Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Andrew Bresticker]

Hi Roger,

On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq-l0cyMroinI0@public.gmane.org> wrote:
> Usage model:
> -----------
>
> - The OTG controller device is assumed to be the parent of
> the host and gadget controller. It must call usb_otg_register()
> before populating the host and gadget devices so that the OTG
> core is aware that it is an OTG device before the host & gadget
> register. The OTG controller must provide struct otg_fsm_ops *
> which will be called by the OTG core depending on OTG bus state.

I'm wondering if the requirement that the OTG controller be the parent
of the USB host/device-controllers makes sense.  For some context, I'm
working on adding dual-role support for Tegra210, specifically on a
system with USB Type-C.  On Tegra, the USB host-controller and USB
device-controller are two separate IP blocks (XUSB host and XUSB
device) with another, separate, IP block (XUSB padctl) for the USB PHY
and OTG support.  In the non-Type-C case, your OTG framework could
work well, though it's debatable as to whether or not the XUSB padctl
device should be a parent to the XUSB host/device-controller devices
(currently it isn't - it's just a PHY provider).  But in the Type-C
case, it's an off-chip embedded controller that determines the
dual-role status of the Type-C port, so the above requirement doesn't
make sense at all.

My idea was to have the OTG/DRD controller explicitly specify its host
and device controllers, so in DT, something like:

otg-controller {
    ...
    device-controller = <&usb_device>;
    host-controller = <&usb_host>;
    ...
};

usb_device: usb-device@.... {
    ...
};

usb_host: usb-host@... {
    ...
};

What do you think?

Thanks,
Andrew

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Peter Chen]

On Mon, Jul 13, 2015 at 12:14:43PM -0700, Andrew Bresticker wrote:
> Hi Roger,
> 
> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq@ti.com> wrote:
> > Usage model:
> > -----------
> >
> > - The OTG controller device is assumed to be the parent of
> > the host and gadget controller. It must call usb_otg_register()
> > before populating the host and gadget devices so that the OTG
> > core is aware that it is an OTG device before the host & gadget
> > register. The OTG controller must provide struct otg_fsm_ops *
> > which will be called by the OTG core depending on OTG bus state.
> 
> I'm wondering if the requirement that the OTG controller be the parent
> of the USB host/device-controllers makes sense.  For some context, I'm
> working on adding dual-role support for Tegra210, specifically on a
> system with USB Type-C.  On Tegra, the USB host-controller and USB
> device-controller are two separate IP blocks (XUSB host and XUSB
> device) with another, separate, IP block (XUSB padctl) for the USB PHY
> and OTG support.  In the non-Type-C case, your OTG framework could
> work well, though it's debatable as to whether or not the XUSB padctl
> device should be a parent to the XUSB host/device-controller devices
> (currently it isn't - it's just a PHY provider).  But in the Type-C
> case, it's an off-chip embedded controller that determines the
> dual-role status of the Type-C port, so the above requirement doesn't
> make sense at all.

Hi Andrew,

I think your problem is how to add your core driver to manage device and
host functionality together, and once you find how (through padctl/type-c
controller) to do it based on current code, it will be clear how to use roger
proposal framework at that time.

Most of current core drivers, we use extcon driver (through gpio) or USB
vbus/id pin (through internal registers) to manager roles.

> 
> My idea was to have the OTG/DRD controller explicitly specify its host
> and device controllers, so in DT, something like:
> 
> otg-controller {
>     ...
>     device-controller = <&usb_device>;
>     host-controller = <&usb_host>;
>     ...
> };
> 
> usb_device: usb-device@.... {
>     ...
> };
> 
> usb_host: usb-host@... {
>     ...
> };
> 
> What do you think?
> 
> Thanks,
> Andrew

-- 

Best Regards,
Peter Chen

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Andrew Bresticker]

