Skip to content

arm64: dts: apple: t6050-j714s: route dcpext1 to the left-front port's DP tunnels - #2006

Merged
DjDeveloperr merged 2 commits into
aurora-silicon:integration/aurorafrom
Isaac-Duarte:j714s/dcpext1-left
Nov 16, 2026
Merged

DjDeveloperr merged 2 commits into
aurora-silicon:integration/aurorafrom
Isaac-Duarte:j714s/dcpext1-left

Conversation

@Isaac-Duarte

@Isaac-Duarte Isaac-Duarte commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

Stacks on #1921. This branch is #1921's j714s-third-dock-display (4c6a7b4) plus two commits. Only the last two are new here:

  1. 3cc9ce453f "arm64: dts: apple: t6050-j714s: route dcpext1 to the left-front port's DP tunnels"
  2. 0e0169b201 "arm64: dts: apple: t6050-j714s: Add the left-front router's DP IN interrupts", folded in from arm64: dts: apple: t6050-j714s: Add the left-front router's DP IN interrupts #2002 (now closed) so that the two don't conflict over the file's header comment.

Review #1921 first; I'll rebase this once it merges.

What

Two commits, one file (arch/arm64/boot/dts/apple/t6050-j714s-display.dtsi), J714s only. No driver change.

1. Route dcpext1 to the left-front port

dcpext1 gets a second tunnel-only route, route@1, to the left-front port (ATC1, &typec1). It mirrors dcpext0 and dcpext2's left routes and dcpext1's right route:

  • PHY index 1 (atc1_phy), as dcpext0 and dcpext2's left routes have it.
  • The left-front crossbar's controls (atc1_dpxbar 0-4, named typec1 and typec1-dpin0..typec1-dpin3). The driver looks these up by the route's reg.
  • Crossbar source 2 (dispext1 port 0), the same as on the right port:
    • Both crossbars are apple,t6050-display-crossbar, whose state is the dispext bit index (2 × dispext + port).
    • dcpext0 and dcpext2 use 0 and 4 on both crossbars.
  • apple,tunnel-only, no mode-switch:
    • hpm1, like hpm2, already has TYPEC_MUX_MAX_DEVS (3) mode-switch providers: atc1_phy, dcpext0 and dcpext2.
    • Tunnels find their port through the endpoint, so the new route links to a new typec1 endpoint@2.

Connector names stay stable. #1921 creates tunnel-only port/pipeline pairs after the others, so the new pair is appended:

connector pair
DP-1..DP-4 unchanged (dcpext0 and dcpext2 × right / left-front)
DP-5 dcpext1 × right (as in #1921)
DP-6 dcpext1 × left-front (new)

2. The left-front router's DP IN interrupts (from #2002)

&acio1 gets its four DP IN interrupts, dpin0..dpin3 (AIC 1635, 1636, 1641, 1642), next to &acio2's and in the same form. Where the values come from:

  • atc1_dpin0..3 in t6050-peripherals.dtsi (the T6050 inventory).
  • J716c's acio1 (t6050-j716c-left-usb4.dtsi) has the same four.
  • They are acio2's (1667, 1668, 1673, 1674) shifted down by 32, like the rest of the two routers' interrupts.

It fixes nothing that is broken, but it does change behaviour. Without a dpinN line, the router driver polls a mapped DP IN every 20 ms (APPLE_DPIN_POLL_MS). The IRQ thread and the poll run the same handler (apple_dpin_irq_dispatch()). With the interrupts:

  • hot plug and IRQ_HPD events arrive without the up to 20 ms poll delay;
  • there is no idle poll per mapped DP IN.

The risk: with a dpinN line present, apple_dpin_irq_start() enables the interrupt instead of the poll. If a line never fired, that DP IN would lose its HPD and IRQ_HPD events after the tunnel came up. The numbers match the ADT and follow acio2's, which work, so I expect them to be right. The hardware evidence so far is thin, though (see Tested), and the To-run item below checks them directly.

The header comment needs no change for this commit: commit 1 already says the left-front port carries USB4 DP tunnels.

Why

A Thunderbolt dock on the left-front port lit only two of three displays. Only dcpext0 and dcpext2 had a route to that port, so DP IN 2 was left waiting:

thunderbolt-apple-acio 2a9ac0000.acio: dpin2: no display pipeline free; waiting for one

On the right port, #1921's dcpext1 route takes the third tunnel. Commit 1 gives the left-front port the same. Commit 2 makes acio1 match acio2 and the T6050 inventory: #1791 added the router without its DP IN interrupts.

