M4 Mac mini reference

HDMI/display bring-up on J773g

This is the short, code-pointer-first description of the working Linux sequence. It is hardware bring-up documentation: do not transplant it into a cold loader or change power/firmware ordering without a recovery-qualified trial. It describes the retained-iBoot DCP path used by the run-59 kernel, not a recipe for starting DCP firmware from a cold loader.

Working order

  1. Keep the display fabric powered. The loader-visible DT keeps ps_disp_fe and ps_dptx_phy available and describes the main dcp@30ae00000, its DART/mailbox, five memory regions, the T8122 DPTX PHY, and GPIOs hdmi-pwren, dp2hdmi-pwren, hdmi-hpd, and dp2hdmi-reset. Linux’s M4 path calls apple_pmp_j713_display_prepare() before touching the retained session. Code: drivers/gpu/drm/apple/iomfb_m4.c:iomfb_m4_init() and drivers/gpu/drm/apple/pmp-display.c; DT/source delta: private evidence receipt (sections for t8132-j773g.dts, t8132.dtsi, and iomfb_m4.c).

  2. Prepare the fixed HDMI route before DCP adoption. In drivers/gpu/drm/apple/dcp.c:dcp_platform_probe_pipe(), a apple,t8132-dcp whose connector is HDMI-A selects the main-external path. dcp_m4_hdmi_prepare() gets dp-phy, runs phy_init() then phy_power_on(), asserts both converter power GPIOs. The T8122 implementation is drivers/phy/apple/dptx.c:dptx_phy_activate_t8122(); it performs the preserved-value PHY writes. Source: the dcp.c and dptx.c sections of the same integration diff above.

  3. Adopt, do not restart, iBoot’s DCP RTKit session. iomfb_m4_init() selects the J713-compatible external profile for this J773g main DCP, validates the main resource, maps the firmware/shared regions, prepares PMP, then opens the retained DCP endpoints. dcp_pin_live_session() deliberately keeps that session alive; iomfb_m4_adopt_external() completes the M4 handoff. The critical endpoint traffic is the DCP RTKit/AFK service set, including system, property, IOMFB, DPTX and OS-log notifications; the AP must acknowledge OS-log type-2 notifications before DCP can report AP QUIESCED. Code: drivers/gpu/drm/apple/iomfb_m4.c, dcp.c, dcp-afk.c, and dcp-oslog.c; contract source: private evidence receipt, rule text around the DCP quiesce/OS-log requirement.

  4. Connect DPTX and establish hotplug. After adoption, dcp_desktop_dptx_connect(dcp, 0) connects port 0. The DPTX endpoint activates the PHY/lanes, trains the link, and reports HPD; the fixed-HDMI DRM wrapper is registered by drivers/gpu/drm/apple/t8132_ext_drm.c:t8132_ext_out_init(). The observed working sequence reaches HPDMonitor: // off -> on only after IOMFB switches the pipeline to normal mode. Evidence: private evidence receipt lines 123-150; code delta: the dcp.c and t8132_ext_drm.c sections of the integration diff.

  5. Mode-set, then swap. DRM reads modes after HPD, performs the atomic modeset through the fixed external connector/CRTC, allocates an IOMFB surface, submits it to DCP, and waits for the firmware swap/completion callback before retiring the old scanout. Follow drivers/gpu/drm/apple/t8132_ext_drm.c into drivers/gpu/drm/apple/iomfb_m4.c (surface/map, swap submission and swap completion functions). The evidence records a normal-mode transition, HPD on, and subsequent pipeline programming. The exact public function names for the final swap on the current rebase are unverified because the cited notes snapshot contains the topic diff rather than the complete topic tree.

Why Aura 7c5ab9a3 is not equivalent

The Linux path powers the PHY and converter, then adopts iBoot’s already running DCP RTKit session and follows the kernel’s endpoint/quiesce/hotplug ordering. Aura branch techmore/aura:j773g/boot-hdmi-20261112@7c5ab9a3d90b instead adds loader-side boot-HDMI work before Linux and was composed with the run-59 DTB and U-Boot. Its payload hangs, so it is not a known-good display initializer. The precise failing Aura operation is unverified; do not infer that the kernel sequence can be shortened to Aura’s ordering. Build composition and isolation source: private evidence receipt on notes branch j773g/boot-hdmi-20261112@bec15a4d6106; result source: the same run’s receipt (unverified on origin/main at publication time).

Known-good boundary

The last proven HDMI kernel is Image SHA-256 13d3d966… behind Stage 2 1838b1f3… (run 59). Treat any change before the HPDMonitor off -> on transition as power/PHY/RTKit/hotplug work, and any change after it as modeset/surface/swap work. Source: private evidence receipt and the run-59 receipt referenced there.

Public code history: Linux HDMI topic commit 396679fe and Aura boot-HDMI commit 7c5ab9a3.

Part of Techmore