Skip to content

Explain why external DisplayPort over USB-C does nothing on Apple Silicon - #289

Open
haripako wants to merge 7 commits into
omacom:quattrofrom
haripako:apple-silicon-dp-altmode-diagnostics
Open

haripako wants to merge 7 commits into
omacom:quattrofrom
haripako:apple-silicon-dp-altmode-diagnostics

Conversation

@haripako

Copy link
Copy Markdown

Plugging an external monitor into an Apple Silicon Mac and getting nothing is
expected on Asahi today, but nothing in Omarchy says so. This adds the
explanation and the two commands to check it. It does not enable the feature —
see the scope note at the end.

The problem

Apple Silicon has two display controllers: dcp for the internal panel and
dcpext for anything external. No t8103 device tree enables dcpext, not
even in asahi-wip-7.2, so there is no controller for an external display to
attach to. Asahi marks DisplayPort alt mode on M1 as work in progress.

Everything below that in the chain is already merged and works: the Type-C PHY
does DP mode, the display crossbar is wired, and the cd321x driver registers the
port altmodes, switches the mux and reads DisplayPort status.

What makes it expensive to diagnose is that every layer fails silently. No
error in dmesg, nothing in the compositor log, and each failure looks exactly
like the one before it from userspace. Two things also actively mislead:

  • A monitor with a USB hub enumerates a USB Billboard Device, which reads
    like a failed alt mode and says nothing either way.
  • With DisplayPort pin assignment C all four lanes carry DP and USB drops to
    2.0, so losing SuperSpeed is what success looks like, not a bad cable.

What this adds

  • omarchy debug dp-altmode — walks the chain in order, stops at the first
    thing missing, and says plainly that the stock state is expected rather than a
    fault. No root, reads only.
  • omarchy-hw-dp-altmode — hidden, for conditionals. Returns true only when
    a USB-C connector carries the displayport phandle and the node it names
    is enabled. Either half alone is a half-configured machine, and reporting
    success there sends someone hunting a cable fault that does not exist.
  • docs/apple-silicon-dp-altmode.md
  • test/shell.d/dp-altmode-test.sh — 7 checks against device tree fixtures.

Both resolve the port by i2c address, never by the portN index. That index
is assigned in i2c probe order and the same physical port was observed as
port1 on one boot and port0 on the next.

Scope: this deliberately does not enable the feature

Making external DisplayPort actually work on M1 needs a patched device tree plus
two out-of-tree kernel modules. A kernel update overwrites the modules and
leaves the DTB frozen at the old version, silently — the display just stops
working after an update, with no error. I do not think that is safe to ship
default-on in a distribution that updates kernels unattended, and it would want
a post-update guard that fails loudly first. Happy to follow up with that as a
separate PR if you want it.

Credit and status, for the record

  • A working implementation, the full evidence and the reasoning are at
    https://github.com/haripako/dp-altmode
  • The kernel-side HPD forwarding in that repo is not my work — it is Janne
    Grunau's and Sven Peter's from the fairydust branch, backported unchanged.
  • One original fix came out of this (the DCP latching ->crashed on a poweroff
    timeout, which permanently rejects every atomic commit). It is not fixed in
    fairydust either, and is being sent to AsahiLinux/linux separately.
  • One bug is unresolved: after unplug plus suspend, a phantom DCPAVSERV
    service blocks reconnection until reboot. It is not proven that suspend is the
    trigger. Written up in that repo.

Testing

test/shell.d/dp-altmode-test.sh passes (7/7). The full shell suite was run:
the only failures are the three that need the omarchy-pkgs sibling checkout
(config, preinstalls, unowned-system-paths), which fail the same way
without this change.

…icon

Plugging a monitor into an M1 and getting nothing is expected on Asahi today,
but nothing says so. No t8103 device tree enables dcpext, the second display
controller, so there is no controller for an external display to attach to.
Everything below that in the chain is already merged and works, and every layer
fails silently: no error in dmesg, nothing in the compositor log, and each
failure looks like the one before it from userspace.

Two things also mislead people here. A monitor with a USB hub enumerates a
Billboard device, which reads like a failed alt mode and says nothing either
way. And with DisplayPort pin assignment C all four lanes carry DP and USB drops
to 2.0, so losing SuperSpeed is what success looks like, not a bad cable.

Adds:

- omarchy-hw-dp-altmode, which returns true only when a USB-C connector carries
  the displayport phandle AND the node it names is enabled. Either half alone is
  a half-configured machine, and reporting success there sends someone hunting a
  cable fault that does not exist.
- omarchy debug dp-altmode, which walks the chain in order and stops at the
  first thing missing, saying plainly that the stock state is expected rather
  than a fault.
- docs/apple-silicon-dp-altmode.md.

Both resolve the port by i2c address, never by the portN index: that index is
assigned in i2c probe order and the same physical port was observed as port1 on
one boot and port0 on the next.

Enabling the feature is deliberately out of scope here. It needs a patched
device tree plus two out-of-tree modules, and a kernel update silently reverts
all of it, so it is not safe to turn on by default in a distribution that
updates kernels unattended.
@malik-na
malik-na requested a review from scottjones August 28, 2026 13:23
@malik-na

Copy link
Copy Markdown
Member

Hey @scottjones
Could you review and test this please? I don't have an external monitor at hand right now.

@scottjones scottjones left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The tooling here is better than the description suggests, and I want the doc — "why does nothing happen when I plug in a monitor" is a real support question. Four things to fix first, one of which sends readers somewhere I do not think we want to send them.

What I verified on an M2 Max (apple,j414c, t6021): omarchy-hw-dp-altmode does not look for a node called dcpext — it follows each USB-PD connector's displayport phandle to the controller it names and checks that node's status. That generalises past M1 nicely. This machine has no dcpext at all, just three dcp@ nodes (two okay, one disabled), and no connector exposes a displayport property, so the detector correctly returns 1. Its test passes and catches an inverted status check when I mutate the detector. omarchy commands --check is clean at 455.

1. The doc never mentions the fairydust branch. Asahi has implemented DP alt mode on the fairydust branch of their downstream tree — DCP, DPXBAR, ATCPHY and ACE, demoed at 39C3 with an M1 Air driving a display over USB-C. It is experimental (one blessed port per machine, cold/hot-plug quirks, "provided strictly as-is", no support until they deem it ready), but it is the canonical work and it is converging upstream.

Right now the "Enabling it" section says M1 needs "a patched device tree plus two out-of-tree kernel modules" and sends readers to github.com/haripako/dp-altmode. A reader comes away thinking the only route is a stranger's patches, when the project itself has a named branch. Please name fairydust, describe its limits honestly, and keep your own work as the additional reference rather than the primary one.

