Silent system hang when userspace probes SD root (`mmcblk1`) via `fe320000.mmc` / `dw_mmc_rockchip` (triggered by fwupd emmc plugin)

[SMinin]

Summary

On ROCK 4C+ running Radxa kernel 6.1.115-debug-r2+ with root on SD (mmcblk1, controller fe320000.mmc), userspace firmware enumeration (fwupd emmc plugin coldplug) leads to a silent full-system hang. No kernel panic or oops is recorded; Synopsys DesignWare hardware watchdog resets the board after ~30 s.

This may indicate an MMC driver / controller bug exposed when fwupd scans the SD block device, but the hang affects all I/O to root and is not recoverable without reset.

Environment

Item	Value
Board	Radxa ROCK 4C+ (radxa,rock-4c-plus)
Kernel	6.1.115-debug-r2+ (#3, Jul 6 2026)
Root	SD mmcblk1p3 ext4
MMC	fe320000.mmc, driver dw_mmc_rockchip
SD device	type SD, ID_PATH=platform-fe320000.mmc

Trigger (userspace)

Any fwupd engine load with udev coldplug, e.g. fwupdmgr refresh, fwupdtool get-devices, or fwupd-refresh.timer.

Last kernel log line before freeze:

fwupd[PID]: FuEngine  failed to add device .../mmcblk1: failed get ffu_capable

Expected

Kernel remains responsive; MMC errors should be reported and I/O should not wedge the entire controller.

Actual

Silent hang; console-ramoops stops; HW WDT reboot ~30 s later. softlockup_panic / hung_task_panic do not trigger in time.

Reproduction

1. Debian 12 on SD root, install fwupd 1.8.12.
2. systemctl start fwupd-refresh.service (serial console recommended).
3. Observe hang after emmc/mmcblk1 log line.

Alternative isolation: capture MMC trace with radxa-mmc-debug.service / dyndbg on dw_mmc_rockchip — prior attempt showed logs may not flush before total freeze.

Request

• Investigate dw_mmc_rockchip + SD slot behavior when userspace enumerates mmcblk1 during fwupd coldplug.
• Determine if kernel should harden against controller lockup on SD root systems.
• Cross-reference fwupd issue emmc plugin coldplug on SD card root device (mmcblk1) causes full system hang on Rockchip (Radxa ROCK 4C+) fwupd/fwupd#10625