Skip to content

drm/apple, soc: apple: remove the T6030 manual external-display path (-1781) - #2432

Open
RyanTheTide wants to merge 1 commit into
integration/aurorafrom
deadweight/t6030-dcpext-manual
Open

RyanTheTide wants to merge 1 commit into
integration/aurorafrom
deadweight/t6030-dcpext-manual

Conversation

@RyanTheTide

@RyanTheTide RyanTheTide commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

Problem

The T6030 (M3 Pro/Max) display gate still carries its old "manual" external-display path next to the native one that replaced it. With apple_t6030_display.dcpext_manual=1 the gate marks the scanout node verified, and dcpext_scanout/dcpext_drm present a test pattern and register a separate DRM device. display-gate.c itself calls this "the earlier diagnostic path". Nothing we ship sets the parameter, so about 1,800 lines only run on a command line nobody uses.

Change

  • display-gate.c: remove the dcpext_manual early_param and the manual branch. The comment block now describes the remaining outcomes, native or refused.
  • Remove dcpext_scanout.[ch] and dcpext_drm.[ch]: the dcpext_pageflip parameter and the dcpext_pattern, dcpext_desktop, dcpext_status and dcpext_present sysfs files.
  • Remove their hooks in dcp.c, dcp-fabric.c, iomfb_v14_7.c and ibootep.c, and the ibootep pattern helpers only they used.
  • Drop two host-test stubs for the removed functions. Those tests build with -Werror, so the stubs would break them.

Net -1,781 lines. Every removed hook keyed on dcp->dcpext_scanout, which only the manual mode could set, so the default native path does not change. The manual path was Chris Kearney's bring-up work (98d767e, 718c6a3, e6c1062).

Kept on purpose: the explicit-start path for a dcpext that isn't native-shaped (iomfb_v14_7_external_start, sysfs dcpext_start). A real boot can still reach it if the native check fails, so it isn't provably dead.

Testing

  • Combined fleet build of all eleven deadweight code branches stacked on 0abc2f3 (DEBUG_INFO off): make Image modules dtbs RC=0, 5,909 modules, modpost reports no undefined symbols, and the only warnings are ones the baseline already had in untouched files.
  • make W=1 drivers/gpu/drm/apple/ drivers/soc/apple/ with the fleet config, before and after: RC=0, same warning set.
  • .modinfo of appledrm.o: only dcpext_pageflip (parm and parmtype) disappears.
  • 34 host tests that read the touched files (dcpext26-, dcp14-, display-sleep-merge, j613-mux-order, apple-dcp/lifecycle and others): identical exit codes before and after.
  • Setter search for dcpext_manual, dcpext_pageflip and the sysfs names: nothing in the kernel tree, fedora/alarm/nixos/aurora-platform, aura or the lab kits.

Still to run

Assisted-by: LLM (Claude:Opus 5.5)

The T6030 display gate has two modes. The default, native, runs the
external display coprocessor through the 14.7 IOMFB service in the main
DRM device. The "manual" mode, which display-gate.c calls "the earlier
diagnostic path", only runs with apple_t6030_display.dcpext_manual=1:
the gate then marks the scanout node apple,t6030-scanout-verified and
dcpext_scanout/dcpext_drm present a test pattern and register a
separate DRM device. Native mode replaced it, and nothing we ship sets
the parameter (kernel configs, distro and platform packages, aura, lab
kits).

Remove the early_param and the manual branch of the gate, the
dcpext_scanout and dcpext_drm files (the dcpext_pageflip parameter and
the dcpext_pattern, dcpext_desktop, dcpext_status and dcpext_present
sysfs files), their hooks in dcp.c, dcp-fabric.c, iomfb_v14_7.c and
ibootep.c, the ibootep pattern helpers only they used, and two host
test stubs for the removed functions. Every removed hook keyed on
dcp->dcpext_scanout, which only the manual mode could set, so the
native path is unchanged.

The manual path was written by Chris Kearney (98d767e,
718c6a3, e6c1062 and later).

Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
@RyanTheTide

Copy link
Copy Markdown
Contributor Author

No blockers.

I checked that the default path does not change. apple,t6030-scanout-verified was only ever written by the manual branch of the gate (display-gate.c, old gate_ext_apply). No DT in the tree carries it. dcpext_scanout_register() returned -ENODEV without it, so dcp->dcpext_scanout stayed NULL. With it NULL, every removed hook (terminal, requested, invalidate, link_restored, fault) was a no-op. The gate's native case is the same as before: scanout_ok was already required, and piodma plus the display DART add run in the same order. The PR still merges cleanly onto current integration/aurora, which is 398 commits ahead of its base. In the merged tree nothing refers to dcpext_scanout, dcpext_drm, dcpext_manual, dcpext_pageflip or the four sysfs names. The removal also takes out the firmware-UUID string compare that gated dcpext_pageflip.

Two small leftovers worth folding in:

  1. drivers/gpu/drm/apple/dcp-fabric-core.h:55: pipeline->terminal is now never written. Only dcp_fabric_snapshot_port() set it, and the struct comes from devm_kzalloc, so it is always false. That leaves two dead branches, dcp-fabric-core.c:379 (the -ESHUTDOWN return) and :444, plus the cases in dcp-fabric-test.c:456 and :1585-1589 that test a state the driver can no longer reach. Either remove the field with its tests or say in the commit message why it stays.
  2. drivers/gpu/drm/apple/dcp-internal.h:574: "Clear for the manual diagnostic path (explicit start)" now names a path this patch removes. Say instead that it is clear when the dcpext is not native-shaped and uses explicit start (dcpext_start).

HW row: J514s or J516s (T6030) on this head with an external display on built-in HDMI and on USB-C. Pass if the gate logs "enabled dcpext%u ... started by the display driver on first use", the mode and picture match the pre-change kernel, 3 hotplug cycles and one s2idle cycle each work, with 0 AVC, 0 failed units and no new dcp/dcpext kernel errors. The code is safe to leave draft without this row, but the row should be posted before merge if a T6030 owner can run it.

Assisted-by: LLM (Claude:Opus 5.5)

@RyanTheTide
RyanTheTide marked this pull request as ready for review November 22, 2026 10:11
@RyanTheTide

Copy link
Copy Markdown
Contributor Author

Re-check at 9f5ce93 against current integration/aurora and the open PRs. The default native path is still unchanged, and nothing in the tree, the distro repos or the boot loader sets apple_t6030_display.dcpext_manual, dcpext_pageflip or the four sysfs files. The two leftovers from my earlier comment (pipeline->terminal and the dcp-internal.h:556 comment) are still open. One new problem:

Assisted-by: LLM (Claude:Opus 5.5)

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.

1 participant