This also makes "no t8103 device tree enables dcpext, not even in asahi-wip-7.2" read as more settled than it is — true of the branches named, misleading about the project, since fairydust is where that changes.

2. The doc never mentions HDMI, and on many of these machines HDMI works. Zero occurrences in 288 lines. This MacBook Pro 14" M2 Max has a live HDMI connector:

card2-HDMI-A-1   disconnected
card2-eDP-1      connected

Someone on a 14"/16" machine reading "external display does nothing on Apple Silicon" would reasonably conclude they cannot drive an external monitor at all, with a working port beside them. One sentence putting HDMI outside the scope fixes it.

3. omarchy-debug-dp-altmode calls the detector by bare name, which fails when run from a checkout rather than an installed PATH:

bin/omarchy-debug-dp-altmode: line 32: omarchy-hw-dp-altmode: command not found
  External DP controller (dcpext): NOT enabled

It prints "NOT enabled" either way, so a missing command and a real negative look identical — the worst failure mode for a diagnostic. Worth resolving through $OMARCHY_PATH/bin or checking the command exists before reporting.

4. The explanation is t8103-specific but the title says "Apple Silicon". On t6021 the shape differs — three dcp nodes, no dcpext, no displayport property on any connector — while the outcome matches. Stating the outcome generally and keeping the t8103 detail as the M1 example would cover both without claiming more than is known.

**Point to Asahi's own work first.** The doc described the only route as a
patched device tree plus out-of-tree modules and sent readers to a third-party
repository, which reads as if the project had nothing. It does: `fairydust`
carries the device tree changes for t8103, t8112 and t60xx plus the `tipd`
change that forwards the hotplug event to DRM, and it is what converges
upstream. Named it, and took it on its own terms — its commits are titled
"dp-altmode dts hacks" and "HACK: Use drm oob hotplug event" by their own
authors, it blesses one specific port per machine, and it is experimental. The
third-party writeup is now explicitly an additional reference that backports the
fairydust change unmodified, not a substitute.

That also removes the "not even in asahi-wip-7.2" claim, which was true of the
branches named and misleading about the project.

**HDMI was never mentioned in 288 lines.** On the 14"/16" MacBook Pros and the
Mac mini it works and appears as its own connector, so "external display does
nothing on Apple Silicon" would tell someone with a working port beside them
that they cannot drive a monitor. Scoped to DisplayPort over USB-C in the doc,
in the command's header comment, and in its first line of output.

**A missing detector read exactly like a real negative.** The diagnostic called
`omarchy-hw-dp-altmode` by bare name, so from a checkout it printed
`command not found` and then "NOT enabled" — the worst failure mode a
diagnostic has. It now resolves the copy next to itself first, so a checkout
answers for itself, then `$OMARCHY_PATH/bin`, then PATH, and exits 1 with a
message if there is none. Two tests cover it.

**Generalised past t8103.** The outcome is stated generally, with M1 as the
worked example and the t6021 shape recorded beside it: no `dcpext` node at all,
several `dcp@` nodes, no connector carrying `displayport`. The output label no
longer says "dcpext", since on t6021 no such node exists.

Suite is 11 checks, up from 7. `omarchy commands --check` passes at 455.
@haripako

Copy link
Copy Markdown
Author

Thanks — all four are fixed in 273e939, and the fairydust one was the right call to push back on.

1. fairydust. You are right that the omission changed what the doc meant. The irony is that the linked repository credits fairydust throughout — its tipd change is the fairydust one, backported unmodified — but none of that reached this doc, so it read as if the project had nothing and a stranger had the only answer.

Named now, and described on its own terms rather than as a recommendation: the device tree changes for t8103, t8112 and t60xx plus the tipd hotplug forwarding, the commits titled "dp-altmode dts hacks" and "HACK: Use drm oob hotplug event" by their own authors, one blessed port per machine, experimental and as-is. The M1 writeup is now explicitly an additional reference that backports the fairydust change unmodified.

That also took out "not even in asahi-wip-7.2", which as you say was true of the branches named and misleading about the project.

One thing your review gave me: the t8103 commit names the blessed port per machine, so the doc now lists them — front-left USB-C on the 13" Pro and Air, next to HDMI on the mini, back-left on the 2-port iMac, back-right-middle on the 4-port. More useful than "one port per machine".

2. HDMI. Fair, and the framing was the problem more than the omission. Scoped in three places: the doc's second paragraph, the command's header comment, and its first line of output, which now says HDMI is a separate path not covered here.

3. The bare-name call. That one was a real defect and your diagnosis of why it was the worst kind was exact — a missing command and a real negative were indistinguishable. It now resolves the copy next to the running script first, so a checkout answers for itself, then $OMARCHY_PATH/bin, then PATH, and exits 1 with a message if there is none. Two tests: one asserts a missing detector exits non-zero and never prints "NOT enabled", the other that a checkout run with PATH=/usr/bin:/bin and OMARCHY_PATH=/nonexistent still reports correctly.

4. Generalising past t8103. The outcome is stated generally now, with M1 as the worked example and your t6021 findings recorded beside it — no dcpext node at all, several dcp@ nodes, no connector carrying displayport. The output label no longer says "dcpext" either, since on t6021 there is no such node to name.

Suite is 11 checks, up from 7. omarchy commands --check passes at 455. Verified on the M1 this was written against; the t6021 path is covered by fixtures rather than hardware, so a second look on your machine would be worth having.

@scottjones scottjones left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the rework — the three things I raised last time are all properly handled now: fairydust is named as Asahi's own branch, HDMI is explicitly scoped out, and the detector lookup no longer reports a missing binary as a negative result. Each has a test. The doc itself is the valuable part of this PR, and the "every layer fails silently" framing is exactly right about why this is so expensive to diagnose.

I verified the rest on an M2 Max (apple,j414c, t6021) running Asahi, and found two things that need fixing before this lands.

1. omarchy-hw-dp-altmode cannot return 0 on any Apple Silicon Mac

The detector requires a displayport phandle property on a USB-C connector node. I grepped every Apple DTB shipped with this machine's kernel (7.1.6):

20 Apple DTBs (t8103, t6000/1/2, t6020/21/22, t8112)
  displayport property:   0 of 20
  usb-c-connector:       20 of 20
  dcpext:                20 of 20

Zero — including t8103-j293.dtb, the exact machine the test fixture models. The property name does not appear anywhere in any of them.

On real hardware the connectors use the standard OF graph binding instead:

soc/i2c@39b040000/usb-pd@38/connector   label = "USB-C Left Rear"
  ports/port@0/endpoint -> soc/usb@702280000/ports/port@0/endpoint
  ports/port@1/endpoint -> soc/phy@703000000/ports/port@0/endpoint

USB controller and Type-C PHY. No DisplayPort endpoint, no displayport property.

