#
…

2026-09-24

02:00yurix has quit [Remote host closed the connection]
02:00yurix has joined #asahi-dev
02:25pb17 has quit [Ping timeout: 480 seconds]
02:28pb17 has joined #asahi-dev
02:48tobhe_ has joined #asahi-dev
02:55tobhe has quit [Ping timeout: 480 seconds]
03:20marvin24 has joined #asahi-dev
03:55nora_ has joined #asahi-dev
03:56nora has quit [Ping timeout: 480 seconds]
04:04saiprasad_ has joined #asahi-dev
04:06zumi_ has joined #asahi-dev
04:10zumi has quit [Ping timeout: 480 seconds]
04:11zumi_ is now known as zumi
04:16pb17 has quit [Ping timeout: 480 seconds]
04:25pb17 has joined #asahi-dev
04:29eluks has quit [Remote host closed the connection]
04:29eluks has joined #asahi-dev
04:59Peiran has joined #asahi-dev
05:31schappi2610[m]integralpilot: it took a while yesterday but it worked. i expanded my grub to boot both kernels: fairydust and stable. Now i encounter the situation that i can use my DP-USB-C monitor on USB-C port 2 but not on Port 1. No big deal for me. in the logs i can see that there is happening a lot when plugging in ;-) USB-C → DP → HPD → EDID/Modi → Modeset → HPD fails → Disconnect → Reconnect → again Modeset
05:36myriaCurrently fairydust blesses a port iirc, so I think this is expected?
05:36JamesCalligeros[m]yes this is expected
05:46WindowPain3 has joined #asahi-dev
05:46WindowPain has quit [Ping timeout: 480 seconds]
05:46WindowPain3 is now known as WindowPain
06:13mildsunrise has quit [Quit: disconnect]
06:14mildsunrise has joined #asahi-dev
06:16elvishjerricco4 has joined #asahi-dev
06:22elvishjerricco has quit [Ping timeout: 480 seconds]
06:22elvishjerricco4 is now known as elvishjerricco
06:31bennofs3 has quit [Ping timeout: 480 seconds]
06:38bennofs3 has joined #asahi-dev
07:18hardy94 has joined #asahi-dev
07:26N_Tempus_04 has joined #asahi-dev
07:56pb17 has quit [Ping timeout: 480 seconds]
07:58pb17 has joined #asahi-dev
08:02N_Tempus_04 has quit [Ping timeout: 480 seconds]
08:06sofusdoes anyone have the data for a 10b interchange surface?
08:07sofusie tile size and so on
08:09sofusi meant everything, because something is wrong
08:13JamesCalligeros[m]What is wrong? dcp works with 10 bit argb
08:20sofusim talking about yuv surfaces, im getting Decomp fail messages
08:31msimonelli2 has quit [Quit: Ping timeout (120 seconds)]
08:32msimonelli2 has joined #asahi-dev
08:35JamesCalligeros[m]Janne had some manifest of pixel formats a little while ago but i cant find it
08:35JamesCalligeros[m]8 bit formats work right?
08:36sofusi have only tested nv12, but yes
08:38saiprasad_ has quit [Remote host closed the connection]
08:38saiprasad_ has joined #asahi-dev
08:39sofusi think im using the wrong format
08:46mattia013 has quit [Remote host closed the connection]
08:47sofusthanks!
08:53integralpilot has joined #asahi-dev
09:07N_Tempus_04 has joined #asahi-dev
09:11sofusyeah p010 is wrong for interchange i think, or at least avd seems to output something else. We should have 10 bpp (ie. packed)
09:12sofusit could actually be the same format as the one i found
09:28sofusyes! after changing tile_bytes calculation, using 10/20 for bpp and changing the dcp format to 'p420' it works!
