SM8750: restore dock DisplayPort output broken by the 7.1 bump

[aanze]

Summary

• What is the goal of this PR? Restore USB-C dock / KVM DisplayPort output on the AYN Odin 3 (SM8750), which has been dead since the 7.0.11 -> 7.1.3 kernel bump (6166516). Plugging a dock gives no video at all — the connector never leaves disconnected and the display shows no sign of detection.

The 7.1 merge window added a "skip duplicates" filter to typec_mux_match() / typec_switch_match() in drivers/usb/typec/mux.c and started passing the output array in as the match data. struct device is the first member of struct typec_mux_dev / struct typec_switch_dev, so container_of() on a NULL device yields NULL. When class_find_device() finds nothing — the normal case while the mux driver has not probed yet, here the wcd939x-usbss retimer on i2c — the dedup loop compares that NULL against an array slot that is also NULL, matches, and returns NULL instead of ERR_PTR(-EPROBE_DEFER). The connector is then built without its mux, permanently, and the DP lanes and the AUX/SBU pair are never routed.

The two dedup arrays are also uninitialised stack, yet every slot is read before fwnode_connection_find_matches() has filled them, so a valid mux can be dropped by a chance comparison against stack garbage.

Kernel patch 0517 bails out with -EPROBE_DEFER before the dedup loop when no device was found, and zero-initialises both arrays. This is not Odin 3 specific: any platform whose Type-C mux probes after the connector is affected.

Testing

• How was this tested? Built a ROCKNIX image for SM8750 and flashed an AYN Odin 3, with a USB-C KVM driving a DisplayPort monitor. A/B against release images on both sides of the bump (7.0.11 works, 7.1.3 does not), same dock and same instrumentation.
• Test results:

Before (7.1.3) — the firmware enters DP alt mode and raises HPD, the driver decodes it correctly and notifies DRM, then the very first DPCD read times out, so the link never trains:

pmic_glink_altmode: NOTIFY svid=0xff01 mux_ctrl=3 pin_assignment=4 hpd_state=1
aux_hpd_bridge:     HPD notify status=1          (connected)
[drm:msm_dp_display_host_phy_init] core_init=1 phy_init=0
[drm:drm_dp_dpcd_probe] dpu_dp_aux: 0x00102 AUX -> (ret=-110)
[drm:msm_dp_bridge_detect] link_ready = false

After (same image + this patch):

DP status : connected
DP enabled: enabled
EDID      : 256 bytes
modes     : 3840x2160  4096x2160
AUX -110  : 0

Picture on the external display, confirmed on device.

Additional Context

• An AUX -> (ret=-110) timeout while every software layer reports success is the signature of the lanes never being routed — worth remembering for similar reports.
• Reached by elimination: every other file of the DP path is byte-identical between 7.0.11 and 7.1.3 (dp_drm.c, dp_display.c, aux-hpd-bridge.c, pmic_glink_altmode.c, wcd939x-usbss.c, phy-qcom-qmp-combo.c, phy-core.c, dispcc-sm8750.c, drm_bridge.c, drm_bridge_connector.c), as is the live device tree, so mux.c was the only behavioural change left.
• The fix belongs upstream in the kernel; carrying it here fixes ROCKNIX users now. Other users have reported no dock video on ROCKNIX, which this would explain.
• Freeze is active (last 7 days of the month) — this is a bugfix and needs the bugfix label, which I do not have permission to set.

---

AI Usage

Did you use AI tools to help write this code? YES