So the positive branch is unreachable and the command returns 1 unconditionally, forever, on every Apple Silicon machine. I confirmed it by building a fixture that mirrors this machine's real DTB shape with a dcpext node present and status = "okay" — the state where the route is enabled — and the detector still returned 1.

The suite passes because make_tree constructs the shape the detector assumes rather than the shape the kernel actually exposes. Worth pointing the fixture at a real DTB layout; the mismatch is invisible otherwise.

2. The t6021 claim in the doc is inverted

The doc says:

M2 Pro/Max (t6021) — there is no dcpext node at all, just several dcp@ nodes

On this t6021 there are three display controllers, and two of them are dcpext:

dcp@289c00000   apple,t6020-dcpext apple,dcpext   status=okay       <- enabled
dcp@315c00000   apple,t6020-dcpext apple,dcpext   status=disabled
dcp@389c00000   apple,t6020-dcp    apple,dcp      status=okay       <- internal panel

dcpext exists and one instance is enabled. What's true is that the node names are all dcp@ — "dcpext-ness" lives in the compatible string. That also means the soc/dcpext*@* glob matches nothing on real hardware, while soc/dcp*@* matches the internal panel controller, so the external-vs-internal half of the check isn't actually being made either.

This one goes beyond the detector: if a dcpext is enabled on this machine and external DP over USB-C still does nothing, then "the second controller is left disabled" isn't the explanation the doc is built around. Worth re-checking that on t8103 too before settling the wording — the compatible string is the thing to read, not the node name.

Suggestion

The honest signal available on every machine is the compatible string plus status, so keying on apple,*dcpext would at least be measuring something real. But given an enabled dcpext here still doesn't produce a working external display, I'm not sure the DTB can answer "does DP alt-mode work" on its own — in which case a detector that always says no is worse than no detector, because it reads as a definitive answer. It would be reasonable to land the doc and the diagnostic on their own and drop omarchy-hw-dp-altmode, or have it report an explicit "cannot determine" state.

Happy to re-test whatever you land on — this machine has the hardware and I still have the DTB comparison set up.

@haripako

haripako commented Sep 2, 2026

Copy link
Copy Markdown
Author

Thanks for going back to the hardware on this one. Points 2 and the glob observation are right and I'm fixing both. On the central claim I disagree, and I think we've found the thing neither of us had stated: you are testing on a stock DTB and I am testing on a patched one. That single fact explains the whole disagreement.

1. The positive branch is reachable — it is taken on this machine

Your grep is correct and I reproduced it: of the 23 Apple Silicon DTBs shipped with 7.1.6 here (t8103, t600x, t602x, t8112), zero carry a displayport property, t8103-j293.dtb included. No argument there.

But that is the stock tree. This M1 (apple,j293, t8103) boots the fairydust device tree change, and its live tree looks like this:

dcp@231c00000   apple,t8103-dcp      status=okay   phandle=0x26   <- internal panel
dcp@271c00000   apple,t8103-dcpext   status=okay   phandle=0x45   <- external
connector "USB-C Left-front"   displayport -> 0x45

$ omarchy-hw-dp-altmode; echo $?
0

Comparing the two DTBs for this exact machine side by side:

dcpext status displayport on a connector detector
stock t8103-j293.dtb disabled absent 1
patched (fairydust DT change) okay present, "USB-C Left-front" 0

So the command does not return 1 unconditionally and forever. It returns 1 on a stock tree — which is the correct answer, because DP alt mode genuinely does not work there — and 0 once the route is actually wired up. That is the distinction the check exists to make, and why it reads the live tree rather than the model.

On why your fixture still returned 1 with dcpext present and status = "okay": the check deliberately requires both halves, the connector's displayport phandle and an enabled target. A fixture with the enabled node but no phandle is the half-configured state, and the suite has a case asserting exactly that it must not count — the comment there reads "either one alone is the state a machine sits in halfway through enabling this, and reporting success there would send someone hunting for a cable fault that does not exist." Your fixture supplied one half.

The suite also already has a stock-state case asserting 1, so the fixtures aren't only modelling the shape that succeeds. Your underlying point still lands though: they are hand-built rather than derived from a real DTB, and I'll re-point them at the real stock and patched shapes so the difference is visible in the test rather than in this thread.

I should have said in the PR that this machine runs a patched tree. Without that, your grep and my results look like a contradiction instead of the two ends of the same check.

2. The dcpext*@* glob is dead code — confirmed on t8103 too

You're right, and it is worse than a t6021 quirk. On this M1 the external controller node is named dcp@271c00000, not dcpext@...:

dcp@271c00000   compatible = "apple,t8103-dcpext", "apple,dcpext"

So soc/dcpext*@* matches nothing here either, and soc/dcp*@* matches both controllers, which means the external-vs-internal distinction isn't coming from the glob at all — it comes from the phandle, which points at the right node. The glob is doing nothing. Keying on the compatible string as you suggest is the correct read, and I'll change it to that.

3. The t6021 wording

Fixing it with your new data. For the record that sentence came verbatim from your first review — I took it at face value and should have marked it as unverified rather than stating it flatly, since I have no t6021 to check it against. Your correction is the better information and it goes in as such: the node names are all dcp@, dcpext-ness lives in the compatible string, and on t6021 there are three controllers of which two are dcpext and one of those is enabled.

4. Whether "the second controller is left disabled" survives

On t6021, your finding stands and I'm not going to claim otherwise from here. On t8103 it is verified: the stock t8103-j293.dtb ships dcp@271c00000 with status = "disabled", and the patched one flips it to okay alongside adding the connector phandle. The doc states the outcome generally with M1 as the worked example, so I'll keep the mechanism attached to t8103 where it's measured and let your t6021 observation stand beside it as the open question it is.

That an enabled dcpext on t6021 still produces nothing is genuinely interesting and I don't have an explanation for it. It may be that the connector never routes to it without the OF-graph endpoint the fairydust series adds.

5. On dropping the detector

I'd rather not. The argument for dropping it is that a detector which always says no is worse than none, and that holds only if it always says no — which is what I don't think is the case. It separates "stock tree, route not wired up" from "route wired up", and that is the question someone plugging in a monitor actually has.

Its summary line already promises only that the route is wired up, not that external display works — deliberately, because on this machine with the route enabled, cold plug brings up a display and hot plug still does not. Reporting "wired up" and being wrong about hot plug would be a different bug; reporting "not enabled" on a stock tree is simply true.

If you'd still prefer a third state I can add one, but I'd want it to mean "this is not an Apple Silicon machine / the tree is unreadable" rather than "cannot determine", since on both trees I can check the answer is determinate.

What I'll push

  • key the controller check on apple,*dcpext compatible rather than the node-name glob
  • re-point the fixtures at real stock and patched DTB shapes
  • fix the t6021 paragraph with your data
  • state in the doc that this machine runs a patched tree, and that on a stock tree the expected result is 1

If you want to reproduce the positive path on the M2 Max, the fairydust DT change for t60xx adds the same two things — connector displayport phandle and dcpext enabled. Happy to hand you the exact before/after node dump from this machine if it helps as a reference shape.

haripako added 2 commits September 2, 2026 23:55
The external display controller is named dcp@ like the internal one on
every machine: t8103 exposes it as dcp@271c00000 with compatible
"apple,t8103-dcpext", and t6020 does the same. The "dcpext@" glob
therefore matched nothing anywhere, and "dcp@" matched both controllers,
so the external-vs-internal distinction was resting entirely on the
phandle rather than on anything the check itself asserted.

Read the compatible string instead, which is where dcpext-ness actually
lives, and drop the dead glob.

Reshape the fixture after the real t8103-j293 tree as the kernel exposes
it: both controllers present, named dcp@, carrying their compatible
strings, with the internal one enabled as it always is. The previous
fixture had only the external node, so a check keyed on the node name
could not be caught failing.

Add a case for a connector whose phandle names the internal controller,
which the node-name glob accepted and the compatible check rejects.

Verified on an M1 (apple,j293, t8103): the detector still exits 0 against
the live patched tree, and the suite is 12 checks.
…d one

The t6021 paragraph was inverted: that machine does have dcpext, two of
its three display controllers are apple,t6020-dcpext, and one of those is
enabled. Corrected from the review, with the consequence stated -- an
enabled external controller there still yields no external display, so a
disabled controller is not the whole explanation on t6021. What both
shapes share is that no connector is wired to the controller at all.

Say plainly that no machine names the node dcpext, since that is what
makes a name-based check look reasonable and be wrong.

Record what the check answers on each kind of tree. On a stock tree it
reports the route as not enabled, which is correct and is where nearly
every machine sits; it reports enabled only once a tree wires a connector
to an enabled external controller. The M1 this was written against runs
such a tree, which is why it can be seen returning both answers. Without
that stated, a stock-tree result and a patched-tree result read as a
contradiction rather than as the check working.

Lead the explanation with the missing connector wiring rather than the
disabled controller, since only the former holds on both SoCs.

Suite is 13 checks.
@haripako

haripako commented Sep 2, 2026

Copy link
Copy Markdown
Author

Pushed as ab8b581e and dab98045. All four are in.

Compatible string instead of the node-name glob. The dcpext@ glob is gone. The check now confirms the node the phandle names is an external controller by reading its compatible, which is the only place that distinction exists on either SoC.

Fixtures reshaped after the real tree. Both controllers are now present and both are named dcp@, carrying their compatible strings, with the internal one enabled as it always is. The old fixture had only the external node, so a name-based check could not be caught failing against it. New case: a connector whose phandle names the internal controller, which the old glob accepted and the compatible check rejects — removing the compatible line fails that case and nothing else.

Worth noting the fixture already used dcp@271c00000 rather than dcpext@..., so the node naming was right; what was missing was the second controller and the compatible strings, which is what left the gap you found.

t6021 corrected, including the consequence you drew: an enabled external controller there still produces no external display, so a disabled controller is not the whole explanation on t6021. The doc now leads with the connector wiring, since that is the part that holds on both shapes, and says plainly that no machine names the node dcpext.

Stock versus patched is now stated. The doc records that on a stock tree the check reports the route as not enabled — correct, and where nearly every machine sits — and that it reports enabled only once a tree wires a connector to an enabled external controller, which the M1 this was written against does. Two greps assert it stays there.

Suite is 13 checks, omarchy commands --check passes at 455, and the detector still exits 0 against the live tree here.

If you re-test on the j414c, a stock tree should give exit 1 from the connector half — no displayport property on any connector — rather than from the controller half, which is the opposite of where it failed before. The offer of the before/after node dump from this machine stands if a reference shape is useful for constructing the positive case there.

@adamkstinson

Copy link
Copy Markdown

Hardware note from a stock Omarchy Mac install, if a t8112 fixture would help.

Machine: Apple MacBook Air 13" M2 2022 (apple,j413 / t8112), Omarchy 4.0.2, linux-asahi 7.1.6.asahi1-1. Stock DTB. hyprctl monitors only ever shows eDP-1 (2560×1664). DRM: card2-eDP-1 connected; no DP-*.

Live tree matches the stock t8112-j413.dtb in this kernel:

dcp@231c00000   apple,t8112-dcp      status=okay       phandle=0x36   <- internal
dcp@271c00000   apple,t8112-dcpext   status=disabled   phandle=0xa7   <- external
phy@23c500000   apple,t8112-dptx-phy status=disabled

USB-C connectors (no displayport property on either, confirmed on the live tree and by walking the shipped t8112-*.dtbs):

i2c label sysfs this boot
usb-pd@38 USB-C Left-back port1 (0-0038)
usb-pd@3f USB-C Left-front port2 (0-003f)
usb-pd@3a (no connector node — MagSafe) port0 (0-003a)

So omarchy-hw-dp-altmode would exit 1 on the connector half first (zero displayport phandles), and the controller half is also disabled. Same shape as the stock t8103 table in the doc.

t8112 stock DTB comparison (this kernel, no displayport on any of them):

DTB machine dcpext dptx-phy
t8112-j413 Air 13" M2 (this box) disabled disabled
t8112-j415 Air 15" M2 disabled disabled
t8112-j493 13" Pro M2 disabled disabled
t8112-j473 M2 Mini okay okay

The Mini line is the useful contrast: dcpext + dptx-phy are on there (HDMI), but USB-C still has no displayport phandle. The dual-half check is right — an enabled dcpext alone must not count as USB-C DP. Worth a sentence in the doc next to the t8103 / t6021 cases, because t8112 currently isn’t named.

Two extra silent-failure gotchas on this box, if you want them in the “looks like a fault and is not” section:

  1. Cheap “DisplayPort adapters” are often DisplayLink, not DP-alt-mode. This one was 17e9:4307 Subosen_3900_Docking. USB hub/Ethernet/audio enumerate; video never becomes a DRM connector. No evdi. Looks exactly like the Asahi gap until you read idVendor.
  2. Layers under the missing route still load and stay quiet: mux_apple_display_crossbar logs Switched dpphy to disconnected state on both muxes; plugging the DisplayLink dock produced phy-apple-atc … Failed to lock pipehandler. Easy to chase. Not the missing displayport phandle.