09:32sofusis userspace supposed to handle the size calculation? In particular the offset to the second plane
09:36jannaudata_offset
09:37mattia013 has joined #asahi-dev
09:42BlueMatt_ has joined #asahi-dev
09:44jannaudoes anyone know if the "NVM EXPRESS DRIVER" MAINTAINERS entry intentionally does not cover Documentation/devicetree/bindings/nvme/
09:46sofusi think thats for something else. At least with gstreamer i think it ignores all v4l2 fields for the drm sizes/pitches
09:47svenjannau: probably because i forgot to add it when I added the initial nvme driver :(
09:48sofusi calculate the offset in the dcp driver right now, but i would assume thats forbidden
09:48BlueMatt has quit [Ping timeout: 480 seconds]
09:48svenours was the first (and possibly only) that uses dt-bindings
09:53jannausofus: then gstreamer is broken and implicitly relies on either having just a single plane, no offset for the chroma plane(s) or "knowing" the right offset for specific hardware
09:54jannausven: indeed, took a while to consider that the apple one is the only one which needs dt-bindings
09:58JamesCalligeros[m]sofus: the drm driver should not be doing special handling for misbehaving clients
09:58JamesCalligeros[m]If the clients are not implementing their consumed APIs properly then they are broken
10:00sofusi found it, its using the pitch 'calculcated by the kms driver' to calculate the offsets according to a comment
10:05zumi_ has joined #asahi-dev
10:05sofusit gets it via DRM_IOCTL_MODE_CREATE_DUMB
10:07saiprasad_ has quit [Remote host closed the connection]
10:10zumi has quit [Ping timeout: 480 seconds]
10:10zumi_ is now known as zumi
10:12jannauafaik DRM_IOCTL_MODE_CREATE_DUMB is not supposed to be work/used with modifiers
10:18sofuslol yeah 'KMS dumb buffers are not suitable for hardware-accelerated rendering nor video decoding'
10:26leioMaybe ndufresne has some thought about the above when he wakes up
10:32jannauI think there were recently discussions to use custom create buffer IOCTLs for kms only devices to support video / compressed buffers
10:32jannauprobably in dri-devel
10:55pb17 has quit [Ping timeout: 480 seconds]
10:56pb17 has joined #asahi-dev
11:14ndufresnesofus: I'm missing some context, but it seems unlikely, specially for when opaque DRM formats are negotiated. But perhaps you mean we don't interpolate offset and second stride, I know there is still few gaps to fill there when drivers uses single allocation (which in v4l is forces to use the single plane interface, the API is really bad)
11:16ndufresneBut we can code HW specific offset calculations in gst v4l plugin, it's much more productive then wait for v4l3
11:18ndufresneLet's say, we got away for now with AFBC since it's a single plane, we need to do more work, rmader was on that recently
11:20chadmedthe tricky thing is that interchange/agx are similar to afbc in this regard. software is supposed to treat them as opaque/single-plane formats but the interface to the display controller requires the compression buffer metadata to be programmed in by us
11:21chadmedbut we need to negotiate the correct pixel format and drm format modifier to make both the codec and display controller work properly so we cant rely on create_dumb and kludging things in the kms driver
11:28sofusndufresne: i saw the HW specifik calculations but im not sure they make a difference when using the kmssink
11:29sofusbut anyway, my main goal was just testing everything works
11:30ndufresneRight, kmssink is on its decay, I don't use it anymore myself, I always go for Wayland (Weston for niche use case, mutter/KDE for desktop work)
11:35sofusi also only learned about it for testing this specifik feature. I wish i can get it to run on wayland
11:38chadmed has quit [Ping timeout: 480 seconds]
11:39chadmed_ has joined #asahi-dev
11:39chadmed_ is now known as chadmed
11:39chadmedit is still useful for testing stuff like this without the interference of a wayland compositor doing unexpected things
11:40chadmedkwin for example still has issues with direct scanout on machines where the gpu and display controller are not the same thing, as it treats this as a multi-gpu setup
11:48ndufresneRight, for testing on Wayland I aim for Weston, which has less magic, more config, but I won't hide that I work with bunch of Weston devs to help me
11:49ndufresneIt would be nice to keep kms bit in shape, but it needs work, and needs HW specific allocation support
11:49N_Tempus_04 has quit [Ping timeout: 480 seconds]
11:50BlueMatt_ is now known as BlueMatt
11:52integralpilot has quit [Quit: integralpilot]
12:13mango-602cf7 has joined #asahi-dev
12:14mango-602cf7Hey. x86-32 emulator author here. I couldn’t get a clear answer out of Google. Is there a per-process, or hypervisor, or any way to get 4k pages on Asahi? I’ve got an EL1 hypervisor going on my MacOS version
12:15mango-602cf7 is now known as originaldave_
12:15originaldave_I ask because I was about to totally drop all the slow 16k page support stuff, but realized I might need it on Asahi
12:18linkmauveoriginaldave_, perhaps have a look at how FEX does it, alongside muvm?
12:18linkmauveHow does your emulator compare to FEX btw?
12:19chaos_princessoriginaldave_: use muvm.
