[PATCH] arm64: dts: rockchip: odroid-m1: use rgmii-id for gmac0

[PATCH] arm64: dts: rockchip: odroid-m1: use rgmii-id for gmac0

[Miguel Jesus Seron Gonzalez]

The gmac0 node uses phy-mode "rgmii" with the MAC delays
tx_delay = <0x4f> and rx_delay = <0x2d>. On some boards this TX delay
does not work at 1000 Mbit/s. Dongjin Kim reported iperf3 throughput
below 200 Mbit/s on REV 1.0 20220613 boards. On a REV 1.0 20211112
board, TX from the board stalls completely in iperf3 and ping loss
grows with frame size (5% at 56 bytes, 20% at 1472 bytes), while RX is
unaffected. Other boards work with the current value, so it seems to
be marginal rather than wrong everywhere.

Instead of tuning the MAC TX delay again, switch to phy-mode
"rgmii-id", so that the RTL8211F PHY adds both RGMII delays, and drop
the MAC delay properties, as suggested by Chukun Pan. Dongjin
confirmed in the same thread that rgmii-id works.

With this change, on the REV 1.0 20211112 board at 1000 Mbit/s full
duplex, there were 3 retransmitted TCP segments out of 24929 (0.012%)
over almost four hours, no RX/TX errors and no ping loss.

Fixes: fd3583267703 ("arm64: dts: rockchip: Add Hardkernel ODROID-M1 board")
Link: https://lore.kernel.org/r/20250819075522.2238643-1-tobetter@gmail.com
Suggested-by: Chukun Pan <amadeus@jmu.edu.cn>
Assisted-by: Claude:claude-opus-5-5
Signed-off-by: Miguel Jesus Seron Gonzalez <info@gluon.es>
---

Notes:
    Sent as a standalone patch because Home Assistant OS needs an upstream
    reference for the same change. Dongjin confirmed in the v2 thread that
    rgmii-id works. Dongjin, if you would rather send your own version,
    I'll withdraw this one.
    
    The debugging, the changelog and the patch were prepared with the help
    of Claude (Anthropic). I reviewed all of it, and the test results above
    are from my own board.

 arch/arm64/boot/dts/rockchip/rk3568-odroid-m1.dts | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/arch/arm64/boot/dts/rockchip/rk3568-odroid-m1.dts b/arch/arm64/boot/dts/rockchip/rk3568-odroid-m1.dts
index 442a2bc43..a3cd21aa0 100644
--- a/arch/arm64/boot/dts/rockchip/rk3568-odroid-m1.dts
+++ b/arch/arm64/boot/dts/rockchip/rk3568-odroid-m1.dts
@@ -200,7 +200,7 @@ &gmac0 {
 	assigned-clock-rates = <0>, <125000000>;
 	clock_in_out = "output";
 	phy-handle = <&rgmii_phy0>;
-	phy-mode = "rgmii";
+	phy-mode = "rgmii-id";
 	phy-supply = <&vcc3v3_sys>;
 	pinctrl-names = "default";
 	pinctrl-0 = <&gmac0_miim
@@ -209,9 +209,6 @@ &gmac0_rx_bus2
 		     &gmac0_rgmii_clk
 		     &gmac0_rgmii_bus>;
 	status = "okay";
-
-	tx_delay = <0x4f>;
-	rx_delay = <0x2d>;
 };
 
 &gpu {

base-commit: c03c1a14507dfaf8b856c2608f9ac216021fc222
-- 
2.53.0

Re: [PATCH] arm64: dts: rockchip: odroid-m1: use rgmii-id for gmac0

[Andrew Lunn]

On Mon, Oct 19, 2026 at 09:22:54PM +0200, Miguel Jesus Seron Gonzalez wrote:
> The gmac0 node uses phy-mode "rgmii" with the MAC delays
> tx_delay = <0x4f> and rx_delay = <0x2d>. On some boards this TX delay
> does not work at 1000 Mbit/s. Dongjin Kim reported iperf3 throughput
> below 200 Mbit/s on REV 1.0 20220613 boards. On a REV 1.0 20211112
> board, TX from the board stalls completely in iperf3 and ping loss
> grows with frame size (5% at 56 bytes, 20% at 1472 bytes), while RX is
> unaffected. Other boards work with the current value, so it seems to
> be marginal rather than wrong everywhere.
> 
> Instead of tuning the MAC TX delay again, switch to phy-mode
> "rgmii-id", so that the RTL8211F PHY adds both RGMII delays, and drop
> the MAC delay properties, as suggested by Chukun Pan. Dongjin
> confirmed in the same thread that rgmii-id works.
> 
> With this change, on the REV 1.0 20211112 board at 1000 Mbit/s full
> duplex, there were 3 retransmitted TCP segments out of 24929 (0.012%)
> over almost four hours, no RX/TX errors and no ping loss.
> 
> Fixes: fd3583267703 ("arm64: dts: rockchip: Add Hardkernel ODROID-M1 board")
> Link: https://lore.kernel.org/r/20250819075522.2238643-1-tobetter@gmail.com
> Suggested-by: Chukun Pan <amadeus@jmu.edu.cn>
> Assisted-by: Claude:claude-opus-5-5
> Signed-off-by: Miguel Jesus Seron Gonzalez <info@gluon.es>

> @@ -200,7 +200,7 @@ &gmac0 {
>  	assigned-clock-rates = <0>, <125000000>;
>  	clock_in_out = "output";
>  	phy-handle = <&rgmii_phy0>;
> -	phy-mode = "rgmii";
> +	phy-mode = "rgmii-id";
>  	phy-supply = <&vcc3v3_sys>;
>  	pinctrl-names = "default";
>  	pinctrl-0 = <&gmac0_miim
> @@ -209,9 +209,6 @@ &gmac0_rx_bus2
>  		     &gmac0_rgmii_clk
>  		     &gmac0_rgmii_bus>;
>  	status = "okay";
> -
> -	tx_delay = <0x4f>;
> -	rx_delay = <0x2d>;

Correct from the DT binding point of few.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>

    Andrew