Re: [PATCH v2 03/22] usb: ulpi: Support device discovery via device properties

[Stephen Boyd]

On Fri, Aug 5, 2016 at 2:40 PM, Stephen Boyd <stephen.boyd@xxxxxxxxxx> wrote:
> Quoting Rob Herring (2016-07-17 19:23:55)
>> On Thu, Jul 07, 2016 at 03:20:54PM -0700, Stephen Boyd wrote:
>> > +-------
>> > +
>> > +usb {
>> > +     compatible = "vendor,usb-controller";
>> > +
>> > +     ulpi {
>> > +             phy {
>> > +                     compatible = "vendor,phy";
>> > +                     ulpi-vendor = /bits/ 16 <0x1d6b>;
>> > +                     ulpi-product = /bits/ 16 <0x0002>;
>> > +             };
>> > +     };
>>
>> I'm still having concerns about describing both phys and devices. If I
>> have a controller with 2 ports and 2 devices attached, I'd have
>> something like this under the USB controller:
>>
>> ulpi {
>>         phy@1 {
>>         };
>>         phy@2 {
>>         };
>> };
>
> My understanding is there would only be one status="ok" node on the ULPI
> bus for the single phy that a usb controller would have. At the least,
> the kernel's ULPI layer only seems to support one ULPI phy for a
> controller right now. So even if there are two ports, it doesn't mean
> there are two phys.
>
>>
>> dev@1 {
>> ...
>> };
>>
>> dev@2 {
>> ...
>> };
>>
>>
>> That doesn't seem the best, but I don't have a better suggestion. Maybe
>> the device nodes need to go under the phy nodes?
>>
>
> What if we moved the dev@1 and dev@2 to another sub node like "ports" or
> "usb-devices"? Legacy code can support having those devices directly
> underneath the usb controller, but future users would always need to put
> them in a different sub-node so that we can easily differentiate the
> different busses that a usb controller node may support?
>
> I'm not sure I see any need to relate the phy to the ports that are on
> the controller, but if that is needed then perhaps you're right and we
> should move the ports underneath the phy. USB core could be modified to
> go through the legacy path or through the phy, if it even exists, to
> find ports.
>
> Do we typically do this for other phy designs like sata or pci? The phy
> always seemed like a parallel thing to the logical bus that the phy is
> used for.

Rob does this sound ok to you?