Hyprland is already preferred / auto on all outputs. Nothing in monitors.lua will create a connector. Real enablement is still fairydust DT + tipd HPD, out of scope here.

Happy to re-run omarchy debug dp-altmode on this Air once the command lands.

malik-na and others added 2 commits September 3, 2026 14:31
Reported from a stock M2 Air (apple,j413) and confirmed here against the
shipped DTBs. t8112 was the shape the doc did not name, and it carries
the clearest evidence for testing both halves rather than one: the M2
mini (j473) ships apple,t8112-dcpext and its dptx-phy both enabled, and
still has no displayport property on any USB-C connector. A check keyed
on an enabled controller alone would report that machine as working.

Record the likely reason an enabled controller shows up at all -- it
appears to serve HDMI, and the machines that have it enabled are the ones
with an HDMI port -- with the caveat that t8103 does not fit cleanly: the
M1 mini has HDMI, ships dcpext disabled, and no t8103 tree carries the
dptx-phy binding at all.

Add the DisplayLink case to the section of things that look like faults.
Adapters sold as DisplayPort are frequently DisplayLink USB graphics,
which fail in exactly the shape this doc describes for an unrelated
reason; the vendor id 17e9 is what tells them apart. Add the kernel
messages from the layers that do work, which read like faults to someone
already hunting for one.

Suite is 15 checks.
@haripako

haripako commented Sep 3, 2026

Copy link
Copy Markdown
Author

@adamkstinson This is exactly the data the PR was missing — thank you. All of it is in as of dceac87a, and I reproduced your table against the DTBs shipped here before writing any of it down.

t8112 is now the third named shape, and j473 carries the argument. A stock machine with apple,t8112-dcpext and its dptx-phy both enabled, and still no displayport property on any USB-C connector, is the clearest evidence in this PR that both halves have to be tested rather than one. A check keyed on an enabled controller alone would report that mini as working. That is a better case for the dual-half design than anything I had, because it comes from an unmodified machine rather than a patched one.

The DisplayLink case is in the section of things that look like faults, with 17e9 named, since the vendor id is the thing that separates it from the gap this doc is about. The kernel messages from the layers that do work went in beside it — mux_apple_display_crossbar reporting Switched dpphy to disconnected state and phy-apple-atc … Failed to lock pipehandler read like faults to someone already hunting for one.

Your reading of what the detector does on your Air is right: it exits 1 on the connector half first, before the controller half is ever reached. The command is in this branch now if you want that re-run.


@scottjones Your open question from the last review — an enabled dcpext on t6021 that still produces nothing — I think this answers, or at least points at the answer.

The enabled controller appears to serve HDMI, not USB-C. On t8112 the only machine of the four with dcpext and dptx-phy enabled is the mini, which is the one with an HDMI port; the three laptops ship both disabled. Your j414c has a live card2-HDMI-A-1, which fits. On that reading the t6021 dcpext is doing its job — it is just not the controller USB-C would need, and nothing wires a connector to one that is.

One caveat, because it does not generalise cleanly. t8103 does not fit: the M1 mini (j274) has HDMI and ships dcpext disabled, and no t8103 device tree carries the dptx-phy binding at all — I checked all five here. So the pattern holds on t8112 and is consistent with t6021, while t8103 looks like it simply predates it rather than contradicting it. The doc states it with that reservation rather than as a conclusion.

If it holds, it also removes the reason to drop the detector. The j473 case is a stock machine where the controller half is satisfied and the answer is still correctly no — so the check is not merely always saying no, it is distinguishing states that genuinely differ, on hardware nobody has patched.

Suite is 15 checks. omarchy commands --check passes at 458, up from 455 because quattro was merged into the branch since the last run.

@adamkstinson

Copy link
Copy Markdown

Follow-up from the same apple,j413 / t8112 Air (13" M2 2022, Omarchy 4.0.2, linux-asahi 7.1.6.asahi1-1): the fairydust route is now wired up on this machine, and USB-C DP alt-mode works. External display is a Dell U2422H at 1920×1080@60, driven from the front-left USB-C port (HDMI adapter / hub that actually does DP alt-mode — not DisplayLink).

This is the patched-tree counterpart of the stock dump I posted earlier. I have not seen a published t8112 field report before, so here is the exact shape and the four commits.

What was applied

Patches pulled verbatim from AsahiLinux/linux fairydust, Janne Grunau. Applied against tag asahi-7.1.6-1 (matches this kernel). Neither piece replaces the kernel package.

# commit what
1 f216a6e787f89b7545e28494f624e0ce997900ea arm64: dts: apple: t8112: Add dp-altmode dts hacks — #define ENABLE_DCPEXT_TYPEC on t8112-j413.dts / j415 / j493 / j473, plus the t8112-jxxx.dtsi block that enables dcpext, sets apple,connector-type = "DP", and puts displayport = <&dcpext> on the USB-C connector
2 1069f56d6225fbd16cea99d5e9988163465f207b HACK: arm64: dts: apple: t8112: Mark ps_atc1_common as always on — ATC phy suspend/resume workaround
3 9b65351d9cf850ec8273f138f04480dc1e93f15a usb: typec: tipd: Track data_status changes for CD321x — debounce HPD from data_status
4 d02d57a6c830ed5f2934490aa6bb35df0db4d376 usb: typec: tipd: HACK: Use drm oob hotplug event — drm_connector_oob_hotplug_event() on the connector's displayport phandle

Install method, same architecture as haripako's t8103 repo:

  1. DTB built from the patched t8112-j413.dts, injected via update-m1n1 DTBS= override (/usr/local/lib/dpalt-dtbs/*.dtb → new boot.bin). Kernel image untouched. Revert is restore boot.bin / drop the DTBS= line and rerun update-m1n1.
  2. tipd (tps6598x-core.ko) rebuilt out-of-tree from the patched drivers/usb/typec/tipd/{core.c,tps6598x.h} against this kernel's headers. Live insmod first; persist under /lib/modules/$(uname -r)/updates/ only after a picture appeared.

Both halves are required. DTB alone produces card2-DP-1 (disconnected) and a booted dcpext; without the tipd HPD forward, a real DP sink still never lights. tipd alone has nothing to point oob_hotplug at.

Live tree after the DT change (vs the stock dump)

dcp@231c00000   apple,t8112-dcp      status=okay       <- internal eDP
dcp@271c00000   apple,t8112-dcpext   status=okay       <- external DP
aliases: dcp -> /soc/dcp@231c00000
         dcpext -> /soc/dcp@271c00000
i2c label displayport phandle
usb-pd@38 USB-C Left-back absent (USB only)
usb-pd@3f USB-C Left-front present → dcpext (dcp@271c00000)
usb-pd@3a MagSafe no connector node

That is exactly the dual-half signal the detector is written for: connector phandle and enabled apple,*dcpext. On a stock j413 both fail; after this, both pass. omarchy-hw-dp-altmode is not on PATH on this install yet, but the live tree is the positive t8112 fixture.

