Skip to content

J873: report live AFK endpoints to Linux at display handoff - #89

Draft
Acelogic wants to merge 3 commits into
mainfrom
j873-afk-handoff
Draft

Acelogic wants to merge 3 commits into
mainfrom
j873-afk-handoff

Conversation

@Acelogic

@Acelogic Acelogic commented Nov 6, 2026 •

Copy link
Copy Markdown
Member

J873 can leave DCP AFK sessions active when handing off to Linux. Starting a fresh session on an already-active endpoint asserts in the display firmware. This change reports endpoints owned by the current Aura instance in the runtime apple,afk-active-endpoints device-tree property, allowing the Linux consumer to await shutdown acknowledgment before initializing them. Companion: aurora-silicon/linux#1019.

The report comes from live endpoint state, matched to the controller through its ADT register range. It is omitted for unknown or partially initialized controllers. Existing scanout, firmware reservations, CDC upload, PMU guards and readback checks are preserved.

The common RAM-chainload path is not yet supported; keep this PR in draft. The substantive change at 94ed02ef passed an LLVM release build, full source-identity audit, a guarded CDC boot on J873, and all 23 initial smoke checks including 12,288 exact GPU pixels. The following commit changes formatting only. On that boot, Aura reused an earlier stage's framebuffer without owning its AFK objects, so it could not publish the report. DCP takeover was not attempted, and the earlier failure remains preserved.

Earlier-stage endpoint state must be propagated or otherwise validated before display takeover can be qualified. A working inherited framebuffer is not proof of endpoint state. HDMI/dual-display operation was not tested with the USB-C-only setup.

Attribution

  • New active-endpoint reporting, the companion Linux handoff guard, crash-buffer diagnosis, and J873 qualification: @Acelogic, with Codex assistance.
  • This extends the existing J873 display/AFK infrastructure from Thomas Glanzmann (@sithglan); see Aura #9, Aura #12, and the Linux consumer in linux-aurora #44. Those original implementations remain credited to their source authors.

Aurora agent policy: remove assistant instruction files that are
not copies of .github/aurora/AGENTS.md on main, and keep the copies on policy
branches identical to it.
@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for this. A few things before it can leave draft. The endpoint report is gated to T8152 (display.c display_active_afk_endpoints: chip_id != T8152; kboot.c only calls it from dt_set_display_t8152 with fixed dcp/dcpext0 paths). Restarting a live AFK endpoint is a DCP protocol issue, so could this walk whichever DCP controllers Aura owns and match DT nodes by register range on any chip? In dt_set_afk_handoff_t8152, a missing /arm-io/dcpext0 makes bail() fail the whole DT; this metadata should be optional. Please name the 0x20 endpoint base. The branch also needs a rebase onto main: please drop the gfya commit, which deletes AGENTS.md/CLAUDE.md/GEMINI.md/copilot-instructions.md that are still on main, and add Signed-off-by. Finally, a boot log where the property is actually published and consumed (the RAM-chainload case) would close the evidence gap. Happy to help test.

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