[techmore]

• Target: omacom/linux-aurora, base integration/aurora (2a76ae3)
• Head: techmore:m4/j773g-boot-v3 (24f75ab), 11 commits, +403/-2, device tree and bindings only

Status: work in progress, not booted

• Built, not booted. This branch builds (full Image modules dtbs, 0 warnings), and dtbs_check is clean on all six T8132 boards. That is a compile result, not a runtime pass: no kernel from this tree has booted on a J773g yet.
• One boot attempt failed. An omacom Aurora kernel (7.2.11) with this series' J773g DT hung before the root filesystem on our J773g. That was v2 of this series plus the cpufreq/sensors commits; v3 differs by commits 9-11 below.
  • One attempt, on 2026-11-02. Each power-cycle into it failed the same way: no network and a black screen.
  • No logs were captured: no journal entry and an empty pstore.
  • The same stage-2 DTB booted a GravityLinux 7.1.13 kernel twice.
  • The initramfs held every module the SSD path needs, so a missing module is ruled out. The cause is not known yet.
• Debugging is underway. Commit 9 reserves a ramoops region as an early persistent-log backend. Whether its contents survive a reboot through this boot chain is not established yet: a test marker did not survive a warm reboot under GravityLinux. Test kernels now start by kexec from the running system, with pstore and the panic options built in, so no boot files change.
• Blocker: commits 2 and 5 are Yureka Lilian's and have no Signed-off-by from her in the history they come from (see Credit). They need her Signed-off-by, or a rewrite from the ADT, before merge.
• Follow-ups: CPU frequency/sensors/fan and the PMP RFC are ready on top of this branch (techmore:m4/j773g-cpufreq-sensors-v3, techmore:m4/j773g-pmp-rfc-v3). They will be opened once this PR settles.

Verification (v3, 2026-11-03)

• Base: integration/aurora 2a76ae3. The rebase from the v2 base 4fdd0fc was clean; commits 1-8 are unchanged.
• DART aperture change (21d3646): needs no change here. It only affects consumers with their own apple,dma-range, and none of these nodes has one.
• dtc 1.8.1: all six t8132-*.dts at every commit, 0 warnings, with avoid_unnecessary_addr_size, alias_paths, unique_unit_address, pci_device_reg and the strict name checks.
• make CHECK_DTBS=y (dt-schema 2026.9): all six T8132 boards clean at this head. Before commits 10 and 11, the J773g PCIe node failed it.
• dt_binding_check on the changed bindings: make dt_binding_check (dt-schema 2026.9) with DT_SCHEMA_FILES set to the 11 bindings this series changes: exit 0, and no warning or error names any of them. The only problems reported are in unrelated example files (Apple DCP, i.MX display).
• checkpatch --strict:
  • The only errors are the missing author Signed-off-by on commits 2 and 5.
  • Warnings: the Co-Authored-By trailer form, GravityLinux commit ids in port notes, and the standard pciclass/pci14e4 PCI compatibles in existing context.
• Build: with this series and the follow-ups on top, a full make LLVM=1 Image modules dtbs on the J773g (rust 1.93.1, the omarchy-pkgs linux-aurora config merged with asahi.config) gives exit 0 and 0 warnings. That's Image, 1917 modules and all six T8132 DTBs.

Summary

integration/aurora has t8132-j773g.dts with a RAM-only boot: CPUs, interrupts, power domains, pins, I2C and the boot framebuffer on the HDMI port. This series adds what the Mac mini needs to run an installed system: the NVMe SSD, the internal PCIe bus with Ethernet and the Wi-Fi/Bluetooth module, and the SMC (power-off, reboot, RTC, power button, and the GPIO that powers the Wi-Fi port). No driver changes: every block works with a driver and fallback compatible this tree already has (apple_nvme_t8132_hw, t602x PCIe, t8110 DART, t6000 SART, asc-mailbox-v4, t8103 SMC/SPMI). USB-C, HDMI (DCP), audio and the GPU are not part of it.

This is the M4 bring-up work invited for this repo. It is J773g-only in effect: the SoC nodes are added disabled, and the J773g board file enables them.

Layers (in commit order)

#	Commit	Author	What it gives
1	dt-bindings: apple: Add the T8132 compatibles the Mac mini boots with	Sean Dolbec	apple,t8132-{dart,sart,asc-mailbox,smc,spmi,pcie}, apple,abbey-pmic with the T8122/T8140 fallbacks
2	arm64: dts: apple: t8132: Add the SMC and its mailbox	Yureka Lilian (ported)	SMC + GPIO controller, disabled
3	arm64: dts: apple: t8132: Add NVMe nodes	Yureka Lilian (ported)	ANS2, SART, ANS mailbox, disabled
4	arm64: dts: apple: t8132: Describe the NVMe controller and NVMMU windows	Sean Dolbec	the separate nvmmu window this tree's T8132 ANS2 code needs (ADT reg[3] 0x485cc0000)
5	arm64: dts: apple: t8132: Add the PCIe controller and its DARTs	Yureka Lilian (ported)	controller, 3 ports with PERST#/CLKREQ#, 3 DARTs, SoC dma-ranges
6	arm64: dts: apple: t8132: Add the nub SPMI controller	Sean Dolbec	the bus of the main PMU
7	arm64: dts: apple: t8132-j773g: Enable the SMC, PMU cells, reboot and RTC	Sean Dolbec	power-off/reboot/RTC/power button
8	arm64: dts: apple: t8132-j773g: Enable the SSD, Wi-Fi, Bluetooth and Ethernet	Sean Dolbec	NVMe, PCIe port 0 (BCM4388, apple,sakhalin) and port 2 (Ethernet), MAC/BD aliases for m1n1
9	arm64: dts: apple: t8132-j773g: Reserve a ramoops region	Niklas Sheth (ported)	a 2 MiB pstore region, intended as an early persistent-log backend (survival across a reboot through this boot chain is not established yet). From GravityLinux 877c574, with his Signed-off-by; the same region is in use on our J773g under GravityLinux
10	dt-bindings: pci: apple,pcie: Describe the three T8132 root ports	Sean Dolbec	T8132 gets the T8140 list (3 ports + 3 PHYs) instead of the T6020 rule's ten regions, which dtbs_check failed
11	arm64: dts: apple: t8132: Power the PCIe ports with pwren-gpios	Sean Dolbec	Wi-Fi port power as pwren-gpios (SMC GPIO 19), which pcie-apple drives itself, instead of a fixed regulator plus pciclass,0604 root-port compatibles; same GPIO and polarity

Credit

Commits 2, 3 and 5 are Yureka Lilian's T8132 nodes as GravityLinux gravity-m4 carries them (be925f3, 79b86cb, 9edf52c + bdafece; cherry-picks of her d7525ca925b6, 9a864cea9329, 19103645439f, c59a7a826661). They keep her authorship and date; the port changes are listed in a [sdolbec: …] note in each. Commits 2 and 5 have no Signed-off-by from Yureka in that history; they are forwarded under DCO (b) and need her Signed-off-by before merge (or a rewrite from the ADT). The NVMMU window follows KB2UKA's J713 ADT reading (Kb2uka/Omarchy-M4-Native). The board wiring in commit 8 follows Yureka's J773g PCIe description; the PMU cells in commit 7 are ours.

What was measured

On our J773g (M4, 10-core GPU, 10 Gb Ethernet), the same nodes run today under GravityLinux gravity-m4 7.1.13 with m1n1-gravity 1.6.1: root on the NVMe SSD, AQC113 Ethernet on atlantic, BCM4388 on brcmfmac/hci_bcm4377, SMC power-off (LED off) and reboot, RTC across reboots, power button events. A dtc comparison of this branch's t8132-j773g.dtb against that kernel's DTB shows the same reg, interrupts, MSI ranges, IOMMU maps, GPIOs and power domains for every node in this series. The differences are node names (mailbox@/iommu@), status properties, the added nvmmu window and the SPMI power domain.

