usb4: add guarded X1E PHY groundwork and top-port retimer test

[ooaklee]

Summary

Integrate the public Qualcomm X1E USB43DP PHY v4 groundwork and add a production-safe, explicitly selected Surface Pro 11 top-port retimer qualification path at 7.2.0-jg-0sp11v20.

The scoped retimer qualification now passes, but this PR does not claim complete USB4 support. Linux had no active Qualcomm host-router PHY consumer during the test, and no USB4 domain/router, dock USB/Thunderbolt enumeration, or USB3/PCIe/DisplayPort tunnel appeared. Both production Denali PS8830 nodes remain guarded by parade,disable-usb4.

• Base: sp11/integration-7.2.x, including the merged pen/touch integration
• Head: 70ddec100fe9
• Companion build, rollback, and runtime evidence: linux-surface-pro-11-oe PR 56 at support head 3dd3b4c

Production safety and experimental selection

• The normal Denali OLED DTB retains parade,disable-usb4 on both PS8830 retimers.
• The experimental DTS inherits the production tree and deletes the guard only under i2c7, mapped to the physical top port. The bottom-port guard remains.
• The experimental DTB is absent from dtb-y and the normal dtbs-list.
• Packaging filters the experimental filename from normal Stubble auto-selection and fails closed if the invariant changes.
• The normal /boot/vmlinuz-* remains guarded.
• A second Stubble EFI image containing only the experimental DTB is packaged at /usr/lib/linux-image-<ABI>/sp11-usb4-top-experimental.efi; the package does not install it under /boot or add a GRUB entry.

SP11 boots the DTB embedded in the Stubble EFI image, so a loose GRUB devicetree line is not an override. The experiment was selected by copying the alternate EFI to a distinct temporary filename and transiently editing only the normal v20 GRUB linux path for one cold boot.

Implementation

Public Qualcomm USB43DP PHY v4 series

The provenance-preserving public series remains split into five commits:

• 076072b40299 — binding and USB4 PHY index;
• b343d089a705 — TBT PHY mode and USB4/TBT3 submodes;
• d2ddf9dce2ef — USB4 PHY instance and lifecycle;
• c6d1b64af29f — Hamoa USB4/TBT3 electrical data; and
• 8564ca38642c — PHY-to-router clocks.

5b5f1d124b7a adds transactional shared-PHY ownership, rollback, and runtime-PM hardening.

Retimer qualification increment

• 75617434c1e6 rejects malformed Enter_USB state and adds one-shot PS8830/QMP diagnostics.
• a57d6d807dbc adds the top-port-only experimental DTS.
• 5321047ea5e2 records the experiment in the unchanged sp11v20 package version.
• 70ddec100fe9 isolates the experiment in its alternate Stubble image.

Clean package result

An Ubuntu 26.04 binary-indep binary-qcom-x1e build completed from exact head 70ddec100fe953712c309067fe2db4d8207facc6 with Local patches: none. All four packages report version 7.2.0-jg-0sp11v20; strict verification of the four packages and two provenance manifests passed.

Packaged-image inspection confirmed:

• normal ARM64 PE: 38 .dtbauto sections and no experimental model marker;
• alternate ARM64 PE: exactly one .dtbauto with the exact experimental model;
• production/experimental guard counts: 2/1;
• normal and alternate .linux payloads are identical; only embedded DTBs differ;
• no experimental package path under /boot; and
• no loose experimental DTB in the package.

Exact package and checksum hashes are recorded in companion ADR0068.

Controlled X1E OLED runtime result

The normal v20 control boot had the ordinary OLED model and two live guards. Touch, pen/pressure/barrel button, and a direct USB3 device passed; direct DisplayPort Alt Mode was not retested because suitable hardware was unavailable. A guarded TS4 attach encountered the intended DT policy rejection.

After a cold boot of the transient alternate image, the exact experimental model and one remaining bottom-port guard were verified before attachment. One top-port attach of the matched CalDigit TS4 and bundled 40 Gb/s cable produced:

ps883x_retimer 5-0008: USB4 mode accepted by retimer; host-router state not established
qcom-qmp-combo-phy fda000.phy: USB4/TBT mux request ignored: no active host-router PHY consumer

The first line is emitted only after ps883x_configure() returns success, so the driver completed its USB4 configuration writes without error. This does not prove the silicon trained a USB4 link or negotiated 40 Gb/s. The second means that a USB4 PHY object existed but had no active host-router PHY consumer at the mux request. It does not establish why initialization was absent or prove that one consumer is the sole remaining blocker.

UCSI reported the Type-C/PD partner, but there was no dock USB/PCI enumeration, DRM connector, Thunderbolt device, or USB4 domain. This validates removal of the retimer-side policy rejection for the isolated top-port experiment; it does not validate a USB4 link, working hub, or tunnel.

A subsequent cold boot of the normal unedited v20 image restored the ordinary OLED model and both production guards.

Windows hardware oracle

The reviewed, redacted same-device Windows result proves that the exact TS4, bundled passive 0.8 m 40 Gb/s cable, and physical top port can form a USB4 connection with USB3, PCIe, and DisplayPort tunnels under Windows. It is a matched hardware success oracle, not Linux source or host-router implementation evidence.

Raw Windows and Linux captures remain private. They do not provide a reviewable host-router implementation, MMIO/resource map, firmware/ring protocol, or decoded PS8830 sideband sequence.

Validation

Gate	Result
git diff --check	Pass
strict scripts/checkpatch.pl	Pass: 0 errors, 0 warnings, 0 checks
focused W=1 build	Pass: ps883x.ko and phy-qcom-qmp-combo.o
X1E production, X1E experimental, and X1P Denali DTBs	Pass
full ARM64 dtbs	Pass; experimental filename absent from normal dtbs-list
production/experimental compiled guard counts	Pass: 2/1
alternate Stubble construction and package isolation	Pass
full same-version qcom-x1e package rebuild	Pass at exact head 70ddec100fe9
guarded control boot and guarded rollback	Pass
top-port PS8830 configuration path	Pass for the bounded retimer-only scope
USB4 link/domain/router and dock tunnels	Not established
direct DisplayPort regression	Not run; suitable hardware unavailable

Remaining boundary

• No production router node, firmware loader, sideband programming, or USB4 PHY consumer is added here.
• The observed inactive-consumer state is a blocker, not proof of its cause or the only remaining blocker.
• Direct DisplayPort regression, both-port/orientation coverage, repeated cold boot, suspend/resume, detach recovery, and security/IOMMU qualification remain required before any production guard or runtime-PM policy change.
• No raw MMIO, debugfs register writes, driver rebinding, firmware-memory dump, or user-triggered power-domain toggle is part of this work.

This remains downstream integration groundwork, not complete in-tree SP11 USB4 support.