[PATCH v3 1/5] arm64: dts: rockchip: Reset the NanoPi Zero2 PHY from the MDIO bus

[Andrey Korshunov]

On Wed Sep 9, 2026 at 10:33 PM CST, 安容 via B4 Relay wrote:
> Describe the reset on the MDIO bus instead. __mdiobus_register() asserts
> and releases the bus level reset-gpios before any device on the bus is
> registered, so the PHY responds whatever state the bootloader left it
> in.

Confirmed. And thanks for catching the deprecated properties I had
suggested - I had not read snps,dwmac.yaml, and dt-validate does not warn
on deprecated ones, so none of the checks I ran would have told me either.

I ran it both ways on a NanoPi R28S - same pin, same RTL8211F on mdio1 -
with two images that differ only in those nine lines of device tree. Both
are 6.18.44 with mainline U-Boot 2026.07 (generic-rk3528).

Reset in the PHY node, the way mainline describes it today:

  [    0.217516] mdio_bus stmmac-0: MDIO device at address 1 is missing.
  [   24.566496] rk_gmac-dwmac ffbe0000.ethernet eth0: cannot attach PHY

and the port stays down. Reset on the MDIO bus, as this patch does it:

  [   14.608938] rk_gmac-dwmac ffbe0000.ethernet eth0: PHY [stmmac-0:01] driver [RTL8211F Gigabit Ethernet] (irq=POLL)
  [   17.770103] rk_gmac-dwmac ffbe0000.ethernet eth0: Link is Up - 1Gbps/Full - flow control rx/tx

So U-Boot 2026.07 does leave the pin asserted, the same as the 2025.10 I
first saw this on - the commit message holds for the current release too,
not just the one I happened to report it from.

The one thing I cannot confirm is the board in the subject: this is an
R28S, not a Zero2, which I do not have.

Tested-by: Andrey Korshunov <razor1702 at gmail.com>

Andrey