uboot: inject per-board ETH PHY reset GPIO from thingino.json

[WLTB-Gino]

The problem

The 2026.07 dwmac_ingenic driver takes the wired-Ethernet PHY reset line from the gmac DT node's ingenic,reset-gpio property. The per-SoC .dtsi can only carry the ISVP reference default (PB28, active-low) - right for the reference board and wrong for most cameras, and there was no per-board override mechanism. Boards whose reset pin differs (e.g. the Jooan Q3R, GPIO 64 / GPC0) never had their line driven: the reset only happens when the property is present, so the PHY comes out of a prior boot in whatever state the last user left it.

Meanwhile the board carried its pin via the legacy CONFIG_GPIO_PHY_RESET make opts + local patch - a mechanism that is dead on the 2026.07 tree (CONFIG_GPIO_PHY_RESET doesn't exist there), so the knob silently stopped working when U-Boot moved to 2026.07.

Boards with no reset line wired at all are NOT supposed to get a random GPIO driven: the upstream driver added an MDIO BMCR soft-reset fallback exactly for them, and a GPIO pointing at an unrelated pin would actively break them. So the injector only acts when the board declares the key.

The change

inject-uboot-eth-dt.sh (in the pattern of inject-uboot-audio-dt.sh): when the board's thingino.json declares gpio.eth_phy_reset (object {pin, active_low}, or a bare int = active-low), rewrite the gmac node's ingenic,reset-gpio in the board's U-Boot leaf .dts at build time. Boards without the key (or without a GMAC node) are left untouched, so every other board is unaffected.

Also:

• the legacy CONFIG_GPIO_PHY_RESET make opts are gated to the 2013.07 tree they belong to (no behavior change - they were already no-ops on 2026.07)
• the kconfig symbols are depends on BR2_THINGINO_UBOOT_VERSION_2013_07, so a future board cannot pick the dead knobs again
• jooan_q3r declares gpio.eth_phy_reset {pin: 64, active_low: true} and drops the dead defconfig lines + local patch

Testing

• shellcheck + shfmt clean; pre-commit hooks pass
• injector exercised standalone on a synthetic leaf: object form → &gpc 0 1, bare int → &gpb 28 1, no key → no-op, no gmac node → clean exit
• not device-tested: needs a Q3R (or another wired-Ethernet board) build + boot to confirm the PHY reset line waveform and link coming up

Real-world validation on hardware is welcome - if someone with a Q3R could build and confirm the RMII link behaves, that would close the loop.

[themactep]

no, we should not mix a userland config into uboot configuration. we need a per-camera dtsi fragments that would define these settings.

[jmertic]

ok @themactep - is there a pattern I could see to replicate what you are suggesting? Happy to take a stab at this.

[themactep]

You need to coordinate with @gtxaspec on that, he is working on changes to dts right now.

[gtxaspec]

I've got some changes in progress, i'll touch base here when ready @jmertic