Skip to content

drm/sysfb: report eDP for framebuffer panels - #1688

Draft
joshuaswarren wants to merge 2 commits into
aurora-silicon:integration/aurorafrom
joshuaswarren:w71/edp-connector
Draft

joshuaswarren wants to merge 2 commits into
aurora-silicon:integration/aurorafrom
joshuaswarren:w71/edp-connector

Conversation

@joshuaswarren

@joshuaswarren joshuaswarren commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

When the boot framebuffer points to an internal panel, simpledrm reports an eDP connector. A framebuffer without a resolved panel phandle keeps the Unknown connector type.

The J714c framebuffer links to the panel with the existing panel binding. The loader still enables the framebuffer. The 302 by 196 mm size comes from J714s and is unconfirmed on J714c.

The two-commit candidate has the same file contents as CI-green head e65fe2640c72. It drops the historical vendor-property commits.

The no-DCP handoff row is still open: card0-eDP-1, no handle-lid-switch inhibitor, and suspend and resume in the journal. The M1 Pro and M2 regression rows follow that check.

@joshuaswarren joshuaswarren changed the title [Joshua's agent] drm/sysfb: eDP connector for the built-in panel, K114/K116 framebuffer property [Joshua's agent] drm/sysfb: simpledrm: eDP connector for an Apple built-in panel, and its binding (draft) Nov 15, 2026
@joshuaswarren

Copy link
Copy Markdown
Contributor Author

GREEN on 667f644
Model: zai/glm-5.3

Checked the split against the diff and the body against both commits. The PR touches exactly the three claimed files - simpledrm.c, the J714c dts hunk carried inside commit 1, and the simple-framebuffer binding - with no K114/K116 dtsi present, matching the split-out claim. Commit 1's provenance is stated the right way: carried unchanged from omacom/linux-aurora (cherry-pick sha recorded), the J714c hunk rebased onto the current framebuffer node and disclosed, no Signed-off-by from ed claimed, and the J714c test result attributed to ed rather than restated as this PR's own validation. The Validated-on section keeps the same discipline: build-only on the M1 at this head, the CHECK_DTBS counts attributed to the earlier body on another machine and explicitly not repeated, the K114c eDP observation attributed to an earlier tree, and the lid test named as not run with its slot plan. The binding commit closes the CHECK_DTBS warning the property would otherwise raise - the right order given #2009 depends on it.

Model: zai/glm-5.3 (live from lane_models.py this turn).

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

GREEN on 667f644
Model: anthropic/claude-sonnet

Independent review. No blocking items. I did not boot anything, so the lid behaviour is not something I can confirm.

What I checked:

  • Files. The PR changes exactly three files: Documentation/devicetree/bindings/display/simple-framebuffer.yaml, arch/arm64/boot/dts/apple/t6050-j714c.dts and drivers/gpu/drm/sysfb/simpledrm.c. It does not touch t6050-k114-k116.dtsi.
  • Build. drivers/gpu/drm/sysfb/simpledrm.o builds with W=1 and no warnings at this head. I used an arm64 defconfig with CONFIG_DRM_SIMPLEDRM=y, gcc 12.2 and the aarch64 cross toolchain, on a depth-limited checkout of the PR head.
  • Authorship. The carried commit keeps ed <edimoldovan@gmail.com> as author, with (cherry picked from commit 9dee12b1266c4…) and a Signed-off-by from the submitter. The PR text says the same.
  • Binding. dt-doc-validate accepts the edited simple-framebuffer.yaml. I built a minimal framebuffer node with apple,internal-panel and validated it against the schema with and without this change. Without it, dt-validate reports 'apple,internal-panel' does not match any of the regexes. With it, there is no message. A disabled node is checked too, which matches the PR text.
  • Driver. of_property_read_bool() on a null of_node returns false, so non-device-tree users of simpledrm still get the Unknown connector as before.

What I could not run: a full dt_binding_check or CHECK_DTBS. On this tree the validator stops on a type conflict for apple,tunable between pci/apple,pcie.yaml and phy/ti,cd2e22.yaml, before it reaches the framebuffer node. That conflict is not in this PR. The node-level check above is what I could run.

Non-blocking notes:

  1. The binding commit message describes the wrong tree. It says the K114c, K114s, K116c and K116s framebuffer nodes carry the property. At this head only the J714c node does. The four K114/K116 nodes get it in arm64: dts: apple: t6050-k114-k116: report the built-in panel as eDP (draft, unbooted) #2009. Say "J714c" here, or say that the others follow in arm64: dts: apple: t6050-k114-k116: report the built-in panel as eDP (draft, unbooted) #2009.
  2. No Signed-off-by from the original author. The PR text says so. If Aurora means to send this upstream, ed's sign-off is needed. The carried message also keeps a Co-Authored-By trailer for an AI model, unchanged from the original commit.
  3. Boot testing is not done. The PR says so: the J714c result in the commit message is the original author's, and the K114c lid test is still to come.

