Re: [PATCH 2/3] dt-bindings: phy: Add lane<n>-mode property to WIZ (SERDES wrapper)

[Jyri Sarha <jsarha@ti.com>]

On 30/12/2019 12:18, Kishon Vijay Abraham I wrote:
> Hi,
> 
> On 30/12/19 3:07 PM, Jyri Sarha wrote:
>> On 24/12/2019 23:31, Rob Herring wrote:
>>> On Fri, Dec 20, 2019 at 5:52 AM Jyri Sarha <jsarha@ti.com> wrote:
>>>>
>>>> On 19/12/2019 21:08, Rob Herring wrote:
>>>>> On Mon, Dec 09, 2019 at 06:22:11PM +0200, Jyri Sarha wrote:
>>>>>> Add property to indicate the usage of SERDES lane controlled by the
>>>>>> WIZ wrapper. The wrapper configuration has some variation depending on
>>>>>> how each lane is going to be used.
>>>>>>
>>>>>> Signed-off-by: Jyri Sarha <jsarha@ti.com>
>>>>>> ---
>>>>>>  .../devicetree/bindings/phy/ti,phy-j721e-wiz.yaml    | 12 ++++++++++++
>>>>>>  1 file changed, 12 insertions(+)
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/phy/ti,phy-j721e-wiz.yaml b/Documentation/devicetree/bindings/phy/ti,phy-j721e-wiz.yaml
>>>>>> index 94e3b4b5ed8e..399725f65278 100644
>>>>>> --- a/Documentation/devicetree/bindings/phy/ti,phy-j721e-wiz.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/phy/ti,phy-j721e-wiz.yaml
>>>>>> @@ -97,6 +97,18 @@ patternProperties:
>>>>>>        Torrent SERDES should follow the bindings specified in
>>>>>>        Documentation/devicetree/bindings/phy/phy-cadence-dp.txt
>>>>>>
>>>>>> +  "^lane[1-4]-mode$":
>>>>>> +    allOf:
>>>>>> +      - $ref: /schemas/types.yaml#/definitions/uint32
>>>>>> +      - enum: [0, 1, 2, 3, 4, 5, 6]
>>>>>> +    description: |
>>>>>> +     Integer describing static lane usage for the lane indicated in
>>>>>> +     the property name. For Sierra there may be properties lane0 and
>>>>>> +     lane1, for Torrent all lane[1-4]-mode properties may be
>>>>>> +     there. The constants to indicate the lane usage are defined in
>>>>>> +     "include/dt-bindings/phy/phy.h". The lane is assumed to be unused
>>>>>> +     if its lane<n>-use property does not exist.
>>>>>
>>>>> The defines were intended to be in 'phys' cells. Does putting both lane
>>>>> and mode in the client 'phys' properties not work?
>>>>>
>>>>
>>>> Let me first check if I understood you. So you are suggesting something
>>>> like this:
>>>>
>>>> dp-phy {
>>>>         #phy-cells = <5>; /* 1 for phy-type and 4 for lanes = 5 */
>>>>         ...
>>>> };
>>>>
>>>> dp-bridge {
>>>>         ...
>>>>         phys = <&dp-phy PHY_TYPE_DP 1 1 0 0>; /* lanes 0 and 1 for DP */
>>>
>>> Yes, but I think the lanes can be a single cell mask. And I'd probably
>>> make that the first cell which is generally "which PHY" and make
>>> type/mode the 2nd cell. I'd look for other users of PHY_TYPE_ defines
>>> and match what they've done if possible.
>>>
>>
>> I see. This will cause some head ache on the driver implementation side,
>> as there is no way for the phy driver to peek the lane use or type from
>> the phy client's device tree node. It also looks to me that the phy
>> API[1] has to be extended quite a bit before the phy client can pass the
>> lane usage information to the phy driver. It will cause some pain to
>> implement the extension without breaking the phy API and causing a nasty
>> cross dependency over all the phy client domains.
>>
>> Also, there is not much point in putting the PHY_TYPE constant to the
>> phy client's node, as normally the phy client driver will know quite
>> well what PHY_TYPE to use. E.g. a SATA driver will always select
>> PHY_TYPE_SATA and a PCIE driver will select PHY_TYPE_PCIE, etc.
>>
>> Kishon, if we have to take this road it also starts to sound like we
>> will have to move the phy client's phandle to point to the phy wrapper
>> node, if we want to keep the actual phy driver wrapper agnostic. Then we
>> can make the wrapper to act like a proxy that forwards the phy_ops calls
>> to the actual phy driver. Luckily the per lane phy-type selection is not
>> a blocker for our j721e DisplayPort functionality.
> 
> WIZ is a PHY wrapper and not a PHY in itself. I'm not inclined in
> modeling WIZ as a PHY and adding an additional level of indirection.
> This can add more challenges w.r.t PHY sequencing and can also lead to
> locking issues. That also doesn't accurately represent the HW bock.
> 

Ok, then assuming the phy wrapper node's lane<n>-mode property can't be
used and if the lane-mode information is only available at the phy
client driver, we must somehow deliver the phy-mode information from the
phy client driver to the phy wrapper driver.

Maybe a way for the phy wrapper driver to request the phy-mode from the
actual phy driver, which in turn gets it from the phy client through
phy_ops set_mode() call-back.

Then there is the extra twist of a single phy driver serving multiple
phy clients using different lanes, but we do not need to cross that
bridge for the current DisplayPort functionality.

Best regards,
Jyri

-- 
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki