Skip to contentSkip to conversation

sunxi phy-mode rgmii-id

RFC Heinrich Schuchardt 2 participants 2 messages Nov 29, 2020 – Nov 30, 2020

lore
Share & tools
lore
#0 lore raw

Hello Joe, hello Jagan,

since Linux patch
bbc4d71d6354 ("net: phy: realtek: fix rtl8211e rx/tx delay config") i.e. since v5.8.15 or v5.9 many if not all Sunxi A64, H6, H5 boards require phy-mode = "rgmii-id" to provide network in Linux

The U-Boot device-tree is the fallback in the UEFI sub-system if no other device-tree is provided. So it would make sense to adjust the U-Boot device-trees too.

The Linux patch has specifically changed the values corresponding to our U-Boot defines:

#define MIIM_RTL8211E_CONFREG_TXD???????????????0x0002 #define MIIM_RTL8211E_CONFREG_RXD???????????????0x0004

At least my experience with the NanoPi Neo 2 and the Pine64 A64 LTS indicates that setting phy-mode = "rgmii-id" in the device-tree does not stop the operation of the network even if the driver is not adjusted.

To avoid breaking boards like Linux did we should change the device-trees before or in sync with drivers/net/phy/realtek.c.

Best regards

Heinrich

#1 lore raw

On 29/11/2020 23:07, Heinrich Schuchardt wrote:

Hi,

Quoted 12 lines

since Linux patch
bbc4d71d6354 ("net: phy: realtek: fix rtl8211e rx/tx delay config") i.e. since v5.8.15 or v5.9 many if not all Sunxi A64, H6, H5 boards require phy-mode = "rgmii-id" to provide network in Linux

The U-Boot device-tree is the fallback in the UEFI sub-system if no other device-tree is provided. So it would make sense to adjust the U-Boot device-trees too.

The Linux patch has specifically changed the values corresponding to our U-Boot defines:

Quoted 6 lines

#define MIIM_RTL8211E_CONFREG_TXD???????????????0x0002 #define MIIM_RTL8211E_CONFREG_RXD???????????????0x0004

At least my experience with the NanoPi Neo 2 and the Pine64 A64 LTS indicates that setting phy-mode = "rgmii-id" in the device-tree does not stop the operation of the network even if the driver is not adjusted.

That is not entirely true, it depends on a patch to not reject other phy-modes in the Linux driver (f1239d8aa84d).
This entered with v5.5, I believe, and was backported to stable kernels since, but is not universally available in older kernels. Especially installers comes to mind ...

But first we need at least the equivalent patch for U-Boot, as I posted here:
https://lists.denx.de/pipermail/u-boot/2020-November/432823.html

Quoted 2 lines

To avoid breaking boards like Linux did we should change the device-trees before or in sync with drivers/net/phy/realtek.c.

Eventually: yes, but at the cost of breaking older kernels. This is really an unfortunate situation, and actually should not have happened:
normally we at least guarantee operation of newer kernels with older DTs.

Cheers,
Andre