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

GREEN on 4b92610
Model: zai/glm-5.3

Tree identity confirmed independently: both heads resolve to tree 4b96fb09bcfcb9438a4d31939cb6175439610dc1, the compare shows the binding commit rewritten with only the yaml in the reword, so nothing but the commit message changed. The reworded message and body attribute the evidence correctly for the split: the J714c node carries apple,internal-panel from the previous commit, the K114c/K114s/K116c/K116s nodes get it in #2009 which lands after, and the CHECK_DTBS numbers are attributed to an earlier form on another machine and not repeated at this head. My previous verdict on the code carries unchanged.

Model: zai/glm-5.3 (live from lane_models.py this turn).

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

GREEN on 4b92610

The tree is unchanged from my review of 667f644. Both heads have tree 4b96fb09bcfcb9438a4d31939cb6175439610dc1 and the same parent, and git diff between them is empty.

The reworded binding commit message is accurate. It says only the J714c node carries the property in the previous commit, and that the K114c, K114s, K116c and K116s nodes get it in a later device-tree change that fails CHECK_DTBS without this binding. I confirmed that last point earlier: a framebuffer node with apple,internal-panel is rejected by the old schema and accepted by the new one. The body's CHECK_DTBS paragraph now says the same, names #2009 and says the check was not repeated at this head.

Model: anthropic/claude-sonnet

joshuaswarren added a commit to joshuaswarren/linux that referenced this pull request Nov 18, 2026
The K114c and K116c panel is a simpledrm framebuffer until DCP drives it,
and its connector shows up as Unknown-1. Omarchy treats a connector that
is not eDP, LVDS or DSI as an external monitor and holds a block inhibitor
on handle-lid-switch, so closing the lid never suspends.

aurora-silicon/linux aurora-silicon#1688 makes simpledrm report eDP for a framebuffer node that
carries apple,internal-panel. Put the property on the shared K114/K116
framebuffer node, as the J714c node already has it.

Compile check only: the K114c dtb builds and dtc shows the property on
framebuffer@0; simpledrm.o builds. Not booted on a K114c yet.

Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for splitting this out, and to ed for the original. One design point.

The binding already has a panel phandle (simple-framebuffer.yaml:126-129), and simpledrm already follows it for width-mm/height-mm (simpledrm.c:191-219). Most Apple laptop trees already set panel = <&panel> on framebuffer0: j313, j314/j316, j414/j416 and the M3 boards.

Could simpledrm report eDP when the framebuffer node has a panel phandle, instead of using a new vendor property? Then J714c (and K114/K116 in #2009) would get a panel node with width-mm/height-mm. The K114c panel node in #2183 could be that target. This keeps the generic binding untouched and makes it the default on every laptop that describes its panel. A panel link is also the strongest argument if this goes to dri-devel.

ed's Signed-off-by on the first commit is needed before merge.

Row: K114c or J714c booted without the DCP handoff. /sys/class/drm shows card0-eDP-1, systemd-inhibit --list shows no handle-lid-switch block, and closing the lid suspends (journal shows suspend entry and exit).

After that, we can run a no-regression row on our M1 Pro.

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

[Joshua's agent]

The design point is right and we will rework the PR along those lines: report eDP when the framebuffer node carries the panel phandle, drop the apple,internal-panel binding addition, and set panel = <&panel> on the J714c framebuffer node. That reuses the existing binding, keeps the generic schema untouched, and makes the eDP connector the default on every laptop that describes its panel instead of one property at a time.

One dependency shapes the order: the phandle needs a panel node to point at. The panel nodes are still open (#2183 for the K114c panel, the K114/K116 set in #2009), so the rework stacks behind them. The driver change itself (phandle implies eDP) is independent and lands with the rework.

ed's Signed-off-by will be on the carried commit before merge - we will ask him to confirm it (here or in the team chat), not add it on his behalf.

The requested row (booted without the DCP handoff: card0-eDP-1 present, no handle-lid-switch inhibitor, lid close suspends and resumes) is queued on a 14-inch M5 Max; we will post the journal lines here when it runs.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks for taking the panel phandle approach. Nothing more from us until the rework is pushed. The head is still 4b92610 with apple,internal-panel. What's missing:

  • the rework: simpledrm reports eDP when the framebuffer node has a panel phandle (simpledrm.c:868), the binding addition is dropped, and the J714c framebuffer gets panel = <&panel> plus a panel node with width-mm/height-mm,
  • ed's Signed-off-by on the carried commit (from ed himself, as you said),
  • the row: K114c or J714c booted without the DCP handoff, showing card0-eDP-1, no handle-lid-switch inhibitor, and the lid suspending and resuming in the journal.

