[PATCH v1 1/9] of: Add NVIDIA Tegra XUSB mailbox binding

[Stephen Warren]

On 06/25/2014 04:37 PM, Andrew Bresticker wrote:
> On Wed, Jun 25, 2014 at 2:42 PM, Stephen Warren <swarren at wwwdotorg.org> wrote:
>> On 06/18/2014 12:16 AM, Andrew Bresticker wrote:
>>> Add device-tree bindings for the Tegra XUSB mailbox which will be used
>>> for communication between the Tegra XHCI controller and the host.
>>
>> Sorry for the slow review.
>>
>>> diff --git a/Documentation/devicetree/bindings/mailbox/nvidia,tegra124-xusb-mbox.txt b/Documentation/devicetree/bindings/mailbox/nvidia,tegra124-xusb-mbox.txt
>>
>>> +NVIDIA Tegra XUSB mailbox
>>> +=========================
>>> +
>>> +The Tegra XUSB mailbox is used by the Tegra XHCI controller's firmware to
>>> +communicate with the host.
>>
>> Isn't the mailbox an internal implementation detail of the XUSB controller.
>>
>> In other words, I'd naively think that there isn't a standalone generic
>> mailbox that can be used by anything, but we just happen to want to use
>> for XUSB. Rather, there's an XUSB controller, and part of the interface
>> to that controller is a mailbox.
> 
> Yes, the mailbox isn't an actual piece of hardware but rather the
> interface through which the XUSB host and AP communicate.
> 
>> As such, I don't think we want a standalone mailbox node in DT. Rather,
>> we should add the required reg and interrupt values into the XUSB DT node.
>>
>> The driver for that XUSB HW module can either:
>>
>> a) Register as both a mailbox driver and an EHCI driver.
>>
>> b) Spawn a child device to instantiate the mailbox driver.
>>
>> Perhaps (b) could be assisted by using the MFD framework?
> 
> So in the RFC series I did something like (a) where the XUSB host
> handled the mailbox interrupt with both the PHY and host could
> registering notifiers to handle the messages.  It was suggested by
> Arnd though that I make a separate mailbox driver.  Instead of having
> a both a host and mailbox node, I could have a single XUSB host node
> and have the mailbox driver bind to that - thoughts?

Yes, that sounds like what I meant by (b) above. I don't think you can
actually have 2 drivers bind to the same DT node though, so it'd have to
work something like:

* XUSB host node causes a platform device to be instantiated
* XUSB host driver probe()s against that
* XUSB host driver's probe() creates a platform device for the mailbox
* XUSB mailbox driver probe()s against that.

Or, perhaps go completely MFD, and have 2 child devices (XUSB host and
XUSB mailbox) instantiated by the MFD parent, which is what is in the DT.