Repository navigation
kboot: T6041: J575c (Mac Studio M4 Max) external display handoff without a panel - #160
Conversation
The J575c (Mac Studio M4 Max) has no panel. iBoot starts its internal DCP and passes a 640x1136 dummy framebuffer, so display_init() takes the dummy-framebuffer path and runs display_configure() with the M1 config (/arm-io/dcp): the DCP finds nothing connected, the external outputs are not reached that way, and the DCP is put to sleep, or (in a later stage) does not answer HELLO at all. Keep the inherited state on the J575c, as on the J614s, T8140 and T8132. Its displays are external and go through the T6041 handoff of the external processors. Assisted-by: LLM Signed-off-by: Roberto Dantas <7robertodantas@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ut a panel The T6041 handoff is written for the J614C and J616C: it hands over the internal DCP, which enables the display subsystem, and then fills each external processor's firmware memory, using the internal DCP's __TEXT reservation. The J575c (Mac Studio M4 Max) has no panel, so there is nothing to hand over internally: a kernel has no panel to drive, and Stage 1 1.0.0b8 has already put the internal DCP to sleep by then. iBoot does leave the external processors running with their firmware loaded. Take the J575c in the T6041 handoff, but never hand its internal DCP over. Hand over each external processor on its own instead: the first one's ADT __TEXT stands in for the panel's, and its reservation is made here. Once one has firmware memory, enable the display subsystem and its DART, and give the subsystem iBoot's dummy framebuffer as its "framebuffer" region, the DRM aperture the other handoffs take from /vram. The kernel's display gate then enables that processor's chain. Tested on a J575c (OS firmware 26.6.2, Stage 1 1.0.0b8) with aurora-silicon/linux's J575c USB-C display tree: dcpext1 is handed over, the display subsystem is enabled, and a monitor on typec1 shows the Linux console at 2560x1440. dcpext0 is refused, as its tree has no firmware memory placeholders. A tree without the dcp/disp0 aliases (every J575c tree so far) is left alone, as before: dcp6_dt_describes() stops there. Assisted-by: LLM Signed-off-by: Roberto Dantas <7robertodantas@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
No remaining blocking source findings in the reviewed/integrated net change. The headless external handoff is J575c-only under the existing exact T6041 firmware/board gate. Per-external-processor memory publication is transactional and the internal-panel path remains unchanged. Matching fixed Linux1546 is a source dependency; no all-port/HDMI or Stage1-built-from-this qualification is claimed. Posted owner typec1 cold/replug evidence is attributed, and approved current-head build/format/Python/policy gates must pass.
Approval does not waive pending CI, dependency order, hardware-safety gates or source-versus-release qualification. No source merge is an installed-loader write or a new hardware test.
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Fixed the actual macOS Pytest failures on the contributor branch, preserving its source and ancestry: both display_init host fixtures now include the referenced T6041 define, and the two wrong-chip scenarios expect the current J616C/J614C/J575C refusal diagnostic. No behavior assertions were removed or weakened. All 65 affected display/inherited-only/handoff tests pass locally, including ASan/UBSan handoff scenarios. Fresh hosted CI remains required on d08fe72; the cancelled sibling matrix jobs from the old failure are not passes. Matching Linux1546 integration is prepared separately, and the existing typec1 coldplug/replug evidence does not qualify other ports, HDMI, or a new installed Stage1.
Remove the one extra indentation space rejected by CI's pinned clang-format 20.1.8. Non-whitespace source bytes and board ordering are unchanged. Signed-off-by: Diljit Singh <hey@djdev.me>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
The format retry reached the checker and found a real one-space alignment error at the dcp_boards_t6041_max terminator. Pushed bcabf7b as a normal child of d08fe72. It removes exactly one indentation byte from src/kboot_display_handoff.c; the board list, ordering and all non-whitespace source bytes are unchanged. The entire changed file passes the same clang-format20.1.8 and repository style used by CI; diff -w is empty and diff-check passes.
All five Pytest matrix rows, build, stage1-clang and policy had passed on the previous head. Fresh current-head CI is still required. Existing source approval and J575c typec1/no-panel hardware evidence keep their original scope; no binary/loader install, hardware experiment or default change occurred here. The paired Linux1546 stays unmerged until this source pair is ready. Aura147 and Linux1472/1473/1476 have now merged after their current checks passed.