Once that is up, we will run a no-regression row on our M1 Pro, which already has panel = <&panel> on its framebuffer.

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

Note

🤖 OpenAI Codex responding on behalf of joshuaswarren

e65fe26 makes simpledrm report eDP when the existing panel phandle resolves. Firmware framebuffers without a panel keep the Unknown connector type. The PR no longer adds apple,internal-panel to the binding.

The J714c DTB resolves /chosen/framebuffer@0 to /panel at 302 by 196 mm. I used the J714s dimensions. The framebuffer stays disabled until the loader sets it up. The framebuffer schema reported no diagnostics, and simpledrm.o cross-compiled.

The no-DCP eDP and lid row, and ed's own sign-off, remain open. The M1 Pro regression row can follow when the same head is ready.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks, e65fe26 is the shape we hoped for. simpledrm reports eDP when the panel phandle resolves, the binding is untouched, and dt-checks and host-tests are green.

Before review:

  • History: the branch still adds apple,internal-panel (1006491, 4b92610) and then removes it (b570505). Please squash it to one simpledrm commit and one J714c dts commit, and update the title and body. ed's code is no longer in the diff, so Suggested-by: ed on the simpledrm commit replaces the need for his Signed-off-by.
  • The J714c size (302 x 196 mm) is the J714s value. Please confirm it in the row that is still owed: K114c or J714c without the DCP handoff, card0-eDP-1, no handle-lid-switch inhibitor, and lid suspend/resume in the journal.

This changes the connector type on every Apple laptop tree with panel = <&panel>, so we'll then run no-regression rows on our M1 Pro and M2.

Report eDP when the framebuffer node points to a panel. Leave the
connector type unchanged when the panel phandle does not resolve.

Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
@joshuaswarren joshuaswarren changed the title [Joshua's agent] drm/sysfb: simpledrm: eDP connector for an Apple built-in panel, and its binding (draft) drm/sysfb: report eDP for framebuffer panels Nov 20, 2026
@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks, 8da6768 and 02d7b6c are the clean two-commit history we asked for, and the title and body match. Two small things:

  • Please add Suggested-by: ed to 8da6768. The idea is his, and that tag replaces the need for his Signed-off-by.
  • t6050-j714c.dts:46-51 adds the panel as a top-level node. On J714s the panel is a child of &dcp (t6050-j714s-gpu-panel.dtsi:80-92), and the arm64: dts: apple: t6050-k114-k116: report the built-in panel as eDP (draft, unbooted) #2009 candidate also puts its panel under &dcp. Please do the same here with a panel: label, so the DCP handoff later reuses this node and does not add a second one.

Still owed, as you say: the no-DCP-handoff row with card0-eDP-1, no handle-lid-switch inhibitor and lid suspend/resume in the journal, which also confirms the 302 x 196 mm size. Our M1 Pro and M2 no-regression rows follow.

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

Note

🤖 OpenAI Codex responding on behalf of joshuaswarren

The J714c panel now sits under &dcp, and the framebuffer points to that single node. The compiled DTB still reports 302 by 196 mm. The framebuffer stays disabled and DCP status is unchanged.

I'll add the Suggested-by: trailer once Joshua confirms. The PR stays draft until the no-handoff eDP, lid inhibitor and suspend/resume checks run on hardware.

@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks, b1cb83a puts the J714c panel under &dcp with the panel: label and the framebuffer points at that one node, which is what we asked for.

One small thing: b1cb83a only moves the node that 02d7b6c added. Please fold it into 02d7b6c when you add the Suggested-by: trailer to 8da6768, so the series stays one simpledrm commit and one J714c dts commit.

Nothing else from us at this head. As you say, the no-handoff hardware row is still owed (card0-eDP-1, no handle-lid-switch inhibitor, lid suspend/resume in the journal, 302 x 196 mm confirmed), and dt-checks/host-tests are still running. Once you mark it ready we will run our M1 Pro and M2 no-regression rows.

Link the boot framebuffer to the panel under DCP. Keep the panel
in one place so the boot framebuffer and a later DCP handoff can
use the same description. The dimensions come from J714s.

Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
@joshuaswarren

Copy link
Copy Markdown
Contributor Author

Note

🤖 OpenAI Codex responding on behalf of joshuaswarren

971e41fb1099 folds the J714c panel move into the DTS commit. The simpledrm commit stays at 8da67681, and the final tree is identical to b1cb83a1. The compiled J714c framebuffer phandle resolves to the 302 × 196 mm panel under DCP.

I’ll add the Suggested-by trailer once Joshua confirms the attribution. The no-handoff hardware check is still open.

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