DRM / Hyprland with the monitor on front-left:

card2-DP-1    connected enabled    1920x1080 (and the rest of the U2422H list)
card2-eDP-1   connected enabled    2560x1664

Monitor DP-1: Dell Inc. DELL U2422H 9GW62K3  1920x1080@60
Monitor eDP-1: 2560x1664@60

Kernel path that actually lights it (tipd HPD → apple-dcp):

#### oob_hotplug status:0x1 ####
apple-dcp 271c00000.dcp: dcp_dptx_connect(port=0)
… display HPD asserted
dcp_hotplug() connected:1 … nr_modes:20
set_digital_out_mode … "1920x1080": 60 148500 …
mode_set_gated: 1920x1080@60 Hz link: 1

oob_hotplug status:0x2 + dcp_dptx_disconnect is the non-sink / unplug case.

What is not this patch (still true, and easy to misread)

  • Front-left USB-C only. Left-back enumerates USB just fine and even shows DP SVID ff01 active; it is not wired to dcpext, so DRM never gets a connector for it.
  • DisplayLink is still not DP alt-mode. Same 17e9:4307 Subosen_3900_Docking as before: USB hub/Ethernet/audio + VIA billboard, oob_hotplug stays 0x2, DP-1 stays disconnected. A Realtek USB-C hub with no HDMI sink attached did the same. A real DP-alt-mode HDMI/DP path on that front-left port is what produced 0x1 and the Dell.
  • Unplug can latch dcpext. Same unresolved fairydust/M1 note: after a successful link, unplug has been reported to leave the display coprocessor wedged until reboot. Not fully characterised here; first bring-up survived.

Happy to dump the patched t8112-j413.dtb node-for-node against stock if you want it as a fixture next to the t8103 pair. The two DTS hunks + two tipd hunks above are the whole change.

@adamkstinson

Copy link
Copy Markdown

One more t8112 gotcha, now that the patched tipd is persisted on this Air.

depmod + /lib/modules/$(uname -r)/updates/tps6598x-core.ko is not enough. Asahi’s initramfs (mkinitcpio asahi + kms hooks) ships stock

usr/lib/modules/…/kernel/drivers/usb/typec/tipd/tps6598x-core.ko

That loads before rootfs. The updates/ copy never replaces it. Symptoms after reboot, even with the fairydust DTB live (DP-1 exists):

  • modinfo tps6598x-core shows updates/… (first file on disk)
  • the running module does not taint, and there is no oob_hotplug in dmesg
  • card2-DP-1 stays disconnected with a real DP-alt-mode sink on Left-front
  • a live rmmod/insmod of the patched core (after login) immediately makes the next plug work

Fix: after depmod, rebuild initramfs so the image contains the updates module and not the stock core:

sudo mkinitcpio -P
lsinitcpio /boot/initramfs-linux-asahi.img | grep tps6598x
# want:
#   …/updates/tps6598x-core.ko
#   …/tipd/tps6598x.ko          # i2c glue, stock, fine
# must not have:
#   …/kernel/drivers/usb/typec/tipd/tps6598x-core.ko

Confirmed on this box after mkinitcpio -P: only updates/tps6598x-core.ko + the glue. Worth a sentence next to any “out-of-tree tipd” instructions, because the README instinct (“USB-C display doesn’t need initramfs”) is wrong on this distro — kms pulls tipd in.

linux-asahi updates will need the module rebuilt and another mkinitcpio -P. Unplug latch (#336) is separate and still reboot-only.

@scottjones scottjones left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Withdrawing my central objection from the last review. I said omarchy-hw-dp-altmode cannot return 0 on any Apple Silicon Mac. You answered that correctly on Sept 2 and I should have replied then rather than leaving a CHANGES_REQUESTED standing over it for three days.

You were right about the cause: I was reading a stock DTB and you were running a patched one, and that single fact accounts for the whole disagreement. My fixture supplied the enabled controller without the connector phandle, which is the half-configured state the check is explicitly written to reject — the suite already had a case asserting exactly that, and I read it as a bug instead of as the design.

What settles it beyond your machine is @adamkstinson's j473: a stock M2 mini shipping apple,t8112-dcpext and its dptx-phy enabled, with no displayport property on any USB-C connector. A controller-only check reports that machine as working. That is the dual-half design justified on hardware nobody has patched, and it is a better argument than anything in my review. His two field reports are the most useful thing in this thread.

t6021 hardware session

You asked twice for this shape and nobody had it. Ran it on the M2 Max (apple,j414c, t6021), stock tree, with an external display physically plugged into the front-left USB-C port:

port2  =  i2c 0-003f  =  usb-pd@3f  =  "USB-C Left Front"    <- cable here
usb-pd@38  USB-C Left Rear    displayport: absent
usb-pd@3b  USB-C Right        displayport: absent
usb-pd@3f  USB-C Left Front   displayport: absent

dcp@289c00000  apple,t6020-dcpext,apple,dcpext   status=okay
dcp@315c00000  apple,t6020-dcpext,apple,dcpext   status=disabled
dcp@389c00000  apple,t6020-dcp,apple,dcp         status=okay

omarchy-hw-dp-altmode -> 1

The partner attaches and advertises zero DP alt-modes, no DRM connector is created, and the only DCP activity in the log is 389c00000.dcp — the internal controller. No oob_hotplug, no dcp_dptx_connect. The detector's answer is correct, and it fails on the connector half first, exactly as you predicted for a stock tree.

Your HDMI reading is consistent here too: this machine has a live card2-HDMI-A-1 alongside the enabled dcpext, which fits "the enabled external controller serves HDMI, and nothing wires a USB-C connector to one." Not proof, but it is the t60xx data point behind the reservation you stated.

Also verified: router integration (omarchy debug dp-altmode and omarchy hw dp-altmode both resolve and return correctly, 1 and 0 respectively), all 15 tests pass, and omarchy commands --check passes at 458 — matching your number exactly.

The one blocker

test/all fails on a single assertion, not ok - bin commands use command helpers. It is bin/omarchy-debug-dp-altmode:27:

[[ -n ${hw_dp_altmode:-} ]] || hw_dp_altmode=$(command -v omarchy-hw-dp-altmode || true)

test/shell.d/bin-style-test.sh:7 forbids raw command -v in bin/ outside a small allowlist.