What the base needs: the left-front port's USB4 router, &acio1, from #1791. #1921's base (7c71d6d) already has it (fbfb50c), so this branch needs nothing else. Commit 1 works without commit 2: the DP INs are then polled.

Tested

J714s (MacBook Pro 14" M5 Pro).

test result routing (kernel log)
Boot DP-6 is there: card1-DP-6 exists in sysfs, and KWin has used DP-6 since the first boot with this DTB
Dock with three displays plugged into the left-front port all three lit DP IN 0 → dcpext0 (2b2e00000), DP IN 1 → dcpext1 (392e00000), DP IN 2 → dcpext2 (302e00000); no no display pipeline free
Unplug and replug the dock on the left-front port all three back same routing
Dock moved to the right port all three lit dcpext1 on DP IN 1, as before
USB-C to DP cable alone on the left-front port (DP alt mode) lit dcpext0 (2b2e00000), not dcpext1
Cable on the left-front port, dock with two displays on the right port, at once (three displays in all)¹ all three lit cable: dcpext0, Type-C PHY 1; dock DP IN 0 → dcpext1 (392e00000), DP IN 1 → dcpext2 (302e00000)
Reversed: dock with two displays on the left-front port, cable on the right port¹ all three lit dock DP IN 0 → dcpext1, DP IN 1 → dcpext2, both Type-C PHY 1; cable: dcpext0, Type-C PHY 2

¹ A later boot of the same kernel and DTB, with the T6050 PHY module from #2008 loaded (a separate fix).

Commit 2's interrupts are not yet shown to fire.

  • In a later boot, after two left-front dock sessions, /proc/interrupts showed 2a9ac0000.acio-dpin0..dpin3 = 0/0/0/0.
  • The right router's DP IN lines also showed 0 in that boot, so a steady dock may simply raise none.
  • An earlier read of dpin0 = 1 came from a different boot and can't be tied to these runs.

All of the runs above had commit 2's interrupts, so the displays came up and stayed up with the poll off. What hasn't been exercised is a sink-side hot plug or IRQ_HPD on a left-front tunnel that is already up.

Build:

  • t6050-j714s.dtb built with W=1 at drm/apple: dcp-fabric: J714s: drive a dock's third display through dcpext1 #1921's head, at commit 1 and at commit 2 (0e0169b201). No new dtc warnings.
  • Commit 2's DTB differs from commit 1's only in acio@2a9ac0000's new interrupts (AIC 1635, 1636, 1641, 1642, level high) and interrupt-names (dpin0..dpin3).
  • Commit 1's decompiled DTB has dcpext1's apple,typec-dptx-phys = <2 1>, apple,typec-mux-indices = <2 2>, both routes apple,tunnel-only, and route@1 and typec1 endpoint@2 linked to each other.
  • Only t6050-j714s.dts includes this file, so no other DTB changes.
  • checkpatch --strict: commit 1 has 0 errors and 1 warning (the quoted kernel log line); commit 2 is clean.

Bindings: checked by hand against apple,dcp.yaml. This uses the same property shapes that dcpext0 and dcpext2 already have, plus #1921's apple,tunnel-only. dt-validate was not run.

Not tested

  • More displays than pipelines, e.g. three on a dock plus a DP alt mode display: only three monitors were available.
  • A fourth dock display (DP IN 3).

To run before merge

  • J714s: a DP alt mode display on the left-front port lights on dcpext0 or dcpext2, never 392e00000.
  • J714s: a dock on the right port and a DP alt mode display on the left-front port at once, and the reverse.
  • J714s: with a dock on the left-front port, power-cycle one of its monitors. Its 2a9ac0000.acio-dpinN count in /proc/interrupts rises and the display comes back.
  • J714s: four displays (three on a dock plus a DP alt mode display), where one has to wait for a pipeline.
  • Rebase onto integration/aurora once drm/apple: dcp-fabric: J714s: drive a dock's third display through dcpext1 #1921 merges, and rebuild the DTB.

Later tidy-up, not for this PR: acio1's and acio2's DP IN interrupts live in t6050-j714s-display.dtsi, while #2017 puts acio0's in the router's own file. They should end up in one place.

@DjDeveloperr DjDeveloperr left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Validation on J714s: no regression on the left-front port; the new route itself not exercised

Our kits k2b to k6 carried this branch at 0e0169b unchanged, with #1921: integration/aurora 514ae1f + #2006 + #2018 (its earlier version) + our follow-up fixes.

Rig:

  • internal panel;
  • Dell S2725QC (3840x2160@120) on the left-front port by DP alt mode (no dock there);
  • Thunderbolt 5 dock on the right port: JTG S123 (1920x1280@60) on DP IN 0, Samsung Odyssey G81SF (3840x2160) on DP IN 1.

