Skip to content

Add normal J514M/T6034 graphics handoff - #24

Open
chriskearney wants to merge 3 commits into
chris/j514c-bringup-20261105-7a53from
chris/j514m-t6034-20261105
Open

chriskearney wants to merge 3 commits into
chris/j514c-bringup-20261105-7a53from
chris/j514m-t6034-20261105

Conversation

@chriskearney

@chriskearney chriskearney commented Nov 5, 2026 •

Copy link
Copy Markdown

Add an exact J514M/T6034 handoff for normal native-graphics boot without enable flags. Keep explicit owner opt-outs and the firmware, resource, powered-identity and display handoff checks. This draft is stacked on #23 and pairs with the T6034 kernel work.

The loader records this machine's powered GPU identity, image metadata, six-channel MCC facts and separate DCP/PMP handoff namespaces. Its boot-clock table reports 400 MHz; that metadata does not independently establish the programmed display rate. The measured GPU has four ten-slot clusters with 30 enabled cores. Other T6034 boards are not admitted by the J514M path. Normal builds include these defaults; measurement-only operation remains a distinct build choice.

Validation at 9488b7b7733190a5f1ceeb34d35120a090e420c6:

  • 170 host tests passed; eight optional unavailable checks skipped. AArch64 compilation and pinned package build passed.
  • The powered-identity diagnostic passed. The unchanged packaged normal loader subsequently booted the corrected T6034 diagnostic kernels, with the exact DCP/PMP UUIDs, handoff markers and transferred tuning values.
  • The latest diagnostic reached native apple-drm, panel timing and completed swaps. The kernel measured one consistent canonical GPU firmware identity, then refused the deliberately unpinned image. The diagnostic still failed because its checker matched the refusal message as startup; its raw failure is preserved. Original kernel and loader were restored and verified.
  • Native GPU rendering and automatic default-boot qualification remain pending. The kernel PR records the unresolved display-clock reporting and power/reference limitations. No full rendering or sustained-load pass is claimed.

No raw firmware image is included. Qualification metadata remains hardware-tested=false until the complete native boot/render path passes. Human DCO certification is pending. No merge or package-channel release is requested.

Scope a normal-boot diagnostic to the J514M and T6034 pair. Reuse the bounded G15C identity power closure with exact restoration, publish SoC-specific identity and firmware-header metadata, and export display clock and memory-controller facts. Leave the GPU and mailbox disabled so the stock kernel remains unbound. Honor explicit owner opt-outs and lost configuration.

Test normal boot, mismatched board/SoC, opt-outs, invalid SGX windows, failed power-on, faulted reads and failed restoration with the production probe and real libfdt. This measurement build is an intermediate diagnostic, not completed native graphics support.

Assisted-by: Codex:GPT-6
Scope normal graphics defaults to the J514M and T6034 pair, with independent owner disables, master disable and lost-config handling. Keep T6031 and other T6034 boards on their separate paths.

The fresh J514M ADT supplies the checked internal DCP and DART addresses and the existing 14.7 DCP/PMP image UUIDs. Add a distinct pinned display handoff, PMP value namespace and firmware UUID list; retain all bounded mapping and power-state checks before publishing the handoff. Linux separately validates the complete GPU firmware identity.

Normal-build and owner-choice host tests pass for J514M while wrong boards and SoCs remain excluded. The display clock is independently measured as 400 MHz and represented by the kernel device tree.

Assisted-by: Codex:GPT-6
Add the exact J514M T6034 GPU profile to the shared G15C setup. Its powered identity and firmware family are measured first with exact power restoration; the setup consumes that identity without additional GPU hardware access. Preserve per-board defaults, owner opt-outs, lost-config handling, DT/ADT windows, UAT reservations and family checks. Export an explicit T6034 list of stand-in power properties.