Deleting that line fixes it, and I verified the deletion rather than guessing:

bin-style-test.sh    after deleting line 27  -> PASS
dp-altmode-test.sh   after deleting line 27  -> PASS   (not load-bearing)

Lines 22-26 already try $SELF_BIN/omarchy-hw-dp-altmode then $OMARCHY_PATH/bin/..., which cover the checkout case and the installed case, and still exit 1 with a message when neither resolves — so the loud-failure behaviour you added for my earlier review survives intact.

Please delete the line rather than widening the allowlist. #343 widens that same single-line regex (mac-(setup|asahi-install)$), so an allowlist entry here would collide with it on one line — the worst kind of merge conflict. Deleting also matches AGENTS.md, which says commands from the default package set are runtime invariants to be invoked directly; the detector ships in this same PR.

For completeness, the full suite locally is 3 of 249 files failing: this one plus config-test.sh and unowned-system-paths-test.sh. Both of the latter pass once OMARCHY_PKGS_PATH is set, which is why CI reports one failure and not three. One run reported "4 of 249" with the same three assertions and a clean rerun gave 3, so there may be a flaky file somewhere — I only saw it once and can't pin it down; mentioning it as an observation, not a claim against this branch.

Two suggestions

Move the display-connectors block above the stock-tree exit 0. Right now the port table, the cable position and the connector list all live after that early exit, so on a stock tree — which is nearly every machine, and the entire audience for this doc — none of it prints.

I originally talked myself out of this on the grounds that every port row reads DisplayPort: no on a stock tree, so the table is not actionable. Then someone with a monitor plugged in read the output and asked "so it didn't detect that I'm plugged in?" That is the gap: the tool knows a cable is in 0-003f, knows no port has DisplayPort wired, and knows no new DRM connector appeared, and it reports none of it. The connector list is the piece that matters most, because it is what separates @adamkstinson's DisplayLink dock from the alt-mode gap — a dock enumerates USB and creates no connector, which looks identical to this from userspace. Even one line naming the port the cable is in would close it.

@adamkstinson's mkinitcpio -P finding is not in the doc. He posted it 2026-09-04 00:41, after your last push at 09-03 18:05. Asahi's initramfs ships the stock tps6598x-core.ko, which shadows the updates/ copy, so a patched tipd silently does nothing after reboot while DP-1 still exists. That is precisely the class of silent failure this document is for, and it will bite anyone following the enablement path.

Delete line 27 and I will approve immediately. Sorry again for sitting on the last review — the disagreement was resolved by your Sept 2 comment and I left it blocking anyway.

@edulmartins

Copy link
Copy Markdown

Hardware validation (stock Asahi) — MacBook Air M1 j313, with USB-C↔USB-C monitor plugged

  • Machine: MacBook Air (M1, 2020) — MacBookAir10,1 / apple,j313 / apple,t8103
  • Kernel: linux-asahi 7.1.6.asahi1-1 (7.1.6-1-1-ARCH)
  • Omarchy: 4.0.3rc1-2
  • Mesa: 26.1.8-1
  • Cable: USB-C to USB-C (native DP alt-mode path, not DisplayLink)
  • Tested from PR head 03533009 (worktree), not from quattro

With monitor plugged

  • Hyprland / DRM: only eDP-1 (internal). No external connector appears; monitor brand/model cannot be read because nothing attaches to DRM.
  • Type-C: port0 shows a partner (data_role=[host], power_role=[source]), so the plug is recognized.
  • On plug, kernel logged a WARNING in atcphy_mux_set (phy_apple_atc) via typec_mux_set / cd321x_update_work — consistent with alt-mode mux activity without a wired external DP path.
$ omarchy-hw-dp-altmode; echo $?
1

$ omarchy debug dp-altmode
External DisplayPort over USB-C
(HDMI, where present, is a separate path and is not covered here.)

  Machine: Apple MacBook Air (M1, 2020)
  DisplayPort-capable display controller: NOT enabled

  This is the expected state on a stock Asahi kernel, and it is why
  nothing appears when you plug a monitor into USB-C. No connector
  points at an enabled display controller, so the plug event has
  nowhere to go. Nothing is broken on your machine and there is no
  setting to change.

  Asahi implements this on their downstream 'fairydust' branch, which
  is experimental and blesses one specific port per machine.

  See docs/apple-silicon-dp-altmode.md.

Live device-tree (unchanged by the plug):

  • dcp@231c00000 → apple,t8103-dcp / status okay (internal)
  • dcp@271c00000 → apple,t8103-dcpext / status disabled
  • usb-pd@38 / usb-pd@3f connectors have no displayport property

So on real j313 stock: detector exit 1, debug message matches reality (plug USB-C monitor → nothing in Hyprland), no false positive. Did not test fairydust/patched DTB.

Happy to re-run after the command -v CI fix lands.

@malik-na

malik-na commented Sep 9, 2026

Copy link
Copy Markdown
Member

The remaining blocker is mechanical and CI is showing it: remove the raw command -v fallback in omarchy-debug-dp-altmode. The two explicit detector paths already cover checkout and installed use.

@theashgen

Copy link
Copy Markdown

Can confirm this also reproduces on MacBook Air M2 (t8112), not just t8103 — so the dcpext gap isn't M1-specific.

Tested on Asahi/Omarchy (ALARM base), booted into the fairydust branch (kernel 7.1.13-fairydust+) specifically to try to get this working — no external display detected over USB-C.

dmesg | grep -i dcp after plugging in the monitor shows only a single apple-dcp instance, which resolves to 2560x1664 (the internal panel's native resolution) with is_main_display: 1. No second dcpext platform device binds at all — matches this PR's description that the device tree has no dcpext node to attach an external display to, rather than a failed mode-negotiation.

Ended up fully removing the fairydust build afterward since it didn't move the needle for this model. Happy to test omarchy debug dp-altmode against this hardware once it lands, or run the haripako/dp-altmode out-of-tree patch if useful data points from an Air (vs. HDMI-equipped Pro/Max/Mini) would help.

@Natetgmaxwell

Copy link
Copy Markdown

Strong support for documenting this — the silent-failure point is exactly right. One premise has gone stale since it was written, though, and it changes what the doc should say.

The feature is reachable today

No t8103 device tree enables dcpext, not even in asahi-wip-7.2, so there is no controller for an external display to attach to. Asahi marks DisplayPort alt mode on M1 as work in progress.

True of the release branches, but Asahi's fairydust branch enables it, and it works. Verified today on t8103/j293 (M1 MacBook Pro 13", Omarchy on Arch ARM), driving a Dell U4323QE at 3840x2160@60 over one USB-C cable:

apple-dcp 271c00000.dcp: display HPD asserted
apple-dcp 271c00000.dcp: dcp_hotplug() connected:1 valid_mode:0 nr_modes:31
apple-dcp 271c00000.dcp: set_digital_out_mode(color:90 timing:52) "3840x2160": 60 533250 ...

hyprctl monitors then shows a normal second output. The t8103 DT hack covers all M1 machines (commit 7eb562930657), and there's an equivalent for t8112 and t600x.

