannau: woo!
#asahi-gpu · 2022-11-10 21:21
jannaualyssa: turns out iova.c works if used properly. proper fix pushed. "leaked" memory is freed on terminating weston
#asahi-gpu · 2022-11-10 17:32
alyssalina: thank u
#asahi-gpu · 2022-11-10 16:38
linaI get the feeling I already know what 49:0 is... I bet it's related to unk_flag and the things computed from it. Possibly unk_378.
#asahi-gpu · 2022-11-10 16:35
linaIf you're in a rush I can try to find some time this sunday maybe
#asahi-gpu · 2022-11-10 16:34
linaalyssa: Got it! I'll add those on wednesday next week if that's okay ^^
#asahi-gpu · 2022-11-10 14:34
alyssaSee my cmdbuf.xml file in asahi/mesa:agx/mrt for notes
#asahi-gpu · 2022-11-10 14:34
alyssaI probably also need "Unk 49:0"
#asahi-gpu · 2022-11-10 14:33
alyssa"Tile width", "Tile height", and "Unk 212" from the IOGPU structure, I need plumbed into Linux
#asahi-gpu · 2022-11-10 14:33
alyssalina: I have some kernel feature reqs (blocking GL3 Mesa work)
#asahi-gpu · 2022-11-10 03:58
alyssathat's like, basically passing
#asahi-gpu · 2022-11-10 03:57
alyssa Failed: 51/940 (5.4%)
#asahi-gpu · 2022-11-10 03:57
alyssa Passed: 889/940 (94.6%)
#asahi-gpu · 2022-11-10 02:46
alyssadown to 60+, okay I'll take the win
#asahi-gpu · 2022-11-10 02:21
alyssamerges some code
#asahi-gpu · 2022-11-10 02:11
alyssaso i'm now 84 commits on top of main ... ugh
#asahi-gpu · 2022-11-10 02:11
alyssaWell, just merged agx/next with lina's stuff
#asahi-gpu · 2022-11-10 01:57
alyssajannau: nice
#asahi-gpu · 2022-11-09 23:31
maximbazblocking dcp unfortunately didn't give any extra information - however thank you for your helpful attitude, it's awesome to learn something new here, and I'll try to collect some more useful info when I get my hands on another device
#asahi-gpu · 2022-11-09 23:23
jannaualyssa, lina: https://github.com/jannau/linux/tree/gpu/rust-wip-dcp first patch is already in asahi-wip, fixes a dart build as module, 2nd commit prevents hitting that warning in dart_map_pages() but of course doesn't fix exhausting the iova space through memory leaks
#asahi-gpu · 2022-11-09 22:52
jannauI wonder if the issues with reserved low iova would go away with unmanaged iommu domains. I suspose it'll remain a problem disp0 which I'd like to keep managed though
#asahi-gpu · 2022-11-09 22:38
jannauI found an ugly workaround: exclude the low reserved iova via aperture_start
#asahi-gpu · 2022-11-09 22:02
jannaumaximbaz: add module_blacklist=apple_dcp to the kernel commandline
#asahi-gpu · 2022-11-09 21:46
maximbazI'm up for some more experiments! May I ask for a hint, how to block dcp, or where to read more about it?
#asahi-gpu · 2022-11-09 21:45
jannaulooks like iova.c can't handle that there's reserved mem at iova 0x0
#asahi-gpu · 2022-11-09 21:32
jannauyou could try to block dcp to see if that gives more information
#asahi-gpu · 2022-11-09 21:16
maximbazgotcha, it's unfortunately the only device I have until weekend, I'll try to get more info then! thanks again for your answers! :)
#asahi-gpu · 2022-11-09 21:13
alyssajannau: having fun? :p
#asahi-gpu · 2022-11-09 21:13
jannauwth! dma_alloc_coherent() fails
#asahi-gpu · 2022-11-09 21:06
jannauI guess the display could also freeze because dcp takes over. it would be helpful if you can boot with m1n1's hypervisor and the virtual uart
#asahi-gpu · 2022-11-09 21:03
maximbazasahi.debug_flags=0xfffffffff, no quiet)... which makes me think somehow the freeze happens before the driver is running 🤔
#asahi-gpu · 2022-11-09 21:03
maximbazthanks jannau, I realized that dev_info!(dev, "MMU: Initializing...\n") - which I see in dmesg on HEAD~1 of linux/gpu/rust-wip just before driver crashes - is not conditional on asahi.debug_flags anyway, and on HEAD I don't see that line on screen after selecting a grub entry before everything freezes (with loglevel=7 and
#asahi-gpu · 2022-11-09 21:00
alyssawhoops
#asahi-gpu · 2022-11-09 20:55
jannaubut fixing it breaks dcp since it prevents now the memmap of the rtkit shmem
#asahi-gpu · 2022-11-09 20:52
sven:)
#asahi-gpu · 2022-11-09 20:52
jannauso, yes, the leak found a bug via the WARN_ON
#asahi-gpu · 2022-11-09 20:51
jannaukernel bug spotted. it reserves the physical address. oversight on my part when I added support for translated reserved regions
#asahi-gpu · 2022-11-09 20:37
alyssa1 kernel and 1 userspace
#asahi-gpu · 2022-11-09 20:36
alyssaSo there are (at least) 2 distinct bugs here then
#asahi-gpu · 2022-11-09 20:36
alyssaOof
#asahi-gpu · 2022-11-09 20:32
jannauit tries to map something in the middle of the boot framebuffer
#asahi-gpu · 2022-11-09 20:16
jannaudart-disp0 stream 0 looks full, except for some small gaps
#asahi-gpu · 2022-11-09 20:14
alyssaMy memory leak is helpful after all!
#asahi-gpu · 2022-11-09 20:14
alyssaahahaha
#asahi-gpu · 2022-11-09 20:11
jannauso possible just a bug in the reserved memory handling on iommu side
#asahi-gpu · 2022-11-09 20:10
jannauso the leak exhaust all iova space and the iommu code tries to use the boot mappings
#asahi-gpu · 2022-11-09 20:06
jannaumaybe an issue with the reserved memory
#asahi-gpu · 2022-11-09 20:05
jannauit is WARN_ON(ptep[i] & APPLE_DART_PTE_VALID) from the inlined dart_init_pte()
#asahi-gpu · 2022-11-09 19:59
jannaureproduced
#asahi-gpu · 2022-11-09 19:48
jannaufunky, it takes ~2 seconds in weston+Xwayland until the glxgears window vanishes

plus ›