#
✕ nick:chadmed
#asahi-dev · 2026-01-03 11:33
chadmedmost of it is iomfb which used to be a kext that got moved to the other side of the iommu because it was so full of security holes
#asahi-dev · 2026-01-03 11:32
chadmedthe uncompressed blob is 9 MB :P
#asahi-dev · 2026-01-03 11:32
chadmed>some
#asahi-dev · 2026-01-03 11:27
chadmedhaving the firmware do all of the really hard stuff is actually quite pleasant. the most annoying part is that we have to bend atomic modesetting around what iomfb wants to do, and this requires some pretty heinous crimes against the atomic api
#asahi-dev · 2026-01-03 11:17
chadmedbut its going to need to go upstream eventually so its gonna need to be good, and we might as well make it do what it's supposed to while we're at it
#asahi-dev · 2026-01-03 11:17
chadmedthe driver is really scary and held together with hopes and prayers so no one really wanted to touch it :p
#asahi-dev · 2026-01-03 09:12
chadmedprobably because thats the sane thing they should have done
#asahi-dev · 2026-01-03 09:12
chadmedoh right, idk why i thought it was its own boot policy knob
#asahi-dev · 2026-01-03 09:10
chadmedif we install stage1 m1n1 as a mach-o do they still get disabled
#asahi-dev · 2026-01-03 08:22
chadmedall the important stuff is there though
#asahi-dev · 2026-01-03 08:22
chadmedurgh the lines got truncated
#asahi-dev · 2026-01-03 01:59
chadmednull pointer deref in asahi driver on asahi-wip
#asahi-dev · 2026-01-03 00:20
chadmedgoodnight, nah already most of the way through :)
#asahi-dev · 2026-01-03 00:15
chadmed(apart from kwin trying to put 1 and 3 on the same plane)
#asahi-dev · 2026-01-03 00:15
chadmedmy scene last night was ar30 srgb on 0, ar24 srgb on 1, nv12 bt.709 on 3 and it all worked correctly
#asahi-dev · 2026-01-03 00:14
chadmedi havent noticed anything like that
#asahi-dev · 2026-01-03 00:13
chadmedi dont particularly care about native x11 anymore if its going to box us into a corner on properly driving this thing
#asahi-dev · 2026-01-03 00:12
chadmedand my understanding of ams may be entirely wrong but i think tihs might be necessary for async/nonblocking commits anyway
#asahi-dev · 2026-01-03 00:12
chadmedand im hoping that if we get the atomic sequencing more correct we can dispense with the vake vblank events entirely but given how long a swap takes im not particularly hopeful
#asahi-dev · 2026-01-03 00:11
chadmedi will wait until kwin is fixed so we can test that theory though of course
#asahi-dev · 2026-01-03 00:11
chadmedin such cases kwin will only ever use it as an underlay and put the primary scene on top of it, so it _shouldnt_ matter what shows through underneath the primary scene
#asahi-dev · 2026-01-03 00:09
chadmedsurface 0 should be safe if we keep it as an immutable primary, at least for kwin
#asahi-dev · 2026-01-03 00:08
chadmedsetting a "vrr" timing mode is not enough
#asahi-dev · 2026-01-03 00:08
chadmeddoesnt really matter right now though since i havent found the bits to flip to make dcp actually drive vrr anyawy
#asahi-dev · 2026-01-03 00:07
chadmedfair enough. i dont think vrr will ever work if we keep delivering fake/mistimed vblank events on "timed out" commits though
#asahi-dev · 2026-01-02 23:58
chadmedi had a clean(er) branch at dcp/mergable that i was going to clean up more tomorrow after i tested https://invent.kde.org/plasma/kwin/-/issues/282 but no stress if you already did
#asahi-dev · 2026-01-02 23:57
chadmednice
#asahi-dev · 2026-01-02 23:53
chadmedworks fine on my branch
#asahi-dev · 2026-01-02 23:53
chadmedwell, more than already was the case
#asahi-dev · 2026-01-02 23:53
chadmedi am in your debt
#asahi-dev · 2026-01-02 12:57
chadmedfirefox will do it now but not mpv
#asahi-dev · 2026-01-02 12:49
chadmedlol it stopped working again
#asahi-dev · 2026-01-02 12:38
chadmedive got my branch of random bs
#asahi-dev · 2026-01-02 12:37
chadmeditll be even sicker if we actually can scan out >2 surfaces and im not just being teased by kwin here :P
#asahi-dev · 2026-01-02 12:26
chadmedits fine for the primary plane
#asahi-dev · 2026-01-02 12:23
chadmedbut surface 0 is Very Special and dcp seems to not like it to have any transparency so idk if we can do anything about that
#asahi-dev · 2026-01-02 12:21
chadmedhaving surface 0 be the immutable primary also might break underlaying but im not game to test that until overlaying works
#asahi-dev · 2026-01-02 12:21
chadmedbut we then have three fungible overlays that kwin doesnt really know what to do with
#asahi-dev · 2026-01-02 12:21
chadmedso zpos=0 always goes on the bottom, which works fine since 0 is the primary plane
#asahi-dev · 2026-01-02 12:20
chadmediomfb_flush doesnt get a random order for the planes. the surfaces in the swap request are ordered bottom to top and we index into that based on the plane's normalised_zpos
#asahi-dev · 2026-01-02 12:19
chadmedand afaik all of this stuff in kwin was built mostly around DCN
#asahi-dev · 2026-01-02 12:18
chadmedthis probably works fine on amd where the overplay planes are severely limited and not fighting for the cursor too
#asahi-dev · 2026-01-02 12:17
chadmedbut if kwin is going to be stupid and try setting multiple planes to the same zpos then that is no good either
#asahi-dev · 2026-01-02 12:16
chadmedwhich wouldnt be an issue, we dont _need_ zpos to be immutable to satisfy iomfb since we map each plane's zpos to the correct surface order anyway
#asahi-dev · 2026-01-02 12:16
chadmedkwin really does not like immutable zpos it seems
#asahi-dev · 2026-01-02 12:14
chadmedit seems quite unreliable
#asahi-dev · 2026-01-02 12:14
chadmedthey just decided to start working
#asahi-dev · 2026-01-02 12:13
chadmednot quite sure tbh i havent changed anything
#asahi-dev · 2026-01-02 12:11
chadmedwell dcp is still displaying three rectangles i suppose so thats progress
#asahi-dev · 2026-01-02 12:10
chadmed /me installs weston

plus ›