Hi Peter,

On Mon, Jul 13, 2015 at 5:59 PM, Peter Chen <peter.chen@freescale.com> wrote:
> On Mon, Jul 13, 2015 at 12:14:43PM -0700, Andrew Bresticker wrote:
>> Hi Roger,
>>
>> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq@ti.com> wrote:
>> > Usage model:
>> > -----------
>> >
>> > - The OTG controller device is assumed to be the parent of
>> > the host and gadget controller. It must call usb_otg_register()
>> > before populating the host and gadget devices so that the OTG
>> > core is aware that it is an OTG device before the host & gadget
>> > register. The OTG controller must provide struct otg_fsm_ops *
>> > which will be called by the OTG core depending on OTG bus state.
>>
>> I'm wondering if the requirement that the OTG controller be the parent
>> of the USB host/device-controllers makes sense.  For some context, I'm
>> working on adding dual-role support for Tegra210, specifically on a
>> system with USB Type-C.  On Tegra, the USB host-controller and USB
>> device-controller are two separate IP blocks (XUSB host and XUSB
>> device) with another, separate, IP block (XUSB padctl) for the USB PHY
>> and OTG support.  In the non-Type-C case, your OTG framework could
>> work well, though it's debatable as to whether or not the XUSB padctl
>> device should be a parent to the XUSB host/device-controller devices
>> (currently it isn't - it's just a PHY provider).  But in the Type-C
>> case, it's an off-chip embedded controller that determines the
>> dual-role status of the Type-C port, so the above requirement doesn't
>> make sense at all.
>
> Hi Andrew,
>
> I think your problem is how to add your core driver to manage device and
> host functionality together, and once you find how (through padctl/type-c
> controller) to do it based on current code, it will be clear how to use roger
> proposal framework at that time.
>
> Most of current core drivers, we use extcon driver (through gpio) or USB
> vbus/id pin (through internal registers) to manager roles.

Right, currently I'm modeling the Type-C controller as an extcon
device and handle the role-changes in the core drivers, but that
doesn't really make sense for the non-Type-C case where we use the
XUSB padctl controller and need a full OTG state-machine.  Roger's new
OTG/DRD framework would fit my situation perfectly since it makes the
host/device-controller drivers independent from all the
OTG/role-changing logic.  The only issue is the requirement that the
OTG/DRD controller be the parent device of the host/device
controllers.

Thanks,
Andrew

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Peter Chen]

On Tue, Jul 14, 2015 at 11:18:30AM -0700, Andrew Bresticker wrote:
> Hi Peter,
> 
> On Mon, Jul 13, 2015 at 5:59 PM, Peter Chen <peter.chen@freescale.com> wrote:
> > On Mon, Jul 13, 2015 at 12:14:43PM -0700, Andrew Bresticker wrote:
> >> Hi Roger,
> >>
> >> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq@ti.com> wrote:
> >> > Usage model:
> >> > -----------
> >> >
> >> > - The OTG controller device is assumed to be the parent of
> >> > the host and gadget controller. It must call usb_otg_register()
> >> > before populating the host and gadget devices so that the OTG
> >> > core is aware that it is an OTG device before the host & gadget
> >> > register. The OTG controller must provide struct otg_fsm_ops *
> >> > which will be called by the OTG core depending on OTG bus state.
> >>
> >> I'm wondering if the requirement that the OTG controller be the parent
> >> of the USB host/device-controllers makes sense.  For some context, I'm
> >> working on adding dual-role support for Tegra210, specifically on a
> >> system with USB Type-C.  On Tegra, the USB host-controller and USB
> >> device-controller are two separate IP blocks (XUSB host and XUSB
> >> device) with another, separate, IP block (XUSB padctl) for the USB PHY
> >> and OTG support.  In the non-Type-C case, your OTG framework could
> >> work well, though it's debatable as to whether or not the XUSB padctl
> >> device should be a parent to the XUSB host/device-controller devices
> >> (currently it isn't - it's just a PHY provider).  But in the Type-C
> >> case, it's an off-chip embedded controller that determines the
> >> dual-role status of the Type-C port, so the above requirement doesn't
> >> make sense at all.
> >
> > Hi Andrew,
> >
> > I think your problem is how to add your core driver to manage device and
> > host functionality together, and once you find how (through padctl/type-c
> > controller) to do it based on current code, it will be clear how to use roger
> > proposal framework at that time.
> >
> > Most of current core drivers, we use extcon driver (through gpio) or USB
> > vbus/id pin (through internal registers) to manager roles.
> 
> Right, currently I'm modeling the Type-C controller as an extcon
> device and handle the role-changes in the core drivers, but that
> doesn't really make sense for the non-Type-C case where we use the
> XUSB padctl controller and need a full OTG state-machine.

The full OTG FSM is only applied if your board needs it, you can
disable it through dts. Jun [1] and Roger's patchset are for it.

> Roger's new
> OTG/DRD framework would fit my situation perfectly since it makes the
> host/device-controller drivers independent from all the
> OTG/role-changing logic.  The only issue is the requirement that the
> OTG/DRD controller be the parent device of the host/device
> controllers.
> 

The core device is the parent for host/device device, the OTG core
just use the pointer of it, Roger does an example using dwc3 [2].