12:22originaldave_mine is LLE/system emulation focused, compared to FEX which is HLE. they’re great folks though, and I’m looking at an interesting hyrid LLE-HLE method for the best of both worlds
12:22pb17 has quit [Ping timeout: 480 seconds]
12:22nickchanfull system emulation is interesting because even without TSO
12:23chadmedndufresne: Weston is not suitable for us as it does not gracefully handle us rejecting swaps when the breeze is blowing or mercury is in retrograde
12:23pb17 has joined #asahi-dev
12:23nickchanthe software MMU emulation is the perf killer of epic proportions
12:23chadmedkwin at least will immediately fall back to software compositing if it tries to do something dcp wouldnt like (e.g. putting a cursor too close to the edge of the screen lmao this display controller suuucks)
12:23linkmauvechadmed, Weston doesn’t work correctly on Asahi?
12:23nickchanmerely from being able to use the hardware mmu via shadow paging in a virtual machine could improve performance
12:23linkmauveDid you open an issue for that?
12:24chadmedno i havent opened an issue for it because its a silly edge case
12:24chadmedweston wont put a cursor on an overlay plane so you need to explicitly expose the plane as a cursor plane (which we dont do in production)
12:27ndufresnechadmed ok, Canadian loves talking about weather, but you lost me about breeze and mercury here !
12:27chadmedlmao sorry im being cheeky, im saying that dcp is extremely temperamental and will crash if something is even slightly outside of what it expects
12:28chadmedsaying that it will crash if a strong breeze is blowing or the planets are orbiting a certain way is saying that it's extremely flakey and unreliable and fussy
12:28ndufresneIf there is anything in Weston that is incompatible with asahi gpu driver, file an issue, I'm sure we can fix it, but I'm not saying its good for desktop, use kde/mutter/wlroot based stuff, but its been better for testing overlay/underlay at low level then kmssink
12:29root[mds] has joined #asahi-dev
12:29chadmedthis is pretty much the only (major) issue ive encountered and like i said it was just because i was doing stuff we would never ship anyway
12:29ndufresneFun fact, I remember working on the display controller of the OpenMoko Freerunner, and one bug requires to place the phone in the freezer for 30min or so and resume from suspend at low temprature to reproduce
12:30saiprasad_ has joined #asahi-dev
12:30svenlol :D
12:31myriaHey, I'm adding some extra supported information to the website (see https://github.com/AsahiLinux/docs/pull/278)
12:32myriaBut question I had, do we want a separate page for the A18, or just lump it in with M4?
12:33chaos_princessthrow it in together, a18 does not deserve a page of it's own :P
12:33nickchana-series and m-series relations are a mess since the A16/A17/M3
12:33chadmedA18 & M4 Series Feature Support or something like that
12:33myriaFair enough, pushing the extra stuff now
12:33chadmedor even just M4 Series Feature Support (incl. A18 Pro)
12:34nickchanto the point m6/a20 cores are materially different from each other
12:35chadmedi feel like theyll converge again
12:36chadmedthese were all decisions that have to have been made ~3 years ago before everything became expensive and on fire and evil
12:37LarstiQndufresne: living in Finland after the openmoko era, you mean regular winter temperatures? ;)
12:39chaos_princessM3 is also a pretty messy generation
12:40nickchanm3pro ecore != m3max ecore 🙃
12:40chadmedtbf this was the case even for m1
12:40chadmednot at any user-facing level but they had different chickens and core IDs
12:41chadmedwhich is why lscpu and kde system info will tell you which soc the core is on
12:41nickchanm1 is close enough
12:41nickchanthe pro/max/ultra variant is a bit different from the PA space though
12:43nickchanI have not seen a shared midr on apple soc since a10/t2
12:43fl0_id_nickchan what's the difference, different chickens and IDs? or actually also different functioning
12:44nickchanfl0_id_: number of FP/SIMD execution units
12:44nickchanit is detailed in the apple silicon cpu optimization guide
12:44nickchanm3 max ecore has 3 units and m3 pro ecore has 2 units
12:44nickchanfunctionally it can be seen from using the performance counters
12:45nickchanm6/a20 is another case of cores having different performance characistics
12:45fl0_id_ah thx. so m3 max ecore prob also uses more power i guess
12:46nickchanm6 clock speed < a20 clock speeds
12:46yukaaddressed the review comment on https://github.com/AsahiLinux/m1n1/pull/636 from the other day & tested on t6000, t6034
12:46yuka(added defines for the return codes)
12:47nickchanmy rtkit crashlog patches are sitting on the list
12:49myriaAny updates on your A7-A11 PMU patches? Saw a new version at the start of the year but nothing since
12:51nickchanthey are going nowhere so i stopped sending them
12:51nickchanIt is still in my downstream tree and covered by my tests
12:51nickchanI may send it again if someone sends the m3 pmu patches
12:52myriaUhhh, I've been meaning to look into them
12:52ndufresneLarstiQ yep, regular temperature, I was also reproducing in Montreal trying to daily that phone (it never worked well though)
12:52myriaThe event code may have changed to be more than a byte so need to trace M3 to see what's happening there
12:54nickchanmost of the changes are the event list
12:54nickchanthe rest of the changes is relatively small
12:55nickchanmyria: cpmu is in oss xnu
12:55myriaIs that free to look at? I'm just worried about RE stuff
12:56myriaAnd licenses
12:56nickchanoss xnu does say m3 has 16bit event codes
12:56chadmedyou can look at the oss xnu for facts about the hardware, which are not copyrightable
12:59nickchanfor m3+ i think the only useful patch to you from my patches is the one that allows each implementation to declare their own event tables
13:00nickchanif events are 16bit then it's quite likely the event table is different
13:02N_Tempus_04 has joined #asahi-dev
13:24yukait's always the same
13:25yukalooking over everything 10x before sending out the patch series
13:25yukathen noticing a mistake right after
13:25yukaI forgot one "arm64: " in the commit message prefixes
13:32JamesCalligeros[m]Yup literally every single tomr
13:32yukaintegralpilot: is your latest dchid state somewhere on github?