Each kit was booted once and read out only; then the default kernel was restored and verified.

What we saw:

  • DP-6 (dcpext1 × left-front) exists in every kit, disconnected. DP-1..DP-4 keep their meaning: the Dell is DP-2 on the stock kernel and in every kit, and the dock's displays are DP-3 (dcpext2 × right) and DP-5 (dcpext1 × right).
  • The left-front DP alt mode display still takes dcpext0, not dcpext1: Type-C DPTX PHY 1, 3840x2160@120, in k3 to k6. In k2b the Dell wasn't up (its port's PD controller came up late in that boot; that path isn't touched here).
  • Final kit (k6): all four displays lit: panel 3024x1964@120, Dell 3840x2160@120, G81SF 3840x2160@120, JTG 1920x1280@60.
  • Faults: 0 no display pipeline free, 0 port already routed, 0 DP tunnel setup failed in all kits.
  • acio1's DP IN interrupts (commit 2): the left-front router probed and the port worked in DP alt mode. No tunnel ran on that port, so the four lines were never needed.

Not tested here: a dock on the left-front port. So neither the new route nor commit 2's interrupt lines carried a display in our runs. /proc/interrupts was not captured.

I haven't pushed anything to this branch.

…s DP tunnels

A Thunderbolt dock on the left-front port lights two displays: DP IN 0
and 1 take dcpext0 and dcpext2, the only pipelines with a route to that
port, and DP IN 2 waits:

  thunderbolt-apple-acio 2a9ac0000.acio: dpin2: no display pipeline free; waiting for one

On the right port, dcpext1's tunnel-only route takes the third tunnel.
Give dcpext1 the same route to the left-front port, as route@1 beside
the right port's route@0, like dcpext0 and dcpext2:

- PHY index 1 (atc1_phy) and the left-front crossbar's controls
  (atc1_dpxbar 0-4, "typec1" and "typec1-dpin0".."typec1-dpin3"), as
  dcpext0 and dcpext2 have them.
- Crossbar source 2 (dispext1 port 0), as on the right port: both
  crossbars are "apple,t6050-display-crossbar", whose state is
  2 * dispext + port; dcpext0 and dcpext2 use 0 and 4 on both.
- apple,tunnel-only and no mode-switch: hpm1, like hpm2, already has
  TYPEC_MUX_MAX_DEVS (3) mode-switch providers (atc1_phy, dcpext0 and
  dcpext2). Tunnels find their port through the endpoint, typec1's new
  endpoint@2.

The DRM connectors keep their names: tunnel-only port/pipeline pairs
are created after the others, so dcpext1 x left-front is DP-6 after
DP-5 (dcpext1 x right).

Signed-off-by: Isaac Duarte <31712685+Isaac-Duarte@users.noreply.github.com>
…errupts

The left-front port's USB4 router (acio1) came with the port's USB4
support, but without the DP IN interrupts that the right port's router
(acio2) has. With no "dpinN" interrupt, the router driver polls a mapped
DP IN's status every 20 ms instead; the interrupt and the poll run the
same handler. This is not a functional fix: with the interrupts, hot plug
and IRQ_HPD events arrive without the up to 20 ms poll delay, and there
is no idle poll per mapped DP IN.

Give acio1 the four DP IN interrupts, as acio2 has. The numbers are those
of atc1-dpin0..3 in the T6050 inventory and of J716c's acio1, and follow
acio2's, 32 lower like the rest of the router's interrupts. The DP IN
blocks there sit at the offsets the driver maps from the router's "rc"
region. The display side is already in place: dcpext0, dcpext1 and
dcpext2 have routes to this port with its DP IN crossbar controls
(typec1-dpin0..3 on atc1_dpxbar).

Signed-off-by: Isaac Duarte <31712685+Isaac-Duarte@users.noreply.github.com>
@DjDeveloperr

Copy link
Copy Markdown
Contributor

Validated on our M5 Pro (J714s): the two commits of this PR, at the same patch-ids as this rebased head, ran together with #1921 in our test kit and in the default candidate built from the same source. All four of the owner's displays lit at native modes (panel 3024x1964@120; Dell S2725QC and Samsung G81SF at 3840x2160@120; JTG S123 at 1920x1280@60). Read-outs at about 60, 150 and 300 s showed every connector connected, with no routing or tunnel failures and no oops, MMU, DART or GPU faults. CI is green on 6997339. Merging.

@DjDeveloperr
DjDeveloperr merged commit a450ca0 into aurora-silicon:integration/aurora Nov 16, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants