Re: [PATCH 3/6] phy: ti: gmii-sel: Add support for CPSW9G GMII SEL in J721e

[Siddharth Vadapalli]

Hello Roger,

On 14/09/22 17:04, Roger Quadros wrote:
> Hi Siddharth,
> 
> On 14/09/2022 12:39, Siddharth Vadapalli wrote:
>> Each of the CPSW9G ports in J721e support additional modes like QSGMII.
>> Add a new compatible for J721e to support the additional modes.
>>
>> In TI's J721e, each of the CPSW9G ethernet interfaces can act as a
>> QSGMII main or QSGMII-SUB port. The QSGMII main interface is responsible
>> for performing auto-negotiation between the MAC and the PHY while the rest
>> of the interfaces are designated as QSGMII-SUB interfaces, indicating that
>> they will not be taking part in the auto-negotiation process.
>>
>> Signed-off-by: Siddharth Vadapalli <s-vadapalli@xxxxxx>
>> ---
>>  drivers/phy/ti/phy-gmii-sel.c | 47 +++++++++++++++++++++++++++--------
>>  1 file changed, 37 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/phy/ti/phy-gmii-sel.c b/drivers/phy/ti/phy-gmii-sel.c
>> index f0b2ba7a9c96..fdb1a7db123d 100644
>> --- a/drivers/phy/ti/phy-gmii-sel.c
>> +++ b/drivers/phy/ti/phy-gmii-sel.c
>> @@ -223,6 +223,13 @@ struct phy_gmii_sel_soc_data phy_gmii_sel_cpsw5g_soc_j7200 = {
>>  	.extra_modes = BIT(PHY_INTERFACE_MODE_QSGMII) | BIT(PHY_INTERFACE_MODE_SGMII),
>>  };
>>  
>> +static const
>> +struct phy_gmii_sel_soc_data phy_gmii_sel_cpsw9g_soc_j721e = {
>> +	.use_of_data = true,
>> +	.regfields = phy_gmii_sel_fields_am654,
>> +	.extra_modes = BIT(PHY_INTERFACE_MODE_QSGMII),
>> +};
>> +
>>  static const struct of_device_id phy_gmii_sel_id_table[] = {
>>  	{
>>  		.compatible	= "ti,am3352-phy-gmii-sel",
>> @@ -248,6 +255,10 @@ static const struct of_device_id phy_gmii_sel_id_table[] = {
>>  		.compatible	= "ti,j7200-cpsw5g-phy-gmii-sel",
>>  		.data		= &phy_gmii_sel_cpsw5g_soc_j7200,
>>  	},
>> +	{
>> +		.compatible	= "ti,j721e-cpsw9g-phy-gmii-sel",
>> +		.data		= &phy_gmii_sel_cpsw9g_soc_j721e,
>> +	},
>>  	{}
>>  };
>>  MODULE_DEVICE_TABLE(of, phy_gmii_sel_id_table);
>> @@ -389,7 +400,7 @@ static int phy_gmii_sel_probe(struct platform_device *pdev)
>>  	struct device_node *node = dev->of_node;
>>  	const struct of_device_id *of_id;
>>  	struct phy_gmii_sel_priv *priv;
>> -	u32 main_ports = 1;
>> +	u32 main_ports[2] = {1, 1};
>>  	int ret;
>>  
>>  	of_id = of_match_node(phy_gmii_sel_id_table, pdev->dev.of_node);
>> @@ -403,15 +414,31 @@ static int phy_gmii_sel_probe(struct platform_device *pdev)
>>  	priv->dev = &pdev->dev;
>>  	priv->soc_data = of_id->data;
>>  	priv->num_ports = priv->soc_data->num_ports;
>> -	of_property_read_u32(node, "ti,qsgmii-main-ports", &main_ports);
>> -	/*
>> -	 * Ensure that main_ports is within bounds. If the property
>> -	 * ti,qsgmii-main-ports is not mentioned, or the value mentioned
>> -	 * is out of bounds, default to 1.
>> -	 */
>> -	if (main_ports < 1 || main_ports > 4)
>> -		main_ports = 1;
>> -	priv->qsgmii_main_ports = PHY_GMII_PORT(main_ports);
>> +	/* Differentiate between J7200 CPSW5G and J721e CPSW9G */
>> +	if (of_device_is_compatible(node, "ti,j7200-cpsw5g-phy-gmii-sel") > 0) {
> 
> Why not just "if (of_device_is_compatible())" ?

Thank you for reviewing the patch. I will fix this in the v2 series.

> 
>> +		of_property_read_u32(node, "ti,qsgmii-main-ports", &main_ports[0]);
>> +		/*
>> +		 * Ensure that main_ports is within bounds. If the property
>> +		 * ti,qsgmii-main-ports is not mentioned, or the value mentioned
>> +		 * is out of bounds, default to 1.
>> +		 */
>> +		if (main_ports[0] < 1 || main_ports[0] > 4)
>> +			main_ports[0] = 1;
> 
> how about printing this issue with dev_err()?

I agree that using dev_err() instead of defaulting to a value is a
better choice here. I had initially planned on defaulting to a value
since this check is a part of the probe function and I had thought that
the phy-mode is not yet known at this point. However, looking at it
again, for the special case where the property "ti,qsgmii-main-ports" is
mentioned in the devicetree node, it is possible to know with certainty
that QSGMII mode is intended and a wrong value has been provided in the
devicetree node. I will add dev_err() in the v2 series, instead of
defaulting to 1 if the check fails.

For the other scenario where "ti,qsgmii-main-ports" is not mentioned in
the devicetree node, I think that defaulting to 1 would be the correct
choice since the intended phy-mode is not yet known at this point.

> 
>> +		priv->qsgmii_main_ports = PHY_GMII_PORT(main_ports[0]);
>> +	} else if (of_device_is_compatible(node, "ti,j721e-cpsw9g-phy-gmii-sel") > 0) {
>> +		of_property_read_u32_array(node, "ti,qsgmii-main-ports", &main_ports[0], 2);
>> +		/*
>> +		 * Ensure that main_ports is within bounds. If the property
>> +		 * ti,qsgmii-main-ports is not mentioned, or the value mentioned
>> +		 * is out of bounds, default to 1.
>> +		 */
>> +		if (main_ports[0] < 1 || main_ports[0] > 8)
>> +			main_ports[0] = 1;
>> +		if (main_ports[1] < 1 || main_ports[1] > 8)
>> +			main_ports[1] = 1;
>> +		priv->qsgmii_main_ports = PHY_GMII_PORT(main_ports[0]);
>> +		priv->qsgmii_main_ports |= PHY_GMII_PORT(main_ports[1]);
>> +	}
> 
> The whole if/else logic can be got rid of if you store num_qsgmii_main_ports in priv data structure
> after obtaining it from of_data.
> 
> Then all the above reduces to
> 	for (i = 0; i < priv->num_qsgmii_main_ports; i++) {
> 		if (main_ports[i] ...)
> 	}
> 
> It will also make it very easy to scale later on for future platforms.

Thank you for the suggestion. I will add the variable "u32
num_qsgmii_main_ports" in "struct phy_gmii_sel_soc_data" and set its
value to 1 for the "phy_gmii_sel_cpsw5g_soc_j7200" compatible and to 2
for the "phy_gmii_sel_cpsw9g_soc_j721e" compatible. I will implement
this in the v2 series.

Regards,
Siddharth.