Use per-boot G15C state names so neither T6031 nor T6034 can be mistaken for the other profile. The kernel must independently match the full firmware digest before startup. Host production-body and default/opt-out suites pass 170 tests with eight unavailable optional cases; Stage2 compiles without a board enable build flag.

Assisted-by: Codex:GPT-6
@scottjones

Copy link
Copy Markdown
Collaborator

Hardware results from our Mac lab on a J514M (14" M3 Max, 30-core GPU), m1n1 built from 9488b7b:

  • stage2 v1.6.1 at 9488b7b in the boot log.
  • Defaults, no flags (j1107-053241-ccf1, kernel from linux-aurora#372): Vulkan 1389 dispatches, 0 bad. GL 5 minutes, 2779 draws, 0 bad. 0 DART faults.
  • With asahi: admit the measured 30-core T6034 G15C contract mesa#32 staged (j1107-054455-d032), Hyprland ran at eDP-1 3024x1964@120.
  • The Mac's boot.bin was checked as stock before, swapped to the candidate, and restored to stock after, with the restore verified.

Also in the combined trial (m1n1 #23 + #24 + #25, linux-aurora AsahiLinux#363 + AsahiLinux#372 + AsahiLinux#388 + AsahiLinux#400, stage2 v1.6.1-473-gd2b8d44b): this J514M ran Hyprland at 3024x1964@120 with Vulkan 694/0 bad. 0 DART faults across the four Macs, no regressions vs stock apart from audio.microphone-signal on the J514M, which fails on stock too.

@greptile-apps

greptile-apps Bot commented Nov 7, 2026 •

Copy link
Copy Markdown

RetriggerView in GreptileConfidence Score: 4/5

[Medium impact] The PR changes normal graphics handoff behavior for one exact Mac model.

Fix the ignored explicit GPU choice before merging.

Findings

  1. P1 Explicit GPU enable is ignored ▶
  2. P2 J516M test now falsely fails ▶

Summary

Adds normal J514M/T6034 graphics handoff:

  • J514M boots get separate T6034 GPU and display handoff choices.
  • The J514M GPU handoff uses a measured T6034 identity.
  • T6034 firmware and hardware facts use separate handoff names.

An explicit GPU enable is still blocked when automatic defaults are off. The existing J516M host-test assertion also needs updating.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[J514M and T6034 checks] --> B[Apply independent defaults]
    B --> C[Display firmware and resource checks]
    B --> D[GPU owner choice]
    D --> E[Powered identity and firmware facts]
    E --> F[GPU resource and identity checks]
    F --> G[Enable GPU or leave disabled]
    C --> H[Publish display handoff or keep framebuffer]
Loading

Reviews (1) · Last reviewed commit: "kboot: populate the measured T6034 G15C ..." · Reviewed by Greptile

Comment thread src/kboot_gpu.c
Comment on lines +2734 to +2736
const char *opt = chosen < 0 ? NULL : fdt_getprop(dt, chosen, "asahi,m3-defaults", &len);
if (opt && (len != 2 || opt[0] != '1' || opt[1]))
return 0;

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Explicit GPU enable is ignored

On J514M, chosen.asahi,m3-defaults=0 plus chosen.asahi,t6034-gpu=1 leaves the GPU disabled despite the explicit enable. dt_set_m3_defaults() preserves explicit choices, but this extra master-switch check skips identity measurement and firmware checks. The later setup sees g15_g15c_fw == NULL and disables the GPU.

Remove this extra check and add a test for this combination.

Comment thread src/kboot_gpu.c
Comment on lines +3189 to +3190
case T6034:
return dt_set_gpu_t6034(dt);

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 J516M test now falsely fails

The existing t6034 host-test scenario still requires the message ADT: GPU: unsupported chip!. With this new dispatch, dt_set_gpu_t6034() correctly returns quietly for J516M, so that assertion always fails. The default host-test run includes this scenario.

Update sc_t6034() to check that J516M stays untouched without requiring the old message.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants