Re: [PATCH 1/2] dt-bindings: phy: ti: phy-gmii-sel: Add bindings for J7200

[Siddharth Vadapalli]

Hello Roger,

On 03/06/22 14:18, Roger Quadros wrote:
> Hi Siddharth,
> 
> On 01/06/2022 14:27, Siddharth Vadapalli wrote:
>> Hello Roger,
>>
>> On 01/06/22 15:08, Roger Quadros wrote:
>>> Siddharth,
>>>
>>> On 01/06/2022 09:01, Siddharth Vadapalli wrote:
>>>> Hello Roger,
>>>>
>>>> On 31/05/22 17:15, Roger Quadros wrote:
>>>>> Hi Siddharth,
>>>>>
>>>>> On 31/05/2022 14:12, Siddharth Vadapalli wrote:
>>>>>> TI's J7200 SoC supports additional PHY modes like QSGMII and SGMII
>>>>>> that are not supported on earlier SoCs. Add a compatible for it.
>>>>>>
>>>>>> Signed-off-by: Siddharth Vadapalli <s-vadapalli@xxxxxx>
>>>>>> ---
>>>>>>  .../mfd/ti,j721e-system-controller.yaml       |  5 ++++
>>>>>>  .../bindings/phy/ti,phy-gmii-sel.yaml         | 24 ++++++++++++++++++-
>>>>>>  2 files changed, 28 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/mfd/ti,j721e-system-controller.yaml b/Documentation/devicetree/bindings/mfd/ti,j721e-system-controller.yaml
>>>>>> index fa86691ebf16..e381ba62a513 100644
>>>>>> --- a/Documentation/devicetree/bindings/mfd/ti,j721e-system-controller.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/mfd/ti,j721e-system-controller.yaml
>>>>>> @@ -48,6 +48,11 @@ patternProperties:
>>>>>>      description:
>>>>>>        This is the SERDES lane control mux.
>>>>>>  
>>>>>> +  "phy@[0-9a-f]+$":
>>>>>> +    type: object
>>>>>> +    description:
>>>>>> +      This is the register to set phy mode through phy-gmii-sel driver.
>>>>>> +
>>>>>
>>>>> Is this really required? The system controller has 100s of different such registers and it is not practical to mention about all.
>>>>
>>>> The property has to be mentioned in order to pass: make dtbs_check.
>>>>
>>>>>
>>>>>>  required:
>>>>>>    - compatible
>>>>>>    - reg
>>>>>> diff --git a/Documentation/devicetree/bindings/phy/ti,phy-gmii-sel.yaml b/Documentation/devicetree/bindings/phy/ti,phy-gmii-sel.yaml
>>>>>> index ff8a6d9eb153..7427758451e7 100644
>>>>>> --- a/Documentation/devicetree/bindings/phy/ti,phy-gmii-sel.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/phy/ti,phy-gmii-sel.yaml
>>>>>> @@ -53,12 +53,21 @@ properties:
>>>>>>        - ti,am43xx-phy-gmii-sel
>>>>>>        - ti,dm814-phy-gmii-sel
>>>>>>        - ti,am654-phy-gmii-sel
>>>>>> +      - ti,j7200-cpsw5g-phy-gmii-sel
>>>>>
>>>>> Why not just "ti,j7200-phy-gmii-sel" so it is consistent naming.
>>>>
>>>> In TI's J7200 device, there are two CPSW MACs, namely CPSW2G and CPSW5G. While
>>>> CPSW5G supports QSGMII mode, CPSW2G does not. Hence, the compatible being added
>>>> with the extra mode (QSGMII) enabled is applicable only for CPSW5G and not for
>>>> CPSW2G. Thus, to highlight this, the word "CPSW5G" has been included in the name
>>>> of the compatible.
>>>
>>> Here we are talking about the PHY driver (phy-gmii-sel) and not the MAC (CPSW2G / CPSW5G)
>>> Does this PHY on J7200 always support QSGMII mode? if yes then embedding "cpsw5g" in compatible is wrong.
>>
>> The PHY on J7200 is part of the Add-On Ethernet card. It is possible to connect
>> RGMII, QSGMII and SGMII PHY. The CPSW5G MAC supports all these modes. With the
>> current patch, I am adding just QSGMII mode as an extra mode, but in a future
>> patch, I will be adding SGMII also as an extra mode. For this reason, CPSW5G is
>> being mentioned in the compatible name, to differentiate supported modes for
>> CPSW2G and CPSW5G. Also, the phy-gmii-sel driver actually configures CPSW MAC
>> registers and not the PHY.
> 
> phy-gmii-sel configures CTRL MMR register right? How does it configure CPSW MAC register?
> 
> Anyways, I just looked at the TRM and there are in fact separate phy-gmii-sel (ENET_CTRL)
> registers for CPSW2g and CPSW5g. So they warrant for separate compatibles as they are
> not identical.

By CPSW MAC registers being configured, I meant that the configuration being
done is for the MAC and not for the PHY. As per the TRM, for CPSW2G, the
CTRLMMR_MCU_ENET_CTRL register is configured and for CPSW5G, the
CTRLMMR_ENETx_CTRL registers are configured, with x ranging from 1 to 4
(corresponding to the 4 ports of CPSW5G). These registers configure the CPSW MAC
(CPSW2G/CPSW5G) and not the PHY. For this reason, I think that it would be
appropriate to use cpsw5g in the compatible name, to indicate which CTRLMMR
registers are being configured.

>>
>>>
>>> You need to use a different compatible in CPSW driver and make sure CPSW2G doesn't initiate QSGMII mode.
>>
>> Yes, I will add a check there too by using a different compatible in the CPSW
>> driver, but shouldn't the phy-gmii-sel driver also have a check to ensure that
>> it doesn't try configuring QSGMII mode for CPSW2G?
> 
> Yes, additional check in phy-gmii-sel driver is fine.
> 
>>
>>>
>>>>
>>>>>
>>>>>>  
>>>>>>    reg:
>>>>>>      maxItems: 1
>>>>>>  
>>>>>>    '#phy-cells': true
>>>>>>  
>>>>>> +  ti,enet-ctrl-qsgmii:
>>>>>> +    $ref: /schemas/types.yaml#/definitions/uint32
>>>>>> +    description: |
>>>>>> +      Required only for QSGMII mode. Bitmask to select the port for
>>>>>> +      QSGMII main mode. Rest of the ports are selected as QSGMII_SUB
>>>>>> +      ports automatically. Any of the 4 CPSW5G ports can act as the
>>>>>> +      main port with the rest of them being the QSGMII_SUB ports.
>>>>>> +
>>>>>
>>>>> This is weird way of doing things.
>>>>>
>>>>> The Ethernet controller driver already knows which mode the port is
>>>>> supposed to operate.
>>>>
>>>> From the ethernet driver perspective, there is no difference between the QSGMII
>>>> or QSGMII-SUB modes and both are treated the same. However, the phy-gmii-sel
>>>> driver configures CPSW MAC registers differently depending on the mode being
> 
> You mean the ENET_CTRL register in CTRL_MMR space?

Yes I am referring to the CTRLMMR_ENETx_CTRL registers as per the J7200 TRM,
corresponding to the CPSW5G MAC.

> 
>>>> QSGMII or QSGMII-SUB. Hence, the ti,enet-ctrl-qsgmii property is used to
>>>> identify the QSGMII main port and the rest are configured in CPSW MAC as
>>>> QSGMII-SUB ports.
>>>>
>>>>>
>>>>> e.g.
>>>>> +&cpsw0_port1 {
>>>>> +	phy-handle = <&cpsw5g_phy0>;
>>>>> +	phy-mode = "qsgmii";
>>>>> +	mac-address = [00 00 00 00 00 00];
>>>>> +	phys = <&cpsw0_phy_gmii_sel 1>;
>>>>> +};
>>>>> +
>>>>> +&cpsw0_port2 {
>>>>> +	phy-handle = <&cpsw5g_phy1>;
>>>>> +	phy-mode = "qsgmii-sub";
>>>>> +	mac-address = [00 00 00 00 00 00];
>>>>> +	phys = <&cpsw0_phy_gmii_sel 2>;
>>>>>
>>>>> And it can convey the mode to the PHY driver via phy_ops->set_mode.
>>>>> So you should be depending on that instead of adding this new property.
>>>>
>>>> QSGMII-SUB is not a standard mode in the Linux kernel. In order to proceed with
>>>> the suggested implementation, a new phy mode named PHY_INTERFACE_MODE_QSGMII_SUB
>>>> has to be introduced to the kernel. Additionally, all existing phy drivers will
>>>> have to be updated to recognize the new phy mode.
>>>>
>>>> Since the QSGMII-SUB mode is TI specific, it was decided that it would be better
>>>> to add a new property in TI specific files for identifying the QSGMII main port
>>>> and treating the rest as QSGMII-SUB ports.
>>>
>>> Who decides which port should be MAIN and which should be SUB? Can all ports be MAIN?
>>> Can all ports be SUB or there has to be at least one MAIN?
>>
>> All 4 ports in CPSW5G have the capability to be the MAIN port, with the only
>> restriction being that only one of them should be the MAIN port at a time. The
>> role of the CPSW5G ports is decided based on what PHY port each of the CPSW5G
>> ports connects to.
> 
> OK, then instead of using bitmask and property being named "ti,enet-ctrl-qsgmii", why not
> just say "ti,qsgmii-main-port" = <main_port_number>;

I plan to send patches for J721e device which has CPSW9G (8 external ports) MAC.
CPSW9G can work with two sets of QSGMII interfaces (4 ports + 4 ports). Thus,
using a bitmask for the QSGMII main port will help identify the QSGMII main port
across both sets of QSGMII interfaces. The bitmask in case of J721e CPSW9G will
consider the first 4 bits for the first interface's 4 ports and the next 4 bits
for the second interface's 4 ports. In this manner, it will be possible to
extend it for 8 port CPSW9G MAC as well, without having to add a new property
for the second QSGMII interface.

> 
> Also do some sanity check when getting that property.

To ensure that multiple QSGMII ports are not declared as the main port, the
"ti,enet-ctrl-qsgmii" property has been declared as an enum: [1,2,4,8]. If a
different value other than the value in enum were to be used, then "make
dtbs_check" would raise an error. This will prevent configuring multiple QSGMII
ports as the main port at once. Also, in the phy-gmii-sel driver, a default
value of 1 is being assigned to the variable that will store the value
corresponding to the ti,enet-ctrl-qsgmii property from the device tree, thereby
treating the first CPSW5G port as the QSGMII main port by default.

Thanks,
Siddharth.