[PATCH 2/3] doc: dt-binding: generic onboard USB HUB

[Ulf Hansson]

On 9 December 2015 at 09:55, Arnd Bergmann <arnd at arndb.de> wrote:
> On Wednesday 09 December 2015 16:12:24 Peter Chen wrote:
>> On Tue, Dec 08, 2015 at 09:24:03PM -0600, Rob Herring wrote:
>> > On Tue, Dec 08, 2015 at 10:58:48AM +0100, Arnd Bergmann wrote:
>> > > On Tuesday 08 December 2015 10:50:49 Philipp Zabel wrote:
>> > > > This something we don't have to define ad-hoc. The hub hangs off an USB
>> > > > controller, right? The "Open Firmware recommended practice: USB"
>> > > > document already describes how to represent USB devices in a generic
>> > > > manner:
>> > > > http://www.firmware.org/1275/bindings/usb/usb-1_0.ps
>> > > >
>> > > > Is there a reason not to reuse this?
>> > > >
>> > > > The usb hub node would be a child of the usb controller node, and it
>> > > > could use
>> > > >         compatible = "usb,class9"; /* bDeviceClass 9 (Hub) */
>> > >
>> > > Good point, I had not thought of that when I looked at the patches.
>> > >
>> > > Yes, let's do this way. I don't know if we ever implemented the simple
>> > > patch to associate a USB device with a device_node, but if not, then
>> > > let's do it now for this driver. A lot of people have asked for it in
>> > > the past.
>> >
>> > Agreed. Also, some hubs have I2C buses as well, but I still think under
>> > the USB bus is the right place.
>> >
>> > However, one complication here is often (probably this case) these
>> > addtional signals need to be controlled before the device enumerates.
>> >
>>
>> Yes, I did not find a way to let the USB bus code handle it, so I had to
>> write a platform driver to do it
>
> Looping in Ulf, he solved the same problem for SDIO devices recently,
> and probably remembers the details best.
>

Thanks Arnd!

Yes, that was a kind of a long outstanding issue we have had for SDIO devices.

Several generic attempts was made to have a framework/library
available to support so called "power sequences" for exactly the same
reasons as above.
Others and myself failed to get those attempts accepted.

Instead, I invented a mmc subsystem specific DT based solution, the
"mmc-pwrseq".

DT documentation:
Documentation/devicetree/bindings/mmc/mmc-pwrseq-simple.txt
Documentation/devicetree/bindings/mmc/mmc.txt

Long story short: The mmc host device may contain a phandle to a
mmc-pwrseq, which will describe the resources needed to power on/off
the SDIO card.

The code is available at:
drivers/mmc/core/pwrseq*

We didn't implement this as platform driver, but that was mostly
because I initially wanted things to be simple. Although, nothing
prevents us from converting to this as a follow up, which would make
the solution a bit less "hacky".

Kind regards
Uffe