Date:   Mon, 6 Jun 2022 15:09:23 +0530
From:   Siddharth Vadapalli <s-vadapalli@...com>
To:     Roger Quadros <rogerq@...nel.org>
CC:     <robh+dt@...nel.org>, <lee.jones@...aro.org>,
        <krzysztof.kozlowski+dt@...aro.org>, <vkoul@...nel.org>,
        <dan.carpenter@...cle.com>, <kishon@...com>,
        <grygorii.strashko@...com>, <devicetree@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>, <linux-phy@...ts.infradead.org>
Subject: Re: [PATCH 1/2] dt-bindings: phy: ti: phy-gmii-sel: Add bindings for
 J7200

Hello Roger,

On 03/06/22 18:06, Roger Quadros wrote:
> Siddharth,
> 
> On 03/06/2022 13:49, Siddharth Vadapalli wrote:
>> 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@...com>
>>>>>>>> ---
>>>>>>>>  .../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.
> 
> Yes, I already agreed that separate compatible is fine :).
> 
>>
>>>>
>>>>>
>>>>> 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
> 
> All I'm saying is that instead of bitmask please use port number to specify main port.
> You can use minimum/maximum to limit the values.
> 
> Take care of limit checking per compatible and converting into bitmask in the driver.

The current series of patches is for J7200 device which supports one QSGMII
interface. However, I plan to post patches for another device (J721e) which has
8 external ports and therefore, can be configured as two sets of QSGMII
interfaces. To identify the two main ports across the two QSGMII interfaces, a
single port number will not be sufficient. Hence, a bitmask has been used, to
avoid adding a new property for the second QSGMII interface's main port, when I
post patches for J721e.

Thanks,
Siddharth.

Powered by blists - more mailing lists