[DjDeveloperr]

The M4 PMP bridge, PMGR policy, PMP report, telemetry and thermal, the display handoff and the G16 GPU path were gated on the M4 Airs (apple_t8132_air(), J713/J715). What they depend on is the firmware. This gates them on the SoC and the qualified OS firmware, 26.6.2, so the J604, J773g, J623 and J624 take the same path as the J713. Board data, such as panel timing and which USB-C ports exist, stays in the board trees and profiles.

The gate

apple_t8132_qualified_firmware() is true on apple,t8132 with /chosen/asahi,os-fw-version = "26.6.2", which aura publishes (iBoot mBoot-18000.161.10). apple_t8132_firmware_path() is that, or an Air. The Air term keeps the J713 and J715 exactly as they are on any firmware. It can go once the Airs only ever run 26.6.2.

Evidence

From work/adt/:

• The J604 ADT as iBoot hands it over on 26.6.2 has the J713's PMP: iop,ascwrap-v6, iop-version 1, interrupts 878/877/880/879, and the same iop-pmp-nub sochot-sensors and lts-ctrl-loop. It has the same gfx-tvm, and sgx and gfx-asc at the J713's addresses. Values over 256 bytes are cut from that dump.
• The J604's 27.0 ADT and the J773g's 26.6.2 ADT carry the J713's soc-device table byte for byte: 34 records, AGX at 12 and DISP at 15, which the bridge checks itself. They also have the same temp-sensor and controller records, and PMGR devices, ps-regs, pwrgate-regs, power-domains and gfx-tvm.
• Only lts-config, lts-clpc-voltage and the J773g's dvfs-disp-min-state differ. aura hands each Mac's own values to the PMP firmware.
• The GPU DVFS tables have the shape G16 qualifies: perf-state-count 16, one table, 15 GPU states. Top clocks differ, and the driver keeps each Mac's own values.

Where a board difference is real

I compiled every M4 tree and checked the PMGR policy against what each board's enabled devices use. The J713 policy differs from the other boards in exactly two places:

• The Airs wire only ATC0/ATC1, so the policy disables the ATC2/ATC3 common, usb and usb_aon domains. The J604 uses ATC3's, and the J773g and J624 use both ports'. Those boards now state these domains okay, as the policy does for ATC0/ATC1.
• The J773g has no camera, and its tree keeps the ISP domains disabled.

The platform handoff now keeps a non-Air tree's explicit status, except on domains the policy gives the firmware (firmware-owned or inherited-on), such as the DCP CPU domains the SoC tree marks "Locked, don't touch it". After the handoff, every M4 tree ends up with the J713's domain states apart from those two differences, and no enabled device sits on a disabled domain. On the Airs the handoff still overrides any status, as the original J713 loader tree needs.

Panel timing stays per board in dcp-m4-profile.h, which has only the J713 and J715. The J604 and the iMacs need their own profiles before their internal displays are adopted. The J773g has no internal panel. The GPU doesn't wait for a display the loader didn't hand over (#1170).

Commits

1. PMP bridge (and with it G16's board check), telemetry, thermal and report readiness on the firmware path. The host harnesses model no /chosen.
2. J604, J773g, J624: state the ATC2/ATC3 domains they wire.
3. PMGR policy handoff and the pmgr-pwrstate check on the firmware path.
4. Display DARTs, dcpext routes and external output sleep handling on the firmware path; the retained display ADT check accepts every M4 target-type. Each still acts only on what the loader hands over.
5. rtkit-helper: the legacy live-MTP path follows apple_t8132_live_mtp(), which includes the J604.
6. t8132-gpu.dtsi: describe the GPU, mailbox and TVM gate for every M4 Mac, with the TVM parent, apple,sleep-retained on ps_gfx and the PMP report link, and drop the J713/J715 copies.

Still Air-only: the PCIe identity-domain default in apple-dart, which isn't firmware-bound and changes working PCIe on other boards.

Testing

• Every M4 dtb builds without dtc warnings. Against integration/aurora 72b828e, the J713 and J715 dtbs differ only by three default status = "okay" lines. dt-validate shows no new warnings on any M4 board. That needs phy/ti,cd2e22.yaml dropped from the schema; CI crashes on apple,tunable for every T8132 dtb.
• Built together with drm/asahi: M3 Pro/Max and M4 GPUs on by default, M4 Max GPU enabled #1170 on this base (asahi profile, Rust on): no warnings. The built J604 dtb has the GPU node enabled. The full host suite passes 195/195 together with drm/asahi: M3 Pro/Max and M4 GPUs on by default, M4 Max GPU enabled #1170 on 72b828e, and on this branch alone before the rebase onto arm64: dts: apple: lay out the M4 trees like the M1/M2 ones #1166.
• Nothing has run on a J604, J773g or iMac. With current aura the GPU stays off there, because aura disables the GPU node on every board but the J713. The aura PR moves that to the firmware too.

Base

Rebased on integration/aurora 72b828e, after #1166. The GPU, mailbox, TVM parent, sleep-retained and PMP report link move from t8132-j713-j715.dtsi into the shared t8132-gpu.dtsi. The J713/J715 dtbs then differ only by three status = "okay" lines that were the default anyway.

🤖 Generated with Claude Code