[jgoyvaerts]

This adds the device tree for the Apple M4 Pro (T6040) and the MacBook Pro (16-inch, M4 Pro, Nov 2024), J616s / Mac16,7, in the layout of the T8140 and T6030 trees:

• t6040.dtsi: the SoC blocks, disabled by default;
• t6040-pmgr.dtsi: their PMGR power domains, added to the PMGR nodes of t6040.dtsi;
• t6040-j616s.dts: the board, which enables what the J616s uses.

The SoC blocks use the apple,t6041-* compatibles; apple,t6040 stays the machine compatible. Values come from the J616s Apple Device Tree (macOS 26.6.2).

The series is ordered so the base can be merged first and each later commit as its driver lands. Every commit builds on its own.

#	Commit	What	Binding from
1	dt-bindings: apple: Add the T6041 entries the J616s device tree uses	T6041 entries for SART, DockChannel, MTP helper, DCP, cluster cpufreq and SPMI; lane 2 of apple,fast-die-effort	this PR
2	arm64: dts: apple: Add the T6040 SoC and the MacBook Pro J616s	CPUs, cpu-map and capacities, AIC, timer, PMGR, pin controllers, I2C, watchdog, PMIC NVMEM, SMC (GPIO, reboot, RTC), keyboard backlight	tip
3	arm64: dts: apple: t6040: Add MTP (keyboard, trackpad) support (Josh Rodd)	MTP, mailbox, DART, DockChannel HID	tip, commit 1
4	arm64: dts: apple: t6040: add cluster cpufreq and OPP tables (CJ Damsleth)	cluster DVFS blocks, OPP tables	commit 1
5	arm64: dts: apple: t6040: Add the cpuidle node	cpuidle	tip
6	arm64: dts: apple: t6040-j616s: Add the SMC sensors and a CPU die thermal zone	SMC hwmon (TCMz, fans, shared sensor lists), thermal zone	tip
7	arm64: dts: apple: t6040-j616s: Add Wi-Fi and Bluetooth	PCIe, its DART, BCM4388	tip
8	arm64: dts: apple: t6040: Add the NVMe controller	ANS2, mailbox, SART	#387, commit 1
9	arm64: dts: apple: t6040: Add the video decoder	AVD, DART	tip
10	arm64: dts: apple: t6040: Add the PMP and link the CPU clusters	PMP, report, DART, mailbox; apple,fast-die-effort	tip, commit 1
11	arm64: dts: apple: t6040: Add the display controller	DCP, mailbox, DARTs, display subsystem, panel (disabled; the loader enables them)	commit 1; the handoff needs #386
12	arm64: dts: apple: t6040: Add the PMC voters, CLPC and SLC planes	PMC, CLPC, MCC	#385
13	arm64: dts: apple: t6040: Add the AOP and its audio DMA	AOP (audio with all five DMA channels, ALS, LAS), mailbox, DART, two ADMACs	AOP series (github.com/jgoyvaerts/linux branch m4pro/aop-t6041); ADMAC on tip
14	arm64: dts: apple: t6040-j616s: Add the speakers	six SN012776 amps, SDZ regulator, sound card	speaker series (github.com/jgoyvaerts/linux branch m4pro/speakers-j616s)
15	arm64: dts: apple: t6040-j616s: Add the headphone jack	CS42L84, jack sound card	jack series (github.com/jgoyvaerts/linux branch m4pro/jack-j616s)
16	arm64: dts: apple: t6040: Add the ISP and the J616s camera	ISP, three DARTs, ISP power domains, IMX958 presets	#384

"tip" = on integration/aurora; #371, #372, #373, #374 and #379 are merged.

