dwc_eth_qos driver for tegra

[Marek Vasut]

On 5/23/22 13:46, Rasmus Villemoes wrote:
> On 23/05/2022 12.57, Marek Vasut wrote:
>> On 5/23/22 11:17, Rasmus Villemoes wrote:
>>> Hi
>>
>> Hi,
>>
>>> I'm looking at switching the dwc_eth_qos driver over to use
>>> dm_eth_phy_connect(). However, I'm a little puzzled by the code for the
>>> tegra variant. The comment at the top of the file, as well as
>>> tegra186.dtsi, says
>>>
>>>     phy-mode = "rgmii";
>>>
>>> But eqos_get_interface_tegra186() returns a hard-coded
>>> PHY_INTERFACE_MODE_MII. Now the commit which introduced the ->interface
>>> abstraction, ac2d4efb16e (net: dwc_eth_qos: add Ethernet stm32mp1
>>> support), and that eqos_get_interface_tegra186() function, changed
>>>
>>> -       eqos->phy = phy_connect(eqos->mii, 0, dev, 0);
>>>
>>> to
>>>
>>> +               eqos->phy = phy_connect(eqos->mii, 0, dev,
>>> +                                       eqos->config->interface(dev));
>>>
>>> and that last hard-coded 0 in the former phy_connect() is indeed
>>> equivalent to PHY_INTERFACE_MODE_MII.
>>>
>>> So which is it? It would be nice if one could just rely on
>>> dm_eth_phy_connect() picking up the correct value from device tree, and
>>> drop all the code which duplicates parsing of phy-mode from the ethernet
>>> driver.
>>
>> linux-2.6$ git grep mii arch/arm64/boot/dts/nvidia/tegra186*
>> arch/arm64/boot/dts/nvidia/tegra186-p3310.dtsi:         phy-mode = "rgmii";
>> arch/arm64/boot/dts/nvidia/tegra186-p3509-0000+p3636-0001.dts: phy-mode
>> = "rgmii-id";
>>
>> So probably RGMII ?
> 
> Well, yes, I also did check the linux device tree files which also says
> rgmii, but that doesn't explain why the U-Boot driver code seems to
> ignore that entirely and use mii hardcoded, both before and after
> ac2d4efb16e.
> 
> So another way of asking: does this driver actually work today, and/or
> has it worked at some point? I assume the answer is yes - after all, the
> very first commit "supports the specific configuration used in NVIDIA's
> Tegra186 chip", but that commit also did that phy_connect() with a last
> argument of 0 aka PHY_INTERFACE_MODE_MII.
> 
> And would it break if one started taking the phy-mode from device tree?
> If so, should device tree be updated to say "mii"?

I think we wait for nvidia to answer all this, I don't have that SoC 
available.