13:41sveneh, i can adjust the messages when picking the series up
13:41svenno need to resend for minor stuff like that
13:42yukayep. but... my poor neurospicy brain
13:54integralpilot has joined #asahi-dev
13:55integralpilotyuka: sorry, v4 draft is on my other laptop and i was about to go to bed. i'll push it tomorrow morning
13:56integralpilotand then list (along with BRCM4388 v1) tomorrow evening
13:56integralpilotunless you have anything you'd like changed
13:58yukano worries
13:58yukano I just wanted to test it
13:59integralpilot has quit []
14:00catred has joined #asahi-dev
14:08linkmauvendufresne, did that Freerunner driver ever get mainlined btw? I took mine out a few years ago and was pretty disappointed I still had to use a 2.6.3x kernel…
14:20yukaI think the dt-check-style ordering requirements have changed in linux-next, model and compatible are now the other way around /o\
14:20yukaso depending on which branch I'm working on, I get errors or not
14:20svenlet’s maybe hold off on strict warnings then
14:21yukaAlso, they newly added a mixed-indent-chars?
14:21yukawhich is not even strict-only
14:21yukaI dislike it
14:23yukabut might need to stop properly aligning the properties with line breaks in the next device trees since that's not really universally possible with tabs-only
14:23steven has quit []
14:24integralpilot has joined #asahi-dev
14:24integralpilot has quit []
14:25steven has joined #asahi-dev
14:36jannauwas the not exactly aligned strict check dropped? it sounds like that would conflict with mixed-indent-chars
14:40jannau80 char line length moved to --mode stricter in next
14:43catred has quit [Remote host closed the connection]
14:47jkkm has quit [Read error: Connection timed out]
14:49jkkm has joined #asahi-dev
14:58hardy94 has quit [Remote host closed the connection]
15:02myrianickchan: Yeah, the cpu tables are definitely different. And I think M3/M4 are different, but M5 seems to just be extra events on top of m4
15:03myriaSame with m6, so it might just be a case of M3 table, then M4/5/6 tables
15:03myriaThough not sure I want the affinities to be an array, there will be too many spaces for it to be an efficient use of memory
15:12gruetzkopf has quit []
15:18minecrell8 has quit [Read error: Connection reset by peer]
15:20minecrell8 has joined #asahi-dev
15:41aditya has joined #asahi-dev
15:48croissant` has joined #asahi-dev
15:53croissant has quit [Ping timeout: 480 seconds]
15:57croissant` has quit []
15:58croissant has joined #asahi-dev
16:14yukasven: for SARTv4 support in Linux, would you a refactor or just copy, paste and adjust the sartv3 defines and functions?
16:14yuka*would you prefer
16:14yukaIt's the same thing as in m1n1, the code is almost the same except different offsets
16:15sveni'm not a big fan on the m1n1 refactor tbh, i'd just copy/paste
16:15yukalet's close the m1n1 PR as well then, since jannau and you are both slightly against it?
16:16svenworks for me
16:25originaldave_ has quit [Read error: Connection reset by peer]
16:28N_Tempus_04 has quit [Ping timeout: 480 seconds]
16:44yukado we follow apple's odd naming of t6041 (not t6040) for t604x base compatibles?
