#
✕ nick:lina
#asahi-dev · 2024-09-21 16:07
linaIt sounds like ARM kind of thought to specify this and never did, even though Apple implemented it officially? It would be nice to know what feature, if any, is supposed to represent this...
#asahi-dev · 2024-09-21 16:06
lina"has_actlr_virtualisation"
#asahi-dev · 2024-09-21 16:06
linahttps://developer.arm.com/documentation/100964/1125/Fast-Models-components/Core-components/AEMvACT
#asahi-dev · 2024-09-21 16:06
linamaz: I'm looking at the TSO virt stuff and I discovered something interesting... it seems on M2 and later Apple implements an architectural ACTLR_EL12 at s3_5_c1_c0_1 instead of their old IMPDEF. But it's not mentioned anywhere in ARM docs except here:
#asahi-dev · 2023-10-07 02:02
linaAfter all when you send the image to other people they see the unmirrored version, mirroring is only for local preview
#asahi-dev · 2023-10-07 02:02
linaSo we don't need to worry about that
#asahi-dev · 2023-10-07 02:01
lina(Some probably have toggles for it but it should never be the default and I think videocall apps always mirror internally)
#asahi-dev · 2023-10-07 02:00
linajanneg: Mirroring is always an application thing, as far as I know no webcam mirrors the image internally
#asahi-dev · 2023-10-06 12:09
linajanneg: ^
#asahi-dev · 2023-10-06 12:09
linapushed the tracer/etc changes to isp/t602x-b
#asahi-dev · 2023-10-06 09:09
lina1080p should definitely be correct though
#asahi-dev · 2023-10-06 09:09
linaAlso I didn't technically test any of this on m1pro/max, I just assumed the resolutions are the same as on m2pro/max since they use the same sensor... but it could be wrong.
#asahi-dev · 2023-10-06 09:08
linaIf too much stuff is broken with extra resolutions we can just limit the 1080p cameras to 1920x1080p only for now, that's a simple device tree change.
#asahi-dev · 2023-10-06 09:07
linaGStreamer works though
#asahi-dev · 2023-10-06 09:07
linaI know for one ffmpeg is broken with vertical resolutions (I have a patch for that)
#asahi-dev · 2023-10-06 09:06
linaDepending on what Firefox is doing exactly, we might be able to change the logic in the driver to prefer some resolutions, or hide the weird vertical ones behind a module argument if they cause too much trouble
#asahi-dev · 2023-10-06 09:06
linaCan you try some webcam test sites?
#asahi-dev · 2023-10-06 09:06
linacy8aer: We support a lot of resolutions on those machines, including vertical ones... maybe Firefox/jitsi is unhappy with that?
#asahi-dev · 2023-10-06 09:05
linaAnd that makes sense since 12.4 didn't support any machines where high IOVAs are used
#asahi-dev · 2023-10-06 09:04
linaShould be easy to fix then!
#asahi-dev · 2023-10-06 09:04
linaeiln: Good thing I added the version to the m1n1 side ^^
#asahi-dev · 2023-10-01 05:58
linaThe only consistent part is one of the addresses is always within one of the firmware allocations, but that could be the actual problem or just a coincidence that it's always the last thing it touched when the problem happens... Either way I didn't see anything wrong with that allocation...
#asahi-dev · 2023-10-01 05:58
linaI've never seen any DART faults related to this and the addresses it prints during the PIODMA errors are always valid...
#asahi-dev · 2023-10-01 05:57
linaUsually it dies on the second stream, sometimes the first, sometimes it lasts until the third. It never dies during probe, and the problem always happens on the first frame or after a few frames rarely. So I think something is wrong with the PIODMA hardware, like something doesn't get reset/initialized properly and it raises an error.
#asahi-dev · 2023-10-01 05:56
linaeiln: I'm not sure. I looked through DART/DAPF config and couldn't find anything interesting. My suspicion right now is something missing in the reset/power sequencing.
#asahi-dev · 2023-10-01 03:22
linaeiln: If nobody else figures out what's up with M2 by Wednesday I'll spend that stream trying to fix it, hopefully that's the last one ^^
#asahi-dev · 2023-09-30 10:30
linaOnly DART difference is streams 14 and 15... but I tried enabling them (just as clones) and it doesn't fault or anything, so I don't think that's it...
#asahi-dev · 2023-09-30 10:21
linaPushed some stuff but I haven't really gotten anywhere... Next things to check are DART registers and whether there's any other weird init stuff (tunables...)
#asahi-dev · 2023-09-30 08:43
linaFixed the tracer for t8112, trying to compare things now...
#asahi-dev · 2023-09-30 08:09
linaHmm... back to square one...
#asahi-dev · 2023-09-30 07:55
linaMaybe it doesn't need save/restore, hmm...
#asahi-dev · 2023-09-30 07:47
linaI think it basically worked by accident when it worked ^^;;
#asahi-dev · 2023-09-30 07:47
linaThat and in general more stuff being wrong with the IOVA offset...
#asahi-dev · 2023-09-30 07:33
linaI think it's the DAPF config getting lost when we power down the DART. We might need save/restore in the DART driver for that...
#asahi-dev · 2023-09-30 07:30
linaLooks like that workaround thing is a red herring...
#asahi-dev · 2023-09-30 07:05
linaI'm so confused...
#asahi-dev · 2023-09-30 06:53
linaWoah... what's up with the DAPF property here...
#asahi-dev · 2023-09-30 06:33
linaThat's sus... in the DART only for t8112 ISP...
#asahi-dev · 2023-09-30 06:33
lina dual-vc-llt-workaround-0 = [0x0000000F00000000, 0x0000001000000000]
#asahi-dev · 2023-09-30 06:30
linaYeah, I repro'd... it's not MLVNR, I need the workaround for that on 1080p sensors as with t602x, but this is something else
#asahi-dev · 2023-09-30 06:19
linaLooking at t8112...
#asahi-dev · 2023-09-30 03:32
linaI think so! Just need to fix up M2 12.4 I think? And test the other machines
#asahi-dev · 2023-09-30 00:08
linaIn good news: It survived all night!
#asahi-dev · 2023-09-30 00:08
linaI wanted to get rid of the iotranslate stuff because I'd seen it crash there and also some really ugly memory corruption stuff (filesystem page cache oopses) and I felt really uncomfortable about doing it that way... ^^;;
#asahi-dev · 2023-09-30 00:07
linaeiln: Yeah, the surface filter is a hack... Though 0xc000 does work in practice ^^
#asahi-dev · 2023-09-29 15:34
linaI should sleep ^^
#asahi-dev · 2023-09-29 15:34
linaActually removed it now anyway, in a fixup
#asahi-dev · 2023-09-29 15:30
linaI'm going to leave it all night starting/killing captures while looping some ffmpeg compiles, if nothing goes horribly wrong I think we're good ^^
#asahi-dev · 2023-09-29 15:21
linaOK, looks much cleaner now ^^
#asahi-dev · 2023-09-29 15:13
linaSo on signals all that breaks...

plus ›