dt: rk3588s-nanopi-m6: add explicit rx_delay=0 for rgmii-rxid gmac1

[Copilot]

NanoPi M6 Ethernet breaks on Linux 7.0 because dwmac-rk.c changed RGMII_RXID handling to use bsp_priv->rx_delay (defaulting to 0x10) instead of hardcoding 0, incorrectly applying SoC-side RX delay when the RTL8211F PHY already handles it internally.

Change

• Add rx_delay = <0x00> to &gmac1 in rk3588s-nanopi-m6.dts, matching all other active RK3588/RK3588S boards using rgmii-rxid (hinlink-h88k, mixtile-core3588e, youyeetoo-r1)

 &gmac1 {
     clock_in_out = "output";
     phy-handle = <&rgmii_phy1>;
     phy-mode = "rgmii-rxid";
     ...
+    rx_delay = <0x00>;
     tx_delay = <0x42>;
     status = "okay";
 };

Notes

The reporter also blacklisted dwmac_rk, which is the actual driver for fe1c0000.ethernet — removing it from the blacklist is required regardless of this DTS fix. The sync_state() pending kernel log messages are a downstream effect of unprobed blacklisted drivers (GPU, VPU, HDMI, Ethernet) and are not themselves the cause of failure.

---

📱 Kick off Copilot coding agent tasks wherever you are with GitHub Mobile, available on iOS and Android.

[MichaIng]

Reason for failing Ethernet was handmade on my particular system, but the missing rx_delay indeed causes a kernel error, after which it falls back to 0x10:

[    5.113392] rk_gmac-dwmac fe1c0000.ethernet: Can not read property: rx_delay.
[    5.114018] rk_gmac-dwmac fe1c0000.ethernet: set rx_delay to 0x10

Copilot's analysis however is wrong: rx_delay is in fact not applied in the end with rgmii-rxid, neither before nor after Linux 7.0, and it does not cause any issues if missing. A recent commit touched that code, but did not practically change something: torvalds/linux@d323769

Reason for the error is that the value is checked unconditionally. But is it ignored in case og rgmii-rxid: https://github.com/torvalds/linux/blob/b10d56b/drivers/net/ethernet/stmicro/stmmac/dwmac-rk.c#L1276-L1285