What was not measured

• Any boot of this tree on a J773g (see Status).
• The nvmmu window with this tree's NVMe driver. Gravity's driver reads the NVMMU through the controller window, which is kept at 0x60000 for that reason.
• pwren-gpios power sequencing of the Wi-Fi port with this tree's pcie-apple. Gravity powered the port through a regulator.

Merge Danger

Door: two-way. Blast radius: J773g boots. Other T8132 boards gain only disabled nodes and the SoC dma-ranges (the same line T8112, T8122 and T6030 have). On J773g, m1n1 puts this DTB into stage 2 for every kernel; it is built to keep booting GravityLinux kernels too (same compatibles, 0x60000 controller window), but the mini loses USB, DCP, audio and GPU nodes until those follow-ups land, so owners of a working mini setup should not switch stage 2 to it yet.

Follow-ups

CPU frequency, SMC sensors and fan (m4/j773g-cpufreq-sensors-v3, after this PR settles), then USB-C (ATC PHY, dwc3, the SPMI Type-C controllers, the front-port hub), DCP/HDMI, audio, and the GPU once a G16 backend is in this tree. The PMP stays an RFC (m4/j773g-pmp-rfc-v3).

🤖 Generated with Claude Code

[techmore]

@yuyuyureka, hi Yureka. This draft ports two of your T8132 device-tree commits for the M4 Mac mini (J773g):

• arm64: dts: apple: t8132: Add the SMC and its mailbox, from your d7525ca925b6 ("t8132 smc") as carried in GravityLinux gravity-m4 be925f37d4ef.
• arm64: dts: apple: t8132: Add the PCIe controller and its DARTs, from your 19103645439f ("t8132 pcie") and c59a7a826661 ("wifi pwren"), as carried in GravityLinux 9edf52c13ea3 + bdafece9eddb.

Both keep your authorship. The changes made while porting are listed in a [sdolbec: …] note in each commit. Neither original has a Signed-off-by, so they are forwarded under DCO (b) for now, and that blocks merging.

Would you be willing to add your sign-off? A reply here with

Signed-off-by: Yureka Lilian <yureka@cyberchaos.dev>

for those two commits is enough; I'll add it to them. If you'd rather not, say so, and I'll rewrite those nodes independently from the ADT instead.

Your third commit here (Add NVMe nodes) already carries your Signed-off-by. Thank you for the T8132 work this builds on.

[techmore]

Status from the J773g bring-up (2026-11-03), for anyone following this draft:

• Boots with a Gravity-shaped DT. GravityLinux's live J773g DT plus one change — the NVMe node's extra nvmmu window (reg-names = "nvme", "nvmmu", "ans", 0x485cc0000/0x60000) — boots the GravityLinux 7.1.13 kernel to userspace with NVMe on the M4 mini. The integration/aurora 7.3 kernel (19f71a142419 + the SMC change below) boots on that same DT, prints, and panics a few seconds in; the panic text has not been captured yet (no serial console on this unit yet, and the display is blanked by then). So the T8132 NVMe T8132 data path needing the nvmmu window is confirmed; what kills 7.3 shortly after is still open.
• The J773g DT in this series still dies before the first printk when booted by either kernel through Limine (dtb_path). Not a dtc/format or size issue (decompile+recompile is byte-identical; padding to 130 KB changes nothing); PCIe/DARTs, display power domains and the built-in-DTB option are ruled out. Bisecting against the Gravity DT continues.
• Restart path: with aurora-silicon's bounded atomic SMC write (c243a5f), a restart on the J773g can fall through macsmc_restart() to apple_wdt, and iBoot then boots macOS (panicmedic). Giving the atomic path 5 s instead of 500 ms made the restart return to Linux. This tree still has the unbounded poll, so it is not affected; I'll send the fix to aurora-silicon.

Keeping the WIP label; still waiting on Yureka's sign-off for the two ported DT commits.