Skip to contentSkip to conversation

dwc_eth_qos driver for tegra

DISCUSSION Rasmus Villemoes 2 participants 4 messages May 23, 2022

lore
Share & tools
lore
#0 lore raw

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.

Rasmus

#1 lore raw
Quoted 1 line · Rasmus Villemoes

Hi

Hi,

Quoted 19 lines

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

(quoted diff) +2 -0
+               eqos->phy = phy_connect(eqos->mii, 0, dev,
+                                       eqos->config->interface(dev));
Quoted 7 lines

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 ?

#2 lore raw
Quoted 32 lines · Marek Vasut
Quoted 1 line · Rasmus Villemoes

Hi

Hi,

Quoted 19 lines

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));

Quoted 7 lines

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.

Quoted 6 lines

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"?

Rasmus

#3 lore raw
Quoted 45 lines · Rasmus Villemoes
Quoted 32 lines · Marek Vasut
Quoted 1 line · Rasmus Villemoes

Hi

Hi,

Quoted 19 lines

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));

Quoted 7 lines

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.

Quoted 6 lines

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.

Quoted 8 lines

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.