Layout notes (where this differs from the T8140/T6030 trees, and why):

• No soc simple-bus. The nodes sit at the root because our boot loader edits several of them by path: the PMGR blocks, the AIC, the DCP, the AOP, the AVD, the NVMe and PCIe nodes, /cpus and /chosen.
  • So nonposted-mmio sits on the blocks known to need it: cpufreq, PMC, MCC, PMP report and DockChannel.
  • The L2 caches are root nodes too.
• PCIe node (commit 7). It is our booted tree's node; only its msi-ranges now names the AIC by label (<&aic AIC_IRQ 2071 IRQ_TYPE_EDGE_RISING 32>). Our loader's Bluetooth admission depends on that node.
• CPU compatibles are apple,brava-e/apple,brava-p (dt-bindings: Apple M4 Pro/Max (T6040/T6041) compatibles #372). The booted tree had arm,armv8; nothing in the kernel or the loader matches CPU compatibles.
• AVD windows. The ADT gives the AVD as one block (0x2_80000000, 0x1404000). The windows use the same code/SRAM/mailbox pattern as T6030/T8140, sized so the mailbox lands at +0x10a8000 (code and SRAM 0x14000 each).
• ADMAC interrupts. Each ADMAC is wired to the first of its four interrupt outputs, so the other three entries are empty, as on T8140.

Hardware tested on: Apple MacBook Pro (16-inch, M4 Pro, Nov 2024), J616s / T6040, macOS 26.6.2 firmware, booted by our own loader. Our daily tree has run Linux 7.3.0-m4.3 daily since 2026-10-22.

How tested: this tree has not been booted as such. Our daily tree is these files plus one private include. The include holds what only our loader and our not-yet-submitted drivers need, and pins the phandles the PCIe node carries. That combined tree was compiled and compared with the booted daily tree (m4/aurora-7.3, 8af0f4d):

• at the DTB level (dtc -I dtb -O dts -s, phandle cells resolved to node paths), and
• at the source level.

Every node path is equal (apart from the dropped keyboard PWM clock node), and the PCIe node is byte-for-byte equal (raw phandles included). Property differences:

• the GPU: our tree carries the Neo-style GPU node of our T6040 GPU series (not yet submitted). That node is not in this PR; it comes with the GPU series as one more commit in this layout;

• the binding and review fixes:

  • PMGR state compatibles, PMP report entry compatibles and #power-domain-cells;
  • cpufreq and SPMI compatibles;
  • the panel's apple,panel-j616 entry;
  • apple,brava-* CPU compatibles;
  • AVD sram window;
  • ADMAC interrupts-extended;
  • the keyboard PWM on clkref (its separate 24 MHz clock node is gone, the only path difference);
  • the ADT names as the SSD/SIO/FPWM power-state labels;
  • the model string.

  None of them changes what the drivers or the loader do. I checked the drivers' match tables and of_device_is_compatible users, the loader source and our packaging. In particular, the AVD driver does not map the SRAM, and the ADMAC keeps its interrupt at output 0;

• cpu-release-addr = <0 0> on the 13 secondary CPUs and reg = <0 0 0 0> on the framebuffer. These are placeholders the loader overwrites on every boot;

• status = "okay" on the nodes the board enables.

The combined tree is 62,250 bytes, under the loader's 98,304-byte board.dtb limit.

Not tested / limits:

• No boot of the public tree alone, nor of the combined tree with the binding fixes. The latter is the next daily test boot. Without the private include, the public tree lacks:

  • our loader's placeholders (direct-smp-*, dcp-map-pool*, headphone-codec-probe);
  • the NVMe write guard;
  • the loader's apple,t6040-* second compatibles.

  A loader that fills /memory, the framebuffer, the CPU release addresses and the PMP's per-unit properties should be able to boot it, but none has.

• Left out because no binding for them exists yet. Our daily tree keeps them in its private include:

  • USB-C: the three ports, MagSafe, and the T6040 ATC PHY and DWC3 with their USB3 tunables;
  • DART address filters: the AOP and ISP DART filters (apple,dapf-entries);
  • Trackpad: the trackpad AFE's SMC power key (apple,afe-smc);
  • DCP: three iomfb properties of our imported display driver.

• No GPU node yet: it follows with the T6040 GPU series. Its operating points will be this unit's values, binned per device, until a loader fills them in.

• The i2c6 retimers have no driver. i2c4 (SD card helper) and i2c8 (HDMI converter) stay disabled.

