Re: Multiple generic PHY instances for DWC3 USB IP

[Felipe Balbi]

Hi,

Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> writes:
> 2018-04-04 14:36 GMT+09:00 Felipe Balbi <felipe.balbi@xxxxxxxxxxxxxxx>:
>>
>> Hi,
>>
>> Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> writes:
>>> Currently, DWC3 core IP (drivers/usb/dwc3/core.c)
>>> can take only one PHY phandle for each of SS, HS.
>>> (phy-names DT property is "usb2-phy" and "usb3-phy" for each)
>>
>> We never had any other requirements :-)
>>
>>> The DWC3 core IP is provided by Synopsys,
>>> but some SoC-dependent parts (a.k.a glue-layer)
>>> are implemented by SoC venders.
>>>
>>> The number of connected PHY instances are SoC-dependent.
>>>
>>> If you look at generic drivers such as
>>>   drivers/usb/host/ehci-platform.c
>>> the driver can handle arbitrary number of PHY instances.
>>>
>>> However, as mentioned above, DWC3 core allows only one PHY phandle
>>> for each SS/HS.
>>> This can result in a strange DT structure.
>>>
>>> For example, Socionext PXs3 SoC is integrated with 2 instances of DWC3.
>>>
>>> The instance 0 of DWC3 is connected with 2 super-speed PHYs.
>>
>> why 2 super-speed phys? Is this a two-port host-only implementation?
>
>
> Socionext SoCs only support the host-mode.
>
>
> The instance 0 has 2 ports.
> In our integration, 1 SS PHY is needed for each port.
> That's why it needs 2 SS PHYs.
>
> Each DWC3 instance is connected with
> multiple HS PHYs and multiple SS PHYs,
> depending on the number of ports.

in that case, you shouldn't need dwc3 at all. A Host-only dwc3 is xHCI
compliant. If you really don't have the gadget block, there's no need
for you to use dwc3. Just use xhci-plat directly.

>>> Is this OK?
>>
>> I don't know, I need a bit more details about your integration :-)
>
>
> I can send a patch.
>
> My concern is the following commit.
> I do not know which parts are using this lookups.

Samsung SoCs, probably ;-)

Anyway, if your IP really is host-only, then you don't need dwc3 for
anything. Just go for xHCI directly. If xHCI needs to be extended when
it comes to PHY, then you can discuss with Mathias Nyman :-)

-- 
balbi