[1] http://www.spinics.net/lists/linux-usb/msg127110.html
[2] http://www.spinics.net/lists/linux-usb/msg126999.html
-- 

Best Regards,
Peter Chen

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Roger Quadros]

Hi Andrew,

On 13/07/15 22:14, Andrew Bresticker wrote:
> Hi Roger,
> 
> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq@ti.com> wrote:
>> Usage model:
>> -----------
>>
>> - The OTG controller device is assumed to be the parent of
>> the host and gadget controller. It must call usb_otg_register()
>> before populating the host and gadget devices so that the OTG
>> core is aware that it is an OTG device before the host & gadget
>> register. The OTG controller must provide struct otg_fsm_ops *
>> which will be called by the OTG core depending on OTG bus state.
> 
> I'm wondering if the requirement that the OTG controller be the parent
> of the USB host/device-controllers makes sense.  For some context, I'm
> working on adding dual-role support for Tegra210, specifically on a
> system with USB Type-C.  On Tegra, the USB host-controller and USB
> device-controller are two separate IP blocks (XUSB host and XUSB
> device) with another, separate, IP block (XUSB padctl) for the USB PHY
> and OTG support.  In the non-Type-C case, your OTG framework could
> work well, though it's debatable as to whether or not the XUSB padctl
> device should be a parent to the XUSB host/device-controller devices
> (currently it isn't - it's just a PHY provider).  But in the Type-C
> case, it's an off-chip embedded controller that determines the
> dual-role status of the Type-C port, so the above requirement doesn't
> make sense at all.
> 
> My idea was to have the OTG/DRD controller explicitly specify its host
> and device controllers, so in DT, something like:
> 
> otg-controller {
>     ...
>     device-controller = <&usb_device>;
>     host-controller = <&usb_host>;
>     ...
> };
> 
> usb_device: usb-device@.... {
>     ...
> };
> 
> usb_host: usb-host@... {
>     ...
> };
> 
> What do you think?

I agree that we need to support your use case but how to do it
is not yet clear to me.

In your above example the otg controller knows what are the host
and gadget controllers but the host/gadget devices don't
know who is their otg controller.

So the problem is that when usb_otg_register_hcd/gadget() is called
we need to get a handle to the otg controller.

One solution I see is to iterate over the registered otg_controller_list
and check if we match the host/gadget controller in there.

Then there is also a possibility that host/gadget controllers get
registered before the OTG controller. Then we can't know for sure if
the host/gadget controller was meant for dual-role operation or not
and it will resort to single role operation.

Any idea to prevent the above situation?

Maybe we need to add some logic in host/gadget cores to check if the port
is meant for dual-role use and defer probe if OTG controller is not yet
registered?

cheers,
-roger

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Andrew Bresticker]

Hi Roger,

On Wed, Jul 15, 2015 at 6:26 AM, Roger Quadros <rogerq-l0cyMroinI0@public.gmane.org> wrote:
> Hi Andrew,
>
> On 13/07/15 22:14, Andrew Bresticker wrote:
>> Hi Roger,
>>
>> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq-l0cyMroinI0@public.gmane.org> wrote:
>>> Usage model:
>>> -----------
>>>
>>> - The OTG controller device is assumed to be the parent of
>>> the host and gadget controller. It must call usb_otg_register()
>>> before populating the host and gadget devices so that the OTG
>>> core is aware that it is an OTG device before the host & gadget
>>> register. The OTG controller must provide struct otg_fsm_ops *
>>> which will be called by the OTG core depending on OTG bus state.
>>
>> I'm wondering if the requirement that the OTG controller be the parent
>> of the USB host/device-controllers makes sense.  For some context, I'm
>> working on adding dual-role support for Tegra210, specifically on a
>> system with USB Type-C.  On Tegra, the USB host-controller and USB
>> device-controller are two separate IP blocks (XUSB host and XUSB
>> device) with another, separate, IP block (XUSB padctl) for the USB PHY
>> and OTG support.  In the non-Type-C case, your OTG framework could
>> work well, though it's debatable as to whether or not the XUSB padctl
>> device should be a parent to the XUSB host/device-controller devices
>> (currently it isn't - it's just a PHY provider).  But in the Type-C
>> case, it's an off-chip embedded controller that determines the
>> dual-role status of the Type-C port, so the above requirement doesn't
>> make sense at all.
>>
>> My idea was to have the OTG/DRD controller explicitly specify its host
>> and device controllers, so in DT, something like:
>>
>> otg-controller {
>>     ...
>>     device-controller = <&usb_device>;
>>     host-controller = <&usb_host>;
>>     ...
>> };
>>
>> usb_device: usb-device@.... {
>>     ...
>> };
>>
>> usb_host: usb-host@... {
>>     ...
>> };
>>
>> What do you think?
>
> I agree that we need to support your use case but how to do it
> is not yet clear to me.
>
> In your above example the otg controller knows what are the host
> and gadget controllers but the host/gadget devices don't
> know who is their otg controller.
>
> So the problem is that when usb_otg_register_hcd/gadget() is called
> we need to get a handle to the otg controller.
>
> One solution I see is to iterate over the registered otg_controller_list
> and check if we match the host/gadget controller in there.
>
> Then there is also a possibility that host/gadget controllers get
> registered before the OTG controller. Then we can't know for sure if
> the host/gadget controller was meant for dual-role operation or not
> and it will resort to single role operation.
>
> Any idea to prevent the above situation?
>
> Maybe we need to add some logic in host/gadget cores to check if the port
> is meant for dual-role use and defer probe if OTG controller is not yet
> registered?