Follow-ups (each needs a test boot or a driver or loader change, so not in this PR):

• opp-microwatt for the CPU OPPs: without it, the cpufreq fallback while the PMP is not running is the slowest state.
• The nub GPIO bank's interrupts: move the omission from the board into t6040.dtsi, with the measured reason (the AOP owns some of its pins and interrupt registers).
• nonposted-mmio on every SoC node, not only the blocks known to need it.
• PMS as the parent of the PMP and PMS_SRAM states (ADT parent).
• base-ns ADMAC dma-channels = 26.
• The AOP unit address (its node is named after its ASC range; renaming needs a loader path change).
• Merge pmgr_isp into the main PMGR node.
• SMC: use "apple,t6041-smc", "apple,t8103-smc". Not done because the DockChannel HID's AFE power-up refuses an SMC without apple,smc (dockchannel-hid.c).
• DockChannel HID: use plain apple,dockchannel-hid. Not done because the daily HID driver keys its ring mode and the AFE power-up on apple,t6041-dockchannel-hid.
• ISP firmware reservations: write isp-firmware-text/-data as boot-loader placeholders (reg = <0 0 0 0>, as isp-common.dtsi). Not done: a zero-size reg next to iommu-addresses makes of_iommu_get_resv_regions set up a translated mapping of the firmware IOVA to physical address 0 (phys.end wraps), so an unfilled placeholder would point the ISP at memory 0. The fix is the IOMMU core treating a zero-size reg as absent, or the loader filling it.

Depends on: per commit, see the table. Open: #384, #385, #386 and #387; not yet opened: the AOP, speaker and jack series.

Credits:

• The apple,t6041-* naming follows Yureka Lilian's acked T604x bindings, merged with dt-bindings: Apple M4 Pro/Max (T6040/T6041) compatibles #372 (commit 573f98b).
• The MTP/DockChannel nodes are Josh Rodd's, from his MacBook Air M4 MTP support. They are carried as his commit, adapted to this layout.
• The cluster cpufreq blocks and OPP tables are CJ Damsleth's, from his J614s tree (https://github.com/damsleth/wallace, commit 90ea03566545). They are carried as his commit, adapted to this layout. The SMC RTC cell location also follows his tree.
• The thermal zone and the CPU capacity method follow Kb2uka's J713 overlays (Omarchy-M4-Native).
• The layout follows aurora's T8140 and T6030 trees.

Base: integration/aurora (8c2774e). Built and checked there:

• make dtbs W=1 for apple/t6040-j616s.dtb gives no warnings. On the same base at W=1, t8140-j700.dtb gives 5 and t6030-j516s.dtb 17. Each commit builds without warnings.
• checkpatch --strict gives three line-length warnings, all in the booted tree's PCIe node, and the "MAINTAINERS" note for the new files.
• dt_binding_check is clean for the six bindings commit 1 touches.
• dtbs_check (dtschema 2026.9) ran against the tip's bindings plus media: apple: isp: add T6040 (M4 Pro) camera support #384, soc: apple: pmc, clpc, mcc: add T6040 (M4 Pro) #385, nvme: apple: add T6040 (M4 Pro) ANS2 support #387, the AOP/speaker/jack series and commit 1. What remains comes from bindings that do not describe the T6041 yet, or from the booted tree's PCIe node:
  • SMC compatible: the SMC keeps its three-entry compatible (see follow-ups);
  • DockChannel HID: the HID keeps its apple,t6041-dockchannel-hid entry (see follow-ups), and its mtp and tp-accel children are not in the binding;
  • nonposted-mmio on the cpufreq, PMC, MCC and PMP report nodes, whose bindings do not list it;
  • PCIe node: reg-names lacks port1/2; the port properties; the Wi-Fi compatible without a PCI ID; the wifi-clkreq pin node name; local-bd-address, which the loader adds;
  • Sound nodes: the dai-tdm-* properties, the CS42L84 sound-name-prefix, and the two-entry SN012776 compatible.

🤖 Generated with Claude Code