It does not require shipping a kernel fork

This seems worth folding into the scope note, because it's a much lighter lift than it looks. Against the packaged linux-asahi 7.1.13.asahi3-1, the entire fairydust delta relevant to this is:

what size
drivers/usb/typec/tipd/{core.c,tps6598x.h} +23 lines — CD321x tracks DP HPD in data_status, calls drm_connector_oob_hotplug_event()
arch/arm64/boot/dts/apple/t8103-jxxx.dtsi +50 lines — enables dcpext + DART + mbox, the dpxbar mux, displayport = <&dcpext> on the connector

The tipd half builds out-of-tree against linux-asahi-headers into updates/; the DT half is a recompiled t8103-j293.dtb packed by update-m1n1. Everything below it already ships in the stock kernel — phy-apple-dptx, mux-apple-display-crossbar, appledrm, and an exported drm_connector_oob_hotplug_event — and CONFIG_MODULE_SIG/CONFIG_MODVERSIONS are both off, so the modules just load. No kernel build.

(One build detail that will bite: compiling the DTB needs Asahi's patched dtc from scripts/dtc. Stock dtc 1.8.1 rejects the float literals in t8103.dtsi — apple,avg-power-ki-only = <7.5> — with a bare syntax error.)

Two more silent failures worth encoding in omarchy debug dp-altmode

Both cost real time and both look identical to "the feature isn't enabled":

  1. The DTB in boot.bin is not the DTB you installed. update-m1n1 selects $(ls -d /lib/modules/*-ARCH | sort -rV | head -1)/dtbs/*.dtb, so a locally built kernel (7.1.13-fairydust+) is skipped by the glob entirely and the stock device tree keeps booting. I believe this is what happened in MacBook Air M2 (t8112): external display over USB-C DP altmode not detected — no dcpext device tree node #384. The check that survives board differences is the hack's fingerprint in the live tree, rather than a node name:

    ls /proc/device-tree/soc/i2c*/usb-pd@*/connector/ | grep displayport

    (On t8112 the DP-alt path uses dcp, not dcpext — per commit bd435201b527, "dcpext is used for the HDMI port" — so grepping dmesg for a second dcpext device gives a false negative there.)

  2. The initramfs shadows the patched module. tps6598x-core loads in early userspace, so whatever is in /boot/initramfs-linux-asahi.img wins and updates/ is never consulted. A kernel upgrade regenerates the initramfs mid-transaction, so the natural ordering — upgrade, then install modules — bakes in the stock module. The DT half works, the monitor asserts HPD correctly, and nothing happens:

    lsinitcpio /boot/initramfs-linux-asahi.img | grep tps6598   # must show the updates/ path

Also worth a mention in the doc

The tps6598x tracepoints are the fastest way to split host-side from monitor-side, and they turn a guessing game into one command:

echo 1 | sudo tee /sys/kernel/tracing/events/tps6598x/*/enable
# replug, then:
sudo cat /sys/kernel/tracing/trace

cd321x_data_status: ...|DP_CONNECTION|HPD_LEVEL, DP pinout D proves cable, OSD settings and PD negotiation are all fine and moves the search host-side. Pinout D also nuances the note above about pin assignment C — with D you keep SuperSpeed on two lanes, so losing SuperSpeed isn't universally "what success looks like".

Lastly, AsahiLinux/linux#622 merged on 2026-09-14 and fixes DP-alt hotplug freezes plus parks the PIPE handler before a Type-C mode switch, which is the phy-apple-atc: Failed to lock pipehandler noise.

Caveat: all of the above is verified on t8103/j293 only. Happy to test patches or a debug command on that hardware.

@malik-na malik-na added lvl 0 Core userspace: display, audio, video, graphics, memory, storage, battery, keyboard, network, BT. US labels Sep 18, 2026
@garymeyerca

Copy link
Copy Markdown

M1 Mac mini (j274/t8103) USB-C display field result — 2026-09-29

I tested a direct USB-C connection from an M1 Mac mini (2020, apple,j274) to an ASUS PA329CRV. This uses the experimental Aurora Thunderbolt/display series, PR #8, rather than the older Fairydust-only path described here. The desktop kernel was 7.1.12-usbcdp+ from that PR's 9632d3ec tip, with a local t8103-jxxx.dtsi alias fix. The booted device tree was a patched t8103-j274.dtb packed into m1n1 v1.6.1, and the kernel command line included thunderbolt_apple.dp_display=1.

Result: the ASUS appeared as DRM USB-2 (card1-USB-2 in the console test, card2-USB-2 in the desktop boot). The kernel read a 256-byte EDID and completed a 3840×2160 @ 60 Hz modeset. The monitor showed a picture; in the subsequent desktop boot Hyprland drove it at 3840×2160 @ 60 Hz, scale 1.6. asahi, appledrm, and thunderbolt_apple loaded, and Hyprland reported no configuration errors.

Two bring-up details may help others diagnose an apparently dead connection:

  1. With a Satechi Multiport Pro and the ASUS both attached, the Satechi's USB-C port acquired the display crossbar route first. The ASUS port's Type-C mux returned -EBUSY. After I unplugged the Satechi and reconnected the ASUS cable, the ASUS acquired the route and worked. The Satechi also enumerated a USB Billboard Device; that alone did not identify the cause.
  2. The first test lacked the m1n1 v1.6.1 USB4 aliases needed by this kernel. Adding usb4_{0,1}_rc, usb4_{0,1}_nhi, and usb4_{0,1}_pcie_adapter to the j274 DTB made the live ACIO nodes receive nonempty apple,tunable-rc, apple,tunable-nhi, and apple,thunderbolt-drom; both ACIO controllers then bound to thunderbolt-apple-acio.

Limit: this proves one direct-display reconnect and one successful desktop boot. It does not establish repeated hotplug, suspend/resume, docked display, or durable installation. The test's recovery service restored the original m1n1 boot bundle after the patched DTB had loaded, so the next ordinary boot has not been shown to retain USB-C display support. GRUB remains on its stock default for that reason.

This may be useful context for the doc's distinction between a wired device-tree route and a working display, and for noting the newer Aurora path alongside Fairydust. I can provide a focused, redacted journal/DT capture if it would help the review.

Field report assembled with Codex assistance; hardware observations were checked on the M1 Mac mini.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

cherry-pick-later Port to omacom/omarchy after the overlay merges; match mainline. lvl 0 Core userspace: display, audio, video, graphics, memory, storage, battery, keyboard, network, BT. US

Projects

None yet

Development

Successfully merging this pull request may close these issues.

9 participants