16:45yukaLiterally the M4 Pro Mac mini (j773s) has compatible = [arm-io,t6041]
17:01ddxtanx has joined #asahi-dev
17:02lion328 has quit [Remote host closed the connection]
17:04lion328 has joined #asahi-dev
17:32ddxtanx has quit [Remote host closed the connection]
17:51aditya has quit [Quit: Connection closed for inactivity]
17:53Overtone[m] has joined #asahi-dev
17:59yukado I call the M5 Max SoC t6051? there is no single mention of t6051 in an M5 Max adt
18:00yukaBut I think it did have a different chip id?
18:01Overtone[m]Is it that important to call it what Apple arbitrary calls it inside the company instad of just saying the public known name?
18:02Overtone[m]just sayin. new to topic and dont know too much.
18:05meow67676767 has quit [Read error: Connection reset by peer]
18:16yuka> do I call the M5 Max SoC t6051? there is no single mention of t6051 in an M5 Max adt
18:16yukaah but this was already the case for M3 Ultra for example
18:18nickchanm5 max and m5 ultra is chip id 0x6050
18:18jannau m1 max has only 'compatible = [pmgr1,t6001]'. it's based on the chip_id
18:19nickchant6051 doesn't seem to exist
18:19yukaHmm okay
18:20meow67676767 has joined #asahi-dev
18:21jannaum5 ultra has the same chip_id as max? this sounds unexpected. I would have also expected that M5 Pro and Max have different chip_ids
18:27mmediouni[m]M5 Pro and Max are different tapeouts for the base tile that contains dram controllers and GPU
18:27mmediouni[m]CPU tile is the same though
18:31yukaM5 Pro and M5 Max cores definitely have different MIDR part numbers
18:31yukabut possibly same chip id
18:31yukaI don't remember
18:32chaos_princesslet's just do t6051 for consistency and hope that apple does not release something that conflicts with that number
18:33jannauI would expect the combination of cpu and base tile to have each their own chip-id, 27.0 Devicetrees from the ipsw have 0 as chip-id
18:35jannaum5 ultra has "clpc,t6050d", dispext has "dispext0,t605x"
18:37jannau`ioreg -l | grep chip-id` from a M5 Max / Ultra should clear things up
18:39hardy94 has joined #asahi-dev
18:54saiprasad_ has quit [Remote host closed the connection]
19:03barusu has quit [Remote host closed the connection]
19:04barusu has joined #asahi-dev
19:21jannauyuka: strict dt-style-check from next improves the results for t603*/t8122*
19:23jannauthe model/compatible property order and 2 new node order issues but the line length and dcache-size property issues are gone
19:25hardy94 has quit [Quit: hardy94]
19:27hardy94 has joined #asahi-dev
19:35yukayeah okay I aligned the model/compatible order to the old version in the patch series I sent out earlier
19:39N_Tempus_04 has joined #asahi-dev
19:47N_Tempus_04 has quit [Ping timeout: 480 seconds]
19:49N_Tempus_04 has joined #asahi-dev
19:54hardy94 has quit [Remote host closed the connection]
20:07fugi5 has quit [Quit: bye]
20:08fugi5 has joined #asahi-dev
20:15yurix has quit [Read error: Connection reset by peer]
20:15yurix has joined #asahi-dev
20:25N_Tempus_04 has quit [Ping timeout: 480 seconds]
20:27picnoir has quit [Read error: Connection reset by peer]
20:28picnoir has joined #asahi-dev
21:09lucy_ has joined #asahi-dev
21:24bennofs3 has quit [Ping timeout: 480 seconds]
21:25jvoisin has quit [Quit: jvoisin]
21:25jvoisin has joined #asahi-dev
21:25bennofs3 has joined #asahi-dev
21:28Ardu0us_ has quit [Quit: %Reality broken: (I)gnore, (R)etry, (A)bort%]
22:10jsk_ has joined #asahi-dev
22:12bennofs3 has quit [Quit: Ping timeout (120 seconds)]
22:12bennofs3 has joined #asahi-dev
22:15jsk has quit [Ping timeout: 480 seconds]
22:15Ardu0us has joined #asahi-dev
22:18jsk_ is now known as jsk
22:21lucy_ has quit [Quit: Konversation terminated!]
22:30nickchan has quit [Quit: ZNC 1.9.1+deb2+b3 - https://znc.in]
22:30nickchan has joined #asahi-dev
23:41distantorigin has quit [Ping timeout: 480 seconds]
23:57distantorigin has joined #asahi-dev
23:57integralpilot has joined #asahi-dev
…