#
#asahi-dev · 2026-11-10 17:21
svenYou’ll also need a notification from tbt tunnel up to dpin but feel free to just hack that for now
#asahi-dev · 2026-11-10 17:21
svenand yes, HPD comes from dpin
#asahi-dev · 2026-11-10 17:21
svenyup! the header file already contains all registers I think so tracing should give you the right order then
#asahi-dev · 2026-11-10 17:07
danield75That makes sense- so I should probably trace the MIMO pokes via M1N1 to the DP-in phy? Assuming I can use the address for atc0-dpin in the ADT (0x701E50000 for the J414). It looks like I'd read the HDP level bit from the DP-in phy instead of the CD321X. Thanks for the info- I'll look around some more and see what I find :)
#asahi-dev · 2026-11-10 15:54
jannauatcN-dpin[01], see https://f.svenpeter.dev/d09f1fa677f4a843e9c9eb42840f8992861131d2474ec1c200760a5c7fd9c13f_dpin.h how it works
#asahi-dev · 2026-11-10 15:50
svenfor tbt tunnels the dp aux channel is tunneled as well and that thing usually sends HPD updates
#asahi-dev · 2026-11-10 15:50
svenit’s very simple and iirc HPD must come from there as well
#asahi-dev · 2026-11-10 15:50
sventhat thing should be in the ADT
#asahi-dev · 2026-11-10 15:34
svenit needs a bunch of MMIO pokes after the tunnel is up
#asahi-dev · 2026-11-10 15:34
sventhere’s a DP-in phy which is very simple
#asahi-dev · 2026-11-10 15:30
jannaunot sure how the squencing there is supposed to work but I'd assume the there have to start things on USB4 side before you can connect/HPD dcp
#asahi-dev · 2026-11-10 15:27
jannauDPTX_APCALL_DEVICE_NOT_RESPONDING (22) indicates that the signal path between dcp and host router and/or display is not up
#asahi-dev · 2026-11-10 15:24
jannauI would expect it to work mostly the same as dp-altmode with selecting dpin0 in the display crossbar, not interacting with atc-phy and then handling everything on usb4 side
#asahi-dev · 2026-11-10 15:14
danieldBased on the fact I had to hack the HPD bit in the typeC driver to get the DCP to even attempt to connect, I'm wondering if the 18,10, 0 AP call sequence indicates some initial progress, or if thunderbolt just isn't set up correctly (I'm aware it isn't supported yet on the fairydust branch). Any thoughts?
#asahi-dev · 2026-11-10 15:13
danieldI have a branch here for the M2 Pro: https://github.com/danieldegrasse/linux/tree/wip/m2-dp-hacks, which gets the following AP calls from the DCP: 20, 1, 20, 18, 10, 0, 22, 24. Based on that, it seems like the *initial* set of calls is matching what Sonoma gets, but then the DCP seems to drop connection (calls 22 and 24)
#asahi-dev · 2026-11-10 15:10
danieldHi- I've been playing with enabling DP alt mode on a thunderbolt dock, and was wondering if someone more familiar with the DCP had any thoughts. I setup Sonoma under the M1N1 hypervisor, and grabbed a trace of the DCPEXT. Based on that, it looks like the AP calls Sonoma gets are: 18, 10, 0, and 21.
#asahi-dev · 2026-11-10 13:37
ayaanahi guys
#asahi-dev · 2026-11-10 12:49
jannauwe even merged the uapi in mesa before v6.16-rc1
#asahi-dev · 2026-11-10 12:48
chadmedthat is what i meant, not capacity
#asahi-dev · 2026-11-10 12:46
jannauI don't see how the compressed formats help with memory pressure. they actual use more memory. they help with memory bandwidth
#asahi-dev · 2026-11-10 12:34
chadmeds/a/an
#asahi-dev · 2026-11-10 12:34
chadmedi might open a MR this weekend then!
#asahi-dev · 2026-11-10 12:32
jannauchadmed: I think landing in v7.4-rc1 is enough. I think we merged the driver itself after rc1 with asahi uapi was out
#asahi-dev · 2026-11-10 12:00
s0ullight[m]Alrightie, I'll polish up my hk improvements and try to upstream. Also will try to get the conversation going regarding fence passing
#asahi-dev · 2026-11-10 11:54
chadmedbut it cant go upstream until the format modifier lands in the drm headers, which should be with 7.4
#asahi-dev · 2026-11-10 11:53
chadmedthe dcp driver already supports it, and i have a local copy of mesa with everything wired up and no cts regressions
#asahi-dev · 2026-11-10 11:52
chadmedbut yes interchange support in mesa will make direct scanout more feasible and also reduce memory pressure during framebuffer copies
#asahi-dev · 2026-11-10 11:52
chadmedim not looking at kwin because c++ is the worst thing since leaded petrol
#asahi-dev · 2026-11-10 11:51
s0ullight[m]chadmed: Is your work also covering this?
#asahi-dev · 2026-11-10 11:51
chadmedbut all of them do support direct scanout of the application's framebuffer without compositing
#asahi-dev · 2026-11-10 11:50
chadmednope, windows fakes it
#asahi-dev · 2026-11-10 11:50
chaos_princessdoes any modern os even do the "classic" fullscreen?
#asahi-dev · 2026-11-10 11:50
s0ullight[m]I think after virtio implicit sync that's the biggest cost for higher resolutions
#asahi-dev · 2026-11-10 11:50
chadmedso everything is massive linear framebuffers
#asahi-dev · 2026-11-10 11:49
chadmedespecially since interchange is not yet supported in mesa (waiting for the format modifier to land in a released kernel)
#asahi-dev · 2026-11-10 11:49
chadmedframebuffer copies are probably a major source of performance degradation yes
#asahi-dev · 2026-11-10 11:48
s0ullight[m]I've been trying to figure out the fps tanking with higher resolutions and just realized it matches macos if the game runs windowed
#asahi-dev · 2026-11-10 11:47
chadmedbut direct scanout for rgb formats is a limitation of the compositor itself now
#asahi-dev · 2026-11-10 11:47
chadmedwell yes ycbcr wont help with that
#asahi-dev · 2026-11-10 11:46
s0ullight[m]Games yes
#asahi-dev · 2026-11-10 11:46
chaos_princessi think the question was more in relation to games, not video
#asahi-dev · 2026-11-10 11:45
chadmedif youre looking for gpu stuff to fix/clean up, you could look into ycbcr image formats on agx :)
#asahi-dev · 2026-11-10 11:45
chadmedkwin still doesnt like doing direct scanout with this sort of setup (dcp is treated as a second gpu) but we're getting there
#asahi-dev · 2026-11-10 11:44
chadmedyes but im working on it
#asahi-dev · 2026-11-10 11:42
s0ullight[m]We still pay for the compositor in Fullscreen apps right?
#asahi-dev · 2026-11-10 11:22
chadmed(well simplest laptop)
#asahi-dev · 2026-11-10 11:22
chadmedthatll teach me for testing on the simplest possible machine...
#asahi-dev · 2026-11-10 11:18
chadmedjannau: oh yes that makes more sense. thanks
#asahi-dev · 2026-11-10 11:13
yuka->dm
#asahi-dev · 2026-11-10 11:12
integralpilotdo you want me to send or wait for you to have some time to think about it so we don't contridict each other

plus ›