In the DT case, I think we could add an "otg-controller" property to
the host and gadget nodes, and in usb_otg_register_{hcd,gadget}()
check for that property and defer probe if the referenced OTG
controller has not yet been registered.  Not sure how to indicate that
a host/gadget is meant for dual-role operation on non-DT platforms
though.

Thanks,
Andrew

Re: [PATCH v3 00/11] USB: OTG/DRD Core functionality

[Roger Quadros]

On 16/07/15 21:29, Andrew Bresticker wrote:
> Hi Roger,
> 
> On Wed, Jul 15, 2015 at 6:26 AM, Roger Quadros <rogerq@ti.com> wrote:
>> Hi Andrew,
>>
>> On 13/07/15 22:14, Andrew Bresticker wrote:
>>> Hi Roger,
>>>
>>> On Wed, Jul 8, 2015 at 3:19 AM, Roger Quadros <rogerq@ti.com> wrote:
>>>> Usage model:
>>>> -----------
>>>>
>>>> - The OTG controller device is assumed to be the parent of
>>>> the host and gadget controller. It must call usb_otg_register()
>>>> before populating the host and gadget devices so that the OTG
>>>> core is aware that it is an OTG device before the host & gadget
>>>> register. The OTG controller must provide struct otg_fsm_ops *
>>>> which will be called by the OTG core depending on OTG bus state.
>>>
>>> I'm wondering if the requirement that the OTG controller be the parent
>>> of the USB host/device-controllers makes sense.  For some context, I'm
>>> working on adding dual-role support for Tegra210, specifically on a
>>> system with USB Type-C.  On Tegra, the USB host-controller and USB
>>> device-controller are two separate IP blocks (XUSB host and XUSB
>>> device) with another, separate, IP block (XUSB padctl) for the USB PHY
>>> and OTG support.  In the non-Type-C case, your OTG framework could
>>> work well, though it's debatable as to whether or not the XUSB padctl
>>> device should be a parent to the XUSB host/device-controller devices
>>> (currently it isn't - it's just a PHY provider).  But in the Type-C
>>> case, it's an off-chip embedded controller that determines the
>>> dual-role status of the Type-C port, so the above requirement doesn't
>>> make sense at all.
>>>
>>> My idea was to have the OTG/DRD controller explicitly specify its host
>>> and device controllers, so in DT, something like:
>>>
>>> otg-controller {
>>>     ...
>>>     device-controller = <&usb_device>;
>>>     host-controller = <&usb_host>;
>>>     ...
>>> };
>>>
>>> usb_device: usb-device@.... {
>>>     ...
>>> };
>>>
>>> usb_host: usb-host@... {
>>>     ...
>>> };
>>>
>>> What do you think?
>>
>> I agree that we need to support your use case but how to do it
>> is not yet clear to me.
>>
>> In your above example the otg controller knows what are the host
>> and gadget controllers but the host/gadget devices don't
>> know who is their otg controller.
>>
>> So the problem is that when usb_otg_register_hcd/gadget() is called
>> we need to get a handle to the otg controller.
>>
>> One solution I see is to iterate over the registered otg_controller_list
>> and check if we match the host/gadget controller in there.
>>
>> Then there is also a possibility that host/gadget controllers get
>> registered before the OTG controller. Then we can't know for sure if
>> the host/gadget controller was meant for dual-role operation or not
>> and it will resort to single role operation.
>>
>> Any idea to prevent the above situation?
>>
>> Maybe we need to add some logic in host/gadget cores to check if the port
>> is meant for dual-role use and defer probe if OTG controller is not yet
>> registered?
> 
> In the DT case, I think we could add an "otg-controller" property to
> the host and gadget nodes, and in usb_otg_register_{hcd,gadget}()
> check for that property and defer probe if the referenced OTG
> controller has not yet been registered.  Not sure how to indicate that
> a host/gadget is meant for dual-role operation on non-DT platforms
> though.
> 

Sounds fine for DT case.

For non DT case I don't see any other way to connect the OTG controller
to the host/gadget controllers other than the parent child relationship.

Maybe for non DT case we can still have that constraint.
We can have this constraint regardless so that DT users who fit
in that model don't need to add 'otg-controller' property.

cheers,
-roger