The J575c's HDMI port runs from the display crossbar's output 4 through the LPDPTX PHY (ADT lpdptx-phy0) to an on-board DP-to-HDMI converter. The converter's pins are the J616c's (ADT dp2hdmi-gpio0: hotplug on AOP GPIO 0x24, power enables on SMC gP0f and gP1a). macOS drives this port from the internal DCP, which aura's Stage 1 puts to sleep on this board. The crossbar has no registers and the PHY's source select picks the engine, so dcpext0 (crossbar input 0) drives it here, as on the M2 Max Mac Studio (t6021-j475c.dts). Describe the PHY with the ADT's PLL calibration and its PMGR domain (DPTX_PHY, pmgr + 0x360), dcpext0's firmware memory placeholders, and dcpext0 as an HDMI-A output on that PHY. All of it is disabled: the boot loader fills dcpext0's memory and the display gate enables the chain. Tested on a J575c with the aura J575c handoff (aurora-silicon/aura#160) and a 2560x1440 monitor on HDMI: the console shows at 2560x1440@59 over four lanes, from a cold boot and after a replug, and together with a second monitor on typec1. With aura 1.0.0b8 the gate logs "display subsystem not enabled" and nothing else changes; with the tree before this commit, the HDMI stage finds no firmware memory and is skipped. Assisted-by: LLM Signed-off-by: Roberto Dantas <7robertodantas@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The J575c's HDMI port runs from the display crossbar's output 4 through the LPDPTX PHY (ADT lpdptx-phy0) to an on-board DP-to-HDMI converter. The converter's pins are the J616c's (ADT dp2hdmi-gpio0: hotplug on AOP GPIO 0x24, power enables on SMC gP0f and gP1a). macOS drives this port from the internal DCP, which aura's Stage 1 puts to sleep on this board. The crossbar has no registers and the PHY's source select picks the engine, so dcpext0 (crossbar input 0) drives it here, as on the M2 Max Mac Studio (t6021-j475c.dts). Describe the PHY with the ADT's PLL calibration and its PMGR domain (DPTX_PHY, pmgr + 0x360), dcpext0's firmware memory placeholders, and dcpext0 as an HDMI-A output on that PHY. All of it is disabled: the boot loader fills dcpext0's memory and the display gate enables the chain. Tested on a J575c with the aura J575c handoff (aurora-silicon/aura#160) and a 2560x1440 monitor on HDMI: the console shows at 2560x1440@59 over four lanes, from a cold boot and after a replug, and together with a second monitor on typec1. With aura 1.0.0b8 the gate logs "display subsystem not enabled" and nothing else changes; with the tree before this commit, the HDMI stage finds no firmware memory and is skipped. Assisted-by: LLM Signed-off-by: Roberto Dantas <7robertodantas@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Written with an AI assistant (Claude) on @7robertodantas's behalf; every hardware step was run and checked on the owner's J575c.
Display output for the J575c (Mac Studio M4 Max), which has no panel. The device tree side is aurora-silicon/linux#1546.
What happens now. iBoot lights no display on the J575c and passes a 640x1136 dummy framebuffer.
display_init()then runsdisplay_configure()with the M1 config (/arm-io/dcp): Stage 1 finds nothing connected and puts the internal DCP to sleep, and Stage 2 gets no HELLO. The T6041 handoff refuses the board (not a J616C), so nothing is handed over.But iBoot does leave the four external processors running with their firmware loaded. From Stage 1's KIS proxy (PMGR and this boot's ADT, read only):
The change.
display: leave the J575c's internal DCP as iBoot left it, as on the J614s, T8140 and T8132.kboot: T6041: hand over the J575c's external display processors without a panel: the J575c joinsdcp_soc_t6041_max, but its internal DCP is never handed over (there is no panel to drive, and the b8 Stage 1 has put it to sleep anyway). Each external processor is handed over on its own: the first one's ADT__TEXTstands in for the panel's reservation. Once one has firmware memory, the display subsystem and its DART are enabled, with iBoot's dummy framebuffer as theframebufferregion (the DRM aperture). The kernel's display gate then enables the chain. J614C/J616C behaviour is unchanged.Tested on a J575c (OS firmware 26.6.2, Stage 1 1.0.0b8, this as an ESP Stage 2 via aura-try) with the linux PR's tree:
and Linux shows the console at 2560x1440 on a monitor on typec1 (USB-C to HDMI adapter), from a cold boot and after a replug. dcpext0 is refused because the tree has no placeholders for it. A tree without the dcp/disp0 aliases is left alone, as before.
Not tested: a Stage 1 built from this (the board keeps the installer's b8), where the internal DCP stays running; the J575c path doesn't hand it over in either case.
🤖 Generated with Claude Code